1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3 * Architecture specific (PPC64) functions for kexec based crash dumps.
4 *
5 * Copyright (C) 2005, IBM Corp.
6 *
7 * Created by: Haren Myneni
8 */
9
10 #include <linux/kernel.h>
11 #include <linux/smp.h>
12 #include <linux/reboot.h>
13 #include <linux/kexec.h>
14 #include <linux/export.h>
15 #include <linux/crash_dump.h>
16 #include <linux/delay.h>
17 #include <linux/irq.h>
18 #include <linux/types.h>
19 #include <linux/libfdt.h>
20 #include <linux/memory.h>
21
22 #include <asm/processor.h>
23 #include <asm/machdep.h>
24 #include <asm/kexec.h>
25 #include <asm/smp.h>
26 #include <asm/setjmp.h>
27 #include <asm/debug.h>
28 #include <asm/interrupt.h>
29 #include <asm/kexec_ranges.h>
30
31 /*
32 * The primary CPU waits a while for all secondary CPUs to enter. This is to
33 * avoid sending an IPI if the secondary CPUs are entering
34 * crash_kexec_secondary on their own (eg via a system reset).
35 *
36 * The secondary timeout has to be longer than the primary. Both timeouts are
37 * in milliseconds.
38 */
39 #define PRIMARY_TIMEOUT 500
40 #define SECONDARY_TIMEOUT 1000
41
42 #define IPI_TIMEOUT 10000
43 #define REAL_MODE_TIMEOUT 10000
44
45 static int time_to_dump;
46
47 /*
48 * In case of system reset, secondary CPUs enter crash_kexec_secondary with out
49 * having to send an IPI explicitly. So, indicate if the crash is via
50 * system reset to avoid sending another IPI.
51 */
52 static int is_via_system_reset;
53
54 /*
55 * crash_wake_offline should be set to 1 by platforms that intend to wake
56 * up offline cpus prior to jumping to a kdump kernel. Currently powernv
57 * sets it to 1, since we want to avoid things from happening when an
58 * offline CPU wakes up due to something like an HMI (malfunction error),
59 * which propagates to all threads.
60 */
61 int crash_wake_offline;
62
63 #define CRASH_HANDLER_MAX 3
64 /* List of shutdown handles */
65 static crash_shutdown_t crash_shutdown_handles[CRASH_HANDLER_MAX];
66 static DEFINE_SPINLOCK(crash_handlers_lock);
67
68 static unsigned long crash_shutdown_buf[JMP_BUF_LEN];
69 static int crash_shutdown_cpu = -1;
70
71 static int handle_fault(struct pt_regs *regs)
72 {
73 if (crash_shutdown_cpu == smp_processor_id())
74 longjmp(crash_shutdown_buf, 1);
75 return 0;
76 }
77
78 #ifdef CONFIG_SMP
79
80 static atomic_t cpus_in_crash;
81 void crash_ipi_callback(struct pt_regs *regs)
82 {
83 static cpumask_t cpus_state_saved = CPU_MASK_NONE;
84
85 int cpu = smp_processor_id();
86
87 hard_irq_disable();
88 if (!cpumask_test_cpu(cpu, &cpus_state_saved)) {
89 crash_save_cpu(regs, cpu);
90 cpumask_set_cpu(cpu, &cpus_state_saved);
91 }
92
93 atomic_inc(&cpus_in_crash);
94 smp_mb__after_atomic();
95
96 /*
97 * Starting the kdump boot.
98 * This barrier is needed to make sure that all CPUs are stopped.
99 */
100 while (!time_to_dump)
101 cpu_relax();
102
103 if (ppc_md.kexec_cpu_down)
104 ppc_md.kexec_cpu_down(1, 1);
105
106 #ifdef CONFIG_PPC64
107 kexec_smp_wait();
108 #else
109 for (;;); /* FIXME */
110 #endif
111
112 /* NOTREACHED */
113 }
114
115 static void crash_kexec_prepare_cpus(void)
116 {
117 unsigned int msecs;
118 volatile unsigned int ncpus = num_online_cpus() - 1;/* Excluding the panic cpu */
119 volatile int tries = 0;
120 int (*old_handler)(struct pt_regs *regs);
121
122 printk(KERN_EMERG "Sending IPI to other CPUs\n");
123
124 if (crash_wake_offline)
125 ncpus = num_present_cpus() - 1;
126
127 /*
128 * If we came in via system reset, secondaries enter via crash_kexec_secondary().
