2 * Copyright (c) 2000-2012 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 <kern/telemetry.h>
95 #include <sys/kdebug.h>
96 #include <prng/random.h>
100 #include <i386/postcode.h>
101 #include <i386/mp_desc.h>
102 #include <i386/proc_reg.h>
104 #include <i386/machine_check.h>
106 #include <mach/i386/syscall_sw.h>
108 #include <libkern/OSDebug.h>
109 #include <i386/cpu_threads.h>
110 #include <machine/pal_routines.h>
112 extern void throttle_lowpri_io(int);
113 extern void kprint_state(x86_saved_state64_t
*saved_state
);
116 * Forward declarations
118 static void user_page_fault_continue(kern_return_t kret
);
119 static void panic_trap(x86_saved_state64_t
*saved_state
);
120 static void set_recovery_ip(x86_saved_state64_t
*saved_state
, vm_offset_t ip
);
122 volatile perfCallback perfTrapHook
= NULL
; /* Pointer to CHUD trap hook routine */
125 /* See <rdar://problem/4613924> */
126 perfCallback tempDTraceTrapHook
= NULL
; /* Pointer to DTrace fbt trap hook routine */
128 extern boolean_t
dtrace_tally_fault(user_addr_t
);
131 extern boolean_t pmap_smep_enabled
;
132 extern boolean_t pmap_smap_enabled
;
135 thread_syscall_return(
138 thread_t thr_act
= current_thread();
142 pal_register_cache_state(thr_act
, DIRTY
);
144 if (thread_is_64bit(thr_act
)) {
145 x86_saved_state64_t
*regs
;
147 regs
= USER_REGS64(thr_act
);
149 code
= (int) (regs
->rax
& SYSCALL_NUMBER_MASK
);
150 is_mach
= (regs
->rax
& SYSCALL_CLASS_MASK
)
151 == (SYSCALL_CLASS_MACH
<< SYSCALL_CLASS_SHIFT
);
152 if (kdebug_enable
&& is_mach
) {
154 KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE
,
155 MACHDBG_CODE(DBG_MACH_EXCP_SC
,code
)|DBG_FUNC_END
,
161 DEBUG_KPRINT_SYSCALL_MACH(
162 "thread_syscall_return: 64-bit mach ret=%u\n",
165 DEBUG_KPRINT_SYSCALL_UNIX(
166 "thread_syscall_return: 64-bit unix ret=%u\n",
170 x86_saved_state32_t
*regs
;
172 regs
= USER_REGS32(thr_act
);
174 code
= ((int) regs
->eax
);
175 is_mach
= (code
< 0);
176 if (kdebug_enable
&& is_mach
) {
178 KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE
,
179 MACHDBG_CODE(DBG_MACH_EXCP_SC
,-code
)|DBG_FUNC_END
,
185 DEBUG_KPRINT_SYSCALL_MACH(
186 "thread_syscall_return: 32-bit mach ret=%u\n",
189 DEBUG_KPRINT_SYSCALL_UNIX(
190 "thread_syscall_return: 32-bit unix ret=%u\n",
194 throttle_lowpri_io(1);
196 thread_exception_return();
202 user_page_fault_continue(
205 thread_t thread
= current_thread();
208 if (thread_is_64bit(thread
)) {
209 x86_saved_state64_t
*uregs
;
211 uregs
= USER_REGS64(thread
);
213 vaddr
= (user_addr_t
)uregs
->cr2
;
215 x86_saved_state32_t
*uregs
;
217 uregs
= USER_REGS32(thread
);
224 pal_dbg_page_fault( thread
, vaddr
, kr
);
226 i386_exception(EXC_BAD_ACCESS
, kr
, vaddr
);
231 * Fault recovery in copyin/copyout routines.
