1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3 * Firmware Assisted dump: A robust mechanism to get reliable kernel crash
4 * dump with assistance from firmware. This approach does not use kexec,
5 * instead firmware assists in booting the kdump kernel while preserving
6 * memory contents. The most of the code implementation has been adapted
7 * from phyp assisted dump implementation written by Linas Vepstas and
8 * Manish Ahuja
9 *
10 * Copyright 2011 IBM Corporation
11 * Author: Mahesh Salgaonkar <mahesh@linux.vnet.ibm.com>
12 */
13
14 #undef DEBUG
15 #define pr_fmt(fmt) "fadump: " fmt
16
17 #include <linux/string.h>
18 #include <linux/memblock.h>
19 #include <linux/delay.h>
20 #include <linux/seq_file.h>
21 #include <linux/crash_dump.h>
22 #include <linux/kobject.h>
23 #include <linux/sysfs.h>
24 #include <linux/slab.h>
25 #include <linux/cma.h>
26 #include <linux/hugetlb.h>
27
28 #include <asm/debugfs.h>
29 #include <asm/page.h>
30 #include <asm/prom.h>
31 #include <asm/fadump.h>
32 #include <asm/fadump-internal.h>
33 #include <asm/setup.h>
34
35 /*
36 * The CPU who acquired the lock to trigger the fadump crash should
37 * wait for other CPUs to enter.
38 *
39 * The timeout is in milliseconds.
40 */
41 #define CRASH_TIMEOUT 500
42
43 static struct fw_dump fw_dump;
44
45 static void __init fadump_reserve_crash_area(u64 base);
46
47 struct kobject *fadump_kobj;
48
49 #ifndef CONFIG_PRESERVE_FA_DUMP
50
51 static atomic_t cpus_in_fadump;
52 static DEFINE_MUTEX(fadump_mutex);
53
54 struct fadump_mrange_info crash_mrange_info = { "crash", NULL, 0, 0, 0, false };
55
56 #define RESERVED_RNGS_SZ 16384 /* 16K - 128 entries */
57 #define RESERVED_RNGS_CNT (RESERVED_RNGS_SZ / \
58 sizeof(struct fadump_memory_range))
59 static struct fadump_memory_range rngs[RESERVED_RNGS_CNT];
60 struct fadump_mrange_info reserved_mrange_info = { "reserved", rngs,
61 RESERVED_RNGS_SZ, 0,
62 RESERVED_RNGS_CNT, true };
63
64 static void __init early_init_dt_scan_reserved_ranges(unsigned long node);
65
66 #ifdef CONFIG_CMA
67 static struct cma *fadump_cma;
68
69 /*
70 * fadump_cma_init() - Initialize CMA area from a fadump reserved memory
71 *
72 * This function initializes CMA area from fadump reserved memory.
73 * The total size of fadump reserved memory covers for boot memory size
74 * + cpu data size + hpte size and metadata.
75 * Initialize only the area equivalent to boot memory size for CMA use.
76 * The reamining portion of fadump reserved memory will be not given
77 * to CMA and pages for thoes will stay reserved. boot memory size is
78 * aligned per CMA requirement to satisy cma_init_reserved_mem() call.
79 * But for some reason even if it fails we still have the memory reservation
80 * with us and we can still continue doing fadump.
81 */
fadump_cma_init(void)82 int __init fadump_cma_init(void)
83 {
84 unsigned long long base, size;
85 int rc;
86
87 if (!fw_dump.fadump_enabled)
88 return 0;
89
90 /*
91 * Do not use CMA if user has provided fadump=nocma kernel parameter.
92 * Return 1 to continue with fadump old behaviour.
93 */
94 if (fw_dump.nocma)
95 return 1;
96
97 base = fw_dump.reserve_dump_area_start;
98 size = fw_dump.boot_memory_size;
99
100 if (!size)
101 return 0;
102
103 rc = cma_init_reserved_mem(base, size, 0, "fadump_cma", &fadump_cma);
104 if (rc) {
105 pr_err("Failed to init cma area for firmware-assisted dump,%d\n", rc);
106 /*
107 * Though the CMA init has failed we still have memory
108 * reservation with us. The reserved memory will be
109 * blocked from production system usage. Hence return 1,
110 * so that we can continue with fadump.
111 */
112 return 1;
113 }
114
115 /*
116 * So we now have successfully initialized cma area for fadump.
117 */
118 pr_info("Initialized 0x%lx bytes cma area at %ldMB from 0x%lx "
119 "bytes of memory reserved for firmware-assisted dump\n",
120 cma_get_size(fadump_cma),
121 (unsigned long)cma_get_base(fadump_cma) >> 20,
122 fw_dump.reserve_dump_area_size);
123 return 1;
124 }
125 #else
fadump_cma_init(void)126 static int __init fadump_cma_init(void) { return 1; }
127 #endif /* CONFIG_CMA */
128
129 /* Scan the Firmware Assisted dump configuration details. */
early_init_dt_scan_fw_dump(unsigned long node,const char * uname,int depth,void * data)130 int __init early_init_dt_scan_fw_dump(unsigned long node, const char *uname,
131 int depth, void *data)
132 {
133 if (depth == 0) {
134 early_init_dt_scan_reserved_ranges(node);
135 return 0;
136 }
137
138 if (depth != 1)
139 return 0;
140
141 if (strcmp(uname, "rtas") == 0) {
142 rtas_fadump_dt_scan(&fw_dump, node);
143 return 1;
144 }
145
146 if (strcmp(uname, "ibm,opal") == 0) {
147 opal_fadump_dt_scan(&fw_dump, node);
148 return 1;
149 }
150
151 return 0;
152 }
153
154 /*
155 * If fadump is registered, check if the memory provided
156 * falls within boot memory area and reserved memory area.
157 */
is_fadump_memory_area(u64 addr,unsigned long size)158 int is_fadump_memory_area(u64 addr, unsigned long size)
159 {
160 u64 d_start, d_end;
161
162 if (!fw_dump.dump_registered)
163 return 0;
164
165 if (!size)
166 return 0;
167
168 d_start = fw_dump.reserve_dump_area_start;
169 d_end = d_start + fw_dump.reserve_dump_area_size;
170 if (((addr + size) > d_start) && (addr <= d_end))
171 return 1;
172
173 return (addr <= fw_dump.boot_mem_top);
174 }
175
should_fadump_crash(void)176 int should_fadump_crash(void)
177 {
178 if (!fw_dump.dump_registered || !fw_dump.fadumphdr_addr)
179 return 0;
180 return 1;
181 }
182
is_fadump_active(void)183 int is_fadump_active(void)
184 {
185 return fw_dump.dump_active;
186 }
187
188 /*
189 * Returns true, if there are no holes in memory area between d_start to d_end,
190 * false otherwise.
191 */
is_fadump_mem_area_contiguous(u64 d_start,u64 d_end)192 static bool is_fadump_mem_area_contiguous(u64 d_start, u64 d_end)
193 {
194 phys_addr_t reg_start, reg_end;
195 bool ret = false;
196 u64 i, start, end;
197
198 for_each_mem_range(i, ®_start, ®_end) {
199 start = max_t(u64, d_start, reg_start);
200 end = min_t(u64, d_end, reg_end);
201 if (d_start < end) {
202 /* Memory hole from d_start to start */
203 if (start > d_start)
204 break;
205
206 if (end == d_end) {
207 ret = true;
208 break;
209 }
210
211 d_start = end + 1;
212 }
213 }
214
215 return ret;
216 }
217
218 /*
219 * Returns true, if there are no holes in boot memory area,
220 * false otherwise.
221 */
is_fadump_boot_mem_contiguous(void)222 bool is_fadump_boot_mem_contiguous(void)
223 {
224 unsigned long d_start, d_end;
225 bool ret = false;
226 int i;
227
228 for (i = 0; i < fw_dump.boot_mem_regs_cnt; i++) {
229 d_start = fw_dump.boot_mem_addr[i];
230 d_end = d_start + fw_dump.boot_mem_sz[i];
231
232 ret = is_fadump_mem_area_contiguous(d_start, d_end);
233 if (!ret)
234 break;
235 }
236
237 return ret;
238 }
239
240 /*
241 * Returns true, if there are no holes in reserved memory area,
242 * false otherwise.
