2 * Copyright (c) 2000-2018 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 <kperf/kperf.h>
97 #include <prng/random.h>
101 #include <i386/postcode.h>
102 #include <i386/mp_desc.h>
103 #include <i386/proc_reg.h>
105 #include <i386/machine_check.h>
107 #include <mach/i386/syscall_sw.h>
109 #include <libkern/OSDebug.h>
110 #include <i386/cpu_threads.h>
111 #include <machine/pal_routines.h>
113 extern void throttle_lowpri_io(int);
114 extern void kprint_state(x86_saved_state64_t
*saved_state
);
117 * Forward declarations
119 static void user_page_fault_continue(kern_return_t kret
);
120 static void panic_trap(x86_saved_state64_t
*saved_state
, uint32_t pl
, kern_return_t fault_result
);
121 static void set_recovery_ip(x86_saved_state64_t
*saved_state
, vm_offset_t ip
);
124 /* See <rdar://problem/4613924> */
125 perfCallback tempDTraceTrapHook
= NULL
; /* Pointer to DTrace fbt trap hook routine */
127 extern boolean_t
dtrace_tally_fault(user_addr_t
);
130 extern boolean_t pmap_smep_enabled
;
131 extern boolean_t pmap_smap_enabled
;
133 __attribute__((noreturn
))
135 thread_syscall_return(
138 thread_t thr_act
= current_thread();
142 pal_register_cache_state(thr_act
, DIRTY
);
144 if (thread_is_64bit_addr(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",
195 #if DEBUG || DEVELOPMENT
196 kern_allocation_name_t
197 prior __assert_only
= thread_get_kernel_state(thr_act
)->allocation_name
;
198 assertf(prior
== NULL
, "thread_set_allocation_name(\"%s\") not cleared", kern_allocation_get_name(prior
));
199 #endif /* DEBUG || DEVELOPMENT */
201 throttle_lowpri_io(1);
203 thread_exception_return();
209 user_page_fault_continue(
212 thread_t thread
= current_thread();
215 if (thread_is_64bit_addr(thread
)) {
216 x86_saved_state64_t
*uregs
;
218 uregs
= USER_REGS64(thread
);
220 vaddr
= (user_addr_t
)uregs
->cr2
;
222 x86_saved_state32_t
*uregs
;
224 uregs
= USER_REGS32(thread
);
231 pal_dbg_page_fault( thread
, vaddr
, kr
);
233 i386_exception(EXC_BAD_ACCESS
, kr
, vaddr
);
238 * Fault recovery in copyin/copyout routines.
241 uintptr_t fault_addr
;
242 uintptr_t recover_addr
;
245 extern struct recovery recover_table
[];
246 extern struct recovery recover_table_end
[];
248 const char * trap_type
[] = {TRAP_NAMES
};
249 unsigned TRAP_TYPES
= sizeof(trap_type
)/sizeof(trap_type
[0]);
251 extern void PE_incoming_interrupt(int interrupt
);
253 #if defined(__x86_64__) && DEBUG
255 kprint_state(x86_saved_state64_t
*saved_state
)
257 kprintf("current_cpu_datap() 0x%lx\n", (uintptr_t)current_cpu_datap());
258 kprintf("Current GS base MSR 0x%llx\n", rdmsr64(MSR_IA32_GS_BASE
));
259 kprintf("Kernel GS base MSR 0x%llx\n", rdmsr64(MSR_IA32_KERNEL_GS_BASE
));
260 kprintf("state at 0x%lx:\n", (uintptr_t) saved_state
);
262 kprintf(" rdi 0x%llx\n", saved_state
->rdi
);
263 kprintf(" rsi 0x%llx\n", saved_state
->rsi
);
264 kprintf(" rdx 0x%llx\n", saved_state
->rdx
);
265 kprintf(" r10 0x%llx\n", saved_state
->r10
);
266 kprintf(" r8 0x%llx\n", saved_state
->r8
);
267 kprintf(" r9 0x%llx\n", saved_state
->r9
);
269 kprintf(" cr2 0x%llx\n", saved_state
->cr2
);
270 kprintf("real cr2 0x%lx\n", get_cr2());
271 kprintf(" r15 0x%llx\n", saved_state
->r15
);
272 kprintf(" r14 0x%llx\n", saved_state
->r14
);
273 kprintf(" r13 0x%llx\n", saved_state
->r13
);
274 kprintf(" r12 0x%llx\n", saved_state
->r12
);
275 kprintf(" r11 0x%llx\n", saved_state
->r11
);
276 kprintf(" rbp 0x%llx\n", saved_state
->rbp
);
277 kprintf(" rbx 0x%llx\n", saved_state
->rbx
);
278 kprintf(" rcx 0x%llx\n", saved_state
->rcx
);
279 kprintf(" rax 0x%llx\n", saved_state
->rax
);
281 kprintf(" gs 0x%x\n", saved_state
->gs
);
282 kprintf(" fs 0x%x\n", saved_state
->fs
);
284 kprintf(" isf.trapno 0x%x\n", saved_state
->isf
.trapno
);
285 kprintf(" isf._pad 0x%x\n", saved_state
->isf
._pad
);
286 kprintf(" isf.trapfn 0x%llx\n", saved_state
->isf
.trapfn
);
287 kprintf(" isf.err 0x%llx\n", saved_state
->isf
.err
);
288 kprintf(" isf.rip 0x%llx\n", saved_state
->isf
.rip
);
289 kprintf(" isf.cs 0x%llx\n", saved_state
->isf
.cs
);
290 kprintf(" isf.rflags 0x%llx\n", saved_state
->isf
.rflags
);
291 kprintf(" isf.rsp 0x%llx\n", saved_state
->isf
.rsp
);
292 kprintf(" isf.ss 0x%llx\n", saved_state
->isf
.ss
);
298 * Non-zero indicates latency assert is enabled and capped at valued
299 * absolute time units.
