xref: /xnu-8796.121.2/osfmk/arm64/sleh.c (revision c54f35ca767986246321eb901baf8f5ff7923f6a)
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/restartable.h>
43 #include <kern/socd_client.h>
44 #include <kern/thread.h>
45 #include <mach/exception.h>
46 #include <mach/arm/traps.h>
47 #include <mach/vm_types.h>
48 #include <mach/machine/thread_status.h>
49 
50 #include <machine/atomic.h>
51 #include <machine/limits.h>
52 
53 #include <pexpert/arm/protos.h>
54 #include <pexpert/arm64/apple_arm64_regs.h>
55 #include <pexpert/arm64/board_config.h>
56 
57 #include <vm/vm_page.h>
58 #include <vm/pmap.h>
59 #include <vm/vm_fault.h>
60 #include <vm/vm_kern.h>
61 
62 #include <sys/errno.h>
63 #include <sys/kdebug.h>
64 #include <kperf/kperf.h>
65 
66 #include <kern/policy_internal.h>
67 #if CONFIG_TELEMETRY
68 #include <kern/telemetry.h>
69 #endif
70 
71 #include <prng/entropy.h>
72 
73 
74 
75 
76 #include <arm64/platform_error_handler.h>
77 
78 #if KASAN_TBI
79 #include <san/kasan.h>
80 #endif /* KASAN_TBI */
81 
82 #if CONFIG_UBSAN_MINIMAL
83 #include <san/ubsan_minimal.h>
84 #endif /* CONFIG_UBSAN_MINIMAL */
85 
86 
87 #ifndef __arm64__
88 #error Should only be compiling for arm64.
89 #endif
90 
91 #define TEST_CONTEXT32_SANITY(context) \
92 	(context->ss.ash.flavor == ARM_SAVED_STATE32 && context->ss.ash.count == ARM_SAVED_STATE32_COUNT && \
93 	 context->ns.nsh.flavor == ARM_NEON_SAVED_STATE32 && context->ns.nsh.count == ARM_NEON_SAVED_STATE32_COUNT)
94 
95 #define TEST_CONTEXT64_SANITY(context) \
96 	(context->ss.ash.flavor == ARM_SAVED_STATE64 && context->ss.ash.count == ARM_SAVED_STATE64_COUNT && \
97 	 context->ns.nsh.flavor == ARM_NEON_SAVED_STATE64 && context->ns.nsh.count == ARM_NEON_SAVED_STATE64_COUNT)
98 
99 #define ASSERT_CONTEXT_SANITY(context) \
100 	assert(TEST_CONTEXT32_SANITY(context) || TEST_CONTEXT64_SANITY(context))
101 
102 
103 #define COPYIN(src, dst, size)                           \
104 	(PSR64_IS_KERNEL(get_saved_state_cpsr(state))) ? \
105 	copyin_kern(src, dst, size) :                    \
106 	copyin(src, dst, size)
107 
108 #define COPYOUT(src, dst, size)                          \
109 	(PSR64_IS_KERNEL(get_saved_state_cpsr(state))) ? \
110 	copyout_kern(src, dst, size)                   : \
111 	copyout(src, dst, size)
112 
113 // Below is for concatenating a string param to a string literal
114 #define STR1(x) #x
115 #define STR(x) STR1(x)
116 
117 #define ARM64_KDBG_CODE_KERNEL (0 << 8)
118 #define ARM64_KDBG_CODE_USER   (1 << 8)
119 #define ARM64_KDBG_CODE_GUEST  (2 << 8)
120 
121 _Static_assert(ARM64_KDBG_CODE_GUEST <= KDBG_CODE_MAX, "arm64 KDBG trace codes out of range");
122 _Static_assert(ARM64_KDBG_CODE_GUEST <= UINT16_MAX, "arm64 KDBG trace codes out of range");
123 
124 void panic_with_thread_kernel_state(const char *msg, arm_saved_state_t *ss) __abortlike;
125 
126 void sleh_synchronous_sp1(arm_context_t *, uint32_t, vm_offset_t) __abortlike;
127 void sleh_synchronous(arm_context_t *, uint32_t, vm_offset_t);
128 
129 
130 
131 void sleh_irq(arm_saved_state_t *);
132 void sleh_fiq(arm_saved_state_t *);
133 void sleh_serror(arm_context_t *context, uint32_t esr, vm_offset_t far);
134 void sleh_invalid_stack(arm_context_t *context, uint32_t esr, vm_offset_t far) __dead2;
135 
136 static void sleh_interrupt_handler_prologue(arm_saved_state_t *, unsigned int type);
137 static void sleh_interrupt_handler_epilogue(void);
138 
139 static void handle_svc(arm_saved_state_t *);
140 static void handle_mach_absolute_time_trap(arm_saved_state_t *);
141 static void handle_mach_continuous_time_trap(arm_saved_state_t *);
142 
143 static void handle_msr_trap(arm_saved_state_t *state, uint32_t esr);
144 #if __has_feature(ptrauth_calls)
145 static void handle_pac_fail(arm_saved_state_t *state, uint32_t esr) __dead2;
146 #endif
147 
148 extern kern_return_t arm_fast_fault(pmap_t, vm_map_address_t, vm_prot_t, bool, bool);
149 
150 static void handle_uncategorized(arm_saved_state_t *);
151 
152 /*
153  * For UBSan trap and continue handling, we must be able to recover
154  * from handle_kernel_breakpoint().
155  */
156 #if !CONFIG_UBSAN_MINIMAL
157 __dead2
158 #endif /* CONFIG_UBSAN_MINIMAL */
159 static void handle_kernel_breakpoint(arm_saved_state_t *, uint32_t);
160 
161 static void handle_breakpoint(arm_saved_state_t *, uint32_t) __dead2;
162 
163 typedef void (*abort_inspector_t)(uint32_t, fault_status_t *, vm_prot_t *);
164 static void inspect_instruction_abort(uint32_t, fault_status_t *, vm_prot_t *);
165 static void inspect_data_abort(uint32_t, fault_status_t *, vm_prot_t *);
166 
167 static int is_vm_fault(fault_status_t);
168 static int is_translation_fault(fault_status_t);
169 static int is_alignment_fault(fault_status_t);
170 
171 typedef void (*abort_handler_t)(arm_saved_state_t *, uint32_t, vm_offset_t, fault_status_t, vm_prot_t, expected_fault_handler_t);
172 static void handle_user_abort(arm_saved_state_t *, uint32_t, vm_offset_t, fault_status_t, vm_prot_t, expected_fault_handler_t);
173 static void handle_kernel_abort(arm_saved_state_t *, uint32_t, vm_offset_t, fault_status_t, vm_prot_t, expected_fault_handler_t);
174 
175 static void handle_pc_align(arm_saved_state_t *ss) __dead2;
176 static void handle_sp_align(arm_saved_state_t *ss) __dead2;
177 static void handle_sw_step_debug(arm_saved_state_t *ss) __dead2;
178 static void handle_wf_trap(arm_saved_state_t *ss) __dead2;
179 static void handle_fp_trap(arm_saved_state_t *ss, uint32_t esr) __dead2;
180 
181 static void handle_watchpoint(vm_offset_t fault_addr) __dead2;
182 
183 static void handle_abort(arm_saved_state_t *, uint32_t, vm_offset_t, abort_inspector_t, abort_handler_t, expected_fault_handler_t);
184 
185 static void handle_user_trapped_instruction32(arm_saved_state_t *, uint32_t esr) __dead2;
186 
187 static void handle_simd_trap(arm_saved_state_t *, uint32_t esr) __dead2;
188 
189 extern void mach_kauth_cred_thread_update(void);
190 void   mach_syscall_trace_exit(unsigned int retval, unsigned int call_number);
191 
192 struct proc;
193 
194 typedef uint32_t arm64_instr_t;
195 
196 extern void
197 unix_syscall(struct arm_saved_state * regs, thread_t thread_act, struct proc * proc);
198 
199 extern void
200 mach_syscall(struct arm_saved_state*);
201 
202 #if CONFIG_DTRACE
203 extern kern_return_t dtrace_user_probe(arm_saved_state_t* regs);
204 extern boolean_t dtrace_tally_fault(user_addr_t);
205 
206 /*
207  * Traps for userland processing. Can't include bsd/sys/fasttrap_isa.h, so copy
208  * and paste the trap instructions
209  * over from that file. Need to keep these in sync!
210  */
211 #define FASTTRAP_ARM32_INSTR 0xe7ffdefc
212 #define FASTTRAP_THUMB32_INSTR 0xdefc
213 #define FASTTRAP_ARM64_INSTR 0xe7eeee7e
214 
215 #define FASTTRAP_ARM32_RET_INSTR 0xe7ffdefb
216 #define FASTTRAP_THUMB32_RET_INSTR 0xdefb
217 #define FASTTRAP_ARM64_RET_INSTR 0xe7eeee7d
218 
219 /* See <rdar://problem/4613924> */
220 perfCallback tempDTraceTrapHook = NULL; /* Pointer to DTrace fbt trap hook routine */
221 #endif
222 
223 
224 
225 extern void arm64_thread_exception_return(void) __dead2;
226 
227 #if defined(APPLETYPHOON)
228 #define CPU_NAME "Typhoon"
229 #elif defined(APPLETWISTER)
230 #define CPU_NAME "Twister"
231 #elif defined(APPLEHURRICANE)
232 #define CPU_NAME "Hurricane"
233 #elif defined(APPLELIGHTNING)
234 #define CPU_NAME "Lightning"
235 #else
236 #define CPU_NAME "Unknown"
237 #endif
238 
239 #if (CONFIG_KERNEL_INTEGRITY && defined(KERNEL_INTEGRITY_WT))
240 #define ESR_WT_SERROR(esr) (((esr) & 0xffffff00) == 0xbf575400)
241 #define ESR_WT_REASON(esr) ((esr) & 0xff)
242 
243 #define WT_REASON_NONE           0
244 #define WT_REASON_INTEGRITY_FAIL 1
245 #define WT_REASON_BAD_SYSCALL    2
246 #define WT_REASON_NOT_LOCKED     3
247 #define WT_REASON_ALREADY_LOCKED 4
248 #define WT_REASON_SW_REQ         5
249 #define WT_REASON_PT_INVALID     6
250 #define WT_REASON_PT_VIOLATION   7
251 #define WT_REASON_REG_VIOLATION  8
252 #endif
253 
254 #if defined(HAS_IPI)
255 void cpu_signal_handler(void);
256 extern unsigned int gFastIPI;
257 #endif /* defined(HAS_IPI) */
258 
259 static arm_saved_state64_t *original_faulting_state = NULL;
260 
261 TUNABLE(bool, fp_exceptions_enabled, "-fp_exceptions", false);
262 
263 extern vm_offset_t static_memory_end;
264 
265 /*
266  * Fault copyio_recovery_entry in copyin/copyout routines.