129 * So, wait a while for the secondary CPUs to enter for that case.
130 * Else, send IPI to all other CPUs.
131 */
132 if (is_via_system_reset)
133 mdelay(PRIMARY_TIMEOUT);
134 else
135 crash_send_ipi(crash_ipi_callback);
136 smp_wmb();
137
138 again:
139 /*
140 * FIXME: Until we will have the way to stop other CPUs reliably,
141 * the crash CPU will send an IPI and wait for other CPUs to
142 * respond.
143 */
144 msecs = IPI_TIMEOUT;
145 while ((atomic_read(&cpus_in_crash) < ncpus) && (--msecs > 0))
146 mdelay(1);
147
148 /* Would it be better to replace the trap vector here? */
149
150 if (atomic_read(&cpus_in_crash) >= ncpus) {
151 printk(KERN_EMERG "IPI complete\n");
152 return;
153 }
154
155 printk(KERN_EMERG "ERROR: %d cpu(s) not responding\n",
156 ncpus - atomic_read(&cpus_in_crash));
157
158 /*
159 * If we have a panic timeout set then we can't wait indefinitely
160 * for someone to activate system reset. We also give up on the
161 * second time through if system reset fail to work.
162 */
163 if ((panic_timeout > 0) || (tries > 0))
164 return;
165
166 /*
167 * A system reset will cause all CPUs to take an 0x100 exception.
168 * The primary CPU returns here via setjmp, and the secondary
169 * CPUs reexecute the crash_kexec_secondary path.
170 */
171 old_handler = __debugger;
172 __debugger = handle_fault;
173 crash_shutdown_cpu = smp_processor_id();
174
175 if (setjmp(crash_shutdown_buf) == 0) {
176 printk(KERN_EMERG "Activate system reset (dumprestart) "
177 "to stop other cpu(s)\n");
178
179 /*
180 * A system reset will force all CPUs to execute the
181 * crash code again. We need to reset cpus_in_crash so we
182 * wait for everyone to do this.
183 */
184 atomic_set(&cpus_in_crash, 0);
185 smp_mb();
186
187 while (atomic_read(&cpus_in_crash) < ncpus)
188 cpu_relax();
189 }
190
191 crash_shutdown_cpu = -1;
192 __debugger = old_handler;
193
194 tries++;
195 goto again;
196 }
197
198 /*
199 * This function will be called by secondary cpus.
200 */
201 void crash_kexec_secondary(struct pt_regs *regs)
202 {
203 unsigned long flags;
204 int msecs = SECONDARY_TIMEOUT;
205
206 local_irq_save(flags);
207
208 /* Wait for the primary crash CPU to signal its progress */
209 while (crashing_cpu < 0) {
210 if (--msecs < 0) {
211 /* No response, kdump image may not have been loaded */
212 local_irq_restore(flags);
213 return;
214 }
215
216 mdelay(1);
217 }
218
219 crash_ipi_callback(regs);
220 }
221
222 #else /* ! CONFIG_SMP */
223
224 static void crash_kexec_prepare_cpus(void)
225 {
226 /*
227 * move the secondaries to us so that we can copy
228 * the new kernel 0-0x100 safely
229 *
230 * do this if kexec in setup.c ?