302 uint64_t interrupt_latency_cap
= 0;
303 boolean_t ilat_assert
= FALSE
;
306 interrupt_latency_tracker_setup(void) {
307 uint32_t ilat_cap_us
;
308 if (PE_parse_boot_argn("interrupt_latency_cap_us", &ilat_cap_us
, sizeof(ilat_cap_us
))) {
309 interrupt_latency_cap
= ilat_cap_us
* NSEC_PER_USEC
;
310 nanoseconds_to_absolutetime(interrupt_latency_cap
, &interrupt_latency_cap
);
312 interrupt_latency_cap
= LockTimeOut
;
314 PE_parse_boot_argn("-interrupt_latency_assert_enable", &ilat_assert
, sizeof(ilat_assert
));
317 void interrupt_reset_latency_stats(void) {
319 for (i
= 0; i
< real_ncpus
; i
++) {
320 cpu_data_ptr
[i
]->cpu_max_observed_int_latency
=
321 cpu_data_ptr
[i
]->cpu_max_observed_int_latency_vector
= 0;
325 void interrupt_populate_latency_stats(char *buf
, unsigned bufsize
) {
326 uint32_t i
, tcpu
= ~0;
327 uint64_t cur_max
= 0;
329 for (i
= 0; i
< real_ncpus
; i
++) {
330 if (cur_max
< cpu_data_ptr
[i
]->cpu_max_observed_int_latency
) {
331 cur_max
= cpu_data_ptr
[i
]->cpu_max_observed_int_latency
;
336 if (tcpu
< real_ncpus
)
337 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 uint32_t interrupt_timer_coalescing_enabled
= 1;
341 uint64_t interrupt_coalesced_timers
;
345 * - local APIC interrupts (IPIs, timers, etc) are handled by the kernel,
346 * - device interrupts go to the platform expert.
349 interrupt(x86_saved_state_t
*state
)
354 boolean_t user_mode
= FALSE
;
356 int cnum
= cpu_number();
357 cpu_data_t
*cdp
= cpu_data_ptr
[cnum
];
358 int itype
= DBG_INTR_TYPE_UNKNOWN
;
360 x86_saved_state64_t
*state64
= saved_state64(state
);
361 rip
= state64
->isf
.rip
;
362 rsp
= state64
->isf
.rsp
;
363 interrupt_num
= state64
->isf
.trapno
;
364 if(state64
->isf
.cs
& 0x03)
367 if (cpu_data_ptr
[cnum
]->lcpu
.package
->num_idle
== topoParms
.nLThreadsPerPackage
)
368 cpu_data_ptr
[cnum
]->cpu_hwIntpexits
[interrupt_num
]++;
370 if (interrupt_num
== (LAPIC_DEFAULT_INTERRUPT_BASE
+ LAPIC_INTERPROCESSOR_INTERRUPT
))
371 itype
= DBG_INTR_TYPE_IPI
;
372 else if (interrupt_num
== (LAPIC_DEFAULT_INTERRUPT_BASE
+ LAPIC_TIMER_INTERRUPT
))
373 itype
= DBG_INTR_TYPE_TIMER
;
375 itype
= DBG_INTR_TYPE_OTHER
;
377 KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE
,
378 MACHDBG_CODE(DBG_MACH_EXCP_INTR
, 0) | DBG_FUNC_START
,
380 (user_mode
? rip
: VM_KERNEL_UNSLIDE(rip
)),
381 user_mode
, itype
, 0);
383 SCHED_STATS_INTERRUPT(current_processor());
386 if (telemetry_needs_record
) {
387 telemetry_mark_curthread(user_mode
, FALSE
);
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
);
401 if (__improbable(get_preemption_level() != ipl
)) {
402 panic("Preemption level altered by interrupt vector 0x%x: initial 0x%x, final: 0x%x\n", interrupt_num
, ipl
, get_preemption_level());
406 if (__improbable(cdp
->cpu_nested_istack
)) {
407 cdp
->cpu_nested_istack_events
++;
410 uint64_t ctime
= mach_absolute_time();
411 uint64_t int_latency
= ctime
- cdp
->cpu_int_event_time
;
412 uint64_t esdeadline
, ehdeadline
;
413 /* Attempt to process deferred timers in the context of
414 * this interrupt, unless interrupt time has already exceeded
415 * TCOAL_ILAT_THRESHOLD.
