1 /*
2 * Copyright (c) 2007-2020 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 <debug.h>
30 #include <mach_kdp.h>
31 #include <kern/kern_stackshot.h>
32
33 #include <kern/thread.h>
34 #include <machine/pmap.h>
35 #include <device/device_types.h>
36
37 #include <mach/vm_param.h>
38 #include <mach/clock_types.h>
39 #include <mach/machine.h>
40 #include <mach/kmod.h>
41 #include <pexpert/boot.h>
42 #include <pexpert/pexpert.h>
43
44 #include <ptrauth.h>
45
46 #include <kern/misc_protos.h>
47 #include <kern/startup.h>
48 #include <kern/clock.h>
49 #include <kern/debug.h>
50 #include <kern/processor.h>
51 #include <kdp/kdp_core.h>
52 #if ALTERNATE_DEBUGGER
53 #include <arm64/alternate_debugger.h>
54 #endif
55 #include <machine/atomic.h>
56 #include <machine/trap.h>
57 #include <kern/spl.h>
58 #include <pexpert/pexpert.h>
59 #include <kdp/kdp_callout.h>
60 #include <kdp/kdp_dyld.h>
61 #include <kdp/kdp_internal.h>
62 #include <kdp/kdp_common.h>
63 #include <uuid/uuid.h>
64 #include <sys/codesign.h>
65 #include <sys/time.h>
66
67 #if CONFIG_SPTM
68 #include <kern/percpu.h>
69 #include <arm64/sptm/pmap/pmap_data.h>
70 #endif
71
72 #include <IOKit/IOPlatformExpert.h>
73 #include <IOKit/IOKitServer.h>
74
75 #include <mach/vm_prot.h>
76 #include <vm/vm_map_xnu.h>
77 #include <vm/pmap.h>
78 #include <vm/vm_shared_region.h>
79 #include <mach/time_value.h>
80 #include <machine/machparam.h> /* for btop */
81
82 #include <console/video_console.h>
83 #include <console/serial_protos.h>
84 #include <arm/cpu_data.h>
85 #include <arm/cpu_data_internal.h>
86 #include <arm/cpu_internal.h>
87 #include <arm/misc_protos.h>
88 #include <libkern/OSKextLibPrivate.h>
89 #include <vm/vm_kern.h>
90 #include <kern/kern_cdata.h>
91 #include <kern/ledger.h>
92
93
94 #if DEVELOPMENT || DEBUG
95 #include <kern/ext_paniclog.h>
96 #endif
97
98 #if CONFIG_EXCLAVES
99 #include <kern/exclaves_panic.h>
100 #include <kern/exclaves_inspection.h>
101 #endif
102
103
104 #if MACH_KDP
105 void kdp_trap(unsigned int, struct arm_saved_state *);
106 #endif
107
108 /*
109 * Increment the PANICLOG_VERSION if you change the format of the panic
110 * log in any way.
111 */
112 #define PANICLOG_VERSION 15
113 static struct kcdata_descriptor kc_panic_data;
114
115 extern char iBoot_version[];
116 #if defined(TARGET_OS_OSX) && defined(__arm64__)
117 extern char iBoot_Stage_2_version[];
118 #endif /* defined(TARGET_OS_OSX) && defined(__arm64__) */
119
120 extern volatile uint32_t debug_enabled;
121 extern unsigned int not_in_kdp;
122
123 extern int copyinframe(vm_address_t fp, uint32_t * frame);
124 extern void kdp_callouts(kdp_event_t event);
125
126 #define MAX_PROCNAME_LEN 32
127 /* #include <sys/proc.h> */
128 struct proc;
129 extern int proc_pid(struct proc *p);
130 extern void proc_name_kdp(struct proc *, char *, int);
131
132 /*
133 * Make sure there's enough space to include the relevant bits in the format required
134 * within the space allocated for the panic version string in the panic header.
135 * The format required by OSAnalytics/DumpPanic is 'Product Version (OS Version)'.
136 */
137 #define PANIC_HEADER_VERSION_FMT_STR "%.14s (%.14s)"
138
139 extern const char version[];
140 extern char osversion[];
141 extern char osproductversion[];
142 extern char osreleasetype[];
143
144 #if defined(XNU_TARGET_OS_BRIDGE)
145 extern char macosproductversion[];
146 extern char macosversion[];
147 #endif
148
149 extern uint8_t gPlatformECID[8];
150 extern uint32_t gPlatformMemoryID;
151
152 extern uint64_t last_hwaccess_thread;
153 extern uint8_t last_hwaccess_type; /* 0 : read, 1 : write. */
154 extern uint8_t last_hwaccess_size;
155 extern uint64_t last_hwaccess_paddr;
156
157 /*Choosing the size for gTargetTypeBuffer as 16 and size for gModelTypeBuffer as 32
158 * since the target name and model name typically doesn't exceed this size */
159 extern char gTargetTypeBuffer[16];
160 extern char gModelTypeBuffer[32];
161
162 extern struct timeval gIOLastSleepTime;
163 extern struct timeval gIOLastWakeTime;
164 extern boolean_t is_clock_configured;
165 extern boolean_t kernelcache_uuid_valid;
166 extern uuid_t kernelcache_uuid;
167 extern uuid_string_t bootsessionuuid_string;
168
169 extern uint64_t roots_installed;
170
171 /* Definitions for frame pointers */
172 #define FP_ALIGNMENT_MASK ((uint32_t)(0x3))
173 #define FP_LR_OFFSET ((uint32_t)4)
174 #define FP_LR_OFFSET64 ((uint32_t)8)
175 #define FP_MAX_NUM_TO_EVALUATE (50)
176
177 /* Timeout for all processors responding to debug crosscall */
178 MACHINE_TIMEOUT_ALWAYS_ENABLED(debug_ack_timeout, "debug-ack", 240000, MACHINE_TIMEOUT_UNIT_TIMEBASE);
179
180 /* Forward functions definitions */
181 void panic_display_times(void);
182 void panic_print_symbol_name(vm_address_t search);
183
184
185 /* Global variables */
186 static uint32_t panic_bt_depth;
187 boolean_t PanicInfoSaved = FALSE;
188 boolean_t force_immediate_debug_halt = FALSE;
189 unsigned int debug_ack_timeout_count = 0;
190 _Atomic unsigned int debugger_sync = 0;
191 _Atomic unsigned int mp_kdp_trap = 0; /* CPUs signalled by the debug CPU will spin on this */
192 _Atomic unsigned int debug_cpus_spinning = 0; /* Number of signalled CPUs still spinning on mp_kdp_trap (in DebuggerXCall). */
193 unsigned int DebugContextCount = 0;
194 bool trap_is_stackshot = false; /* Whether the trap is for a stackshot */
195
196 #if defined(__arm64__)
197 uint8_t PE_smc_stashed_x86_system_state = 0xFF;
198 uint8_t PE_smc_stashed_x86_power_state = 0xFF;
199 uint8_t PE_smc_stashed_x86_efi_boot_state = 0xFF;
200 uint8_t PE_smc_stashed_x86_shutdown_cause = 0xFF;
201 uint64_t PE_smc_stashed_x86_prev_power_transitions = UINT64_MAX;
202 uint32_t PE_pcie_stashed_link_state = UINT32_MAX;
203 uint64_t PE_nvram_stashed_x86_macos_slide = UINT64_MAX;
204 #endif
205
206
207 static void
do_print_backtrace_internal(pmap_t pmap,vm_offset_t topfp,const char * cur_marker,boolean_t is_64_bit,boolean_t print_kexts_in_backtrace)208 do_print_backtrace_internal(pmap_t pmap, vm_offset_t topfp, const char *cur_marker,
209 boolean_t is_64_bit, boolean_t print_kexts_in_backtrace)
210 {
