xref: /xnu-8020.101.4/osfmk/arm64/sleh.c (revision e7776783b89a353188416a9a346c6cdb4928faad)
1 /*
2  * Copyright (c) 2012-2020 Apple Inc. All rights reserved.
3  *
4  * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
5  *
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.
14  *
15  * Please obtain a copy of the License at
16  * http://www.opensource.apple.com/apsl/ and read it before using this file.
17  *
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.
25  *
26  * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
27  */
28 
29 #include <arm/caches_internal.h>
30 #include <arm/cpu_data.h>
31 #include <arm/cpu_data_internal.h>
32 #include <arm/misc_protos.h>
33 #include <arm/thread.h>
34 #include <arm/rtclock.h>
35 #include <arm/trap.h> /* for IS_ARM_GDB_TRAP() et al */
36 #include <arm64/proc_reg.h>
37 #include <arm64/machine_machdep.h>
38 #include <arm64/monotonic.h>
39 #include <arm64/instructions.h>
40 
41 #include <kern/debug.h>
42 #include <kern/socd_client.h>
43 #include <kern/thread.h>
44 #include <mach/exception.h>
45 #include <mach/arm/traps.h>
46 #include <mach/vm_types.h>
47 #include <mach/machine/thread_status.h>
48 
49 #include <machine/atomic.h>
50 #include <machine/limits.h>
51 
52 #include <pexpert/arm/protos.h>
53 
54 #include <vm/vm_page.h>
55 #include <vm/pmap.h>
56 #include <vm/vm_fault.h>
57 #include <vm/vm_kern.h>
58 
59 #include <sys/errno.h>
60 #include <sys/kdebug.h>
61 #include <kperf/kperf.h>
62 
63 #include <kern/policy_internal.h>
64 #if CONFIG_TELEMETRY
65 #include <kern/telemetry.h>
66 #endif
67 
68 #include <prng/entropy.h>
69 
70 
71 
72 
73 
74 #if CONFIG_KERNEL_TBI && KASAN_TBI
75 #include <san/kasan.h>
76 #endif /* CONFIG_KERNEL_TBI && KASAN_TBI */
77 
78 #if CONFIG_UBSAN_MINIMAL
79 #include <san/ubsan_minimal.h>
80 #endif /* CONFIG_UBSAN_MINIMAL */
81 
82 #ifndef __arm64__
83 #error Should only be compiling for arm64.
84 #endif
85 
86 #define TEST_CONTEXT32_SANITY(context) \
87 	(context->ss.ash.flavor == ARM_SAVED_STATE32 && context->ss.ash.count == ARM_SAVED_STATE32_COUNT && \
88 	 context->ns.nsh.flavor == ARM_NEON_SAVED_STATE32 && context->ns.nsh.count == ARM_NEON_SAVED_STATE32_COUNT)
89 
90 #define TEST_CONTEXT64_SANITY(context) \
91 	(context->ss.ash.flavor == ARM_SAVED_STATE64 && context->ss.ash.count == ARM_SAVED_STATE64_COUNT && \
92 	 context->ns.nsh.flavor == ARM_NEON_SAVED_STATE64 && context->ns.nsh.count == ARM_NEON_SAVED_STATE64_COUNT)
93 
94 #define ASSERT_CONTEXT_SANITY(context) \
95 	assert(TEST_CONTEXT32_SANITY(context) || TEST_CONTEXT64_SANITY(context))
96 
97 
98 #define COPYIN(src, dst, size)                           \
99 	(PSR64_IS_KERNEL(get_saved_state_cpsr(state))) ? \
100 	copyin_kern(src, dst, size) :                    \
101 	copyin(src, dst, size)
102 
103 #define COPYOUT(src, dst, size)                          \
104 	(PSR64_IS_KERNEL(get_saved_state_cpsr(state))) ? \
105 	copyout_kern(src, dst, size)                   : \
106 	copyout(src, dst, size)
107 
108 // Below is for concatenating a string param to a string literal
109 #define STR1(x) #x
110 #define STR(x) STR1(x)
111 
112 #define ARM64_KDBG_CODE_KERNEL (0 << 8)
113 #define ARM64_KDBG_CODE_USER   (1 << 8)
114 #define ARM64_KDBG_CODE_GUEST  (2 << 8)
115 
116 _Static_assert(ARM64_KDBG_CODE_GUEST <= KDBG_CODE_MAX, "arm64 KDBG trace codes out of range");
117 _Static_assert(ARM64_KDBG_CODE_GUEST <= UINT16_MAX, "arm64 KDBG trace codes out of range");
118 
119 void panic_with_thread_kernel_state(const char *msg, arm_saved_state_t *ss) __abortlike;
120 
121 void sleh_synchronous_sp1(arm_context_t *, uint32_t, vm_offset_t) __abortlike;
122 void sleh_synchronous(arm_context_t *, uint32_t, vm_offset_t);
123 
124 
125 
126 void sleh_irq(arm_saved_state_t *);
127 void sleh_fiq(arm_saved_state_t *);
128 void sleh_serror(arm_context_t *context, uint32_t esr, vm_offset_t far);
129 void sleh_invalid_stack(arm_context_t *context, uint32_t esr, vm_offset_t far) __dead2;
130 
131 static void sleh_interrupt_handler_prologue(arm_saved_state_t *, unsigned int type);
132 static void sleh_interrupt_handler_epilogue(void);
133 
134 static void handle_svc(arm_saved_state_t *);
135 static void handle_mach_absolute_time_trap(arm_saved_state_t *);
136 static void handle_mach_continuous_time_trap(arm_saved_state_t *);
137 
138 static void handle_msr_trap(arm_saved_state_t *state, uint32_t esr);
139 #ifdef __ARM_ARCH_8_6__
140 static void handle_pac_fail(arm_saved_state_t *state, uint32_t esr) __dead2;
141 #endif
142 
143 extern kern_return_t arm_fast_fault(pmap_t, vm_map_address_t, vm_prot_t, bool, bool);
144 
145 static void handle_uncategorized(arm_saved_state_t *);
146 
147 /*
148  * For UBSan trap and continue handling, we must be able to recover
149  * from handle_kernel_breakpoint().
150  */
151 #if !CONFIG_UBSAN_MINIMAL
152 __dead2
153 #endif /* CONFIG_UBSAN_MINIMAL */
154 static void handle_kernel_breakpoint(arm_saved_state_t *, uint32_t);
155 
156 static void handle_breakpoint(arm_saved_state_t *, uint32_t) __dead2;
157 
158 typedef void (*abort_inspector_t)(uint32_t, fault_status_t *, vm_prot_t *);
159 static void inspect_instruction_abort(uint32_t, fault_status_t *, vm_prot_t *);
160 static void inspect_data_abort(uint32_t, fault_status_t *, vm_prot_t *);
161 
162 static int is_vm_fault(fault_status_t);
163 static int is_translation_fault(fault_status_t);
164 static int is_alignment_fault(fault_status_t);
165 
166 typedef void (*abort_handler_t)(arm_saved_state_t *, uint32_t, vm_offset_t, fault_status_t, vm_prot_t, vm_offset_t, expected_fault_handler_t);
167 static void handle_user_abort(arm_saved_state_t *, uint32_t, vm_offset_t, fault_status_t, vm_prot_t, vm_offset_t, expected_fault_handler_t);
168 static void handle_kernel_abort(arm_saved_state_t *, uint32_t, vm_offset_t, fault_status_t, vm_prot_t, vm_offset_t, expected_fault_handler_t);
169 
170 static void handle_pc_align(arm_saved_state_t *ss) __dead2;
171 static void handle_sp_align(arm_saved_state_t *ss) __dead2;
172 static void handle_sw_step_debug(arm_saved_state_t *ss) __dead2;
173 static void handle_wf_trap(arm_saved_state_t *ss) __dead2;
174 static void handle_fp_trap(arm_saved_state_t *ss, uint32_t esr) __dead2;
175 
176 static void handle_watchpoint(vm_offset_t fault_addr) __dead2;
177 
178 static void handle_abort(arm_saved_state_t *, uint32_t, vm_offset_t, vm_offset_t, abort_inspector_t, abort_handler_t, expected_fault_handler_t);
179 
180 static void handle_user_trapped_instruction32(arm_saved_state_t *, uint32_t esr) __dead2;
181 
182 static void handle_simd_trap(arm_saved_state_t *, uint32_t esr) __dead2;
183 
184 extern void mach_kauth_cred_thread_update(void);
185 void   mach_syscall_trace_exit(unsigned int retval, unsigned int call_number);
186 
187 struct proc;
188 
189 typedef uint32_t arm64_instr_t;
190 
191 extern void
192 unix_syscall(struct arm_saved_state * regs, thread_t thread_act, struct proc * proc);
193 
194 extern void
195 mach_syscall(struct arm_saved_state*);
196 
197 #if CONFIG_DTRACE
198 extern kern_return_t dtrace_user_probe(arm_saved_state_t* regs);
199 extern boolean_t dtrace_tally_fault(user_addr_t);
200 
201 /*
202  * Traps for userland processing. Can't include bsd/sys/fasttrap_isa.h, so copy
203  * and paste the trap instructions
204  * over from that file. Need to keep these in sync!
