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 <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
, uint32_t pl
, kern_return_t fault_result
);
120 static void set_recovery_ip(x86_saved_state64_t
*saved_state
, vm_offset_t ip
);
123 /* See <rdar://problem/4613924> */
124 perfCallback tempDTraceTrapHook
= NULL
; /* Pointer to DTrace fbt trap hook routine */
126 extern boolean_t
dtrace_tally_fault(user_addr_t
);
129 extern boolean_t pmap_smep_enabled
;
130 extern boolean_t pmap_smap_enabled
;
132 __attribute__((noreturn
))
134 thread_syscall_return(
137 thread_t thr_act
= current_thread();
141 pal_register_cache_state(thr_act
, DIRTY
);
143 if (thread_is_64bit(thr_act
)) {
144 x86_saved_state64_t
*regs
;
146 regs
= USER_REGS64(thr_act
);
148 code
= (int) (regs
->rax
& SYSCALL_NUMBER_MASK
);
149 is_mach
= (regs
->rax
& SYSCALL_CLASS_MASK
)
150 == (SYSCALL_CLASS_MACH
<< SYSCALL_CLASS_SHIFT
);
151 if (kdebug_enable
&& is_mach
) {
153 KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE
,
154 MACHDBG_CODE(DBG_MACH_EXCP_SC
,code
)|DBG_FUNC_END
,
160 DEBUG_KPRINT_SYSCALL_MACH(
161 "thread_syscall_return: 64-bit mach ret=%u\n",
164 DEBUG_KPRINT_SYSCALL_UNIX(
165 "thread_syscall_return: 64-bit unix ret=%u\n",
169 x86_saved_state32_t
*regs
;
171 regs
= USER_REGS32(thr_act
);
173 code
= ((int) regs
->eax
);
174 is_mach
= (code
< 0);
175 if (kdebug_enable
&& is_mach
) {
177 KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE
,
178 MACHDBG_CODE(DBG_MACH_EXCP_SC
,-code
)|DBG_FUNC_END
,
184 DEBUG_KPRINT_SYSCALL_MACH(
185 "thread_syscall_return: 32-bit mach ret=%u\n",
188 DEBUG_KPRINT_SYSCALL_UNIX(
189 "thread_syscall_return: 32-bit unix ret=%u\n",
194 #if DEBUG || DEVELOPMENT
195 kern_allocation_name_t
196 prior __assert_only
= thread_get_kernel_state(thr_act
)->allocation_name
;
197 assertf(prior
== NULL
, "thread_set_allocation_name(\"%s\") not cleared", kern_allocation_get_name(prior
));
198 #endif /* DEBUG || DEVELOPMENT */
200 throttle_lowpri_io(1);
202 thread_exception_return();
208 user_page_fault_continue(
211 thread_t thread
= current_thread();
214 if (thread_is_64bit(thread
)) {
215 x86_saved_state64_t
*uregs
;
217 uregs
= USER_REGS64(thread
);
219 vaddr
= (user_addr_t
)uregs
->cr2
;
221 x86_saved_state32_t
*uregs
;
223 uregs
= USER_REGS32(thread
);
230 pal_dbg_page_fault( thread
, vaddr
, kr
);
232 i386_exception(EXC_BAD_ACCESS
, kr
, vaddr
);
237 * Fault recovery in copyin/copyout routines.
240 uintptr_t fault_addr
;
241 uintptr_t recover_addr
;
244 extern struct recovery recover_table
[];
245 extern struct recovery recover_table_end
[];
247 const char * trap_type
[] = {TRAP_NAMES
};
248 unsigned TRAP_TYPES
= sizeof(trap_type
)/sizeof(trap_type
[0]);
250 extern void PE_incoming_interrupt(int interrupt
);
252 #if defined(__x86_64__) && DEBUG
254 kprint_state(x86_saved_state64_t
*saved_state
)
256 kprintf("current_cpu_datap() 0x%lx\n", (uintptr_t)current_cpu_datap());
257 kprintf("Current GS base MSR 0x%llx\n", rdmsr64(MSR_IA32_GS_BASE
));
258 kprintf("Kernel GS base MSR 0x%llx\n", rdmsr64(MSR_IA32_KERNEL_GS_BASE
));
259 kprintf("state at 0x%lx:\n", (uintptr_t) saved_state
);
261 kprintf(" rdi 0x%llx\n", saved_state
->rdi
);
262 kprintf(" rsi 0x%llx\n", saved_state
->rsi
);
263 kprintf(" rdx 0x%llx\n", saved_state
->rdx
);
264 kprintf(" r10 0x%llx\n", saved_state
->r10
);
265 kprintf(" r8 0x%llx\n", saved_state
->r8
);
266 kprintf(" r9 0x%llx\n", saved_state
->r9
);
268 kprintf(" cr2 0x%llx\n", saved_state
->cr2
);
269 kprintf("real cr2 0x%lx\n", get_cr2());
270 kprintf(" r15 0x%llx\n", saved_state
->r15
);
271 kprintf(" r14 0x%llx\n", saved_state
->r14
);
272 kprintf(" r13 0x%llx\n", saved_state