211 unsigned int i = 0;
212 addr64_t lr = 0;
213 addr64_t fp = topfp;
214 addr64_t fp_for_ppn = 0;
215 ppnum_t ppn = (ppnum_t)NULL;
216 vm_offset_t raddrs[FP_MAX_NUM_TO_EVALUATE] = { 0 };
217 bool dump_kernel_stack = (fp >= VM_MIN_KERNEL_ADDRESS);
218
219 do {
220 if ((fp == 0) || ((fp & FP_ALIGNMENT_MASK) != 0)) {
221 break;
222 }
223
224 if ((!dump_kernel_stack) && (fp >= VM_MIN_KERNEL_ADDRESS)) {
225 break;
226 }
227
228 /*
229 * Check to see if current address will result in a different
230 * ppn than previously computed (to avoid recomputation) via
231 * (addr) ^ fp_for_ppn) >> PAGE_SHIFT)
232 */
233 if ((((fp + FP_LR_OFFSET) ^ fp_for_ppn) >> PAGE_SHIFT) != 0x0U) {
234 ppn = pmap_find_phys(pmap, fp + FP_LR_OFFSET);
235 fp_for_ppn = fp + (is_64_bit ? FP_LR_OFFSET64 : FP_LR_OFFSET);
236 }
237 if (ppn != (ppnum_t)NULL) {
238 if (is_64_bit) {
239 lr = ml_phys_read_double_64(((((vm_offset_t)ppn) << PAGE_SHIFT)) | ((fp + FP_LR_OFFSET64) & PAGE_MASK));
240 #if defined(HAS_APPLE_PAC)
241 /* return addresses on stack will be signed by arm64e ABI */
242 lr = (addr64_t) ptrauth_strip((void *)lr, ptrauth_key_return_address);
243 #endif
244 } else {
245 lr = ml_phys_read_word(((((vm_offset_t)ppn) << PAGE_SHIFT)) | ((fp + FP_LR_OFFSET) & PAGE_MASK));
246 }
247 } else {
248 if (is_64_bit) {
249 paniclog_append_noflush("%s\t Could not read LR from frame at 0x%016llx\n", cur_marker, fp + FP_LR_OFFSET64);
250 } else {
251 paniclog_append_noflush("%s\t Could not read LR from frame at 0x%08x\n", cur_marker, (uint32_t)(fp + FP_LR_OFFSET));
252 }
253 break;
254 }
255 if (((fp ^ fp_for_ppn) >> PAGE_SHIFT) != 0x0U) {
256 ppn = pmap_find_phys(pmap, fp);
257 fp_for_ppn = fp;
258 }
259 if (ppn != (ppnum_t)NULL) {
260 if (is_64_bit) {
261 fp = ml_phys_read_double_64(((((vm_offset_t)ppn) << PAGE_SHIFT)) | (fp & PAGE_MASK));
262 #if defined(HAS_APPLE_PAC)
263 /* frame pointers on stack will be signed by arm64e ABI */
264 fp = (addr64_t) ptrauth_strip((void *)fp, ptrauth_key_frame_pointer);
265 #endif
266 } else {
267 fp = ml_phys_read_word(((((vm_offset_t)ppn) << PAGE_SHIFT)) | (fp & PAGE_MASK));
268 }
269 } else {
270 if (is_64_bit) {
271 paniclog_append_noflush("%s\t Could not read FP from frame at 0x%016llx\n", cur_marker, fp);
272 } else {
273 paniclog_append_noflush("%s\t Could not read FP from frame at 0x%08x\n", cur_marker, (uint32_t)fp);
274 }
275 break;
276 }
277 /*
278 * Counter 'i' may == FP_MAX_NUM_TO_EVALUATE when running one
279 * extra round to check whether we have all frames in order to
280 * indicate (in)complete backtrace below. This happens in a case
281 * where total frame count and FP_MAX_NUM_TO_EVALUATE are equal.
282 * Do not capture anything.
283 */
284 if (i < FP_MAX_NUM_TO_EVALUATE && lr) {
285 if (is_64_bit) {
286 paniclog_append_noflush("%s\t lr: 0x%016llx fp: 0x%016llx\n", cur_marker, lr, fp);
287 } else {
288 paniclog_append_noflush("%s\t lr: 0x%08x fp: 0x%08x\n", cur_marker, (uint32_t)lr, (uint32_t)fp);
289 }
290 raddrs[i] = lr;
291 }
292 } while ((++i <= FP_MAX_NUM_TO_EVALUATE) && (fp != topfp));
293
294 if (i > FP_MAX_NUM_TO_EVALUATE && fp != 0) {
295 paniclog_append_noflush("Backtrace continues...\n");
296 }
297
298 if (print_kexts_in_backtrace && i > 0) {
299 kmod_panic_dump(&raddrs[0], i);
300 }
301 }
302
303 #define SANE_TASK_LIMIT 256
304 #define TOP_RUNNABLE_LIMIT 5
305 #define PANICLOG_UUID_BUF_SIZE 256
306
307 extern void panic_print_vnodes(void);
308
309 static void
panic_display_tpidrs(void)310 panic_display_tpidrs(void)
311 {
312 #if defined(__arm64__)
313 paniclog_append_noflush("TPIDRx_ELy = {1: 0x%016llx 0: 0x%016llx 0ro: 0x%016llx }\n",
314 __builtin_arm_rsr64("TPIDR_EL1"), __builtin_arm_rsr64("TPIDR_EL0"),
315 __builtin_arm_rsr64("TPIDRRO_EL0"));
316 #endif //defined(__arm64__)
317 }
318
319
320
321 static void
panic_display_hung_cpus_help(void)322 panic_display_hung_cpus_help(void)
323 {
324 #if defined(__arm64__)
325 const uint32_t pcsr_offset = 0x90;
326
327 /*
328 * Print some info that might help in cases where nothing
329 * else does
330 */
331 const ml_topology_info_t *info = ml_get_topology_info();
332 if (info) {
333 unsigned i, retry;
334
335 for (i = 0; i < info->num_cpus; i++) {
336 ml_topology_cpu_t *cpu = &info->cpus[i];
337 char cluster_name[16], cluster_letter;
338
339 switch (cpu->cluster_type) {
340 case CLUSTER_TYPE_E:
341 cluster_letter = 'E';
342 break;
343 case CLUSTER_TYPE_P:
344 cluster_letter = 'P';
345 break;
346 default:
347 cluster_letter = '?';
348 }
349 snprintf(cluster_name, sizeof(cluster_name), "%cACC%d", cluster_letter, cpu->cluster_id);
350
351 if (!PE_cpu_power_check_kdp(i)) {
352 paniclog_append_noflush("CORE %u [%s] is offline, skipping\n", i, cluster_name);
353 continue;
354 }
355 if (cpu->cpu_UTTDBG_regs) {
356 volatile uint64_t *pcsr = (volatile uint64_t*)(cpu->cpu_UTTDBG_regs + pcsr_offset);
357 volatile uint32_t *pcsrTrigger = (volatile uint32_t*)pcsr;
358 uint64_t pc = 0;
359
360 // a number of retries are needed till this works
361 for (retry = 1024; retry && !pc; retry--) {
362 //a 32-bit read is required to make a PC sample be produced, else we'll only get a zero
363 (void)*pcsrTrigger;
364 pc = *pcsr;
365 }
366
367 //postprocessing (same as astris does)
368 if (pc >> 48) {
369 pc |= 0xffff000000000000ull;
370 }
371 paniclog_append_noflush("CORE %u [%s] recently retired instr at 0x%016llx\n", i, cluster_name, pc);
372 }
373 }
374 }
375 #endif //defined(__arm64__)
376 }
377
378
379
380 static void
panic_display_pvhs_locked(void)381 panic_display_pvhs_locked(void)
382 {
383 }
384
385 static void
panic_display_pvh_to_lock(void)386 panic_display_pvh_to_lock(void)
387 {
388 }
389
390 static void
panic_display_last_pc_lr(void)391 panic_display_last_pc_lr(void)
392 {
393 #if defined(__arm64__)
394 const int max_cpu = ml_get_max_cpu_number();
395
396 for (int cpu = 0; cpu <= max_cpu; cpu++) {
397 cpu_data_t *current_cpu_datap = cpu_datap(cpu);
398
399 if (current_cpu_datap == NULL) {
400 continue;
401 }
402
403 if (current_cpu_datap == getCpuDatap()) {
404 /**
405 * Skip printing the PC/LR if this is the CPU
406 * that initiated the panic.