243 */
is_fadump_reserved_mem_contiguous(void)244 bool is_fadump_reserved_mem_contiguous(void)
245 {
246 u64 d_start, d_end;
247
248 d_start = fw_dump.reserve_dump_area_start;
249 d_end = d_start + fw_dump.reserve_dump_area_size;
250 return is_fadump_mem_area_contiguous(d_start, d_end);
251 }
252
253 /* Print firmware assisted dump configurations for debugging purpose. */
fadump_show_config(void)254 static void fadump_show_config(void)
255 {
256 int i;
257
258 pr_debug("Support for firmware-assisted dump (fadump): %s\n",
259 (fw_dump.fadump_supported ? "present" : "no support"));
260
261 if (!fw_dump.fadump_supported)
262 return;
263
264 pr_debug("Fadump enabled : %s\n",
265 (fw_dump.fadump_enabled ? "yes" : "no"));
266 pr_debug("Dump Active : %s\n",
267 (fw_dump.dump_active ? "yes" : "no"));
268 pr_debug("Dump section sizes:\n");
269 pr_debug(" CPU state data size: %lx\n", fw_dump.cpu_state_data_size);
270 pr_debug(" HPTE region size : %lx\n", fw_dump.hpte_region_size);
271 pr_debug(" Boot memory size : %lx\n", fw_dump.boot_memory_size);
272 pr_debug(" Boot memory top : %llx\n", fw_dump.boot_mem_top);
273 pr_debug("Boot memory regions cnt: %llx\n", fw_dump.boot_mem_regs_cnt);
274 for (i = 0; i < fw_dump.boot_mem_regs_cnt; i++) {
275 pr_debug("[%03d] base = %llx, size = %llx\n", i,
276 fw_dump.boot_mem_addr[i], fw_dump.boot_mem_sz[i]);
277 }
278 }
279
280 /**
281 * fadump_calculate_reserve_size(): reserve variable boot area 5% of System RAM
282 *
283 * Function to find the largest memory size we need to reserve during early
284 * boot process. This will be the size of the memory that is required for a
285 * kernel to boot successfully.
286 *
287 * This function has been taken from phyp-assisted dump feature implementation.
288 *
289 * returns larger of 256MB or 5% rounded down to multiples of 256MB.
290 *
291 * TODO: Come up with better approach to find out more accurate memory size
292 * that is required for a kernel to boot successfully.
293 *
294 */
fadump_calculate_reserve_size(void)295 static __init u64 fadump_calculate_reserve_size(void)
296 {
297 u64 base, size, bootmem_min;
298 int ret;
299
300 if (fw_dump.reserve_bootvar)
301 pr_warn("'fadump_reserve_mem=' parameter is deprecated in favor of 'crashkernel=' parameter.\n");
302
303 /*
304 * Check if the size is specified through crashkernel= cmdline
305 * option. If yes, then use that but ignore base as fadump reserves
306 * memory at a predefined offset.
307 */
308 ret = parse_crashkernel(boot_command_line, memblock_phys_mem_size(),
309 &size, &base);
310 if (ret == 0 && size > 0) {
311 unsigned long max_size;
312
313 if (fw_dump.reserve_bootvar)
314 pr_info("Using 'crashkernel=' parameter for memory reservation.\n");
315
316 fw_dump.reserve_bootvar = (unsigned long)size;
317
318 /*
319 * Adjust if the boot memory size specified is above
320 * the upper limit.
321 */
322 max_size = memblock_phys_mem_size() / MAX_BOOT_MEM_RATIO;
323 if (fw_dump.reserve_bootvar > max_size) {
324 fw_dump.reserve_bootvar = max_size;
325 pr_info("Adjusted boot memory size to %luMB\n",
326 (fw_dump.reserve_bootvar >> 20));
327 }
328
329 return fw_dump.reserve_bootvar;
330 } else if (fw_dump.reserve_bootvar) {
331 /*
332 * 'fadump_reserve_mem=' is being used to reserve memory
333 * for firmware-assisted dump.
334 */
335 return fw_dump.reserve_bootvar;
336 }
337
338 /* divide by 20 to get 5% of value */
339 size = memblock_phys_mem_size() / 20;
340
341 /* round it down in multiples of 256 */
342 size = size & ~0x0FFFFFFFUL;
343
344 /* Truncate to memory_limit. We don't want to over reserve the memory.*/
345 if (memory_limit && size > memory_limit)
346 size = memory_limit;
347
348 bootmem_min = fw_dump.ops->fadump_get_bootmem_min();
349 return (size > bootmem_min ? size : bootmem_min);
350 }
351
352 /*
353 * Calculate the total memory size required to be reserved for
354 * firmware-assisted dump registration.
355 */
get_fadump_area_size(void)356 static unsigned long get_fadump_area_size(void)
357 {
358 unsigned long size = 0;
359
360 size += fw_dump.cpu_state_data_size;
361 size += fw_dump.hpte_region_size;
362 size += fw_dump.boot_memory_size;
363 size += sizeof(struct fadump_crash_info_header);
364 size += sizeof(struct elfhdr); /* ELF core header.*/
365 size += sizeof(struct elf_phdr); /* place holder for cpu notes */
366 /* Program headers for crash memory regions. */
367 size += sizeof(struct elf_phdr) * (memblock_num_regions(memory) + 2);
368
369 size = PAGE_ALIGN(size);
370
371 /* This is to hold kernel metadata on platforms that support it */
372 size += (fw_dump.ops->fadump_get_metadata_size ?
373 fw_dump.ops->fadump_get_metadata_size() : 0);
374 return size;
375 }
376
add_boot_mem_region(unsigned long rstart,unsigned long rsize)377 static int __init add_boot_mem_region(unsigned long rstart,
378 unsigned long rsize)
379 {
380 int i = fw_dump.boot_mem_regs_cnt++;
381
382 if (fw_dump.boot_mem_regs_cnt > FADUMP_MAX_MEM_REGS) {
383 fw_dump.boot_mem_regs_cnt = FADUMP_MAX_MEM_REGS;
384 return 0;
385 }
386
387 pr_debug("Added boot memory range[%d] [%#016lx-%#016lx)\n",
388 i, rstart, (rstart + rsize));
389 fw_dump.boot_mem_addr[i] = rstart;
390 fw_dump.boot_mem_sz[i] = rsize;
391 return 1;
392 }
393
394 /*
395 * Firmware usually has a hard limit on the data it can copy per region.
396 * Honour that by splitting a memory range into multiple regions.
397 */
add_boot_mem_regions(unsigned long mstart,unsigned long msize)398 static int __init add_boot_mem_regions(unsigned long mstart,
399 unsigned long msize)
400 {
401 unsigned long rstart, rsize, max_size;
402 int ret = 1;
403
404 rstart = mstart;
405 max_size = fw_dump.max_copy_size ? fw_dump.max_copy_size : msize;
406 while (msize) {
407 if (msize > max_size)
408 rsize = max_size;
409 else
410 rsize = msize;
411
412 ret = add_boot_mem_region(rstart, rsize);
413 if (!ret)
414 break;
415
416 msize -= rsize;
417 rstart += rsize;
418 }
419
420 return ret;
421 }
422
fadump_get_boot_mem_regions(void)423 static int __init fadump_get_boot_mem_regions(void)
424 {
425 unsigned long size, cur_size, hole_size, last_end;
426 unsigned long mem_size = fw_dump.boot_memory_size;
427 phys_addr_t reg_start, reg_end;
428 int ret = 1;
429 u64 i;
430
431 fw_dump.boot_mem_regs_cnt = 0;
432
433 last_end = 0;
434 hole_size = 0;
435 cur_size = 0;
436 for_each_mem_range(i, ®_start, ®_end) {
437 size = reg_end - reg_start;
438 hole_size += (reg_start - last_end);
439
440 if ((cur_size + size) >= mem_size) {
441 size = (mem_size - cur_size);
442 ret = add_boot_mem_regions(reg_start, size);
443 break;
444 }
445
446 mem_size -= size;
447 cur_size += size;
448 ret = add_boot_mem_regions(reg_start, size);
449 if (!ret)
450 break;
451
452 last_end = reg_end;
453 }
454 fw_dump.boot_mem_top = PAGE_ALIGN(fw_dump.boot_memory_size + hole_size);
455
456 return ret;
457 }
458
459 /*
460 * Returns true, if the given range overlaps with reserved memory ranges
461 * starting at idx. Also, updates idx to index of overlapping memory range
462 * with the given memory range.