->r13
);
273 kprintf(" r12 0x%llx\n", saved_state
->r12
);
274 kprintf(" r11 0x%llx\n", saved_state
->r11
);
275 kprintf(" rbp 0x%llx\n", saved_state
->rbp
);
276 kprintf(" rbx 0x%llx\n", saved_state
->rbx
);
277 kprintf(" rcx 0x%llx\n", saved_state
->rcx
);
278 kprintf(" rax 0x%llx\n", saved_state
->rax
);
280 kprintf(" gs 0x%x\n", saved_state
->gs
);
281 kprintf(" fs 0x%x\n", saved_state
->fs
);
283 kprintf(" isf.trapno 0x%x\n", saved_state
->isf
.trapno
);
284 kprintf(" isf._pad 0x%x\n", saved_state
->isf
._pad
);
285 kprintf(" isf.trapfn 0x%llx\n", saved_state
->isf
.trapfn
);
286 kprintf(" isf.err 0x%llx\n", saved_state
->isf
.err
);
287 kprintf(" isf.rip 0x%llx\n", saved_state
->isf
.rip
);
288 kprintf(" isf.cs 0x%llx\n", saved_state
->isf
.cs
);
289 kprintf(" isf.rflags 0x%llx\n", saved_state
->isf
.rflags
);
290 kprintf(" isf.rsp 0x%llx\n", saved_state
->isf
.rsp
);
291 kprintf(" isf.ss 0x%llx\n", saved_state
->isf
.ss
);
297 * Non-zero indicates latency assert is enabled and capped at valued
298 * absolute time units.
301 uint64_t interrupt_latency_cap
= 0;
302 boolean_t ilat_assert
= FALSE
;
305 interrupt_latency_tracker_setup(void) {
306 uint32_t ilat_cap_us
;
307 if (PE_parse_boot_argn("interrupt_latency_cap_us", &ilat_cap_us
, sizeof(ilat_cap_us
))) {
308 interrupt_latency_cap
= ilat_cap_us
* NSEC_PER_USEC
;
309 nanoseconds_to_absolutetime(interrupt_latency_cap
, &interrupt_latency_cap
);
311 interrupt_latency_cap
= LockTimeOut
;
313 PE_parse_boot_argn("-interrupt_latency_assert_enable", &ilat_assert
, sizeof(ilat_assert
));
316 void interrupt_reset_latency_stats(void) {
318 for (i
= 0; i
< real_ncpus
; i
++) {
319 cpu_data_ptr
[i
]->cpu_max_observed_int_latency
=
320 cpu_data_ptr
[i
]->cpu_max_observed_int_latency_vector
= 0;
324 void interrupt_populate_latency_stats(char *buf
, unsigned bufsize
) {
325 uint32_t i
, tcpu
= ~0;
326 uint64_t cur_max
= 0;
328 for (i
= 0; i
< real_ncpus
; i
++) {
329 if (cur_max
< cpu_data_ptr
[i
]->cpu_max_observed_int_latency
) {
330 cur_max
= cpu_data_ptr
[i
]->cpu_max_observed_int_latency
;
335 if (tcpu
< real_ncpus
)
336 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
);
339 uint32_t interrupt_timer_coalescing_enabled
= 1;
340 uint64_t interrupt_coalesced_timers
;
344 * - local APIC interrupts (IPIs, timers, etc) are handled by the kernel,
345 * - device interrupts go to the platform expert.
348 interrupt(x86_saved_state_t
*state
)
353 boolean_t user_mode
= FALSE
;
355 int cnum
= cpu_number();
356 cpu_data_t
*cdp
= cpu_data_ptr
[cnum
];
357 int itype
= DBG_INTR_TYPE_UNKNOWN
;
359 x86_saved_state64_t
*state64
= saved_state64(state
);
360 rip
= state64
->isf
.rip
;
361 rsp
= state64
->isf
.rsp
;
362 interrupt_num
= state64
->isf
.trapno
;
363 if(state64
->isf
.cs
& 0x03)
366 if (cpu_data_ptr
[cnum
]->lcpu
.package
->num_idle
== topoParms
.nLThreadsPerPackage
)
367 cpu_data_ptr
[cnum
]->cpu_hwIntpexits
[interrupt_num
]++;
369 if (interrupt_num
== (LAPIC_DEFAULT_INTERRUPT_BASE
+ LAPIC_INTERPROCESSOR_INTERRUPT
))
370 itype
= DBG_INTR_TYPE_IPI
;
371 else if (interrupt_num
== (LAPIC_DEFAULT_INTERRUPT_BASE
+ LAPIC_TIMER_INTERRUPT
))
372 itype
= DBG_INTR_TYPE_TIMER
;
374 itype
= DBG_INTR_TYPE_OTHER
;
376 KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE
,
377 MACHDBG_CODE(DBG_MACH_EXCP_INTR
, 0) | DBG_FUNC_START
,
379 (user_mode
? rip
: VM_KERNEL_UNSLIDE(rip
)),
380 user_mode
, itype
, 0);
382 SCHED_STATS_INTERRUPT(current_processor());
385 if (telemetry_needs_record
) {
386 telemetry_mark_curthread(user_mode
);
390 ipl
= get_preemption_level();
393 * Handle local APIC interrupts
394 * else call platform expert for devices.
