Loading...
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#include <linux/debugfs.h>
28#include <linux/of.h>
29#include <linux/of_fdt.h>
30
31#include <asm/page.h>
32#include <asm/fadump.h>
33#include <asm/fadump-internal.h>
34#include <asm/setup.h>
35#include <asm/interrupt.h>
36
37/*
38 * The CPU who acquired the lock to trigger the fadump crash should
39 * wait for other CPUs to enter.
40 *
41 * The timeout is in milliseconds.
42 */
43#define CRASH_TIMEOUT 500
44
45static struct fw_dump fw_dump;
46
47static void __init fadump_reserve_crash_area(u64 base);
48
49#ifndef CONFIG_PRESERVE_FA_DUMP
50
51static struct kobject *fadump_kobj;
52
53static atomic_t cpus_in_fadump;
54static DEFINE_MUTEX(fadump_mutex);
55
56static struct fadump_mrange_info crash_mrange_info = { "crash", NULL, 0, 0, 0, false };
57
58#define RESERVED_RNGS_SZ 16384 /* 16K - 128 entries */
59#define RESERVED_RNGS_CNT (RESERVED_RNGS_SZ / \
60 sizeof(struct fadump_memory_range))
61static struct fadump_memory_range rngs[RESERVED_RNGS_CNT];
62static struct fadump_mrange_info
63reserved_mrange_info = { "reserved", rngs, RESERVED_RNGS_SZ, 0, RESERVED_RNGS_CNT, true };
64
65static void __init early_init_dt_scan_reserved_ranges(unsigned long node);
66
67#ifdef CONFIG_CMA
68static struct cma *fadump_cma;
69
70/*
71 * fadump_cma_init() - Initialize CMA area from a fadump reserved memory
72 *
73 * This function initializes CMA area from fadump reserved memory.
74 * The total size of fadump reserved memory covers for boot memory size
75 * + cpu data size + hpte size and metadata.
76 * Initialize only the area equivalent to boot memory size for CMA use.
77 * The remaining portion of fadump reserved memory will be not given
78 * to CMA and pages for those will stay reserved. boot memory size is
79 * aligned per CMA requirement to satisy cma_init_reserved_mem() call.
80 * But for some reason even if it fails we still have the memory reservation
81 * with us and we can still continue doing fadump.
82 */
83static int __init fadump_cma_init(void)
84{
85 unsigned long long base, size;
86 int rc;
87
88 if (!fw_dump.fadump_enabled)
89 return 0;
90
91 /*
92 * Do not use CMA if user has provided fadump=nocma kernel parameter.
93 * Return 1 to continue with fadump old behaviour.
94 */
95 if (fw_dump.nocma)
96 return 1;
97
98 base = fw_dump.reserve_dump_area_start;
99 size = fw_dump.boot_memory_size;
100
101 if (!size)
102 return 0;
103
104 rc = cma_init_reserved_mem(base, size, 0, "fadump_cma", &fadump_cma);
105 if (rc) {
106 pr_err("Failed to init cma area for firmware-assisted dump,%d\n", rc);
107 /*
108 * Though the CMA init has failed we still have memory
109 * reservation with us. The reserved memory will be
110 * blocked from production system usage. Hence return 1,
111 * so that we can continue with fadump.
112 */
113 return 1;
114 }
115
116 /*
117 * If CMA activation fails, keep the pages reserved, instead of
118 * exposing them to buddy allocator. Same as 'fadump=nocma' case.
119 */
120 cma_reserve_pages_on_error(fadump_cma);
121
122 /*
123 * So we now have successfully initialized cma area for fadump.
124 */
125 pr_info("Initialized 0x%lx bytes cma area at %ldMB from 0x%lx "
126 "bytes of memory reserved for firmware-assisted dump\n",
127 cma_get_size(fadump_cma),
128 (unsigned long)cma_get_base(fadump_cma) >> 20,
129 fw_dump.reserve_dump_area_size);
130 return 1;
131}
132#else
133static int __init fadump_cma_init(void) { return 1; }
134#endif /* CONFIG_CMA */
135
136/* Scan the Firmware Assisted dump configuration details. */
137int __init early_init_dt_scan_fw_dump(unsigned long node, const char *uname,
138 int depth, void *data)
139{
140 if (depth == 0) {
141 early_init_dt_scan_reserved_ranges(node);
142 return 0;
143 }
144
145 if (depth != 1)
146 return 0;
147
148 if (strcmp(uname, "rtas") == 0) {
149 rtas_fadump_dt_scan(&fw_dump, node);
150 return 1;
151 }
152
153 if (strcmp(uname, "ibm,opal") == 0) {
154 opal_fadump_dt_scan(&fw_dump, node);
155 return 1;
156 }
157
158 return 0;
159}
160
161/*
162 * If fadump is registered, check if the memory provided
163 * falls within boot memory area and reserved memory area.
164 */
165int is_fadump_memory_area(u64 addr, unsigned long size)
166{
167 u64 d_start, d_end;
168
169 if (!fw_dump.dump_registered)
170 return 0;
171
172 if (!size)
173 return 0;
174
175 d_start = fw_dump.reserve_dump_area_start;
176 d_end = d_start + fw_dump.reserve_dump_area_size;
177 if (((addr + size) > d_start) && (addr <= d_end))
178 return 1;
179
180 return (addr <= fw_dump.boot_mem_top);
181}
182
183int should_fadump_crash(void)
184{
185 if (!fw_dump.dump_registered || !fw_dump.fadumphdr_addr)
186 return 0;
187 return 1;
188}
189
190int is_fadump_active(void)
191{
192 return fw_dump.dump_active;
193}
194
195/*
196 * Returns true, if there are no holes in memory area between d_start to d_end,
197 * false otherwise.
198 */
199static bool is_fadump_mem_area_contiguous(u64 d_start, u64 d_end)
200{
201 phys_addr_t reg_start, reg_end;
202 bool ret = false;
203 u64 i, start, end;
204
205 for_each_mem_range(i, ®_start, ®_end) {
206 start = max_t(u64, d_start, reg_start);
207 end = min_t(u64, d_end, reg_end);
208 if (d_start < end) {
209 /* Memory hole from d_start to start */
210 if (start > d_start)
211 break;
212
213 if (end == d_end) {
214 ret = true;
215 break;
216 }
217
218 d_start = end + 1;
219 }
220 }
221
222 return ret;
223}
224
225/*
226 * Returns true, if there are no holes in boot memory area,
227 * false otherwise.
228 */
229bool is_fadump_boot_mem_contiguous(void)
230{
231 unsigned long d_start, d_end;
232 bool ret = false;
233 int i;
234
235 for (i = 0; i < fw_dump.boot_mem_regs_cnt; i++) {
236 d_start = fw_dump.boot_mem_addr[i];
237 d_end = d_start + fw_dump.boot_mem_sz[i];
238
239 ret = is_fadump_mem_area_contiguous(d_start, d_end);
240 if (!ret)
241 break;
242 }
243
244 return ret;
245}
246
247/*
248 * Returns true, if there are no holes in reserved memory area,
249 * false otherwise.
250 */
251bool is_fadump_reserved_mem_contiguous(void)
252{
253 u64 d_start, d_end;
254
255 d_start = fw_dump.reserve_dump_area_start;
256 d_end = d_start + fw_dump.reserve_dump_area_size;
257 return is_fadump_mem_area_contiguous(d_start, d_end);
258}
259
260/* Print firmware assisted dump configurations for debugging purpose. */
261static void __init fadump_show_config(void)
262{
263 int i;
264
265 pr_debug("Support for firmware-assisted dump (fadump): %s\n",
266 (fw_dump.fadump_supported ? "present" : "no support"));
267
268 if (!fw_dump.fadump_supported)
269 return;
270
271 pr_debug("Fadump enabled : %s\n",
272 (fw_dump.fadump_enabled ? "yes" : "no"));
273 pr_debug("Dump Active : %s\n",
274 (fw_dump.dump_active ? "yes" : "no"));
275 pr_debug("Dump section sizes:\n");
276 pr_debug(" CPU state data size: %lx\n", fw_dump.cpu_state_data_size);
277 pr_debug(" HPTE region size : %lx\n", fw_dump.hpte_region_size);
278 pr_debug(" Boot memory size : %lx\n", fw_dump.boot_memory_size);
279 pr_debug(" Boot memory top : %llx\n", fw_dump.boot_mem_top);
280 pr_debug("Boot memory regions cnt: %llx\n", fw_dump.boot_mem_regs_cnt);
281 for (i = 0; i < fw_dump.boot_mem_regs_cnt; i++) {
282 pr_debug("[%03d] base = %llx, size = %llx\n", i,
283 fw_dump.boot_mem_addr[i], fw_dump.boot_mem_sz[i]);
284 }
285}
286
287/**
288 * fadump_calculate_reserve_size(): reserve variable boot area 5% of System RAM
289 *
290 * Function to find the largest memory size we need to reserve during early
291 * boot process. This will be the size of the memory that is required for a
292 * kernel to boot successfully.
293 *
294 * This function has been taken from phyp-assisted dump feature implementation.
295 *
296 * returns larger of 256MB or 5% rounded down to multiples of 256MB.
297 *
298 * TODO: Come up with better approach to find out more accurate memory size
299 * that is required for a kernel to boot successfully.
300 *
301 */
302static __init u64 fadump_calculate_reserve_size(void)
303{
304 u64 base, size, bootmem_min;
305 int ret;
306
307 if (fw_dump.reserve_bootvar)
308 pr_warn("'fadump_reserve_mem=' parameter is deprecated in favor of 'crashkernel=' parameter.\n");
309
310 /*
311 * Check if the size is specified through crashkernel= cmdline
312 * option. If yes, then use that but ignore base as fadump reserves
313 * memory at a predefined offset.
314 */
315 ret = parse_crashkernel(boot_command_line, memblock_phys_mem_size(),
316 &size, &base, NULL, NULL);
317 if (ret == 0 && size > 0) {
318 unsigned long max_size;
319
320 if (fw_dump.reserve_bootvar)
321 pr_info("Using 'crashkernel=' parameter for memory reservation.\n");
322
323 fw_dump.reserve_bootvar = (unsigned long)size;
324
325 /*
326 * Adjust if the boot memory size specified is above
327 * the upper limit.
328 */
329 max_size = memblock_phys_mem_size() / MAX_BOOT_MEM_RATIO;
330 if (fw_dump.reserve_bootvar > max_size) {
331 fw_dump.reserve_bootvar = max_size;
332 pr_info("Adjusted boot memory size to %luMB\n",
333 (fw_dump.reserve_bootvar >> 20));
334 }
335
336 return fw_dump.reserve_bootvar;
337 } else if (fw_dump.reserve_bootvar) {
338 /*
339 * 'fadump_reserve_mem=' is being used to reserve memory
340 * for firmware-assisted dump.
341 */
342 return fw_dump.reserve_bootvar;
343 }
344
345 /* divide by 20 to get 5% of value */
346 size = memblock_phys_mem_size() / 20;
347
348 /* round it down in multiples of 256 */
349 size = size & ~0x0FFFFFFFUL;
350
351 /* Truncate to memory_limit. We don't want to over reserve the memory.*/
352 if (memory_limit && size > memory_limit)
353 size = memory_limit;
354
355 bootmem_min = fw_dump.ops->fadump_get_bootmem_min();
356 return (size > bootmem_min ? size : bootmem_min);
357}
358
359/*
360 * Calculate the total memory size required to be reserved for
361 * firmware-assisted dump registration.
362 */
363static unsigned long __init get_fadump_area_size(void)
364{
365 unsigned long size = 0;
366
367 size += fw_dump.cpu_state_data_size;
368 size += fw_dump.hpte_region_size;
369 /*
370 * Account for pagesize alignment of boot memory area destination address.
371 * This faciliates in mmap reading of first kernel's memory.
372 */
373 size = PAGE_ALIGN(size);
374 size += fw_dump.boot_memory_size;
375 size += sizeof(struct fadump_crash_info_header);
376 size += sizeof(struct elfhdr); /* ELF core header.*/
377 size += sizeof(struct elf_phdr); /* place holder for cpu notes */
378 /* Program headers for crash memory regions. */
379 size += sizeof(struct elf_phdr) * (memblock_num_regions(memory) + 2);
380
381 size = PAGE_ALIGN(size);
382
383 /* This is to hold kernel metadata on platforms that support it */
384 size += (fw_dump.ops->fadump_get_metadata_size ?
385 fw_dump.ops->fadump_get_metadata_size() : 0);
386 return size;
387}
388
389static int __init add_boot_mem_region(unsigned long rstart,
390 unsigned long rsize)
391{
392 int i = fw_dump.boot_mem_regs_cnt++;
393
394 if (fw_dump.boot_mem_regs_cnt > FADUMP_MAX_MEM_REGS) {
395 fw_dump.boot_mem_regs_cnt = FADUMP_MAX_MEM_REGS;
396 return 0;
397 }
398
399 pr_debug("Added boot memory range[%d] [%#016lx-%#016lx)\n",
400 i, rstart, (rstart + rsize));
401 fw_dump.boot_mem_addr[i] = rstart;
402 fw_dump.boot_mem_sz[i] = rsize;
403 return 1;
404}
405
406/*
407 * Firmware usually has a hard limit on the data it can copy per region.
408 * Honour that by splitting a memory range into multiple regions.
409 */
410static int __init add_boot_mem_regions(unsigned long mstart,
411 unsigned long msize)
412{
413 unsigned long rstart, rsize, max_size;
414 int ret = 1;
415
416 rstart = mstart;
417 max_size = fw_dump.max_copy_size ? fw_dump.max_copy_size : msize;
418 while (msize) {
419 if (msize > max_size)
420 rsize = max_size;
421 else
422 rsize = msize;
423
424 ret = add_boot_mem_region(rstart, rsize);
425 if (!ret)
426 break;
427
428 msize -= rsize;
429 rstart += rsize;
430 }
431
432 return ret;
433}
434
435static int __init fadump_get_boot_mem_regions(void)
436{
437 unsigned long size, cur_size, hole_size, last_end;
438 unsigned long mem_size = fw_dump.boot_memory_size;
439 phys_addr_t reg_start, reg_end;
440 int ret = 1;
441 u64 i;
442
443 fw_dump.boot_mem_regs_cnt = 0;
444
445 last_end = 0;
446 hole_size = 0;
447 cur_size = 0;
448 for_each_mem_range(i, ®_start, ®_end) {
449 size = reg_end - reg_start;
450 hole_size += (reg_start - last_end);
451
452 if ((cur_size + size) >= mem_size) {
453 size = (mem_size - cur_size);
454 ret = add_boot_mem_regions(reg_start, size);
455 break;
456 }
457
458 mem_size -= size;
459 cur_size += size;
460 ret = add_boot_mem_regions(reg_start, size);
461 if (!ret)
462 break;
463
464 last_end = reg_end;
465 }
466 fw_dump.boot_mem_top = PAGE_ALIGN(fw_dump.boot_memory_size + hole_size);
467
468 return ret;
469}
470
471/*
472 * Returns true, if the given range overlaps with reserved memory ranges
473 * starting at idx. Also, updates idx to index of overlapping memory range
474 * with the given memory range.
475 * False, otherwise.
476 */
477static bool __init overlaps_reserved_ranges(u64 base, u64 end, int *idx)
478{
479 bool ret = false;
480 int i;
481
482 for (i = *idx; i < reserved_mrange_info.mem_range_cnt; i++) {
483 u64 rbase = reserved_mrange_info.mem_ranges[i].base;
484 u64 rend = rbase + reserved_mrange_info.mem_ranges[i].size;
485
486 if (end <= rbase)
487 break;
488
489 if ((end > rbase) && (base < rend)) {
490 *idx = i;
491 ret = true;
492 break;
493 }
494 }
495
496 return ret;
497}
498
499/*
500 * Locate a suitable memory area to reserve memory for FADump. While at it,
501 * lookup reserved-ranges & avoid overlap with them, as they are used by F/W.