463 * False, otherwise.
464 */
overlaps_reserved_ranges(u64 base,u64 end,int * idx)465 static bool overlaps_reserved_ranges(u64 base, u64 end, int *idx)
466 {
467 bool ret = false;
468 int i;
469
470 for (i = *idx; i < reserved_mrange_info.mem_range_cnt; i++) {
471 u64 rbase = reserved_mrange_info.mem_ranges[i].base;
472 u64 rend = rbase + reserved_mrange_info.mem_ranges[i].size;
473
474 if (end <= rbase)
475 break;
476
477 if ((end > rbase) && (base < rend)) {
478 *idx = i;
479 ret = true;
480 break;
481 }
482 }
483
484 return ret;
485 }
486
487 /*
488 * Locate a suitable memory area to reserve memory for FADump. While at it,
489 * lookup reserved-ranges & avoid overlap with them, as they are used by F/W.
490 */
fadump_locate_reserve_mem(u64 base,u64 size)491 static u64 __init fadump_locate_reserve_mem(u64 base, u64 size)
492 {
493 struct fadump_memory_range *mrngs;
494 phys_addr_t mstart, mend;
495 int idx = 0;
496 u64 i, ret = 0;
497
498 mrngs = reserved_mrange_info.mem_ranges;
499 for_each_free_mem_range(i, NUMA_NO_NODE, MEMBLOCK_NONE,
500 &mstart, &mend, NULL) {
501 pr_debug("%llu) mstart: %llx, mend: %llx, base: %llx\n",
502 i, mstart, mend, base);
503
504 if (mstart > base)
505 base = PAGE_ALIGN(mstart);
506
507 while ((mend > base) && ((mend - base) >= size)) {
508 if (!overlaps_reserved_ranges(base, base+size, &idx)) {
509 ret = base;
510 goto out;
511 }
512
513 base = mrngs[idx].base + mrngs[idx].size;
514 base = PAGE_ALIGN(base);
515 }
516 }
517
518 out:
519 return ret;
520 }
521
fadump_reserve_mem(void)522 int __init fadump_reserve_mem(void)
523 {
524 u64 base, size, mem_boundary, bootmem_min;
525 int ret = 1;
526
527 if (!fw_dump.fadump_enabled)
528 return 0;
529
530 if (!fw_dump.fadump_supported) {
531 pr_info("Firmware-Assisted Dump is not supported on this hardware\n");
532 goto error_out;
533 }
534
535 /*
536 * Initialize boot memory size
537 * If dump is active then we have already calculated the size during
538 * first kernel.
539 */
540 if (!fw_dump.dump_active) {
541 fw_dump.boot_memory_size =
542 PAGE_ALIGN(fadump_calculate_reserve_size());
543 #ifdef CONFIG_CMA
544 if (!fw_dump.nocma) {
545 fw_dump.boot_memory_size =
546 ALIGN(fw_dump.boot_memory_size,
547 FADUMP_CMA_ALIGNMENT);
548 }
549 #endif
550
551 bootmem_min = fw_dump.ops->fadump_get_bootmem_min();
552 if (fw_dump.boot_memory_size < bootmem_min) {
553 pr_err("Can't enable fadump with boot memory size (0x%lx) less than 0x%llx\n",
554 fw_dump.boot_memory_size, bootmem_min);
555 goto error_out;
556 }
557
558 if (!fadump_get_boot_mem_regions()) {
559 pr_err("Too many holes in boot memory area to enable fadump\n");
560 goto error_out;
561 }
562 }
563
564 /*
565 * Calculate the memory boundary.
566 * If memory_limit is less than actual memory boundary then reserve
567 * the memory for fadump beyond the memory_limit and adjust the
568 * memory_limit accordingly, so that the running kernel can run with
569 * specified memory_limit.
570 */
571 if (memory_limit && memory_limit < memblock_end_of_DRAM()) {
572 size = get_fadump_area_size();
573 if ((memory_limit + size) < memblock_end_of_DRAM())
574 memory_limit += size;
575 else
576 memory_limit = memblock_end_of_DRAM();
577 printk(KERN_INFO "Adjusted memory_limit for firmware-assisted"
578 " dump, now %#016llx\n", memory_limit);
579 }
580 if (memory_limit)
581 mem_boundary = memory_limit;
582 else
583 mem_boundary = memblock_end_of_DRAM();
584
585 base = fw_dump.boot_mem_top;
586 size = get_fadump_area_size();
587 fw_dump.reserve_dump_area_size = size;
588 if (fw_dump.dump_active) {
589 pr_info("Firmware-assisted dump is active.\n");
590
591 #ifdef CONFIG_HUGETLB_PAGE
592 /*
593 * FADump capture kernel doesn't care much about hugepages.
594 * In fact, handling hugepages in capture kernel is asking for
595 * trouble. So, disable HugeTLB support when fadump is active.
596 */
597 hugetlb_disabled = true;
598 #endif
599 /*
600 * If last boot has crashed then reserve all the memory
601 * above boot memory size so that we don't touch it until
602 * dump is written to disk by userspace tool. This memory
603 * can be released for general use by invalidating fadump.
604 */
605 fadump_reserve_crash_area(base);
606
607 pr_debug("fadumphdr_addr = %#016lx\n", fw_dump.fadumphdr_addr);
608 pr_debug("Reserve dump area start address: 0x%lx\n",
609 fw_dump.reserve_dump_area_start);
610 } else {
611 /*
612 * Reserve memory at an offset closer to bottom of the RAM to
613 * minimize the impact of memory hot-remove operation.
614 */
615 base = fadump_locate_reserve_mem(base, size);
616
617 if (!base || (base + size > mem_boundary)) {
618 pr_err("Failed to find memory chunk for reservation!\n");
619 goto error_out;
620 }
621 fw_dump.reserve_dump_area_start = base;
622
623 /*
624 * Calculate the kernel metadata address and register it with
625 * f/w if the platform supports.
626 */
627 if (fw_dump.ops->fadump_setup_metadata &&
628 (fw_dump.ops->fadump_setup_metadata(&fw_dump) < 0))
629 goto error_out;
630
631 if (memblock_reserve(base, size)) {
632 pr_err("Failed to reserve memory!\n");
633 goto error_out;
634 }
635
636 pr_info("Reserved %lldMB of memory at %#016llx (System RAM: %lldMB)\n",
637 (size >> 20), base, (memblock_phys_mem_size() >> 20));
638
639 ret = fadump_cma_init();
640 }
641
642 return ret;
643 error_out:
644 fw_dump.fadump_enabled = 0;
645 return 0;
646 }
647
648 /* Look for fadump= cmdline option. */
early_fadump_param(char * p)649 static int __init early_fadump_param(char *p)
650 {
651 if (!p)
652 return 1;
653
654 if (strncmp(p, "on", 2) == 0)
655 fw_dump.fadump_enabled = 1;
656 else if (strncmp(p, "off", 3) == 0)
657 fw_dump.fadump_enabled = 0;
658 else if (strncmp(p, "nocma", 5) == 0) {
659 fw_dump.fadump_enabled = 1;
660 fw_dump.nocma = 1;
661 }
662
663 return 0;
664 }
665 early_param("fadump", early_fadump_param);
666
667 /*
668 * Look for fadump_reserve_mem= cmdline option
669 * TODO: Remove references to 'fadump_reserve_mem=' parameter,
670 * the sooner 'crashkernel=' parameter is accustomed to.