234 uintptr_t fault_addr
;
235 uintptr_t recover_addr
;
238 extern struct recovery recover_table
[];
239 extern struct recovery recover_table_end
[];
241 const char * trap_type
[] = {TRAP_NAMES
};
242 unsigned TRAP_TYPES
= sizeof(trap_type
)/sizeof(trap_type
[0]);
244 extern void PE_incoming_interrupt(int interrupt
);
246 #if defined(__x86_64__) && DEBUG
248 kprint_state(x86_saved_state64_t
*saved_state
)
250 kprintf("current_cpu_datap() 0x%lx\n", (uintptr_t)current_cpu_datap());
251 kprintf("Current GS base MSR 0x%llx\n", rdmsr64(MSR_IA32_GS_BASE
));
252 kprintf("Kernel GS base MSR 0x%llx\n", rdmsr64(MSR_IA32_KERNEL_GS_BASE
));
253 kprintf("state at 0x%lx:\n", (uintptr_t) saved_state
);
255 kprintf(" rdi 0x%llx\n", saved_state
->rdi
);
256 kprintf(" rsi 0x%llx\n", saved_state
->rsi
);
257 kprintf(" rdx 0x%llx\n", saved_state
->rdx
);
258 kprintf(" r10 0x%llx\n", saved_state
->r10
);
259 kprintf(" r8 0x%llx\n", saved_state
->r8
);
260 kprintf(" r9 0x%llx\n", saved_state
->r9
);
262 kprintf(" cr2 0x%llx\n", saved_state
->cr2
);
263 kprintf("real cr2 0x%lx\n", get_cr2());
264 kprintf(" r15 0x%llx\n", saved_state
->r15
);
265 kprintf(" r14 0x%llx\n", saved_state
->r14
);
266 kprintf(" r13 0x%llx\n", saved_state
->r13
);
267 kprintf(" r12 0x%llx\n", saved_state
->r12
);
268 kprintf(" r11 0x%llx\n", saved_state
->r11
);
269 kprintf(" rbp 0x%llx\n", saved_state
->rbp
);
270 kprintf(" rbx 0x%llx\n", saved_state
->rbx
);
271 kprintf(" rcx 0x%llx\n", saved_state
->rcx
);
272 kprintf(" rax 0x%llx\n", saved_state
->rax
);
274 kprintf(" gs 0x%x\n", saved_state
->gs
);
275 kprintf(" fs 0x%x\n", saved_state
->fs
);
277 kprintf(" isf.trapno 0x%x\n", saved_state
->isf
.trapno
);
278 kprintf(" isf._pad 0x%x\n", saved_state
->isf
._pad
);
279 kprintf(" isf.trapfn 0x%llx\n", saved_state
->isf
.trapfn
);
280 kprintf(" isf.err 0x%llx\n", saved_state
->isf
.err
);
281 kprintf(" isf.rip 0x%llx\n", saved_state
->isf
.rip
);
282 kprintf(" isf.cs 0x%llx\n", saved_state
->isf
.cs
);
283 kprintf(" isf.rflags 0x%llx\n", saved_state
->isf
.rflags
);
284 kprintf(" isf.rsp 0x%llx\n", saved_state
->isf
.rsp
);
285 kprintf(" isf.ss 0x%llx\n", saved_state
->isf
.ss
);
291 * Non-zero indicates latency assert is enabled and capped at valued
292 * absolute time units.
295 uint64_t interrupt_latency_cap
= 0;
296 boolean_t ilat_assert
= FALSE
;
299 interrupt_latency_tracker_setup(void) {
300 uint32_t ilat_cap_us
;
301 if (PE_parse_boot_argn("interrupt_latency_cap_us", &ilat_cap_us
, sizeof(ilat_cap_us
))) {
302 interrupt_latency_cap
= ilat_cap_us
* NSEC_PER_USEC
;
303 nanoseconds_to_absolutetime(interrupt_latency_cap
, &interrupt_latency_cap
);
305 interrupt_latency_cap
= LockTimeOut
;
307 PE_parse_boot_argn("-interrupt_latency_assert_enable", &ilat_assert
, sizeof(ilat_assert
));
310 void interrupt_reset_latency_stats(void) {
312 for (i
= 0; i
< real_ncpus
; i
++) {
313 cpu_data_ptr
[i
]->cpu_max_observed_int_latency
=
314 cpu_data_ptr
[i
]->cpu_max_observed_int_latency_vector
= 0;
318 void interrupt_populate_latency_stats(char *buf
, unsigned bufsize
) {
319 uint32_t i
, tcpu
= ~0;
320 uint64_t cur_max
= 0;
322 for (i
= 0; i
< real_ncpus
; i
++) {
323 if (cur_max
< cpu_data_ptr
[i
]->cpu_max_observed_int_latency
) {
324 cur_max
= cpu_data_ptr
[i
]->cpu_max_observed_int_latency
;
329 if (tcpu
< real_ncpus
)
330 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
);
333 uint32_t interrupt_timer_coalescing_enabled
= 1;
334 uint64_t interrupt_coalesced_timers
;
338 * - local APIC interrupts (IPIs, timers, etc) are handled by the kernel,
339 * - device interrupts go to the platform expert.
342 interrupt(x86_saved_state_t
*state
)
347 boolean_t user_mode
= FALSE
;
349 int cnum
= cpu_number();
350 cpu_data_t
*cdp
= cpu_data_ptr
[cnum
];
353 if (is_saved_state64(state
) == TRUE
) {
354 x86_saved_state64_t
*state64
;
356 state64
= saved_state64(state
);
357 rip
= state64
->isf
.rip
;
358 rsp
= state64
->isf
.rsp
;
359 interrupt_num
= state64
->isf
.trapno
;
361 if(state64
->isf
.cs
& 0x03)
365 x86_saved_state32_t
*state32
;
367 state32
= saved_state32(state
);
368 if (state32
->cs
& 0x03)
372 interrupt_num
= state32
->trapno
;
375 if (cpu_data_ptr
[cnum
]->lcpu
.package
->num_idle
== topoParms
.nLThreadsPerPackage
)
376 cpu_data_ptr
[cnum
]->cpu_hwIntpexits
[interrupt_num
]++;
378 if (interrupt_num
== (LAPIC_DEFAULT_INTERRUPT_BASE
+ LAPIC_INTERPROCESSOR_INTERRUPT
))
380 else if (interrupt_num
== (LAPIC_DEFAULT_INTERRUPT_BASE
+ LAPIC_TIMER_INTERRUPT
))
385 KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE
,
386 MACHDBG_CODE(DBG_MACH_EXCP_INTR
, 0) | DBG_FUNC_START
,
388 (user_mode
? rip
: VM_KERNEL_UNSLIDE(rip
)),
389 user_mode
, itype
, 0);
391 SCHED_STATS_INTERRUPT(current_processor());
394 if (telemetry_needs_record
395 && (current_task() != kernel_task
)
396 #if CONFIG_SCHED_IDLE_IN_PLACE
397 && ((current_thread()->state
& TH_IDLE
) == 0) /* idle-in-place should be treated like the idle thread */
400 telemetry_mark_curthread(user_mode
);
404 ipl
= get_preemption_level();
407 * Handle local APIC interrupts
408 * else call platform expert for devices.