396 if (!lapic_interrupt(interrupt_num
, state
)) {
397 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 if (__improbable(cdp
->cpu_nested_istack
)) {
406 cdp
->cpu_nested_istack_events
++;
409 uint64_t ctime
= mach_absolute_time();
410 uint64_t int_latency
= ctime
- cdp
->cpu_int_event_time
;
411 uint64_t esdeadline
, ehdeadline
;
412 /* Attempt to process deferred timers in the context of
413 * this interrupt, unless interrupt time has already exceeded
414 * TCOAL_ILAT_THRESHOLD.
416 #define TCOAL_ILAT_THRESHOLD (30000ULL)
418 if ((int_latency
< TCOAL_ILAT_THRESHOLD
) &&
419 interrupt_timer_coalescing_enabled
) {
420 esdeadline
= cdp
->rtclock_timer
.queue
.earliest_soft_deadline
;
421 ehdeadline
= cdp
->rtclock_timer
.deadline
;
422 if ((ctime
>= esdeadline
) && (ctime
< ehdeadline
)) {
423 interrupt_coalesced_timers
++;
424 TCOAL_DEBUG(0x88880000 | DBG_FUNC_START
, ctime
, esdeadline
, ehdeadline
, interrupt_coalesced_timers
, 0);
426 TCOAL_DEBUG(0x88880000 | DBG_FUNC_END
, ctime
, esdeadline
, interrupt_coalesced_timers
, 0, 0);
428 TCOAL_DEBUG(0x77770000, ctime
, cdp
->rtclock_timer
.queue
.earliest_soft_deadline
, cdp
->rtclock_timer
.deadline
, interrupt_coalesced_timers
, 0);
432 if (__improbable(ilat_assert
&& (int_latency
> interrupt_latency_cap
) && !machine_timeout_suspended())) {
433 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
);
436 if (__improbable(int_latency
> cdp
->cpu_max_observed_int_latency
)) {
437 cdp
->cpu_max_observed_int_latency
= int_latency
;
438 cdp
->cpu_max_observed_int_latency_vector
= interrupt_num
;
443 * Having serviced the interrupt first, look at the interrupted stack depth.
446 uint64_t depth
= cdp
->cpu_kernel_stack
447 + sizeof(struct thread_kernel_state
)
448 + sizeof(struct i386_exception_link
*)
450 if (__improbable(depth
> kernel_stack_depth_max
)) {
451 kernel_stack_depth_max
= (vm_offset_t
)depth
;
452 KERNEL_DEBUG_CONSTANT(
453 MACHDBG_CODE(DBG_MACH_SCHED
, MACH_STACK_DEPTH
),
454 (long) depth
, (long) VM_KERNEL_UNSLIDE(rip
), 0, 0, 0);
458 if (cnum
== master_cpu
)
459 ml_entropy_collect();
461 KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE
,
462 MACHDBG_CODE(DBG_MACH_EXCP_INTR
, 0) | DBG_FUNC_END
,
463 interrupt_num
, 0, 0, 0, 0);
465 assert(ml_get_interrupts_enabled() == FALSE
);
471 long dr7
= 0x400; /* magic dr7 reset value; 32 bit on i386, 64 bit on x86_64 */
472 __asm__
volatile("mov %0,%%dr7" : : "r" (dr7
));
475 unsigned kdp_has_active_watchpoints
= 0;
476 #define NO_WATCHPOINTS (!kdp_has_active_watchpoints)
478 #define NO_WATCHPOINTS 1
481 * Trap from kernel mode. Only page-fault errors are recoverable,
482 * and then only in special circumstances. All other errors are
483 * fatal. Return value indicates if trap was handled.
488 x86_saved_state_t
*state
,
491 x86_saved_state64_t
*saved_state
;
495 vm_map_t map
= 0; /* protected by T_PAGE_FAULT */
496 kern_return_t result
= KERN_FAILURE
;
497 kern_return_t fault_result
= KERN_SUCCESS
;
503 #if NCOPY_WINDOWS > 0
504 int fault_in_copy_window
= -1;
507 int trap_pl
= get_preemption_level();
509 thread
= current_thread();
511 if (__improbable(is_saved_state32(state
)))
512 panic("kernel_trap(%p) with 32-bit state", state
);
513 saved_state
= saved_state64(state
);
515 /* Record cpu where state was captured */
516 saved_state
->isf
.cpu
= cpu_number();
518 vaddr
= (user_addr_t
)saved_state
->cr2
;
519 type
= saved_state
->isf
.trapno
;
520 code
= (int)(saved_state
->isf
.err
& 0xffff);
521 intr
= (saved_state
->isf
.rflags
& EFL_IF
) != 0; /* state of ints at trap */
522 kern_ip
= (vm_offset_t
)saved_state
->isf
.rip
;
524 is_user
= (vaddr
< VM_MAX_USER_PAGE_ADDRESS
);
528 * Is there a DTrace hook?