407 */
408 paniclog_append_noflush("CORE %u is the one that panicked. Check the full backtrace for details.\n", cpu);
409 continue;
410 }
411
412 paniclog_append_noflush("CORE %u: PC=0x%016llx, LR=0x%016llx, FP=0x%016llx\n", cpu,
413 current_cpu_datap->ipi_pc, (uint64_t)VM_KERNEL_STRIP_PTR(current_cpu_datap->ipi_lr),
414 (uint64_t)VM_KERNEL_STRIP_PTR(current_cpu_datap->ipi_fp));
415 }
416 #endif
417 }
418
419 #if CONFIG_EXCLAVES
420 static void
panic_report_exclaves_stackshot(void)421 panic_report_exclaves_stackshot(void)
422 {
423 if (exclaves_panic_ss_status == EXCLAVES_PANIC_STACKSHOT_FOUND) {
424 paniclog_append_noflush("** Exclaves panic stackshot found\n");
425 } else if (exclaves_panic_ss_status == EXCLAVES_PANIC_STACKSHOT_NOT_FOUND) {
426 paniclog_append_noflush("** Exclaves panic stackshot not found\n");
427 } else if (exclaves_panic_ss_status == EXCLAVES_PANIC_STACKSHOT_DECODE_FAILED) {
428 paniclog_append_noflush("!! Exclaves panic stackshot decode failed !!\n");
429 }
430 }
431 #endif /* CONFIG_EXCLAVES */
432
433 __attribute__((always_inline))
434 static inline void
print_backtrace_internal(thread_t thread,bool filesetKC)435 print_backtrace_internal(thread_t thread, bool filesetKC)
436 {
437 uintptr_t cur_fp = (uintptr_t)__builtin_frame_address(0);
438 const char *nohilite_thread_marker = "\t";
439
440 #if defined(HAS_APPLE_PAC)
441 cur_fp = (addr64_t)ptrauth_strip((void *)cur_fp, ptrauth_key_frame_pointer);
442 #endif
443
444 if (cur_fp < VM_MAX_KERNEL_ADDRESS) {
445 paniclog_append_noflush("Panicked thread: %p, backtrace: 0x%llx, tid: %llu\n",
446 thread, (addr64_t)cur_fp, thread_tid(thread));
447 #if __LP64__
448 do_print_backtrace_internal(kernel_pmap, cur_fp, nohilite_thread_marker, TRUE, filesetKC);
449 #else
450 do_print_backtrace_internal(kernel_pmap, cur_fp, nohilite_thread_marker, FALSE, filesetKC);
451 #endif
452 } else {
453 paniclog_append_noflush("Could not print panicked thread backtrace:"
454 "frame pointer outside kernel vm.\n");
455 }
456 }
457
458 static bool
is_filesetKC(void)459 is_filesetKC(void)
460 {
461 kc_format_t kc_format;
462 bool filesetKC = false;
463
464 __unused bool result = PE_get_primary_kc_format(&kc_format);
465 assert(result == true);
466 filesetKC = kc_format == KCFormatFileset;
467 return filesetKC;
468 }
469
470
471 static void
do_print_all_panic_info(const char * message,uint64_t panic_options,const char * panic_initiator)472 do_print_all_panic_info(const char *message, uint64_t panic_options, const char *panic_initiator)
473 {
474 int logversion = PANICLOG_VERSION;
475 thread_t cur_thread = current_thread();
476 task_t task;
477 struct proc *proc;
478 int print_vnodes = 0;
479
480 /* end_marker_bytes set to 200 for printing END marker + stackshot summary info always */
481 int bytes_traced = 0, bytes_remaining = 0, end_marker_bytes = 200;
482 int bytes_uncompressed = 0;
483 uint64_t bytes_used = 0ULL;
484 int err = 0;
485 char *stackshot_begin_loc = NULL;
486 bool filesetKC = is_filesetKC();
487 uint32_t panic_initiator_len = 0;
488 #if CONFIG_EXT_PANICLOG
489 uint32_t ext_paniclog_bytes = 0;
490 #endif
491
492 if (panic_bt_depth != 0) {
493 return;
494 }
495 panic_bt_depth++;
496
497 /* Truncate panic string to 1200 bytes */
498 paniclog_append_noflush("Debugger message: %.1200s\n", message);
499 if (debug_enabled) {
500 paniclog_append_noflush("Device: %s\n",
501 ('\0' != gTargetTypeBuffer[0]) ? gTargetTypeBuffer : "Not set yet");
502 paniclog_append_noflush("Hardware Model: %s\n",
503 ('\0' != gModelTypeBuffer[0]) ? gModelTypeBuffer:"Not set yet");
504 paniclog_append_noflush("ECID: %02X%02X%02X%02X%02X%02X%02X%02X\n", gPlatformECID[7],
505 gPlatformECID[6], gPlatformECID[5], gPlatformECID[4], gPlatformECID[3],
506 gPlatformECID[2], gPlatformECID[1], gPlatformECID[0]);
507 if (last_hwaccess_thread) {
508 paniclog_append_noflush("AppleHWAccess Thread: 0x%llx\n", last_hwaccess_thread);
509 if (!last_hwaccess_size) {
510 paniclog_append_noflush("AppleHWAccess last access: no access data, this is unexpected.\n");
511 } else {
512 const char *typ = last_hwaccess_type ? "write" : "read";
513 paniclog_append_noflush("AppleHWAccess last access: %s of size %u at address 0x%llx\n", typ, last_hwaccess_size, last_hwaccess_paddr);
514 }
515 }
516 paniclog_append_noflush("Boot args: %s\n", PE_boot_args());
517 }
518 paniclog_append_noflush("Memory ID: 0x%x\n", gPlatformMemoryID);
519 paniclog_append_noflush("OS release type: %.256s\n",
520 ('\0' != osreleasetype[0]) ? osreleasetype : "Not set yet");
521 paniclog_append_noflush("OS version: %.256s\n",
522 ('\0' != osversion[0]) ? osversion : "Not set yet");
523 #if defined(XNU_TARGET_OS_BRIDGE)
524 paniclog_append_noflush("macOS version: %.256s\n",
525 ('\0' != macosversion[0]) ? macosversion : "Not set");
526 #endif
527 paniclog_append_noflush("Kernel version: %.512s\n", version);
528
529 #if CONFIG_EXCLAVES
530 exclaves_panic_append_info();
531 #endif
532
533 if (kernelcache_uuid_valid) {
534 if (filesetKC) {
535 paniclog_append_noflush("Fileset Kernelcache UUID: ");
536 } else {
537 paniclog_append_noflush("KernelCache UUID: ");
538 }
539 for (size_t index = 0; index < sizeof(uuid_t); index++) {
540 paniclog_append_noflush("%02X", kernelcache_uuid[index]);
541 }
542 paniclog_append_noflush("\n");
543 }
544 panic_display_kernel_uuid();
545
546 if (bootsessionuuid_string[0] != '\0') {
547 paniclog_append_noflush("Boot session UUID: %s\n", bootsessionuuid_string);
548 } else {
549 paniclog_append_noflush("Boot session UUID not yet initialized\n");
550 }
551
552 paniclog_append_noflush("iBoot version: %.128s\n", iBoot_version);
553 #if defined(TARGET_OS_OSX) && defined(__arm64__)
554 paniclog_append_noflush("iBoot Stage 2 version: %.128s\n", iBoot_Stage_2_version);
555 #endif /* defined(TARGET_OS_OSX) && defined(__arm64__) */
556
557 paniclog_append_noflush("secure boot?: %s\n", debug_enabled ? "NO": "YES");
558 paniclog_append_noflush("roots installed: %lld\n", roots_installed);
559 #if defined(XNU_TARGET_OS_BRIDGE)
560 paniclog_append_noflush("x86 EFI Boot State: ");
561 if (PE_smc_stashed_x86_efi_boot_state != 0xFF) {
562 paniclog_append_noflush("0x%x\n", PE_smc_stashed_x86_efi_boot_state);
563 } else {
564 paniclog_append_noflush("not available\n");
565 }
566 paniclog_append_noflush("x86 System State: ");
567 if (PE_smc_stashed_x86_system_state != 0xFF) {
568 paniclog_append_noflush("0x%x\n", PE_smc_stashed_x86_system_state);
569 } else {
570 paniclog_append_noflush("not available\n");
571 }
572 paniclog_append_noflush("x86 Power State: ");
573 if (PE_smc_stashed_x86_power_state != 0xFF) {
574 paniclog_append_noflush("0x%x\n", PE_smc_stashed_x86_power_state);
575 } else {
576 paniclog_append_noflush("not available\n");
577 }
578 paniclog_append_noflush("x86 Shutdown Cause: ");
579 if (PE_smc_stashed_x86_shutdown_cause != 0xFF) {
580 paniclog_append_noflush("0x%x\n", PE_smc_stashed_x86_shutdown_cause);
581 } else {
582 paniclog_append_noflush("not available\n");
583 }