205  */
206 #define FASTTRAP_ARM32_INSTR 0xe7ffdefc
207 #define FASTTRAP_THUMB32_INSTR 0xdefc
208 #define FASTTRAP_ARM64_INSTR 0xe7eeee7e
209 
210 #define FASTTRAP_ARM32_RET_INSTR 0xe7ffdefb
211 #define FASTTRAP_THUMB32_RET_INSTR 0xdefb
212 #define FASTTRAP_ARM64_RET_INSTR 0xe7eeee7d
213 
214 /* See <rdar://problem/4613924> */
215 perfCallback tempDTraceTrapHook = NULL; /* Pointer to DTrace fbt trap hook routine */
216 #endif
217 
218 
219 
220 extern void arm64_thread_exception_return(void) __dead2;
221 
222 #if defined(APPLETYPHOON)
223 #define CPU_NAME "Typhoon"
224 #elif defined(APPLETWISTER)
225 #define CPU_NAME "Twister"
226 #elif defined(APPLEHURRICANE)
227 #define CPU_NAME "Hurricane"
228 #elif defined(APPLELIGHTNING)
229 #define CPU_NAME "Lightning"
230 #else
231 #define CPU_NAME "Unknown"
232 #endif
233 
234 #if (CONFIG_KERNEL_INTEGRITY && defined(KERNEL_INTEGRITY_WT))
235 #define ESR_WT_SERROR(esr) (((esr) & 0xffffff00) == 0xbf575400)
236 #define ESR_WT_REASON(esr) ((esr) & 0xff)
237 
238 #define WT_REASON_NONE           0
239 #define WT_REASON_INTEGRITY_FAIL 1
240 #define WT_REASON_BAD_SYSCALL    2
241 #define WT_REASON_NOT_LOCKED     3
242 #define WT_REASON_ALREADY_LOCKED 4
243 #define WT_REASON_SW_REQ         5
244 #define WT_REASON_PT_INVALID     6
245 #define WT_REASON_PT_VIOLATION   7
246 #define WT_REASON_REG_VIOLATION  8
247 #endif
248 
249 #if defined(HAS_IPI)
250 void cpu_signal_handler(void);
251 extern unsigned int gFastIPI;
252 #endif /* defined(HAS_IPI) */
253 
254 static arm_saved_state64_t *original_faulting_state = NULL;
255 
256 TUNABLE(bool, fp_exceptions_enabled, "-fp_exceptions", false);
257 
258 extern vm_offset_t static_memory_end;
259 
260 static inline int
is_vm_fault(fault_status_t status)261 is_vm_fault(fault_status_t status)
262 {
263 	switch (status) {
264 	case FSC_TRANSLATION_FAULT_L0:
265 	case FSC_TRANSLATION_FAULT_L1:
266 	case FSC_TRANSLATION_FAULT_L2:
267 	case FSC_TRANSLATION_FAULT_L3:
268 	case FSC_ACCESS_FLAG_FAULT_L1:
269 	case FSC_ACCESS_FLAG_FAULT_L2:
270 	case FSC_ACCESS_FLAG_FAULT_L3:
271 	case FSC_PERMISSION_FAULT_L1:
272 	case FSC_PERMISSION_FAULT_L2:
273 	case FSC_PERMISSION_FAULT_L3:
274 		return TRUE;
275 	default:
276 		return FALSE;
277 	}
278 }
279 
280 static inline int
is_translation_fault(fault_status_t status)281 is_translation_fault(fault_status_t status)
282 {
283 	switch (status) {
284 	case FSC_TRANSLATION_FAULT_L0:
285 	case FSC_TRANSLATION_FAULT_L1:
286 	case FSC_TRANSLATION_FAULT_L2:
287 	case FSC_TRANSLATION_FAULT_L3:
288 		return TRUE;
289 	default:
290 		return FALSE;
291 	}
292 }
293 
294 static inline int
is_permission_fault(fault_status_t status)295 is_permission_fault(fault_status_t status)
296 {
297 	switch (status) {
298 	case FSC_PERMISSION_FAULT_L1:
299 	case FSC_PERMISSION_FAULT_L2:
300 	case FSC_PERMISSION_FAULT_L3:
301 		return TRUE;
302 	default:
303 		return FALSE;
304 	}
305 }
306 
307 static inline int
is_alignment_fault(fault_status_t status)308 is_alignment_fault(fault_status_t status)
309 {
310 	return status == FSC_ALIGNMENT_FAULT;
311 }
312 
313 static inline int
is_parity_error(fault_status_t status)314 is_parity_error(fault_status_t status)
315 {
316 	switch (status) {
317 	case FSC_SYNC_PARITY:
318 	case FSC_ASYNC_PARITY:
319 	case FSC_SYNC_PARITY_TT_L1:
320 	case FSC_SYNC_PARITY_TT_L2:
321 	case FSC_SYNC_PARITY_TT_L3:
322 		return TRUE;
323 	default:
324 		return FALSE;
325 	}
326 }
327 
328 __dead2 __unused
329 static void
arm64_implementation_specific_error(arm_saved_state_t * state,uint32_t esr,vm_offset_t far)330 arm64_implementation_specific_error(arm_saved_state_t *state, uint32_t esr, vm_offset_t far)
331 {
332 #pragma unused (state, esr, far)
333 	panic_plain("Unhandled implementation specific error\n");
334 }
335 
336 #if CONFIG_KERNEL_INTEGRITY
337 #pragma clang diagnostic push
338 #pragma clang diagnostic ignored "-Wunused-parameter"
339 static void
kernel_integrity_error_handler(uint32_t esr,vm_offset_t far)340 kernel_integrity_error_handler(uint32_t esr, vm_offset_t far)
341 {
342 #if defined(KERNEL_INTEGRITY_WT)
343 #if (DEVELOPMENT || DEBUG)
344 	if (ESR_WT_SERROR(esr)) {
345 		switch (ESR_WT_REASON(esr)) {
346 		case WT_REASON_INTEGRITY_FAIL:
347 			panic_plain("Kernel integrity, violation in frame 0x%016lx.", far);
348 		case WT_REASON_BAD_SYSCALL:
349 			panic_plain("Kernel integrity, bad syscall.");
350 		case WT_REASON_NOT_LOCKED:
351 			panic_plain("Kernel integrity, not locked.");
352 		case WT_REASON_ALREADY_LOCKED:
353 			panic_plain("Kernel integrity, already locked.");
354 		case WT_REASON_SW_REQ:
355 			panic_plain("Kernel integrity, software request.");
356 		case WT_REASON_PT_INVALID:
357 			panic_plain("Kernel integrity, encountered invalid TTE/PTE while "
358 			    "walking 0x%016lx.", far);
359 		case WT_REASON_PT_VIOLATION:
360 			panic_plain("Kernel integrity, violation in mapping 0x%016lx.",
361 			    far);
362 		case WT_REASON_REG_VIOLATION:
363 			panic_plain("Kernel integrity, violation in system register %d.",
364 			    (unsigned) far);
365 		default:
366 			panic_plain("Kernel integrity, unknown (esr=0x%08x).", esr);
367 		}
368 	}
369 #else
370 	if (ESR_WT_SERROR(esr)) {
371 		panic_plain("SError esr: 0x%08x far: 0x%016lx.", esr, far);
372 	}
373 #endif
374 #endif
375 }
376 #pragma clang diagnostic pop
377 #endif
378 
379 static void
arm64_platform_error(arm_saved_state_t * state,uint32_t esr,vm_offset_t far)380 arm64_platform_error(arm_saved_state_t *state, uint32_t esr, vm_offset_t far)
381 {
382 #if CONFIG_KERNEL_INTEGRITY
383 	kernel_integrity_error_handler(esr, far);
384 #endif
385 
386 	cpu_data_t *cdp = getCpuDatap();
387 
388 	if (PE_handle_platform_error(far)) {
389 		return;
390 	} else if (cdp->platform_error_handler != NULL) {
391 		cdp->platform_error_handler(cdp->cpu_id, far);
392 	} else {
393 		arm64_implementation_specific_error(state, esr, far);
394 	}
395 }
396 
397 void
panic_with_thread_kernel_state(const char * msg,arm_saved_state_t * ss)398 panic_with_thread_kernel_state(const char *msg, arm_saved_state_t *ss)
399 {
400 	boolean_t ss_valid;
401 
402 	ss_valid = is_saved_state64(ss);
403 	arm_saved_state64_t *state = saved_state64(ss);
404 
405 	os_atomic_cmpxchg(&original_faulting_state, NULL, state, seq_cst);
406 
407 	// rdar://80659177
408 	// Read SoCD tracepoints up to twice — once the first time we call panic and
409 	// another time if we encounter a nested panic after that.
410 	static int twice = 2;
411 	if (twice > 0) {
412 		twice--;
413 		SOCD_TRACE_XNU(KERNEL_STATE_PANIC, ADDR(state->pc),
414 		    PACK_LSB(VALUE(state->lr), VALUE(ss_valid)),
415 		    PACK_2X32(VALUE(state->esr), VALUE(state->cpsr)),
416 		    VALUE(state->far));
417 	}
418 
419 	panic_plain("%s at pc 0x%016llx, lr 0x%016llx (saved state: %p%s)\n"
420 	    "\t  x0:  0x%016llx x1:  0x%016llx  x2:  0x%016llx  x3:  0x%016llx\n"
421 	    "\t  x4:  0x%016llx x5:  0x%016llx  x6:  0x%016llx  x7:  0x%016llx\n"
422 	    "\t  x8:  0x%016llx x9:  0x%016llx  x10: 0x%016llx  x11: 0x%016llx\n"
423 	    "\t  x12: 0x%016llx x13: 0x%016llx  x14: 0x%016llx  x15: 0x%016llx\n"
424 	    "\t  x16: 0x%016llx x17: 0x%016llx  x18: 0x%016llx  x19: 0x%016llx\n"
425 	    "\t  x20: 0x%016llx x21: 0x%016llx  x22: 0x%016llx  x23: 0x%016llx\n"
426 	    "\t  x24: 0x%016llx x25: 0x%016llx  x26: 0x%016llx  x27: 0x%016llx\n"
427 	    "\t  x28: 0x%016llx fp:  0x%016llx  lr:  0x%016llx  sp:  0x%016llx\n"
428 	    "\t  pc:  0x%016llx cpsr: 0x%08x         esr: 0x%08x          far: 0x%016llx\n",
429 	    msg, state->pc, state->lr, ss, (ss_valid ? "" : " INVALID"),
430 	    state->x[0], state->x[1], state->x[2], state->x[3],
431 	    state->x[4], state->x[5], state->x[6], state->x[7],
432 	    state->x[8], state->x[9], state->x[10], state->x[11],
433 	    state->x[12], state->x[13], state->x[14], state->x[15],
434 	    state->x[16], state->x[17], state->x[18], state->x[19],
435 	    state->x[20], state->x[21], state->x[22], state->x[23],
436 	    state->x[24], state->x[25], state->x[26], state->x[27],
437 	    state->x[28], state->fp, state->lr, state->sp,
438 	    state->pc, state->cpsr, state->esr, state->far);
439 }
440 
441 void
sleh_synchronous_sp1(arm_context_t * context,uint32_t esr,vm_offset_t far __unused)442 sleh_synchronous_sp1(arm_context_t *context, uint32_t esr, vm_offset_t far __unused)
443 {
444 	esr_exception_class_t  class = ESR_EC(esr);
445 	arm_saved_state_t    * state = &context->ss;
446 
447 	switch (class) {
448 	case ESR_EC_UNCATEGORIZED:
449 	{
450 		uint32_t instr = *((uint32_t*)get_saved_state_pc(state));
451 		if (IS_ARM_GDB_TRAP(instr)) {
452 			DebuggerCall(EXC_BREAKPOINT, state);
453 		}
454 	}
455 		OS_FALLTHROUGH; // panic if we return from the debugger
456 	default:
457 		panic_with_thread_kernel_state("Synchronous exception taken while SP1 selected", state);
458 	}
459 }
460 
461 
462 __attribute__((noreturn))
463 void
thread_exception_return()464 thread_exception_return()
465 {
466 	thread_t thread = current_thread();
467 	if (thread->machine.exception_trace_code != 0) {
468 		KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE,
469 		    MACHDBG_CODE(DBG_MACH_EXCP_SYNC_ARM, thread->machine.exception_trace_code) | DBG_FUNC_END, 0, 0, 0, 0, 0);
470 		thread->machine.exception_trace_code = 0;
471 	}
472 
473 	arm64_thread_exception_return();
474 	__builtin_unreachable();
475 }
476 
477 /*
478  * check whether task vtimers are running and set thread and CPU BSD AST
479  *
480  * must be called with interrupts masked so updates of fields are atomic
481  * must be emitted inline to avoid generating an FBT probe on the exception path
482  *
483  */
484 __attribute__((__always_inline__))
485 static inline void
task_vtimer_check(thread_t thread)486 task_vtimer_check(thread_t thread)
487 {
488 	task_t task = get_threadtask_early(thread);
489 
490 	if (__improbable(task != NULL && task->vtimers)) {
491 		thread->ast |= AST_BSD;
492 		thread->machine.CpuDatap->cpu_pending_ast |= AST_BSD;
493 	}
494 }
495 
496 #if MACH_ASSERT
497 /**
498  * A version of get_preemption_level() that works in early boot.
499  *
500  * If an exception is raised in early boot before the initial thread has been
501  * set up, then calling get_preemption_level() in the SLEH will trigger an
502  * infinitely-recursing exception. This function handles this edge case.
503  */
504 static inline int
sleh_get_preemption_level(void)505 sleh_get_preemption_level(void)
506 {
507 	if (__improbable(current_thread() == NULL)) {
508 		return 0;
509 	}
510 	return get_preemption_level();
511 }
512 #endif // MACH_ASSERT
513 
514 void
sleh_synchronous(arm_context_t * context,uint32_t esr,vm_offset_t far)515 sleh_synchronous(arm_context_t *context, uint32_t esr, vm_offset_t far)
516 {
517 	esr_exception_class_t  class   = ESR_EC(esr);
518 	arm_saved_state_t    * state   = &context->ss;
519 	vm_offset_t            recover = 0;
520 	thread_t               thread  = current_thread();
521 #if MACH_ASSERT
522 	int                    preemption_level = sleh_get_preemption_level();
523 #endif
524 	expected_fault_handler_t expected_fault_handler = NULL;
525 #ifdef CONFIG_XNUPOST
526 	expected_fault_handler_t saved_expected_fault_handler = NULL;
527 	uintptr_t saved_expected_fault_addr = 0;
528 #endif /* CONFIG_XNUPOST */
529 
530 	ASSERT_CONTEXT_SANITY(context);
531 
532 	task_vtimer_check(thread);
533 
534 #if CONFIG_DTRACE
535 	/*
536 	 * Handle kernel DTrace probes as early as possible to minimize the likelihood
537 	 * that this path will itself trigger a DTrace probe, which would lead to infinite
538 	 * probe recursion.