410 if (!lapic_interrupt(interrupt_num
, state
)) {
411 PE_incoming_interrupt(interrupt_num
);
414 if (__improbable(get_preemption_level() != ipl
)) {
415 panic("Preemption level altered by interrupt vector 0x%x: initial 0x%x, final: 0x%x\n", interrupt_num
, ipl
, get_preemption_level());
419 if (__improbable(cdp
->cpu_nested_istack
)) {
420 cdp
->cpu_nested_istack_events
++;
423 uint64_t ctime
= mach_absolute_time();
424 uint64_t int_latency
= ctime
- cdp
->cpu_int_event_time
;
425 uint64_t esdeadline
, ehdeadline
;
426 /* Attempt to process deferred timers in the context of
427 * this interrupt, unless interrupt time has already exceeded
428 * TCOAL_ILAT_THRESHOLD.
430 #define TCOAL_ILAT_THRESHOLD (30000ULL)
432 if ((int_latency
< TCOAL_ILAT_THRESHOLD
) &&
433 interrupt_timer_coalescing_enabled
) {
434 esdeadline
= cdp
->rtclock_timer
.queue
.earliest_soft_deadline
;
435 ehdeadline
= cdp
->rtclock_timer
.deadline
;
436 if ((ctime
>= esdeadline
) && (ctime
< ehdeadline
)) {
437 interrupt_coalesced_timers
++;
438 TCOAL_DEBUG(0x88880000 | DBG_FUNC_START
, ctime
, esdeadline
, ehdeadline
, interrupt_coalesced_timers
, 0);
440 TCOAL_DEBUG(0x88880000 | DBG_FUNC_END
, ctime
, esdeadline
, interrupt_coalesced_timers
, 0, 0);
442 TCOAL_DEBUG(0x77770000, ctime
, cdp
->rtclock_timer
.queue
.earliest_soft_deadline
, cdp
->rtclock_timer
.deadline
, interrupt_coalesced_timers
, 0);
446 if (__improbable(ilat_assert
&& (int_latency
> interrupt_latency_cap
) && !machine_timeout_suspended())) {
447 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
, cdp
->cpu_prior_signals
, cdp
->cpu_signals
);
450 if (__improbable(int_latency
> cdp
->cpu_max_observed_int_latency
)) {
451 cdp
->cpu_max_observed_int_latency
= int_latency
;
452 cdp
->cpu_max_observed_int_latency_vector
= interrupt_num
;
457 * Having serviced the interrupt first, look at the interrupted stack depth.
460 uint64_t depth
= cdp
->cpu_kernel_stack
461 + sizeof(struct x86_kernel_state
)
462 + sizeof(struct i386_exception_link
*)
464 if (__improbable(depth
> kernel_stack_depth_max
)) {
465 kernel_stack_depth_max
= (vm_offset_t
)depth
;
466 KERNEL_DEBUG_CONSTANT(
467 MACHDBG_CODE(DBG_MACH_SCHED
, MACH_STACK_DEPTH
),
468 (long) depth
, (long) VM_KERNEL_UNSLIDE(rip
), 0, 0, 0);
472 if (cnum
== master_cpu
)
473 ml_entropy_collect();
475 KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE
,
476 MACHDBG_CODE(DBG_MACH_EXCP_INTR
, 0) | DBG_FUNC_END
,
477 interrupt_num
, 0, 0, 0, 0);
484 long dr7
= 0x400; /* magic dr7 reset value; 32 bit on i386, 64 bit on x86_64 */
485 __asm__
volatile("mov %0,%%dr7" : : "r" (dr7
));
488 unsigned kdp_has_active_watchpoints
= 0;
489 #define NO_WATCHPOINTS (!kdp_has_active_watchpoints)
491 #define NO_WATCHPOINTS 1
494 * Trap from kernel mode. Only page-fault errors are recoverable,
495 * and then only in special circumstances. All other errors are
496 * fatal. Return value indicates if trap was handled.