671 */
early_fadump_reserve_mem(char * p)672 static int __init early_fadump_reserve_mem(char *p)
673 {
674 if (p)
675 fw_dump.reserve_bootvar = memparse(p, &p);
676 return 0;
677 }
678 early_param("fadump_reserve_mem", early_fadump_reserve_mem);
679
crash_fadump(struct pt_regs * regs,const char * str)680 void crash_fadump(struct pt_regs *regs, const char *str)
681 {
682 unsigned int msecs;
683 struct fadump_crash_info_header *fdh = NULL;
684 int old_cpu, this_cpu;
685 /* Do not include first CPU */
686 unsigned int ncpus = num_online_cpus() - 1;
687
688 if (!should_fadump_crash())
689 return;
690
691 /*
692 * old_cpu == -1 means this is the first CPU which has come here,
693 * go ahead and trigger fadump.
694 *
695 * old_cpu != -1 means some other CPU has already on it's way
696 * to trigger fadump, just keep looping here.
697 */
698 this_cpu = smp_processor_id();
699 old_cpu = cmpxchg(&crashing_cpu, -1, this_cpu);
700
701 if (old_cpu != -1) {
702 atomic_inc(&cpus_in_fadump);
703
704 /*
705 * We can't loop here indefinitely. Wait as long as fadump
706 * is in force. If we race with fadump un-registration this
707 * loop will break and then we go down to normal panic path
708 * and reboot. If fadump is in force the first crashing
709 * cpu will definitely trigger fadump.
710 */
711 while (fw_dump.dump_registered)
712 cpu_relax();
713 return;
714 }
715
716 fdh = __va(fw_dump.fadumphdr_addr);
717 fdh->crashing_cpu = crashing_cpu;
718 crash_save_vmcoreinfo();
719
720 if (regs)
721 fdh->regs = *regs;
722 else
723 ppc_save_regs(&fdh->regs);
724
725 fdh->online_mask = *cpu_online_mask;
726
727 /*
728 * If we came in via system reset, wait a while for the secondary
729 * CPUs to enter.
730 */
731 if (TRAP(&(fdh->regs)) == 0x100) {
732 msecs = CRASH_TIMEOUT;
733 while ((atomic_read(&cpus_in_fadump) < ncpus) && (--msecs > 0))
734 mdelay(1);
735 }
736
737 fw_dump.ops->fadump_trigger(fdh, str);
738 }
739
fadump_regs_to_elf_notes(u32 * buf,struct pt_regs * regs)740 u32 *fadump_regs_to_elf_notes(u32 *buf, struct pt_regs *regs)
741 {
742 struct elf_prstatus prstatus;
743
744 memset(&prstatus, 0, sizeof(prstatus));
745 /*
746 * FIXME: How do i get PID? Do I really need it?
747 * prstatus.pr_pid = ????
748 */
749 elf_core_copy_kernel_regs(&prstatus.pr_reg, regs);
750 buf = append_elf_note(buf, CRASH_CORE_NOTE_NAME, NT_PRSTATUS,
751 &prstatus, sizeof(prstatus));
752 return buf;
753 }
754
fadump_update_elfcore_header(char * bufp)755 void fadump_update_elfcore_header(char *bufp)
756 {
757 struct elf_phdr *phdr;
758
759 bufp += sizeof(struct elfhdr);
760
761 /* First note is a place holder for cpu notes info. */
762 phdr = (struct elf_phdr *)bufp;
763
764 if (phdr->p_type == PT_NOTE) {
765 phdr->p_paddr = __pa(fw_dump.cpu_notes_buf_vaddr);
766 phdr->p_offset = phdr->p_paddr;
767 phdr->p_filesz = fw_dump.cpu_notes_buf_size;
768 phdr->p_memsz = fw_dump.cpu_notes_buf_size;
769 }
770 return;
771 }
772
fadump_alloc_buffer(unsigned long size)773 static void *fadump_alloc_buffer(unsigned long size)
774 {
775 unsigned long count, i;
776 struct page *page;
777 void *vaddr;
778
779 vaddr = alloc_pages_exact(size, GFP_KERNEL | __GFP_ZERO);
780 if (!vaddr)
781 return NULL;
782
783 count = PAGE_ALIGN(size) / PAGE_SIZE;
784 page = virt_to_page(vaddr);
785 for (i = 0; i < count; i++)
786 mark_page_reserved(page + i);
787 return vaddr;
788 }
789
fadump_free_buffer(unsigned long vaddr,unsigned long size)790 static void fadump_free_buffer(unsigned long vaddr, unsigned long size)
791 {
792 free_reserved_area((void *)vaddr, (void *)(vaddr + size), -1, NULL);
793 }
794
fadump_setup_cpu_notes_buf(u32 num_cpus)795 s32 fadump_setup_cpu_notes_buf(u32 num_cpus)
796 {
797 /* Allocate buffer to hold cpu crash notes. */
798 fw_dump.cpu_notes_buf_size = num_cpus * sizeof(note_buf_t);
799 fw_dump.cpu_notes_buf_size = PAGE_ALIGN(fw_dump.cpu_notes_buf_size);
800 fw_dump.cpu_notes_buf_vaddr =
801 (unsigned long)fadump_alloc_buffer(fw_dump.cpu_notes_buf_size);
802 if (!fw_dump.cpu_notes_buf_vaddr) {
803 pr_err("Failed to allocate %ld bytes for CPU notes buffer\n",
804 fw_dump.cpu_notes_buf_size);
805 return -ENOMEM;
806 }
807
808 pr_debug("Allocated buffer for cpu notes of size %ld at 0x%lx\n",
809 fw_dump.cpu_notes_buf_size,
810 fw_dump.cpu_notes_buf_vaddr);
811 return 0;
812 }
813
fadump_free_cpu_notes_buf(void)814 void fadump_free_cpu_notes_buf(void)
815 {
816 if (!fw_dump.cpu_notes_buf_vaddr)
817 return;
818
819 fadump_free_buffer(fw_dump.cpu_notes_buf_vaddr,
820 fw_dump.cpu_notes_buf_size);
821 fw_dump.cpu_notes_buf_vaddr = 0;
822 fw_dump.cpu_notes_buf_size = 0;
823 }
824
fadump_free_mem_ranges(struct fadump_mrange_info * mrange_info)825 static void fadump_free_mem_ranges(struct fadump_mrange_info *mrange_info)
826 {
827 if (mrange_info->is_static) {
828 mrange_info->mem_range_cnt = 0;
829 return;
830 }
831
832 kfree(mrange_info->mem_ranges);
833 memset((void *)((u64)mrange_info + RNG_NAME_SZ), 0,
834 (sizeof(struct fadump_mrange_info) - RNG_NAME_SZ));
835 }
836
837 /*
838 * Allocate or reallocate mem_ranges array in incremental units
839 * of PAGE_SIZE.
840 */
fadump_alloc_mem_ranges(struct fadump_mrange_info * mrange_info)841 static int fadump_alloc_mem_ranges(struct fadump_mrange_info *mrange_info)
842 {
843 struct fadump_memory_range *new_array;
844 u64 new_size;
845
846 new_size = mrange_info->mem_ranges_sz + PAGE_SIZE;
847 pr_debug("Allocating %llu bytes of memory for %s memory ranges\n",
848 new_size, mrange_info->name);
849
850 new_array = krealloc(mrange_info->mem_ranges, new_size, GFP_KERNEL);
851 if (new_array == NULL) {
852 pr_err("Insufficient memory for setting up %s memory ranges\n",
853 mrange_info->name);
854 fadump_free_mem_ranges(mrange_info);
855 return -ENOMEM;
856 }
857
858 mrange_info->mem_ranges = new_array;
859 mrange_info->mem_ranges_sz = new_size;
860 mrange_info->max_mem_ranges = (new_size /
861 sizeof(struct fadump_memory_range));
862 return 0;
863 }
fadump_add_mem_range(struct fadump_mrange_info * mrange_info,u64 base,u64 end)864 static inline int fadump_add_mem_range(struct fadump_mrange_info *mrange_info,
865 u64 base, u64 end)
866 {
867 struct fadump_memory_range *mem_ranges = mrange_info->mem_ranges;
868 bool is_adjacent = false;
869 u64 start, size;
870
871 if (base == end)
872 return 0;
873
874 /*
875 * Fold adjacent memory ranges to bring down the memory ranges/
876 * PT_LOAD segments count.