267  *
268  * Offets are expressed in bytes from &copy_recovery_table
269  */
270 struct copyio_recovery_entry {
271 	ptrdiff_t cre_start;
272 	ptrdiff_t cre_end;
273 	ptrdiff_t cre_recovery;
274 };
275 
276 extern struct copyio_recovery_entry copyio_recover_table[];
277 extern struct copyio_recovery_entry copyio_recover_table_end[];
278 
279 static inline ptrdiff_t
copyio_recovery_offset(uintptr_t addr)280 copyio_recovery_offset(uintptr_t addr)
281 {
282 	return (ptrdiff_t)(addr - (uintptr_t)copyio_recover_table);
283 }
284 
285 static inline uintptr_t
copyio_recovery_addr(ptrdiff_t offset)286 copyio_recovery_addr(ptrdiff_t offset)
287 {
288 	return (uintptr_t)copyio_recover_table + (uintptr_t)offset;
289 }
290 
291 static inline struct copyio_recovery_entry *
find_copyio_recovery_entry(arm_saved_state_t * state)292 find_copyio_recovery_entry(arm_saved_state_t *state)
293 {
294 	ptrdiff_t offset = copyio_recovery_offset(get_saved_state_pc(state));
295 	struct copyio_recovery_entry *e;
296 
297 	for (e = copyio_recover_table; e < copyio_recover_table_end; e++) {
298 		if (offset >= e->cre_start && offset < e->cre_end) {
299 			return e;
300 		}
301 	}
302 
303 	return NULL;
304 }
305 
306 static inline uintptr_t
copyio_recovery_get_recover_addr(arm_saved_state_t * state)307 copyio_recovery_get_recover_addr(
308 	arm_saved_state_t              *state)
309 {
310 	struct copyio_recovery_entry *e = find_copyio_recovery_entry(state);
311 	if (e == NULL) {
312 		panic("copyio recovery: couldn't find a range for %p",
313 		    (void *)get_saved_state_pc(state));
314 	}
315 	return copyio_recovery_addr(e->cre_recovery);
316 }
317 
318 static inline int
is_vm_fault(fault_status_t status)319 is_vm_fault(fault_status_t status)
320 {
321 	switch (status) {
322 	case FSC_TRANSLATION_FAULT_L0:
323 	case FSC_TRANSLATION_FAULT_L1:
324 	case FSC_TRANSLATION_FAULT_L2:
325 	case FSC_TRANSLATION_FAULT_L3:
326 	case FSC_ACCESS_FLAG_FAULT_L1:
327 	case FSC_ACCESS_FLAG_FAULT_L2:
328 	case FSC_ACCESS_FLAG_FAULT_L3:
329 	case FSC_PERMISSION_FAULT_L1:
330 	case FSC_PERMISSION_FAULT_L2:
331 	case FSC_PERMISSION_FAULT_L3:
332 		return TRUE;
333 	default:
334 		return FALSE;
335 	}
336 }
337 
338 static inline int
is_translation_fault(fault_status_t status)339 is_translation_fault(fault_status_t status)
340 {
341 	switch (status) {
342 	case FSC_TRANSLATION_FAULT_L0:
343 	case FSC_TRANSLATION_FAULT_L1:
344 	case FSC_TRANSLATION_FAULT_L2:
345 	case FSC_TRANSLATION_FAULT_L3:
346 		return TRUE;
347 	default:
348 		return FALSE;
349 	}
350 }
351 
352 static inline int
is_permission_fault(fault_status_t status)353 is_permission_fault(fault_status_t status)
354 {
355 	switch (status) {
356 	case FSC_PERMISSION_FAULT_L1:
357 	case FSC_PERMISSION_FAULT_L2:
358 	case FSC_PERMISSION_FAULT_L3:
359 		return TRUE;
360 	default:
361 		return FALSE;
362 	}
363 }
364 
365 static inline int
is_alignment_fault(fault_status_t status)366 is_alignment_fault(fault_status_t status)
367 {
368 	return status == FSC_ALIGNMENT_FAULT;
369 }
370 
371 static inline int
is_parity_error(fault_status_t status)372 is_parity_error(fault_status_t status)
373 {
374 	switch (status) {
375 	/*
376 	 * TODO: According to ARM ARM, Async Parity (0b011001) is a DFSC that is
377 	 * only applicable to AArch32 HSR register. Can this be removed?
378 	 */
379 	case FSC_ASYNC_PARITY:
380 	case FSC_SYNC_PARITY:
381 	case FSC_SYNC_PARITY_TT_L1:
382 	case FSC_SYNC_PARITY_TT_L2:
383 	case FSC_SYNC_PARITY_TT_L3:
384 		return TRUE;
385 	default:
386 		return FALSE;
387 	}
388 }
389 
390 
391 __dead2 __unused
392 static void
arm64_implementation_specific_error(arm_saved_state_t * state,uint32_t esr,vm_offset_t far)393 arm64_implementation_specific_error(arm_saved_state_t *state, uint32_t esr, vm_offset_t far)
394 {
395 #pragma unused (state, esr, far)
396 	panic_plain("Unhandled implementation specific error\n");
397 }
398 
399 #if CONFIG_KERNEL_INTEGRITY
400 #pragma clang diagnostic push
401 #pragma clang diagnostic ignored "-Wunused-parameter"
402 static void
kernel_integrity_error_handler(uint32_t esr,vm_offset_t far)403 kernel_integrity_error_handler(uint32_t esr, vm_offset_t far)
404 {
405 #if defined(KERNEL_INTEGRITY_WT)
406 #if (DEVELOPMENT || DEBUG)
407 	if (ESR_WT_SERROR(esr)) {
408 		switch (ESR_WT_REASON(esr)) {
409 		case WT_REASON_INTEGRITY_FAIL:
410 			panic_plain("Kernel integrity, violation in frame 0x%016lx.", far);
411 		case WT_REASON_BAD_SYSCALL:
412 			panic_plain("Kernel integrity, bad syscall.");
413 		case WT_REASON_NOT_LOCKED:
414 			panic_plain("Kernel integrity, not locked.");
415 		case WT_REASON_ALREADY_LOCKED:
416 			panic_plain("Kernel integrity, already locked.");
417 		case WT_REASON_SW_REQ:
418 			panic_plain("Kernel integrity, software request.");
419 		case WT_REASON_PT_INVALID:
420 			panic_plain("Kernel integrity, encountered invalid TTE/PTE while "
421 			    "walking 0x%016lx.", far);
422 		case WT_REASON_PT_VIOLATION:
423 			panic_plain("Kernel integrity, violation in mapping 0x%016lx.",
424 			    far);
425 		case WT_REASON_REG_VIOLATION:
426 			panic_plain("Kernel integrity, violation in system register %d.",
427 			    (unsigned) far);
428 		default:
429 			panic_plain("Kernel integrity, unknown (esr=0x%08x).", esr);
430 		}
431 	}
432 #else
433 	if (ESR_WT_SERROR(esr)) {
434 		panic_plain("SError esr: 0x%08x far: 0x%016lx.", esr, far);
435 	}
436 #endif
437 #endif
438 }
439 #pragma clang diagnostic pop
440 #endif
441 
442 static void
arm64_platform_error(arm_saved_state_t * state,uint32_t esr,vm_offset_t far,platform_error_source_t source)443 arm64_platform_error(arm_saved_state_t *state, uint32_t esr, vm_offset_t far, platform_error_source_t source)
444 {
445 #if CONFIG_KERNEL_INTEGRITY
446 	kernel_integrity_error_handler(esr, far);
447 #endif
448 
449 	(void)source;
450 	cpu_data_t *cdp = getCpuDatap();
451 
452 	if (PE_handle_platform_error(far)) {
453 		return;
454 	} else if (cdp->platform_error_handler != NULL) {
455 		cdp->platform_error_handler(cdp->cpu_id, far);
456 	} else {
457 		arm64_implementation_specific_error(state, esr, far);
458 	}
459 }
460 
461 void
panic_with_thread_kernel_state(const char * msg,arm_saved_state_t * ss)462 panic_with_thread_kernel_state(const char *msg, arm_saved_state_t *ss)
463 {
464 	boolean_t ss_valid;
465 
466 	ss_valid = is_saved_state64(ss);
467 	arm_saved_state64_t *state = saved_state64(ss);
468 
469 	os_atomic_cmpxchg(&original_faulting_state, NULL, state, seq_cst);
470 
471 	// rdar://80659177
472 	// Read SoCD tracepoints up to twice — once the first time we call panic and
473 	// another time if we encounter a nested panic after that.
474 	static int twice = 2;
475 	if (twice > 0) {
476 		twice--;
477 		SOCD_TRACE_XNU(KERNEL_STATE_PANIC, ADDR(state->pc),
478 		    PACK_LSB(VALUE(state->lr), VALUE(ss_valid)),
479 		    PACK_2X32(VALUE(state->esr), VALUE(state->cpsr)),
480 		    VALUE(state->far));
481 	}
482 
483 	panic_plain("%s at pc 0x%016llx, lr 0x%016llx (saved state: %p%s)\n"
484 	    "\t  x0:  0x%016llx x1:  0x%016llx  x2:  0x%016llx  x3:  0x%016llx\n"
485 	    "\t  x4:  0x%016llx x5:  0x%016llx  x6:  0x%016llx  x7:  0x%016llx\n"
486 	    "\t  x8:  0x%016llx x9:  0x%016llx  x10: 0x%016llx  x11: 0x%016llx\n"
487 	    "\t  x12: 0x%016llx x13: 0x%016llx  x14: 0x%016llx  x15: 0x%016llx\n"
488 	    "\t  x16: 0x%016llx x17: 0x%016llx  x18: 0x%016llx  x19: 0x%016llx\n"
489 	    "\t  x20: 0x%016llx x21: 0x%016llx  x22: 0x%016llx  x23: 0x%016llx\n"
490 	    "\t  x24: 0x%016llx x25: 0x%016llx  x26: 0x%016llx  x27: 0x%016llx\n"
491 	    "\t  x28: 0x%016llx fp:  0x%016llx  lr:  0x%016llx  sp:  0x%016llx\n"
492 	    "\t  pc:  0x%016llx cpsr: 0x%08x         esr: 0x%08x          far: 0x%016llx\n",
493 	    msg, state->pc, state->lr, ss, (ss_valid ? "" : " INVALID"),
494 	    state->x[0], state->x[1], state->x[2], state->x[3],
495 	    state->x[4], state->x[5], state->x[6], state->x[7],
496 	    state->x[8], state->x[9], state->x[10], state->x[11],
497 	    state->x[12], state->x[13], state->x[14], state->x[15],
498 	    state->x[16], state->x[17], state->x[18], state->x[19],
499 	    state->x[20], state->x[21], state->x[22], state->x[23],
500 	    state->x[24], state->x[25], state->x[26], state->x[27],
501 	    state->x[28], state->fp, state->lr, state->sp,
502 	    state->pc, state->cpsr, state->esr, state->far);
503 }
504 
505 void
sleh_synchronous_sp1(arm_context_t * context,uint32_t esr,vm_offset_t far __unused)506 sleh_synchronous_sp1(arm_context_t *context, uint32_t esr, vm_offset_t far __unused)
507 {
508 	esr_exception_class_t  class = ESR_EC(esr);
509 	arm_saved_state_t    * state = &context->ss;
510 
511 	switch (class) {
512 	case ESR_EC_UNCATEGORIZED:
513 	{
514 		uint32_t instr = *((uint32_t*)get_saved_state_pc(state));
515 		if (IS_ARM_GDB_TRAP(instr)) {
516 			DebuggerCall(EXC_BREAKPOINT, state);
517 		}
518 	}
519 		OS_FALLTHROUGH; // panic if we return from the debugger
520 	default:
521 		panic_with_thread_kernel_state("Synchronous exception taken while SP1 selected", state);
522 	}
523 }
524 
525 
526 __attribute__((noreturn))
527 void
thread_exception_return()528 thread_exception_return()
529 {
530 	thread_t thread = current_thread();
531 	if (thread->machine.exception_trace_code != 0) {
532 		KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE,
533 		    MACHDBG_CODE(DBG_MACH_EXCP_SYNC_ARM, thread->machine.exception_trace_code) | DBG_FUNC_END, 0, 0, 0, 0, 0);
534 		thread->machine.exception_trace_code = 0;
535 	}
536 
537 #if KASAN_TBI
538 	kasan_unpoison_curstack(true);
539 #endif /* KASAN_TBI */
540 	arm64_thread_exception_return();
541 	__builtin_unreachable();
542 }
543 
544 /*
545  * check whether task vtimers are running and set thread and CPU BSD AST
546  *
547  * must be called with interrupts masked so updates of fields are atomic
548  * must be emitted inline to avoid generating an FBT probe on the exception path
549  *
550  */
551 __attribute__((__always_inline__))
552 static inline void
task_vtimer_check(thread_t thread)553 task_vtimer_check(thread_t thread)
554 {
555 	task_t task = get_threadtask_early(thread);
556 
557 	if (__improbable(task != NULL && task->vtimers)) {
558 		thread_ast_set(thread, AST_BSD);
559 		thread->machine.CpuDatap->cpu_pending_ast |= AST_BSD;
560 	}
561 }
562 
563 #if MACH_ASSERT
564 /**
565  * A version of get_preemption_level() that works in early boot.