417 #define TCOAL_ILAT_THRESHOLD (30000ULL)
419 if ((int_latency
< TCOAL_ILAT_THRESHOLD
) &&
420 interrupt_timer_coalescing_enabled
) {
421 esdeadline
= cdp
->rtclock_timer
.queue
.earliest_soft_deadline
;
422 ehdeadline
= cdp
->rtclock_timer
.deadline
;
423 if ((ctime
>= esdeadline
) && (ctime
< ehdeadline
)) {
424 interrupt_coalesced_timers
++;
425 TCOAL_DEBUG(0x88880000 | DBG_FUNC_START
, ctime
, esdeadline
, ehdeadline
, interrupt_coalesced_timers
, 0);
427 TCOAL_DEBUG(0x88880000 | DBG_FUNC_END
, ctime
, esdeadline
, interrupt_coalesced_timers
, 0, 0);
429 TCOAL_DEBUG(0x77770000, ctime
, cdp
->rtclock_timer
.queue
.earliest_soft_deadline
, cdp
->rtclock_timer
.deadline
, interrupt_coalesced_timers
, 0);
433 if (__improbable(ilat_assert
&& (int_latency
> interrupt_latency_cap
) && !machine_timeout_suspended())) {
434 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
);
437 if (__improbable(int_latency
> cdp
->cpu_max_observed_int_latency
)) {
438 cdp
->cpu_max_observed_int_latency
= int_latency
;
439 cdp
->cpu_max_observed_int_latency_vector
= interrupt_num
;
444 * Having serviced the interrupt first, look at the interrupted stack depth.
447 uint64_t depth
= cdp
->cpu_kernel_stack
448 + sizeof(struct thread_kernel_state
)
449 + sizeof(struct i386_exception_link
*)
451 if (__improbable(depth
> kernel_stack_depth_max
)) {
452 kernel_stack_depth_max
= (vm_offset_t
)depth
;
453 KERNEL_DEBUG_CONSTANT(
454 MACHDBG_CODE(DBG_MACH_SCHED
, MACH_STACK_DEPTH
),
455 (long) depth
, (long) VM_KERNEL_UNSLIDE(rip
), 0, 0, 0);
459 if (cnum
== master_cpu
)
460 ml_entropy_collect();
466 KDBG_RELEASE(MACHDBG_CODE(DBG_MACH_EXCP_INTR
, 0) | DBG_FUNC_END
,
469 assert(ml_get_interrupts_enabled() == FALSE
);
475 long dr7
= 0x400; /* magic dr7 reset value; 32 bit on i386, 64 bit on x86_64 */
476 __asm__
volatile("mov %0,%%dr7" : : "r" (dr7
));
479 unsigned kdp_has_active_watchpoints
= 0;
480 #define NO_WATCHPOINTS (!kdp_has_active_watchpoints)
482 #define NO_WATCHPOINTS 1
485 * Trap from kernel mode. Only page-fault errors are recoverable,
486 * and then only in special circumstances. All other errors are
487 * fatal. Return value indicates if trap was handled.
492 x86_saved_state_t
*state
,
495 x86_saved_state64_t
*saved_state
;
499 vm_map_t map
= 0; /* protected by T_PAGE_FAULT */
500 kern_return_t result
= KERN_FAILURE
;
501 kern_return_t fault_result
= KERN_SUCCESS
;
507 #if NCOPY_WINDOWS > 0
508 int fault_in_copy_window
= -1;
511 int trap_pl
= get_preemption_level();
513 thread
= current_thread();
515 if (__improbable(is_saved_state32(state
)))
516 panic("kernel_trap(%p) with 32-bit state", state
);
517 saved_state
= saved_state64(state
);
519 /* Record cpu where state was captured */
520 saved_state
->isf
.cpu
= cpu_number();
522 vaddr
= (user_addr_t
)saved_state
->cr2
;
523 type
= saved_state
->isf
.trapno
;
524 code
= (int)(saved_state
->isf
.err
& 0xffff);
525 intr
= (saved_state
->isf
.rflags
& EFL_IF
) != 0; /* state of ints at trap */
526 kern_ip
= (vm_offset_t
)saved_state
->isf
.rip
;
528 is_user
= (vaddr
< VM_MAX_USER_PAGE_ADDRESS
);
532 * Is there a DTrace hook?