231 */
232 #ifdef CONFIG_PPC64
233 smp_release_cpus();
234 #else
235 /* FIXME */
236 #endif
237 }
238
239 void crash_kexec_secondary(struct pt_regs *regs)
240 {
241 }
242 #endif /* CONFIG_SMP */
243
244 /* wait for all the CPUs to hit real mode but timeout if they don't come in */
245 #if defined(CONFIG_SMP) && defined(CONFIG_PPC64)
246 noinstr static void __maybe_unused crash_kexec_wait_realmode(int cpu)
247 {
248 unsigned int msecs;
249 int i;
250
251 msecs = REAL_MODE_TIMEOUT;
252 for (i=0; i < nr_cpu_ids && msecs > 0; i++) {
253 if (i == cpu)
254 continue;
255
256 while (paca_ptrs[i]->kexec_state < KEXEC_STATE_REAL_MODE) {
257 barrier();
258 if (!cpu_possible(i) || !cpu_online(i) || (msecs <= 0))
259 break;
260 msecs--;
261 mdelay(1);
262 }
263 }
264 mb();
265 }
266 #else
267 static inline void crash_kexec_wait_realmode(int cpu) {}
268 #endif /* CONFIG_SMP && CONFIG_PPC64 */
269
270 void crash_kexec_prepare(void)
271 {
272 /* Avoid hardlocking with irresponsive CPU holding logbuf_lock */
273 printk_deferred_enter();
274
275 /*
276 * This function is only called after the system
277 * has panicked or is otherwise in a critical state.
278 * The minimum amount of code to allow a kexec'd kernel
279 * to run successfully needs to happen here.
280 *
281 * In practice this means stopping other cpus in
282 * an SMP system.
283 * The kernel is broken so disable interrupts.
284 */
285 hard_irq_disable();
286
287 /*
288 * Make a note of crashing cpu. Will be used in machine_kexec
289 * such that another IPI will not be sent.
290 */
291 crashing_cpu = smp_processor_id();
292
293 crash_kexec_prepare_cpus();
294 }
295
296 /*
297 * Register a function to be called on shutdown. Only use this if you
298 * can't reset your device in the second kernel.
299 */
300 int crash_shutdown_register(crash_shutdown_t handler)
301 {
302 unsigned int i, rc;
303
304 spin_lock(&crash_handlers_lock);
305 for (i = 0 ; i < CRASH_HANDLER_MAX; i++)
306 if (!crash_shutdown_handles[i]) {
307 /* Insert handle at first empty entry */
308 crash_shutdown_handles[i] = handler;
309 rc = 0;
310 break;
311 }
312
313 if (i == CRASH_HANDLER_MAX) {
314 printk(KERN_ERR "Crash shutdown handles full, "
315 "not registered.\n");
316 rc = 1;
317 }
318
319 spin_unlock(&crash_handlers_lock);
320 return rc;
321 }
322 EXPORT_SYMBOL(crash_shutdown_register);
323
324 int crash_shutdown_unregister(crash_shutdown_t handler)
325 {
326 unsigned int i, rc;
327
328 spin_lock(&crash_handlers_lock);
329 for (i = 0 ; i < CRASH_HANDLER_MAX; i++)
330 if (crash_shutdown_handles[i] == handler)
331 break;
332
333 if (i == CRASH_HANDLER_MAX) {
334 printk(KERN_ERR "Crash shutdown handle not found\n");
335 rc = 1;
336 } else {
337 /* Shift handles down */
338 for (; i < (CRASH_HANDLER_MAX - 1); i++)
339 crash_shutdown_handles[i] =
340 crash_shutdown_handles[i+1];
341 /*
342 * Reset last entry to NULL now that it has been shifted down,
343 * this will allow new handles to be added here.
344 */
345 crash_shutdown_handles[i] = NULL;
346 rc = 0;
347 }
348
349 spin_unlock(&crash_handlers_lock);
350 return rc;
351 }
352 EXPORT_SYMBOL(crash_shutdown_unregister);
353
354 void default_machine_crash_shutdown(struct pt_regs *regs)
355 {
356 volatile unsigned int i;
357 int (*old_handler)(struct pt_regs *regs);
358
359 if (TRAP(regs) == INTERRUPT_SYSTEM_RESET)
360 is_via_system_reset = 1;
361
362 crash_smp_send_stop();
363
364 crash_save_cpu(regs, crashing_cpu);
365
366 time_to_dump = 1;
367
368 crash_kexec_wait_realmode(crashing_cpu);
369
370 machine_kexec_mask_interrupts();
371
372 /*
373 * Call registered shutdown routines safely. Swap out
374 * __debugger_fault_handler, and replace on exit.
