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