584 paniclog_append_noflush("x86 Previous Power Transitions: ");
585 if (PE_smc_stashed_x86_prev_power_transitions != UINT64_MAX) {
586 paniclog_append_noflush("0x%llx\n", PE_smc_stashed_x86_prev_power_transitions);
587 } else {
588 paniclog_append_noflush("not available\n");
589 }
590 paniclog_append_noflush("PCIeUp link state: ");
591 if (PE_pcie_stashed_link_state != UINT32_MAX) {
592 paniclog_append_noflush("0x%x\n", PE_pcie_stashed_link_state);
593 } else {
594 paniclog_append_noflush("not available\n");
595 }
596 paniclog_append_noflush("macOS kernel slide: ");
597 if (PE_nvram_stashed_x86_macos_slide != UINT64_MAX) {
598 paniclog_append_noflush("%#llx\n", PE_nvram_stashed_x86_macos_slide);
599 } else {
600 paniclog_append_noflush("not available\n");
601 }
602 #endif
603 if (panic_data_buffers != NULL) {
604 paniclog_append_noflush("%s data: ", panic_data_buffers->producer_name);
605 uint8_t *panic_buffer_data = (uint8_t *) panic_data_buffers->buf;
606 for (int i = 0; i < panic_data_buffers->len; i++) {
607 paniclog_append_noflush("%02X", panic_buffer_data[i]);
608 }
609 paniclog_append_noflush("\n");
610 }
611 paniclog_append_noflush("Paniclog version: %d\n", logversion);
612
613 panic_display_kernel_aslr();
614 panic_display_times();
615 panic_display_zalloc();
616 panic_display_hung_cpus_help();
617 panic_display_tpidrs();
618
619
620 panic_display_pvhs_locked();
621 panic_display_pvh_to_lock();
622 panic_display_last_pc_lr();
623 #if CONFIG_ECC_LOGGING
624 panic_display_ecc_errors();
625 #endif /* CONFIG_ECC_LOGGING */
626 panic_display_compressor_stats();
627
628 #if DEVELOPMENT || DEBUG
629 if (cs_debug_unsigned_exec_failures != 0 || cs_debug_unsigned_mmap_failures != 0) {
630 paniclog_append_noflush("Unsigned code exec failures: %u\n", cs_debug_unsigned_exec_failures);
631 paniclog_append_noflush("Unsigned code mmap failures: %u\n", cs_debug_unsigned_mmap_failures);
632 }
633 #endif
634
635 // Highlight threads that used high amounts of CPU in the panic log if requested (historically requested for watchdog panics)
636 if (panic_options & DEBUGGER_OPTION_PRINT_CPU_USAGE_PANICLOG) {
637 thread_t top_runnable[5] = {0};
638 thread_t thread;
639 int total_cpu_usage = 0;
640
641 print_vnodes = 1;
642
643
644 for (thread = (thread_t)queue_first(&threads);
645 PANIC_VALIDATE_PTR(thread) && !queue_end(&threads, (queue_entry_t)thread);
646 thread = (thread_t)queue_next(&thread->threads)) {
647 total_cpu_usage += thread->cpu_usage;
648
649 // Look for the 5 runnable threads with highest priority
650 if (thread->state & TH_RUN) {
651 int k;
652 thread_t comparison_thread = thread;
653
654 for (k = 0; k < TOP_RUNNABLE_LIMIT; k++) {
655 if (top_runnable[k] == 0) {
656 top_runnable[k] = comparison_thread;
657 break;
658 } else if (comparison_thread->sched_pri > top_runnable[k]->sched_pri) {
659 thread_t temp = top_runnable[k];
660 top_runnable[k] = comparison_thread;
661 comparison_thread = temp;
662 } // if comparison thread has higher priority than previously saved thread
663 } // loop through highest priority runnable threads
664 } // Check if thread is runnable
665 } // Loop through all threads
666
667 // Print the relevant info for each thread identified
668 paniclog_append_noflush("Total cpu_usage: %d\n", total_cpu_usage);
669 paniclog_append_noflush("Thread task pri cpu_usage\n");
670
671 for (int i = 0; i < TOP_RUNNABLE_LIMIT; i++) {
672 if (top_runnable[i] &&
673 panic_get_thread_proc_task(top_runnable[i], &task, &proc) && proc) {
674 char name[MAX_PROCNAME_LEN + 1];
675 proc_name_kdp(proc, name, sizeof(name));
676 paniclog_append_noflush("%p %s %d %d\n",
677 top_runnable[i], name, top_runnable[i]->sched_pri, top_runnable[i]->cpu_usage);
678 }
679 } // Loop through highest priority runnable threads
680 paniclog_append_noflush("\n");
681 }
682
683 // print current task info
684 if (panic_get_thread_proc_task(cur_thread, &task, &proc)) {
685 if (PANIC_VALIDATE_PTR(task->map) &&
686 PANIC_VALIDATE_PTR(task->map->pmap)) {
687 ledger_amount_t resident = 0;
688 if (task != kernel_task) {
689 ledger_get_balance(task->ledger, task_ledgers.phys_mem, &resident);
690 resident >>= VM_MAP_PAGE_SHIFT(task->map);
691 }
692 paniclog_append_noflush("Panicked task %p: %lld pages, %d threads: ",
693 task, resident, task->thread_count);
694 } else {
695 paniclog_append_noflush("Panicked task %p: %d threads: ",
696 task, task->thread_count);
697 }
698
699 if (proc) {
700 char name[MAX_PROCNAME_LEN + 1];
701 proc_name_kdp(proc, name, sizeof(name));
702 paniclog_append_noflush("pid %d: %s", proc_pid(proc), name);
703 } else {
704 paniclog_append_noflush("unknown task");
705 }
706
707 paniclog_append_noflush("\n");
708 }
709
710 print_backtrace_internal(cur_thread, filesetKC);
711
712 paniclog_append_noflush("\n");
713 dump_cpu_event_log(&paniclog_append_noflush);
714
715 paniclog_append_noflush("\n");
716 if (filesetKC) {
717 kext_dump_panic_lists(&paniclog_append_noflush);
718 paniclog_append_noflush("\n");
719 }
720 panic_info->eph_panic_log_len = PE_get_offset_into_panic_region(debug_buf_ptr) - panic_info->eph_panic_log_offset;
721 /* set the os version data in the panic header in the format 'Product Version (OS Version)' (only if they have been set) */
722 if ((osversion[0] != '\0') && (osproductversion[0] != '\0')) {
723 snprintf((char *)&panic_info->eph_os_version, sizeof(panic_info->eph_os_version), PANIC_HEADER_VERSION_FMT_STR,
724 osproductversion, osversion);
725 }
726 #if defined(XNU_TARGET_OS_BRIDGE)
727 if ((macosversion[0] != '\0') && (macosproductversion[0] != '\0')) {
728 snprintf((char *)&panic_info->eph_macos_version, sizeof(panic_info->eph_macos_version), PANIC_HEADER_VERSION_FMT_STR,
729 macosproductversion, macosversion);
730 }
731 #endif
732 if (bootsessionuuid_string[0] != '\0') {
733 memcpy(panic_info->eph_bootsessionuuid_string, bootsessionuuid_string,
734 sizeof(panic_info->eph_bootsessionuuid_string));
735 }
736 panic_info->eph_roots_installed = roots_installed;
737
738 if (panic_initiator != NULL) {
739 bytes_remaining = debug_buf_size - (unsigned int)((uintptr_t)debug_buf_ptr - (uintptr_t)debug_buf_base);
740 // If panic_initiator isn't null, safely copy up to MAX_PANIC_INITIATOR_SIZE
741 panic_initiator_len = strnlen(panic_initiator, MAX_PANIC_INITIATOR_SIZE);
742 // Calculate the bytes to write, accounting for remaining buffer space, and ensuring the lowest size we can have is 0
743 panic_initiator_len = MAX(0, MIN(panic_initiator_len, bytes_remaining));
744 panic_info->eph_panic_initiator_offset = (panic_initiator_len != 0) ? PE_get_offset_into_panic_region(debug_buf_ptr) : 0;
745 panic_info->eph_panic_initiator_len = panic_initiator_len;
746 memcpy(debug_buf_ptr, panic_initiator, panic_initiator_len);
747 debug_buf_ptr += panic_initiator_len;
748 }
749
750 if (debug_ack_timeout_count) {
751 panic_info->eph_panic_flags |= EMBEDDED_PANIC_HEADER_FLAG_STACKSHOT_FAILED_DEBUGGERSYNC;
752 panic_info->eph_other_log_offset = PE_get_offset_into_panic_region(debug_buf_ptr);