877 */
878 if (mrange_info->mem_range_cnt) {
879 start = mem_ranges[mrange_info->mem_range_cnt - 1].base;
880 size = mem_ranges[mrange_info->mem_range_cnt - 1].size;
881
882 /*
883 * Boot memory area needs separate PT_LOAD segment(s) as it
884 * is moved to a different location at the time of crash.
885 * So, fold only if the region is not boot memory area.
886 */
887 if ((start + size) == base && start >= fw_dump.boot_mem_top)
888 is_adjacent = true;
889 }
890 if (!is_adjacent) {
891 /* resize the array on reaching the limit */
892 if (mrange_info->mem_range_cnt == mrange_info->max_mem_ranges) {
893 int ret;
894
895 if (mrange_info->is_static) {
896 pr_err("Reached array size limit for %s memory ranges\n",
897 mrange_info->name);
898 return -ENOSPC;
899 }
900
901 ret = fadump_alloc_mem_ranges(mrange_info);
902 if (ret)
903 return ret;
904
905 /* Update to the new resized array */
906 mem_ranges = mrange_info->mem_ranges;
907 }
908
909 start = base;
910 mem_ranges[mrange_info->mem_range_cnt].base = start;
911 mrange_info->mem_range_cnt++;
912 }
913
914 mem_ranges[mrange_info->mem_range_cnt - 1].size = (end - start);
915 pr_debug("%s_memory_range[%d] [%#016llx-%#016llx], %#llx bytes\n",
916 mrange_info->name, (mrange_info->mem_range_cnt - 1),
917 start, end - 1, (end - start));
918 return 0;
919 }
920
fadump_exclude_reserved_area(u64 start,u64 end)921 static int fadump_exclude_reserved_area(u64 start, u64 end)
922 {
923 u64 ra_start, ra_end;
924 int ret = 0;
925
926 ra_start = fw_dump.reserve_dump_area_start;
927 ra_end = ra_start + fw_dump.reserve_dump_area_size;
928
929 if ((ra_start < end) && (ra_end > start)) {
930 if ((start < ra_start) && (end > ra_end)) {
931 ret = fadump_add_mem_range(&crash_mrange_info,
932 start, ra_start);
933 if (ret)
934 return ret;
935
936 ret = fadump_add_mem_range(&crash_mrange_info,
937 ra_end, end);
938 } else if (start < ra_start) {
939 ret = fadump_add_mem_range(&crash_mrange_info,
940 start, ra_start);
941 } else if (ra_end < end) {
942 ret = fadump_add_mem_range(&crash_mrange_info,
943 ra_end, end);
944 }
945 } else
946 ret = fadump_add_mem_range(&crash_mrange_info, start, end);
947
948 return ret;
949 }
950
fadump_init_elfcore_header(char * bufp)951 static int fadump_init_elfcore_header(char *bufp)
952 {
953 struct elfhdr *elf;
954
955 elf = (struct elfhdr *) bufp;
956 bufp += sizeof(struct elfhdr);
957 memcpy(elf->e_ident, ELFMAG, SELFMAG);
958 elf->e_ident[EI_CLASS] = ELF_CLASS;
959 elf->e_ident[EI_DATA] = ELF_DATA;
960 elf->e_ident[EI_VERSION] = EV_CURRENT;
961 elf->e_ident[EI_OSABI] = ELF_OSABI;
962 memset(elf->e_ident+EI_PAD, 0, EI_NIDENT-EI_PAD);
963 elf->e_type = ET_CORE;
964 elf->e_machine = ELF_ARCH;
965 elf->e_version = EV_CURRENT;
966 elf->e_entry = 0;
967 elf->e_phoff = sizeof(struct elfhdr);
968 elf->e_shoff = 0;
969 #if defined(_CALL_ELF)
970 elf->e_flags = _CALL_ELF;
971 #else
972 elf->e_flags = 0;
973 #endif
974 elf->e_ehsize = sizeof(struct elfhdr);
975 elf->e_phentsize = sizeof(struct elf_phdr);
976 elf->e_phnum = 0;
977 elf->e_shentsize = 0;
978 elf->e_shnum = 0;
979 elf->e_shstrndx = 0;
980
981 return 0;
982 }
983
984 /*
985 * Traverse through memblock structure and setup crash memory ranges. These
986 * ranges will be used create PT_LOAD program headers in elfcore header.
987 */
fadump_setup_crash_memory_ranges(void)988 static int fadump_setup_crash_memory_ranges(void)
989 {
990 u64 i, start, end;
991 int ret;
992
993 pr_debug("Setup crash memory ranges.\n");
994 crash_mrange_info.mem_range_cnt = 0;
995
996 /*
997 * Boot memory region(s) registered with firmware are moved to
998 * different location at the time of crash. Create separate program
999 * header(s) for this memory chunk(s) with the correct offset.
1000 */
1001 for (i = 0; i < fw_dump.boot_mem_regs_cnt; i++) {
1002 start = fw_dump.boot_mem_addr[i];
1003 end = start + fw_dump.boot_mem_sz[i];
1004 ret = fadump_add_mem_range(&crash_mrange_info, start, end);
1005 if (ret)
1006 return ret;
1007 }
1008
1009 for_each_mem_range(i, &start, &end) {
1010 /*
1011 * skip the memory chunk that is already added
1012 * (0 through boot_memory_top).
1013 */
1014 if (start < fw_dump.boot_mem_top) {
1015 if (end > fw_dump.boot_mem_top)
1016 start = fw_dump.boot_mem_top;
1017 else
1018 continue;
1019 }
1020
1021 /* add this range excluding the reserved dump area. */
1022 ret = fadump_exclude_reserved_area(start, end);
1023 if (ret)
1024 return ret;
1025 }
1026
1027 return 0;
1028 }
1029
1030 /*
1031 * If the given physical address falls within the boot memory region then
1032 * return the relocated address that points to the dump region reserved
1033 * for saving initial boot memory contents.
1034 */
fadump_relocate(unsigned long paddr)1035 static inline unsigned long fadump_relocate(unsigned long paddr)
1036 {
1037 unsigned long raddr, rstart, rend, rlast, hole_size;
1038 int i;
1039
1040 hole_size = 0;
1041 rlast = 0;
1042 raddr = paddr;
1043 for (i = 0; i < fw_dump.boot_mem_regs_cnt; i++) {
1044 rstart = fw_dump.boot_mem_addr[i];
1045 rend = rstart + fw_dump.boot_mem_sz[i];
1046 hole_size += (rstart - rlast);
1047
1048 if (paddr >= rstart && paddr < rend) {
1049 raddr += fw_dump.boot_mem_dest_addr - hole_size;
1050 break;
1051 }
1052
1053 rlast = rend;
1054 }
1055
1056 pr_debug("vmcoreinfo: paddr = 0x%lx, raddr = 0x%lx\n", paddr, raddr);
1057 return raddr;
1058 }
1059
fadump_create_elfcore_headers(char * bufp)1060 static int fadump_create_elfcore_headers(char *bufp)
1061 {
1062 unsigned long long raddr, offset;
1063 struct elf_phdr *phdr;
1064 struct elfhdr *elf;
1065 int i, j;
1066
1067 fadump_init_elfcore_header(bufp);
1068 elf = (struct elfhdr *)bufp;
1069 bufp += sizeof(struct elfhdr);
1070
1071 /*
1072 * setup ELF PT_NOTE, place holder for cpu notes info. The notes info
1073 * will be populated during second kernel boot after crash. Hence
1074 * this PT_NOTE will always be the first elf note.
1075 *
1076 * NOTE: Any new ELF note addition should be placed after this note.