530 if (__improbable(tempDTraceTrapHook
!= NULL
)) {
531 if (tempDTraceTrapHook(type
, state
, lo_spp
, 0) == KERN_SUCCESS
) {
533 * If it succeeds, we are done...
538 #endif /* CONFIG_DTRACE */
541 * we come here with interrupts off as we don't want to recurse
542 * on preemption below. but we do want to re-enable interrupts
543 * as soon we possibly can to hold latency down
545 if (__improbable(T_PREEMPT
== type
)) {
548 KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE
,
549 (MACHDBG_CODE(DBG_MACH_EXCP_KTRAP_x86
, type
)) | DBG_FUNC_NONE
,
550 0, 0, 0, VM_KERNEL_UNSLIDE(kern_ip
), 0);
554 user_addr_t kd_vaddr
= is_user
? vaddr
: VM_KERNEL_UNSLIDE(vaddr
);
555 KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE
,
556 (MACHDBG_CODE(DBG_MACH_EXCP_KTRAP_x86
, type
)) | DBG_FUNC_NONE
,
557 (unsigned)(kd_vaddr
>> 32), (unsigned)kd_vaddr
, is_user
,
558 VM_KERNEL_UNSLIDE(kern_ip
), 0);
561 if (T_PAGE_FAULT
== type
) {
563 * assume we're faulting in the kernel map
567 if (__probable(thread
!= THREAD_NULL
&& thread
->map
!= kernel_map
)) {
568 #if NCOPY_WINDOWS > 0
569 vm_offset_t copy_window_base
;
573 kvaddr
= (vm_offset_t
)vaddr
;
575 * must determine if fault occurred in
576 * the copy window while pre-emption is
577 * disabled for this processor so that
578 * we only need to look at the window
579 * associated with this processor
581 copy_window_base
= current_cpu_datap()->cpu_copywindow_base
;
583 if (kvaddr
>= copy_window_base
&& kvaddr
< (copy_window_base
+ (NBPDE
* NCOPY_WINDOWS
)) ) {
585 window_index
= (int)((kvaddr
- copy_window_base
) / NBPDE
);
587 if (thread
->machine
.copy_window
[window_index
].user_base
!= (user_addr_t
)-1) {
589 kvaddr
-= (copy_window_base
+ (NBPDE
* window_index
));
590 vaddr
= thread
->machine
.copy_window
[window_index
].user_base
+ kvaddr
;
593 fault_in_copy_window
= window_index
;
597 if (__probable(vaddr
< VM_MAX_USER_PAGE_ADDRESS
)) {
598 /* fault occurred in userspace */
601 /* Intercept a potential Supervisor Mode Execute
602 * Protection fault. These criteria identify
603 * both NX faults and SMEP faults, but both
604 * are fatal. We avoid checking PTEs (racy).
605 * (The VM could just redrive a SMEP fault, hence
608 if (__improbable((code
== (T_PF_PROT
| T_PF_EXECUTE
)) &&
609 (pmap_smep_enabled
) && (saved_state
->isf
.rip
== vaddr
))) {
614 * Additionally check for SMAP faults...
615 * which are characterized by page-present and
616 * the AC bit unset (i.e. not from copyin/out path).
618 if (__improbable(code
& T_PF_PROT
&&
620 (saved_state
->isf
.rflags
& EFL_AC
) == 0)) {
625 * If we're not sharing cr3 with the user
626 * and we faulted in copyio,
627 * then switch cr3 here and dismiss the fault.
630 (thread
->machine
.specFlags
&CopyIOActive
) &&
631 map
->pmap
->pm_cr3
!= get_cr3_base()) {
632 pmap_assert(current_cpu_datap()->cpu_pmap_pcid_enabled
== FALSE
);
633 set_cr3_raw(map
->pmap
->pm_cr3
);
636 if (__improbable(vaddr
< PAGE_SIZE
) &&
637 ((thread
->machine
.specFlags
& CopyIOActive
) == 0)) {
645 (void) ml_set_interrupts_enabled(intr
);
657 case T_FLOATING_POINT_ERROR
:
661 case T_SSE_FLOAT_ERROR
:
665 case T_INVALID_OPCODE
:
670 if ((saved_state
->isf
.rflags
& EFL_TF
) == 0 && NO_WATCHPOINTS
)
672 /* We've somehow encountered a debug
673 * register match that does not belong
674 * to the kernel debugger.
675 * This isn't supposed to happen.
686 if (thread
!= THREAD_NULL
&& thread
->options
& TH_OPT_DTRACE
) { /* Executing under dtrace_probe? */
687 if (dtrace_tally_fault(vaddr
)) { /* Should a fault under dtrace be ignored? */
689 * DTrace has "anticipated" the possibility of this fault, and has
690 * established the suitable recovery state. Drop down now into the
691 * recovery handling code in "case T_GENERAL_PROTECTION:".