753 paniclog_append_noflush("!! debugger synchronization failed, no stackshot !!\n");
754 } else if (panic_stackshot_active()) {
755 panic_info->eph_panic_flags |= EMBEDDED_PANIC_HEADER_FLAG_STACKSHOT_FAILED_NESTED;
756 panic_info->eph_other_log_offset = PE_get_offset_into_panic_region(debug_buf_ptr);
757 paniclog_append_noflush("!! panicked during stackshot, skipping panic stackshot !!\n");
758 } else {
759 /* Align the stackshot buffer to an 8-byte address (especially important for armv7k devices) */
760 debug_buf_ptr += (8 - ((uintptr_t)debug_buf_ptr % 8));
761 stackshot_begin_loc = debug_buf_ptr;
762
763 bytes_remaining = debug_buf_size - (unsigned int)((uintptr_t)stackshot_begin_loc - (uintptr_t)debug_buf_base);
764 err = kcdata_memory_static_init(&kc_panic_data, (mach_vm_address_t)debug_buf_ptr,
765 KCDATA_BUFFER_BEGIN_COMPRESSED, bytes_remaining - end_marker_bytes,
766 KCFLAG_USE_MEMCOPY);
767 if (err == KERN_SUCCESS) {
768 uint64_t stackshot_flags = (STACKSHOT_GET_GLOBAL_MEM_STATS | STACKSHOT_SAVE_LOADINFO | STACKSHOT_KCDATA_FORMAT |
769 STACKSHOT_ENABLE_BT_FAULTING | STACKSHOT_ENABLE_UUID_FAULTING | STACKSHOT_FROM_PANIC | STACKSHOT_DO_COMPRESS |
770 STACKSHOT_DISABLE_LATENCY_INFO | STACKSHOT_NO_IO_STATS | STACKSHOT_THREAD_WAITINFO | STACKSHOT_GET_DQ |
771 STACKSHOT_COLLECT_SHAREDCACHE_LAYOUT);
772
773 err = kcdata_init_compress(&kc_panic_data, KCDATA_BUFFER_BEGIN_STACKSHOT, kdp_memcpy, KCDCT_ZLIB);
774 if (err != KERN_SUCCESS) {
775 panic_info->eph_panic_flags |= EMBEDDED_PANIC_HEADER_FLAG_COMPRESS_FAILED;
776 stackshot_flags &= ~STACKSHOT_DO_COMPRESS;
777 }
778 if (filesetKC) {
779 stackshot_flags |= STACKSHOT_SAVE_KEXT_LOADINFO;
780 }
781
782 kdp_snapshot_preflight(-1, stackshot_begin_loc, bytes_remaining - end_marker_bytes,
783 stackshot_flags, &kc_panic_data, 0, 0);
784 err = do_panic_stackshot(NULL);
785 bytes_traced = kdp_stack_snapshot_bytes_traced();
786 if (bytes_traced > 0 && !err) {
787 debug_buf_ptr += bytes_traced;
788 panic_info->eph_panic_flags |= EMBEDDED_PANIC_HEADER_FLAG_STACKSHOT_SUCCEEDED;
789 panic_info->eph_stackshot_offset = PE_get_offset_into_panic_region(stackshot_begin_loc);
790 panic_info->eph_stackshot_len = bytes_traced;
791
792 panic_info->eph_other_log_offset = PE_get_offset_into_panic_region(debug_buf_ptr);
793 #if CONFIG_EXCLAVES
794 panic_report_exclaves_stackshot();
795 #endif /* CONFIG_EXCLAVES */
796 if (stackshot_flags & STACKSHOT_DO_COMPRESS) {
797 panic_info->eph_panic_flags |= EMBEDDED_PANIC_HEADER_FLAG_STACKSHOT_DATA_COMPRESSED;
798 bytes_uncompressed = kdp_stack_snapshot_bytes_uncompressed();
799 paniclog_append_noflush("\n** Stackshot Succeeded ** Bytes Traced %d (Uncompressed %d) **\n", bytes_traced, bytes_uncompressed);
800 } else {
801 paniclog_append_noflush("\n** Stackshot Succeeded ** Bytes Traced %d **\n", bytes_traced);
802 }
803 } else {
804 bytes_used = kcdata_memory_get_used_bytes(&kc_panic_data);
805 #if CONFIG_EXCLAVES
806 panic_report_exclaves_stackshot();
807 #endif /* CONFIG_EXCLAVES */
808 if (bytes_used > 0) {
809 /* Zero out the stackshot data */
810 bzero(stackshot_begin_loc, bytes_used);
811 panic_info->eph_panic_flags |= EMBEDDED_PANIC_HEADER_FLAG_STACKSHOT_FAILED_INCOMPLETE;
812
813 panic_info->eph_other_log_offset = PE_get_offset_into_panic_region(debug_buf_ptr);
814 paniclog_append_noflush("\n** Stackshot Incomplete ** Bytes Filled %llu, err %d **\n", bytes_used, err);
815 } else {
816 bzero(stackshot_begin_loc, bytes_used);
817 panic_info->eph_panic_flags |= EMBEDDED_PANIC_HEADER_FLAG_STACKSHOT_FAILED_ERROR;
818
819 panic_info->eph_other_log_offset = PE_get_offset_into_panic_region(debug_buf_ptr);
820 paniclog_append_noflush("\n!! Stackshot Failed !! Bytes Traced %d, err %d\n", bytes_traced, err);
821 }
822 }
823 } else {
824 panic_info->eph_panic_flags |= EMBEDDED_PANIC_HEADER_FLAG_STACKSHOT_FAILED_ERROR;
825 panic_info->eph_other_log_offset = PE_get_offset_into_panic_region(debug_buf_ptr);
826 paniclog_append_noflush("\n!! Stackshot Failed !!\nkcdata_memory_static_init returned %d", err);
827 }
828 }
829
830 #if CONFIG_EXT_PANICLOG
831 // Write ext paniclog at the end of the paniclog region.
832 ext_paniclog_bytes = ext_paniclog_write_panicdata();
833 panic_info->eph_ext_paniclog_offset = (ext_paniclog_bytes != 0) ?
834 PE_get_offset_into_panic_region((debug_buf_base + debug_buf_size) - ext_paniclog_bytes) :
835 0;
836 panic_info->eph_ext_paniclog_len = ext_paniclog_bytes;
837 #endif
838
839 assert(panic_info->eph_other_log_offset != 0);
840
841 if (print_vnodes != 0) {
842 panic_print_vnodes();
843 }
844
845 panic_bt_depth--;
846 }
847
848 /*
849 * Entry to print_all_panic_info is serialized by the debugger lock
850 */
851 static void
print_all_panic_info(const char * message,uint64_t panic_options,const char * panic_initiator)852 print_all_panic_info(const char *message, uint64_t panic_options, const char *panic_initiator)
853 {
854 unsigned int initial_not_in_kdp = not_in_kdp;
855
856 cpu_data_t * cpu_data_ptr = getCpuDatap();
857
858 assert(cpu_data_ptr->PAB_active == FALSE);
859 cpu_data_ptr->PAB_active = TRUE;
860
861 /*
862 * Because print all backtraces uses the pmap routines, it needs to
863 * avoid taking pmap locks. Right now, this is conditionalized on
864 * not_in_kdp.
865 */
866 not_in_kdp = 0;
867 do_print_all_panic_info(message, panic_options, panic_initiator);
868
869 not_in_kdp = initial_not_in_kdp;
870
871 cpu_data_ptr->PAB_active = FALSE;
872 }
873
874 void
print_curr_backtrace(void)875 print_curr_backtrace(void)
876 {
877 print_backtrace_internal(current_thread(), is_filesetKC());
878 }
879
880 void
panic_display_times()881 panic_display_times()
882 {
883 if (kdp_clock_is_locked()) {
884 paniclog_append_noflush("Warning: clock is locked. Can't get time\n");
885 return;
886 }
887
888 extern lck_ticket_t clock_lock;
889 extern lck_grp_t clock_lock_grp;
890
891 if ((is_clock_configured) && (lck_ticket_lock_try(&clock_lock, &clock_lock_grp))) {
892 clock_sec_t secs, boot_secs;
893 clock_usec_t usecs, boot_usecs;
894
895 lck_ticket_unlock(&clock_lock);
896
897 clock_get_calendar_microtime(&secs, &usecs);
898 clock_get_boottime_microtime(&boot_secs, &boot_usecs);
899
900 paniclog_append_noflush("mach_absolute_time: 0x%llx\n", mach_absolute_time());
901 paniclog_append_noflush("Epoch Time: sec usec\n");
902 paniclog_append_noflush(" Boot : 0x%08x 0x%08x\n", (unsigned int)boot_secs, (unsigned int)boot_usecs);
903 paniclog_append_noflush(" Sleep : 0x%08x 0x%08x\n", (unsigned int)gIOLastSleepTime.tv_sec, (unsigned int)gIOLastSleepTime.tv_usec);
904 paniclog_append_noflush(" Wake : 0x%08x 0x%08x\n", (unsigned int)gIOLastWakeTime.tv_sec, (unsigned int)gIOLastWakeTime.tv_usec);
905 paniclog_append_noflush(" Calendar: 0x%08x 0x%08x\n\n", (unsigned int)secs, (unsigned int)usecs);
906 }
907 }
908
909 void
panic_print_symbol_name(vm_address_t search)910 panic_print_symbol_name(vm_address_t search)
911 {
912 #pragma unused(search)
913 // empty stub. Really only used on x86_64.