534 if (__improbable(tempDTraceTrapHook
!= NULL
)) {
535 if (tempDTraceTrapHook(type
, state
, lo_spp
, 0) == KERN_SUCCESS
) {
537 * If it succeeds, we are done...
542 #endif /* CONFIG_DTRACE */
545 * we come here with interrupts off as we don't want to recurse
546 * on preemption below. but we do want to re-enable interrupts
547 * as soon we possibly can to hold latency down
549 if (__improbable(T_PREEMPT
== type
)) {
552 KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE
,
553 (MACHDBG_CODE(DBG_MACH_EXCP_KTRAP_x86
, type
)) | DBG_FUNC_NONE
,
554 0, 0, 0, VM_KERNEL_UNSLIDE(kern_ip
), 0);
558 user_addr_t kd_vaddr
= is_user
? vaddr
: VM_KERNEL_UNSLIDE(vaddr
);
559 KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE
,
560 (MACHDBG_CODE(DBG_MACH_EXCP_KTRAP_x86
, type
)) | DBG_FUNC_NONE
,
561 (unsigned)(kd_vaddr
>> 32), (unsigned)kd_vaddr
, is_user
,
562 VM_KERNEL_UNSLIDE(kern_ip
), 0);
565 if (T_PAGE_FAULT
== type
) {
567 * assume we're faulting in the kernel map
571 if (__probable(thread
!= THREAD_NULL
&& thread
->map
!= kernel_map
)) {
572 #if NCOPY_WINDOWS > 0
573 vm_offset_t copy_window_base
;
577 kvaddr
= (vm_offset_t
)vaddr
;
579 * must determine if fault occurred in
580 * the copy window while pre-emption is
581 * disabled for this processor so that
582 * we only need to look at the window
583 * associated with this processor
585 copy_window_base
= current_cpu_datap()->cpu_copywindow_base
;
587 if (kvaddr
>= copy_window_base
&& kvaddr
< (copy_window_base
+ (NBPDE
* NCOPY_WINDOWS
)) ) {
589 window_index
= (int)((kvaddr
- copy_window_base
) / NBPDE
);
591 if (thread
->machine
.copy_window
[window_index
].user_base
!= (user_addr_t
)-1) {
593 kvaddr
-= (copy_window_base
+ (NBPDE
* window_index
));
594 vaddr
= thread
->machine
.copy_window
[window_index
].user_base
+ kvaddr
;
597 fault_in_copy_window
= window_index
;
601 if (__probable(vaddr
< VM_MAX_USER_PAGE_ADDRESS
)) {
602 /* 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
)) &&
613 (pmap_smep_enabled
) && (saved_state
->isf
.rip
== vaddr
))) {
618 * Additionally check for SMAP faults...
619 * which are characterized by page-present and
620 * the AC bit unset (i.e. not from copyin/out path).
622 if (__improbable(code
& T_PF_PROT
&&
624 (saved_state
->isf
.rflags
& EFL_AC
) == 0)) {
629 * If we're not sharing cr3 with the user
630 * and we faulted in copyio,
631 * then switch cr3 here and dismiss the fault.
634 (thread
->machine
.specFlags
&CopyIOActive
) &&
635 map
->pmap
->pm_cr3
!= get_cr3_base()) {
636 pmap_assert(current_cpu_datap()->cpu_pmap_pcid_enabled
== FALSE
);
637 set_cr3_raw(map
->pmap
->pm_cr3
);
640 if (__improbable(vaddr
< PAGE_SIZE
) &&
641 ((thread
->machine
.specFlags
& CopyIOActive
) == 0)) {
649 (void) ml_set_interrupts_enabled(intr
);
661 case T_FLOATING_POINT_ERROR
:
665 case T_SSE_FLOAT_ERROR
:
669 case T_INVALID_OPCODE
:
674 if ((saved_state
->isf
.rflags
& EFL_TF
) == 0 && NO_WATCHPOINTS
)
676 /* We've somehow encountered a debug
677 * register match that does not belong
678 * to the kernel debugger.
679 * This isn't supposed to happen.
690 if (thread
!= THREAD_NULL
&& thread
->options
& TH_OPT_DTRACE
) { /* Executing under dtrace_probe? */
691 if (dtrace_tally_fault(vaddr
)) { /* Should a fault under dtrace be ignored? */
693 * DTrace has "anticipated" the possibility of this fault, and has
694 * established the suitable recovery state. Drop down now into the
695 * recovery handling code in "case T_GENERAL_PROTECTION:".