696 #endif /* CONFIG_DTRACE */
700 if (code
& T_PF_WRITE
)
701 prot
|= VM_PROT_WRITE
;
702 if (code
& T_PF_EXECUTE
)
703 prot
|= VM_PROT_EXECUTE
;
705 fault_result
= result
= vm_fault(map
,
708 FALSE
, VM_KERN_MEMORY_NONE
,
709 THREAD_UNINT
, NULL
, 0);
711 if (result
== KERN_SUCCESS
) {
712 #if NCOPY_WINDOWS > 0
713 if (fault_in_copy_window
!= -1) {
714 ml_set_interrupts_enabled(FALSE
);
715 copy_window_fault(thread
, map
,
716 fault_in_copy_window
);
717 (void) ml_set_interrupts_enabled(intr
);
719 #endif /* NCOPY_WINDOWS > 0 */
727 #endif /* CONFIG_DTRACE */
729 case T_GENERAL_PROTECTION
:
731 * If there is a failure recovery address
732 * for this fault, go there.
734 for (rp
= recover_table
; rp
< recover_table_end
; rp
++) {
735 if (kern_ip
== rp
->fault_addr
) {
736 set_recovery_ip(saved_state
, rp
->recover_addr
);
742 * Check thread recovery address also.
744 if (thread
!= THREAD_NULL
&& thread
->recover
) {
745 set_recovery_ip(saved_state
, thread
->recover
);
750 * Unanticipated page-fault errors in kernel
757 * Exception 15 is reserved but some chips may generate it
758 * spuriously. Seen at startup on AMD Athlon-64.
761 kprintf("kernel_trap() ignoring spurious trap 15\n");
765 /* Ensure that the i386_kernel_state at the base of the
766 * current thread's stack (if any) is synchronized with the
767 * context at the moment of the trap, to facilitate
768 * access through the debugger.
770 sync_iss_to_iks(state
);
772 if (kdp_i386_trap(type
, saved_state
, result
, (vm_offset_t
)vaddr
))
777 panic_trap(saved_state
, trap_pl
, fault_result
);
784 set_recovery_ip(x86_saved_state64_t
*saved_state
, vm_offset_t ip
)
786 saved_state
->isf
.rip
= ip
;
790 panic_trap(x86_saved_state64_t
*regs
, uint32_t pl
, kern_return_t fault_result
)
792 const char *trapname
= "Unknown";
793 pal_cr_t cr0
, cr2
, cr3
, cr4
;
794 boolean_t potential_smep_fault
= FALSE
, potential_kernel_NX_fault
= FALSE
;
795 boolean_t potential_smap_fault
= FALSE
;
797 pal_get_control_registers( &cr0
, &cr2
, &cr3
, &cr4
);
798 assert(ml_get_interrupts_enabled() == FALSE
);
799 current_cpu_datap()->cpu_fatal_trap_state
= regs
;
801 * Issue an I/O port read if one has been requested - this is an
802 * event logic analyzers can use as a trigger point.
804 panic_io_port_read();
806 kprintf("CPU %d panic trap number 0x%x, rip 0x%016llx\n",
807 cpu_number(), regs
->isf
.trapno
, regs
->isf
.rip
);
808 kprintf("cr0 0x%016llx cr2 0x%016llx cr3 0x%016llx cr4 0x%016llx\n",
811 if (regs
->isf
.trapno
< TRAP_TYPES
)
812 trapname
= trap_type
[regs
->isf
.trapno
];
814 if ((regs
->isf
.trapno
== T_PAGE_FAULT
) && (regs
->isf
.err
== (T_PF_PROT
| T_PF_EXECUTE
)) && (regs
->isf
.rip
== regs
->cr2
)) {
815 if (pmap_smep_enabled
&& (regs
->isf
.rip
< VM_MAX_USER_PAGE_ADDRESS
)) {
816 potential_smep_fault
= TRUE
;
817 } else if (regs
->isf
.rip
>= VM_MIN_KERNEL_AND_KEXT_ADDRESS
) {
818 potential_kernel_NX_fault
= TRUE
;
820 } else if (pmap_smap_enabled
&&
821 regs
->isf
.trapno
== T_PAGE_FAULT
&&
822 regs
->isf
.err
& T_PF_PROT
&&
823 regs
->cr2
< VM_MAX_USER_PAGE_ADDRESS
&&
824 regs
->isf
.rip
>= VM_MIN_KERNEL_AND_KEXT_ADDRESS
) {
825 potential_smap_fault
= TRUE
;
829 panic("Kernel trap at 0x%016llx, type %d=%s, registers:\n"
830 "CR0: 0x%016llx, CR2: 0x%016llx, CR3: 0x%016llx, CR4: 0x%016llx\n"