1077 */
1078 phdr = (struct elf_phdr *)bufp;
1079 bufp += sizeof(struct elf_phdr);
1080 phdr->p_type = PT_NOTE;
1081 phdr->p_flags = 0;
1082 phdr->p_vaddr = 0;
1083 phdr->p_align = 0;
1084
1085 phdr->p_offset = 0;
1086 phdr->p_paddr = 0;
1087 phdr->p_filesz = 0;
1088 phdr->p_memsz = 0;
1089
1090 (elf->e_phnum)++;
1091
1092 /* setup ELF PT_NOTE for vmcoreinfo */
1093 phdr = (struct elf_phdr *)bufp;
1094 bufp += sizeof(struct elf_phdr);
1095 phdr->p_type = PT_NOTE;
1096 phdr->p_flags = 0;
1097 phdr->p_vaddr = 0;
1098 phdr->p_align = 0;
1099
1100 phdr->p_paddr = fadump_relocate(paddr_vmcoreinfo_note());
1101 phdr->p_offset = phdr->p_paddr;
1102 phdr->p_memsz = phdr->p_filesz = VMCOREINFO_NOTE_SIZE;
1103
1104 /* Increment number of program headers. */
1105 (elf->e_phnum)++;
1106
1107 /* setup PT_LOAD sections. */
1108 j = 0;
1109 offset = 0;
1110 raddr = fw_dump.boot_mem_addr[0];
1111 for (i = 0; i < crash_mrange_info.mem_range_cnt; i++) {
1112 u64 mbase, msize;
1113
1114 mbase = crash_mrange_info.mem_ranges[i].base;
1115 msize = crash_mrange_info.mem_ranges[i].size;
1116 if (!msize)
1117 continue;
1118
1119 phdr = (struct elf_phdr *)bufp;
1120 bufp += sizeof(struct elf_phdr);
1121 phdr->p_type = PT_LOAD;
1122 phdr->p_flags = PF_R|PF_W|PF_X;
1123 phdr->p_offset = mbase;
1124
1125 if (mbase == raddr) {
1126 /*
1127 * The entire real memory region will be moved by
1128 * firmware to the specified destination_address.
1129 * Hence set the correct offset.
1130 */
1131 phdr->p_offset = fw_dump.boot_mem_dest_addr + offset;
1132 if (j < (fw_dump.boot_mem_regs_cnt - 1)) {
1133 offset += fw_dump.boot_mem_sz[j];
1134 raddr = fw_dump.boot_mem_addr[++j];
1135 }
1136 }
1137
1138 phdr->p_paddr = mbase;
1139 phdr->p_vaddr = (unsigned long)__va(mbase);
1140 phdr->p_filesz = msize;
1141 phdr->p_memsz = msize;
1142 phdr->p_align = 0;
1143
1144 /* Increment number of program headers. */
1145 (elf->e_phnum)++;
1146 }
1147 return 0;
1148 }
1149
init_fadump_header(unsigned long addr)1150 static unsigned long init_fadump_header(unsigned long addr)
1151 {
1152 struct fadump_crash_info_header *fdh;
1153
1154 if (!addr)
1155 return 0;
1156
1157 fdh = __va(addr);
1158 addr += sizeof(struct fadump_crash_info_header);
1159
1160 memset(fdh, 0, sizeof(struct fadump_crash_info_header));
1161 fdh->magic_number = FADUMP_CRASH_INFO_MAGIC;
1162 fdh->elfcorehdr_addr = addr;
1163 /* We will set the crashing cpu id in crash_fadump() during crash. */
1164 fdh->crashing_cpu = FADUMP_CPU_UNKNOWN;
1165
1166 return addr;
1167 }
1168
register_fadump(void)1169 static int register_fadump(void)
1170 {
1171 unsigned long addr;
1172 void *vaddr;
1173 int ret;
1174
1175 /*
1176 * If no memory is reserved then we can not register for firmware-
1177 * assisted dump.
1178 */
1179 if (!fw_dump.reserve_dump_area_size)
1180 return -ENODEV;
1181
1182 ret = fadump_setup_crash_memory_ranges();
1183 if (ret)
1184 return ret;
1185
1186 addr = fw_dump.fadumphdr_addr;
1187
1188 /* Initialize fadump crash info header. */
1189 addr = init_fadump_header(addr);
1190 vaddr = __va(addr);
1191
1192 pr_debug("Creating ELF core headers at %#016lx\n", addr);
1193 fadump_create_elfcore_headers(vaddr);
1194
1195 /* register the future kernel dump with firmware. */
1196 pr_debug("Registering for firmware-assisted kernel dump...\n");
1197 return fw_dump.ops->fadump_register(&fw_dump);
1198 }
1199
fadump_cleanup(void)1200 void fadump_cleanup(void)
1201 {
1202 if (!fw_dump.fadump_supported)
1203 return;
1204
1205 /* Invalidate the registration only if dump is active. */
1206 if (fw_dump.dump_active) {
1207 pr_debug("Invalidating firmware-assisted dump registration\n");
1208 fw_dump.ops->fadump_invalidate(&fw_dump);
1209 } else if (fw_dump.dump_registered) {
1210 /* Un-register Firmware-assisted dump if it was registered. */
1211 fw_dump.ops->fadump_unregister(&fw_dump);
1212 fadump_free_mem_ranges(&crash_mrange_info);
1213 }
1214
1215 if (fw_dump.ops->fadump_cleanup)
1216 fw_dump.ops->fadump_cleanup(&fw_dump);
1217 }
1218
fadump_free_reserved_memory(unsigned long start_pfn,unsigned long end_pfn)1219 static void fadump_free_reserved_memory(unsigned long start_pfn,
1220 unsigned long end_pfn)
1221 {
1222 unsigned long pfn;
1223 unsigned long time_limit = jiffies + HZ;
1224
1225 pr_info("freeing reserved memory (0x%llx - 0x%llx)\n",
1226 PFN_PHYS(start_pfn), PFN_PHYS(end_pfn));
1227
1228 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
1229 free_reserved_page(pfn_to_page(pfn));
1230
1231 if (time_after(jiffies, time_limit)) {
1232 cond_resched();
1233 time_limit = jiffies + HZ;
1234 }
1235 }
1236 }
1237
1238 /*
1239 * Skip memory holes and free memory that was actually reserved.
1240 */
fadump_release_reserved_area(u64 start,u64 end)1241 static void fadump_release_reserved_area(u64 start, u64 end)
1242 {
1243 unsigned long reg_spfn, reg_epfn;
1244 u64 tstart, tend, spfn, epfn;
1245 int i;
1246
1247 spfn = PHYS_PFN(start);
1248 epfn = PHYS_PFN(end);
1249
1250 for_each_mem_pfn_range(i, MAX_NUMNODES, ®_spfn, ®_epfn, NULL) {
1251 tstart = max_t(u64, spfn, reg_spfn);
1252 tend = min_t(u64, epfn, reg_epfn);
1253
1254 if (tstart < tend) {
1255 fadump_free_reserved_memory(tstart, tend);
1256
1257 if (tend == epfn)
1258 break;
1259
1260 spfn = tend;
1261 }
1262 }
1263 }
1264
1265 /*
1266 * Sort the mem ranges in-place and merge adjacent ranges
1267 * to minimize the memory ranges count.
1268 */
sort_and_merge_mem_ranges(struct fadump_mrange_info * mrange_info)1269 static void sort_and_merge_mem_ranges(struct fadump_mrange_info *mrange_info)
1270 {
1271 struct fadump_memory_range *mem_ranges;
1272 struct fadump_memory_range tmp_range;
1273 u64 base, size;
1274 int i, j, idx;
1275
1276 if (!reserved_mrange_info.mem_range_cnt)
1277 return;
1278
1279 /* Sort the memory ranges */
1280 mem_ranges = mrange_info->mem_ranges;
1281 for (i = 0; i < mrange_info->mem_range_cnt; i++) {
1282 idx = i;
1283 for (j = (i + 1); j < mrange_info->mem_range_cnt; j++) {
1284 if (mem_ranges[idx].base > mem_ranges[j].base)
1285 idx = j;
1286 }
1287 if (idx != i) {
1288 tmp_range = mem_ranges[idx];
1289 mem_ranges[idx] = mem_ranges[i];
1290 mem_ranges[i] = tmp_range;
1291 }
1292 }
1293
1294 /* Merge adjacent reserved ranges */
1295 idx = 0;
1296 for (i = 1; i < mrange_info->mem_range_cnt; i++) {
1297 base = mem_ranges[i-1].base;
1298 size = mem_ranges[i-1].size;
1299 if (mem_ranges[i].base == (base + size))
1300 mem_ranges[idx].size += mem_ranges[i].size;
1301 else {
1302 idx++;
1303 if (i == idx)
1304 continue;
1305
1306 mem_ranges[idx] = mem_ranges[i];
1307 }
1308 }
1309 mrange_info->mem_range_cnt = idx + 1;
1310 }
1311
1312 /*
1313 * Scan reserved-ranges to consider them while reserving/releasing
1314 * memory for FADump.