700 #endif /* CONFIG_DTRACE */
704 if (code
& T_PF_WRITE
)
705 prot
|= VM_PROT_WRITE
;
706 if (code
& T_PF_EXECUTE
)
707 prot
|= VM_PROT_EXECUTE
;
709 fault_result
= result
= vm_fault(map
,
712 FALSE
, VM_KERN_MEMORY_NONE
,
713 THREAD_UNINT
, NULL
, 0);
715 if (result
== KERN_SUCCESS
) {
716 #if NCOPY_WINDOWS > 0
717 if (fault_in_copy_window
!= -1) {
718 ml_set_interrupts_enabled(FALSE
);
719 copy_window_fault(thread
, map
,
720 fault_in_copy_window
);
721 (void) ml_set_interrupts_enabled(intr
);
723 #endif /* NCOPY_WINDOWS > 0 */
731 #endif /* CONFIG_DTRACE */
733 case T_GENERAL_PROTECTION
:
735 * If there is a failure recovery address
736 * for this fault, go there.
738 for (rp
= recover_table
; rp
< recover_table_end
; rp
++) {
739 if (kern_ip
== rp
->fault_addr
) {
740 set_recovery_ip(saved_state
, rp
->recover_addr
);
746 * Check thread recovery address also.
748 if (thread
!= THREAD_NULL
&& thread
->recover
) {
749 set_recovery_ip(saved_state
, thread
->recover
);
754 * Unanticipated page-fault errors in kernel
761 * Exception 15 is reserved but some chips may generate it
762 * spuriously. Seen at startup on AMD Athlon-64.
765 kprintf("kernel_trap() ignoring spurious trap 15\n");
769 /* Ensure that the i386_kernel_state at the base of the
770 * current thread's stack (if any) is synchronized with the
771 * context at the moment of the trap, to facilitate
772 * access through the debugger.
774 sync_iss_to_iks(state
);
776 if (kdp_i386_trap(type
, saved_state
, result
, (vm_offset_t
)vaddr
))
781 panic_trap(saved_state
, trap_pl
, fault_result
);
788 set_recovery_ip(x86_saved_state64_t
*saved_state
, vm_offset_t ip
)
790 saved_state
->isf
.rip
= ip
;
794 panic_trap(x86_saved_state64_t
*regs
, uint32_t pl
, kern_return_t fault_result
)
796 const char *trapname
= "Unknown";
797 pal_cr_t cr0
, cr2
, cr3
, cr4
;
798 boolean_t potential_smep_fault
= FALSE
, potential_kernel_NX_fault
= FALSE
;
799 boolean_t potential_smap_fault
= FALSE
;
801 pal_get_control_registers( &cr0
, &cr2
, &cr3
, &cr4
);
802 assert(ml_get_interrupts_enabled() == FALSE
);
803 current_cpu_datap()->cpu_fatal_trap_state
= regs
;
805 * Issue an I/O port read if one has been requested - this is an
806 * event logic analyzers can use as a trigger point.
808 panic_io_port_read();
810 kprintf("CPU %d panic trap number 0x%x, rip 0x%016llx\n",
811 cpu_number(), regs
->isf
.trapno
, regs
->isf
.rip
);
812 kprintf("cr0 0x%016llx cr2 0x%016llx cr3 0x%016llx cr4 0x%016llx\n",
815 if (regs
->isf
.trapno
< TRAP_TYPES
)
816 trapname
= trap_type
[regs
->isf
.trapno
];
818 if ((regs
->isf
.trapno
== T_PAGE_FAULT
) && (regs
->isf
.err
== (T_PF_PROT
| T_PF_EXECUTE
)) && (regs
->isf
.rip
== regs
->cr2
)) {
819 if (pmap_smep_enabled
&& (regs
->isf
.rip
< VM_MAX_USER_PAGE_ADDRESS
)) {
820 potential_smep_fault
= TRUE
;
821 } else if (regs
->isf
.rip
>= VM_MIN_KERNEL_AND_KEXT_ADDRESS
) {
822 potential_kernel_NX_fault
= TRUE
;
824 } else if (pmap_smap_enabled
&&
825 regs
->isf
.trapno
== T_PAGE_FAULT
&&
826 regs
->isf
.err
& T_PF_PROT
&&
827 regs
->cr2
< VM_MAX_USER_PAGE_ADDRESS
&&
828 regs
->isf
.rip
>= VM_MIN_KERNEL_AND_KEXT_ADDRESS
) {
829 potential_smap_fault
= TRUE
;
833 panic("Kernel trap at 0x%016llx, type %d=%s, registers:\n"
834 "CR0: 0x%016llx, CR2: 0x%016llx, CR3: 0x%016llx, CR4: 0x%016llx\n"