914 return;
915 }
916
917 void
SavePanicInfo(const char * message,__unused void * panic_data,uint64_t panic_options,const char * panic_initiator)918 SavePanicInfo(
919 const char *message, __unused void *panic_data, uint64_t panic_options, const char* panic_initiator)
920 {
921 /*
922 * This should be initialized by the time we get here, but
923 * if it is not, asserting about it will be of no use (it will
924 * come right back to here), so just loop right here and now.
925 * This prevents early-boot panics from becoming recursive and
926 * thus makes them easier to debug. If you attached to a device
927 * and see your PC here, look down a few frames to see your
928 * early-boot panic there.
929 */
930 while (!panic_info || panic_info->eph_panic_log_offset == 0) {
931 // rdar://87170225 (PanicHardening: audit panic code for naked spinloops)
932 // rdar://88094367 (Add test hooks for panic at different stages in XNU)
933 ;
934 }
935
936 if (panic_options & DEBUGGER_OPTION_PANICLOGANDREBOOT) {
937 panic_info->eph_panic_flags |= EMBEDDED_PANIC_HEADER_FLAG_BUTTON_RESET_PANIC;
938 }
939
940 if (panic_options & DEBUGGER_OPTION_COMPANION_PROC_INITIATED_PANIC) {
941 panic_info->eph_panic_flags |= EMBEDDED_PANIC_HEADER_FLAG_COMPANION_PROC_INITIATED_PANIC;
942 }
943
944 if (panic_options & DEBUGGER_OPTION_INTEGRATED_COPROC_INITIATED_PANIC) {
945 panic_info->eph_panic_flags |= EMBEDDED_PANIC_HEADER_FLAG_INTEGRATED_COPROC_INITIATED_PANIC;
946 }
947
948 if (panic_options & DEBUGGER_OPTION_USERSPACE_INITIATED_PANIC) {
949 panic_info->eph_panic_flags |= EMBEDDED_PANIC_HEADER_FLAG_USERSPACE_INITIATED_PANIC;
950 }
951
952 #if defined(XNU_TARGET_OS_BRIDGE)
953 panic_info->eph_x86_power_state = PE_smc_stashed_x86_power_state;
954 panic_info->eph_x86_efi_boot_state = PE_smc_stashed_x86_efi_boot_state;
955 panic_info->eph_x86_system_state = PE_smc_stashed_x86_system_state;
956 #endif
957
958 /*
959 * On newer targets, panic data is stored directly into the iBoot panic region.
960 * If we re-enter SavePanicInfo (e.g. on a double panic) on such a target, update the
961 * panic CRC so that iBoot can hopefully find *something* useful in the panic region.
962 */
963 if (PanicInfoSaved && (debug_buf_base >= (char*)gPanicBase) && (debug_buf_base < (char*)gPanicBase + gPanicSize)) {
964 unsigned int pi_size = (unsigned int)(debug_buf_ptr - gPanicBase);
965 PE_update_panic_crc((unsigned char*)gPanicBase, &pi_size);
966 PE_sync_panic_buffers(); // extra precaution; panic path likely isn't reliable if we're here
967 }
968
969 if (PanicInfoSaved || (debug_buf_size == 0)) {
970 return;
971 }
972
973 PanicInfoSaved = TRUE;
974
975
976 print_all_panic_info(message, panic_options, panic_initiator);
977
978 assert(panic_info->eph_panic_log_len != 0);
979 panic_info->eph_other_log_len = PE_get_offset_into_panic_region(debug_buf_ptr) - panic_info->eph_other_log_offset;
980
981 PEHaltRestart(kPEPanicSync);
982
983 /*
984 * Notifies registered IOPlatformPanicAction callbacks
985 * (which includes one to disable the memcache) and flushes
986 * the buffer contents from the cache
987 */
988 paniclog_flush();
989 }
990
991 void
paniclog_flush()992 paniclog_flush()
993 {
994 unsigned int panicbuf_length = 0;
995
996 panicbuf_length = (unsigned int)(debug_buf_ptr - gPanicBase);
997 if (!debug_buf_ptr || !panicbuf_length) {
998 return;
999 }
1000
1001 /*
1002 * Updates the log length of the last part of the panic log.
1003 */
1004 panic_info->eph_other_log_len = PE_get_offset_into_panic_region(debug_buf_ptr) - panic_info->eph_other_log_offset;
1005
1006 /*
1007 * Updates the metadata at the beginning of the panic buffer,
1008 * updates the CRC.
1009 */
1010 PE_update_panic_crc((unsigned char *)gPanicBase, &panicbuf_length);
1011
1012 /*
1013 * This is currently unused by platform KEXTs on embedded but is
1014 * kept for compatibility with the published IOKit interfaces.
1015 */
1016 PESavePanicInfo((unsigned char *)gPanicBase, panicbuf_length);
1017
1018 PE_sync_panic_buffers();
1019 }
1020
1021 #if CONFIG_SPTM
1022 /*
1023 * Patch thread state to appear as if a debugger stop IPI occurred, when a thread
1024 * is parked in SPTM panic loop. This allows stackshot to proceed as usual.
1025 */
1026 static void
DebuggerPatchupThreadState(int cpu,xnu_saved_registers_t * regp)1027 DebuggerPatchupThreadState(
1028 int cpu, xnu_saved_registers_t *regp)
1029 {
1030 cpu_data_t *target_cpu_datap;
1031 arm_saved_state_t *statep;
1032 vm_offset_t kstackptr;
1033
1034 target_cpu_datap = (cpu_data_t *)CpuDataEntries[cpu].cpu_data_vaddr;
1035 statep = target_cpu_datap->cpu_active_thread->machine.kpcb;
1036 kstackptr = (vm_offset_t)target_cpu_datap->cpu_active_thread->machine.kstackptr;
1037
1038 target_cpu_datap->ipi_pc = regp->pc;
1039 target_cpu_datap->ipi_lr = regp->lr;
1040 target_cpu_datap->ipi_fp = regp->fp;
1041
1042 if (statep != NULL) {
1043 statep->ss_64.fp = regp->fp;
1044 statep->ss_64.lr = regp->lr;
1045 statep->ss_64.sp = regp->sp;
1046 statep->ss_64.pc = regp->pc;
1047 } else if ((void *)kstackptr != NULL) {
1048 arm_kernel_saved_state_t *kstatep = (arm_kernel_saved_state_t *)kstackptr;
1049 kstatep->fp = regp->fp;
1050 kstatep->lr = regp->lr;
1051 kstatep->sp = regp->sp;
1052 }
1053 }
1054 #endif
1055
1056 /*
1057 * @function DebuggerXCallEnter
1058 *
1059 * @abstract IPI other cores so this core can run in a single-threaded context.
1060 *
1061 * @discussion This function should be called with the debugger lock held. It
1062 * signals the other cores to go into a busy loop so this core can run in a
1063 * single-threaded context and inspect kernel memory.
1064 *
1065 * @param proceed_on_sync_failure If true, then go ahead and try to debug even
1066 * if we can't synch with the other cores. This is inherently unsafe and should
1067 * only be used if the kernel is going down in flames anyway.
1068 *
1069 * @param is_stackshot If true, this is a stackshot request.
1070 *
1071 * @result returns KERN_OPERATION_TIMED_OUT if synchronization times out and
1072 * proceed_on_sync_failure is false.
1073 */
1074 kern_return_t
DebuggerXCallEnter(boolean_t proceed_on_sync_failure,bool is_stackshot)1075 DebuggerXCallEnter(
1076 boolean_t proceed_on_sync_failure, bool is_stackshot)
1077 {
1078 uint64_t max_mabs_time, current_mabs_time;
1079 int cpu;
1080 int timeout_cpu = -1;
1081 int max_cpu;
1082 unsigned int sync_pending;
1083 cpu_data_t *target_cpu_datap;
1084 cpu_data_t *cpu_data_ptr = getCpuDatap();
1085
1086 /* Check for nested debugger entry. */
1087 cpu_data_ptr->debugger_active++;
1088 if (cpu_data_ptr->debugger_active != 1) {
1089 return KERN_SUCCESS;
1090 }
1091
1092 /*
1093 * If debugger_sync is not 0, someone responded excessively late to the last
1094 * debug request (we zero the sync variable in the return function). Zero it
1095 * again here. This should prevent us from getting out of sync (heh) and
1096 * timing out on every entry to the debugger if we timeout once.