539 	 */
540 	if (__improbable((class == ESR_EC_UNCATEGORIZED) && tempDTraceTrapHook &&
541 	    (tempDTraceTrapHook(EXC_BAD_INSTRUCTION, state, 0, 0) == KERN_SUCCESS))) {
542 		return;
543 	}
544 #endif
545 	bool is_user = PSR64_IS_USER(get_saved_state_cpsr(state));
546 
547 	/*
548 	 * Use KERNEL_DEBUG_CONSTANT_IST here to avoid producing tracepoints
549 	 * that would disclose the behavior of PT_DENY_ATTACH processes.
550 	 */
551 	if (is_user) {
552 		thread->machine.exception_trace_code = (uint16_t)(ARM64_KDBG_CODE_USER | class);
553 		KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE,
554 		    MACHDBG_CODE(DBG_MACH_EXCP_SYNC_ARM, thread->machine.exception_trace_code) | DBG_FUNC_START,
555 		    esr, far, get_saved_state_pc(state), 0, 0);
556 	} else {
557 		KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE,
558 		    MACHDBG_CODE(DBG_MACH_EXCP_SYNC_ARM, ARM64_KDBG_CODE_KERNEL | class) | DBG_FUNC_START,
559 		    esr, VM_KERNEL_ADDRHIDE(far), VM_KERNEL_UNSLIDE(get_saved_state_pc(state)), 0, 0);
560 	}
561 
562 	if (__improbable(ESR_INSTR_IS_2BYTES(esr))) {
563 		/*
564 		 * We no longer support 32-bit, which means no 2-byte
565 		 * instructions.
566 		 */
567 		if (is_user) {
568 			panic("Exception on 2-byte instruction, "
569 			    "context=%p, esr=%#x, far=%p",
570 			    context, esr, (void *)far);
571 		} else {
572 			panic_with_thread_kernel_state("Exception on 2-byte instruction", state);
573 		}
574 	}
575 
576 	/* Don't run exception handler with recover handler set in case of double fault */
577 	if (thread->recover) {
578 		recover = thread->recover;
579 		thread->recover = (vm_offset_t)NULL;
580 	}
581 
582 #ifdef CONFIG_XNUPOST
583 	if (thread->machine.expected_fault_handler != NULL) {
584 		saved_expected_fault_handler = thread->machine.expected_fault_handler;
585 		saved_expected_fault_addr = thread->machine.expected_fault_addr;
586 
587 		thread->machine.expected_fault_handler = NULL;
588 		thread->machine.expected_fault_addr = 0;
589 
590 		if (saved_expected_fault_addr == far) {
591 			expected_fault_handler = saved_expected_fault_handler;
592 		}
593 	}
594 #endif /* CONFIG_XNUPOST */
595 
596 	/* Inherit the interrupt masks from previous context */
597 	if (SPSR_INTERRUPTS_ENABLED(get_saved_state_cpsr(state))) {
598 		ml_set_interrupts_enabled(TRUE);
599 	}
600 
601 	switch (class) {
602 	case ESR_EC_SVC_64:
603 		if (!is_saved_state64(state) || !is_user) {
604 			panic("Invalid SVC_64 context");
605 		}
606 
607 		handle_svc(state);
608 		break;
609 
610 	case ESR_EC_DABORT_EL0:
611 		handle_abort(state, esr, far, recover, inspect_data_abort, handle_user_abort, expected_fault_handler);
612 		break;
613 
614 	case ESR_EC_MSR_TRAP:
615 		handle_msr_trap(state, esr);
616 		break;
617 
618 #ifdef __ARM_ARCH_8_6__
619 	case ESR_EC_PAC_FAIL:
620 		handle_pac_fail(state, esr);
621 		__builtin_unreachable();
622 
623 #endif /* __ARM_ARCH_8_6__ */
624 
625 	case ESR_EC_IABORT_EL0:
626 		handle_abort(state, esr, far, recover, inspect_instruction_abort, handle_user_abort, expected_fault_handler);
627 		break;
628 
629 	case ESR_EC_IABORT_EL1:
630 #ifdef CONFIG_XNUPOST
631 		if ((expected_fault_handler != NULL) && expected_fault_handler(state)) {
632 			break;
633 		}
634 #endif /* CONFIG_XNUPOST */
635 
636 		panic_with_thread_kernel_state("Kernel instruction fetch abort", state);
637 
638 	case ESR_EC_PC_ALIGN:
639 		handle_pc_align(state);
640 		__builtin_unreachable();
641 
642 	case ESR_EC_DABORT_EL1:
643 		handle_abort(state, esr, far, recover, inspect_data_abort, handle_kernel_abort, expected_fault_handler);
644 		break;
645 
646 	case ESR_EC_UNCATEGORIZED:
647 		assert(!ESR_ISS(esr));
648 
649 		handle_uncategorized(&context->ss);
650 		break;
651 
652 	case ESR_EC_SP_ALIGN:
653 		handle_sp_align(state);
654 		__builtin_unreachable();
655 
656 	case ESR_EC_BKPT_AARCH32:
657 		handle_breakpoint(state, esr);
658 		__builtin_unreachable();
659 
660 	case ESR_EC_BRK_AARCH64:
661 		if (PSR64_IS_KERNEL(get_saved_state_cpsr(state))) {
662 			handle_kernel_breakpoint(state, esr);
663 #if CONFIG_UBSAN_MINIMAL
664 			/* UBSan breakpoints are recoverable */
665 			break;
666 #endif /* CONFIG_UBSAN_MINIMAL */
667 		} else {
668 			handle_breakpoint(state, esr);
669 			__builtin_unreachable();
670 		}
671 
672 	case ESR_EC_BKPT_REG_MATCH_EL0:
673 		if (FSC_DEBUG_FAULT == ISS_SSDE_FSC(esr)) {
674 			handle_breakpoint(state, esr);
675 		}
676 		panic("Unsupported Class %u event code. state=%p class=%u esr=%u far=%p",
677 		    class, state, class, esr, (void *)far);
678 		__builtin_unreachable();
679 
680 	case ESR_EC_BKPT_REG_MATCH_EL1:
681 		panic_with_thread_kernel_state("Hardware Breakpoint Debug exception from kernel. Panic (by design)", state);
682 		__builtin_unreachable();
683 
684 	case ESR_EC_SW_STEP_DEBUG_EL0:
685 		if (FSC_DEBUG_FAULT == ISS_SSDE_FSC(esr)) {
686 			handle_sw_step_debug(state);
687 		}
688 		panic("Unsupported Class %u event code. state=%p class=%u esr=%u far=%p",
689 		    class, state, class, esr, (void *)far);
690 		__builtin_unreachable();
691 
692 	case ESR_EC_SW_STEP_DEBUG_EL1:
693 		panic_with_thread_kernel_state("Software Step Debug exception from kernel. Panic (by design)", state);
694 		__builtin_unreachable();
695 
696 	case ESR_EC_WATCHPT_MATCH_EL0:
697 		if (FSC_DEBUG_FAULT == ISS_SSDE_FSC(esr)) {
698 			handle_watchpoint(far);
699 		}
700 		panic("Unsupported Class %u event code. state=%p class=%u esr=%u far=%p",
701 		    class, state, class, esr, (void *)far);
702 		__builtin_unreachable();
703 
704 	case ESR_EC_WATCHPT_MATCH_EL1:
705 		/*
706 		 * If we hit a watchpoint in kernel mode, probably in a copyin/copyout which we don't want to
707 		 * abort.  Turn off watchpoints and keep going; we'll turn them back on in return_from_exception..
708 		 */
709 		if (FSC_DEBUG_FAULT == ISS_SSDE_FSC(esr)) {
710 			arm_debug_set(NULL);
711 			break; /* return to first level handler */
712 		}
713 		panic("Unsupported Class %u event code. state=%p class=%u esr=%u far=%p",
714 		    class, state, class, esr, (void *)far);
715 		__builtin_unreachable();
716 
717 	case ESR_EC_TRAP_SIMD_FP:
718 		handle_simd_trap(state, esr);
719 		__builtin_unreachable();
720 
721 	case ESR_EC_ILLEGAL_INSTR_SET:
722 		if (EXCB_ACTION_RERUN !=
723 		    ex_cb_invoke(EXCB_CLASS_ILLEGAL_INSTR_SET, far)) {
724 			// instruction is not re-executed
725 			panic("Illegal instruction set exception. state=%p class=%u esr=%u far=%p spsr=0x%x",
726 			    state, class, esr, (void *)far, get_saved_state_cpsr(state));
727 		}
728 		// must clear this fault in PSR to re-run
729 		mask_saved_state_cpsr(state, 0, PSR64_IL);
730 		break;
731 
732 	case ESR_EC_MCR_MRC_CP15_TRAP:
733 	case ESR_EC_MCRR_MRRC_CP15_TRAP:
734 	case ESR_EC_MCR_MRC_CP14_TRAP:
735 	case ESR_EC_LDC_STC_CP14_TRAP:
736 	case ESR_EC_MCRR_MRRC_CP14_TRAP:
737 		handle_user_trapped_instruction32(state, esr);
738 		__builtin_unreachable();
739 
740 	case ESR_EC_WFI_WFE:
741 		// Use of WFI or WFE instruction when they have been disabled for EL0
742 		handle_wf_trap(state);
743 		__builtin_unreachable();
744 
745 	case ESR_EC_FLOATING_POINT_64:
746 		handle_fp_trap(state, esr);
747 		__builtin_unreachable();
748 
749 	default:
750 		handle_uncategorized(state);
751 	}
752 
753 #ifdef CONFIG_XNUPOST
754 	if (saved_expected_fault_handler != NULL) {
755 		thread->machine.expected_fault_handler = saved_expected_fault_handler;
756 		thread->machine.expected_fault_addr = saved_expected_fault_addr;
757 	}
758 #endif /* CONFIG_XNUPOST */
759 
760 	if (recover) {
761 		thread->recover = recover;
762 	}
763 	if (is_user) {
764 		KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE,
765 		    MACHDBG_CODE(DBG_MACH_EXCP_SYNC_ARM, thread->machine.exception_trace_code) | DBG_FUNC_END,
766 		    esr, far, get_saved_state_pc(state), 0, 0);
767 		thread->machine.exception_trace_code = 0;
768 	} else {
769 		KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE,
770 		    MACHDBG_CODE(DBG_MACH_EXCP_SYNC_ARM, ARM64_KDBG_CODE_KERNEL | class) | DBG_FUNC_END,
771 		    esr, VM_KERNEL_ADDRHIDE(far), VM_KERNEL_UNSLIDE(get_saved_state_pc(state)), 0, 0);
772 	}
773 #if MACH_ASSERT
774 	if (preemption_level != sleh_get_preemption_level()) {
775 		panic("synchronous exception changed preemption level from %d to %d", preemption_level, sleh_get_preemption_level());
776 	}
777 #endif
778 }
779 
780 /*
781  * Uncategorized exceptions are a catch-all for general execution errors.
782  * ARM64_TODO: For now, we assume this is for undefined instruction exceptions.