501 x86_saved_state_t
*state
,
504 x86_saved_state64_t
*saved_state
;
508 vm_map_t map
= 0; /* protected by T_PAGE_FAULT */
509 kern_return_t result
= KERN_FAILURE
;
516 #if NCOPY_WINDOWS > 0
517 int fault_in_copy_window
= -1;
521 thread
= current_thread();
523 if (__improbable(is_saved_state32(state
)))
524 panic("kernel_trap(%p) with 32-bit state", state
);
525 saved_state
= saved_state64(state
);
527 /* Record cpu where state was captured */
528 saved_state
->isf
.cpu
= cpu_number();
530 vaddr
= (user_addr_t
)saved_state
->cr2
;
531 type
= saved_state
->isf
.trapno
;
532 code
= (int)(saved_state
->isf
.err
& 0xffff);
533 intr
= (saved_state
->isf
.rflags
& EFL_IF
) != 0; /* state of ints at trap */
534 kern_ip
= (vm_offset_t
)saved_state
->isf
.rip
;
536 myast
= ast_pending();
538 perfASTCallback astfn
= perfASTHook
;
539 if (__improbable(astfn
!= NULL
)) {
540 if (*myast
& AST_CHUD_ALL
)
541 astfn(AST_CHUD_ALL
, myast
);
543 *myast
&= ~AST_CHUD_ALL
;
548 * Is there a DTrace hook?
550 if (__improbable(tempDTraceTrapHook
!= NULL
)) {
551 if (tempDTraceTrapHook(type
, state
, lo_spp
, 0) == KERN_SUCCESS
) {
553 * If it succeeds, we are done...
558 #endif /* CONFIG_DTRACE */
561 * we come here with interrupts off as we don't want to recurse
562 * on preemption below. but we do want to re-enable interrupts
563 * as soon we possibly can to hold latency down
565 if (__improbable(T_PREEMPT
== type
)) {
566 ast_taken(AST_PREEMPTION
, FALSE
);
568 KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE
,
569 (MACHDBG_CODE(DBG_MACH_EXCP_KTRAP_x86
, type
)) | DBG_FUNC_NONE
,
570 0, 0, 0, VM_KERNEL_UNSLIDE(kern_ip
), 0);
574 if (T_PAGE_FAULT
== type
) {
576 * assume we're faulting in the kernel map
580 if (__probable(thread
!= THREAD_NULL
&& thread
->map
!= kernel_map
)) {
581 #if NCOPY_WINDOWS > 0
582 vm_offset_t copy_window_base
;
586 kvaddr
= (vm_offset_t
)vaddr
;
588 * must determine if fault occurred in
589 * the copy window while pre-emption is
590 * disabled for this processor so that
591 * we only need to look at the window
592 * associated with this processor
594 copy_window_base
= current_cpu_datap()->cpu_copywindow_base
;
596 if (kvaddr
>= copy_window_base
&& kvaddr
< (copy_window_base
+ (NBPDE
* NCOPY_WINDOWS
)) ) {
598 window_index
= (int)((kvaddr
- copy_window_base
) / NBPDE
);
600 if (thread
->machine
.copy_window
[window_index
].user_base
!= (user_addr_t
)-1) {
602 kvaddr
-= (copy_window_base
+ (NBPDE
* window_index
));
603 vaddr
= thread
->machine
.copy_window
[window_index
].user_base
+ kvaddr
;
606 fault_in_copy_window
= window_index
;
611 if (__probable(vaddr
< VM_MAX_USER_PAGE_ADDRESS
)) {
612 /* fault occurred in userspace */
616 /* Intercept a potential Supervisor Mode Execute
617 * Protection fault. These criteria identify
618 * both NX faults and SMEP faults, but both
619 * are fatal. We avoid checking PTEs (racy).
620 * (The VM could just redrive a SMEP fault, hence
623 if (__improbable((code
== (T_PF_PROT
| T_PF_EXECUTE
)) && (pmap_smep_enabled
) && (saved_state
->isf
.rip
== vaddr
))) {
628 * Additionally check for SMAP faults...
629 * which are characterized by page-present and
630 * the AC bit unset (i.e. not from copyin/out path).
632 if (__improbable(code
& T_PF_PROT
&&
634 (saved_state
->isf
.rflags
& EFL_AC
) == 0)) {
639 * If we're not sharing cr3 with the user
640 * and we faulted in copyio,
641 * then switch cr3 here and dismiss the fault.
644 (thread
->machine
.specFlags
&CopyIOActive
) &&
645 map
->pmap
->pm_cr3
!= get_cr3_base()) {
646 pmap_assert(current_cpu_datap()->cpu_pmap_pcid_enabled
== FALSE
);
647 set_cr3_raw(map
->pmap
->pm_cr3
);
655 user_addr_t kd_vaddr
= is_user
? vaddr
: VM_KERNEL_UNSLIDE(vaddr
);
656 KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE
,
657 (MACHDBG_CODE(DBG_MACH_EXCP_KTRAP_x86
, type
)) | DBG_FUNC_NONE
,
658 (unsigned)(kd_vaddr
>> 32), (unsigned)kd_vaddr
, is_user
,
659 VM_KERNEL_UNSLIDE(kern_ip
), 0);
662 (void) ml_set_interrupts_enabled(intr
);
674 case T_FLOATING_POINT_ERROR
:
678 case T_SSE_FLOAT_ERROR
:
682 if ((saved_state
->isf
.rflags
& EFL_TF
) == 0 && NO_WATCHPOINTS
)
684 /* We've somehow encountered a debug
685 * register match that does not belong
686 * to the kernel debugger.