1097 */
1098
1099 os_atomic_store(&debugger_sync, 0, relaxed);
1100 os_atomic_store(&mp_kdp_trap, 1, relaxed);
1101 os_atomic_store(&debug_cpus_spinning, 0, relaxed);
1102 trap_is_stackshot = is_stackshot;
1103
1104
1105 /*
1106 * Try to signal all CPUs (except ourselves, of course). Use debugger_sync to
1107 * synchronize with every CPU that we appeared to signal successfully (cpu_signal
1108 * is not synchronous).
1109 */
1110 max_cpu = ml_get_max_cpu_number();
1111
1112 boolean_t immediate_halt = FALSE;
1113 if (proceed_on_sync_failure && force_immediate_debug_halt) {
1114 immediate_halt = TRUE;
1115 }
1116
1117 if (!immediate_halt) {
1118 for (cpu = 0; cpu <= max_cpu; cpu++) {
1119 target_cpu_datap = (cpu_data_t *)CpuDataEntries[cpu].cpu_data_vaddr;
1120
1121 if ((target_cpu_datap == NULL) || (target_cpu_datap == cpu_data_ptr)) {
1122 continue;
1123 }
1124
1125 kern_return_t ret = cpu_signal(target_cpu_datap, SIGPdebug, (void *)NULL, NULL);
1126 if (ret == KERN_SUCCESS) {
1127 os_atomic_inc(&debugger_sync, relaxed);
1128 os_atomic_inc(&debug_cpus_spinning, relaxed);
1129 } else {
1130 kprintf("%s: cpu_signal failed. cpu=%d ret=%d proceed=%d\n", __func__, cpu, ret, proceed_on_sync_failure);
1131 }
1132 }
1133
1134 max_mabs_time = os_atomic_load(&debug_ack_timeout, relaxed);
1135
1136 if (max_mabs_time > 0) {
1137 current_mabs_time = mach_absolute_time();
1138 max_mabs_time += current_mabs_time;
1139 assert(max_mabs_time > current_mabs_time);
1140 }
1141
1142 /*
1143 * Wait for DEBUG_ACK_TIMEOUT ns for a response from everyone we IPI'd. If we
1144 * timeout, that is simply too bad; we don't have a true NMI, and one CPU may be
1145 * uninterruptibly spinning on someone else. The best we can hope for is that
1146 * all other CPUs have either responded or are spinning in a context that is
1147 * debugger safe.
1148 */
1149 do {
1150 current_mabs_time = mach_absolute_time();
1151 sync_pending = os_atomic_load(&debugger_sync, acquire);
1152 } while ((sync_pending != 0) && (max_mabs_time == 0 || current_mabs_time < max_mabs_time));
1153 }
1154
1155 if (!immediate_halt && max_mabs_time > 0 && current_mabs_time >= max_mabs_time) {
1156 /*
1157 * We timed out trying to IPI the other CPUs. Skip counting any CPUs that
1158 * are offline; then we must account for the remainder, either counting
1159 * them as halted, or trying to dbgwrap them to get them to halt in the
1160 * case where the system is going down and we are running a dev fused
1161 * device.
1162 */
1163 __builtin_arm_dmb(DMB_ISH);
1164 for (cpu = 0; cpu <= max_cpu; cpu++) {
1165 target_cpu_datap = (cpu_data_t *)CpuDataEntries[cpu].cpu_data_vaddr;
1166
1167 if ((target_cpu_datap == NULL) || (target_cpu_datap == cpu_data_ptr)) {
1168 continue;
1169 }
1170 if (!(target_cpu_datap->cpu_signal & SIGPdebug)) {
1171 continue;
1172 }
1173 if (processor_array[cpu]->state <= PROCESSOR_PENDING_OFFLINE) {
1174 int dbg_sync_count;
1175
1176 /*
1177 * This is a processor that was successfully sent a SIGPdebug signal
1178 * but which hasn't acknowledged it because it went offline with
1179 * interrupts disabled before the IPI was delivered, so count it
1180 * as halted here.
1181 */
1182 dbg_sync_count = os_atomic_dec(&debugger_sync, relaxed);
1183 kprintf("%s>found CPU %d offline, debugger_sync=%d\n", __FUNCTION__, cpu, dbg_sync_count);
1184 continue;
1185 }
1186 kprintf("%s>Debugger synch pending on cpu %d\n", __FUNCTION__, cpu);
1187 timeout_cpu = cpu;
1188 #if CONFIG_SPTM
1189 if (proceed_on_sync_failure) {
1190 /*
1191 * If a core is spinning in the SPTM panic loop, consider it
1192 * as sync'd, and try to patch up the thread state from the
1193 * SPTM callee saved registers.
1194 */
1195 bool sptm_panic_loop = false;
1196 vm_offset_t base = other_percpu_base(cpu);
1197 pmap_sptm_percpu_data_t *sptm_pcpu = PERCPU_GET_WITH_BASE(base, pmap_sptm_percpu);
1198 uint64_t sptm_cpuid = sptm_pcpu->sptm_cpu_id;
1199
1200 if (sptm_get_cpu_state(sptm_cpuid, CPUSTATE_PANIC_SPIN, &sptm_panic_loop)
1201 == SPTM_SUCCESS && sptm_panic_loop) {
1202 xnu_saved_registers_t regs;
1203
1204 if (sptm_copy_callee_saved_state(sptm_cpuid, ®s)
1205 == LIBSPTM_SUCCESS) {
1206 DebuggerPatchupThreadState(cpu, ®s);
1207 }
1208
1209 kprintf("%s>found CPU %d in SPTM\n", __FUNCTION__, cpu);
1210 os_atomic_dec(&debugger_sync, relaxed);
1211 }
1212 }
1213 #endif
1214 }
1215
1216 if (debugger_sync == 0) {
1217 return KERN_SUCCESS;
1218 } else if (!proceed_on_sync_failure) {
1219 panic("%s>Debugger synch pending on cpu %d\n",
1220 __FUNCTION__, timeout_cpu);
1221 }
1222 }
1223 if (immediate_halt || (max_mabs_time > 0 && current_mabs_time >= max_mabs_time)) {
1224 if (immediate_halt) {
1225 __builtin_arm_dmb(DMB_ISH);
1226 }
1227 for (cpu = 0; cpu <= max_cpu; cpu++) {
1228 target_cpu_datap = (cpu_data_t *)CpuDataEntries[cpu].cpu_data_vaddr;
1229
1230 if ((target_cpu_datap == NULL) || (target_cpu_datap == cpu_data_ptr)) {
1231 continue;
1232 }
1233 paniclog_append_noflush("Attempting to forcibly halt cpu %d\n", cpu);
1234 dbgwrap_status_t halt_status = ml_dbgwrap_halt_cpu(cpu, 0);
1235 if (halt_status < 0) {
1236 paniclog_append_noflush("cpu %d failed to halt with error %d: %s\n", cpu, halt_status, ml_dbgwrap_strerror(halt_status));
1237 } else {
1238 if (halt_status > 0) {
1239 paniclog_append_noflush("cpu %d halted with warning %d: %s\n", cpu, halt_status, ml_dbgwrap_strerror(halt_status));
1240 }
1241 target_cpu_datap->halt_status = CPU_HALTED;
1242 }
1243 }
1244 for (cpu = 0; cpu <= max_cpu; cpu++) {
1245 target_cpu_datap = (cpu_data_t *)CpuDataEntries[cpu].cpu_data_vaddr;
1246
1247 if ((target_cpu_datap == NULL) || (target_cpu_datap == cpu_data_ptr)) {
1248 continue;
1249 }
1250 dbgwrap_status_t halt_status = ml_dbgwrap_halt_cpu_with_state(cpu,
1251 NSEC_PER_SEC, &target_cpu_datap->halt_state);
1252 if ((halt_status < 0) || (halt_status == DBGWRAP_WARN_CPU_OFFLINE)) {
1253 paniclog_append_noflush("Unable to obtain state for cpu %d with status %d: %s\n", cpu, halt_status, ml_dbgwrap_strerror(halt_status));
1254 debug_ack_timeout_count++;
1255 } else {
1256 paniclog_append_noflush("cpu %d successfully halted\n", cpu);
1257 target_cpu_datap->halt_status = CPU_HALTED_WITH_STATE;
1258 }
1259 }
1260 if (immediate_halt) {
1261 paniclog_append_noflush("Immediate halt requested on all cores\n");
1262 } else {
1263 paniclog_append_noflush("Debugger synchronization timed out; timeout %llu nanoseconds\n",
1264 os_atomic_load(&debug_ack_timeout, relaxed));
1265 }
1266 }
1267 return KERN_SUCCESS;
1268 }
1269
1270 /*
1271 * @function DebuggerXCallReturn
1272 *
1273 * @abstract Resume normal multicore operation after DebuggerXCallEnter()
1274 *
1275 * @discussion This function should be called with debugger lock held.