783  */
784 static void
handle_uncategorized(arm_saved_state_t * state)785 handle_uncategorized(arm_saved_state_t *state)
786 {
787 	exception_type_t           exception = EXC_BAD_INSTRUCTION;
788 	mach_exception_data_type_t codes[2]  = {EXC_ARM_UNDEFINED};
789 	mach_msg_type_number_t     numcodes  = 2;
790 	uint32_t                   instr     = 0;
791 
792 	COPYIN(get_saved_state_pc(state), (char *)&instr, sizeof(instr));
793 
794 #if CONFIG_DTRACE
795 
796 	if (PSR64_IS_USER64(get_saved_state_cpsr(state))) {
797 		/*
798 		 * For a 64bit user process, we care about all 4 bytes of the
799 		 * instr.
800 		 */
801 		if (instr == FASTTRAP_ARM64_INSTR || instr == FASTTRAP_ARM64_RET_INSTR) {
802 			if (dtrace_user_probe(state) == KERN_SUCCESS) {
803 				return;
804 			}
805 		}
806 	} else if (PSR64_IS_USER32(get_saved_state_cpsr(state))) {
807 		/*
808 		 * For a 32bit user process, we check for thumb mode, in
809 		 * which case we only care about a 2 byte instruction length.
810 		 * For non-thumb mode, we care about all 4 bytes of the instructin.
811 		 */
812 		if (get_saved_state_cpsr(state) & PSR64_MODE_USER32_THUMB) {
813 			if (((uint16_t)instr == FASTTRAP_THUMB32_INSTR) ||
814 			    ((uint16_t)instr == FASTTRAP_THUMB32_RET_INSTR)) {
815 				if (dtrace_user_probe(state) == KERN_SUCCESS) {
816 					return;
817 				}
818 			}
819 		} else {
820 			if ((instr == FASTTRAP_ARM32_INSTR) ||
821 			    (instr == FASTTRAP_ARM32_RET_INSTR)) {
822 				if (dtrace_user_probe(state) == KERN_SUCCESS) {
823 					return;
824 				}
825 			}
826 		}
827 	}
828 
829 #endif /* CONFIG_DTRACE */
830 
831 	if (PSR64_IS_KERNEL(get_saved_state_cpsr(state))) {
832 		if (IS_ARM_GDB_TRAP(instr)) {
833 			boolean_t interrupt_state;
834 			exception = EXC_BREAKPOINT;
835 
836 			interrupt_state = ml_set_interrupts_enabled(FALSE);
837 
838 			/* Save off the context here (so that the debug logic
839 			 * can see the original state of this thread).
840 			 */
841 			current_thread()->machine.kpcb = state;
842 
843 			/* Hop into the debugger (typically either due to a
844 			 * fatal exception, an explicit panic, or a stackshot
845 			 * request.
846 			 */
847 			DebuggerCall(exception, state);
848 
849 			current_thread()->machine.kpcb = NULL;
850 			(void) ml_set_interrupts_enabled(interrupt_state);
851 			return;
852 		} else {
853 			panic("Undefined kernel instruction: pc=%p instr=%x", (void*)get_saved_state_pc(state), instr);
854 		}
855 	}
856 
857 	/*
858 	 * Check for GDB breakpoint via illegal opcode.
859 	 */
860 	if (IS_ARM_GDB_TRAP(instr)) {
861 		exception = EXC_BREAKPOINT;
862 		codes[0] = EXC_ARM_BREAKPOINT;
863 		codes[1] = instr;
864 	} else {
865 		codes[1] = instr;
866 	}
867 
868 	exception_triage(exception, codes, numcodes);
869 	__builtin_unreachable();
870 }
871 
872 #if __has_feature(ptrauth_calls)
873 static const uint16_t ptrauth_brk_comment_base = 0xc470;
874 
875 static inline bool
brk_comment_is_ptrauth(uint16_t comment)876 brk_comment_is_ptrauth(uint16_t comment)
877 {
878 	return comment >= ptrauth_brk_comment_base &&
879 	       comment <= ptrauth_brk_comment_base + ptrauth_key_asdb;
880 }
881 
882 static inline const char *
ptrauth_key_to_string(ptrauth_key key)883 ptrauth_key_to_string(ptrauth_key key)
884 {
885 	switch (key) {
886 	case ptrauth_key_asia:
887 		return "IA";
888 	case ptrauth_key_asib:
889 		return "IB";
890 	case ptrauth_key_asda:
891 		return "DA";
892 	case ptrauth_key_asdb:
893 		return "DB";
894 	default:
895 		__builtin_unreachable();
896 	}
897 }
898 #endif /* __has_feature(ptrauth_calls) */
899 
900 #if CONFIG_KERNEL_TBI && KASAN_TBI
901 static inline bool
brk_comment_is_kasan_failure(uint16_t comment)902 brk_comment_is_kasan_failure(uint16_t comment)
903 {
904 	return comment >= KASAN_TBI_ESR_BASE &&
905 	       comment <= KASAN_TBI_ESR_TOP;
906 }
907 #endif /* CONFIG_KERNEL_TBI && KASAN_TBI */
908 
909 #if CONFIG_UBSAN_MINIMAL
910 static inline bool
brk_comment_is_ubsan(uint16_t comment)911 brk_comment_is_ubsan(uint16_t comment)
912 {
913 	return comment >= UBSAN_MINIMAL_TRAPS_START &&
914 	       comment < UBSAN_MINIMAL_TRAPS_END;
915 }
916 #endif /* CONFIG_UBSAN_MINIMAL */
917 
918 static void
handle_kernel_breakpoint(arm_saved_state_t * state,uint32_t esr)919 handle_kernel_breakpoint(arm_saved_state_t *state, uint32_t esr)
920 {
921 	uint16_t comment = ISS_BRK_COMMENT(esr);
922 
923 #if __has_feature(ptrauth_calls)
924 	if (brk_comment_is_ptrauth(comment)) {
925 #define MSG_FMT "Break 0x%04X instruction exception from kernel. Ptrauth failure with %s key resulted in 0x%016llx"
926 		char msg[strlen(MSG_FMT)
927 		- strlen("0x%04X") + strlen("0xFFFF")
928 		- strlen("%s") + strlen("IA")
929 		- strlen("0x%016llx") + strlen("0xFFFFFFFFFFFFFFFF")
930 		+ 1];
931 		ptrauth_key key = (ptrauth_key)(comment - ptrauth_brk_comment_base);
932 		const char *key_str = ptrauth_key_to_string(key);
933 		snprintf(msg, sizeof(msg), MSG_FMT, comment, key_str, saved_state64(state)->x[16]);
934 
935 		panic_with_thread_kernel_state(msg, state);
936 		__builtin_unreachable();
937 #undef MSG_FMT
938 	}
939 #endif /* __has_feature(ptrauth_calls) */
940 
941 #if CONFIG_KERNEL_TBI && KASAN_TBI
942 	if (brk_comment_is_kasan_failure(comment)) {
943 		kasan_handle_brk_failure(saved_state64(state)->x[0], comment);
944 		__builtin_unreachable();
945 	}
946 #endif /* CONFIG_KERNEL_TBI && KASAN_TBI */
947 
948 #if CONFIG_UBSAN_MINIMAL
949 	if (brk_comment_is_ubsan(comment)) {
950 		ubsan_handle_brk_trap(comment, get_saved_state_pc(state),
951 		    get_saved_state_fp(state));
952 		add_saved_state_pc(state, 4);
953 		return;
954 	}
955 #endif /* CONFIG_UBSAN_MINIMAL */
956 
957 #define MSG_FMT "Break 0x%04X instruction exception from kernel. Panic (by design)"
958 	char msg[strlen(MSG_FMT) - strlen("0x%04X") + strlen("0xFFFF") + 1];
959 	snprintf(msg, sizeof(msg), MSG_FMT, comment);
960 #undef MSG_FMT
961 
962 	panic_with_thread_kernel_state(msg, state);
963 	__builtin_unreachable();
964 }
965 
966 static void
handle_breakpoint(arm_saved_state_t * state,uint32_t esr __unused)967 handle_breakpoint(arm_saved_state_t *state, uint32_t esr __unused)
968 {
969 	exception_type_t           exception = EXC_BREAKPOINT;
970 	mach_exception_data_type_t codes[2]  = {EXC_ARM_BREAKPOINT};
971 	mach_msg_type_number_t     numcodes  = 2;
972 
973 #if __has_feature(ptrauth_calls) && !__ARM_ARCH_8_6__
974 	if (ESR_EC(esr) == ESR_EC_BRK_AARCH64 &&
975 	    brk_comment_is_ptrauth(ISS_BRK_COMMENT(esr))) {
976 		exception |= EXC_PTRAUTH_BIT;
977 	}
978 #endif /* __has_feature(ptrauth_calls) && !__ARM_ARCH_8_6__ */
979 
980 	codes[1] = get_saved_state_pc(state);
981 	exception_triage(exception, codes, numcodes);
982 	__builtin_unreachable();
983 }
984 
985 static void
handle_watchpoint(vm_offset_t fault_addr)986 handle_watchpoint(vm_offset_t fault_addr)
987 {
988 	exception_type_t           exception = EXC_BREAKPOINT;
989 	mach_exception_data_type_t codes[2]  = {EXC_ARM_DA_DEBUG};
990 	mach_msg_type_number_t     numcodes  = 2;
991 
992 	codes[1] = fault_addr;
993 	exception_triage(exception, codes, numcodes);
994 	__builtin_unreachable();
995 }
996 
997 static void
handle_abort(arm_saved_state_t * state,uint32_t esr,vm_offset_t fault_addr,vm_offset_t recover,abort_inspector_t inspect_abort,abort_handler_t handler,expected_fault_handler_t expected_fault_handler)998 handle_abort(arm_saved_state_t *state, uint32_t esr, vm_offset_t fault_addr, vm_offset_t recover,
999     abort_inspector_t inspect_abort, abort_handler_t handler, expected_fault_handler_t expected_fault_handler)
1000 {
1001 	fault_status_t fault_code;
1002 	vm_prot_t      fault_type;
1003 
1004 	inspect_abort(ESR_ISS(esr), &fault_code, &fault_type);
1005 	handler(state, esr, fault_addr, fault_code, fault_type, recover, expected_fault_handler);
1006 }
1007 
1008 static void
inspect_instruction_abort(uint32_t iss,fault_status_t * fault_code,vm_prot_t * fault_type)1009 inspect_instruction_abort(uint32_t iss, fault_status_t *fault_code, vm_prot_t *fault_type)
1010 {
1011 	getCpuDatap()->cpu_stat.instr_ex_cnt++;
1012 	*fault_code = ISS_IA_FSC(iss);
1013 	*fault_type = (VM_PROT_READ | VM_PROT_EXECUTE);
1014 }
1015 
1016 static void
inspect_data_abort(uint32_t iss,fault_status_t * fault_code,vm_prot_t * fault_type)1017 inspect_data_abort(uint32_t iss, fault_status_t *fault_code, vm_prot_t *fault_type)
1018 {
1019 	getCpuDatap()->cpu_stat.data_ex_cnt++;
1020 	*fault_code = ISS_DA_FSC(iss);
1021 
1022 	/*
1023 	 * Cache maintenance operations always report faults as write access.
1024 	 * Change these to read access, unless they report a permission fault.
1025 	 * Only certain cache maintenance operations (e.g. 'dc ivac') require write
1026 	 * access to the mapping, but if a cache maintenance operation that only requires
1027 	 * read access generates a permission fault, then we will not be able to handle
1028 	 * the fault regardless of whether we treat it as a read or write fault.