502 */
503static u64 __init fadump_locate_reserve_mem(u64 base, u64 size)
504{
505 struct fadump_memory_range *mrngs;
506 phys_addr_t mstart, mend;
507 int idx = 0;
508 u64 i, ret = 0;
509
510 mrngs = reserved_mrange_info.mem_ranges;
511 for_each_free_mem_range(i, NUMA_NO_NODE, MEMBLOCK_NONE,
512 &mstart, &mend, NULL) {
513 pr_debug("%llu) mstart: %llx, mend: %llx, base: %llx\n",
514 i, mstart, mend, base);
515
516 if (mstart > base)
517 base = PAGE_ALIGN(mstart);
518
519 while ((mend > base) && ((mend - base) >= size)) {
520 if (!overlaps_reserved_ranges(base, base+size, &idx)) {
521 ret = base;
522 goto out;
523 }
524
525 base = mrngs[idx].base + mrngs[idx].size;
526 base = PAGE_ALIGN(base);
527 }
528 }
529
530out:
531 return ret;
532}
533
534int __init fadump_reserve_mem(void)
535{
536 u64 base, size, mem_boundary, bootmem_min;
537 int ret = 1;
538
539 if (!fw_dump.fadump_enabled)
540 return 0;
541
542 if (!fw_dump.fadump_supported) {
543 pr_info("Firmware-Assisted Dump is not supported on this hardware\n");
544 goto error_out;
545 }
546
547 /*
548 * Initialize boot memory size
549 * If dump is active then we have already calculated the size during
550 * first kernel.
551 */
552 if (!fw_dump.dump_active) {
553 fw_dump.boot_memory_size =
554 PAGE_ALIGN(fadump_calculate_reserve_size());
555#ifdef CONFIG_CMA
556 if (!fw_dump.nocma) {
557 fw_dump.boot_memory_size =
558 ALIGN(fw_dump.boot_memory_size,
559 CMA_MIN_ALIGNMENT_BYTES);
560 }
561#endif
562
563 bootmem_min = fw_dump.ops->fadump_get_bootmem_min();
564 if (fw_dump.boot_memory_size < bootmem_min) {
565 pr_err("Can't enable fadump with boot memory size (0x%lx) less than 0x%llx\n",
566 fw_dump.boot_memory_size, bootmem_min);
567 goto error_out;
568 }
569
570 if (!fadump_get_boot_mem_regions()) {
571 pr_err("Too many holes in boot memory area to enable fadump\n");
572 goto error_out;
573 }
574 }
575
576 /*
577 * Calculate the memory boundary.
578 * If memory_limit is less than actual memory boundary then reserve
579 * the memory for fadump beyond the memory_limit and adjust the
580 * memory_limit accordingly, so that the running kernel can run with
581 * specified memory_limit.
582 */
583 if (memory_limit && memory_limit < memblock_end_of_DRAM()) {
584 size = get_fadump_area_size();
585 if ((memory_limit + size) < memblock_end_of_DRAM())
586 memory_limit += size;
587 else
588 memory_limit = memblock_end_of_DRAM();
589 printk(KERN_INFO "Adjusted memory_limit for firmware-assisted"
590 " dump, now %#016llx\n", memory_limit);
591 }
592 if (memory_limit)
593 mem_boundary = memory_limit;
594 else
595 mem_boundary = memblock_end_of_DRAM();
596
597 base = fw_dump.boot_mem_top;
598 size = get_fadump_area_size();
599 fw_dump.reserve_dump_area_size = size;
600 if (fw_dump.dump_active) {
601 pr_info("Firmware-assisted dump is active.\n");
602
603#ifdef CONFIG_HUGETLB_PAGE
604 /*
605 * FADump capture kernel doesn't care much about hugepages.
606 * In fact, handling hugepages in capture kernel is asking for
607 * trouble. So, disable HugeTLB support when fadump is active.
608 */
609 hugetlb_disabled = true;
610#endif
611 /*
612 * If last boot has crashed then reserve all the memory
613 * above boot memory size so that we don't touch it until
614 * dump is written to disk by userspace tool. This memory
615 * can be released for general use by invalidating fadump.
616 */
617 fadump_reserve_crash_area(base);
618
619 pr_debug("fadumphdr_addr = %#016lx\n", fw_dump.fadumphdr_addr);
620 pr_debug("Reserve dump area start address: 0x%lx\n",
621 fw_dump.reserve_dump_area_start);
622 } else {
623 /*
624 * Reserve memory at an offset closer to bottom of the RAM to
625 * minimize the impact of memory hot-remove operation.
626 */
627 base = fadump_locate_reserve_mem(base, size);
628
629 if (!base || (base + size > mem_boundary)) {
630 pr_err("Failed to find memory chunk for reservation!\n");
631 goto error_out;
632 }
633 fw_dump.reserve_dump_area_start = base;
634
635 /*
636 * Calculate the kernel metadata address and register it with
637 * f/w if the platform supports.
638 */
639 if (fw_dump.ops->fadump_setup_metadata &&
640 (fw_dump.ops->fadump_setup_metadata(&fw_dump) < 0))
641 goto error_out;
642
643 if (memblock_reserve(base, size)) {
644 pr_err("Failed to reserve memory!\n");
645 goto error_out;
646 }
647
648 pr_info("Reserved %lldMB of memory at %#016llx (System RAM: %lldMB)\n",
649 (size >> 20), base, (memblock_phys_mem_size() >> 20));
650
651 ret = fadump_cma_init();
652 }
653
654 return ret;
655error_out:
656 fw_dump.fadump_enabled = 0;
657 fw_dump.reserve_dump_area_size = 0;
658 return 0;
659}
660
661/* Look for fadump= cmdline option. */
662static int __init early_fadump_param(char *p)
663{
664 if (!p)
665 return 1;
666
667 if (strncmp(p, "on", 2) == 0)
668 fw_dump.fadump_enabled = 1;
669 else if (strncmp(p, "off", 3) == 0)
670 fw_dump.fadump_enabled = 0;
671 else if (strncmp(p, "nocma", 5) == 0) {
672 fw_dump.fadump_enabled = 1;
673 fw_dump.nocma = 1;
674 }
675
676 return 0;
677}
678early_param("fadump", early_fadump_param);
679
680/*
681 * Look for fadump_reserve_mem= cmdline option
682 * TODO: Remove references to 'fadump_reserve_mem=' parameter,
683 * the sooner 'crashkernel=' parameter is accustomed to.
684 */
685static int __init early_fadump_reserve_mem(char *p)
686{
687 if (p)
688 fw_dump.reserve_bootvar = memparse(p, &p);
689 return 0;
690}
691early_param("fadump_reserve_mem", early_fadump_reserve_mem);
692
693void crash_fadump(struct pt_regs *regs, const char *str)
694{
695 unsigned int msecs;
696 struct fadump_crash_info_header *fdh = NULL;
697 int old_cpu, this_cpu;
698 /* Do not include first CPU */
699 unsigned int ncpus = num_online_cpus() - 1;
700
701 if (!should_fadump_crash())
702 return;
703
704 /*
705 * old_cpu == -1 means this is the first CPU which has come here,
706 * go ahead and trigger fadump.
707 *
708 * old_cpu != -1 means some other CPU has already on it's way
709 * to trigger fadump, just keep looping here.
710 */
711 this_cpu = smp_processor_id();
712 old_cpu = cmpxchg(&crashing_cpu, -1, this_cpu);
713
714 if (old_cpu != -1) {
715 atomic_inc(&cpus_in_fadump);
716
717 /*
718 * We can't loop here indefinitely. Wait as long as fadump
719 * is in force. If we race with fadump un-registration this
720 * loop will break and then we go down to normal panic path
721 * and reboot. If fadump is in force the first crashing
722 * cpu will definitely trigger fadump.
723 */
724 while (fw_dump.dump_registered)
725 cpu_relax();
726 return;
727 }
728
729 fdh = __va(fw_dump.fadumphdr_addr);
730 fdh->crashing_cpu = crashing_cpu;
731 crash_save_vmcoreinfo();
732
733 if (regs)
734 fdh->regs = *regs;
735 else
736 ppc_save_regs(&fdh->regs);
737
738 fdh->cpu_mask = *cpu_online_mask;
739
740 /*
741 * If we came in via system reset, wait a while for the secondary
742 * CPUs to enter.
743 */
744 if (TRAP(&(fdh->regs)) == INTERRUPT_SYSTEM_RESET) {
745 msecs = CRASH_TIMEOUT;
746 while ((atomic_read(&cpus_in_fadump) < ncpus) && (--msecs > 0))
747 mdelay(1);
748 }
749
750 fw_dump.ops->fadump_trigger(fdh, str);
751}
752
753u32 *__init fadump_regs_to_elf_notes(u32 *buf, struct pt_regs *regs)
754{
755 struct elf_prstatus prstatus;
756
757 memset(&prstatus, 0, sizeof(prstatus));
758 /*
759 * FIXME: How do i get PID? Do I really need it?
760 * prstatus.pr_pid = ????
761 */
762 elf_core_copy_regs(&prstatus.pr_reg, regs);
763 buf = append_elf_note(buf, CRASH_CORE_NOTE_NAME, NT_PRSTATUS,
764 &prstatus, sizeof(prstatus));
765 return buf;
766}
767
768void __init fadump_update_elfcore_header(char *bufp)
769{
770 struct elf_phdr *phdr;
771
772 bufp += sizeof(struct elfhdr);
773
774 /* First note is a place holder for cpu notes info. */
775 phdr = (struct elf_phdr *)bufp;
776
777 if (phdr->p_type == PT_NOTE) {
778 phdr->p_paddr = __pa(fw_dump.cpu_notes_buf_vaddr);
779 phdr->p_offset = phdr->p_paddr;
780 phdr->p_filesz = fw_dump.cpu_notes_buf_size;
781 phdr->p_memsz = fw_dump.cpu_notes_buf_size;
782 }
783 return;
784}
785
786static void *__init fadump_alloc_buffer(unsigned long size)
787{
788 unsigned long count, i;
789 struct page *page;
790 void *vaddr;
791
792 vaddr = alloc_pages_exact(size, GFP_KERNEL | __GFP_ZERO);
793 if (!vaddr)
794 return NULL;
795
796 count = PAGE_ALIGN(size) / PAGE_SIZE;
797 page = virt_to_page(vaddr);
798 for (i = 0; i < count; i++)
799 mark_page_reserved(page + i);
800 return vaddr;
801}
802
803static void fadump_free_buffer(unsigned long vaddr, unsigned long size)
804{
805 free_reserved_area((void *)vaddr, (void *)(vaddr + size), -1, NULL);
806}
807
808s32 __init fadump_setup_cpu_notes_buf(u32 num_cpus)
809{
810 /* Allocate buffer to hold cpu crash notes. */
811 fw_dump.cpu_notes_buf_size = num_cpus * sizeof(note_buf_t);
812 fw_dump.cpu_notes_buf_size = PAGE_ALIGN(fw_dump.cpu_notes_buf_size);
813 fw_dump.cpu_notes_buf_vaddr =
814 (unsigned long)fadump_alloc_buffer(fw_dump.cpu_notes_buf_size);
815 if (!fw_dump.cpu_notes_buf_vaddr) {
816 pr_err("Failed to allocate %ld bytes for CPU notes buffer\n",
817 fw_dump.cpu_notes_buf_size);
818 return -ENOMEM;
819 }
820
821 pr_debug("Allocated buffer for cpu notes of size %ld at 0x%lx\n",
822 fw_dump.cpu_notes_buf_size,
823 fw_dump.cpu_notes_buf_vaddr);
824 return 0;
825}
826
827void fadump_free_cpu_notes_buf(void)
828{
829 if (!fw_dump.cpu_notes_buf_vaddr)
830 return;
831
832 fadump_free_buffer(fw_dump.cpu_notes_buf_vaddr,
833 fw_dump.cpu_notes_buf_size);
834 fw_dump.cpu_notes_buf_vaddr = 0;
835 fw_dump.cpu_notes_buf_size = 0;
836}
837
838static void fadump_free_mem_ranges(struct fadump_mrange_info *mrange_info)
839{
840 if (mrange_info->is_static) {
841 mrange_info->mem_range_cnt = 0;
842 return;
843 }
844
845 kfree(mrange_info->mem_ranges);
846 memset((void *)((u64)mrange_info + RNG_NAME_SZ), 0,
847 (sizeof(struct fadump_mrange_info) - RNG_NAME_SZ));
848}
849
850/*
851 * Allocate or reallocate mem_ranges array in incremental units
852 * of PAGE_SIZE.
853 */
854static int fadump_alloc_mem_ranges(struct fadump_mrange_info *mrange_info)
855{
856 struct fadump_memory_range *new_array;
857 u64 new_size;
858
859 new_size = mrange_info->mem_ranges_sz + PAGE_SIZE;
860 pr_debug("Allocating %llu bytes of memory for %s memory ranges\n",
861 new_size, mrange_info->name);
862
863 new_array = krealloc(mrange_info->mem_ranges, new_size, GFP_KERNEL);
864 if (new_array == NULL) {
865 pr_err("Insufficient memory for setting up %s memory ranges\n",
866 mrange_info->name);
867 fadump_free_mem_ranges(mrange_info);
868 return -ENOMEM;
869 }
870
871 mrange_info->mem_ranges = new_array;
872 mrange_info->mem_ranges_sz = new_size;
873 mrange_info->max_mem_ranges = (new_size /
874 sizeof(struct fadump_memory_range));
875 return 0;
876}
877static inline int fadump_add_mem_range(struct fadump_mrange_info *mrange_info,
878 u64 base, u64 end)
879{
880 struct fadump_memory_range *mem_ranges = mrange_info->mem_ranges;
881 bool is_adjacent = false;
882 u64 start, size;
883
884 if (base == end)
885 return 0;
886
887 /*
888 * Fold adjacent memory ranges to bring down the memory ranges/
889 * PT_LOAD segments count.
890 */
891 if (mrange_info->mem_range_cnt) {
892 start = mem_ranges[mrange_info->mem_range_cnt - 1].base;
893 size = mem_ranges[mrange_info->mem_range_cnt - 1].size;
894
895 /*
896 * Boot memory area needs separate PT_LOAD segment(s) as it
897 * is moved to a different location at the time of crash.
898 * So, fold only if the region is not boot memory area.