566  *
567  * If an exception is raised in early boot before the initial thread has been
568  * set up, then calling get_preemption_level() in the SLEH will trigger an
569  * infinitely-recursing exception. This function handles this edge case.
570  */
571 static inline int
sleh_get_preemption_level(void)572 sleh_get_preemption_level(void)
573 {
574 	if (__improbable(current_thread() == NULL)) {
575 		return 0;
576 	}
577 	return get_preemption_level();
578 }
579 #endif // MACH_ASSERT
580 
581 static inline bool
is_platform_error(uint32_t esr)582 is_platform_error(uint32_t esr)
583 {
584 	esr_exception_class_t class = ESR_EC(esr);
585 	uint32_t iss = ESR_ISS(esr);
586 	fault_status_t fault_code;
587 
588 	if (class == ESR_EC_DABORT_EL0 || class == ESR_EC_DABORT_EL1) {
589 		fault_code = ISS_DA_FSC(iss);
590 	} else if (class == ESR_EC_IABORT_EL0 || class == ESR_EC_IABORT_EL1) {
591 		fault_code = ISS_IA_FSC(iss);
592 	} else {
593 		return false;
594 	}
595 
596 	return fault_code == FSC_SYNC_PARITY;
597 }
598 
599 void
sleh_synchronous(arm_context_t * context,uint32_t esr,vm_offset_t far)600 sleh_synchronous(arm_context_t *context, uint32_t esr, vm_offset_t far)
601 {
602 	esr_exception_class_t  class   = ESR_EC(esr);
603 	arm_saved_state_t    * state   = &context->ss;
604 	thread_t               thread  = current_thread();
605 #if MACH_ASSERT
606 	int                    preemption_level = sleh_get_preemption_level();
607 #endif
608 	expected_fault_handler_t expected_fault_handler = NULL;
609 #ifdef CONFIG_XNUPOST
610 	expected_fault_handler_t saved_expected_fault_handler = NULL;
611 	uintptr_t saved_expected_fault_addr = 0;
612 #endif /* CONFIG_XNUPOST */
613 
614 	ASSERT_CONTEXT_SANITY(context);
615 
616 	task_vtimer_check(thread);
617 
618 #if CONFIG_DTRACE
619 	/*
620 	 * Handle kernel DTrace probes as early as possible to minimize the likelihood
621 	 * that this path will itself trigger a DTrace probe, which would lead to infinite
622 	 * probe recursion.
623 	 */
624 	if (__improbable((class == ESR_EC_UNCATEGORIZED) && tempDTraceTrapHook &&
625 	    (tempDTraceTrapHook(EXC_BAD_INSTRUCTION, state, 0, 0) == KERN_SUCCESS))) {
626 		return;
627 	}
628 #endif
629 	bool is_user = PSR64_IS_USER(get_saved_state_cpsr(state));
630 
631 	/*
632 	 * Use KERNEL_DEBUG_CONSTANT_IST here to avoid producing tracepoints
633 	 * that would disclose the behavior of PT_DENY_ATTACH processes.
634 	 */
635 	if (is_user) {
636 		thread->machine.exception_trace_code = (uint16_t)(ARM64_KDBG_CODE_USER | class);
637 		KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE,
638 		    MACHDBG_CODE(DBG_MACH_EXCP_SYNC_ARM, thread->machine.exception_trace_code) | DBG_FUNC_START,
639 		    esr, far, get_saved_state_pc(state), 0, 0);
640 	} else {
641 		KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE,
642 		    MACHDBG_CODE(DBG_MACH_EXCP_SYNC_ARM, ARM64_KDBG_CODE_KERNEL | class) | DBG_FUNC_START,
643 		    esr, VM_KERNEL_ADDRHIDE(far), VM_KERNEL_UNSLIDE(get_saved_state_pc(state)), 0, 0);
644 	}
645 
646 	if (__improbable(ESR_INSTR_IS_2BYTES(esr))) {
647 		/*
648 		 * We no longer support 32-bit, which means no 2-byte
649 		 * instructions.
650 		 */
651 		if (is_user) {
652 			panic("Exception on 2-byte instruction, "
653 			    "context=%p, esr=%#x, far=%p",
654 			    context, esr, (void *)far);
655 		} else {
656 			panic_with_thread_kernel_state("Exception on 2-byte instruction", state);
657 		}
658 	}
659 
660 #ifdef CONFIG_XNUPOST
661 	if (thread->machine.expected_fault_handler != NULL) {
662 		saved_expected_fault_handler = thread->machine.expected_fault_handler;
663 		saved_expected_fault_addr = thread->machine.expected_fault_addr;
664 
665 		thread->machine.expected_fault_handler = NULL;
666 		thread->machine.expected_fault_addr = 0;
667 
668 		if (saved_expected_fault_addr == far) {
669 			expected_fault_handler = saved_expected_fault_handler;
670 		}
671 	}
672 #endif /* CONFIG_XNUPOST */
673 
674 	if (__improbable(is_platform_error(esr))) {
675 		/*
676 		 * Must gather error info in platform error handler before
677 		 * thread is preempted to another core/cluster to guarantee
678 		 * accurate error details
679 		 */
680 
681 		arm64_platform_error(state, esr, far, PLAT_ERR_SRC_SYNC);
682 		return;
683 	}
684 
685 	if (is_user && class == ESR_EC_DABORT_EL0) {
686 		thread_reset_pcs_will_fault(thread);
687 	}
688 
689 	/* Inherit the interrupt masks from previous context */
690 	if (SPSR_INTERRUPTS_ENABLED(get_saved_state_cpsr(state))) {
691 		ml_set_interrupts_enabled(TRUE);
692 	}
693 
694 	switch (class) {
695 	case ESR_EC_SVC_64:
696 		if (!is_saved_state64(state) || !is_user) {
697 			panic("Invalid SVC_64 context");
698 		}
699 
700 		handle_svc(state);
701 		break;
702 
703 	case ESR_EC_DABORT_EL0:
704 		handle_abort(state, esr, far, inspect_data_abort, handle_user_abort, expected_fault_handler);
705 		break;
706 
707 	case ESR_EC_MSR_TRAP:
708 		handle_msr_trap(state, esr);
709 		break;
710 /**
711  * Some APPLEVIRTUALPLATFORM targets do not specify armv8.6, but it's still possible for
712  * them to be hosted by a host that implements ARM_FPAC. There's no way for such a host
713  * to disable it or trap it without substantial performance penalty. Therefore, the FPAC
714  * handler here needs to be built into the guest kernels to prevent the exception to fall
715  * through.
716  */
717 #if __has_feature(ptrauth_calls)
718 	case ESR_EC_PAC_FAIL:
719 		handle_pac_fail(state, esr);
720 		__builtin_unreachable();
721 
722 #endif /* __has_feature(ptrauth_calls) */
723 
724 	case ESR_EC_IABORT_EL0:
725 		handle_abort(state, esr, far, inspect_instruction_abort, handle_user_abort, expected_fault_handler);
726 		break;
727 
728 	case ESR_EC_IABORT_EL1:
729 #ifdef CONFIG_XNUPOST
730 		if ((expected_fault_handler != NULL) && expected_fault_handler(state)) {
731 			break;
732 		}
733 #endif /* CONFIG_XNUPOST */
734 
735 		panic_with_thread_kernel_state("Kernel instruction fetch abort", state);
736 
737 	case ESR_EC_PC_ALIGN:
738 		handle_pc_align(state);
739 		__builtin_unreachable();
740 
741 	case ESR_EC_DABORT_EL1:
742 		handle_abort(state, esr, far, inspect_data_abort, handle_kernel_abort, expected_fault_handler);
743 		break;
744 
745 	case ESR_EC_UNCATEGORIZED:
746 		assert(!ESR_ISS(esr));
747 
748 		handle_uncategorized(&context->ss);
749 		break;
750 
751 	case ESR_EC_SP_ALIGN:
752 		handle_sp_align(state);
753 		__builtin_unreachable();
754 
755 	case ESR_EC_BKPT_AARCH32:
756 		handle_breakpoint(state, esr);
757 		__builtin_unreachable();
758 
759 	case ESR_EC_BRK_AARCH64:
760 		if (PSR64_IS_KERNEL(get_saved_state_cpsr(state))) {
761 			handle_kernel_breakpoint(state, esr);
762 #if CONFIG_UBSAN_MINIMAL
763 			/* UBSan breakpoints are recoverable */
764 			break;
765 #endif /* CONFIG_UBSAN_MINIMAL */
766 		} else {
767 			handle_breakpoint(state, esr);
768 			__builtin_unreachable();
769 		}
770 
771 	case ESR_EC_BKPT_REG_MATCH_EL0:
772 		if (FSC_DEBUG_FAULT == ISS_SSDE_FSC(esr)) {
773 			handle_breakpoint(state, esr);
774 		}
775 		panic("Unsupported Class %u event code. state=%p class=%u esr=%u far=%p",
776 		    class, state, class, esr, (void *)far);
777 		__builtin_unreachable();
778 
779 	case ESR_EC_BKPT_REG_MATCH_EL1:
780 		panic_with_thread_kernel_state("Hardware Breakpoint Debug exception from kernel. Panic (by design)", state);
781 		__builtin_unreachable();
782 
783 	case ESR_EC_SW_STEP_DEBUG_EL0:
784 		if (FSC_DEBUG_FAULT == ISS_SSDE_FSC(esr)) {
785 			handle_sw_step_debug(state);
786 		}
787 		panic("Unsupported Class %u event code. state=%p class=%u esr=%u far=%p",
788 		    class, state, class, esr, (void *)far);
789 		__builtin_unreachable();
790 
791 	case ESR_EC_SW_STEP_DEBUG_EL1:
792 		panic_with_thread_kernel_state("Software Step Debug exception from kernel. Panic (by design)", state);
793 		__builtin_unreachable();
794 
795 	case ESR_EC_WATCHPT_MATCH_EL0:
796 		if (FSC_DEBUG_FAULT == ISS_SSDE_FSC(esr)) {
797 			handle_watchpoint(far);
798 		}
799 		panic("Unsupported Class %u event code. state=%p class=%u esr=%u far=%p",
800 		    class, state, class, esr, (void *)far);
801 		__builtin_unreachable();
802 
803 	case ESR_EC_WATCHPT_MATCH_EL1:
804 		/*
805 		 * If we hit a watchpoint in kernel mode, probably in a copyin/copyout which we don't want to
806 		 * abort.  Turn off watchpoints and keep going; we'll turn them back on in return_from_exception..