1029 	 */
1030 	if ((iss & ISS_DA_WNR) && (!(iss & ISS_DA_CM) || is_permission_fault(*fault_code))) {
1031 		*fault_type = (VM_PROT_READ | VM_PROT_WRITE);
1032 	} else {
1033 		*fault_type = (VM_PROT_READ);
1034 	}
1035 }
1036 
1037 #if __has_feature(ptrauth_calls)
1038 #ifdef __ARM_ARCH_8_6__
1039 static inline uint64_t
fault_addr_bitmask(unsigned int bit_from,unsigned int bit_to)1040 fault_addr_bitmask(unsigned int bit_from, unsigned int bit_to)
1041 {
1042 	return ((1ULL << (bit_to - bit_from + 1)) - 1) << bit_from;
1043 }
1044 #else
1045 static inline bool
fault_addr_bit(vm_offset_t fault_addr,unsigned int bit)1046 fault_addr_bit(vm_offset_t fault_addr, unsigned int bit)
1047 {
1048 	return (bool)((fault_addr >> bit) & 1);
1049 }
1050 #endif /* __ARM_ARCH_8_6__ */
1051 
1052 /**
1053  * Determines whether a fault address taken at EL0 contains a PAC error code
1054  * corresponding to the specified kind of ptrauth key.
1055  */
1056 static bool
user_fault_addr_matches_pac_error_code(vm_offset_t fault_addr,bool data_key)1057 user_fault_addr_matches_pac_error_code(vm_offset_t fault_addr, bool data_key)
1058 {
1059 	bool instruction_tbi = !(get_tcr() & TCR_TBID0_TBI_DATA_ONLY);
1060 	bool tbi = data_key || __improbable(instruction_tbi);
1061 #ifdef __ARM_ARCH_8_6__
1062 	/*
1063 	 * EnhancedPAC2 CPUs don't encode error codes at fixed positions, so
1064 	 * treat all non-canonical address bits like potential poison bits.
1065 	 */
1066 	uint64_t mask = fault_addr_bitmask(T0SZ_BOOT, 54);
1067 	if (!tbi) {
1068 		mask |= fault_addr_bitmask(56, 63);
1069 	}
1070 	return (fault_addr & mask) != 0;
1071 #else /* !__ARM_ARCH_8_6__ */
1072 	unsigned int poison_shift;
1073 	if (tbi) {
1074 		poison_shift = 53;
1075 	} else {
1076 		poison_shift = 61;
1077 	}
1078 
1079 	/* PAC error codes are always in the form key_number:NOT(key_number) */
1080 	bool poison_bit_1 = fault_addr_bit(fault_addr, poison_shift);
1081 	bool poison_bit_2 = fault_addr_bit(fault_addr, poison_shift + 1);
1082 	return poison_bit_1 != poison_bit_2;
1083 #endif /* __ARM_ARCH_8_6__ */
1084 }
1085 #endif /* __has_feature(ptrauth_calls) */
1086 
1087 static void
handle_pc_align(arm_saved_state_t * ss)1088 handle_pc_align(arm_saved_state_t *ss)
1089 {
1090 	exception_type_t exc;
1091 	mach_exception_data_type_t codes[2];
1092 	mach_msg_type_number_t numcodes = 2;
1093 
1094 	if (!PSR64_IS_USER(get_saved_state_cpsr(ss))) {
1095 		panic_with_thread_kernel_state("PC alignment exception from kernel.", ss);
1096 	}
1097 
1098 	exc = EXC_BAD_ACCESS;
1099 #if __has_feature(ptrauth_calls)
1100 	if (user_fault_addr_matches_pac_error_code(get_saved_state_pc(ss), false)) {
1101 		exc |= EXC_PTRAUTH_BIT;
1102 	}
1103 #endif /* __has_feature(ptrauth_calls) */
1104 
1105 	codes[0] = EXC_ARM_DA_ALIGN;
1106 	codes[1] = get_saved_state_pc(ss);
1107 
1108 	exception_triage(exc, codes, numcodes);
1109 	__builtin_unreachable();
1110 }
1111 
1112 static void
handle_sp_align(arm_saved_state_t * ss)1113 handle_sp_align(arm_saved_state_t *ss)
1114 {
1115 	exception_type_t exc;
1116 	mach_exception_data_type_t codes[2];
1117 	mach_msg_type_number_t numcodes = 2;
1118 
1119 	if (!PSR64_IS_USER(get_saved_state_cpsr(ss))) {
1120 		panic_with_thread_kernel_state("SP alignment exception from kernel.", ss);
1121 	}
1122 
1123 	exc = EXC_BAD_ACCESS;
1124 #if __has_feature(ptrauth_calls)
1125 	if (user_fault_addr_matches_pac_error_code(get_saved_state_sp(ss), true)) {
1126 		exc |= EXC_PTRAUTH_BIT;
1127 	}
1128 #endif /* __has_feature(ptrauth_calls) */
1129 
1130 	codes[0] = EXC_ARM_SP_ALIGN;
1131 	codes[1] = get_saved_state_sp(ss);
1132 
1133 	exception_triage(exc, codes, numcodes);
1134 	__builtin_unreachable();
1135 }
1136 
1137 static void
handle_wf_trap(arm_saved_state_t * state)1138 handle_wf_trap(arm_saved_state_t *state)
1139 {
1140 	exception_type_t exc;
1141 	mach_exception_data_type_t codes[2];
1142 	mach_msg_type_number_t numcodes = 2;
1143 	uint32_t instr = 0;
1144 
1145 	COPYIN(get_saved_state_pc(state), (char *)&instr, sizeof(instr));
1146 
1147 	exc = EXC_BAD_INSTRUCTION;
1148 	codes[0] = EXC_ARM_UNDEFINED;
1149 	codes[1] = instr;
1150 
1151 	exception_triage(exc, codes, numcodes);
1152 	__builtin_unreachable();
1153 }
1154 
1155 static void
handle_fp_trap(arm_saved_state_t * state,uint32_t esr)1156 handle_fp_trap(arm_saved_state_t *state, uint32_t esr)
1157 {
1158 	exception_type_t exc = EXC_ARITHMETIC;
1159 	mach_exception_data_type_t codes[2];
1160 	mach_msg_type_number_t numcodes = 2;
1161 	uint32_t instr = 0;
1162 
1163 	if (PSR64_IS_KERNEL(get_saved_state_cpsr(state))) {
1164 		panic_with_thread_kernel_state("Floating point exception from kernel", state);
1165 	}
1166 
1167 	COPYIN(get_saved_state_pc(state), (char *)&instr, sizeof(instr));
1168 	codes[1] = instr;
1169 
1170 	/* The floating point trap flags are only valid if TFV is set. */
1171 	if (!fp_exceptions_enabled) {
1172 		exc = EXC_BAD_INSTRUCTION;
1173 		codes[0] = EXC_ARM_UNDEFINED;
1174 	} else if (!(esr & ISS_FP_TFV)) {
1175 		codes[0] = EXC_ARM_FP_UNDEFINED;
1176 	} else if (esr & ISS_FP_UFF) {
1177 		codes[0] = EXC_ARM_FP_UF;
1178 	} else if (esr & ISS_FP_OFF) {
1179 		codes[0] = EXC_ARM_FP_OF;
1180 	} else if (esr & ISS_FP_IOF) {
1181 		codes[0] = EXC_ARM_FP_IO;
1182 	} else if (esr & ISS_FP_DZF) {
1183 		codes[0] = EXC_ARM_FP_DZ;
1184 	} else if (esr & ISS_FP_IDF) {
1185 		codes[0] = EXC_ARM_FP_ID;
1186 	} else if (esr & ISS_FP_IXF) {
1187 		codes[0] = EXC_ARM_FP_IX;
1188 	} else {
1189 		panic("Unrecognized floating point exception, state=%p, esr=%#x", state, esr);
1190 	}
1191 
1192 	exception_triage(exc, codes, numcodes);
1193 	__builtin_unreachable();
1194 }
1195 
1196 
1197 
1198 /*
1199  * handle_alignment_fault_from_user:
1200  *   state: Saved state
1201  *
1202  * Attempts to deal with an alignment fault from userspace (possibly by
1203  * emulating the faulting instruction).  If emulation failed due to an
1204  * unservicable fault, the ESR for that fault will be stored in the
1205  * recovery_esr field of the thread by the exception code.
1206  *
1207  * Returns:
1208  *   -1:     Emulation failed (emulation of state/instr not supported)
1209  *   0:      Successfully emulated the instruction
1210  *   EFAULT: Emulation failed (probably due to permissions)
1211  *   EINVAL: Emulation failed (probably due to a bad address)
1212  */
1213 
1214 
1215 static int
handle_alignment_fault_from_user(arm_saved_state_t * state,kern_return_t * vmfr)1216 handle_alignment_fault_from_user(arm_saved_state_t *state, kern_return_t *vmfr)
1217 {
1218 	int ret = -1;
1219 
1220 #pragma unused (state)
1221 #pragma unused (vmfr)
1222 
1223 	return ret;
1224 }
1225 
1226 
1227 static void
handle_sw_step_debug(arm_saved_state_t * state)1228 handle_sw_step_debug(arm_saved_state_t *state)
1229 {
1230 	thread_t thread = current_thread();
1231 	exception_type_t exc;
1232 	mach_exception_data_type_t codes[2];
1233 	mach_msg_type_number_t numcodes = 2;
1234 
1235 	if (!PSR64_IS_USER(get_saved_state_cpsr(state))) {
1236 		panic_with_thread_kernel_state("SW_STEP_DEBUG exception from kernel.", state);
1237 	}
1238 
1239 	// Disable single step and unmask interrupts (in the saved state, anticipating next exception return)
1240 	if (thread->machine.DebugData != NULL) {
1241 		thread->machine.DebugData->uds.ds64.mdscr_el1 &= ~0x1;
1242 	} else {
1243 		panic_with_thread_kernel_state("SW_STEP_DEBUG exception thread DebugData is NULL.", state);
1244 	}
1245 
1246 	mask_saved_state_cpsr(thread->machine.upcb, 0, PSR64_SS | DAIF_ALL);
1247 
1248 	// Special encoding for gdb single step event on ARM
1249 	exc = EXC_BREAKPOINT;
1250 	codes[0] = 1;
1251 	codes[1] = 0;
1252 
1253 	exception_triage(exc, codes, numcodes);
1254 	__builtin_unreachable();
1255 }
1256 
1257 static void
set_saved_state_pc_to_recovery_handler(arm_saved_state_t * iss,vm_offset_t recover)1258 set_saved_state_pc_to_recovery_handler(arm_saved_state_t *iss, vm_offset_t recover)
1259 {
1260 #if defined(HAS_APPLE_PAC)
1261 	thread_t thread = current_thread();
1262 	const uintptr_t disc = ptrauth_blend_discriminator(&thread->recover, PAC_DISCRIMINATOR_RECOVER);
1263 	const char *panic_msg = "Illegal thread->recover value %p";
1264 
1265 	MANIPULATE_SIGNED_THREAD_STATE(iss,
1266 	    // recover = (vm_offset_t)ptrauth_auth_data((void *)recover, ptrauth_key_function_pointer,
1267 	    //     ptrauth_blend_discriminator(&thread->recover, PAC_DISCRIMINATOR_RECOVER));
1268 	    "mov	x1, %[recover]		\n"
1269 	    "mov	x6, %[disc]		\n"
1270 	    "autia	x1, x6			\n"
1271 #if !__ARM_ARCH_8_6__
1272 	    // if (recover != (vm_offset_t)ptrauth_strip((void *)recover, ptrauth_key_function_pointer)) {
1273 	    "mov	x6, x1			\n"
1274 	    "xpaci	x6			\n"
1275 	    "cmp	x1, x6			\n"
1276 	    "beq	1f			\n"
1277 	    //         panic("Illegal thread->recover value %p", (void *)recover);
1278 	    "mov	x0, %[panic_msg]	\n"
1279 	    "bl		_panic			\n"
1280 	    // }
1281 	    "1:					\n"
1282 #endif /* !__ARM_ARCH_8_6__ */
1283 	    "str	x1, [x0, %[SS64_PC]]	\n",
1284 	    [recover]     "r"(recover),
1285 	    [disc]        "r"(disc),
1286 	    [panic_msg]   "r"(panic_msg)
1287 	    );
1288 #else
1289 	set_saved_state_pc(iss, recover);
1290 #endif
1291 }
1292 
1293 static void