831 "RAX: 0x%016llx, RBX: 0x%016llx, RCX: 0x%016llx, RDX: 0x%016llx\n"
832 "RSP: 0x%016llx, RBP: 0x%016llx, RSI: 0x%016llx, RDI: 0x%016llx\n"
833 "R8: 0x%016llx, R9: 0x%016llx, R10: 0x%016llx, R11: 0x%016llx\n"
834 "R12: 0x%016llx, R13: 0x%016llx, R14: 0x%016llx, R15: 0x%016llx\n"
835 "RFL: 0x%016llx, RIP: 0x%016llx, CS: 0x%016llx, SS: 0x%016llx\n"
836 "Fault CR2: 0x%016llx, Error code: 0x%016llx, Fault CPU: 0x%x%s%s%s%s, PL: %d, VF: %d\n",
837 regs
->isf
.rip
, regs
->isf
.trapno
, trapname
,
839 regs
->rax
, regs
->rbx
, regs
->rcx
, regs
->rdx
,
840 regs
->isf
.rsp
, regs
->rbp
, regs
->rsi
, regs
->rdi
,
841 regs
->r8
, regs
->r9
, regs
->r10
, regs
->r11
,
842 regs
->r12
, regs
->r13
, regs
->r14
, regs
->r15
,
843 regs
->isf
.rflags
, regs
->isf
.rip
, regs
->isf
.cs
& 0xFFFF,
844 regs
->isf
.ss
& 0xFFFF,regs
->cr2
, regs
->isf
.err
, regs
->isf
.cpu
,
845 virtualized
? " VMM" : "",
846 potential_kernel_NX_fault
? " Kernel NX fault" : "",
847 potential_smep_fault
? " SMEP/User NX fault" : "",
848 potential_smap_fault
? " SMAP fault" : "",
852 * This next statement is not executed,
853 * but it's needed to stop the compiler using tail call optimization
854 * for the panic call - which confuses the subsequent backtrace.
860 extern kern_return_t
dtrace_user_probe(x86_saved_state_t
*);
865 uint32_t fsigns
, fsigcs
;
869 * Trap from user mode.
873 x86_saved_state_t
*saved_state
)
877 mach_exception_code_t code
;
878 mach_exception_subcode_t subcode
;
882 thread_t thread
= current_thread();
885 unsigned long dr6
= 0; /* 32 bit for i386, 64 bit for x86_64 */
887 assert((is_saved_state32(saved_state
) && !thread_is_64bit(thread
)) ||
888 (is_saved_state64(saved_state
) && thread_is_64bit(thread
)));
890 if (is_saved_state64(saved_state
)) {
891 x86_saved_state64_t
*regs
;
893 regs
= saved_state64(saved_state
);
895 /* Record cpu where state was captured */
896 regs
->isf
.cpu
= cpu_number();
898 type
= regs
->isf
.trapno
;
899 err
= (int)regs
->isf
.err
& 0xffff;
900 vaddr
= (user_addr_t
)regs
->cr2
;
901 rip
= (user_addr_t
)regs
->isf
.rip
;
903 x86_saved_state32_t
*regs
;
905 regs
= saved_state32(saved_state
);
907 /* Record cpu where state was captured */
908 regs
->cpu
= cpu_number();
911 err
= regs
->err
& 0xffff;
912 vaddr
= (user_addr_t
)regs
->cr2
;
913 rip
= (user_addr_t
)regs
->eip
;
916 if ((type
== T_DEBUG
) && thread
->machine
.ids
) {
917 unsigned long clear
= 0;
918 /* Stash and clear this processor's DR6 value, in the event
919 * this was a debug register match
921 __asm__
volatile ("mov %%db6, %0" : "=r" (dr6
));
922 __asm__
volatile ("mov %0, %%db6" : : "r" (clear
));
927 KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE
,
928 (MACHDBG_CODE(DBG_MACH_EXCP_UTRAP_x86
, type
)) | DBG_FUNC_NONE
,
929 (unsigned)(vaddr
>>32), (unsigned)vaddr
,
930 (unsigned)(rip
>>32), (unsigned)rip
, 0);
938 * DTrace does not consume all user traps, only INT_3's for now.
939 * Avoid needlessly calling tempDTraceTrapHook here, and let the
940 * INT_3 case handle them.
944 DEBUG_KPRINT_SYSCALL_MASK(1,
945 "user_trap: type=0x%x(%s) err=0x%x cr2=%p rip=%p\n",
946 type
, trap_type
[type
], err
, (void *)(long) vaddr
, (void *)(long) rip
);
951 exc
= EXC_ARITHMETIC
;
959 * Update the PCB with this processor's DR6 value
960 * in the event this was a debug register match.