807 		 */
808 		if (FSC_DEBUG_FAULT == ISS_SSDE_FSC(esr)) {
809 			arm_debug_set(NULL);
810 			break; /* return to first level handler */
811 		}
812 		panic("Unsupported Class %u event code. state=%p class=%u esr=%u far=%p",
813 		    class, state, class, esr, (void *)far);
814 		__builtin_unreachable();
815 
816 	case ESR_EC_TRAP_SIMD_FP:
817 		handle_simd_trap(state, esr);
818 		__builtin_unreachable();
819 
820 	case ESR_EC_ILLEGAL_INSTR_SET:
821 		if (EXCB_ACTION_RERUN !=
822 		    ex_cb_invoke(EXCB_CLASS_ILLEGAL_INSTR_SET, far)) {
823 			// instruction is not re-executed
824 			panic("Illegal instruction set exception. state=%p class=%u esr=%u far=%p spsr=0x%x",
825 			    state, class, esr, (void *)far, get_saved_state_cpsr(state));
826 		}
827 		// must clear this fault in PSR to re-run
828 		mask_saved_state_cpsr(state, 0, PSR64_IL);
829 		break;
830 
831 	case ESR_EC_MCR_MRC_CP15_TRAP:
832 	case ESR_EC_MCRR_MRRC_CP15_TRAP:
833 	case ESR_EC_MCR_MRC_CP14_TRAP:
834 	case ESR_EC_LDC_STC_CP14_TRAP:
835 	case ESR_EC_MCRR_MRRC_CP14_TRAP:
836 		handle_user_trapped_instruction32(state, esr);
837 		__builtin_unreachable();
838 
839 	case ESR_EC_WFI_WFE:
840 		// Use of WFI or WFE instruction when they have been disabled for EL0
841 		handle_wf_trap(state);
842 		__builtin_unreachable();
843 
844 	case ESR_EC_FLOATING_POINT_64:
845 		handle_fp_trap(state, esr);
846 		__builtin_unreachable();
847 
848 	default:
849 		handle_uncategorized(state);
850 	}
851 
852 #ifdef CONFIG_XNUPOST
853 	if (saved_expected_fault_handler != NULL) {
854 		thread->machine.expected_fault_handler = saved_expected_fault_handler;
855 		thread->machine.expected_fault_addr = saved_expected_fault_addr;
856 	}
857 #endif /* CONFIG_XNUPOST */
858 
859 	if (is_user) {
860 		KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE,
861 		    MACHDBG_CODE(DBG_MACH_EXCP_SYNC_ARM, thread->machine.exception_trace_code) | DBG_FUNC_END,
862 		    esr, far, get_saved_state_pc(state), 0, 0);
863 		thread->machine.exception_trace_code = 0;
864 	} else {
865 		KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE,
866 		    MACHDBG_CODE(DBG_MACH_EXCP_SYNC_ARM, ARM64_KDBG_CODE_KERNEL | class) | DBG_FUNC_END,
867 		    esr, VM_KERNEL_ADDRHIDE(far), VM_KERNEL_UNSLIDE(get_saved_state_pc(state)), 0, 0);
868 	}
869 #if MACH_ASSERT
870 	if (preemption_level != sleh_get_preemption_level()) {
871 		panic("synchronous exception changed preemption level from %d to %d", preemption_level, sleh_get_preemption_level());
872 	}
873 #endif
874 }
875 
876 /*
877  * Uncategorized exceptions are a catch-all for general execution errors.
878  * ARM64_TODO: For now, we assume this is for undefined instruction exceptions.
879  */
880 static void
handle_uncategorized(arm_saved_state_t * state)881 handle_uncategorized(arm_saved_state_t *state)
882 {
883 	exception_type_t           exception = EXC_BAD_INSTRUCTION;
884 	mach_exception_data_type_t codes[2]  = {EXC_ARM_UNDEFINED};
885 	mach_msg_type_number_t     numcodes  = 2;
886 	uint32_t                   instr     = 0;
887 
888 	COPYIN(get_saved_state_pc(state), (char *)&instr, sizeof(instr));
889 
890 #if CONFIG_DTRACE
891 
892 	if (PSR64_IS_USER64(get_saved_state_cpsr(state))) {
893 		/*
894 		 * For a 64bit user process, we care about all 4 bytes of the
895 		 * instr.
896 		 */
897 		if (instr == FASTTRAP_ARM64_INSTR || instr == FASTTRAP_ARM64_RET_INSTR) {
898 			if (dtrace_user_probe(state) == KERN_SUCCESS) {
899 				return;
900 			}
901 		}
902 	} else if (PSR64_IS_USER32(get_saved_state_cpsr(state))) {
903 		/*
904 		 * For a 32bit user process, we check for thumb mode, in
905 		 * which case we only care about a 2 byte instruction length.
906 		 * For non-thumb mode, we care about all 4 bytes of the instructin.
907 		 */
908 		if (get_saved_state_cpsr(state) & PSR64_MODE_USER32_THUMB) {
909 			if (((uint16_t)instr == FASTTRAP_THUMB32_INSTR) ||
910 			    ((uint16_t)instr == FASTTRAP_THUMB32_RET_INSTR)) {
911 				if (dtrace_user_probe(state) == KERN_SUCCESS) {
912 					return;
913 				}
914 			}
915 		} else {
916 			if ((instr == FASTTRAP_ARM32_INSTR) ||
917 			    (instr == FASTTRAP_ARM32_RET_INSTR)) {
918 				if (dtrace_user_probe(state) == KERN_SUCCESS) {
919 					return;
920 				}
921 			}
922 		}
923 	}
924 
925 #endif /* CONFIG_DTRACE */
926 
927 	if (PSR64_IS_KERNEL(get_saved_state_cpsr(state))) {
928 		if (IS_ARM_GDB_TRAP(instr)) {
929 			boolean_t interrupt_state;
930 			exception = EXC_BREAKPOINT;
931 
932 			interrupt_state = ml_set_interrupts_enabled(FALSE);
933 
934 			/* Save off the context here (so that the debug logic
935 			 * can see the original state of this thread).
936 			 */
937 			current_thread()->machine.kpcb = state;
938 
939 			/* Hop into the debugger (typically either due to a
940 			 * fatal exception, an explicit panic, or a stackshot
941 			 * request.
942 			 */
943 			DebuggerCall(exception, state);
944 
945 			current_thread()->machine.kpcb = NULL;
946 			(void) ml_set_interrupts_enabled(interrupt_state);
947 			return;
948 		} else {
949 			panic("Undefined kernel instruction: pc=%p instr=%x", (void*)get_saved_state_pc(state), instr);
950 		}
951 	}
952 
953 	/*
954 	 * Check for GDB breakpoint via illegal opcode.
955 	 */
956 	if (IS_ARM_GDB_TRAP(instr)) {
957 		exception = EXC_BREAKPOINT;
958 		codes[0] = EXC_ARM_BREAKPOINT;
959 		codes[1] = instr;
960 	} else {
961 		codes[1] = instr;
962 	}
963 
964 	exception_triage(exception, codes, numcodes);
965 	__builtin_unreachable();
966 }
967 
968 #if __has_feature(ptrauth_calls)
969 static const uint16_t ptrauth_brk_comment_base = 0xc470;
970 
971 static inline bool
brk_comment_is_ptrauth(uint16_t comment)972 brk_comment_is_ptrauth(uint16_t comment)
973 {
974 	return comment >= ptrauth_brk_comment_base &&
975 	       comment <= ptrauth_brk_comment_base + ptrauth_key_asdb;
976 }
977 
978 static inline const char *
ptrauth_key_to_string(ptrauth_key key)979 ptrauth_key_to_string(ptrauth_key key)
980 {
981 	switch (key) {
982 	case ptrauth_key_asia:
983 		return "IA";
984 	case ptrauth_key_asib:
985 		return "IB";
986 	case ptrauth_key_asda:
987 		return "DA";
988 	case ptrauth_key_asdb:
989 		return "DB";
990 	default:
991 		__builtin_unreachable();
992 	}
993 }
994 #endif /* __has_feature(ptrauth_calls) */
995 
996 #if KASAN_TBI
997 static inline bool
brk_comment_is_kasan_failure(uint16_t comment)998 brk_comment_is_kasan_failure(uint16_t comment)
999 {
1000 	return comment >= KASAN_TBI_ESR_BASE &&
1001 	       comment <= KASAN_TBI_ESR_TOP;
1002 }
1003 #endif /* KASAN_TBI */
1004 
1005 #if CONFIG_UBSAN_MINIMAL
1006 static inline bool
brk_comment_is_ubsan(uint16_t comment)1007 brk_comment_is_ubsan(uint16_t comment)
1008 {
1009 	return comment >= UBSAN_MINIMAL_TRAPS_START &&
1010 	       comment < UBSAN_MINIMAL_TRAPS_END;
1011 }
1012 #endif /* CONFIG_UBSAN_MINIMAL */
1013 
1014 static void
handle_kernel_breakpoint(arm_saved_state_t * state,uint32_t esr)1015 handle_kernel_breakpoint(arm_saved_state_t *state, uint32_t esr)
1016 {
1017 	uint16_t comment = ISS_BRK_COMMENT(esr);
1018 
1019 #if __has_feature(ptrauth_calls)
1020 	if (brk_comment_is_ptrauth(comment)) {
1021 #define MSG_FMT "Break 0x%04X instruction exception from kernel. Ptrauth failure with %s key resulted in 0x%016llx"
1022 		char msg[strlen(MSG_FMT)
1023 		- strlen("0x%04X") + strlen("0xFFFF")
1024 		- strlen("%s") + strlen("IA")
1025 		- strlen("0x%016llx") + strlen("0xFFFFFFFFFFFFFFFF")
1026 		+ 1];
1027 		ptrauth_key key = (ptrauth_key)(comment - ptrauth_brk_comment_base);
1028 		const char *key_str = ptrauth_key_to_string(key);
1029 		snprintf(msg, sizeof(msg), MSG_FMT, comment, key_str, saved_state64(state)->x[16]);
1030 
1031 		panic_with_thread_kernel_state(msg, state);
1032 		__builtin_unreachable();
1033 #undef MSG_FMT
1034 	}
1035 #endif /* __has_feature(ptrauth_calls) */
1036 
1037 #if KASAN_TBI
1038 	if (brk_comment_is_kasan_failure(comment)) {
1039 		kasan_handle_brk_failure(saved_state64(state)->x[0], comment);
1040 		__builtin_unreachable();
1041 	}
1042 #endif /* KASAN_TBI */
1043 
1044 #if CONFIG_UBSAN_MINIMAL
1045 	if (brk_comment_is_ubsan(comment)) {
1046 		ubsan_handle_brk_trap(comment, get_saved_state_pc(state),
1047 		    get_saved_state_fp(state));
1048 		add_saved_state_pc(state, 4);
1049 		return;
1050 	}
1051 #endif /* CONFIG_UBSAN_MINIMAL */
1052 
1053 #define MSG_FMT "Break 0x%04X instruction exception from kernel. Panic (by design)"
1054 	char msg[strlen(MSG_FMT) - strlen("0x%04X") + strlen("0xFFFF") + 1];
1055 	snprintf(msg, sizeof(msg), MSG_FMT, comment);
1056 #undef MSG_FMT
1057 
1058 	panic_with_thread_kernel_state(msg, state);
1059 	__builtin_unreachable();
1060 }
1061 
1062 static void
handle_breakpoint(arm_saved_state_t * state,uint32_t esr __unused)1063 handle_breakpoint(arm_saved_state_t *state, uint32_t esr __unused)
1064 {
1065 	exception_type_t           exception = EXC_BREAKPOINT;
1066 	mach_exception_data_type_t codes[2]  = {EXC_ARM_BREAKPOINT};
1067 	mach_msg_type_number_t     numcodes  = 2;
1068 
1069 #if __has_feature(ptrauth_calls)
1070 	if (ESR_EC(esr) == ESR_EC_BRK_AARCH64 &&
1071 	    brk_comment_is_ptrauth(ISS_BRK_COMMENT(esr))) {
1072 		exception |= EXC_PTRAUTH_BIT;
1073 	}
1074 #endif /* __has_feature(ptrauth_calls) */
1075 
1076 	codes[1] = get_saved_state_pc(state);
1077 	exception_triage(exception, codes, numcodes);
1078 	__builtin_unreachable();
1079 }
1080 
1081 static void
handle_watchpoint(vm_offset_t fault_addr)1082 handle_watchpoint(vm_offset_t fault_addr)
1083 {
1084 	exception_type_t           exception = EXC_BREAKPOINT;
1085 	mach_exception_data_type_t codes[2]  = {EXC_ARM_DA_DEBUG};
1086 	mach_msg_type_number_t     numcodes  = 2;
1087 
1088 	codes[1] = fault_addr;
1089 	exception_triage(exception, codes, numcodes);
1090 	__builtin_unreachable();
1091 }
1092 
1093 static void
handle_abort(arm_saved_state_t * state,uint32_t esr,vm_offset_t fault_addr,abort_inspector_t inspect_abort,abort_handler_t handler,expected_fault_handler_t expected_fault_handler)1094 handle_abort(arm_saved_state_t *state, uint32_t esr, vm_offset_t fault_addr,
1095     abort_inspector_t inspect_abort, abort_handler_t handler, expected_fault_handler_t expected_fault_handler)
1096 {
1097 	fault_status_t fault_code;
1098 	vm_prot_t      fault_type;
1099 
1100 	inspect_abort(ESR_ISS(esr), &fault_code, &fault_type);
1101 	handler(state, esr, fault_addr, fault_code, fault_type, expected_fault_handler);
1102 }
1103 
1104 static void
inspect_instruction_abort(uint32_t iss,fault_status_t * fault_code,vm_prot_t * fault_type)1105 inspect_instruction_abort(uint32_t iss, fault_status_t *fault_code, vm_prot_t *fault_type)
1106 {
1107 	getCpuDatap()->cpu_stat.instr_ex_cnt++;
1108 	*fault_code = ISS_IA_FSC(iss);
1109 	*fault_type = (VM_PROT_READ | VM_PROT_EXECUTE);
1110 }
1111 
1112 static void
inspect_data_abort(uint32_t iss,fault_status_t * fault_code,vm_prot_t * fault_type)1113 inspect_data_abort(uint32_t iss, fault_status_t *fault_code, vm_prot_t *fault_type)
1114 {
1115 	getCpuDatap()->cpu_stat.data_ex_cnt++;
1116 	*fault_code = ISS_DA_FSC(iss);
1117 
1118 	/*
1119 	 * Cache maintenance operations always report faults as write access.