899 */
900 if ((start + size) == base && start >= fw_dump.boot_mem_top)
901 is_adjacent = true;
902 }
903 if (!is_adjacent) {
904 /* resize the array on reaching the limit */
905 if (mrange_info->mem_range_cnt == mrange_info->max_mem_ranges) {
906 int ret;
907
908 if (mrange_info->is_static) {
909 pr_err("Reached array size limit for %s memory ranges\n",
910 mrange_info->name);
911 return -ENOSPC;
912 }
913
914 ret = fadump_alloc_mem_ranges(mrange_info);
915 if (ret)
916 return ret;
917
918 /* Update to the new resized array */
919 mem_ranges = mrange_info->mem_ranges;
920 }
921
922 start = base;
923 mem_ranges[mrange_info->mem_range_cnt].base = start;
924 mrange_info->mem_range_cnt++;
925 }
926
927 mem_ranges[mrange_info->mem_range_cnt - 1].size = (end - start);
928 pr_debug("%s_memory_range[%d] [%#016llx-%#016llx], %#llx bytes\n",
929 mrange_info->name, (mrange_info->mem_range_cnt - 1),
930 start, end - 1, (end - start));
931 return 0;
932}
933
934static int fadump_exclude_reserved_area(u64 start, u64 end)
935{
936 u64 ra_start, ra_end;
937 int ret = 0;
938
939 ra_start = fw_dump.reserve_dump_area_start;
940 ra_end = ra_start + fw_dump.reserve_dump_area_size;
941
942 if ((ra_start < end) && (ra_end > start)) {
943 if ((start < ra_start) && (end > ra_end)) {
944 ret = fadump_add_mem_range(&crash_mrange_info,
945 start, ra_start);
946 if (ret)
947 return ret;
948
949 ret = fadump_add_mem_range(&crash_mrange_info,
950 ra_end, end);
951 } else if (start < ra_start) {
952 ret = fadump_add_mem_range(&crash_mrange_info,
953 start, ra_start);
954 } else if (ra_end < end) {
955 ret = fadump_add_mem_range(&crash_mrange_info,
956 ra_end, end);
957 }
958 } else
959 ret = fadump_add_mem_range(&crash_mrange_info, start, end);
960
961 return ret;
962}
963
964static int fadump_init_elfcore_header(char *bufp)
965{
966 struct elfhdr *elf;
967
968 elf = (struct elfhdr *) bufp;
969 bufp += sizeof(struct elfhdr);
970 memcpy(elf->e_ident, ELFMAG, SELFMAG);
971 elf->e_ident[EI_CLASS] = ELF_CLASS;
972 elf->e_ident[EI_DATA] = ELF_DATA;
973 elf->e_ident[EI_VERSION] = EV_CURRENT;
974 elf->e_ident[EI_OSABI] = ELF_OSABI;
975 memset(elf->e_ident+EI_PAD, 0, EI_NIDENT-EI_PAD);
976 elf->e_type = ET_CORE;
977 elf->e_machine = ELF_ARCH;
978 elf->e_version = EV_CURRENT;
979 elf->e_entry = 0;
980 elf->e_phoff = sizeof(struct elfhdr);
981 elf->e_shoff = 0;
982
983 if (IS_ENABLED(CONFIG_PPC64_ELF_ABI_V2))
984 elf->e_flags = 2;
985 else if (IS_ENABLED(CONFIG_PPC64_ELF_ABI_V1))
986 elf->e_flags = 1;
987 else
988 elf->e_flags = 0;
989
990 elf->e_ehsize = sizeof(struct elfhdr);
991 elf->e_phentsize = sizeof(struct elf_phdr);
992 elf->e_phnum = 0;
993 elf->e_shentsize = 0;
994 elf->e_shnum = 0;
995 elf->e_shstrndx = 0;
996
997 return 0;
998}
999
1000/*
1001 * Traverse through memblock structure and setup crash memory ranges. These
1002 * ranges will be used create PT_LOAD program headers in elfcore header.
1003 */
1004static int fadump_setup_crash_memory_ranges(void)
1005{
1006 u64 i, start, end;
1007 int ret;
1008
1009 pr_debug("Setup crash memory ranges.\n");
1010 crash_mrange_info.mem_range_cnt = 0;
1011
1012 /*
1013 * Boot memory region(s) registered with firmware are moved to
1014 * different location at the time of crash. Create separate program
1015 * header(s) for this memory chunk(s) with the correct offset.
1016 */
1017 for (i = 0; i < fw_dump.boot_mem_regs_cnt; i++) {
1018 start = fw_dump.boot_mem_addr[i];
1019 end = start + fw_dump.boot_mem_sz[i];
1020 ret = fadump_add_mem_range(&crash_mrange_info, start, end);
1021 if (ret)
1022 return ret;
1023 }
1024
1025 for_each_mem_range(i, &start, &end) {
1026 /*
1027 * skip the memory chunk that is already added
1028 * (0 through boot_memory_top).
1029 */
1030 if (start < fw_dump.boot_mem_top) {
1031 if (end > fw_dump.boot_mem_top)
1032 start = fw_dump.boot_mem_top;
1033 else
1034 continue;
1035 }
1036
1037 /* add this range excluding the reserved dump area. */
1038 ret = fadump_exclude_reserved_area(start, end);
1039 if (ret)
1040 return ret;
1041 }
1042
1043 return 0;
1044}
1045
1046/*
1047 * If the given physical address falls within the boot memory region then
1048 * return the relocated address that points to the dump region reserved
1049 * for saving initial boot memory contents.
1050 */
1051static inline unsigned long fadump_relocate(unsigned long paddr)
1052{
1053 unsigned long raddr, rstart, rend, rlast, hole_size;
1054 int i;
1055
1056 hole_size = 0;
1057 rlast = 0;
1058 raddr = paddr;
1059 for (i = 0; i < fw_dump.boot_mem_regs_cnt; i++) {
1060 rstart = fw_dump.boot_mem_addr[i];
1061 rend = rstart + fw_dump.boot_mem_sz[i];
1062 hole_size += (rstart - rlast);
1063
1064 if (paddr >= rstart && paddr < rend) {
1065 raddr += fw_dump.boot_mem_dest_addr - hole_size;
1066 break;
1067 }
1068
1069 rlast = rend;
1070 }
1071
1072 pr_debug("vmcoreinfo: paddr = 0x%lx, raddr = 0x%lx\n", paddr, raddr);
1073 return raddr;
1074}
1075
1076static int fadump_create_elfcore_headers(char *bufp)
1077{
1078 unsigned long long raddr, offset;
1079 struct elf_phdr *phdr;
1080 struct elfhdr *elf;
1081 int i, j;
1082
1083 fadump_init_elfcore_header(bufp);
1084 elf = (struct elfhdr *)bufp;
1085 bufp += sizeof(struct elfhdr);
1086
1087 /*
1088 * setup ELF PT_NOTE, place holder for cpu notes info. The notes info
1089 * will be populated during second kernel boot after crash. Hence
1090 * this PT_NOTE will always be the first elf note.
1091 *
1092 * NOTE: Any new ELF note addition should be placed after this note.
1093 */
1094 phdr = (struct elf_phdr *)bufp;
1095 bufp += sizeof(struct elf_phdr);
1096 phdr->p_type = PT_NOTE;
1097 phdr->p_flags = 0;
1098 phdr->p_vaddr = 0;
1099 phdr->p_align = 0;
1100
1101 phdr->p_offset = 0;
1102 phdr->p_paddr = 0;
1103 phdr->p_filesz = 0;
1104 phdr->p_memsz = 0;
1105
1106 (elf->e_phnum)++;
1107
1108 /* setup ELF PT_NOTE for vmcoreinfo */
1109 phdr = (struct elf_phdr *)bufp;
1110 bufp += sizeof(struct elf_phdr);
1111 phdr->p_type = PT_NOTE;
1112 phdr->p_flags = 0;
1113 phdr->p_vaddr = 0;
1114 phdr->p_align = 0;
1115
1116 phdr->p_paddr = fadump_relocate(paddr_vmcoreinfo_note());
1117 phdr->p_offset = phdr->p_paddr;
1118 phdr->p_memsz = phdr->p_filesz = VMCOREINFO_NOTE_SIZE;
1119
1120 /* Increment number of program headers. */
1121 (elf->e_phnum)++;
1122
1123 /* setup PT_LOAD sections. */
1124 j = 0;
1125 offset = 0;
1126 raddr = fw_dump.boot_mem_addr[0];
1127 for (i = 0; i < crash_mrange_info.mem_range_cnt; i++) {
1128 u64 mbase, msize;
1129
1130 mbase = crash_mrange_info.mem_ranges[i].base;
1131 msize = crash_mrange_info.mem_ranges[i].size;
1132 if (!msize)
1133 continue;
1134
1135 phdr = (struct elf_phdr *)bufp;
1136 bufp += sizeof(struct elf_phdr);
1137 phdr->p_type = PT_LOAD;
1138 phdr->p_flags = PF_R|PF_W|PF_X;
1139 phdr->p_offset = mbase;
1140
1141 if (mbase == raddr) {
1142 /*
1143 * The entire real memory region will be moved by
1144 * firmware to the specified destination_address.
1145 * Hence set the correct offset.
1146 */
1147 phdr->p_offset = fw_dump.boot_mem_dest_addr + offset;
1148 if (j < (fw_dump.boot_mem_regs_cnt - 1)) {
1149 offset += fw_dump.boot_mem_sz[j];
1150 raddr = fw_dump.boot_mem_addr[++j];
1151 }
1152 }
1153
1154 phdr->p_paddr = mbase;
1155 phdr->p_vaddr = (unsigned long)__va(mbase);
1156 phdr->p_filesz = msize;
1157 phdr->p_memsz = msize;
1158 phdr->p_align = 0;
1159
1160 /* Increment number of program headers. */
1161 (elf->e_phnum)++;
1162 }
1163 return 0;
1164}
1165
1166static unsigned long init_fadump_header(unsigned long addr)
1167{
1168 struct fadump_crash_info_header *fdh;
1169
1170 if (!addr)
1171 return 0;
1172
1173 fdh = __va(addr);
1174 addr += sizeof(struct fadump_crash_info_header);
1175
1176 memset(fdh, 0, sizeof(struct fadump_crash_info_header));
1177 fdh->magic_number = FADUMP_CRASH_INFO_MAGIC;
1178 fdh->elfcorehdr_addr = addr;
1179 /* We will set the crashing cpu id in crash_fadump() during crash. */
1180 fdh->crashing_cpu = FADUMP_CPU_UNKNOWN;
1181 /*
1182 * When LPAR is terminated by PYHP, ensure all possible CPUs'
1183 * register data is processed while exporting the vmcore.
1184 */
1185 fdh->cpu_mask = *cpu_possible_mask;
1186
1187 return addr;
1188}
1189
1190static int register_fadump(void)
1191{
1192 unsigned long addr;
1193 void *vaddr;
1194 int ret;
1195
1196 /*
1197 * If no memory is reserved then we can not register for firmware-
1198 * assisted dump.
1199 */
1200 if (!fw_dump.reserve_dump_area_size)
1201 return -ENODEV;
1202
1203 ret = fadump_setup_crash_memory_ranges();
1204 if (ret)
1205 return ret;
1206
1207 addr = fw_dump.fadumphdr_addr;
1208
1209 /* Initialize fadump crash info header. */
1210 addr = init_fadump_header(addr);
1211 vaddr = __va(addr);
1212
1213 pr_debug("Creating ELF core headers at %#016lx\n", addr);
1214 fadump_create_elfcore_headers(vaddr);
1215
1216 /* register the future kernel dump with firmware. */
1217 pr_debug("Registering for firmware-assisted kernel dump...\n");
1218 return fw_dump.ops->fadump_register(&fw_dump);
1219}
1220
1221void fadump_cleanup(void)
1222{
1223 if (!fw_dump.fadump_supported)
1224 return;
1225
1226 /* Invalidate the registration only if dump is active. */
1227 if (fw_dump.dump_active) {
1228 pr_debug("Invalidating firmware-assisted dump registration\n");
1229 fw_dump.ops->fadump_invalidate(&fw_dump);
1230 } else if (fw_dump.dump_registered) {
1231 /* Un-register Firmware-assisted dump if it was registered. */
1232 fw_dump.ops->fadump_unregister(&fw_dump);
1233 fadump_free_mem_ranges(&crash_mrange_info);
1234 }
1235
1236 if (fw_dump.ops->fadump_cleanup)
1237 fw_dump.ops->fadump_cleanup(&fw_dump);
1238}
1239
1240static void fadump_free_reserved_memory(unsigned long start_pfn,
1241 unsigned long end_pfn)
1242{
1243 unsigned long pfn;
1244 unsigned long time_limit = jiffies + HZ;
1245
1246 pr_info("freeing reserved memory (0x%llx - 0x%llx)\n",
1247 PFN_PHYS(start_pfn), PFN_PHYS(end_pfn));
1248
1249 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
1250 free_reserved_page(pfn_to_page(pfn));
1251
1252 if (time_after(jiffies, time_limit)) {
1253 cond_resched();
1254 time_limit = jiffies + HZ;
1255 }
1256 }
1257}
1258
1259/*
1260 * Skip memory holes and free memory that was actually reserved.
1261 */
1262static void fadump_release_reserved_area(u64 start, u64 end)
1263{
1264 unsigned long reg_spfn, reg_epfn;
1265 u64 tstart, tend, spfn, epfn;
1266 int i;
1267
1268 spfn = PHYS_PFN(start);
1269 epfn = PHYS_PFN(end);
1270
1271 for_each_mem_pfn_range(i, MAX_NUMNODES, ®_spfn, ®_epfn, NULL) {
1272 tstart = max_t(u64, spfn, reg_spfn);
1273 tend = min_t(u64, epfn, reg_epfn);
1274
1275 if (tstart < tend) {
1276 fadump_free_reserved_memory(tstart, tend);
1277
1278 if (tend == epfn)
1279 break;
1280
1281 spfn = tend;
1282 }
1283 }
1284}
1285
1286/*
1287 * Sort the mem ranges in-place and merge adjacent ranges
1288 * to minimize the memory ranges count.
1289 */
1290static void sort_and_merge_mem_ranges(struct fadump_mrange_info *mrange_info)
1291{
1292 struct fadump_memory_range *mem_ranges;
1293 u64 base, size;
1294 int i, j, idx;
1295
1296 if (!reserved_mrange_info.mem_range_cnt)
1297 return;
1298
1299 /* Sort the memory ranges */
1300 mem_ranges = mrange_info->mem_ranges;
1301 for (i = 0; i < mrange_info->mem_range_cnt; i++) {
1302 idx = i;
1303 for (j = (i + 1); j < mrange_info->mem_range_cnt; j++) {
1304 if (mem_ranges[idx].base > mem_ranges[j].base)
1305 idx = j;
1306 }
1307 if (idx != i)
1308 swap(mem_ranges[idx], mem_ranges[i]);
1309 }
1310
1311 /* Merge adjacent reserved ranges */
1312 idx = 0;
1313 for (i = 1; i < mrange_info->mem_range_cnt; i++) {
1314 base = mem_ranges[i-1].base;
1315 size = mem_ranges[i-1].size;
1316 if (mem_ranges[i].base == (base + size))
1317 mem_ranges[idx].size += mem_ranges[i].size;
1318 else {
1319 idx++;
1320 if (i == idx)
1321 continue;
1322
1323 mem_ranges[idx] = mem_ranges[i];
1324 }
1325 }
1326 mrange_info->mem_range_cnt = idx + 1;
1327}
1328
1329/*
1330 * Scan reserved-ranges to consider them while reserving/releasing
1331 * memory for FADump.
1332 */
1333static void __init early_init_dt_scan_reserved_ranges(unsigned long node)
1334{
1335 const __be32 *prop;
1336 int len, ret = -1;
1337 unsigned long i;
1338
1339 /* reserved-ranges already scanned */
1340 if (reserved_mrange_info.mem_range_cnt != 0)
1341 return;
1342
1343 prop = of_get_flat_dt_prop(node, "reserved-ranges", &len);
1344 if (!prop)
1345 return;
1346
1347 /*
1348 * Each reserved range is an (address,size) pair, 2 cells each,
1349 * totalling 4 cells per range.
1350 */
1351 for (i = 0; i < len / (sizeof(*prop) * 4); i++) {
1352 u64 base, size;
1353
1354 base = of_read_number(prop + (i * 4) + 0, 2);
1355 size = of_read_number(prop + (i * 4) + 2, 2);
1356
1357 if (size) {
1358 ret = fadump_add_mem_range(&reserved_mrange_info,
1359 base, base + size);
1360 if (ret < 0) {
1361 pr_warn("some reserved ranges are ignored!\n");
1362 break;
1363 }
1364 }
1365 }
1366
1367 /* Compact reserved ranges */
1368 sort_and_merge_mem_ranges(&reserved_mrange_info);
1369}
1370
1371/*
1372 * Release the memory that was reserved during early boot to preserve the
1373 * crash'ed kernel's memory contents except reserved dump area (permanent
1374 * reservation) and reserved ranges used by F/W. The released memory will
1375 * be available for general use.
1376 */
1377static void fadump_release_memory(u64 begin, u64 end)
1378{
1379 u64 ra_start, ra_end, tstart;
1380 int i, ret;
1381
1382 ra_start = fw_dump.reserve_dump_area_start;
1383 ra_end = ra_start + fw_dump.reserve_dump_area_size;
1384
1385 /*
1386 * If reserved ranges array limit is hit, overwrite the last reserved
1387 * memory range with reserved dump area to ensure it is excluded from
1388 * the memory being released (reused for next FADump registration).