375 */
376 old_handler = __debugger_fault_handler;
377 __debugger_fault_handler = handle_fault;
378 crash_shutdown_cpu = smp_processor_id();
379 for (i = 0; i < CRASH_HANDLER_MAX && crash_shutdown_handles[i]; i++) {
380 if (setjmp(crash_shutdown_buf) == 0) {
381 /*
382 * Insert syncs and delay to ensure
383 * instructions in the dangerous region don't
384 * leak away from this protected region.
385 */
386 asm volatile("sync; isync");
387 /* dangerous region */
388 crash_shutdown_handles[i]();
389 asm volatile("sync; isync");
390 }
391 }
392 crash_shutdown_cpu = -1;
393 __debugger_fault_handler = old_handler;
394
395 if (ppc_md.kexec_cpu_down)
396 ppc_md.kexec_cpu_down(1, 0);
397 }
398
399 #ifdef CONFIG_CRASH_HOTPLUG
400 #undef pr_fmt
401 #define pr_fmt(fmt) "crash hp: " fmt
402
403 /*
404 * Advertise preferred elfcorehdr size to userspace via
405 * /sys/kernel/crash_elfcorehdr_size sysfs interface.
406 */
407 unsigned int arch_crash_get_elfcorehdr_size(void)
408 {
409 unsigned long phdr_cnt;
410
411 /* A program header for possible CPUs + vmcoreinfo */
412 phdr_cnt = num_possible_cpus() + 1;
413 if (IS_ENABLED(CONFIG_MEMORY_HOTPLUG))
414 phdr_cnt += CONFIG_CRASH_MAX_MEMORY_RANGES;
415
416 return sizeof(struct elfhdr) + (phdr_cnt * sizeof(Elf64_Phdr));
417 }
418
419 /**
420 * update_crash_elfcorehdr() - Recreate the elfcorehdr and replace it with old
421 * elfcorehdr in the kexec segment array.
422 * @image: the active struct kimage
423 * @mn: struct memory_notify data handler
424 */
425 static void update_crash_elfcorehdr(struct kimage *image, struct memory_notify *mn)
426 {
427 int ret;
428 struct crash_mem *cmem = NULL;
429 struct kexec_segment *ksegment;
430 void *ptr, *mem, *elfbuf = NULL;
431 unsigned long elfsz, memsz, base_addr, size;
432
433 ksegment = &image->segment[image->elfcorehdr_index];
434 mem = (void *) ksegment->mem;
435 memsz = ksegment->memsz;
436
437 ret = get_crash_memory_ranges(&cmem);
438 if (ret) {
439 pr_err("Failed to get crash mem range\n");
440 return;
441 }
442
443 /*
444 * The hot unplugged memory is part of crash memory ranges,
445 * remove it here.
446 */
447 if (image->hp_action == KEXEC_CRASH_HP_REMOVE_MEMORY) {
448 base_addr = PFN_PHYS(mn->start_pfn);
449 size = mn->nr_pages * PAGE_SIZE;
450 ret = remove_mem_range(&cmem, base_addr, size);
451 if (ret) {
452 pr_err("Failed to remove hot-unplugged memory from crash memory ranges\n");
453 goto out;
454 }
455 }
456
457 ret = crash_prepare_elf64_headers(cmem, false, &elfbuf, &elfsz);
458 if (ret) {
459 pr_err("Failed to prepare elf header\n");
460 goto out;
461 }
462
463 /*
464 * It is unlikely that kernel hit this because elfcorehdr kexec
465 * segment (memsz) is built with addition space to accommodate growing
466 * number of crash memory ranges while loading the kdump kernel. It is
467 * Just to avoid any unforeseen case.