1120 	 * Change these to read access, unless they report a permission fault.
1121 	 * Only certain cache maintenance operations (e.g. 'dc ivac') require write
1122 	 * access to the mapping, but if a cache maintenance operation that only requires
1123 	 * read access generates a permission fault, then we will not be able to handle
1124 	 * the fault regardless of whether we treat it as a read or write fault.
1125 	 */
1126 	if ((iss & ISS_DA_WNR) && (!(iss & ISS_DA_CM) || is_permission_fault(*fault_code))) {
1127 		*fault_type = (VM_PROT_READ | VM_PROT_WRITE);
1128 	} else {
1129 		*fault_type = (VM_PROT_READ);
1130 	}
1131 }
1132 
1133 #if __has_feature(ptrauth_calls)
1134 static inline uint64_t
fault_addr_bitmask(unsigned int bit_from,unsigned int bit_to)1135 fault_addr_bitmask(unsigned int bit_from, unsigned int bit_to)
1136 {
1137 	return ((1ULL << (bit_to - bit_from + 1)) - 1) << bit_from;
1138 }
1139 
1140 static inline bool
fault_addr_bit(vm_offset_t fault_addr,unsigned int bit)1141 fault_addr_bit(vm_offset_t fault_addr, unsigned int bit)
1142 {
1143 	return (bool)((fault_addr >> bit) & 1);
1144 }
1145 
1146 extern int gARM_FEAT_PAuth2;
1147 
1148 /**
1149  * Determines whether a fault address taken at EL0 contains a PAC error code
1150  * corresponding to the specified kind of ptrauth key.
1151  */
1152 static bool
user_fault_addr_matches_pac_error_code(vm_offset_t fault_addr,bool data_key)1153 user_fault_addr_matches_pac_error_code(vm_offset_t fault_addr, bool data_key)
1154 {
1155 	bool instruction_tbi = !(get_tcr() & TCR_TBID0_TBI_DATA_ONLY);
1156 	bool tbi = data_key || __improbable(instruction_tbi);
1157 
1158 	if (gARM_FEAT_PAuth2) {
1159 		/*
1160 		 * EnhancedPAC2 CPUs don't encode error codes at fixed positions, so
1161 		 * treat all non-canonical address bits like potential poison bits.
1162 		 */
1163 		uint64_t mask = fault_addr_bitmask(T0SZ_BOOT, 54);
1164 		if (!tbi) {
1165 			mask |= fault_addr_bitmask(56, 63);
1166 		}
1167 		return (fault_addr & mask) != 0;
1168 	} else {
1169 		unsigned int poison_shift;
1170 		if (tbi) {
1171 			poison_shift = 53;
1172 		} else {
1173 			poison_shift = 61;
1174 		}
1175 
1176 		/* PAC error codes are always in the form key_number:NOT(key_number) */
1177 		bool poison_bit_1 = fault_addr_bit(fault_addr, poison_shift);
1178 		bool poison_bit_2 = fault_addr_bit(fault_addr, poison_shift + 1);
1179 		return poison_bit_1 != poison_bit_2;
1180 	}
1181 }
1182 #endif /* __has_feature(ptrauth_calls) */
1183 
1184 static void
handle_pc_align(arm_saved_state_t * ss)1185 handle_pc_align(arm_saved_state_t *ss)
1186 {
1187 	exception_type_t exc;
1188 	mach_exception_data_type_t codes[2];
1189 	mach_msg_type_number_t numcodes = 2;
1190 
1191 	if (!PSR64_IS_USER(get_saved_state_cpsr(ss))) {
1192 		panic_with_thread_kernel_state("PC alignment exception from kernel.", ss);
1193 	}
1194 
1195 	exc = EXC_BAD_ACCESS;
1196 #if __has_feature(ptrauth_calls)
1197 	if (user_fault_addr_matches_pac_error_code(get_saved_state_pc(ss), false)) {
1198 		exc |= EXC_PTRAUTH_BIT;
1199 	}
1200 #endif /* __has_feature(ptrauth_calls) */
1201 
1202 	codes[0] = EXC_ARM_DA_ALIGN;
1203 	codes[1] = get_saved_state_pc(ss);
1204 
1205 	exception_triage(exc, codes, numcodes);
1206 	__builtin_unreachable();
1207 }
1208 
1209 static void
handle_sp_align(arm_saved_state_t * ss)1210 handle_sp_align(arm_saved_state_t *ss)
1211 {
1212 	exception_type_t exc;
1213 	mach_exception_data_type_t codes[2];
1214 	mach_msg_type_number_t numcodes = 2;
1215 
1216 	if (!PSR64_IS_USER(get_saved_state_cpsr(ss))) {
1217 		panic_with_thread_kernel_state("SP alignment exception from kernel.", ss);
1218 	}
1219 
1220 	exc = EXC_BAD_ACCESS;
1221 #if __has_feature(ptrauth_calls)
1222 	if (user_fault_addr_matches_pac_error_code(get_saved_state_sp(ss), true)) {
1223 		exc |= EXC_PTRAUTH_BIT;
1224 	}
1225 #endif /* __has_feature(ptrauth_calls) */
1226 
1227 	codes[0] = EXC_ARM_SP_ALIGN;
1228 	codes[1] = get_saved_state_sp(ss);
1229 
1230 	exception_triage(exc, codes, numcodes);
1231 	__builtin_unreachable();
1232 }
1233 
1234 static void
handle_wf_trap(arm_saved_state_t * state)1235 handle_wf_trap(arm_saved_state_t *state)
1236 {
1237 	exception_type_t exc;
1238 	mach_exception_data_type_t codes[2];
1239 	mach_msg_type_number_t numcodes = 2;
1240 	uint32_t instr = 0;
1241 
1242 	COPYIN(get_saved_state_pc(state), (char *)&instr, sizeof(instr));
1243 
1244 	exc = EXC_BAD_INSTRUCTION;
1245 	codes[0] = EXC_ARM_UNDEFINED;
1246 	codes[1] = instr;
1247 
1248 	exception_triage(exc, codes, numcodes);
1249 	__builtin_unreachable();
1250 }
1251 
1252 static void
handle_fp_trap(arm_saved_state_t * state,uint32_t esr)1253 handle_fp_trap(arm_saved_state_t *state, uint32_t esr)
1254 {
1255 	exception_type_t exc = EXC_ARITHMETIC;
1256 	mach_exception_data_type_t codes[2];
1257 	mach_msg_type_number_t numcodes = 2;
1258 	uint32_t instr = 0;
1259 
1260 	if (PSR64_IS_KERNEL(get_saved_state_cpsr(state))) {
1261 		panic_with_thread_kernel_state("Floating point exception from kernel", state);
1262 	}
1263 
1264 	COPYIN(get_saved_state_pc(state), (char *)&instr, sizeof(instr));
1265 	codes[1] = instr;
1266 
1267 	/* The floating point trap flags are only valid if TFV is set. */
1268 	if (!fp_exceptions_enabled) {
1269 		exc = EXC_BAD_INSTRUCTION;
1270 		codes[0] = EXC_ARM_UNDEFINED;
1271 	} else if (!(esr & ISS_FP_TFV)) {
1272 		codes[0] = EXC_ARM_FP_UNDEFINED;
1273 	} else if (esr & ISS_FP_UFF) {
1274 		codes[0] = EXC_ARM_FP_UF;
1275 	} else if (esr & ISS_FP_OFF) {
1276 		codes[0] = EXC_ARM_FP_OF;
1277 	} else if (esr & ISS_FP_IOF) {
1278 		codes[0] = EXC_ARM_FP_IO;
1279 	} else if (esr & ISS_FP_DZF) {
1280 		codes[0] = EXC_ARM_FP_DZ;
1281 	} else if (esr & ISS_FP_IDF) {
1282 		codes[0] = EXC_ARM_FP_ID;
1283 	} else if (esr & ISS_FP_IXF) {
1284 		codes[0] = EXC_ARM_FP_IX;
1285 	} else {
1286 		panic("Unrecognized floating point exception, state=%p, esr=%#x", state, esr);
1287 	}
1288 
1289 	exception_triage(exc, codes, numcodes);
1290 	__builtin_unreachable();
1291 }
1292 
1293 
1294 
1295 /*
1296  * handle_alignment_fault_from_user:
1297  *   state: Saved state
1298  *
1299  * Attempts to deal with an alignment fault from userspace (possibly by
1300  * emulating the faulting instruction).  If emulation failed due to an
1301  * unservicable fault, the ESR for that fault will be stored in the
1302  * recovery_esr field of the thread by the exception code.