1389 */
1390 if (reserved_mrange_info.mem_range_cnt ==
1391 reserved_mrange_info.max_mem_ranges)
1392 reserved_mrange_info.mem_range_cnt--;
1393
1394 ret = fadump_add_mem_range(&reserved_mrange_info, ra_start, ra_end);
1395 if (ret != 0)
1396 return;
1397
1398 /* Get the reserved ranges list in order first. */
1399 sort_and_merge_mem_ranges(&reserved_mrange_info);
1400
1401 /* Exclude reserved ranges and release remaining memory */
1402 tstart = begin;
1403 for (i = 0; i < reserved_mrange_info.mem_range_cnt; i++) {
1404 ra_start = reserved_mrange_info.mem_ranges[i].base;
1405 ra_end = ra_start + reserved_mrange_info.mem_ranges[i].size;
1406
1407 if (tstart >= ra_end)
1408 continue;
1409
1410 if (tstart < ra_start)
1411 fadump_release_reserved_area(tstart, ra_start);
1412 tstart = ra_end;
1413 }
1414
1415 if (tstart < end)
1416 fadump_release_reserved_area(tstart, end);
1417}
1418
1419static void fadump_invalidate_release_mem(void)
1420{
1421 mutex_lock(&fadump_mutex);
1422 if (!fw_dump.dump_active) {
1423 mutex_unlock(&fadump_mutex);
1424 return;
1425 }
1426
1427 fadump_cleanup();
1428 mutex_unlock(&fadump_mutex);
1429
1430 fadump_release_memory(fw_dump.boot_mem_top, memblock_end_of_DRAM());
1431 fadump_free_cpu_notes_buf();
1432
1433 /*
1434 * Setup kernel metadata and initialize the kernel dump
1435 * memory structure for FADump re-registration.
1436 */
1437 if (fw_dump.ops->fadump_setup_metadata &&
1438 (fw_dump.ops->fadump_setup_metadata(&fw_dump) < 0))
1439 pr_warn("Failed to setup kernel metadata!\n");
1440 fw_dump.ops->fadump_init_mem_struct(&fw_dump);
1441}
1442
1443static ssize_t release_mem_store(struct kobject *kobj,
1444 struct kobj_attribute *attr,
1445 const char *buf, size_t count)
1446{
1447 int input = -1;
1448
1449 if (!fw_dump.dump_active)
1450 return -EPERM;
1451
1452 if (kstrtoint(buf, 0, &input))
1453 return -EINVAL;
1454
1455 if (input == 1) {
1456 /*
1457 * Take away the '/proc/vmcore'. We are releasing the dump
1458 * memory, hence it will not be valid anymore.
1459 */
1460#ifdef CONFIG_PROC_VMCORE
1461 vmcore_cleanup();
1462#endif
1463 fadump_invalidate_release_mem();
1464
1465 } else
1466 return -EINVAL;
1467 return count;
1468}
1469
1470/* Release the reserved memory and disable the FADump */
1471static void __init unregister_fadump(void)
1472{
1473 fadump_cleanup();
1474 fadump_release_memory(fw_dump.reserve_dump_area_start,
1475 fw_dump.reserve_dump_area_size);
1476 fw_dump.fadump_enabled = 0;
1477 kobject_put(fadump_kobj);
1478}
1479
1480static ssize_t enabled_show(struct kobject *kobj,
1481 struct kobj_attribute *attr,
1482 char *buf)
1483{
1484 return sprintf(buf, "%d\n", fw_dump.fadump_enabled);
1485}
1486
1487static ssize_t mem_reserved_show(struct kobject *kobj,
1488 struct kobj_attribute *attr,
1489 char *buf)
1490{
1491 return sprintf(buf, "%ld\n", fw_dump.reserve_dump_area_size);
1492}
1493
1494static ssize_t registered_show(struct kobject *kobj,
1495 struct kobj_attribute *attr,
1496 char *buf)
1497{
1498 return sprintf(buf, "%d\n", fw_dump.dump_registered);
1499}
1500
1501static ssize_t registered_store(struct kobject *kobj,
1502 struct kobj_attribute *attr,
1503 const char *buf, size_t count)
1504{
1505 int ret = 0;
1506 int input = -1;
1507
1508 if (!fw_dump.fadump_enabled || fw_dump.dump_active)
1509 return -EPERM;
1510
1511 if (kstrtoint(buf, 0, &input))
1512 return -EINVAL;
1513
1514 mutex_lock(&fadump_mutex);
1515
1516 switch (input) {
1517 case 0:
1518 if (fw_dump.dump_registered == 0) {
1519 goto unlock_out;
1520 }
1521
1522 /* Un-register Firmware-assisted dump */
1523 pr_debug("Un-register firmware-assisted dump\n");
1524 fw_dump.ops->fadump_unregister(&fw_dump);
1525 break;
1526 case 1:
1527 if (fw_dump.dump_registered == 1) {
1528 /* Un-register Firmware-assisted dump */
1529 fw_dump.ops->fadump_unregister(&fw_dump);
1530 }
1531 /* Register Firmware-assisted dump */
1532 ret = register_fadump();
1533 break;
1534 default:
1535 ret = -EINVAL;
1536 break;
1537 }
1538
1539unlock_out:
1540 mutex_unlock(&fadump_mutex);
1541 return ret < 0 ? ret : count;
1542}
1543
1544static int fadump_region_show(struct seq_file *m, void *private)
1545{
1546 if (!fw_dump.fadump_enabled)
1547 return 0;
1548
1549 mutex_lock(&fadump_mutex);
1550 fw_dump.ops->fadump_region_show(&fw_dump, m);
1551 mutex_unlock(&fadump_mutex);
1552 return 0;
1553}
1554
1555static struct kobj_attribute release_attr = __ATTR_WO(release_mem);
1556static struct kobj_attribute enable_attr = __ATTR_RO(enabled);
1557static struct kobj_attribute register_attr = __ATTR_RW(registered);
1558static struct kobj_attribute mem_reserved_attr = __ATTR_RO(mem_reserved);
1559
1560static struct attribute *fadump_attrs[] = {
1561 &enable_attr.attr,
1562 ®ister_attr.attr,
1563 &mem_reserved_attr.attr,
1564 NULL,
1565};
1566
1567ATTRIBUTE_GROUPS(fadump);
1568
1569DEFINE_SHOW_ATTRIBUTE(fadump_region);
1570
1571static void __init fadump_init_files(void)
1572{
1573 int rc = 0;
1574
1575 fadump_kobj = kobject_create_and_add("fadump", kernel_kobj);
1576 if (!fadump_kobj) {
1577 pr_err("failed to create fadump kobject\n");
1578 return;
1579 }
1580
1581 debugfs_create_file("fadump_region", 0444, arch_debugfs_dir, NULL,
1582 &fadump_region_fops);
1583
1584 if (fw_dump.dump_active) {
1585 rc = sysfs_create_file(fadump_kobj, &release_attr.attr);
1586 if (rc)
1587 pr_err("unable to create release_mem sysfs file (%d)\n",
1588 rc);
1589 }
1590
1591 rc = sysfs_create_groups(fadump_kobj, fadump_groups);
1592 if (rc) {
1593 pr_err("sysfs group creation failed (%d), unregistering FADump",
1594 rc);
1595 unregister_fadump();
1596 return;
1597 }
1598
1599 /*
1600 * The FADump sysfs are moved from kernel_kobj to fadump_kobj need to
1601 * create symlink at old location to maintain backward compatibility.
1602 *
1603 * - fadump_enabled -> fadump/enabled
1604 * - fadump_registered -> fadump/registered
1605 * - fadump_release_mem -> fadump/release_mem
1606 */
1607 rc = compat_only_sysfs_link_entry_to_kobj(kernel_kobj, fadump_kobj,
1608 "enabled", "fadump_enabled");
1609 if (rc) {
1610 pr_err("unable to create fadump_enabled symlink (%d)", rc);
1611 return;
1612 }
1613
1614 rc = compat_only_sysfs_link_entry_to_kobj(kernel_kobj, fadump_kobj,
1615 "registered",
1616 "fadump_registered");
1617 if (rc) {
1618 pr_err("unable to create fadump_registered symlink (%d)", rc);
1619 sysfs_remove_link(kernel_kobj, "fadump_enabled");
1620 return;
1621 }
1622
1623 if (fw_dump.dump_active) {
1624 rc = compat_only_sysfs_link_entry_to_kobj(kernel_kobj,
1625 fadump_kobj,
1626 "release_mem",
1627 "fadump_release_mem");
1628 if (rc)
1629 pr_err("unable to create fadump_release_mem symlink (%d)",
1630 rc);
1631 }
1632 return;
1633}
1634
1635/*
1636 * Prepare for firmware-assisted dump.
1637 */
1638int __init setup_fadump(void)
1639{
1640 if (!fw_dump.fadump_supported)
1641 return 0;
1642
1643 fadump_init_files();
1644 fadump_show_config();
1645
1646 if (!fw_dump.fadump_enabled)
1647 return 1;
1648
1649 /*
1650 * If dump data is available then see if it is valid and prepare for
1651 * saving it to the disk.
1652 */
1653 if (fw_dump.dump_active) {
1654 /*
1655 * if dump process fails then invalidate the registration
1656 * and release memory before proceeding for re-registration.
1657 */
1658 if (fw_dump.ops->fadump_process(&fw_dump) < 0)
1659 fadump_invalidate_release_mem();
1660 }
1661 /* Initialize the kernel dump memory structure and register with f/w */
1662 else if (fw_dump.reserve_dump_area_size) {
1663 fw_dump.ops->fadump_init_mem_struct(&fw_dump);
1664 register_fadump();
1665 }
1666
1667 /*
1668 * In case of panic, fadump is triggered via ppc_panic_event()
1669 * panic notifier. Setting crash_kexec_post_notifiers to 'true'
1670 * lets panic() function take crash friendly path before panic
1671 * notifiers are invoked.
1672 */
1673 crash_kexec_post_notifiers = true;
1674
1675 return 1;
1676}
1677/*
1678 * Use subsys_initcall_sync() here because there is dependency with
1679 * crash_save_vmcoreinfo_init(), which must run first to ensure vmcoreinfo initialization
1680 * is done before registering with f/w.
1681 */
1682subsys_initcall_sync(setup_fadump);
1683#else /* !CONFIG_PRESERVE_FA_DUMP */
1684
1685/* Scan the Firmware Assisted dump configuration details. */
1686int __init early_init_dt_scan_fw_dump(unsigned long node, const char *uname,
1687 int depth, void *data)
1688{
1689 if ((depth != 1) || (strcmp(uname, "ibm,opal") != 0))
1690 return 0;
1691
1692 opal_fadump_dt_scan(&fw_dump, node);
1693 return 1;
1694}
1695
1696/*
1697 * When dump is active but PRESERVE_FA_DUMP is enabled on the kernel,
1698 * preserve crash data. The subsequent memory preserving kernel boot
1699 * is likely to process this crash data.
1700 */
1701int __init fadump_reserve_mem(void)
1702{
1703 if (fw_dump.dump_active) {
1704 /*
1705 * If last boot has crashed then reserve all the memory
1706 * above boot memory to preserve crash data.
1707 */
1708 pr_info("Preserving crash data for processing in next boot.\n");
1709 fadump_reserve_crash_area(fw_dump.boot_mem_top);
1710 } else
1711 pr_debug("FADump-aware kernel..\n");
1712
1713 return 1;
1714}
1715#endif /* CONFIG_PRESERVE_FA_DUMP */
1716
1717/* Preserve everything above the base address */
1718static void __init fadump_reserve_crash_area(u64 base)
1719{
1720 u64 i, mstart, mend, msize;
1721
1722 for_each_mem_range(i, &mstart, &mend) {
1723 msize = mend - mstart;
1724
1725 if ((mstart + msize) < base)
1726 continue;
1727
1728 if (mstart < base) {
1729 msize -= (base - mstart);
1730 mstart = base;
1731 }
1732
1733 pr_info("Reserving %lluMB of memory at %#016llx for preserving crash data",
1734 (msize >> 20), mstart);
1735 memblock_reserve(mstart, msize);
1736 }
1737}
1738
1739unsigned long __init arch_reserved_kernel_pages(void)
1740{
1741 return memblock_reserved_size() / PAGE_SIZE;
1742}
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#include <linux/debugfs.h>
28#include <linux/of.h>
29#include <linux/of_fdt.h>
30
31#include <asm/page.h>
32#include <asm/fadump.h>
33#include <asm/fadump-internal.h>
34#include <asm/setup.h>
35#include <asm/interrupt.h>
36
37/*
38 * The CPU who acquired the lock to trigger the fadump crash should
39 * wait for other CPUs to enter.
40 *
41 * The timeout is in milliseconds.
42 */
43#define CRASH_TIMEOUT 500
44
45static struct fw_dump fw_dump;
46
47static void __init fadump_reserve_crash_area(u64 base);
48
49#ifndef CONFIG_PRESERVE_FA_DUMP
50
51static struct kobject *fadump_kobj;
52
53static atomic_t cpus_in_fadump;
54static DEFINE_MUTEX(fadump_mutex);
55
56#define RESERVED_RNGS_SZ 16384 /* 16K - 128 entries */
57#define RESERVED_RNGS_CNT (RESERVED_RNGS_SZ / \
58 sizeof(struct fadump_memory_range))
59static struct fadump_memory_range rngs[RESERVED_RNGS_CNT];
60static struct fadump_mrange_info
61reserved_mrange_info = { "reserved", rngs, RESERVED_RNGS_SZ, 0, RESERVED_RNGS_CNT, true };
62
63static void __init early_init_dt_scan_reserved_ranges(unsigned long node);
64
65#ifdef CONFIG_CMA
66static struct cma *fadump_cma;
67
68/*
69 * fadump_cma_init() - Initialize CMA area from a fadump reserved memory
70 *
71 * This function initializes CMA area from fadump reserved memory.
72 * The total size of fadump reserved memory covers for boot memory size
73 * + cpu data size + hpte size and metadata.
74 * Initialize only the area equivalent to boot memory size for CMA use.
75 * The remaining portion of fadump reserved memory will be not given
76 * to CMA and pages for those will stay reserved. boot memory size is
77 * aligned per CMA requirement to satisy cma_init_reserved_mem() call.
78 * But for some reason even if it fails we still have the memory reservation
79 * with us and we can still continue doing fadump.
80 */
81void __init fadump_cma_init(void)
82{
83 unsigned long long base, size, end;
84 int rc;
85
86 if (!fw_dump.fadump_supported || !fw_dump.fadump_enabled ||
87 fw_dump.dump_active)
88 return;
89 /*
90 * Do not use CMA if user has provided fadump=nocma kernel parameter.
91 */
92 if (fw_dump.nocma || !fw_dump.boot_memory_size)
93 return;
94
95 /*
96 * [base, end) should be reserved during early init in
97 * fadump_reserve_mem(). No need to check this here as
98 * cma_init_reserved_mem() already checks for overlap.
99 * Here we give the aligned chunk of this reserved memory to CMA.
100 */
101 base = fw_dump.reserve_dump_area_start;
102 size = fw_dump.boot_memory_size;
103 end = base + size;
104
105 base = ALIGN(base, CMA_MIN_ALIGNMENT_BYTES);
106 end = ALIGN_DOWN(end, CMA_MIN_ALIGNMENT_BYTES);
107 size = end - base;
108
109 if (end <= base) {
110 pr_warn("%s: Too less memory to give to CMA\n", __func__);
111 return;
112 }
113
114 rc = cma_init_reserved_mem(base, size, 0, "fadump_cma", &fadump_cma);
115 if (rc) {
116 pr_err("Failed to init cma area for firmware-assisted dump,%d\n", rc);
117 /*
118 * Though the CMA init has failed we still have memory
119 * reservation with us. The reserved memory will be
120 * blocked from production system usage. Hence return 1,
121 * so that we can continue with fadump.