962 pcb
= THREAD_TO_PCB(thread
);
965 * We can get and set the status register
966 * in 32-bit mode even on a 64-bit thread
967 * because the high order bits are not
970 if (thread_is_64bit(thread
)) {
971 x86_debug_state64_t
*ids
= pcb
->ids
;
973 } else { /* 32 bit thread */
974 x86_debug_state32_t
*ids
= pcb
->ids
;
975 ids
->dr6
= (uint32_t) dr6
;
978 exc
= EXC_BREAKPOINT
;
984 if (dtrace_user_probe(saved_state
) == KERN_SUCCESS
)
985 return; /* If it succeeds, we are done... */
987 exc
= EXC_BREAKPOINT
;
992 exc
= EXC_ARITHMETIC
;
993 code
= EXC_I386_INTO
;
996 case T_OUT_OF_BOUNDS
:
998 code
= EXC_I386_BOUND
;
1001 case T_INVALID_OPCODE
:
1002 #if !defined(RC_HIDE_XNU_J137)
1003 fpUDflt(rip
); /* May return from exception directly */
1005 exc
= EXC_BAD_INSTRUCTION
;
1006 code
= EXC_I386_INVOP
;
1014 fpextovrflt(); /* Propagates exception directly, doesn't return */
1017 case T_INVALID_TSS
: /* invalid TSS == iret with NT flag set */
1018 exc
= EXC_BAD_INSTRUCTION
;
1019 code
= EXC_I386_INVTSSFLT
;
1023 case T_SEGMENT_NOT_PRESENT
:
1024 exc
= EXC_BAD_INSTRUCTION
;
1025 code
= EXC_I386_SEGNPFLT
;
1030 exc
= EXC_BAD_INSTRUCTION
;
1031 code
= EXC_I386_STKFLT
;
1035 case T_GENERAL_PROTECTION
:
1037 * There's a wide range of circumstances which generate this
1038 * class of exception. From user-space, many involve bad
1039 * addresses (such as a non-canonical 64-bit address).
1040 * So we map this to EXC_BAD_ACCESS (and thereby SIGSEGV).
1041 * The trouble is cr2 doesn't contain the faulting address;
1042 * we'd need to decode the faulting instruction to really
1043 * determine this. We'll leave that to debuggers.
1044 * However, attempted execution of privileged instructions
1045 * (e.g. cli) also generate GP faults and so we map these to
1046 * to EXC_BAD_ACCESS (and thence SIGSEGV) also - rather than
1047 * EXC_BAD_INSTRUCTION which is more accurate. We just can't
1050 exc
= EXC_BAD_ACCESS
;
1051 code
= EXC_I386_GPFLT
;
1057 prot
= VM_PROT_READ
;
1059 if (err
& T_PF_WRITE
)
1060 prot
|= VM_PROT_WRITE
;
1061 if (__improbable(err
& T_PF_EXECUTE
))
1062 prot
|= VM_PROT_EXECUTE
;
1063 #if DEVELOPMENT || DEBUG
1065 fsig
= thread_fpsimd_hash(thread
);
1070 kret
= vm_fault(thread
->map
,
1072 prot
, FALSE
, VM_KERN_MEMORY_NONE
,
1073 THREAD_ABORTSAFE
, NULL
, 0);
1074 #if DEVELOPMENT || DEBUG
1076 uint32_t fsig2
= thread_fpsimd_hash(thread
);
1081 if (fsig
!= fsig2
) {
1082 panic("FP/SIMD state hash mismatch across fault thread: %p 0x%x->0x%x", thread
, fsig
, fsig2
);
1090 if (__probable((kret
== KERN_SUCCESS
) || (kret
== KERN_ABORTED
))) {
1091 thread_exception_return();
1095 user_page_fault_continue(kret
);
1099 case T_SSE_FLOAT_ERROR
:
1100 fpSSEexterrflt(); /* Propagates exception directly, doesn't return */
1104 case T_FLOATING_POINT_ERROR
:
1105 fpexterrflt(); /* Propagates exception directly, doesn't return */
1110 if (dtrace_user_probe(saved_state
) == KERN_SUCCESS
)
1111 return; /* If it succeeds, we are done... */
1114 * If we get an INT 0x7f when we do not expect to,
1115 * treat it as an illegal instruction
1117 exc
= EXC_BAD_INSTRUCTION
;
1118 code
= EXC_I386_INVOP
;
1122 panic("Unexpected user trap, type %d", type
);
1125 /* Note: Codepaths that directly return from user_trap() have pending
1126 * ASTs processed in locore
1128 i386_exception(exc
, code
, subcode
);
1133 * Handle exceptions for i386.
1135 * If we are an AT bus machine, we must turn off the AST for a
1136 * delayed floating-point exception.
1138 * If we are providing floating-point emulation, we may have
1139 * to retrieve the real register values from the floating point
1145 mach_exception_code_t code
,
1146 mach_exception_subcode_t subcode
)
1148 mach_exception_data_type_t codes
[EXCEPTION_CODE_MAX
];
1150 DEBUG_KPRINT_SYSCALL_MACH("i386_exception: exc=%d code=0x%llx subcode=0x%llx\n",
1151 exc
, code
, subcode
);
1152 codes
[0] = code
; /* new exception interface */
1154 exception_triage(exc
, codes
, 2);
1159 /* Synchronize a thread's x86_kernel_state (if any) with the given
1160 * x86_saved_state_t obtained from the trap/IPI handler; called in
1161 * kernel_trap() prior to entering the debugger, and when receiving
1162 * an "MP_KDP" IPI. Called with null saved_state if an incoming IPI
1163 * was detected from the kernel while spinning with interrupts masked.