1303  *
1304  * Returns:
1305  *   -1:     Emulation failed (emulation of state/instr not supported)
1306  *   0:      Successfully emulated the instruction
1307  *   EFAULT: Emulation failed (probably due to permissions)
1308  *   EINVAL: Emulation failed (probably due to a bad address)
1309  */
1310 
1311 
1312 static int
handle_alignment_fault_from_user(arm_saved_state_t * state,kern_return_t * vmfr)1313 handle_alignment_fault_from_user(arm_saved_state_t *state, kern_return_t *vmfr)
1314 {
1315 	int ret = -1;
1316 
1317 #pragma unused (state)
1318 #pragma unused (vmfr)
1319 
1320 	return ret;
1321 }
1322 
1323 
1324 static void
handle_sw_step_debug(arm_saved_state_t * state)1325 handle_sw_step_debug(arm_saved_state_t *state)
1326 {
1327 	thread_t thread = current_thread();
1328 	exception_type_t exc;
1329 	mach_exception_data_type_t codes[2];
1330 	mach_msg_type_number_t numcodes = 2;
1331 
1332 	if (!PSR64_IS_USER(get_saved_state_cpsr(state))) {
1333 		panic_with_thread_kernel_state("SW_STEP_DEBUG exception from kernel.", state);
1334 	}
1335 
1336 	// Disable single step and unmask interrupts (in the saved state, anticipating next exception return)
1337 	if (thread->machine.DebugData != NULL) {
1338 		thread->machine.DebugData->uds.ds64.mdscr_el1 &= ~0x1;
1339 	} else {
1340 		panic_with_thread_kernel_state("SW_STEP_DEBUG exception thread DebugData is NULL.", state);
1341 	}
1342 
1343 	mask_saved_state_cpsr(thread->machine.upcb, 0, PSR64_SS | DAIF_ALL);
1344 
1345 	// Special encoding for gdb single step event on ARM
1346 	exc = EXC_BREAKPOINT;
1347 	codes[0] = 1;
1348 	codes[1] = 0;
1349 
1350 	exception_triage(exc, codes, numcodes);
1351 	__builtin_unreachable();
1352 }
1353 
1354 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,expected_fault_handler_t expected_fault_handler)1355 handle_user_abort(arm_saved_state_t *state, uint32_t esr, vm_offset_t fault_addr,
1356     fault_status_t fault_code, vm_prot_t fault_type, expected_fault_handler_t expected_fault_handler)
1357 {
1358 	exception_type_t           exc      = EXC_BAD_ACCESS;
1359 	mach_exception_data_type_t codes[2];
1360 	mach_msg_type_number_t     numcodes = 2;
1361 	thread_t                   thread   = current_thread();
1362 
1363 	(void)esr;
1364 	(void)expected_fault_handler;
1365 
1366 	if (ml_at_interrupt_context()) {
1367 		panic_with_thread_kernel_state("Apparently on interrupt stack when taking user abort!\n", state);
1368 	}
1369 
1370 	thread->iotier_override = THROTTLE_LEVEL_NONE; /* Reset IO tier override before handling abort from userspace */
1371 
1372 	if (!is_vm_fault(fault_code) &&
1373 	    thread->t_rr_state.trr_fault_state != TRR_FAULT_NONE) {
1374 		thread_reset_pcs_done_faulting(thread);
1375 	}
1376 
1377 	if (is_vm_fault(fault_code)) {
1378 		vm_map_t        map = thread->map;
1379 		vm_offset_t     vm_fault_addr = fault_addr;
1380 		kern_return_t   result = KERN_FAILURE;
1381 
1382 		assert(map != kernel_map);
1383 
1384 		if (!(fault_type & VM_PROT_EXECUTE)) {
1385 			vm_fault_addr = tbi_clear(fault_addr);
1386 		}
1387 
1388 		/* check to see if it is just a pmap ref/modify fault */
1389 		if (!is_translation_fault(fault_code)) {
1390 			result = arm_fast_fault(map->pmap,
1391 			    vm_fault_addr,
1392 			    fault_type, (fault_code == FSC_ACCESS_FLAG_FAULT_L3), TRUE);
1393 		}
1394 		if (result != KERN_SUCCESS) {
1395 
1396 			{
1397 				/* We have to fault the page in */
1398 				result = vm_fault(map, vm_fault_addr, fault_type,
1399 				    /* change_wiring */ FALSE, VM_KERN_MEMORY_NONE, THREAD_ABORTSAFE,
1400 				    /* caller_pmap */ NULL, /* caller_pmap_addr */ 0);
1401 			}
1402 		}
1403 		if (thread->t_rr_state.trr_fault_state != TRR_FAULT_NONE) {
1404 			thread_reset_pcs_done_faulting(thread);
1405 		}
1406 		if (result == KERN_SUCCESS || result == KERN_ABORTED) {
1407 			return;
1408 		}
1409 
1410 		/*
1411 		 * vm_fault() should never return KERN_FAILURE for page faults from user space.
1412 		 * If it does, we're leaking preemption disables somewhere in the kernel.
1413 		 */
1414 		if (__improbable(result == KERN_FAILURE)) {
1415 			panic("vm_fault() KERN_FAILURE from user fault on thread %p", thread);
1416 		}
1417 
1418 		codes[0] = result;
1419 	} else if (is_alignment_fault(fault_code)) {
1420 		kern_return_t vmfkr = KERN_SUCCESS;
1421 		thread->machine.recover_esr = 0;
1422 		thread->machine.recover_far = 0;
1423 		int result = handle_alignment_fault_from_user(state, &vmfkr);
1424 		if (result == 0) {
1425 			/* Successfully emulated, or instruction
1426 			 * copyin() for decode/emulation failed.
1427 			 * Continue, or redrive instruction.
1428 			 */
1429 			thread_exception_return();
1430 		} else if (((result == EFAULT) || (result == EINVAL)) &&
1431 		    (thread->machine.recover_esr == 0)) {
1432 			/*
1433 			 * If we didn't actually take a fault, but got one of
1434 			 * these errors, then we failed basic sanity checks of
1435 			 * the fault address.  Treat this as an invalid
1436 			 * address.
1437 			 */
1438 			codes[0] = KERN_INVALID_ADDRESS;
1439 		} else if ((result == EFAULT) &&
1440 		    (thread->machine.recover_esr)) {
1441 			/*
1442 			 * Since alignment aborts are prioritized
1443 			 * ahead of translation aborts, the misaligned
1444 			 * atomic emulation flow may have triggered a
1445 			 * VM pagefault, which the VM could not resolve.
1446 			 * Report the VM fault error in codes[]
1447 			 */
1448 
1449 			codes[0] = vmfkr;
1450 			assertf(vmfkr != KERN_SUCCESS, "Unexpected vmfkr 0x%x", vmfkr);
1451 			/* Cause ESR_EC to reflect an EL0 abort */
1452 			thread->machine.recover_esr &= ~ESR_EC_MASK;
1453 			thread->machine.recover_esr |= (ESR_EC_DABORT_EL0 << ESR_EC_SHIFT);
1454 			set_saved_state_esr(thread->machine.upcb, thread->machine.recover_esr);
1455 			set_saved_state_far(thread->machine.upcb, thread->machine.recover_far);
1456 			fault_addr = thread->machine.recover_far;
1457 		} else {
1458 			/* This was just an unsupported alignment
1459 			 * exception. Misaligned atomic emulation
1460 			 * timeouts fall in this category.
1461 			 */
1462 			codes[0] = EXC_ARM_DA_ALIGN;
1463 		}
1464 	} else if (is_parity_error(fault_code)) {
1465 #if defined(APPLE_ARM64_ARCH_FAMILY)
1466 		/*
1467 		 * Platform errors are handled in sleh_sync before interrupts are enabled.
1468 		 */
1469 #else
1470 		panic("User parity error.");
1471 #endif
1472 	} else {
1473 		codes[0] = KERN_FAILURE;
1474 	}
1475 
1476 	codes[1] = fault_addr;
1477 #if __has_feature(ptrauth_calls)
1478 	bool is_data_abort = (ESR_EC(esr) == ESR_EC_DABORT_EL0);
1479 	if (user_fault_addr_matches_pac_error_code(fault_addr, is_data_abort)) {
1480 		exc |= EXC_PTRAUTH_BIT;
1481 	}
1482 #endif /* __has_feature(ptrauth_calls) */
1483 	exception_triage(exc, codes, numcodes);
1484 	__builtin_unreachable();
1485 }
1486 
1487 static void
handle_kernel_abort_recover(arm_saved_state_t * state,uint32_t esr,vm_offset_t fault_addr,thread_t thread)1488 handle_kernel_abort_recover(
1489 	arm_saved_state_t              *state,
1490 	uint32_t                        esr,
1491 	vm_offset_t                     fault_addr,
1492 	thread_t                        thread)
1493 {
1494 	thread->machine.recover_esr = esr;
1495 	thread->machine.recover_far = fault_addr;
1496 #if defined(HAS_APPLE_PAC)
1497 	MANIPULATE_SIGNED_THREAD_STATE(state,
1498 	    "mov	x1, %[pc]		\n"
1499 	    "str	x1, [x0, %[SS64_PC]]	\n",
1500 	    [pc] "r"(copyio_recovery_get_recover_addr(state))
1501 	    );
1502 #else
1503 	saved_state64(state)->pc = copyio_recovery_get_recover_addr(state);
1504 #endif
1505 }
1506 
1507 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,expected_fault_handler_t expected_fault_handler)1508 handle_kernel_abort(arm_saved_state_t *state, uint32_t esr, vm_offset_t fault_addr,
1509     fault_status_t fault_code, vm_prot_t fault_type, expected_fault_handler_t expected_fault_handler)
1510 {
1511 	thread_t thread = current_thread();
1512 	bool recover = find_copyio_recovery_entry(state) != 0;
1513 
1514 #ifndef CONFIG_XNUPOST
1515 	(void)expected_fault_handler;
1516 #endif /* CONFIG_XNUPOST */
1517 
1518 #if CONFIG_DTRACE
1519 	if (is_vm_fault(fault_code) && thread->t_dtrace_inprobe) { /* Executing under dtrace_probe? */
1520 		if (dtrace_tally_fault(fault_addr)) { /* Should a fault under dtrace be ignored? */
1521 			/*
1522 			 * Point to next instruction, or recovery handler if set.
1523 			 */
1524 			if (recover) {
1525 				handle_kernel_abort_recover(state, esr, fault_addr, thread);
1526 			} else {
1527 				add_saved_state_pc(state, 4);
1528 			}
1529 			return;
1530 		} else {
1531 			panic_with_thread_kernel_state("Unexpected page fault under dtrace_probe", state);
1532 		}
1533 	}
1534 #endif
1535 
1536 	if (ml_at_interrupt_context()) {
1537 		panic_with_thread_kernel_state("Unexpected abort while on interrupt stack.", state);
1538 	}
1539 
1540 	if (is_vm_fault(fault_code)) {
1541 		kern_return_t result = KERN_FAILURE;
1542 		vm_map_t      map;
1543 		int           interruptible;
1544 
1545 		/*
1546 		 * Ensure no faults in the physical aperture. This could happen if
1547 		 * a page table is incorrectly allocated from the read only region
1548 		 * when running with KTRR.
1549 		 */
1550 
1551 #ifdef CONFIG_XNUPOST
1552 		if (expected_fault_handler && expected_fault_handler(state)) {
1553 			return;
1554 		}
1555 #endif /* CONFIG_XNUPOST */
1556 
1557 		if (fault_addr >= gVirtBase && fault_addr < static_memory_end) {
1558 			panic_with_thread_kernel_state("Unexpected fault in kernel static region\n", state);
1559 		}
1560 
1561 		if (VM_KERNEL_ADDRESS(fault_addr) || thread == THREAD_NULL || recover == 0) {
1562 			/*
1563 			 * If no recovery handler is supplied, always drive the fault against
1564 			 * the kernel map.  If the fault was taken against a userspace VA, indicating
1565 			 * an unprotected access to user address space, vm_fault() should fail and
1566 			 * ultimately lead to a panic here.