handle_user_abort(arm_saved_state_t * state,uint32_t esr,vm_offset_t fault_addr,fault_status_t fault_code,vm_prot_t fault_type,vm_offset_t recover,expected_fault_handler_t expected_fault_handler)1294 handle_user_abort(arm_saved_state_t *state, uint32_t esr, vm_offset_t fault_addr,
1295     fault_status_t fault_code, vm_prot_t fault_type, vm_offset_t recover, expected_fault_handler_t expected_fault_handler)
1296 {
1297 	exception_type_t           exc      = EXC_BAD_ACCESS;
1298 	mach_exception_data_type_t codes[2];
1299 	mach_msg_type_number_t     numcodes = 2;
1300 	thread_t                   thread   = current_thread();
1301 
1302 	(void)esr;
1303 	(void)expected_fault_handler;
1304 
1305 	if (ml_at_interrupt_context()) {
1306 		panic_with_thread_kernel_state("Apparently on interrupt stack when taking user abort!\n", state);
1307 	}
1308 
1309 	thread->iotier_override = THROTTLE_LEVEL_NONE; /* Reset IO tier override before handling abort from userspace */
1310 
1311 	if (is_vm_fault(fault_code)) {
1312 		vm_map_t        map = thread->map;
1313 		vm_offset_t     vm_fault_addr = fault_addr;
1314 		kern_return_t   result = KERN_FAILURE;
1315 
1316 		assert(map != kernel_map);
1317 
1318 		if (!(fault_type & VM_PROT_EXECUTE)) {
1319 			vm_fault_addr = tbi_clear(fault_addr);
1320 		}
1321 
1322 #if CONFIG_DTRACE
1323 		if (thread->t_dtrace_inprobe) { /* Executing under dtrace_probe? */
1324 			if (dtrace_tally_fault(vm_fault_addr)) { /* Should a user mode fault under dtrace be ignored? */
1325 				if (recover) {
1326 					thread->machine.recover_esr = esr;
1327 					thread->machine.recover_far = vm_fault_addr;
1328 					set_saved_state_pc_to_recovery_handler(state, recover);
1329 				} else {
1330 					panic_with_thread_kernel_state("copyin/out has no recovery point", state);
1331 				}
1332 				return;
1333 			} else {
1334 				panic_with_thread_kernel_state("Unexpected UMW page fault under dtrace_probe", state);
1335 			}
1336 		}
1337 #else
1338 		(void)recover;
1339 #endif
1340 
1341 		/* check to see if it is just a pmap ref/modify fault */
1342 		if (!is_translation_fault(fault_code)) {
1343 			result = arm_fast_fault(map->pmap,
1344 			    vm_fault_addr,
1345 			    fault_type, (fault_code == FSC_ACCESS_FLAG_FAULT_L3), TRUE);
1346 		}
1347 		if (result == KERN_SUCCESS) {
1348 			return;
1349 		}
1350 
1351 		{
1352 			/* We have to fault the page in */
1353 			result = vm_fault(map, vm_fault_addr, fault_type,
1354 			    /* change_wiring */ FALSE, VM_KERN_MEMORY_NONE, THREAD_ABORTSAFE,
1355 			    /* caller_pmap */ NULL, /* caller_pmap_addr */ 0);
1356 		}
1357 		if (result == KERN_SUCCESS || result == KERN_ABORTED) {
1358 			return;
1359 		}
1360 
1361 		/*
1362 		 * vm_fault() should never return KERN_FAILURE for page faults from user space.
1363 		 * If it does, we're leaking preemption disables somewhere in the kernel.
1364 		 */
1365 		if (__improbable(result == KERN_FAILURE)) {
1366 			panic("vm_fault() KERN_FAILURE from user fault on thread %p", thread);
1367 		}
1368 
1369 		codes[0] = result;
1370 	} else if (is_alignment_fault(fault_code)) {
1371 		kern_return_t vmfkr = KERN_SUCCESS;
1372 		thread->machine.recover_esr = 0;
1373 		thread->machine.recover_far = 0;
1374 		int result = handle_alignment_fault_from_user(state, &vmfkr);
1375 		if (result == 0) {
1376 			/* Successfully emulated, or instruction
1377 			 * copyin() for decode/emulation failed.
1378 			 * Continue, or redrive instruction.
1379 			 */
1380 			thread_exception_return();
1381 		} else if (((result == EFAULT) || (result == EINVAL)) &&
1382 		    (thread->machine.recover_esr == 0)) {
1383 			/*
1384 			 * If we didn't actually take a fault, but got one of
1385 			 * these errors, then we failed basic sanity checks of
1386 			 * the fault address.  Treat this as an invalid
1387 			 * address.
1388 			 */
1389 			codes[0] = KERN_INVALID_ADDRESS;
1390 		} else if ((result == EFAULT) &&
1391 		    (thread->machine.recover_esr)) {
1392 			/*
1393 			 * Since alignment aborts are prioritized
1394 			 * ahead of translation aborts, the misaligned
1395 			 * atomic emulation flow may have triggered a
1396 			 * VM pagefault, which the VM could not resolve.
1397 			 * Report the VM fault error in codes[]
1398 			 */
1399 
1400 			codes[0] = vmfkr;
1401 			assertf(vmfkr != KERN_SUCCESS, "Unexpected vmfkr 0x%x", vmfkr);
1402 			/* Cause ESR_EC to reflect an EL0 abort */
1403 			thread->machine.recover_esr &= ~ESR_EC_MASK;
1404 			thread->machine.recover_esr |= (ESR_EC_DABORT_EL0 << ESR_EC_SHIFT);
1405 			set_saved_state_esr(thread->machine.upcb, thread->machine.recover_esr);
1406 			set_saved_state_far(thread->machine.upcb, thread->machine.recover_far);
1407 			fault_addr = thread->machine.recover_far;
1408 		} else {
1409 			/* This was just an unsupported alignment
1410 			 * exception. Misaligned atomic emulation
1411 			 * timeouts fall in this category.
1412 			 */
1413 			codes[0] = EXC_ARM_DA_ALIGN;
1414 		}
1415 	} else if (is_parity_error(fault_code)) {
1416 #if defined(APPLE_ARM64_ARCH_FAMILY)
1417 		if (fault_code == FSC_SYNC_PARITY) {
1418 			arm64_platform_error(state, esr, fault_addr);
1419 			return;
1420 		}
1421 #else
1422 		panic("User parity error.");
1423 #endif
1424 	} else {
1425 		codes[0] = KERN_FAILURE;
1426 	}
1427 
1428 	codes[1] = fault_addr;
1429 #if __has_feature(ptrauth_calls)
1430 	bool is_data_abort = (ESR_EC(esr) == ESR_EC_DABORT_EL0);
1431 	if (user_fault_addr_matches_pac_error_code(fault_addr, is_data_abort)) {
1432 		exc |= EXC_PTRAUTH_BIT;
1433 	}
1434 #endif /* __has_feature(ptrauth_calls) */
1435 	exception_triage(exc, codes, numcodes);
1436 	__builtin_unreachable();
1437 }
1438 
1439 #if __ARM_PAN_AVAILABLE__
1440 static int
is_pan_fault(arm_saved_state_t * state,uint32_t esr,vm_offset_t fault_addr,fault_status_t fault_code)1441 is_pan_fault(arm_saved_state_t *state, uint32_t esr, vm_offset_t fault_addr, fault_status_t fault_code)
1442 {
1443 	// PAN (Privileged Access Never) fault occurs for data read/write in EL1 to
1444 	// virtual address that is readable/writeable from both EL1 and EL0
1445 
1446 	// To check for PAN fault, we evaluate if the following conditions are true:
1447 	// 1. This is a permission fault
1448 	// 2. PAN is enabled
1449 	// 3. AT instruction (on which PAN has no effect) on the same faulting address
1450 	// succeeds
1451 
1452 	vm_offset_t pa;
1453 
1454 	if (!(is_permission_fault(fault_code) && get_saved_state_cpsr(state) & PSR64_PAN)) {
1455 		return FALSE;
1456 	}
1457 
1458 	if (esr & ISS_DA_WNR) {
1459 		pa = mmu_kvtop_wpreflight(fault_addr);
1460 	} else {
1461 		pa = mmu_kvtop(fault_addr);
1462 	}
1463 	return (pa)? TRUE: FALSE;
1464 }
1465 #endif
1466 
1467 static void
handle_kernel_abort(arm_saved_state_t * state,uint32_t esr,vm_offset_t fault_addr,fault_status_t fault_code,vm_prot_t fault_type,vm_offset_t recover,expected_fault_handler_t expected_fault_handler)1468 handle_kernel_abort(arm_saved_state_t *state, uint32_t esr, vm_offset_t fault_addr,
1469     fault_status_t fault_code, vm_prot_t fault_type, vm_offset_t recover, expected_fault_handler_t expected_fault_handler)
1470 {
1471 	thread_t thread = current_thread();
1472 	(void)esr;
1473 
1474 #ifndef CONFIG_XNUPOST
1475 	(void)expected_fault_handler;
1476 #endif /* CONFIG_XNUPOST */
1477 
1478 #if CONFIG_DTRACE
1479 	if (is_vm_fault(fault_code) && thread->t_dtrace_inprobe) { /* Executing under dtrace_probe? */
1480 		if (dtrace_tally_fault(fault_addr)) { /* Should a fault under dtrace be ignored? */
1481 			/*
1482 			 * Point to next instruction, or recovery handler if set.
1483 			 */
1484 			if (recover) {
1485 				thread->machine.recover_esr = esr;
1486 				thread->machine.recover_far = fault_addr;
1487 				set_saved_state_pc_to_recovery_handler(state, recover);
1488 			} else {
1489 				add_saved_state_pc(state, 4);
1490 			}
1491 			return;
1492 		} else {
1493 			panic_with_thread_kernel_state("Unexpected page fault under dtrace_probe", state);
1494 		}
1495 	}
1496 #endif
1497 
1498 	if (ml_at_interrupt_context()) {
1499 		panic_with_thread_kernel_state("Unexpected abort while on interrupt stack.", state);
1500 	}
1501 
1502 	if (is_vm_fault(fault_code)) {
1503 		kern_return_t result = KERN_FAILURE;
1504 		vm_map_t      map;
1505 		int           interruptible;
1506 
1507 		/*
1508 		 * Ensure no faults in the physical aperture. This could happen if
1509 		 * a page table is incorrectly allocated from the read only region
1510 		 * when running with KTRR.
1511 		 */
1512 
1513 #ifdef CONFIG_XNUPOST
1514 		if (expected_fault_handler && expected_fault_handler(state)) {
1515 			return;
1516 		}
1517 #endif /* CONFIG_XNUPOST */
1518 
1519 		if (fault_addr >= gVirtBase && fault_addr < static_memory_end) {
1520 			panic_with_thread_kernel_state("Unexpected fault in kernel static region\n", state);
1521 		}
1522 
1523 		if (VM_KERNEL_ADDRESS(fault_addr) || thread == THREAD_NULL) {
1524 			map = kernel_map;
1525 			interruptible = THREAD_UNINT;
1526 		} else {
1527 			map = thread->map;
1528 
1529 			/**
1530 			 * In the case that the recovery handler is set (e.g., during copyio
1531 			 * and dtrace probes), we don't want the vm_fault() operation to be
1532 			 * aborted early. Those code paths can't handle restarting the
1533 			 * vm_fault() operation so don't allow it to return early without
1534 			 * creating the wanted mapping.