687 * This isn't supposed to happen.
700 if (thread
!= THREAD_NULL
&& thread
->options
& TH_OPT_DTRACE
) { /* Executing under dtrace_probe? */
701 if (dtrace_tally_fault(vaddr
)) { /* Should a fault under dtrace be ignored? */
703 * DTrace has "anticipated" the possibility of this fault, and has
704 * established the suitable recovery state. Drop down now into the
705 * recovery handling code in "case T_GENERAL_PROTECTION:".
710 #endif /* CONFIG_DTRACE */
714 if (code
& T_PF_WRITE
)
715 prot
|= VM_PROT_WRITE
;
716 if (code
& T_PF_EXECUTE
)
717 prot
|= VM_PROT_EXECUTE
;
719 result
= vm_fault(map
,
720 vm_map_trunc_page(vaddr
,
724 THREAD_UNINT
, NULL
, 0);
726 if (result
== KERN_SUCCESS
) {
727 #if NCOPY_WINDOWS > 0
728 if (fault_in_copy_window
!= -1) {
729 ml_set_interrupts_enabled(FALSE
);
730 copy_window_fault(thread
, map
,
731 fault_in_copy_window
);
732 (void) ml_set_interrupts_enabled(intr
);
734 #endif /* NCOPY_WINDOWS > 0 */
742 #endif /* CONFIG_DTRACE */
744 case T_GENERAL_PROTECTION
:
746 * If there is a failure recovery address
747 * for this fault, go there.
749 for (rp
= recover_table
; rp
< recover_table_end
; rp
++) {
750 if (kern_ip
== rp
->fault_addr
) {
751 set_recovery_ip(saved_state
, rp
->recover_addr
);
757 * Check thread recovery address also.
759 if (thread
!= THREAD_NULL
&& thread
->recover
) {
760 set_recovery_ip(saved_state
, thread
->recover
);
765 * Unanticipated page-fault errors in kernel
772 * Exception 15 is reserved but some chips may generate it
773 * spuriously. Seen at startup on AMD Athlon-64.
776 kprintf("kernel_trap() ignoring spurious trap 15\n");
780 /* Ensure that the i386_kernel_state at the base of the
781 * current thread's stack (if any) is synchronized with the
782 * context at the moment of the trap, to facilitate
783 * access through the debugger.
785 sync_iss_to_iks(state
);
787 if (current_debugger
!= KDB_CUR_DB
) {
788 if (kdp_i386_trap(type
, saved_state
, result
, (vm_offset_t
)vaddr
))
794 panic_trap(saved_state
);
802 set_recovery_ip(x86_saved_state64_t
*saved_state
, vm_offset_t ip
)
804 saved_state
->isf
.rip
= ip
;
811 panic_trap(x86_saved_state64_t
*regs
)
813 const char *trapname
= "Unknown";
814 pal_cr_t cr0
, cr2
, cr3
, cr4
;
815 boolean_t potential_smep_fault
= FALSE
, potential_kernel_NX_fault
= FALSE
;
816 boolean_t potential_smap_fault
= FALSE
;
818 pal_get_control_registers( &cr0
, &cr2
, &cr3
, &cr4
);
819 assert(ml_get_interrupts_enabled() == FALSE
);
820 current_cpu_datap()->cpu_fatal_trap_state
= regs
;
822 * Issue an I/O port read if one has been requested - this is an
823 * event logic analyzers can use as a trigger point.