122 */
123 return;
124 }
125
126 /*
127 * If CMA activation fails, keep the pages reserved, instead of
128 * exposing them to buddy allocator. Same as 'fadump=nocma' case.
129 */
130 cma_reserve_pages_on_error(fadump_cma);
131
132 /*
133 * So we now have successfully initialized cma area for fadump.
134 */
135 pr_info("Initialized [0x%llx, %luMB] cma area from [0x%lx, %luMB] "
136 "bytes of memory reserved for firmware-assisted dump\n",
137 cma_get_base(fadump_cma), cma_get_size(fadump_cma) >> 20,
138 fw_dump.reserve_dump_area_start,
139 fw_dump.boot_memory_size >> 20);
140 return;
141}
142#endif /* CONFIG_CMA */
143
144/*
145 * Additional parameters meant for capture kernel are placed in a dedicated area.
146 * If this is capture kernel boot, append these parameters to bootargs.
147 */
148void __init fadump_append_bootargs(void)
149{
150 char *append_args;
151 size_t len;
152
153 if (!fw_dump.dump_active || !fw_dump.param_area_supported || !fw_dump.param_area)
154 return;
155
156 if (fw_dump.param_area < fw_dump.boot_mem_top) {
157 if (memblock_reserve(fw_dump.param_area, COMMAND_LINE_SIZE)) {
158 pr_warn("WARNING: Can't use additional parameters area!\n");
159 fw_dump.param_area = 0;
160 return;
161 }
162 }
163
164 append_args = (char *)fw_dump.param_area;
165 len = strlen(boot_command_line);
166
167 /*
168 * Too late to fail even if cmdline size exceeds. Truncate additional parameters
169 * to cmdline size and proceed anyway.
170 */
171 if (len + strlen(append_args) >= COMMAND_LINE_SIZE - 1)
172 pr_warn("WARNING: Appending parameters exceeds cmdline size. Truncating!\n");
173
174 pr_debug("Cmdline: %s\n", boot_command_line);
175 snprintf(boot_command_line + len, COMMAND_LINE_SIZE - len, " %s", append_args);
176 pr_info("Updated cmdline: %s\n", boot_command_line);
177}
178
179/* Scan the Firmware Assisted dump configuration details. */
180int __init early_init_dt_scan_fw_dump(unsigned long node, const char *uname,
181 int depth, void *data)
182{
183 if (depth == 0) {
184 early_init_dt_scan_reserved_ranges(node);
185 return 0;
186 }
187
188 if (depth != 1)
189 return 0;
190
191 if (strcmp(uname, "rtas") == 0) {
192 rtas_fadump_dt_scan(&fw_dump, node);
193 return 1;
194 }
195
196 if (strcmp(uname, "ibm,opal") == 0) {
197 opal_fadump_dt_scan(&fw_dump, node);
198 return 1;
199 }
200
201 return 0;
202}
203
204/*
205 * If fadump is registered, check if the memory provided
206 * falls within boot memory area and reserved memory area.
207 */
208int is_fadump_memory_area(u64 addr, unsigned long size)
209{
210 u64 d_start, d_end;
211
212 if (!fw_dump.dump_registered)
213 return 0;
214
215 if (!size)
216 return 0;
217
218 d_start = fw_dump.reserve_dump_area_start;
219 d_end = d_start + fw_dump.reserve_dump_area_size;
220 if (((addr + size) > d_start) && (addr <= d_end))
221 return 1;
222
223 return (addr <= fw_dump.boot_mem_top);
224}
225
226int should_fadump_crash(void)
227{
228 if (!fw_dump.dump_registered || !fw_dump.fadumphdr_addr)
229 return 0;
230 return 1;
231}
232
233int is_fadump_active(void)
234{
235 return fw_dump.dump_active;
236}
237
238/*
239 * Returns true, if there are no holes in memory area between d_start to d_end,
240 * false otherwise.
241 */
242static bool is_fadump_mem_area_contiguous(u64 d_start, u64 d_end)
243{
244 phys_addr_t reg_start, reg_end;
245 bool ret = false;
246 u64 i, start, end;
247
248 for_each_mem_range(i, ®_start, ®_end) {
249 start = max_t(u64, d_start, reg_start);
250 end = min_t(u64, d_end, reg_end);
251 if (d_start < end) {
252 /* Memory hole from d_start to start */
253 if (start > d_start)
254 break;
255
256 if (end == d_end) {
257 ret = true;
258 break;
259 }
260
261 d_start = end + 1;
262 }
263 }
264
265 return ret;
266}
267
268/*
269 * Returns true, if there are no holes in reserved memory area,
270 * false otherwise.
271 */
272bool is_fadump_reserved_mem_contiguous(void)
273{
274 u64 d_start, d_end;
275
276 d_start = fw_dump.reserve_dump_area_start;
277 d_end = d_start + fw_dump.reserve_dump_area_size;
278 return is_fadump_mem_area_contiguous(d_start, d_end);
279}
280
281/* Print firmware assisted dump configurations for debugging purpose. */
282static void __init fadump_show_config(void)
283{
284 int i;
285
286 pr_debug("Support for firmware-assisted dump (fadump): %s\n",
287 (fw_dump.fadump_supported ? "present" : "no support"));
288
289 if (!fw_dump.fadump_supported)
290 return;
291
292 pr_debug("Fadump enabled : %s\n",
293 (fw_dump.fadump_enabled ? "yes" : "no"));
294 pr_debug("Dump Active : %s\n",
295 (fw_dump.dump_active ? "yes" : "no"));
296 pr_debug("Dump section sizes:\n");
297 pr_debug(" CPU state data size: %lx\n", fw_dump.cpu_state_data_size);
298 pr_debug(" HPTE region size : %lx\n", fw_dump.hpte_region_size);
299 pr_debug(" Boot memory size : %lx\n", fw_dump.boot_memory_size);
300 pr_debug(" Boot memory top : %llx\n", fw_dump.boot_mem_top);
301 pr_debug("Boot memory regions cnt: %llx\n", fw_dump.boot_mem_regs_cnt);
302 for (i = 0; i < fw_dump.boot_mem_regs_cnt; i++) {
303 pr_debug("[%03d] base = %llx, size = %llx\n", i,
304 fw_dump.boot_mem_addr[i], fw_dump.boot_mem_sz[i]);
305 }
306}
307
308/**
309 * fadump_calculate_reserve_size(): reserve variable boot area 5% of System RAM
310 *
311 * Function to find the largest memory size we need to reserve during early
312 * boot process. This will be the size of the memory that is required for a
313 * kernel to boot successfully.
314 *
315 * This function has been taken from phyp-assisted dump feature implementation.
316 *
317 * returns larger of 256MB or 5% rounded down to multiples of 256MB.
318 *
319 * TODO: Come up with better approach to find out more accurate memory size
320 * that is required for a kernel to boot successfully.
321 *
322 */
323static __init u64 fadump_calculate_reserve_size(void)
324{
325 u64 base, size, bootmem_min;
326 int ret;
327
328 if (fw_dump.reserve_bootvar)
329 pr_warn("'fadump_reserve_mem=' parameter is deprecated in favor of 'crashkernel=' parameter.\n");
330
331 /*
332 * Check if the size is specified through crashkernel= cmdline
333 * option. If yes, then use that but ignore base as fadump reserves
334 * memory at a predefined offset.
335 */
336 ret = parse_crashkernel(boot_command_line, memblock_phys_mem_size(),
337 &size, &base, NULL, NULL);
338 if (ret == 0 && size > 0) {
339 unsigned long max_size;
340
341 if (fw_dump.reserve_bootvar)
342 pr_info("Using 'crashkernel=' parameter for memory reservation.\n");
343
344 fw_dump.reserve_bootvar = (unsigned long)size;
345
346 /*
347 * Adjust if the boot memory size specified is above
348 * the upper limit.
349 */
350 max_size = memblock_phys_mem_size() / MAX_BOOT_MEM_RATIO;
351 if (fw_dump.reserve_bootvar > max_size) {
352 fw_dump.reserve_bootvar = max_size;
353 pr_info("Adjusted boot memory size to %luMB\n",
354 (fw_dump.reserve_bootvar >> 20));
355 }
356
357 return fw_dump.reserve_bootvar;
358 } else if (fw_dump.reserve_bootvar) {
359 /*
360 * 'fadump_reserve_mem=' is being used to reserve memory
361 * for firmware-assisted dump.
362 */
363 return fw_dump.reserve_bootvar;
364 }
365
366 /* divide by 20 to get 5% of value */
367 size = memblock_phys_mem_size() / 20;
368
369 /* round it down in multiples of 256 */
370 size = size & ~0x0FFFFFFFUL;
371
372 /* Truncate to memory_limit. We don't want to over reserve the memory.*/
373 if (memory_limit && size > memory_limit)
374 size = memory_limit;
375
376 bootmem_min = fw_dump.ops->fadump_get_bootmem_min();
377 return (size > bootmem_min ? size : bootmem_min);
378}
379
380/*
381 * Calculate the total memory size required to be reserved for
382 * firmware-assisted dump registration.
383 */
384static unsigned long __init get_fadump_area_size(void)
385{
386 unsigned long size = 0;
387
388 size += fw_dump.cpu_state_data_size;
389 size += fw_dump.hpte_region_size;
390 /*
391 * Account for pagesize alignment of boot memory area destination address.
392 * This faciliates in mmap reading of first kernel's memory.
393 */
394 size = PAGE_ALIGN(size);
395 size += fw_dump.boot_memory_size;
396 size += sizeof(struct fadump_crash_info_header);
397
398 /* This is to hold kernel metadata on platforms that support it */
399 size += (fw_dump.ops->fadump_get_metadata_size ?
400 fw_dump.ops->fadump_get_metadata_size() : 0);
401 return size;
402}
403
404static int __init add_boot_mem_region(unsigned long rstart,
405 unsigned long rsize)
406{
407 int max_boot_mem_rgns = fw_dump.ops->fadump_max_boot_mem_rgns();
408 int i = fw_dump.boot_mem_regs_cnt++;
409
410 if (fw_dump.boot_mem_regs_cnt > max_boot_mem_rgns) {
411 fw_dump.boot_mem_regs_cnt = max_boot_mem_rgns;
412 return 0;
413 }
414
415 pr_debug("Added boot memory range[%d] [%#016lx-%#016lx)\n",
416 i, rstart, (rstart + rsize));
417 fw_dump.boot_mem_addr[i] = rstart;
418 fw_dump.boot_mem_sz[i] = rsize;
419 return 1;
420}
421
422/*
423 * Firmware usually has a hard limit on the data it can copy per region.
424 * Honour that by splitting a memory range into multiple regions.
425 */
426static int __init add_boot_mem_regions(unsigned long mstart,
427 unsigned long msize)
428{
429 unsigned long rstart, rsize, max_size;
430 int ret = 1;
431
432 rstart = mstart;
433 max_size = fw_dump.max_copy_size ? fw_dump.max_copy_size : msize;
434 while (msize) {
435 if (msize > max_size)
436 rsize = max_size;
437 else
438 rsize = msize;
439
440 ret = add_boot_mem_region(rstart, rsize);
441 if (!ret)
442 break;
443
444 msize -= rsize;
445 rstart += rsize;
446 }
447
448 return ret;
449}
450
451static int __init fadump_get_boot_mem_regions(void)
452{
453 unsigned long size, cur_size, hole_size, last_end;
454 unsigned long mem_size = fw_dump.boot_memory_size;
455 phys_addr_t reg_start, reg_end;
456 int ret = 1;
457 u64 i;
458
459 fw_dump.boot_mem_regs_cnt = 0;
460
461 last_end = 0;
462 hole_size = 0;
463 cur_size = 0;
464 for_each_mem_range(i, ®_start, ®_end) {
465 size = reg_end - reg_start;
466 hole_size += (reg_start - last_end);
467
468 if ((cur_size + size) >= mem_size) {
469 size = (mem_size - cur_size);
470 ret = add_boot_mem_regions(reg_start, size);
471 break;
472 }
473
474 mem_size -= size;
475 cur_size += size;
476 ret = add_boot_mem_regions(reg_start, size);
477 if (!ret)
478 break;
479
480 last_end = reg_end;
481 }
482 fw_dump.boot_mem_top = PAGE_ALIGN(fw_dump.boot_memory_size + hole_size);
483
484 return ret;
485}
486
487/*
488 * Returns true, if the given range overlaps with reserved memory ranges
489 * starting at idx. Also, updates idx to index of overlapping memory range
490 * with the given memory range.
491 * False, otherwise.
492 */
493static bool __init overlaps_reserved_ranges(u64 base, u64 end, int *idx)
494{
495 bool ret = false;
496 int i;
497
498 for (i = *idx; i < reserved_mrange_info.mem_range_cnt; i++) {
499 u64 rbase = reserved_mrange_info.mem_ranges[i].base;
500 u64 rend = rbase + reserved_mrange_info.mem_ranges[i].size;
501
502 if (end <= rbase)
503 break;
504
505 if ((end > rbase) && (base < rend)) {
506 *idx = i;
507 ret = true;
508 break;
509 }
510 }
511
512 return ret;
513}
514
515/*
516 * Locate a suitable memory area to reserve memory for FADump. While at it,
517 * lookup reserved-ranges & avoid overlap with them, as they are used by F/W.
518 */
519static u64 __init fadump_locate_reserve_mem(u64 base, u64 size)
520{
521 struct fadump_memory_range *mrngs;
522 phys_addr_t mstart, mend;
523 int idx = 0;
524 u64 i, ret = 0;
525
526 mrngs = reserved_mrange_info.mem_ranges;
527 for_each_free_mem_range(i, NUMA_NO_NODE, MEMBLOCK_NONE,
528 &mstart, &mend, NULL) {
529 pr_debug("%llu) mstart: %llx, mend: %llx, base: %llx\n",
530 i, mstart, mend, base);
531
532 if (mstart > base)
533 base = PAGE_ALIGN(mstart);
534
535 while ((mend > base) && ((mend - base) >= size)) {
536 if (!overlaps_reserved_ranges(base, base+size, &idx)) {
537 ret = base;
538 goto out;
539 }
540
541 base = mrngs[idx].base + mrngs[idx].size;
542 base = PAGE_ALIGN(base);
543 }
544 }
545
546out:
547 return ret;
548}
549
550int __init fadump_reserve_mem(void)
551{
552 u64 base, size, mem_boundary, bootmem_min;
553 int ret = 1;
554
555 if (!fw_dump.fadump_enabled)
556 return 0;
557
558 if (!fw_dump.fadump_supported) {
559 pr_info("Firmware-Assisted Dump is not supported on this hardware\n");
560 goto error_out;
561 }
562
563 /*
564 * Initialize boot memory size
565 * If dump is active then we have already calculated the size during
566 * first kernel.
567 */
568 if (!fw_dump.dump_active) {
569 fw_dump.boot_memory_size =
570 PAGE_ALIGN(fadump_calculate_reserve_size());
571
572 bootmem_min = fw_dump.ops->fadump_get_bootmem_min();
573 if (fw_dump.boot_memory_size < bootmem_min) {
574 pr_err("Can't enable fadump with boot memory size (0x%lx) less than 0x%llx\n",
575 fw_dump.boot_memory_size, bootmem_min);
576 goto error_out;
577 }
578
579 if (!fadump_get_boot_mem_regions()) {
580 pr_err("Too many holes in boot memory area to enable fadump\n");
581 goto error_out;
582 }
583 }
584
585 if (memory_limit)
586 mem_boundary = memory_limit;
587 else
588 mem_boundary = memblock_end_of_DRAM();
589
590 base = fw_dump.boot_mem_top;
591 size = get_fadump_area_size();
592 fw_dump.reserve_dump_area_size = size;
593 if (fw_dump.dump_active) {
594 pr_info("Firmware-assisted dump is active.\n");
595
596#ifdef CONFIG_HUGETLB_PAGE
597 /*
598 * FADump capture kernel doesn't care much about hugepages.