1535 			 */
1536 			interruptible = (recover) ? THREAD_UNINT : THREAD_ABORTSAFE;
1537 		}
1538 
1539 		/* check to see if it is just a pmap ref/modify fault */
1540 		if (!is_translation_fault(fault_code)) {
1541 			result = arm_fast_fault(map->pmap,
1542 			    fault_addr,
1543 			    fault_type, (fault_code == FSC_ACCESS_FLAG_FAULT_L3), FALSE);
1544 			if (result == KERN_SUCCESS) {
1545 				return;
1546 			}
1547 		}
1548 
1549 		if (result != KERN_PROTECTION_FAILURE) {
1550 			/*
1551 			 *  We have to "fault" the page in.
1552 			 */
1553 			result = vm_fault(map, fault_addr, fault_type,
1554 			    /* change_wiring */ FALSE, VM_KERN_MEMORY_NONE, interruptible,
1555 			    /* caller_pmap */ NULL, /* caller_pmap_addr */ 0);
1556 		}
1557 
1558 		if (result == KERN_SUCCESS) {
1559 			return;
1560 		}
1561 
1562 		/*
1563 		 *  If we have a recover handler, invoke it now.
1564 		 */
1565 		if (recover) {
1566 			thread->machine.recover_esr = esr;
1567 			thread->machine.recover_far = fault_addr;
1568 			set_saved_state_pc_to_recovery_handler(state, recover);
1569 			return;
1570 		}
1571 
1572 #if __ARM_PAN_AVAILABLE__
1573 		if (is_pan_fault(state, esr, fault_addr, fault_code)) {
1574 			panic_with_thread_kernel_state("Privileged access never abort.", state);
1575 		}
1576 #endif
1577 	} else if (is_alignment_fault(fault_code)) {
1578 		if (recover) {
1579 			thread->machine.recover_esr = esr;
1580 			thread->machine.recover_far = fault_addr;
1581 			set_saved_state_pc_to_recovery_handler(state, recover);
1582 			return;
1583 		}
1584 		panic_with_thread_kernel_state("Unaligned kernel data abort.", state);
1585 	} else if (is_parity_error(fault_code)) {
1586 #if defined(APPLE_ARM64_ARCH_FAMILY)
1587 		if (fault_code == FSC_SYNC_PARITY) {
1588 			arm64_platform_error(state, esr, fault_addr);
1589 			return;
1590 		}
1591 #else
1592 		panic_with_thread_kernel_state("Kernel parity error.", state);
1593 #endif
1594 	} else {
1595 		kprintf("Unclassified kernel abort (fault_code=0x%x)\n", fault_code);
1596 	}
1597 
1598 	panic_with_thread_kernel_state("Kernel data abort.", state);
1599 }
1600 
1601 extern void syscall_trace(struct arm_saved_state * regs);
1602 
1603 static void
handle_svc(arm_saved_state_t * state)1604 handle_svc(arm_saved_state_t *state)
1605 {
1606 	int      trap_no = get_saved_state_svc_number(state);
1607 	thread_t thread  = current_thread();
1608 	struct   proc *p;
1609 
1610 #define handle_svc_kprintf(x...) /* kprintf("handle_svc: " x) */
1611 
1612 #define TRACE_SYSCALL 1
1613 #if TRACE_SYSCALL
1614 	syscall_trace(state);
1615 #endif
1616 
1617 	thread->iotier_override = THROTTLE_LEVEL_NONE; /* Reset IO tier override before handling SVC from userspace */
1618 
1619 	if (trap_no == (int)PLATFORM_SYSCALL_TRAP_NO) {
1620 		platform_syscall(state);
1621 		panic("Returned from platform_syscall()?");
1622 	}
1623 
1624 	mach_kauth_cred_thread_update();
1625 
1626 	if (trap_no < 0) {
1627 		switch (trap_no) {
1628 		case MACH_ARM_TRAP_ABSTIME:
1629 			handle_mach_absolute_time_trap(state);
1630 			return;
1631 		case MACH_ARM_TRAP_CONTTIME:
1632 			handle_mach_continuous_time_trap(state);
1633 			return;
1634 		}
1635 
1636 		/* Counting perhaps better in the handler, but this is how it's been done */
1637 		thread->syscalls_mach++;
1638 		mach_syscall(state);
1639 	} else {
1640 		/* Counting perhaps better in the handler, but this is how it's been done */
1641 		thread->syscalls_unix++;
1642 		p = get_bsdthreadtask_info(thread);
1643 
1644 		assert(p);
1645 
1646 		unix_syscall(state, thread, p);
1647 	}
1648 }
1649 
1650 static void
handle_mach_absolute_time_trap(arm_saved_state_t * state)1651 handle_mach_absolute_time_trap(arm_saved_state_t *state)
1652 {
1653 	uint64_t now = mach_absolute_time();
1654 	saved_state64(state)->x[0] = now;
1655 }
1656 
1657 static void
handle_mach_continuous_time_trap(arm_saved_state_t * state)1658 handle_mach_continuous_time_trap(arm_saved_state_t *state)
1659 {
1660 	uint64_t now = mach_continuous_time();
1661 	saved_state64(state)->x[0] = now;
1662 }
1663 
1664 
1665 __attribute__((noreturn))
1666 static void
handle_msr_trap(arm_saved_state_t * state,uint32_t esr)1667 handle_msr_trap(arm_saved_state_t *state, uint32_t esr)
1668 {
1669 	exception_type_t           exception = EXC_BAD_INSTRUCTION;
1670 	mach_exception_data_type_t codes[2]  = {EXC_ARM_UNDEFINED};
1671 	mach_msg_type_number_t     numcodes  = 2;
1672 	uint32_t                   instr     = 0;
1673 
1674 	if (!is_saved_state64(state)) {
1675 		panic("MSR/MRS trap (ESR 0x%x) from 32-bit state", esr);
1676 	}
1677 
1678 	if (PSR64_IS_KERNEL(get_saved_state_cpsr(state))) {
1679 		panic("MSR/MRS trap (ESR 0x%x) from kernel", esr);
1680 	}
1681 
1682 	COPYIN(get_saved_state_pc(state), (char *)&instr, sizeof(instr));
1683 	codes[1] = instr;
1684 
1685 	exception_triage(exception, codes, numcodes);
1686 	__builtin_unreachable();
1687 }
1688 
1689 #ifdef __ARM_ARCH_8_6__
1690 static void
autxx_instruction_extract_reg(uint32_t instr,char reg[4])1691 autxx_instruction_extract_reg(uint32_t instr, char reg[4])
1692 {
1693 	unsigned int rd = ARM64_INSTR_AUTxx_RD_GET(instr);
1694 	switch (rd) {
1695 	case 29:
1696 		strncpy(reg, "fp", 4);
1697 		return;
1698 
1699 	case 30:
1700 		strncpy(reg, "lr", 4);
1701 		return;
1702 
1703 	case 31:
1704 		strncpy(reg, "xzr", 4);
1705 		return;
1706 
1707 	default:
1708 		snprintf(reg, 4, "x%u", rd);
1709 		return;
1710 	}
1711 }
1712 
1713 static const char *
autix_system_instruction_extract_reg(uint32_t instr)1714 autix_system_instruction_extract_reg(uint32_t instr)
1715 {
1716 	unsigned int crm_op2 = ARM64_INSTR_AUTIx_SYSTEM_CRM_OP2_GET(instr);
1717 	if (crm_op2 == ARM64_INSTR_AUTIx_SYSTEM_CRM_OP2_AUTIA1716 ||
1718 	    crm_op2 == ARM64_INSTR_AUTIx_SYSTEM_CRM_OP2_AUTIB1716) {
1719 		return "x17";
1720 	} else {
1721 		return "lr";
1722 	}
1723 }
1724 
1725 static void
handle_pac_fail(arm_saved_state_t * state,uint32_t esr)1726 handle_pac_fail(arm_saved_state_t *state, uint32_t esr)
1727 {
1728 	exception_type_t           exception = EXC_BAD_ACCESS | EXC_PTRAUTH_BIT;
1729 	mach_exception_data_type_t codes[2]  = {EXC_ARM_PAC_FAIL};
1730 	mach_msg_type_number_t     numcodes  = 2;
1731 	uint32_t                   instr     = 0;
1732 
1733 	if (!is_saved_state64(state)) {
1734 		panic("PAC failure (ESR 0x%x) from 32-bit state", esr);
1735 	}
1736 
1737 	COPYIN(get_saved_state_pc(state), (char *)&instr, sizeof(instr));
1738 
1739 	if (PSR64_IS_KERNEL(get_saved_state_cpsr(state))) {
1740 #define GENERIC_PAC_FAILURE_MSG_FMT "PAC failure from kernel with %s key"
1741 #define AUTXX_MSG_FMT GENERIC_PAC_FAILURE_MSG_FMT " while authing %s"
1742 #define GENERIC_MSG_FMT GENERIC_PAC_FAILURE_MSG_FMT
1743 
1744 		char msg[strlen(AUTXX_MSG_FMT)
1745 		- strlen("%s") + strlen("IA")
1746 		- strlen("%s") + strlen("xzr")
1747 		+ 1];
1748 		ptrauth_key key = (ptrauth_key)(esr & 0x3);
1749 		const char *key_str = ptrauth_key_to_string(key);
1750 
1751 		if (ARM64_INSTR_IS_AUTxx(instr)) {
1752 			char reg[4];
1753 			autxx_instruction_extract_reg(instr, reg);
1754 			snprintf(msg, sizeof(msg), AUTXX_MSG_FMT, key_str, reg);
1755 		} else if (ARM64_INSTR_IS_AUTIx_SYSTEM(instr)) {
1756 			const char *reg = autix_system_instruction_extract_reg(instr);
1757 			snprintf(msg, sizeof(msg), AUTXX_MSG_FMT, key_str, reg);
1758 		} else {
1759 			snprintf(msg, sizeof(msg), GENERIC_MSG_FMT, key_str);
1760 		}
1761 		panic_with_thread_kernel_state(msg, state);
1762 	}
1763 
1764 	codes[1] = instr;
1765 
1766 	exception_triage(exception, codes, numcodes);
1767 	__builtin_unreachable();
1768 }
1769 #endif /* __ARM_ARCH_8_6__ */
1770 
1771 static void
handle_user_trapped_instruction32(arm_saved_state_t * state,uint32_t esr)1772 handle_user_trapped_instruction32(arm_saved_state_t *state, uint32_t esr)
1773 {
1774 	exception_type_t           exception = EXC_BAD_INSTRUCTION;
1775 	mach_exception_data_type_t codes[2]  = {EXC_ARM_UNDEFINED};
1776 	mach_msg_type_number_t     numcodes  = 2;
1777 	uint32_t                   instr;
1778 
1779 	if (is_saved_state64(state)) {
1780 		panic("ESR (0x%x) for instruction trapped from U32, but saved state is 64-bit.", esr);
1781 	}
1782 
1783 	if (PSR64_IS_KERNEL(get_saved_state_cpsr(state))) {
1784 		panic("ESR (0x%x) for instruction trapped from U32, actually came from kernel?", esr);
1785 	}
1786 
1787 	COPYIN(get_saved_state_pc(state), (char *)&instr, sizeof(instr));
1788 	codes[1] = instr;
1789 
1790 	exception_triage(exception, codes, numcodes);
1791 	__builtin_unreachable();
1792 }
1793 
1794 static void
handle_simd_trap(arm_saved_state_t * state,uint32_t esr)1795 handle_simd_trap(arm_saved_state_t *state, uint32_t esr)
1796 {
1797 	exception_type_t           exception = EXC_BAD_INSTRUCTION;
1798 	mach_exception_data_type_t codes[2]  = {EXC_ARM_UNDEFINED};
1799 	mach_msg_type_number_t     numcodes  = 2;
1800 	uint32_t                   instr     = 0;
1801 
1802 	if (PSR64_IS_KERNEL(get_saved_state_cpsr(state))) {
1803 		panic("ESR (0x%x) for SIMD trap from userland, actually came from kernel?", esr);
1804 	}
1805 
1806 	COPYIN(get_saved_state_pc(state), (char *)&instr, sizeof(instr));
1807 	codes[1] = instr;
1808 
1809 	exception_triage(exception, codes, numcodes);
1810 	__builtin_unreachable();
1811 }
1812 
1813 void
sleh_irq(arm_saved_state_t * state)1814 sleh_irq(arm_saved_state_t *state)
1815 {
1816 	cpu_data_t * cdp __unused             = getCpuDatap();
1817 #if MACH_ASSERT
1818 	int preemption_level = sleh_get_preemption_level();
1819 #endif
1820 
1821 
1822 	sleh_interrupt_handler_prologue(state, DBG_INTR_TYPE_OTHER);
1823 
1824 #if USE_APPLEARMSMP
1825 	PE_handle_ext_interrupt();
1826 #else
1827 	/* Run the registered interrupt handler. */
1828 	cdp->interrupt_handler(cdp->interrupt_target,
1829 	    cdp->interrupt_refCon,
1830 	    cdp->interrupt_nub,
1831 	    cdp->interrupt_source);
1832 #endif
1833 
1834 	entropy_collect();
1835 
1836 
1837 	sleh_interrupt_handler_epilogue();
1838 #if MACH_ASSERT
1839 	if (preemption_level != sleh_get_preemption_level()) {
1840 		panic("irq handler %p changed preemption level from %d to %d", cdp->interrupt_handler, preemption_level, sleh_get_preemption_level());
1841 	}
1842 #endif
1843 }
1844 
1845 void
sleh_fiq(arm_saved_state_t * state)1846 sleh_fiq(arm_saved_state_t *state)
1847 {
1848 	unsigned int type   = DBG_INTR_TYPE_UNKNOWN;
1849 #if MACH_ASSERT
1850 	int preemption_level = sleh_get_preemption_level();
1851 #endif
1852 
1853 #if MONOTONIC_FIQ
1854 	uint64_t pmcr0 = 0, upmsr = 0;
1855 #endif /* MONOTONIC_FIQ */
1856 
1857 #if defined(HAS_IPI)
1858 	boolean_t    is_ipi = FALSE;
1859 	uint64_t     ipi_sr = 0;
1860 
1861 	if (gFastIPI) {
1862 		MRS(ipi_sr, "S3_5_C15_C1_1");
1863 
1864 		if (ipi_sr & 1) {
1865 			is_ipi = TRUE;
1866 		}
1867 	}
1868 
1869 	if (is_ipi) {
1870 		type = DBG_INTR_TYPE_IPI;
1871 	} else
1872 #endif /* defined(HAS_IPI) */
1873 	if (ml_get_timer_pending()) {
1874 		type = DBG_INTR_TYPE_TIMER;
1875 	}
1876 #if MONOTONIC_FIQ
1877 	/* Consult the PMI sysregs last, after IPI/timer
1878 	 * classification.