835 "RAX: 0x%016llx, RBX: 0x%016llx, RCX: 0x%016llx, RDX: 0x%016llx\n"
836 "RSP: 0x%016llx, RBP: 0x%016llx, RSI: 0x%016llx, RDI: 0x%016llx\n"
837 "R8: 0x%016llx, R9: 0x%016llx, R10: 0x%016llx, R11: 0x%016llx\n"
838 "R12: 0x%016llx, R13: 0x%016llx, R14: 0x%016llx, R15: 0x%016llx\n"
839 "RFL: 0x%016llx, RIP: 0x%016llx, CS: 0x%016llx, SS: 0x%016llx\n"
840 "Fault CR2: 0x%016llx, Error code: 0x%016llx, Fault CPU: 0x%x%s%s%s%s, PL: %d, VF: %d\n",
841 regs
->isf
.rip
, regs
->isf
.trapno
, trapname
,
843 regs
->rax
, regs
->rbx
, regs
->rcx
, regs
->rdx
,
844 regs
->isf
.rsp
, regs
->rbp
, regs
->rsi
, regs
->rdi
,
845 regs
->r8
, regs
->r9
, regs
->r10
, regs
->r11
,
846 regs
->r12
, regs
->r13
, regs
->r14
, regs
->r15
,
847 regs
->isf
.rflags
, regs
->isf
.rip
, regs
->isf
.cs
& 0xFFFF,
848 regs
->isf
.ss
& 0xFFFF,regs
->cr2
, regs
->isf
.err
, regs
->isf
.cpu
,
849 virtualized
? " VMM" : "",
850 potential_kernel_NX_fault
? " Kernel NX fault" : "",
851 potential_smep_fault
? " SMEP/User NX fault" : "",
852 potential_smap_fault
? " SMAP fault" : "",
856 * This next statement is not executed,
857 * but it's needed to stop the compiler using tail call optimization
858 * for the panic call - which confuses the subsequent backtrace.
864 extern kern_return_t
dtrace_user_probe(x86_saved_state_t
*);
869 uint32_t fsigns
, fsigcs
;
873 * Trap from user mode.
877 x86_saved_state_t
*saved_state
)
881 mach_exception_code_t code
;
882 mach_exception_subcode_t subcode
;
886 thread_t thread
= current_thread();
889 unsigned long dr6
= 0; /* 32 bit for i386, 64 bit for x86_64 */
891 assert((is_saved_state32(saved_state
) && !thread_is_64bit_addr(thread
)) ||
892 (is_saved_state64(saved_state
) && thread_is_64bit_addr(thread
)));
894 if (is_saved_state64(saved_state
)) {
895 x86_saved_state64_t
*regs
;
897 regs
= saved_state64(saved_state
);
899 /* Record cpu where state was captured */
900 regs
->isf
.cpu
= cpu_number();
902 type
= regs
->isf
.trapno
;
903 err
= (int)regs
->isf
.err
& 0xffff;
904 vaddr
= (user_addr_t
)regs
->cr2
;
905 rip
= (user_addr_t
)regs
->isf
.rip
;
907 x86_saved_state32_t
*regs
;
909 regs
= saved_state32(saved_state
);
911 /* Record cpu where state was captured */
912 regs
->cpu
= cpu_number();
915 err
= regs
->err
& 0xffff;
916 vaddr
= (user_addr_t
)regs
->cr2
;
917 rip
= (user_addr_t
)regs
->eip
;
920 if ((type
== T_DEBUG
) && thread
->machine
.ids
) {
921 unsigned long clear
= 0;
922 /* Stash and clear this processor's DR6 value, in the event
923 * this was a debug register match
925 __asm__
volatile ("mov %%db6, %0" : "=r" (dr6
));
926 __asm__
volatile ("mov %0, %%db6" : : "r" (clear
));
931 KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE
,
932 (MACHDBG_CODE(DBG_MACH_EXCP_UTRAP_x86
, type
)) | DBG_FUNC_NONE
,
933 (unsigned)(vaddr
>>32), (unsigned)vaddr
,
934 (unsigned)(rip
>>32), (unsigned)rip
, 0);
942 * DTrace does not consume all user traps, only INT_3's for now.
943 * Avoid needlessly calling tempDTraceTrapHook here, and let the
944 * INT_3 case handle them.
948 DEBUG_KPRINT_SYSCALL_MASK(1,
949 "user_trap: type=0x%x(%s) err=0x%x cr2=%p rip=%p\n",
950 type
, trap_type
[type
], err
, (void *)(long) vaddr
, (void *)(long) rip
);
955 exc
= EXC_ARITHMETIC
;
963 * Update the PCB with this processor's DR6 value
964 * in the event this was a debug register match.