1567 			 */
1568 			map = kernel_map;
1569 			interruptible = THREAD_UNINT;
1570 		} else {
1571 			map = thread->map;
1572 
1573 			/**
1574 			 * In the case that the recovery handler is set (e.g., during copyio
1575 			 * and dtrace probes), we don't want the vm_fault() operation to be
1576 			 * aborted early. Those code paths can't handle restarting the
1577 			 * vm_fault() operation so don't allow it to return early without
1578 			 * creating the wanted mapping.
1579 			 */
1580 			interruptible = (recover) ? THREAD_UNINT : THREAD_ABORTSAFE;
1581 		}
1582 
1583 		/* check to see if it is just a pmap ref/modify fault */
1584 		if (!is_translation_fault(fault_code)) {
1585 			result = arm_fast_fault(map->pmap,
1586 			    fault_addr,
1587 			    fault_type, (fault_code == FSC_ACCESS_FLAG_FAULT_L3), FALSE);
1588 			if (result == KERN_SUCCESS) {
1589 				return;
1590 			}
1591 		}
1592 
1593 		if (result != KERN_PROTECTION_FAILURE) {
1594 			/*
1595 			 *  We have to "fault" the page in.
1596 			 */
1597 			result = vm_fault(map, fault_addr, fault_type,
1598 			    /* change_wiring */ FALSE, VM_KERN_MEMORY_NONE, interruptible,
1599 			    /* caller_pmap */ NULL, /* caller_pmap_addr */ 0);
1600 		}
1601 
1602 		if (result == KERN_SUCCESS) {
1603 			return;
1604 		}
1605 
1606 		/*
1607 		 *  If we have a recover handler, invoke it now.
1608 		 */
1609 		if (recover) {
1610 			handle_kernel_abort_recover(state, esr, fault_addr, thread);
1611 			return;
1612 		}
1613 	} else if (is_alignment_fault(fault_code)) {
1614 		if (recover) {
1615 			handle_kernel_abort_recover(state, esr, fault_addr, thread);
1616 			return;
1617 		}
1618 		panic_with_thread_kernel_state("Unaligned kernel data abort.", state);
1619 	} else if (is_parity_error(fault_code)) {
1620 #if defined(APPLE_ARM64_ARCH_FAMILY)
1621 		/*
1622 		 * Platform errors are handled in sleh_sync before interrupts are enabled.
1623 		 */
1624 #else
1625 		panic_with_thread_kernel_state("Kernel parity error.", state);
1626 #endif
1627 	} else {
1628 		kprintf("Unclassified kernel abort (fault_code=0x%x)\n", fault_code);
1629 	}
1630 
1631 	panic_with_thread_kernel_state("Kernel data abort.", state);
1632 }
1633 
1634 extern void syscall_trace(struct arm_saved_state * regs);
1635 
1636 static void
handle_svc(arm_saved_state_t * state)1637 handle_svc(arm_saved_state_t *state)
1638 {
1639 	int      trap_no = get_saved_state_svc_number(state);
1640 	thread_t thread  = current_thread();
1641 	struct   proc *p;
1642 
1643 #define handle_svc_kprintf(x...) /* kprintf("handle_svc: " x) */
1644 
1645 #define TRACE_SYSCALL 1
1646 #if TRACE_SYSCALL
1647 	syscall_trace(state);
1648 #endif
1649 
1650 	thread->iotier_override = THROTTLE_LEVEL_NONE; /* Reset IO tier override before handling SVC from userspace */
1651 
1652 	if (trap_no == (int)PLATFORM_SYSCALL_TRAP_NO) {
1653 		platform_syscall(state);
1654 		panic("Returned from platform_syscall()?");
1655 	}
1656 
1657 	mach_kauth_cred_thread_update();
1658 
1659 	if (trap_no < 0) {
1660 		switch (trap_no) {
1661 		case MACH_ARM_TRAP_ABSTIME:
1662 			handle_mach_absolute_time_trap(state);
1663 			return;
1664 		case MACH_ARM_TRAP_CONTTIME:
1665 			handle_mach_continuous_time_trap(state);
1666 			return;
1667 		}
1668 
1669 		/* Counting perhaps better in the handler, but this is how it's been done */
1670 		thread->syscalls_mach++;
1671 		mach_syscall(state);
1672 	} else {
1673 		/* Counting perhaps better in the handler, but this is how it's been done */
1674 		thread->syscalls_unix++;
1675 		p = get_bsdthreadtask_info(thread);
1676 
1677 		assert(p);
1678 
1679 		unix_syscall(state, thread, p);
1680 	}
1681 }
1682 
1683 static void
handle_mach_absolute_time_trap(arm_saved_state_t * state)1684 handle_mach_absolute_time_trap(arm_saved_state_t *state)
1685 {
1686 	uint64_t now = mach_absolute_time();
1687 	saved_state64(state)->x[0] = now;
1688 }
1689 
1690 static void
handle_mach_continuous_time_trap(arm_saved_state_t * state)1691 handle_mach_continuous_time_trap(arm_saved_state_t *state)
1692 {
1693 	uint64_t now = mach_continuous_time();
1694 	saved_state64(state)->x[0] = now;
1695 }
1696 
1697 
1698 __attribute__((noreturn))
1699 static void
handle_msr_trap(arm_saved_state_t * state,uint32_t esr)1700 handle_msr_trap(arm_saved_state_t *state, uint32_t esr)
1701 {
1702 	exception_type_t           exception = EXC_BAD_INSTRUCTION;
1703 	mach_exception_data_type_t codes[2]  = {EXC_ARM_UNDEFINED};
1704 	mach_msg_type_number_t     numcodes  = 2;
1705 	uint32_t                   instr     = 0;
1706 
1707 	if (!is_saved_state64(state)) {
1708 		panic("MSR/MRS trap (ESR 0x%x) from 32-bit state", esr);
1709 	}
1710 
1711 	if (PSR64_IS_KERNEL(get_saved_state_cpsr(state))) {
1712 		panic("MSR/MRS trap (ESR 0x%x) from kernel", esr);
1713 	}
1714 
1715 	COPYIN(get_saved_state_pc(state), (char *)&instr, sizeof(instr));
1716 	codes[1] = instr;
1717 
1718 	exception_triage(exception, codes, numcodes);
1719 	__builtin_unreachable();
1720 }
1721 
1722 #if __has_feature(ptrauth_calls)
1723 static void
autxx_instruction_extract_reg(uint32_t instr,char reg[4])1724 autxx_instruction_extract_reg(uint32_t instr, char reg[4])
1725 {
1726 	unsigned int rd = ARM64_INSTR_AUTxx_RD_GET(instr);
1727 	switch (rd) {
1728 	case 29:
1729 		strncpy(reg, "fp", 4);
1730 		return;
1731 
1732 	case 30:
1733 		strncpy(reg, "lr", 4);
1734 		return;
1735 
1736 	case 31:
1737 		strncpy(reg, "xzr", 4);
1738 		return;
1739 
1740 	default:
1741 		snprintf(reg, 4, "x%u", rd);
1742 		return;
1743 	}
1744 }
1745 
1746 static const char *
autix_system_instruction_extract_reg(uint32_t instr)1747 autix_system_instruction_extract_reg(uint32_t instr)
1748 {
1749 	unsigned int crm_op2 = ARM64_INSTR_AUTIx_SYSTEM_CRM_OP2_GET(instr);
1750 	if (crm_op2 == ARM64_INSTR_AUTIx_SYSTEM_CRM_OP2_AUTIA1716 ||
1751 	    crm_op2 == ARM64_INSTR_AUTIx_SYSTEM_CRM_OP2_AUTIB1716) {
1752 		return "x17";
1753 	} else {
1754 		return "lr";
1755 	}
1756 }
1757 
1758 static void
handle_pac_fail(arm_saved_state_t * state,uint32_t esr)1759 handle_pac_fail(arm_saved_state_t *state, uint32_t esr)
1760 {
1761 	exception_type_t           exception = EXC_BAD_ACCESS | EXC_PTRAUTH_BIT;
1762 	mach_exception_data_type_t codes[2]  = {EXC_ARM_PAC_FAIL};
1763 	mach_msg_type_number_t     numcodes  = 2;
1764 	uint32_t                   instr     = 0;
1765 
1766 	if (!is_saved_state64(state)) {
1767 		panic("PAC failure (ESR 0x%x) from 32-bit state", esr);
1768 	}
1769 
1770 	COPYIN(get_saved_state_pc(state), (char *)&instr, sizeof(instr));
1771 
1772 	if (PSR64_IS_KERNEL(get_saved_state_cpsr(state))) {
1773 #define GENERIC_PAC_FAILURE_MSG_FMT "PAC failure from kernel with %s key"
1774 #define AUTXX_MSG_FMT GENERIC_PAC_FAILURE_MSG_FMT " while authing %s"
1775 #define GENERIC_MSG_FMT GENERIC_PAC_FAILURE_MSG_FMT
1776 
1777 		char msg[strlen(AUTXX_MSG_FMT)
1778 		- strlen("%s") + strlen("IA")
1779 		- strlen("%s") + strlen("xzr")
1780 		+ 1];
1781 		ptrauth_key key = (ptrauth_key)(esr & 0x3);
1782 		const char *key_str = ptrauth_key_to_string(key);
1783 
1784 		if (ARM64_INSTR_IS_AUTxx(instr)) {
1785 			char reg[4];
1786 			autxx_instruction_extract_reg(instr, reg);
1787 			snprintf(msg, sizeof(msg), AUTXX_MSG_FMT, key_str, reg);
1788 		} else if (ARM64_INSTR_IS_AUTIx_SYSTEM(instr)) {
1789 			const char *reg = autix_system_instruction_extract_reg(instr);
1790 			snprintf(msg, sizeof(msg), AUTXX_MSG_FMT, key_str, reg);
1791 		} else {
1792 			snprintf(msg, sizeof(msg), GENERIC_MSG_FMT, key_str);
1793 		}
1794 		panic_with_thread_kernel_state(msg, state);
1795 	}
1796 
1797 	codes[1] = instr;
1798 
1799 	exception_triage(exception, codes, numcodes);
1800 	__builtin_unreachable();
1801 }
1802 #endif /* __has_feature(ptrauth_calls) */
1803 
1804 static void
handle_user_trapped_instruction32(arm_saved_state_t * state,uint32_t esr)1805 handle_user_trapped_instruction32(arm_saved_state_t *state, uint32_t esr)
1806 {
1807 	exception_type_t           exception = EXC_BAD_INSTRUCTION;
1808 	mach_exception_data_type_t codes[2]  = {EXC_ARM_UNDEFINED};
1809 	mach_msg_type_number_t     numcodes  = 2;
1810 	uint32_t                   instr;
1811 
1812 	if (is_saved_state64(state)) {
1813 		panic("ESR (0x%x) for instruction trapped from U32, but saved state is 64-bit.", esr);
1814 	}
1815 
1816 	if (PSR64_IS_KERNEL(get_saved_state_cpsr(state))) {
1817 		panic("ESR (0x%x) for instruction trapped from U32, actually came from kernel?", esr);
1818 	}
1819 
1820 	COPYIN(get_saved_state_pc(state), (char *)&instr, sizeof(instr));
1821 	codes[1] = instr;
1822 
1823 	exception_triage(exception, codes, numcodes);
1824 	__builtin_unreachable();
1825 }
1826 
1827 static void
handle_simd_trap(arm_saved_state_t * state,uint32_t esr)1828 handle_simd_trap(arm_saved_state_t *state, uint32_t esr)
1829 {
1830 	exception_type_t           exception = EXC_BAD_INSTRUCTION;
1831 	mach_exception_data_type_t codes[2]  = {EXC_ARM_UNDEFINED};
1832 	mach_msg_type_number_t     numcodes  = 2;
1833 	uint32_t                   instr     = 0;
1834 
1835 	if (PSR64_IS_KERNEL(get_saved_state_cpsr(state))) {
1836 		panic("ESR (0x%x) for SIMD trap from userland, actually came from kernel?", esr);
1837 	}
1838 
1839 	COPYIN(get_saved_state_pc(state), (char *)&instr, sizeof(instr));
1840 	codes[1] = instr;
1841 
1842 	exception_triage(exception, codes, numcodes);
1843 	__builtin_unreachable();
1844 }
1845 
1846 void
sleh_irq(arm_saved_state_t * state)1847 sleh_irq(arm_saved_state_t *state)
1848 {
1849 	cpu_data_t * cdp __unused             = getCpuDatap();
1850 #if MACH_ASSERT
1851 	int preemption_level = sleh_get_preemption_level();
1852 #endif
1853 
1854 
1855 	sleh_interrupt_handler_prologue(state, DBG_INTR_TYPE_OTHER);
1856 
1857 #if USE_APPLEARMSMP
1858 	PE_handle_ext_interrupt();
1859 #else
1860 	/* Run the registered interrupt handler. */
1861 	cdp->interrupt_handler(cdp->interrupt_target,
1862 	    cdp->interrupt_refCon,
1863 	    cdp->interrupt_nub,
1864 	    cdp->interrupt_source);
1865 #endif
1866 
1867 	entropy_collect();
1868 
1869 
1870 	sleh_interrupt_handler_epilogue();
1871 #if MACH_ASSERT
1872 	if (preemption_level != sleh_get_preemption_level()) {
1873 		panic("irq handler %p changed preemption level from %d to %d", cdp->interrupt_handler, preemption_level, sleh_get_preemption_level());
1874 	}
1875 #endif
1876 }
1877 
1878 void
sleh_fiq(arm_saved_state_t * state)1879 sleh_fiq(arm_saved_state_t *state)
1880 {
1881 	unsigned int type   = DBG_INTR_TYPE_UNKNOWN;
1882 #if MACH_ASSERT
1883 	int preemption_level = sleh_get_preemption_level();
1884 #endif
1885 
1886 #if MONOTONIC_FIQ
1887 	uint64_t pmcr0 = 0, upmsr = 0;
1888 #endif /* MONOTONIC_FIQ */
1889 
1890 #if defined(HAS_IPI)
1891 	boolean_t    is_ipi = FALSE;
1892 	uint64_t     ipi_sr = 0;
1893 
1894 	if (gFastIPI) {
1895 		MRS(ipi_sr, "S3_5_C15_C1_1");
1896 
1897 		if (ipi_sr & ARM64_IPISR_IPI_PENDING) {
1898 			is_ipi = TRUE;
1899 		}
1900 	}
1901 
1902 	if (is_ipi) {
1903 		type = DBG_INTR_TYPE_IPI;
1904 	} else
1905 #endif /* defined(HAS_IPI) */
1906 	if (ml_get_timer_pending()) {
1907 		type = DBG_INTR_TYPE_TIMER;
1908 	}
1909 #if MONOTONIC_FIQ
1910 	/* Consult the PMI sysregs last, after IPI/timer
1911 	 * classification.