1315 */
early_init_dt_scan_reserved_ranges(unsigned long node)1316 static void __init early_init_dt_scan_reserved_ranges(unsigned long node)
1317 {
1318 const __be32 *prop;
1319 int len, ret = -1;
1320 unsigned long i;
1321
1322 /* reserved-ranges already scanned */
1323 if (reserved_mrange_info.mem_range_cnt != 0)
1324 return;
1325
1326 prop = of_get_flat_dt_prop(node, "reserved-ranges", &len);
1327 if (!prop)
1328 return;
1329
1330 /*
1331 * Each reserved range is an (address,size) pair, 2 cells each,
1332 * totalling 4 cells per range.
1333 */
1334 for (i = 0; i < len / (sizeof(*prop) * 4); i++) {
1335 u64 base, size;
1336
1337 base = of_read_number(prop + (i * 4) + 0, 2);
1338 size = of_read_number(prop + (i * 4) + 2, 2);
1339
1340 if (size) {
1341 ret = fadump_add_mem_range(&reserved_mrange_info,
1342 base, base + size);
1343 if (ret < 0) {
1344 pr_warn("some reserved ranges are ignored!\n");
1345 break;
1346 }
1347 }
1348 }
1349
1350 /* Compact reserved ranges */
1351 sort_and_merge_mem_ranges(&reserved_mrange_info);
1352 }
1353
1354 /*
1355 * Release the memory that was reserved during early boot to preserve the
1356 * crash'ed kernel's memory contents except reserved dump area (permanent
1357 * reservation) and reserved ranges used by F/W. The released memory will
1358 * be available for general use.
1359 */
fadump_release_memory(u64 begin,u64 end)1360 static void fadump_release_memory(u64 begin, u64 end)
1361 {
1362 u64 ra_start, ra_end, tstart;
1363 int i, ret;
1364
1365 ra_start = fw_dump.reserve_dump_area_start;
1366 ra_end = ra_start + fw_dump.reserve_dump_area_size;
1367
1368 /*
1369 * If reserved ranges array limit is hit, overwrite the last reserved
1370 * memory range with reserved dump area to ensure it is excluded from
1371 * the memory being released (reused for next FADump registration).
1372 */
1373 if (reserved_mrange_info.mem_range_cnt ==
1374 reserved_mrange_info.max_mem_ranges)
1375 reserved_mrange_info.mem_range_cnt--;
1376
1377 ret = fadump_add_mem_range(&reserved_mrange_info, ra_start, ra_end);
1378 if (ret != 0)
1379 return;
1380
1381 /* Get the reserved ranges list in order first. */
1382 sort_and_merge_mem_ranges(&reserved_mrange_info);
1383
1384 /* Exclude reserved ranges and release remaining memory */
1385 tstart = begin;
1386 for (i = 0; i < reserved_mrange_info.mem_range_cnt; i++) {
1387 ra_start = reserved_mrange_info.mem_ranges[i].base;
1388 ra_end = ra_start + reserved_mrange_info.mem_ranges[i].size;
1389
1390 if (tstart >= ra_end)
1391 continue;
1392
1393 if (tstart < ra_start)
1394 fadump_release_reserved_area(tstart, ra_start);
1395 tstart = ra_end;
1396 }
1397
1398 if (tstart < end)
1399 fadump_release_reserved_area(tstart, end);
1400 }
1401
fadump_invalidate_release_mem(void)1402 static void fadump_invalidate_release_mem(void)
1403 {
1404 mutex_lock(&fadump_mutex);
1405 if (!fw_dump.dump_active) {
1406 mutex_unlock(&fadump_mutex);
1407 return;
1408 }
1409
1410 fadump_cleanup();
1411 mutex_unlock(&fadump_mutex);
1412
1413 fadump_release_memory(fw_dump.boot_mem_top, memblock_end_of_DRAM());
1414 fadump_free_cpu_notes_buf();
1415
1416 /*
1417 * Setup kernel metadata and initialize the kernel dump
1418 * memory structure for FADump re-registration.
1419 */
1420 if (fw_dump.ops->fadump_setup_metadata &&
1421 (fw_dump.ops->fadump_setup_metadata(&fw_dump) < 0))
1422 pr_warn("Failed to setup kernel metadata!\n");
1423 fw_dump.ops->fadump_init_mem_struct(&fw_dump);
1424 }
1425
release_mem_store(struct kobject * kobj,struct kobj_attribute * attr,const char * buf,size_t count)1426 static ssize_t release_mem_store(struct kobject *kobj,
1427 struct kobj_attribute *attr,
1428 const char *buf, size_t count)
1429 {
1430 int input = -1;
1431
1432 if (!fw_dump.dump_active)
1433 return -EPERM;
1434
1435 if (kstrtoint(buf, 0, &input))
1436 return -EINVAL;
1437
1438 if (input == 1) {
1439 /*
1440 * Take away the '/proc/vmcore'. We are releasing the dump
1441 * memory, hence it will not be valid anymore.