966 pcb
= THREAD_TO_PCB(thread
);
969 * We can get and set the status register
970 * in 32-bit mode even on a 64-bit thread
971 * because the high order bits are not
974 if (thread_is_64bit_addr(thread
)) {
975 x86_debug_state64_t
*ids
= pcb
->ids
;
977 } else { /* 32 bit thread */
978 x86_debug_state32_t
*ids
= pcb
->ids
;
979 ids
->dr6
= (uint32_t) dr6
;
982 exc
= EXC_BREAKPOINT
;
988 if (dtrace_user_probe(saved_state
) == KERN_SUCCESS
)
989 return; /* If it succeeds, we are done... */
991 exc
= EXC_BREAKPOINT
;
996 exc
= EXC_ARITHMETIC
;
997 code
= EXC_I386_INTO
;
1000 case T_OUT_OF_BOUNDS
:
1002 code
= EXC_I386_BOUND
;
1005 case T_INVALID_OPCODE
:
1006 #if !defined(RC_HIDE_XNU_J137)
1007 fpUDflt(rip
); /* May return from exception directly */
1009 exc
= EXC_BAD_INSTRUCTION
;
1010 code
= EXC_I386_INVOP
;
1018 fpextovrflt(); /* Propagates exception directly, doesn't return */
1021 case T_INVALID_TSS
: /* invalid TSS == iret with NT flag set */
1022 exc
= EXC_BAD_INSTRUCTION
;
1023 code
= EXC_I386_INVTSSFLT
;
1027 case T_SEGMENT_NOT_PRESENT
:
1028 exc
= EXC_BAD_INSTRUCTION
;
1029 code
= EXC_I386_SEGNPFLT
;
1034 exc
= EXC_BAD_INSTRUCTION
;
1035 code
= EXC_I386_STKFLT
;
1039 case T_GENERAL_PROTECTION
:
1041 * There's a wide range of circumstances which generate this
1042 * class of exception. From user-space, many involve bad
1043 * addresses (such as a non-canonical 64-bit address).
1044 * So we map this to EXC_BAD_ACCESS (and thereby SIGSEGV).
1045 * The trouble is cr2 doesn't contain the faulting address;
1046 * we'd need to decode the faulting instruction to really
1047 * determine this. We'll leave that to debuggers.
1048 * However, attempted execution of privileged instructions
1049 * (e.g. cli) also generate GP faults and so we map these to
1050 * to EXC_BAD_ACCESS (and thence SIGSEGV) also - rather than
1051 * EXC_BAD_INSTRUCTION which is more accurate. We just can't
1054 exc
= EXC_BAD_ACCESS
;
1055 code
= EXC_I386_GPFLT
;
1061 prot
= VM_PROT_READ
;
1063 if (err
& T_PF_WRITE
)
1064 prot
|= VM_PROT_WRITE
;
1065 if (__improbable(err
& T_PF_EXECUTE
))
1066 prot
|= VM_PROT_EXECUTE
;
1067 #if DEVELOPMENT || DEBUG
1069 fsig
= thread_fpsimd_hash(thread
);
1074 kret
= vm_fault(thread
->map
,
1076 prot
, FALSE
, VM_KERN_MEMORY_NONE
,
1077 THREAD_ABORTSAFE
, NULL
, 0);
1078 #if DEVELOPMENT || DEBUG
1080 uint32_t fsig2
= thread_fpsimd_hash(thread
);
1085 if (fsig
!= fsig2
) {
1086 panic("FP/SIMD state hash mismatch across fault thread: %p 0x%x->0x%x", thread
, fsig
, fsig2
);
1094 if (__probable((kret
== KERN_SUCCESS
) || (kret
== KERN_ABORTED
))) {
1095 thread_exception_return();
1099 user_page_fault_continue(kret
);
1103 case T_SSE_FLOAT_ERROR
:
1104 fpSSEexterrflt(); /* Propagates exception directly, doesn't return */
1108 case T_FLOATING_POINT_ERROR
:
1109 fpexterrflt(); /* Propagates exception directly, doesn't return */
1114 if (dtrace_user_probe(saved_state
) == KERN_SUCCESS
)
1115 return; /* If it succeeds, we are done... */
1118 * If we get an INT 0x7f when we do not expect to,
1119 * treat it as an illegal instruction
1121 exc
= EXC_BAD_INSTRUCTION
;
1122 code
= EXC_I386_INVOP
;
1126 panic("Unexpected user trap, type %d", type
);
1129 /* Note: Codepaths that directly return from user_trap() have pending
1130 * ASTs processed in locore
1132 i386_exception(exc
, code
, subcode
);
1137 * Handle exceptions for i386.
1139 * If we are an AT bus machine, we must turn off the AST for a
1140 * delayed floating-point exception.