825 panic_io_port_read();
827 kprintf("panic trap number 0x%x, rip 0x%016llx\n",
828 regs
->isf
.trapno
, regs
->isf
.rip
);
829 kprintf("cr0 0x%016llx cr2 0x%016llx cr3 0x%016llx cr4 0x%016llx\n",
832 if (regs
->isf
.trapno
< TRAP_TYPES
)
833 trapname
= trap_type
[regs
->isf
.trapno
];
835 if ((regs
->isf
.trapno
== T_PAGE_FAULT
) && (regs
->isf
.err
== (T_PF_PROT
| T_PF_EXECUTE
)) && (regs
->isf
.rip
== regs
->cr2
)) {
836 if (pmap_smep_enabled
&& (regs
->isf
.rip
< VM_MAX_USER_PAGE_ADDRESS
)) {
837 potential_smep_fault
= TRUE
;
838 } else if (regs
->isf
.rip
>= VM_MIN_KERNEL_AND_KEXT_ADDRESS
) {
839 potential_kernel_NX_fault
= TRUE
;
841 } else if (pmap_smap_enabled
&&
842 regs
->isf
.trapno
== T_PAGE_FAULT
&&
843 regs
->isf
.err
& T_PF_PROT
&&
844 regs
->cr2
< VM_MAX_USER_PAGE_ADDRESS
&&
845 regs
->isf
.rip
>= VM_MIN_KERNEL_AND_KEXT_ADDRESS
) {
846 potential_smap_fault
= TRUE
;
850 panic("Kernel trap at 0x%016llx, type %d=%s, registers:\n"
851 "CR0: 0x%016llx, CR2: 0x%016llx, CR3: 0x%016llx, CR4: 0x%016llx\n"
852 "RAX: 0x%016llx, RBX: 0x%016llx, RCX: 0x%016llx, RDX: 0x%016llx\n"
853 "RSP: 0x%016llx, RBP: 0x%016llx, RSI: 0x%016llx, RDI: 0x%016llx\n"
854 "R8: 0x%016llx, R9: 0x%016llx, R10: 0x%016llx, R11: 0x%016llx\n"
855 "R12: 0x%016llx, R13: 0x%016llx, R14: 0x%016llx, R15: 0x%016llx\n"
856 "RFL: 0x%016llx, RIP: 0x%016llx, CS: 0x%016llx, SS: 0x%016llx\n"
857 "Fault CR2: 0x%016llx, Error code: 0x%016llx, Fault CPU: 0x%x%s%s%s%s\n",
858 regs
->isf
.rip
, regs
->isf
.trapno
, trapname
,
860 regs
->rax
, regs
->rbx
, regs
->rcx
, regs
->rdx
,
861 regs
->isf
.rsp
, regs
->rbp
, regs
->rsi
, regs
->rdi
,
862 regs
->r8
, regs
->r9
, regs
->r10
, regs
->r11
,
863 regs
->r12
, regs
->r13
, regs
->r14
, regs
->r15
,
864 regs
->isf
.rflags
, regs
->isf
.rip
, regs
->isf
.cs
& 0xFFFF,
865 regs
->isf
.ss
& 0xFFFF,regs
->cr2
, regs
->isf
.err
, regs
->isf
.cpu
,
866 virtualized
? " VMM" : "",
867 potential_kernel_NX_fault
? " Kernel NX fault" : "",
868 potential_smep_fault
? " SMEP/User NX fault" : "",
869 potential_smap_fault
? " SMAP fault" : "");
871 * This next statement is not executed,
872 * but it's needed to stop the compiler using tail call optimization
873 * for the panic call - which confuses the subsequent backtrace.
879 extern kern_return_t
dtrace_user_probe(x86_saved_state_t
*);
883 * Trap from user mode.
887 x86_saved_state_t
*saved_state
)
891 mach_exception_code_t code
;
892 mach_exception_subcode_t subcode
;
896 thread_t thread
= current_thread();
900 unsigned long dr6
= 0; /* 32 bit for i386, 64 bit for x86_64 */
902 assert((is_saved_state32(saved_state
) && !thread_is_64bit(thread
)) ||
903 (is_saved_state64(saved_state
) && thread_is_64bit(thread
)));
905 if (is_saved_state64(saved_state
)) {
906 x86_saved_state64_t
*regs
;
908 regs
= saved_state64(saved_state
);
910 /* Record cpu where state was captured */
911 regs
->isf
.cpu
= cpu_number();
913 type
= regs
->isf
.trapno
;
914 err
= (int)regs
->isf
.err
& 0xffff;
915 vaddr
= (user_addr_t
)regs
->cr2
;
916 rip
= (user_addr_t
)regs
->isf
.rip
;
918 x86_saved_state32_t
*regs
;
920 regs
= saved_state32(saved_state
);
922 /* Record cpu where state was captured */
923 regs
->cpu
= cpu_number();
926 err
= regs
->err
& 0xffff;
927 vaddr
= (user_addr_t
)regs
->cr2
;
928 rip
= (user_addr_t
)regs
->eip
;
931 if ((type
== T_DEBUG
) && thread
->machine
.ids
) {
932 unsigned long clear
= 0;
933 /* Stash and clear this processor's DR6 value, in the event
934 * this was a debug register match
936 __asm__
volatile ("mov %%db6, %0" : "=r" (dr6
));
937 __asm__
volatile ("mov %0, %%db6" : : "r" (clear
));
942 KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE
,
943 (MACHDBG_CODE(DBG_MACH_EXCP_UTRAP_x86
, type
)) | DBG_FUNC_NONE
,
944 (unsigned)(vaddr
>>32), (unsigned)vaddr
,
945 (unsigned)(rip
>>32), (unsigned)rip
, 0);
952 kprintf("user_trap(0x%08x) type=%d vaddr=0x%016llx\n",
953 saved_state
, type
, vaddr
);
956 perfASTCallback astfn
= perfASTHook
;
957 if (__improbable(astfn
!= NULL
)) {
958 myast
= ast_pending();
959 if (*myast
& AST_CHUD_ALL
) {
960 astfn(AST_CHUD_ALL
, myast
);
964 /* Is there a hook? */
965 perfCallback fn
= perfTrapHook
;
966 if (__improbable(fn
!= NULL
)) {
967 if (fn(type
, saved_state
, 0, 0) == KERN_SUCCESS
)
968 return; /* If it succeeds, we are done... */
973 * DTrace does not consume all user traps, only INT_3's for now.