1442 */
1443 #ifdef CONFIG_PROC_VMCORE
1444 vmcore_cleanup();
1445 #endif
1446 fadump_invalidate_release_mem();
1447
1448 } else
1449 return -EINVAL;
1450 return count;
1451 }
1452
1453 /* Release the reserved memory and disable the FADump */
unregister_fadump(void)1454 static void unregister_fadump(void)
1455 {
1456 fadump_cleanup();
1457 fadump_release_memory(fw_dump.reserve_dump_area_start,
1458 fw_dump.reserve_dump_area_size);
1459 fw_dump.fadump_enabled = 0;
1460 kobject_put(fadump_kobj);
1461 }
1462
enabled_show(struct kobject * kobj,struct kobj_attribute * attr,char * buf)1463 static ssize_t enabled_show(struct kobject *kobj,
1464 struct kobj_attribute *attr,
1465 char *buf)
1466 {
1467 return sprintf(buf, "%d\n", fw_dump.fadump_enabled);
1468 }
1469
mem_reserved_show(struct kobject * kobj,struct kobj_attribute * attr,char * buf)1470 static ssize_t mem_reserved_show(struct kobject *kobj,
1471 struct kobj_attribute *attr,
1472 char *buf)
1473 {
1474 return sprintf(buf, "%ld\n", fw_dump.reserve_dump_area_size);
1475 }
1476
registered_show(struct kobject * kobj,struct kobj_attribute * attr,char * buf)1477 static ssize_t registered_show(struct kobject *kobj,
1478 struct kobj_attribute *attr,
1479 char *buf)
1480 {
1481 return sprintf(buf, "%d\n", fw_dump.dump_registered);
1482 }
1483
registered_store(struct kobject * kobj,struct kobj_attribute * attr,const char * buf,size_t count)1484 static ssize_t registered_store(struct kobject *kobj,
1485 struct kobj_attribute *attr,
1486 const char *buf, size_t count)
1487 {
1488 int ret = 0;
1489 int input = -1;
1490
1491 if (!fw_dump.fadump_enabled || fw_dump.dump_active)
1492 return -EPERM;
1493
1494 if (kstrtoint(buf, 0, &input))
1495 return -EINVAL;
1496
1497 mutex_lock(&fadump_mutex);
1498
1499 switch (input) {
1500 case 0:
1501 if (fw_dump.dump_registered == 0) {
1502 goto unlock_out;
1503 }
1504
1505 /* Un-register Firmware-assisted dump */
1506 pr_debug("Un-register firmware-assisted dump\n");
1507 fw_dump.ops->fadump_unregister(&fw_dump);
1508 break;
1509 case 1:
1510 if (fw_dump.dump_registered == 1) {
1511 /* Un-register Firmware-assisted dump */
1512 fw_dump.ops->fadump_unregister(&fw_dump);
1513 }
1514 /* Register Firmware-assisted dump */
1515 ret = register_fadump();
1516 break;
1517 default:
1518 ret = -EINVAL;
1519 break;
1520 }
1521
1522 unlock_out:
1523 mutex_unlock(&fadump_mutex);
1524 return ret < 0 ? ret : count;
1525 }
1526
fadump_region_show(struct seq_file * m,void * private)1527 static int fadump_region_show(struct seq_file *m, void *private)
1528 {
1529 if (!fw_dump.fadump_enabled)
1530 return 0;
1531
1532 mutex_lock(&fadump_mutex);
1533 fw_dump.ops->fadump_region_show(&fw_dump, m);
1534 mutex_unlock(&fadump_mutex);
1535 return 0;
1536 }
1537
1538 static struct kobj_attribute release_attr = __ATTR_WO(release_mem);
1539 static struct kobj_attribute enable_attr = __ATTR_RO(enabled);
1540 static struct kobj_attribute register_attr = __ATTR_RW(registered);
1541 static struct kobj_attribute mem_reserved_attr = __ATTR_RO(mem_reserved);
1542
1543 static struct attribute *fadump_attrs[] = {
1544 &enable_attr.attr,
1545 ®ister_attr.attr,
1546 &mem_reserved_attr.attr,
1547 NULL,
1548 };
1549
1550 ATTRIBUTE_GROUPS(fadump);
1551
1552 DEFINE_SHOW_ATTRIBUTE(fadump_region);
1553
fadump_init_files(void)1554 static void fadump_init_files(void)
1555 {
1556 int rc = 0;
1557
1558 fadump_kobj = kobject_create_and_add("fadump", kernel_kobj);
1559 if (!fadump_kobj) {
1560 pr_err("failed to create fadump kobject\n");
1561 return;
1562 }
1563
1564 debugfs_create_file("fadump_region", 0444, powerpc_debugfs_root, NULL,
1565 &fadump_region_fops);
1566
1567 if (fw_dump.dump_active) {
1568 rc = sysfs_create_file(fadump_kobj, &release_attr.attr);
1569 if (rc)
1570 pr_err("unable to create release_mem sysfs file (%d)\n",
1571 rc);
1572 }
1573
1574 rc = sysfs_create_groups(fadump_kobj, fadump_groups);
1575 if (rc) {
1576 pr_err("sysfs group creation failed (%d), unregistering FADump",
1577 rc);
1578 unregister_fadump();
1579 return;
1580 }
1581
1582 /*
1583 * The FADump sysfs are moved from kernel_kobj to fadump_kobj need to
1584 * create symlink at old location to maintain backward compatibility.
1585 *
1586 * - fadump_enabled -> fadump/enabled
1587 * - fadump_registered -> fadump/registered
1588 * - fadump_release_mem -> fadump/release_mem
1589 */
1590 rc = compat_only_sysfs_link_entry_to_kobj(kernel_kobj, fadump_kobj,
1591 "enabled", "fadump_enabled");
1592 if (rc) {
1593 pr_err("unable to create fadump_enabled symlink (%d)", rc);
1594 return;
1595 }
1596
1597 rc = compat_only_sysfs_link_entry_to_kobj(kernel_kobj, fadump_kobj,
1598 "registered",
1599 "fadump_registered");
1600 if (rc) {
1601 pr_err("unable to create fadump_registered symlink (%d)", rc);
1602 sysfs_remove_link(kernel_kobj, "fadump_enabled");
1603 return;
1604 }
1605
1606 if (fw_dump.dump_active) {
1607 rc = compat_only_sysfs_link_entry_to_kobj(kernel_kobj,
1608 fadump_kobj,
1609 "release_mem",
1610 "fadump_release_mem");
1611 if (rc)
1612 pr_err("unable to create fadump_release_mem symlink (%d)",
1613 rc);
1614 }
1615 return;
1616 }
1617
1618 /*
1619 * Prepare for firmware-assisted dump.
1620 */
setup_fadump(void)1621 int __init setup_fadump(void)
1622 {
1623 if (!fw_dump.fadump_supported)
1624 return 0;
1625
1626 fadump_init_files();
1627 fadump_show_config();
1628
1629 if (!fw_dump.fadump_enabled)
1630 return 1;
1631
1632 /*
1633 * If dump data is available then see if it is valid and prepare for
1634 * saving it to the disk.
1635 */
1636 if (fw_dump.dump_active) {
1637 /*
1638 * if dump process fails then invalidate the registration
1639 * and release memory before proceeding for re-registration.
1640 */
1641 if (fw_dump.ops->fadump_process(&fw_dump) < 0)
1642 fadump_invalidate_release_mem();
1643 }
1644 /* Initialize the kernel dump memory structure for FAD registration. */
1645 else if (fw_dump.reserve_dump_area_size)
1646 fw_dump.ops->fadump_init_mem_struct(&fw_dump);
1647
1648 /*
1649 * In case of panic, fadump is triggered via ppc_panic_event()
1650 * panic notifier. Setting crash_kexec_post_notifiers to 'true'
1651 * lets panic() function take crash friendly path before panic
1652 * notifiers are invoked.
1653 */
1654 crash_kexec_post_notifiers = true;
1655
1656 return 1;
1657 }
1658 subsys_initcall(setup_fadump);
1659 #else /* !CONFIG_PRESERVE_FA_DUMP */
1660
1661 /* Scan the Firmware Assisted dump configuration details. */
early_init_dt_scan_fw_dump(unsigned long node,const char * uname,int depth,void * data)1662 int __init early_init_dt_scan_fw_dump(unsigned long node, const char *uname,
1663 int depth, void *data)
1664 {
1665 if ((depth != 1) || (strcmp(uname, "ibm,opal") != 0))
1666 return 0;
1667
1668 opal_fadump_dt_scan(&fw_dump, node);
1669 return 1;
1670 }
1671
1672 /*
1673 * When dump is active but PRESERVE_FA_DUMP is enabled on the kernel,
1674 * preserve crash data. The subsequent memory preserving kernel boot
1675 * is likely to process this crash data.
1676 */
fadump_reserve_mem(void)1677 int __init fadump_reserve_mem(void)
1678 {
1679 if (fw_dump.dump_active) {
1680 /*
1681 * If last boot has crashed then reserve all the memory
1682 * above boot memory to preserve crash data.
1683 */
1684 pr_info("Preserving crash data for processing in next boot.\n");
1685 fadump_reserve_crash_area(fw_dump.boot_mem_top);
1686 } else
1687 pr_debug("FADump-aware kernel..\n");
1688
1689 return 1;
1690 }
1691 #endif /* CONFIG_PRESERVE_FA_DUMP */
1692
1693 /* Preserve everything above the base address */
fadump_reserve_crash_area(u64 base)1694 static void __init fadump_reserve_crash_area(u64 base)
1695 {
1696 u64 i, mstart, mend, msize;
1697
1698 for_each_mem_range(i, &mstart, &mend) {
1699 msize = mend - mstart;
1700
1701 if ((mstart + msize) < base)
1702 continue;
1703
1704 if (mstart < base) {
1705 msize -= (base - mstart);
1706 mstart = base;
1707 }
1708
1709 pr_info("Reserving %lluMB of memory at %#016llx for preserving crash data",
1710 (msize >> 20), mstart);
1711 memblock_reserve(mstart, msize);
1712 }
1713 }
1714
arch_reserved_kernel_pages(void)1715 unsigned long __init arch_reserved_kernel_pages(void)
1716 {
1717 return memblock_reserved_size() / PAGE_SIZE;
1718 }
1719