1276 */
1277 void
DebuggerXCallReturn(void)1278 DebuggerXCallReturn(
1279 void)
1280 {
1281 cpu_data_t *cpu_data_ptr = getCpuDatap();
1282 uint64_t max_mabs_time, current_mabs_time;
1283
1284 cpu_data_ptr->debugger_active--;
1285 if (cpu_data_ptr->debugger_active != 0) {
1286 return;
1287 }
1288
1289 os_atomic_store(&mp_kdp_trap, 0, release);
1290 os_atomic_store(&debugger_sync, 0, relaxed);
1291
1292 max_mabs_time = os_atomic_load(&debug_ack_timeout, relaxed);
1293
1294 if (max_mabs_time > 0) {
1295 current_mabs_time = mach_absolute_time();
1296 max_mabs_time += current_mabs_time;
1297 assert(max_mabs_time > current_mabs_time);
1298 }
1299
1300 /*
1301 * Wait for other CPUs to stop spinning on mp_kdp_trap (see DebuggerXCall).
1302 * It's possible for one or more CPUs to not decrement debug_cpus_spinning,
1303 * since they may be stuck somewhere else with interrupts disabled.
1304 * Wait for DEBUG_ACK_TIMEOUT ns for a response and move on if we don't get it.
1305 *
1306 * Note that the same is done in DebuggerXCallEnter, when we wait for other
1307 * CPUS to update debugger_sync. If we time out, let's hope for all CPUs to be
1308 * spinning in a debugger-safe context
1309 */
1310 while ((os_atomic_load_exclusive(&debug_cpus_spinning, acquire) != 0) &&
1311 (max_mabs_time == 0 || current_mabs_time < max_mabs_time)) {
1312 __builtin_arm_wfe();
1313 current_mabs_time = mach_absolute_time();
1314 }
1315 os_atomic_clear_exclusive();
1316
1317 // checking debug_ack_timeout != 0 is a workaround for rdar://124242354
1318 if (current_mabs_time >= max_mabs_time && os_atomic_load(&debug_ack_timeout, relaxed) != 0) {
1319 panic("Resuming from debugger synchronization failed: waited %llu nanoseconds\n", os_atomic_load(&debug_ack_timeout, relaxed));
1320 }
1321 }
1322
1323 extern void wait_while_mp_kdp_trap(bool check_SIGPdebug);
1324 /*
1325 * Spin while mp_kdp_trap is set.
1326 *
1327 * processor_offline() calls this with check_SIGPdebug=true
1328 * to break out of the spin loop if the cpu has SIGPdebug
1329 * pending.
1330 */
1331 void
wait_while_mp_kdp_trap(bool check_SIGPdebug)1332 wait_while_mp_kdp_trap(bool check_SIGPdebug)
1333 {
1334 bool found_mp_kdp_trap = false;
1335 bool found_SIGPdebug = false;
1336
1337 while (os_atomic_load_exclusive(&mp_kdp_trap, acquire) != 0) {
1338 found_mp_kdp_trap = true;
1339 if (check_SIGPdebug && cpu_has_SIGPdebug_pending()) {
1340 found_SIGPdebug = true;
1341 break;
1342 }
1343 __builtin_arm_wfe();
1344 }
1345 os_atomic_clear_exclusive();
1346
1347 if (check_SIGPdebug && found_mp_kdp_trap) {
1348 kprintf("%s>found_mp_kdp_trap=true found_SIGPdebug=%s\n", __FUNCTION__, found_SIGPdebug ? "true" : "false");
1349 }
1350 }
1351
1352 void
DebuggerXCall(void * ctx)1353 DebuggerXCall(
1354 void *ctx)
1355 {
1356 boolean_t save_context = FALSE;
1357 vm_offset_t kstackptr = 0;
1358 arm_saved_state_t *regs = (arm_saved_state_t *) ctx;
1359
1360 if (regs != NULL) {
1361 #if defined(__arm64__)
1362 current_cpu_datap()->ipi_pc = (uint64_t)get_saved_state_pc(regs);
1363 current_cpu_datap()->ipi_lr = (uint64_t)get_saved_state_lr(regs);
1364 current_cpu_datap()->ipi_fp = (uint64_t)get_saved_state_fp(regs);
1365 save_context = PSR64_IS_KERNEL(get_saved_state_cpsr(regs));
1366 #endif
1367 }
1368
1369 kstackptr = (vm_offset_t)current_thread()->machine.kstackptr;
1370
1371 #if defined(__arm64__)
1372 arm_kernel_saved_state_t *state = (arm_kernel_saved_state_t *)kstackptr;
1373
1374 if (save_context) {
1375 /* Save the interrupted context before acknowledging the signal */
1376 current_thread()->machine.kpcb = regs;
1377 } else if (regs) {
1378 /* zero old state so machine_trace_thread knows not to backtrace it */
1379 state->fp = 0;
1380 state->pc_was_in_userspace = true;
1381 state->lr = 0;
1382 state->sp = 0;
1383 state->ssbs = 0;
1384 state->uao = 0;
1385 state->dit = 0;
1386 }
1387 #endif
1388
1389 /*
1390 * When running in serial mode, the core capturing the dump may hold interrupts disabled
1391 * for a time longer than the timeout. That path includes logic to reset the timestamp
1392 * so that we do not eventually trigger the interrupt timeout assert().
1393 *
1394 * Here we check whether other cores have already gone over the timeout at this point
1395 * before spinning, so we at least cover the IPI reception path. After spinning, however,
1396 * we reset the timestamp so as to avoid hitting the interrupt timeout assert().
1397 */
1398 if ((serialmode & SERIALMODE_OUTPUT) || trap_is_stackshot) {
1399 ml_interrupt_masked_debug_end();
1400 }
1401
1402 /*
1403 * Before we decrement debugger sync, do stackshot preflight work (if applicable).
1404 * Namely, we want to signal that we're available to do stackshot work, and we need to
1405 * signal so before the stackshot-calling CPU starts work.
1406 */
1407
1408 if (trap_is_stackshot) {
1409 stackshot_cpu_preflight();
1410 }
1411
1412 os_atomic_dec(&debugger_sync, release);
1413
1414 /* If we trapped because we're doing a stackshot, do our work first. */
1415 if (trap_is_stackshot) {
1416 stackshot_aux_cpu_entry();
1417 }
1418
1419
1420 wait_while_mp_kdp_trap(false);
1421
1422 /**
1423 * Alert the triggering CPU that this CPU is done spinning. The CPU that
1424 * signalled all of the other CPUs will wait (in DebuggerXCallReturn) for
1425 * all of the CPUs to exit the above loop before continuing.
1426 */
1427 os_atomic_dec(&debug_cpus_spinning, release);
1428
1429 #if SCHED_HYGIENE_DEBUG
1430 /*
1431 * We also abandon the measurement for preemption disable
1432 * timeouts, if any. Normally, time in interrupt handlers would be
1433 * subtracted from preemption disable time, and this will happen
1434 * up to this point here, but since we here "end" the interrupt
1435 * handler prematurely (from the point of view of interrupt masked
1436 * debugging), the time spinning would otherwise still be
1437 * attributed to preemption disable time, and potentially trigger
1438 * an event, which could be a panic.
1439 */
1440 abandon_preemption_disable_measurement();
1441
1442 if ((serialmode & SERIALMODE_OUTPUT) || trap_is_stackshot) {
1443 ml_interrupt_masked_debug_start((void *)current_thread()->machine.int_handler_addr, current_thread()->machine.int_type);
1444 }
1445 #endif /* SCHED_HYGIENE_DEBUG */
1446
1447 #if defined(__arm64__)
1448 current_thread()->machine.kpcb = NULL;
1449 #endif /* defined(__arm64__) */
1450
1451 /* Any cleanup for our pushed context should go here */
1452 }
1453
1454 void
DebuggerCall(unsigned int reason,void * ctx)1455 DebuggerCall(
1456 unsigned int reason,
1457 void *ctx)
1458 {
1459 #if !MACH_KDP
1460 #pragma unused(reason,ctx)
1461 #endif /* !MACH_KDP */
1462
1463 #if ALTERNATE_DEBUGGER
1464 alternate_debugger_enter();
1465 #endif
1466
1467 #if MACH_KDP
1468 kdp_trap(reason, (struct arm_saved_state *)ctx);
1469 #else
1470 /* TODO: decide what to do if no debugger config */
1471 #endif
1472 }
1473
1474 boolean_t
bootloader_valid_page(ppnum_t ppn)1475 bootloader_valid_page(ppnum_t ppn)
1476 {
1477 return pmap_bootloader_page(ppn);
1478 }
1479