599 * In fact, handling hugepages in capture kernel is asking for
600 * trouble. So, disable HugeTLB support when fadump is active.
601 */
602 hugetlb_disabled = true;
603#endif
604 /*
605 * If last boot has crashed then reserve all the memory
606 * above boot memory size so that we don't touch it until
607 * dump is written to disk by userspace tool. This memory
608 * can be released for general use by invalidating fadump.
609 */
610 fadump_reserve_crash_area(base);
611
612 pr_debug("fadumphdr_addr = %#016lx\n", fw_dump.fadumphdr_addr);
613 pr_debug("Reserve dump area start address: 0x%lx\n",
614 fw_dump.reserve_dump_area_start);
615 } else {
616 /*
617 * Reserve memory at an offset closer to bottom of the RAM to
618 * minimize the impact of memory hot-remove operation.
619 */
620 base = fadump_locate_reserve_mem(base, size);
621
622 if (!base || (base + size > mem_boundary)) {
623 pr_err("Failed to find memory chunk for reservation!\n");
624 goto error_out;
625 }
626 fw_dump.reserve_dump_area_start = base;
627
628 /*
629 * Calculate the kernel metadata address and register it with
630 * f/w if the platform supports.
631 */
632 if (fw_dump.ops->fadump_setup_metadata &&
633 (fw_dump.ops->fadump_setup_metadata(&fw_dump) < 0))
634 goto error_out;
635
636 if (memblock_reserve(base, size)) {
637 pr_err("Failed to reserve memory!\n");
638 goto error_out;
639 }
640
641 pr_info("Reserved %lldMB of memory at %#016llx (System RAM: %lldMB)\n",
642 (size >> 20), base, (memblock_phys_mem_size() >> 20));
643 }
644
645 return ret;
646error_out:
647 fw_dump.fadump_enabled = 0;
648 fw_dump.reserve_dump_area_size = 0;
649 return 0;
650}
651
652/* Look for fadump= cmdline option. */
653static int __init early_fadump_param(char *p)
654{
655 if (!p)
656 return 1;
657
658 if (strncmp(p, "on", 2) == 0)
659 fw_dump.fadump_enabled = 1;
660 else if (strncmp(p, "off", 3) == 0)
661 fw_dump.fadump_enabled = 0;
662 else if (strncmp(p, "nocma", 5) == 0) {
663 fw_dump.fadump_enabled = 1;
664 fw_dump.nocma = 1;
665 }
666
667 return 0;
668}
669early_param("fadump", early_fadump_param);
670
671/*
672 * Look for fadump_reserve_mem= cmdline option
673 * TODO: Remove references to 'fadump_reserve_mem=' parameter,
674 * the sooner 'crashkernel=' parameter is accustomed to.
675 */
676static int __init early_fadump_reserve_mem(char *p)
677{
678 if (p)
679 fw_dump.reserve_bootvar = memparse(p, &p);
680 return 0;
681}
682early_param("fadump_reserve_mem", early_fadump_reserve_mem);
683
684void crash_fadump(struct pt_regs *regs, const char *str)
685{
686 unsigned int msecs;
687 struct fadump_crash_info_header *fdh = NULL;
688 int old_cpu, this_cpu;
689 /* Do not include first CPU */
690 unsigned int ncpus = num_online_cpus() - 1;
691
692 if (!should_fadump_crash())
693 return;
694
695 /*
696 * old_cpu == -1 means this is the first CPU which has come here,
697 * go ahead and trigger fadump.
698 *
699 * old_cpu != -1 means some other CPU has already on its way
700 * to trigger fadump, just keep looping here.
701 */
702 this_cpu = smp_processor_id();
703 old_cpu = cmpxchg(&crashing_cpu, -1, this_cpu);
704
705 if (old_cpu != -1) {
706 atomic_inc(&cpus_in_fadump);
707
708 /*
709 * We can't loop here indefinitely. Wait as long as fadump
710 * is in force. If we race with fadump un-registration this
711 * loop will break and then we go down to normal panic path
712 * and reboot. If fadump is in force the first crashing
713 * cpu will definitely trigger fadump.
714 */
715 while (fw_dump.dump_registered)
716 cpu_relax();
717 return;
718 }
719
720 fdh = __va(fw_dump.fadumphdr_addr);
721 fdh->crashing_cpu = crashing_cpu;
722 crash_save_vmcoreinfo();
723
724 if (regs)
725 fdh->regs = *regs;
726 else
727 ppc_save_regs(&fdh->regs);
728
729 fdh->cpu_mask = *cpu_online_mask;
730
731 /*
732 * If we came in via system reset, wait a while for the secondary
733 * CPUs to enter.
734 */
735 if (TRAP(&(fdh->regs)) == INTERRUPT_SYSTEM_RESET) {
736 msecs = CRASH_TIMEOUT;
737 while ((atomic_read(&cpus_in_fadump) < ncpus) && (--msecs > 0))
738 mdelay(1);
739 }
740
741 fw_dump.ops->fadump_trigger(fdh, str);
742}
743
744u32 *__init fadump_regs_to_elf_notes(u32 *buf, struct pt_regs *regs)
745{
746 struct elf_prstatus prstatus;
747
748 memset(&prstatus, 0, sizeof(prstatus));
749 /*
750 * FIXME: How do i get PID? Do I really need it?
751 * prstatus.pr_pid = ????
752 */
753 elf_core_copy_regs(&prstatus.pr_reg, regs);
754 buf = append_elf_note(buf, CRASH_CORE_NOTE_NAME, NT_PRSTATUS,
755 &prstatus, sizeof(prstatus));
756 return buf;
757}
758
759void __init fadump_update_elfcore_header(char *bufp)
760{
761 struct elf_phdr *phdr;
762
763 bufp += sizeof(struct elfhdr);
764
765 /* First note is a place holder for cpu notes info. */
766 phdr = (struct elf_phdr *)bufp;
767
768 if (phdr->p_type == PT_NOTE) {
769 phdr->p_paddr = __pa(fw_dump.cpu_notes_buf_vaddr);
770 phdr->p_offset = phdr->p_paddr;
771 phdr->p_filesz = fw_dump.cpu_notes_buf_size;
772 phdr->p_memsz = fw_dump.cpu_notes_buf_size;
773 }
774 return;
775}
776
777static void *__init fadump_alloc_buffer(unsigned long size)
778{
779 unsigned long count, i;
780 struct page *page;
781 void *vaddr;
782
783 vaddr = alloc_pages_exact(size, GFP_KERNEL | __GFP_ZERO);
784 if (!vaddr)
785 return NULL;
786
787 count = PAGE_ALIGN(size) / PAGE_SIZE;
788 page = virt_to_page(vaddr);
789 for (i = 0; i < count; i++)
790 mark_page_reserved(page + i);
791 return vaddr;
792}
793
794static void fadump_free_buffer(unsigned long vaddr, unsigned long size)
795{
796 free_reserved_area((void *)vaddr, (void *)(vaddr + size), -1, NULL);
797}
798
799s32 __init fadump_setup_cpu_notes_buf(u32 num_cpus)
800{
801 /* Allocate buffer to hold cpu crash notes. */
802 fw_dump.cpu_notes_buf_size = num_cpus * sizeof(note_buf_t);
803 fw_dump.cpu_notes_buf_size = PAGE_ALIGN(fw_dump.cpu_notes_buf_size);
804 fw_dump.cpu_notes_buf_vaddr =
805 (unsigned long)fadump_alloc_buffer(fw_dump.cpu_notes_buf_size);
806 if (!fw_dump.cpu_notes_buf_vaddr) {
807 pr_err("Failed to allocate %ld bytes for CPU notes buffer\n",
808 fw_dump.cpu_notes_buf_size);
809 return -ENOMEM;
810 }
811
812 pr_debug("Allocated buffer for cpu notes of size %ld at 0x%lx\n",
813 fw_dump.cpu_notes_buf_size,
814 fw_dump.cpu_notes_buf_vaddr);
815 return 0;
816}
817
818void fadump_free_cpu_notes_buf(void)
819{
820 if (!fw_dump.cpu_notes_buf_vaddr)
821 return;
822
823 fadump_free_buffer(fw_dump.cpu_notes_buf_vaddr,
824 fw_dump.cpu_notes_buf_size);
825 fw_dump.cpu_notes_buf_vaddr = 0;
826 fw_dump.cpu_notes_buf_size = 0;
827}
828
829static void fadump_free_mem_ranges(struct fadump_mrange_info *mrange_info)
830{
831 if (mrange_info->is_static) {
832 mrange_info->mem_range_cnt = 0;
833 return;
834 }
835
836 kfree(mrange_info->mem_ranges);
837 memset((void *)((u64)mrange_info + RNG_NAME_SZ), 0,
838 (sizeof(struct fadump_mrange_info) - RNG_NAME_SZ));
839}
840
841/*
842 * Allocate or reallocate mem_ranges array in incremental units
843 * of PAGE_SIZE.
844 */
845static int fadump_alloc_mem_ranges(struct fadump_mrange_info *mrange_info)
846{
847 struct fadump_memory_range *new_array;
848 u64 new_size;
849
850 new_size = mrange_info->mem_ranges_sz + PAGE_SIZE;
851 pr_debug("Allocating %llu bytes of memory for %s memory ranges\n",
852 new_size, mrange_info->name);
853
854 new_array = krealloc(mrange_info->mem_ranges, new_size, GFP_KERNEL);
855 if (new_array == NULL) {
856 pr_err("Insufficient memory for setting up %s memory ranges\n",
857 mrange_info->name);
858 fadump_free_mem_ranges(mrange_info);
859 return -ENOMEM;
860 }
861
862 mrange_info->mem_ranges = new_array;
863 mrange_info->mem_ranges_sz = new_size;
864 mrange_info->max_mem_ranges = (new_size /
865 sizeof(struct fadump_memory_range));
866 return 0;
867}
868static inline int fadump_add_mem_range(struct fadump_mrange_info *mrange_info,
869 u64 base, u64 end)
870{
871 struct fadump_memory_range *mem_ranges = mrange_info->mem_ranges;
872 bool is_adjacent = false;
873 u64 start, size;
874
875 if (base == end)
876 return 0;
877
878 /*
879 * Fold adjacent memory ranges to bring down the memory ranges/
880 * PT_LOAD segments count.
881 */
882 if (mrange_info->mem_range_cnt) {
883 start = mem_ranges[mrange_info->mem_range_cnt - 1].base;
884 size = mem_ranges[mrange_info->mem_range_cnt - 1].size;
885
886 /*
887 * Boot memory area needs separate PT_LOAD segment(s) as it
888 * is moved to a different location at the time of crash.
889 * So, fold only if the region is not boot memory area.
890 */
891 if ((start + size) == base && start >= fw_dump.boot_mem_top)
892 is_adjacent = true;
893 }
894 if (!is_adjacent) {
895 /* resize the array on reaching the limit */
896 if (mrange_info->mem_range_cnt == mrange_info->max_mem_ranges) {
897 int ret;
898
899 if (mrange_info->is_static) {
900 pr_err("Reached array size limit for %s memory ranges\n",
901 mrange_info->name);
902 return -ENOSPC;
903 }
904
905 ret = fadump_alloc_mem_ranges(mrange_info);
906 if (ret)
907 return ret;
908
909 /* Update to the new resized array */
910 mem_ranges = mrange_info->mem_ranges;
911 }
912
913 start = base;
914 mem_ranges[mrange_info->mem_range_cnt].base = start;
915 mrange_info->mem_range_cnt++;
916 }
917
918 mem_ranges[mrange_info->mem_range_cnt - 1].size = (end - start);
919 pr_debug("%s_memory_range[%d] [%#016llx-%#016llx], %#llx bytes\n",
920 mrange_info->name, (mrange_info->mem_range_cnt - 1),
921 start, end - 1, (end - start));
922 return 0;
923}
924
925static int fadump_init_elfcore_header(char *bufp)
926{
927 struct elfhdr *elf;
928
929 elf = (struct elfhdr *) bufp;
930 bufp += sizeof(struct elfhdr);
931 memcpy(elf->e_ident, ELFMAG, SELFMAG);
932 elf->e_ident[EI_CLASS] = ELF_CLASS;
933 elf->e_ident[EI_DATA] = ELF_DATA;
934 elf->e_ident[EI_VERSION] = EV_CURRENT;
935 elf->e_ident[EI_OSABI] = ELF_OSABI;
936 memset(elf->e_ident+EI_PAD, 0, EI_NIDENT-EI_PAD);
937 elf->e_type = ET_CORE;
938 elf->e_machine = ELF_ARCH;
939 elf->e_version = EV_CURRENT;
940 elf->e_entry = 0;
941 elf->e_phoff = sizeof(struct elfhdr);
942 elf->e_shoff = 0;
943
944 if (IS_ENABLED(CONFIG_PPC64_ELF_ABI_V2))
945 elf->e_flags = 2;
946 else if (IS_ENABLED(CONFIG_PPC64_ELF_ABI_V1))
947 elf->e_flags = 1;
948 else
949 elf->e_flags = 0;
950
951 elf->e_ehsize = sizeof(struct elfhdr);
952 elf->e_phentsize = sizeof(struct elf_phdr);
953 elf->e_phnum = 0;
954 elf->e_shentsize = 0;
955 elf->e_shnum = 0;
956 elf->e_shstrndx = 0;
957
958 return 0;
959}
960
961/*
962 * If the given physical address falls within the boot memory region then
963 * return the relocated address that points to the dump region reserved
964 * for saving initial boot memory contents.
965 */
966static inline unsigned long fadump_relocate(unsigned long paddr)
967{
968 unsigned long raddr, rstart, rend, rlast, hole_size;
969 int i;
970
971 hole_size = 0;
972 rlast = 0;
973 raddr = paddr;
974 for (i = 0; i < fw_dump.boot_mem_regs_cnt; i++) {
975 rstart = fw_dump.boot_mem_addr[i];
976 rend = rstart + fw_dump.boot_mem_sz[i];
977 hole_size += (rstart - rlast);
978
979 if (paddr >= rstart && paddr < rend) {
980 raddr += fw_dump.boot_mem_dest_addr - hole_size;
981 break;
982 }
983
984 rlast = rend;
985 }
986
987 pr_debug("vmcoreinfo: paddr = 0x%lx, raddr = 0x%lx\n", paddr, raddr);
988 return raddr;
989}
990
991static void __init populate_elf_pt_load(struct elf_phdr *phdr, u64 start,
992 u64 size, unsigned long long offset)
993{
994 phdr->p_align = 0;
995 phdr->p_memsz = size;
996 phdr->p_filesz = size;
997 phdr->p_paddr = start;
998 phdr->p_offset = offset;
999 phdr->p_type = PT_LOAD;
1000 phdr->p_flags = PF_R|PF_W|PF_X;
1001 phdr->p_vaddr = (unsigned long)__va(start);
1002}
1003
1004static void __init fadump_populate_elfcorehdr(struct fadump_crash_info_header *fdh)
1005{
1006 char *bufp;
1007 struct elfhdr *elf;
1008 struct elf_phdr *phdr;
1009 u64 boot_mem_dest_offset;
1010 unsigned long long i, ra_start, ra_end, ra_size, mstart, mend;
1011
1012 bufp = (char *) fw_dump.elfcorehdr_addr;
1013 fadump_init_elfcore_header(bufp);
1014 elf = (struct elfhdr *)bufp;
1015 bufp += sizeof(struct elfhdr);
1016
1017 /*
1018 * Set up ELF PT_NOTE, a placeholder for CPU notes information.
1019 * The notes info will be populated later by platform-specific code.
1020 * Hence, this PT_NOTE will always be the first ELF note.
1021 *
1022 * NOTE: Any new ELF note addition should be placed after this note.