1912 	 */
1913 	else if (mt_pmi_pending(&pmcr0, &upmsr)) {
1914 		type = DBG_INTR_TYPE_PMI;
1915 	}
1916 #endif /* MONOTONIC_FIQ */
1917 
1918 	sleh_interrupt_handler_prologue(state, type);
1919 
1920 #if APPLEVIRTUALPLATFORM
1921 	uint64_t iar = __builtin_arm_rsr64("ICC_IAR0_EL1");
1922 #endif
1923 
1924 #if defined(HAS_IPI)
1925 	if (type == DBG_INTR_TYPE_IPI) {
1926 		/*
1927 		 * Order is important here: we must ack the IPI by writing IPI_SR
1928 		 * before we call cpu_signal_handler().  Otherwise, there will be
1929 		 * a window between the completion of pending-signal processing in
1930 		 * cpu_signal_handler() and the ack during which a newly-issued
1931 		 * IPI to this CPU may be lost.  ISB is required to ensure the msr
1932 		 * is retired before execution of cpu_signal_handler().
1933 		 */
1934 		MSR("S3_5_C15_C1_1", ARM64_IPISR_IPI_PENDING);
1935 		__builtin_arm_isb(ISB_SY);
1936 		cpu_signal_handler();
1937 	} else
1938 #endif /* defined(HAS_IPI) */
1939 #if MONOTONIC_FIQ
1940 	if (type == DBG_INTR_TYPE_PMI) {
1941 		INTERRUPT_MASKED_DEBUG_START(mt_fiq, DBG_INTR_TYPE_PMI);
1942 		mt_fiq(getCpuDatap(), pmcr0, upmsr);
1943 		INTERRUPT_MASKED_DEBUG_END();
1944 	} else
1945 #endif /* MONOTONIC_FIQ */
1946 	{
1947 		/*
1948 		 * We don't know that this is a timer, but we don't have insight into
1949 		 * the other interrupts that go down this path.
1950 		 */
1951 
1952 		cpu_data_t *cdp = getCpuDatap();
1953 
1954 		cdp->cpu_decrementer = -1; /* Large */
1955 
1956 		/*
1957 		 * ARM64_TODO: whether we're coming from userland is ignored right now.
1958 		 * We can easily thread it through, but not bothering for the
1959 		 * moment (AArch32 doesn't either).
1960 		 */
1961 		INTERRUPT_MASKED_DEBUG_START(rtclock_intr, DBG_INTR_TYPE_TIMER);
1962 		rtclock_intr(TRUE);
1963 		INTERRUPT_MASKED_DEBUG_END();
1964 	}
1965 
1966 #if APPLEVIRTUALPLATFORM
1967 	if (iar != GIC_SPURIOUS_IRQ) {
1968 		__builtin_arm_wsr64("ICC_EOIR0_EL1", iar);
1969 		__builtin_arm_isb(ISB_SY);
1970 	}
1971 #endif
1972 
1973 	sleh_interrupt_handler_epilogue();
1974 #if MACH_ASSERT
1975 	if (preemption_level != sleh_get_preemption_level()) {
1976 		panic("fiq type %u changed preemption level from %d to %d", type, preemption_level, sleh_get_preemption_level());
1977 	}
1978 #endif
1979 }
1980 
1981 void
sleh_serror(arm_context_t * context,uint32_t esr,vm_offset_t far)1982 sleh_serror(arm_context_t *context, uint32_t esr, vm_offset_t far)
1983 {
1984 	task_vtimer_check(current_thread());
1985 
1986 	KDBG_RELEASE(MACHDBG_CODE(DBG_MACH_EXCP_SERR_ARM, 0) | DBG_FUNC_START,
1987 	    esr, VM_KERNEL_ADDRHIDE(far));
1988 	arm_saved_state_t *state = &context->ss;
1989 #if MACH_ASSERT
1990 	int preemption_level = sleh_get_preemption_level();
1991 #endif
1992 
1993 
1994 	ASSERT_CONTEXT_SANITY(context);
1995 	arm64_platform_error(state, esr, far, PLAT_ERR_SRC_ASYNC);
1996 #if MACH_ASSERT
1997 	if (preemption_level != sleh_get_preemption_level()) {
1998 		panic("serror changed preemption level from %d to %d", preemption_level, sleh_get_preemption_level());
1999 	}
2000 #endif
2001 	KDBG_RELEASE(MACHDBG_CODE(DBG_MACH_EXCP_SERR_ARM, 0) | DBG_FUNC_END,
2002 	    esr, VM_KERNEL_ADDRHIDE(far));
2003 }
2004 
2005 void
mach_syscall_trace_exit(unsigned int retval,unsigned int call_number)2006 mach_syscall_trace_exit(unsigned int retval,
2007     unsigned int call_number)
2008 {
2009 	KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE,
2010 	    MACHDBG_CODE(DBG_MACH_EXCP_SC, (call_number)) |
2011 	    DBG_FUNC_END, retval, 0, 0, 0, 0);
2012 }
2013 
2014 __attribute__((noreturn))
2015 void
thread_syscall_return(kern_return_t error)2016 thread_syscall_return(kern_return_t error)
2017 {
2018 	thread_t thread;
2019 	struct arm_saved_state *state;
2020 
2021 	thread = current_thread();
2022 	state = get_user_regs(thread);
2023 
2024 	assert(is_saved_state64(state));
2025 	saved_state64(state)->x[0] = error;
2026 
2027 #if MACH_ASSERT
2028 	kern_allocation_name_t
2029 	prior __assert_only = thread_get_kernel_state(thread)->allocation_name;
2030 	assertf(prior == NULL, "thread_set_allocation_name(\"%s\") not cleared", kern_allocation_get_name(prior));
2031 #endif /* MACH_ASSERT */
2032 
2033 	if (kdebug_enable) {
2034 		/* Invert syscall number (negative for a mach syscall) */
2035 		mach_syscall_trace_exit(error, (-1) * get_saved_state_svc_number(state));
2036 	}
2037 
2038 	thread_exception_return();
2039 }
2040 
2041 void
syscall_trace(struct arm_saved_state * regs __unused)2042 syscall_trace(
2043 	struct arm_saved_state * regs __unused)
2044 {
2045 	/* kprintf("syscall: %d\n", saved_state64(regs)->x[16]);  */
2046 }
2047 
2048 static void
sleh_interrupt_handler_prologue(arm_saved_state_t * state,unsigned int type)2049 sleh_interrupt_handler_prologue(arm_saved_state_t *state, unsigned int type)
2050 {
2051 	boolean_t is_user = PSR64_IS_USER(get_saved_state_cpsr(state));
2052 
2053 	task_vtimer_check(current_thread());
2054 
2055 	uint64_t pc = is_user ? get_saved_state_pc(state) :
2056 	    VM_KERNEL_UNSLIDE(get_saved_state_pc(state));
2057 
2058 	KDBG_RELEASE(MACHDBG_CODE(DBG_MACH_EXCP_INTR, 0) | DBG_FUNC_START,
2059 	    0, pc, is_user, type);
2060 
2061 #if CONFIG_TELEMETRY
2062 	if (telemetry_needs_record) {
2063 		telemetry_mark_curthread(is_user, FALSE);
2064 	}
2065 #endif /* CONFIG_TELEMETRY */
2066 }
2067 
2068 static void
sleh_interrupt_handler_epilogue(void)2069 sleh_interrupt_handler_epilogue(void)
2070 {
2071 #if KPERF
2072 	kperf_interrupt();
2073 #endif /* KPERF */
2074 	KDBG_RELEASE(MACHDBG_CODE(DBG_MACH_EXCP_INTR, 0) | DBG_FUNC_END);
2075 }
2076 
2077 void
sleh_invalid_stack(arm_context_t * context,uint32_t esr __unused,vm_offset_t far __unused)2078 sleh_invalid_stack(arm_context_t *context, uint32_t esr __unused, vm_offset_t far __unused)
2079 {
2080 	thread_t thread = current_thread();
2081 	vm_offset_t kernel_stack_bottom, sp;
2082 
2083 	sp = get_saved_state_sp(&context->ss);
2084 	kernel_stack_bottom = round_page(thread->machine.kstackptr) - KERNEL_STACK_SIZE;
2085 
2086 	if ((sp < kernel_stack_bottom) && (sp >= (kernel_stack_bottom - PAGE_SIZE))) {
2087 		panic_with_thread_kernel_state("Invalid kernel stack pointer (probable overflow).", &context->ss);
2088 	}
2089 
2090 	panic_with_thread_kernel_state("Invalid kernel stack pointer (probable corruption).", &context->ss);
2091 }
2092 
2093