468 */
469 if (elfsz > memsz) {
470 pr_err("Updated crash elfcorehdr elfsz %lu > memsz %lu", elfsz, memsz);
471 goto out;
472 }
473
474 ptr = __va(mem);
475 if (ptr) {
476 /* Temporarily invalidate the crash image while it is replaced */
477 xchg(&kexec_crash_image, NULL);
478
479 /* Replace the old elfcorehdr with newly prepared elfcorehdr */
480 memcpy((void *)ptr, elfbuf, elfsz);
481
482 /* The crash image is now valid once again */
483 xchg(&kexec_crash_image, image);
484 }
485 out:
486 kvfree(cmem);
487 kvfree(elfbuf);
488 }
489
490 /**
491 * get_fdt_index - Loop through the kexec segment array and find
492 * the index of the FDT segment.
493 * @image: a pointer to kexec_crash_image
494 *
495 * Returns the index of FDT segment in the kexec segment array
496 * if found; otherwise -1.
497 */
498 static int get_fdt_index(struct kimage *image)
499 {
500 void *ptr;
501 unsigned long mem;
502 int i, fdt_index = -1;
503
504 /* Find the FDT segment index in kexec segment array. */
505 for (i = 0; i < image->nr_segments; i++) {
506 mem = image->segment[i].mem;
507 ptr = __va(mem);
508
509 if (ptr && fdt_magic(ptr) == FDT_MAGIC) {
510 fdt_index = i;
511 break;
512 }
513 }
514
515 return fdt_index;
516 }
517
518 /**
519 * update_crash_fdt - updates the cpus node of the crash FDT.
520 *
521 * @image: a pointer to kexec_crash_image
522 */
523 static void update_crash_fdt(struct kimage *image)
524 {
525 void *fdt;
526 int fdt_index;
527
528 fdt_index = get_fdt_index(image);
529 if (fdt_index < 0) {
530 pr_err("Unable to locate FDT segment.\n");
531 return;
532 }
533
534 fdt = __va((void *)image->segment[fdt_index].mem);
535
536 /* Temporarily invalidate the crash image while it is replaced */
537 xchg(&kexec_crash_image, NULL);
538
539 /* update FDT to reflect changes in CPU resources */
540 if (update_cpus_node(fdt))
541 pr_err("Failed to update crash FDT");
542
543 /* The crash image is now valid once again */
544 xchg(&kexec_crash_image, image);
545 }
546
547 int arch_crash_hotplug_support(struct kimage *image, unsigned long kexec_flags)
548 {
549 #ifdef CONFIG_KEXEC_FILE
550 if (image->file_mode)
551 return 1;
552 #endif
553 return kexec_flags & KEXEC_CRASH_HOTPLUG_SUPPORT;
554 }
555
556 /**
557 * arch_crash_handle_hotplug_event - Handle crash CPU/Memory hotplug events to update the
558 * necessary kexec segments based on the hotplug event.
559 * @image: a pointer to kexec_crash_image
560 * @arg: struct memory_notify handler for memory hotplug case and NULL for CPU hotplug case.
561 *
562 * Update the kdump image based on the type of hotplug event, represented by image->hp_action.
563 * CPU add: Update the FDT segment to include the newly added CPU.
564 * CPU remove: No action is needed, with the assumption that it's okay to have offline CPUs
565 * part of the FDT.
566 * Memory add/remove: No action is taken as this is not yet supported.
567 */
568 void arch_crash_handle_hotplug_event(struct kimage *image, void *arg)
569 {
570 struct memory_notify *mn;
571
572 switch (image->hp_action) {
573 case KEXEC_CRASH_HP_REMOVE_CPU:
574 return;
575
576 case KEXEC_CRASH_HP_ADD_CPU:
577 update_crash_fdt(image);
578 break;
579
580 case KEXEC_CRASH_HP_REMOVE_MEMORY:
581 case KEXEC_CRASH_HP_ADD_MEMORY:
582 mn = (struct memory_notify *)arg;
583 update_crash_elfcorehdr(image, mn);
584 return;
585 default:
586 pr_warn_once("Unknown hotplug action\n");
587 }
588 }
589 #endif /* CONFIG_CRASH_HOTPLUG */
590
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.