1023 */
1024 phdr = (struct elf_phdr *)bufp;
1025 bufp += sizeof(struct elf_phdr);
1026 phdr->p_type = PT_NOTE;
1027 phdr->p_flags = 0;
1028 phdr->p_vaddr = 0;
1029 phdr->p_align = 0;
1030 phdr->p_offset = 0;
1031 phdr->p_paddr = 0;
1032 phdr->p_filesz = 0;
1033 phdr->p_memsz = 0;
1034 /* Increment number of program headers. */
1035 (elf->e_phnum)++;
1036
1037 /* setup ELF PT_NOTE for vmcoreinfo */
1038 phdr = (struct elf_phdr *)bufp;
1039 bufp += sizeof(struct elf_phdr);
1040 phdr->p_type = PT_NOTE;
1041 phdr->p_flags = 0;
1042 phdr->p_vaddr = 0;
1043 phdr->p_align = 0;
1044 phdr->p_paddr = phdr->p_offset = fdh->vmcoreinfo_raddr;
1045 phdr->p_memsz = phdr->p_filesz = fdh->vmcoreinfo_size;
1046 /* Increment number of program headers. */
1047 (elf->e_phnum)++;
1048
1049 /*
1050 * Setup PT_LOAD sections. first include boot memory regions
1051 * and then add rest of the memory regions.
1052 */
1053 boot_mem_dest_offset = fw_dump.boot_mem_dest_addr;
1054 for (i = 0; i < fw_dump.boot_mem_regs_cnt; i++) {
1055 phdr = (struct elf_phdr *)bufp;
1056 bufp += sizeof(struct elf_phdr);
1057 populate_elf_pt_load(phdr, fw_dump.boot_mem_addr[i],
1058 fw_dump.boot_mem_sz[i],
1059 boot_mem_dest_offset);
1060 /* Increment number of program headers. */
1061 (elf->e_phnum)++;
1062 boot_mem_dest_offset += fw_dump.boot_mem_sz[i];
1063 }
1064
1065 /* Memory reserved for fadump in first kernel */
1066 ra_start = fw_dump.reserve_dump_area_start;
1067 ra_size = get_fadump_area_size();
1068 ra_end = ra_start + ra_size;
1069
1070 phdr = (struct elf_phdr *)bufp;
1071 for_each_mem_range(i, &mstart, &mend) {
1072 /* Boot memory regions already added, skip them now */
1073 if (mstart < fw_dump.boot_mem_top) {
1074 if (mend > fw_dump.boot_mem_top)
1075 mstart = fw_dump.boot_mem_top;
1076 else
1077 continue;
1078 }
1079
1080 /* Handle memblock regions overlaps with fadump reserved area */
1081 if ((ra_start < mend) && (ra_end > mstart)) {
1082 if ((mstart < ra_start) && (mend > ra_end)) {
1083 populate_elf_pt_load(phdr, mstart, ra_start - mstart, mstart);
1084 /* Increment number of program headers. */
1085 (elf->e_phnum)++;
1086 bufp += sizeof(struct elf_phdr);
1087 phdr = (struct elf_phdr *)bufp;
1088 populate_elf_pt_load(phdr, ra_end, mend - ra_end, ra_end);
1089 } else if (mstart < ra_start) {
1090 populate_elf_pt_load(phdr, mstart, ra_start - mstart, mstart);
1091 } else if (ra_end < mend) {
1092 populate_elf_pt_load(phdr, ra_end, mend - ra_end, ra_end);
1093 }
1094 } else {
1095 /* No overlap with fadump reserved memory region */
1096 populate_elf_pt_load(phdr, mstart, mend - mstart, mstart);
1097 }
1098
1099 /* Increment number of program headers. */
1100 (elf->e_phnum)++;
1101 bufp += sizeof(struct elf_phdr);
1102 phdr = (struct elf_phdr *) bufp;
1103 }
1104}
1105
1106static unsigned long init_fadump_header(unsigned long addr)
1107{
1108 struct fadump_crash_info_header *fdh;
1109
1110 if (!addr)
1111 return 0;
1112
1113 fdh = __va(addr);
1114 addr += sizeof(struct fadump_crash_info_header);
1115
1116 memset(fdh, 0, sizeof(struct fadump_crash_info_header));
1117 fdh->magic_number = FADUMP_CRASH_INFO_MAGIC;
1118 fdh->version = FADUMP_HEADER_VERSION;
1119 /* We will set the crashing cpu id in crash_fadump() during crash. */
1120 fdh->crashing_cpu = FADUMP_CPU_UNKNOWN;
1121
1122 /*
1123 * The physical address and size of vmcoreinfo are required in the
1124 * second kernel to prepare elfcorehdr.
1125 */
1126 fdh->vmcoreinfo_raddr = fadump_relocate(paddr_vmcoreinfo_note());
1127 fdh->vmcoreinfo_size = VMCOREINFO_NOTE_SIZE;
1128
1129
1130 fdh->pt_regs_sz = sizeof(struct pt_regs);
1131 /*
1132 * When LPAR is terminated by PYHP, ensure all possible CPUs'
1133 * register data is processed while exporting the vmcore.
1134 */
1135 fdh->cpu_mask = *cpu_possible_mask;
1136 fdh->cpu_mask_sz = sizeof(struct cpumask);
1137
1138 return addr;
1139}
1140
1141static int register_fadump(void)
1142{
1143 unsigned long addr;
1144
1145 /*
1146 * If no memory is reserved then we can not register for firmware-
1147 * assisted dump.
1148 */
1149 if (!fw_dump.reserve_dump_area_size)
1150 return -ENODEV;
1151
1152 addr = fw_dump.fadumphdr_addr;
1153
1154 /* Initialize fadump crash info header. */
1155 addr = init_fadump_header(addr);
1156
1157 /* register the future kernel dump with firmware. */
1158 pr_debug("Registering for firmware-assisted kernel dump...\n");
1159 return fw_dump.ops->fadump_register(&fw_dump);
1160}
1161
1162void fadump_cleanup(void)
1163{
1164 if (!fw_dump.fadump_supported)
1165 return;
1166
1167 /* Invalidate the registration only if dump is active. */
1168 if (fw_dump.dump_active) {
1169 pr_debug("Invalidating firmware-assisted dump registration\n");
1170 fw_dump.ops->fadump_invalidate(&fw_dump);
1171 } else if (fw_dump.dump_registered) {
1172 /* Un-register Firmware-assisted dump if it was registered. */
1173 fw_dump.ops->fadump_unregister(&fw_dump);
1174 }
1175
1176 if (fw_dump.ops->fadump_cleanup)
1177 fw_dump.ops->fadump_cleanup(&fw_dump);
1178}
1179
1180static void fadump_free_reserved_memory(unsigned long start_pfn,
1181 unsigned long end_pfn)
1182{
1183 unsigned long pfn;
1184 unsigned long time_limit = jiffies + HZ;
1185
1186 pr_info("freeing reserved memory (0x%llx - 0x%llx)\n",
1187 PFN_PHYS(start_pfn), PFN_PHYS(end_pfn));
1188
1189 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
1190 free_reserved_page(pfn_to_page(pfn));
1191
1192 if (time_after(jiffies, time_limit)) {
1193 cond_resched();
1194 time_limit = jiffies + HZ;
1195 }
1196 }
1197}
1198
1199/*
1200 * Skip memory holes and free memory that was actually reserved.
1201 */
1202static void fadump_release_reserved_area(u64 start, u64 end)
1203{
1204 unsigned long reg_spfn, reg_epfn;
1205 u64 tstart, tend, spfn, epfn;
1206 int i;
1207
1208 spfn = PHYS_PFN(start);
1209 epfn = PHYS_PFN(end);
1210
1211 for_each_mem_pfn_range(i, MAX_NUMNODES, ®_spfn, ®_epfn, NULL) {
1212 tstart = max_t(u64, spfn, reg_spfn);
1213 tend = min_t(u64, epfn, reg_epfn);
1214
1215 if (tstart < tend) {
1216 fadump_free_reserved_memory(tstart, tend);
1217
1218 if (tend == epfn)
1219 break;
1220
1221 spfn = tend;
1222 }
1223 }
1224}
1225
1226/*
1227 * Sort the mem ranges in-place and merge adjacent ranges
1228 * to minimize the memory ranges count.
1229 */
1230static void sort_and_merge_mem_ranges(struct fadump_mrange_info *mrange_info)
1231{
1232 struct fadump_memory_range *mem_ranges;
1233 u64 base, size;
1234 int i, j, idx;
1235
1236 if (!reserved_mrange_info.mem_range_cnt)
1237 return;
1238
1239 /* Sort the memory ranges */
1240 mem_ranges = mrange_info->mem_ranges;
1241 for (i = 0; i < mrange_info->mem_range_cnt; i++) {
1242 idx = i;
1243 for (j = (i + 1); j < mrange_info->mem_range_cnt; j++) {
1244 if (mem_ranges[idx].base > mem_ranges[j].base)
1245 idx = j;
1246 }
1247 if (idx != i)
1248 swap(mem_ranges[idx], mem_ranges[i]);
1249 }
1250
1251 /* Merge adjacent reserved ranges */
1252 idx = 0;
1253 for (i = 1; i < mrange_info->mem_range_cnt; i++) {
1254 base = mem_ranges[i-1].base;
1255 size = mem_ranges[i-1].size;
1256 if (mem_ranges[i].base == (base + size))
1257 mem_ranges[idx].size += mem_ranges[i].size;
1258 else {
1259 idx++;
1260 if (i == idx)
1261 continue;
1262
1263 mem_ranges[idx] = mem_ranges[i];
1264 }
1265 }
1266 mrange_info->mem_range_cnt = idx + 1;
1267}
1268
1269/*
1270 * Scan reserved-ranges to consider them while reserving/releasing
1271 * memory for FADump.
1272 */
1273static void __init early_init_dt_scan_reserved_ranges(unsigned long node)
1274{
1275 const __be32 *prop;
1276 int len, ret = -1;
1277 unsigned long i;
1278
1279 /* reserved-ranges already scanned */
1280 if (reserved_mrange_info.mem_range_cnt != 0)
1281 return;
1282
1283 prop = of_get_flat_dt_prop(node, "reserved-ranges", &len);
1284 if (!prop)
1285 return;
1286
1287 /*
1288 * Each reserved range is an (address,size) pair, 2 cells each,
1289 * totalling 4 cells per range.
1290 */
1291 for (i = 0; i < len / (sizeof(*prop) * 4); i++) {
1292 u64 base, size;
1293
1294 base = of_read_number(prop + (i * 4) + 0, 2);
1295 size = of_read_number(prop + (i * 4) + 2, 2);
1296
1297 if (size) {
1298 ret = fadump_add_mem_range(&reserved_mrange_info,
1299 base, base + size);
1300 if (ret < 0) {
1301 pr_warn("some reserved ranges are ignored!\n");
1302 break;
1303 }
1304 }
1305 }
1306
1307 /* Compact reserved ranges */
1308 sort_and_merge_mem_ranges(&reserved_mrange_info);
1309}
1310
1311/*
1312 * Release the memory that was reserved during early boot to preserve the
1313 * crash'ed kernel's memory contents except reserved dump area (permanent
1314 * reservation) and reserved ranges used by F/W. The released memory will
1315 * be available for general use.
1316 */
1317static void fadump_release_memory(u64 begin, u64 end)
1318{
1319 u64 ra_start, ra_end, tstart;
1320 int i, ret;
1321
1322 ra_start = fw_dump.reserve_dump_area_start;
1323 ra_end = ra_start + fw_dump.reserve_dump_area_size;
1324
1325 /*
1326 * If reserved ranges array limit is hit, overwrite the last reserved
1327 * memory range with reserved dump area to ensure it is excluded from
1328 * the memory being released (reused for next FADump registration).
1329 */
1330 if (reserved_mrange_info.mem_range_cnt ==
1331 reserved_mrange_info.max_mem_ranges)
1332 reserved_mrange_info.mem_range_cnt--;
1333
1334 ret = fadump_add_mem_range(&reserved_mrange_info, ra_start, ra_end);
1335 if (ret != 0)
1336 return;
1337
1338 /* Get the reserved ranges list in order first. */
1339 sort_and_merge_mem_ranges(&reserved_mrange_info);
1340
1341 /* Exclude reserved ranges and release remaining memory */
1342 tstart = begin;
1343 for (i = 0; i < reserved_mrange_info.mem_range_cnt; i++) {
1344 ra_start = reserved_mrange_info.mem_ranges[i].base;
1345 ra_end = ra_start + reserved_mrange_info.mem_ranges[i].size;
1346
1347 if (tstart >= ra_end)
1348 continue;
1349
1350 if (tstart < ra_start)
1351 fadump_release_reserved_area(tstart, ra_start);
1352 tstart = ra_end;
1353 }
1354
1355 if (tstart < end)
1356 fadump_release_reserved_area(tstart, end);
1357}
1358
1359static void fadump_free_elfcorehdr_buf(void)
1360{
1361 if (fw_dump.elfcorehdr_addr == 0 || fw_dump.elfcorehdr_size == 0)
1362 return;
1363
1364 /*
1365 * Before freeing the memory of `elfcorehdr`, reset the global
1366 * `elfcorehdr_addr` to prevent modules like `vmcore` from accessing
1367 * invalid memory.
1368 */
1369 elfcorehdr_addr = ELFCORE_ADDR_ERR;
1370 fadump_free_buffer(fw_dump.elfcorehdr_addr, fw_dump.elfcorehdr_size);
1371 fw_dump.elfcorehdr_addr = 0;
1372 fw_dump.elfcorehdr_size = 0;
1373}
1374
1375static void fadump_invalidate_release_mem(void)
1376{
1377 mutex_lock(&fadump_mutex);
1378 if (!fw_dump.dump_active) {
1379 mutex_unlock(&fadump_mutex);
1380 return;
1381 }
1382
1383 fadump_cleanup();
1384 mutex_unlock(&fadump_mutex);
1385
1386 fadump_free_elfcorehdr_buf();
1387 fadump_release_memory(fw_dump.boot_mem_top, memblock_end_of_DRAM());
1388 fadump_free_cpu_notes_buf();
1389
1390 /*
1391 * Setup kernel metadata and initialize the kernel dump
1392 * memory structure for FADump re-registration.
1393 */
1394 if (fw_dump.ops->fadump_setup_metadata &&
1395 (fw_dump.ops->fadump_setup_metadata(&fw_dump) < 0))
1396 pr_warn("Failed to setup kernel metadata!\n");
1397 fw_dump.ops->fadump_init_mem_struct(&fw_dump);
1398}
1399
1400static ssize_t release_mem_store(struct kobject *kobj,
1401 struct kobj_attribute *attr,
1402 const char *buf, size_t count)
1403{
1404 int input = -1;
1405
1406 if (!fw_dump.dump_active)
1407 return -EPERM;
1408
1409 if (kstrtoint(buf, 0, &input))
1410 return -EINVAL;
1411
1412 if (input == 1) {
1413 /*
1414 * Take away the '/proc/vmcore'. We are releasing the dump
1415 * memory, hence it will not be valid anymore.
1416 */
1417#ifdef CONFIG_PROC_VMCORE
1418 vmcore_cleanup();
1419#endif
1420 fadump_invalidate_release_mem();
1421
1422 } else
1423 return -EINVAL;
1424 return count;
1425}
1426
1427/* Release the reserved memory and disable the FADump */
1428static void __init unregister_fadump(void)
1429{
1430 fadump_cleanup();
1431 fadump_release_memory(fw_dump.reserve_dump_area_start,
1432 fw_dump.reserve_dump_area_size);
1433 fw_dump.fadump_enabled = 0;
1434 kobject_put(fadump_kobj);
1435}
1436
1437static ssize_t enabled_show(struct kobject *kobj,
1438 struct kobj_attribute *attr,
1439 char *buf)
1440{
1441 return sprintf(buf, "%d\n", fw_dump.fadump_enabled);
1442}
1443
1444/*
1445 * /sys/kernel/fadump/hotplug_ready sysfs node returns 1, which inidcates
1446 * to usersapce that fadump re-registration is not required on memory
1447 * hotplug events.