1167 sync_iss_to_iks(x86_saved_state_t
*saved_state
)
1169 struct x86_kernel_state
*iks
= NULL
;
1171 boolean_t record_active_regs
= FALSE
;
1173 /* The PAL may have a special way to sync registers */
1174 if (saved_state
&& saved_state
->flavor
== THREAD_STATE_NONE
)
1175 pal_get_kern_regs( saved_state
);
1177 if (current_thread() != NULL
&&
1178 (kstack
= current_thread()->kernel_stack
) != 0) {
1179 x86_saved_state64_t
*regs
= saved_state64(saved_state
);
1181 iks
= STACK_IKS(kstack
);
1183 /* Did we take the trap/interrupt in kernel mode? */
1184 if (saved_state
== NULL
|| /* NULL => polling in kernel */
1185 regs
== USER_REGS64(current_thread()))
1186 record_active_regs
= TRUE
;
1188 iks
->k_rbx
= regs
->rbx
;
1189 iks
->k_rsp
= regs
->isf
.rsp
;
1190 iks
->k_rbp
= regs
->rbp
;
1191 iks
->k_r12
= regs
->r12
;
1192 iks
->k_r13
= regs
->r13
;
1193 iks
->k_r14
= regs
->r14
;
1194 iks
->k_r15
= regs
->r15
;
1195 iks
->k_rip
= regs
->isf
.rip
;
1199 if (record_active_regs
== TRUE
) {
1200 /* Show the trap handler path */
1201 __asm__
volatile("movq %%rbx, %0" : "=m" (iks
->k_rbx
));
1202 __asm__
volatile("movq %%rsp, %0" : "=m" (iks
->k_rsp
));
1203 __asm__
volatile("movq %%rbp, %0" : "=m" (iks
->k_rbp
));
1204 __asm__
volatile("movq %%r12, %0" : "=m" (iks
->k_r12
));
1205 __asm__
volatile("movq %%r13, %0" : "=m" (iks
->k_r13
));
1206 __asm__
volatile("movq %%r14, %0" : "=m" (iks
->k_r14
));
1207 __asm__
volatile("movq %%r15, %0" : "=m" (iks
->k_r15
));
1208 /* "Current" instruction pointer */
1209 __asm__
volatile("leaq 1f(%%rip), %%rax; mov %%rax, %0\n1:"
1217 * This is used by the NMI interrupt handler (from mp.c) to
1218 * uncondtionally sync the trap handler context to the IKS
1219 * irrespective of whether the NMI was fielded in kernel
1223 sync_iss_to_iks_unconditionally(__unused x86_saved_state_t
*saved_state
) {
1224 struct x86_kernel_state
*iks
;
1227 if ((kstack
= current_thread()->kernel_stack
) != 0) {
1228 iks
= STACK_IKS(kstack
);
1229 /* Display the trap handler path */
1230 __asm__
volatile("movq %%rbx, %0" : "=m" (iks
->k_rbx
));
1231 __asm__
volatile("movq %%rsp, %0" : "=m" (iks
->k_rsp
));
1232 __asm__
volatile("movq %%rbp, %0" : "=m" (iks
->k_rbp
));
1233 __asm__
volatile("movq %%r12, %0" : "=m" (iks
->k_r12
));
1234 __asm__
volatile("movq %%r13, %0" : "=m" (iks
->k_r13
));
1235 __asm__
volatile("movq %%r14, %0" : "=m" (iks
->k_r14
));
1236 __asm__
volatile("movq %%r15, %0" : "=m" (iks
->k_r15
));
1237 /* "Current" instruction pointer */
1238 __asm__
volatile("leaq 1f(%%rip), %%rax; mov %%rax, %0\n1:" : "=m" (iks
->k_rip
)::"rax");
1247 extern void thread_exception_return_internal(void) __dead2
;
1249 void thread_exception_return(void) {
1250 thread_t thread
= current_thread();
1251 ml_set_interrupts_enabled(FALSE
);
1252 if (thread_is_64bit(thread
) != task_has_64BitAddr(thread
->task
)) {
1253 panic("Task/thread bitness mismatch %p %p, task: %d, thread: %d", thread
, thread
->task
, thread_is_64bit(thread
), task_has_64BitAddr(thread
->task
));
1256 if (thread_is_64bit(thread
)) {
1257 if ((gdt_desc_p(USER64_CS
)->access
& ACC_PL_U
) == 0) {
1258 panic("64-GDT mismatch %p, descriptor: %p", thread
, gdt_desc_p(USER64_CS
));
1261 if ((gdt_desc_p(USER_CS
)->access
& ACC_PL_U
) == 0) {
1262 panic("32-GDT mismatch %p, descriptor: %p", thread
, gdt_desc_p(USER_CS
));
1266 thread_exception_return_internal();