1879 	 */
1880 	else if (mt_pmi_pending(&pmcr0, &upmsr)) {
1881 		type = DBG_INTR_TYPE_PMI;
1882 	}
1883 #endif /* MONOTONIC_FIQ */
1884 
1885 	sleh_interrupt_handler_prologue(state, type);
1886 
1887 #if APPLEVIRTUALPLATFORM
1888 	uint64_t iar = __builtin_arm_rsr64("ICC_IAR0_EL1");
1889 #endif
1890 
1891 #if defined(HAS_IPI)
1892 	if (is_ipi) {
1893 		/*
1894 		 * Order is important here: we must ack the IPI by writing IPI_SR
1895 		 * before we call cpu_signal_handler().  Otherwise, there will be
1896 		 * a window between the completion of pending-signal processing in
1897 		 * cpu_signal_handler() and the ack during which a newly-issued
1898 		 * IPI to this CPU may be lost.  ISB is required to ensure the msr
1899 		 * is retired before execution of cpu_signal_handler().
1900 		 */
1901 		MSR("S3_5_C15_C1_1", ipi_sr);
1902 		__builtin_arm_isb(ISB_SY);
1903 		cpu_signal_handler();
1904 	} else
1905 #endif /* defined(HAS_IPI) */
1906 #if MONOTONIC_FIQ
1907 	if (type == DBG_INTR_TYPE_PMI) {
1908 		INTERRUPT_MASKED_DEBUG_START(mt_fiq, DBG_INTR_TYPE_PMI);
1909 		mt_fiq(getCpuDatap(), pmcr0, upmsr);
1910 		INTERRUPT_MASKED_DEBUG_END();
1911 	} else
1912 #endif /* MONOTONIC_FIQ */
1913 	{
1914 		/*
1915 		 * We don't know that this is a timer, but we don't have insight into
1916 		 * the other interrupts that go down this path.
1917 		 */
1918 
1919 		cpu_data_t *cdp = getCpuDatap();
1920 
1921 		cdp->cpu_decrementer = -1; /* Large */
1922 
1923 		/*
1924 		 * ARM64_TODO: whether we're coming from userland is ignored right now.
1925 		 * We can easily thread it through, but not bothering for the
1926 		 * moment (AArch32 doesn't either).
1927 		 */
1928 		INTERRUPT_MASKED_DEBUG_START(rtclock_intr, DBG_INTR_TYPE_TIMER);
1929 		rtclock_intr(TRUE);
1930 		INTERRUPT_MASKED_DEBUG_END();
1931 	}
1932 
1933 #if APPLEVIRTUALPLATFORM
1934 	if (iar != GIC_SPURIOUS_IRQ) {
1935 		__builtin_arm_wsr64("ICC_EOIR0_EL1", iar);
1936 		__builtin_arm_isb(ISB_SY);
1937 	}
1938 #endif
1939 
1940 	sleh_interrupt_handler_epilogue();
1941 #if MACH_ASSERT
1942 	if (preemption_level != sleh_get_preemption_level()) {
1943 		panic("fiq type %u changed preemption level from %d to %d", type, preemption_level, sleh_get_preemption_level());
1944 	}
1945 #endif
1946 }
1947 
1948 void
sleh_serror(arm_context_t * context,uint32_t esr,vm_offset_t far)1949 sleh_serror(arm_context_t *context, uint32_t esr, vm_offset_t far)
1950 {
1951 	task_vtimer_check(current_thread());
1952 
1953 	KDBG_RELEASE(MACHDBG_CODE(DBG_MACH_EXCP_SERR_ARM, 0) | DBG_FUNC_START,
1954 	    esr, VM_KERNEL_ADDRHIDE(far));
1955 	arm_saved_state_t *state = &context->ss;
1956 #if MACH_ASSERT
1957 	int preemption_level = sleh_get_preemption_level();
1958 #endif
1959 
1960 
1961 	ASSERT_CONTEXT_SANITY(context);
1962 	arm64_platform_error(state, esr, far);
1963 #if MACH_ASSERT
1964 	if (preemption_level != sleh_get_preemption_level()) {
1965 		panic("serror changed preemption level from %d to %d", preemption_level, sleh_get_preemption_level());
1966 	}
1967 #endif
1968 	KDBG_RELEASE(MACHDBG_CODE(DBG_MACH_EXCP_SERR_ARM, 0) | DBG_FUNC_END,
1969 	    esr, VM_KERNEL_ADDRHIDE(far));
1970 }
1971 
1972 void
mach_syscall_trace_exit(unsigned int retval,unsigned int call_number)1973 mach_syscall_trace_exit(unsigned int retval,
1974     unsigned int call_number)
1975 {
1976 	KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE,
1977 	    MACHDBG_CODE(DBG_MACH_EXCP_SC, (call_number)) |
1978 	    DBG_FUNC_END, retval, 0, 0, 0, 0);
1979 }
1980 
1981 __attribute__((noreturn))
1982 void
thread_syscall_return(kern_return_t error)1983 thread_syscall_return(kern_return_t error)
1984 {
1985 	thread_t thread;
1986 	struct arm_saved_state *state;
1987 
1988 	thread = current_thread();
1989 	state = get_user_regs(thread);
1990 
1991 	assert(is_saved_state64(state));
1992 	saved_state64(state)->x[0] = error;
1993 
1994 #if MACH_ASSERT
1995 	kern_allocation_name_t
1996 	prior __assert_only = thread_get_kernel_state(thread)->allocation_name;
1997 	assertf(prior == NULL, "thread_set_allocation_name(\"%s\") not cleared", kern_allocation_get_name(prior));
1998 #endif /* MACH_ASSERT */
1999 
2000 	if (kdebug_enable) {
2001 		/* Invert syscall number (negative for a mach syscall) */
2002 		mach_syscall_trace_exit(error, (-1) * get_saved_state_svc_number(state));
2003 	}
2004 
2005 	thread_exception_return();
2006 }
2007 
2008 void
syscall_trace(struct arm_saved_state * regs __unused)2009 syscall_trace(
2010 	struct arm_saved_state * regs __unused)
2011 {
2012 	/* kprintf("syscall: %d\n", saved_state64(regs)->x[16]);  */
2013 }
2014 
2015 static void
sleh_interrupt_handler_prologue(arm_saved_state_t * state,unsigned int type)2016 sleh_interrupt_handler_prologue(arm_saved_state_t *state, unsigned int type)
2017 {
2018 	boolean_t is_user = PSR64_IS_USER(get_saved_state_cpsr(state));
2019 
2020 	task_vtimer_check(current_thread());
2021 
2022 	uint64_t pc = is_user ? get_saved_state_pc(state) :
2023 	    VM_KERNEL_UNSLIDE(get_saved_state_pc(state));
2024 
2025 	KDBG_RELEASE(MACHDBG_CODE(DBG_MACH_EXCP_INTR, 0) | DBG_FUNC_START,
2026 	    0, pc, is_user, type);
2027 
2028 #if CONFIG_TELEMETRY
2029 	if (telemetry_needs_record) {
2030 		telemetry_mark_curthread(is_user, FALSE);
2031 	}
2032 #endif /* CONFIG_TELEMETRY */
2033 }
2034 
2035 static void
sleh_interrupt_handler_epilogue(void)2036 sleh_interrupt_handler_epilogue(void)
2037 {
2038 #if KPERF
2039 	kperf_interrupt();
2040 #endif /* KPERF */
2041 	KDBG_RELEASE(MACHDBG_CODE(DBG_MACH_EXCP_INTR, 0) | DBG_FUNC_END);
2042 }
2043 
2044 void
sleh_invalid_stack(arm_context_t * context,uint32_t esr __unused,vm_offset_t far __unused)2045 sleh_invalid_stack(arm_context_t *context, uint32_t esr __unused, vm_offset_t far __unused)
2046 {
2047 	thread_t thread = current_thread();
2048 	vm_offset_t kernel_stack_bottom, sp;
2049 
2050 	sp = get_saved_state_sp(&context->ss);
2051 	kernel_stack_bottom = round_page(thread->machine.kstackptr) - KERNEL_STACK_SIZE;
2052 
2053 	if ((sp < kernel_stack_bottom) && (sp >= (kernel_stack_bottom - PAGE_SIZE))) {
2054 		panic_with_thread_kernel_state("Invalid kernel stack pointer (probable overflow).", &context->ss);
2055 	}
2056 
2057 	panic_with_thread_kernel_state("Invalid kernel stack pointer (probable corruption).", &context->ss);
2058 }
2059 
2060