1142 * If we are providing floating-point emulation, we may have
1143 * to retrieve the real register values from the floating point
1149 mach_exception_code_t code
,
1150 mach_exception_subcode_t subcode
)
1152 mach_exception_data_type_t codes
[EXCEPTION_CODE_MAX
];
1154 DEBUG_KPRINT_SYSCALL_MACH("i386_exception: exc=%d code=0x%llx subcode=0x%llx\n",
1155 exc
, code
, subcode
);
1156 codes
[0] = code
; /* new exception interface */
1158 exception_triage(exc
, codes
, 2);
1163 /* Synchronize a thread's x86_kernel_state (if any) with the given
1164 * x86_saved_state_t obtained from the trap/IPI handler; called in
1165 * kernel_trap() prior to entering the debugger, and when receiving
1166 * an "MP_KDP" IPI. Called with null saved_state if an incoming IPI
1167 * was detected from the kernel while spinning with interrupts masked.
1171 sync_iss_to_iks(x86_saved_state_t
*saved_state
)
1173 struct x86_kernel_state
*iks
= NULL
;
1175 boolean_t record_active_regs
= FALSE
;
1177 /* The PAL may have a special way to sync registers */
1178 if (saved_state
&& saved_state
->flavor
== THREAD_STATE_NONE
)
1179 pal_get_kern_regs( saved_state
);
1181 if (current_thread() != NULL
&&
1182 (kstack
= current_thread()->kernel_stack
) != 0) {
1183 x86_saved_state64_t
*regs
= saved_state64(saved_state
);
1185 iks
= STACK_IKS(kstack
);
1187 /* Did we take the trap/interrupt in kernel mode? */
1188 if (saved_state
== NULL
|| /* NULL => polling in kernel */
1189 regs
== USER_REGS64(current_thread()))
1190 record_active_regs
= TRUE
;
1192 iks
->k_rbx
= regs
->rbx
;
1193 iks
->k_rsp
= regs
->isf
.rsp
;
1194 iks
->k_rbp
= regs
->rbp
;
1195 iks
->k_r12
= regs
->r12
;
1196 iks
->k_r13
= regs
->r13
;
1197 iks
->k_r14
= regs
->r14
;
1198 iks
->k_r15
= regs
->r15
;
1199 iks
->k_rip
= regs
->isf
.rip
;
1203 if (record_active_regs
== TRUE
) {
1204 /* Show the trap handler path */
1205 __asm__
volatile("movq %%rbx, %0" : "=m" (iks
->k_rbx
));
1206 __asm__
volatile("movq %%rsp, %0" : "=m" (iks
->k_rsp
));
1207 __asm__
volatile("movq %%rbp, %0" : "=m" (iks
->k_rbp
));
1208 __asm__
volatile("movq %%r12, %0" : "=m" (iks
->k_r12
));
1209 __asm__
volatile("movq %%r13, %0" : "=m" (iks
->k_r13
));
1210 __asm__
volatile("movq %%r14, %0" : "=m" (iks
->k_r14
));
1211 __asm__
volatile("movq %%r15, %0" : "=m" (iks
->k_r15
));
1212 /* "Current" instruction pointer */
1213 __asm__
volatile("leaq 1f(%%rip), %%rax; mov %%rax, %0\n1:"
1221 * This is used by the NMI interrupt handler (from mp.c) to
1222 * uncondtionally sync the trap handler context to the IKS
1223 * irrespective of whether the NMI was fielded in kernel
1227 sync_iss_to_iks_unconditionally(__unused x86_saved_state_t
*saved_state
) {
1228 struct x86_kernel_state
*iks
;
1231 if ((kstack
= current_thread()->kernel_stack
) != 0) {
1232 iks
= STACK_IKS(kstack
);
1233 /* Display 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:" : "=m" (iks
->k_rip
)::"rax");
1251 extern void thread_exception_return_internal(void) __dead2
;
1253 void thread_exception_return(void) {
1254 thread_t thread
= current_thread();
1255 ml_set_interrupts_enabled(FALSE
);
1256 if (thread_is_64bit_addr(thread
) != task_has_64Bit_addr(thread
->task
)) {
1257 panic("Task/thread bitness mismatch %p %p, task: %d, thread: %d", thread
, thread
->task
, thread_is_64bit_addr(thread
), task_has_64Bit_addr(thread
->task
));
1260 if (thread_is_64bit_addr(thread
)) {
1261 if ((gdt_desc_p(USER64_CS
)->access
& ACC_PL_U
) == 0) {
1262 panic("64-GDT mismatch %p, descriptor: %p", thread
, gdt_desc_p(USER64_CS
));
1265 if ((gdt_desc_p(USER_CS
)->access
& ACC_PL_U
) == 0) {
1266 panic("32-GDT mismatch %p, descriptor: %p", thread
, gdt_desc_p(USER_CS
));
1270 thread_exception_return_internal();