974 * Avoid needlessly calling tempDTraceTrapHook here, and let the
975 * INT_3 case handle them.
979 DEBUG_KPRINT_SYSCALL_MASK(1,
980 "user_trap: type=0x%x(%s) err=0x%x cr2=%p rip=%p\n",
981 type
, trap_type
[type
], err
, (void *)(long) vaddr
, (void *)(long) rip
);
986 exc
= EXC_ARITHMETIC
;
994 * Update the PCB with this processor's DR6 value
995 * in the event this was a debug register match.
997 pcb
= THREAD_TO_PCB(thread
);
1000 * We can get and set the status register
1001 * in 32-bit mode even on a 64-bit thread
1002 * because the high order bits are not
1005 if (thread_is_64bit(thread
)) {
1006 x86_debug_state64_t
*ids
= pcb
->ids
;
1008 } else { /* 32 bit thread */
1009 x86_debug_state32_t
*ids
= pcb
->ids
;
1010 ids
->dr6
= (uint32_t) dr6
;
1013 exc
= EXC_BREAKPOINT
;
1014 code
= EXC_I386_SGL
;
1019 if (dtrace_user_probe(saved_state
) == KERN_SUCCESS
)
1020 return; /* If it succeeds, we are done... */
1022 exc
= EXC_BREAKPOINT
;
1023 code
= EXC_I386_BPT
;
1027 exc
= EXC_ARITHMETIC
;
1028 code
= EXC_I386_INTO
;
1031 case T_OUT_OF_BOUNDS
:
1033 code
= EXC_I386_BOUND
;
1036 case T_INVALID_OPCODE
:
1037 exc
= EXC_BAD_INSTRUCTION
;
1038 code
= EXC_I386_INVOP
;
1046 fpextovrflt(); /* Propagates exception directly, doesn't return */
1049 case T_INVALID_TSS
: /* invalid TSS == iret with NT flag set */
1050 exc
= EXC_BAD_INSTRUCTION
;
1051 code
= EXC_I386_INVTSSFLT
;
1055 case T_SEGMENT_NOT_PRESENT
:
1056 exc
= EXC_BAD_INSTRUCTION
;
1057 code
= EXC_I386_SEGNPFLT
;
1062 exc
= EXC_BAD_INSTRUCTION
;
1063 code
= EXC_I386_STKFLT
;
1067 case T_GENERAL_PROTECTION
:
1069 * There's a wide range of circumstances which generate this
1070 * class of exception. From user-space, many involve bad
1071 * addresses (such as a non-canonical 64-bit address).
1072 * So we map this to EXC_BAD_ACCESS (and thereby SIGSEGV).
1073 * The trouble is cr2 doesn't contain the faulting address;
1074 * we'd need to decode the faulting instruction to really
1075 * determine this. We'll leave that to debuggers.
1076 * However, attempted execution of privileged instructions
1077 * (e.g. cli) also generate GP faults and so we map these to
1078 * to EXC_BAD_ACCESS (and thence SIGSEGV) also - rather than
1079 * EXC_BAD_INSTRUCTION which is more accurate. We just can't
1082 exc
= EXC_BAD_ACCESS
;
1083 code
= EXC_I386_GPFLT
;
1089 prot
= VM_PROT_READ
;
1091 if (err
& T_PF_WRITE
)
1092 prot
|= VM_PROT_WRITE
;
1093 if (__improbable(err
& T_PF_EXECUTE
))
1094 prot
|= VM_PROT_EXECUTE
;
1095 kret
= vm_fault(thread
->map
,
1096 vm_map_trunc_page(vaddr
,
1099 THREAD_ABORTSAFE
, NULL
, 0);
1101 if (__probable((kret
== KERN_SUCCESS
) || (kret
== KERN_ABORTED
))) {
1102 thread_exception_return();
1106 user_page_fault_continue(kret
);
1110 case T_SSE_FLOAT_ERROR
:
1111 fpSSEexterrflt(); /* Propagates exception directly, doesn't return */
1115 case T_FLOATING_POINT_ERROR
:
1116 fpexterrflt(); /* Propagates exception directly, doesn't return */
1121 if (dtrace_user_probe(saved_state
) == KERN_SUCCESS
)
1122 return; /* If it succeeds, we are done... */
1125 * If we get an INT 0x7f when we do not expect to,
1126 * treat it as an illegal instruction
1128 exc
= EXC_BAD_INSTRUCTION
;
1129 code
= EXC_I386_INVOP
;
1133 panic("Unexpected user trap, type %d", type
);
1136 /* Note: Codepaths that directly return from user_trap() have pending
1137 * ASTs processed in locore
1139 i386_exception(exc
, code
, subcode
);
1145 * Handle AST traps for i386.