1448 */
1449static ssize_t hotplug_ready_show(struct kobject *kobj,
1450 struct kobj_attribute *attr,
1451 char *buf)
1452{
1453 return sprintf(buf, "%d\n", 1);
1454}
1455
1456static ssize_t mem_reserved_show(struct kobject *kobj,
1457 struct kobj_attribute *attr,
1458 char *buf)
1459{
1460 return sprintf(buf, "%ld\n", fw_dump.reserve_dump_area_size);
1461}
1462
1463static ssize_t registered_show(struct kobject *kobj,
1464 struct kobj_attribute *attr,
1465 char *buf)
1466{
1467 return sprintf(buf, "%d\n", fw_dump.dump_registered);
1468}
1469
1470static ssize_t bootargs_append_show(struct kobject *kobj,
1471 struct kobj_attribute *attr,
1472 char *buf)
1473{
1474 return sprintf(buf, "%s\n", (char *)__va(fw_dump.param_area));
1475}
1476
1477static ssize_t bootargs_append_store(struct kobject *kobj,
1478 struct kobj_attribute *attr,
1479 const char *buf, size_t count)
1480{
1481 char *params;
1482
1483 if (!fw_dump.fadump_enabled || fw_dump.dump_active)
1484 return -EPERM;
1485
1486 if (count >= COMMAND_LINE_SIZE)
1487 return -EINVAL;
1488
1489 /*
1490 * Fail here instead of handling this scenario with
1491 * some silly workaround in capture kernel.
1492 */
1493 if (saved_command_line_len + count >= COMMAND_LINE_SIZE) {
1494 pr_err("Appending parameters exceeds cmdline size!\n");
1495 return -ENOSPC;
1496 }
1497
1498 params = __va(fw_dump.param_area);
1499 strscpy_pad(params, buf, COMMAND_LINE_SIZE);
1500 /* Remove newline character at the end. */
1501 if (params[count-1] == '\n')
1502 params[count-1] = '\0';
1503
1504 return count;
1505}
1506
1507static ssize_t registered_store(struct kobject *kobj,
1508 struct kobj_attribute *attr,
1509 const char *buf, size_t count)
1510{
1511 int ret = 0;
1512 int input = -1;
1513
1514 if (!fw_dump.fadump_enabled || fw_dump.dump_active)
1515 return -EPERM;
1516
1517 if (kstrtoint(buf, 0, &input))
1518 return -EINVAL;
1519
1520 mutex_lock(&fadump_mutex);
1521
1522 switch (input) {
1523 case 0:
1524 if (fw_dump.dump_registered == 0) {
1525 goto unlock_out;
1526 }
1527
1528 /* Un-register Firmware-assisted dump */
1529 pr_debug("Un-register firmware-assisted dump\n");
1530 fw_dump.ops->fadump_unregister(&fw_dump);
1531 break;
1532 case 1:
1533 if (fw_dump.dump_registered == 1) {
1534 /* Un-register Firmware-assisted dump */
1535 fw_dump.ops->fadump_unregister(&fw_dump);
1536 }
1537 /* Register Firmware-assisted dump */
1538 ret = register_fadump();
1539 break;
1540 default:
1541 ret = -EINVAL;
1542 break;
1543 }
1544
1545unlock_out:
1546 mutex_unlock(&fadump_mutex);
1547 return ret < 0 ? ret : count;
1548}
1549
1550static int fadump_region_show(struct seq_file *m, void *private)
1551{
1552 if (!fw_dump.fadump_enabled)
1553 return 0;
1554
1555 mutex_lock(&fadump_mutex);
1556 fw_dump.ops->fadump_region_show(&fw_dump, m);
1557 mutex_unlock(&fadump_mutex);
1558 return 0;
1559}
1560
1561static struct kobj_attribute release_attr = __ATTR_WO(release_mem);
1562static struct kobj_attribute enable_attr = __ATTR_RO(enabled);
1563static struct kobj_attribute register_attr = __ATTR_RW(registered);
1564static struct kobj_attribute mem_reserved_attr = __ATTR_RO(mem_reserved);
1565static struct kobj_attribute hotplug_ready_attr = __ATTR_RO(hotplug_ready);
1566static struct kobj_attribute bootargs_append_attr = __ATTR_RW(bootargs_append);
1567
1568static struct attribute *fadump_attrs[] = {
1569 &enable_attr.attr,
1570 ®ister_attr.attr,
1571 &mem_reserved_attr.attr,
1572 &hotplug_ready_attr.attr,
1573 NULL,
1574};
1575
1576ATTRIBUTE_GROUPS(fadump);
1577
1578DEFINE_SHOW_ATTRIBUTE(fadump_region);
1579
1580static void __init fadump_init_files(void)
1581{
1582 int rc = 0;
1583
1584 fadump_kobj = kobject_create_and_add("fadump", kernel_kobj);
1585 if (!fadump_kobj) {
1586 pr_err("failed to create fadump kobject\n");
1587 return;
1588 }
1589
1590 if (fw_dump.param_area) {
1591 rc = sysfs_create_file(fadump_kobj, &bootargs_append_attr.attr);
1592 if (rc)
1593 pr_err("unable to create bootargs_append sysfs file (%d)\n", rc);
1594 }
1595
1596 debugfs_create_file("fadump_region", 0444, arch_debugfs_dir, NULL,
1597 &fadump_region_fops);
1598
1599 if (fw_dump.dump_active) {
1600 rc = sysfs_create_file(fadump_kobj, &release_attr.attr);
1601 if (rc)
1602 pr_err("unable to create release_mem sysfs file (%d)\n",
1603 rc);
1604 }
1605
1606 rc = sysfs_create_groups(fadump_kobj, fadump_groups);
1607 if (rc) {
1608 pr_err("sysfs group creation failed (%d), unregistering FADump",
1609 rc);
1610 unregister_fadump();
1611 return;
1612 }
1613
1614 /*
1615 * The FADump sysfs are moved from kernel_kobj to fadump_kobj need to
1616 * create symlink at old location to maintain backward compatibility.
1617 *
1618 * - fadump_enabled -> fadump/enabled
1619 * - fadump_registered -> fadump/registered
1620 * - fadump_release_mem -> fadump/release_mem
1621 */
1622 rc = compat_only_sysfs_link_entry_to_kobj(kernel_kobj, fadump_kobj,
1623 "enabled", "fadump_enabled");
1624 if (rc) {
1625 pr_err("unable to create fadump_enabled symlink (%d)", rc);
1626 return;
1627 }
1628
1629 rc = compat_only_sysfs_link_entry_to_kobj(kernel_kobj, fadump_kobj,
1630 "registered",
1631 "fadump_registered");
1632 if (rc) {
1633 pr_err("unable to create fadump_registered symlink (%d)", rc);
1634 sysfs_remove_link(kernel_kobj, "fadump_enabled");
1635 return;
1636 }
1637
1638 if (fw_dump.dump_active) {
1639 rc = compat_only_sysfs_link_entry_to_kobj(kernel_kobj,
1640 fadump_kobj,
1641 "release_mem",
1642 "fadump_release_mem");
1643 if (rc)
1644 pr_err("unable to create fadump_release_mem symlink (%d)",
1645 rc);
1646 }
1647 return;
1648}
1649
1650static int __init fadump_setup_elfcorehdr_buf(void)
1651{
1652 int elf_phdr_cnt;
1653 unsigned long elfcorehdr_size;
1654
1655 /*
1656 * Program header for CPU notes comes first, followed by one for
1657 * vmcoreinfo, and the remaining program headers correspond to
1658 * memory regions.
1659 */
1660 elf_phdr_cnt = 2 + fw_dump.boot_mem_regs_cnt + memblock_num_regions(memory);
1661 elfcorehdr_size = sizeof(struct elfhdr) + (elf_phdr_cnt * sizeof(struct elf_phdr));
1662 elfcorehdr_size = PAGE_ALIGN(elfcorehdr_size);
1663
1664 fw_dump.elfcorehdr_addr = (u64)fadump_alloc_buffer(elfcorehdr_size);
1665 if (!fw_dump.elfcorehdr_addr) {
1666 pr_err("Failed to allocate %lu bytes for elfcorehdr\n",
1667 elfcorehdr_size);
1668 return -ENOMEM;
1669 }
1670 fw_dump.elfcorehdr_size = elfcorehdr_size;
1671 return 0;
1672}
1673
1674/*
1675 * Check if the fadump header of crashed kernel is compatible with fadump kernel.
1676 *
1677 * It checks the magic number, endianness, and size of non-primitive type
1678 * members of fadump header to ensure safe dump collection.
1679 */
1680static bool __init is_fadump_header_compatible(struct fadump_crash_info_header *fdh)
1681{
1682 if (fdh->magic_number == FADUMP_CRASH_INFO_MAGIC_OLD) {
1683 pr_err("Old magic number, can't process the dump.\n");
1684 return false;
1685 }
1686
1687 if (fdh->magic_number != FADUMP_CRASH_INFO_MAGIC) {
1688 if (fdh->magic_number == swab64(FADUMP_CRASH_INFO_MAGIC))
1689 pr_err("Endianness mismatch between the crashed and fadump kernels.\n");
1690 else
1691 pr_err("Fadump header is corrupted.\n");
1692
1693 return false;
1694 }
1695
1696 /*
1697 * Dump collection is not safe if the size of non-primitive type members
1698 * of the fadump header do not match between crashed and fadump kernel.
1699 */
1700 if (fdh->pt_regs_sz != sizeof(struct pt_regs) ||
1701 fdh->cpu_mask_sz != sizeof(struct cpumask)) {
1702 pr_err("Fadump header size mismatch.\n");
1703 return false;
1704 }
1705
1706 return true;
1707}
1708
1709static void __init fadump_process(void)
1710{
1711 struct fadump_crash_info_header *fdh;
1712
1713 fdh = (struct fadump_crash_info_header *) __va(fw_dump.fadumphdr_addr);
1714 if (!fdh) {
1715 pr_err("Crash info header is empty.\n");
1716 goto err_out;
1717 }
1718
1719 /* Avoid processing the dump if fadump header isn't compatible */
1720 if (!is_fadump_header_compatible(fdh))
1721 goto err_out;
1722
1723 /* Allocate buffer for elfcorehdr */
1724 if (fadump_setup_elfcorehdr_buf())
1725 goto err_out;
1726
1727 fadump_populate_elfcorehdr(fdh);
1728
1729 /* Let platform update the CPU notes in elfcorehdr */
1730 if (fw_dump.ops->fadump_process(&fw_dump) < 0)
1731 goto err_out;
1732
1733 /*
1734 * elfcorehdr is now ready to be exported.
1735 *
1736 * set elfcorehdr_addr so that vmcore module will export the
1737 * elfcorehdr through '/proc/vmcore'.
1738 */
1739 elfcorehdr_addr = virt_to_phys((void *)fw_dump.elfcorehdr_addr);
1740 return;
1741
1742err_out:
1743 fadump_invalidate_release_mem();
1744}
1745
1746/*
1747 * Reserve memory to store additional parameters to be passed
1748 * for fadump/capture kernel.
1749 */
1750void __init fadump_setup_param_area(void)
1751{
1752 phys_addr_t range_start, range_end;
1753
1754 if (!fw_dump.param_area_supported || fw_dump.dump_active)
1755 return;
1756
1757 /* This memory can't be used by PFW or bootloader as it is shared across kernels */
1758 if (early_radix_enabled()) {
1759 /*
1760 * Anywhere in the upper half should be good enough as all memory
1761 * is accessible in real mode.
1762 */
1763 range_start = memblock_end_of_DRAM() / 2;
1764 range_end = memblock_end_of_DRAM();
1765 } else {
1766 /*
1767 * Passing additional parameters is supported for hash MMU only
1768 * if the first memory block size is 768MB or higher.
1769 */
1770 if (ppc64_rma_size < 0x30000000)
1771 return;
1772
1773 /*
1774 * 640 MB to 768 MB is not used by PFW/bootloader. So, try reserving
1775 * memory for passing additional parameters in this range to avoid
1776 * being stomped on by PFW/bootloader.
1777 */
1778 range_start = 0x2A000000;
1779 range_end = range_start + 0x4000000;
1780 }
1781
1782 fw_dump.param_area = memblock_phys_alloc_range(COMMAND_LINE_SIZE,
1783 COMMAND_LINE_SIZE,
1784 range_start,
1785 range_end);
1786 if (!fw_dump.param_area) {
1787 pr_warn("WARNING: Could not setup area to pass additional parameters!\n");
1788 return;
1789 }
1790
1791 memset((void *)fw_dump.param_area, 0, COMMAND_LINE_SIZE);
1792}
1793
1794/*
1795 * Prepare for firmware-assisted dump.
1796 */
1797int __init setup_fadump(void)
1798{
1799 if (!fw_dump.fadump_supported)
1800 return 0;
1801
1802 fadump_init_files();
1803 fadump_show_config();
1804
1805 if (!fw_dump.fadump_enabled)
1806 return 1;
1807
1808 /*
1809 * If dump data is available then see if it is valid and prepare for
1810 * saving it to the disk.
1811 */
1812 if (fw_dump.dump_active) {
1813 fadump_process();
1814 }
1815 /* Initialize the kernel dump memory structure and register with f/w */
1816 else if (fw_dump.reserve_dump_area_size) {
1817 fw_dump.ops->fadump_init_mem_struct(&fw_dump);
1818 register_fadump();
1819 }
1820
1821 /*
1822 * In case of panic, fadump is triggered via ppc_panic_event()
1823 * panic notifier. Setting crash_kexec_post_notifiers to 'true'
1824 * lets panic() function take crash friendly path before panic
1825 * notifiers are invoked.
1826 */
1827 crash_kexec_post_notifiers = true;
1828
1829 return 1;
1830}
1831/*
1832 * Use subsys_initcall_sync() here because there is dependency with
1833 * crash_save_vmcoreinfo_init(), which must run first to ensure vmcoreinfo initialization
1834 * is done before registering with f/w.
1835 */
1836subsys_initcall_sync(setup_fadump);
1837#else /* !CONFIG_PRESERVE_FA_DUMP */
1838
1839/* Scan the Firmware Assisted dump configuration details. */
1840int __init early_init_dt_scan_fw_dump(unsigned long node, const char *uname,
1841 int depth, void *data)
1842{
1843 if ((depth != 1) || (strcmp(uname, "ibm,opal") != 0))
1844 return 0;
1845
1846 opal_fadump_dt_scan(&fw_dump, node);
1847 return 1;
1848}
1849
1850/*
1851 * When dump is active but PRESERVE_FA_DUMP is enabled on the kernel,
1852 * preserve crash data. The subsequent memory preserving kernel boot
1853 * is likely to process this crash data.
1854 */
1855int __init fadump_reserve_mem(void)
1856{
1857 if (fw_dump.dump_active) {
1858 /*
1859 * If last boot has crashed then reserve all the memory
1860 * above boot memory to preserve crash data.
1861 */
1862 pr_info("Preserving crash data for processing in next boot.\n");
1863 fadump_reserve_crash_area(fw_dump.boot_mem_top);
1864 } else
1865 pr_debug("FADump-aware kernel..\n");
1866
1867 return 1;
1868}
1869#endif /* CONFIG_PRESERVE_FA_DUMP */
1870
1871/* Preserve everything above the base address */
1872static void __init fadump_reserve_crash_area(u64 base)
1873{
1874 u64 i, mstart, mend, msize;
1875
1876 for_each_mem_range(i, &mstart, &mend) {
1877 msize = mend - mstart;
1878
1879 if ((mstart + msize) < base)
1880 continue;
1881
1882 if (mstart < base) {
1883 msize -= (base - mstart);
1884 mstart = base;
1885 }
1886
1887 pr_info("Reserving %lluMB of memory at %#016llx for preserving crash data",
1888 (msize >> 20), mstart);
1889 memblock_reserve(mstart, msize);
1890 }
1891}