1148 extern void log_thread_action (thread_t
, char *);
1151 i386_astintr(int preemption
)
1153 ast_t mask
= AST_ALL
;
1157 mask
= AST_PREEMPTION
;
1167 * Handle exceptions for i386.
1169 * If we are an AT bus machine, we must turn off the AST for a
1170 * delayed floating-point exception.
1172 * If we are providing floating-point emulation, we may have
1173 * to retrieve the real register values from the floating point
1179 mach_exception_code_t code
,
1180 mach_exception_subcode_t subcode
)
1182 mach_exception_data_type_t codes
[EXCEPTION_CODE_MAX
];
1184 DEBUG_KPRINT_SYSCALL_MACH("i386_exception: exc=%d code=0x%llx subcode=0x%llx\n",
1185 exc
, code
, subcode
);
1186 codes
[0] = code
; /* new exception interface */
1188 exception_triage(exc
, codes
, 2);
1193 /* Synchronize a thread's x86_kernel_state (if any) with the given
1194 * x86_saved_state_t obtained from the trap/IPI handler; called in
1195 * kernel_trap() prior to entering the debugger, and when receiving
1196 * an "MP_KDP" IPI. Called with null saved_state if an incoming IPI
1197 * was detected from the kernel while spinning with interrupts masked.
1201 sync_iss_to_iks(x86_saved_state_t
*saved_state
)
1203 struct x86_kernel_state
*iks
;
1205 boolean_t record_active_regs
= FALSE
;
1207 /* The PAL may have a special way to sync registers */
1208 if (saved_state
&& saved_state
->flavor
== THREAD_STATE_NONE
)
1209 pal_get_kern_regs( saved_state
);
1211 if ((kstack
= current_thread()->kernel_stack
) != 0) {
1212 x86_saved_state64_t
*regs
= saved_state64(saved_state
);
1214 iks
= STACK_IKS(kstack
);
1216 /* Did we take the trap/interrupt in kernel mode? */
1217 if (saved_state
== NULL
|| /* NULL => polling in kernel */
1218 regs
== USER_REGS64(current_thread()))
1219 record_active_regs
= TRUE
;
1221 iks
->k_rbx
= regs
->rbx
;
1222 iks
->k_rsp
= regs
->isf
.rsp
;
1223 iks
->k_rbp
= regs
->rbp
;
1224 iks
->k_r12
= regs
->r12
;
1225 iks
->k_r13
= regs
->r13
;
1226 iks
->k_r14
= regs
->r14
;
1227 iks
->k_r15
= regs
->r15
;
1228 iks
->k_rip
= regs
->isf
.rip
;
1232 if (record_active_regs
== TRUE
) {
1233 /* Show the trap handler path */
1234 __asm__
volatile("movq %%rbx, %0" : "=m" (iks
->k_rbx
));
1235 __asm__
volatile("movq %%rsp, %0" : "=m" (iks
->k_rsp
));
1236 __asm__
volatile("movq %%rbp, %0" : "=m" (iks
->k_rbp
));
1237 __asm__
volatile("movq %%r12, %0" : "=m" (iks
->k_r12
));
1238 __asm__
volatile("movq %%r13, %0" : "=m" (iks
->k_r13
));
1239 __asm__
volatile("movq %%r14, %0" : "=m" (iks
->k_r14
));
1240 __asm__
volatile("movq %%r15, %0" : "=m" (iks
->k_r15
));
1241 /* "Current" instruction pointer */
1242 __asm__
volatile("leaq 1f(%%rip), %%rax; mov %%rax, %0\n1:"
1250 * This is used by the NMI interrupt handler (from mp.c) to
1251 * uncondtionally sync the trap handler context to the IKS
1252 * irrespective of whether the NMI was fielded in kernel
1256 sync_iss_to_iks_unconditionally(__unused x86_saved_state_t
*saved_state
) {
1257 struct x86_kernel_state
*iks
;
1260 if ((kstack
= current_thread()->kernel_stack
) != 0) {
1261 iks
= STACK_IKS(kstack
);
1262 /* Display the trap handler path */
1263 __asm__
volatile("movq %%rbx, %0" : "=m" (iks
->k_rbx
));
1264 __asm__
volatile("movq %%rsp, %0" : "=m" (iks
->k_rsp
));
1265 __asm__
volatile("movq %%rbp, %0" : "=m" (iks
->k_rbp
));
1266 __asm__
volatile("movq %%r12, %0" : "=m" (iks
->k_r12
));
1267 __asm__
volatile("movq %%r13, %0" : "=m" (iks
->k_r13
));
1268 __asm__
volatile("movq %%r14, %0" : "=m" (iks
->k_r14
));
1269 __asm__
volatile("movq %%r15, %0" : "=m" (iks
->k_r15
));
1270 /* "Current" instruction pointer */
1271 __asm__
volatile("leaq 1f(%%rip), %%rax; mov %%rax, %0\n1:" : "=m" (iks
->k_rip
)::"rax");