Linux Audio

Check our new training course

Loading...
v5.9
   1// SPDX-License-Identifier: GPL-2.0-or-later
   2/*
   3 * Firmware Assisted dump: A robust mechanism to get reliable kernel crash
   4 * dump with assistance from firmware. This approach does not use kexec,
   5 * instead firmware assists in booting the kdump kernel while preserving
   6 * memory contents. The most of the code implementation has been adapted
   7 * from phyp assisted dump implementation written by Linas Vepstas and
   8 * Manish Ahuja
   9 *
  10 * Copyright 2011 IBM Corporation
  11 * Author: Mahesh Salgaonkar <mahesh@linux.vnet.ibm.com>
  12 */
  13
  14#undef DEBUG
  15#define pr_fmt(fmt) "fadump: " fmt
  16
  17#include <linux/string.h>
  18#include <linux/memblock.h>
  19#include <linux/delay.h>
  20#include <linux/seq_file.h>
  21#include <linux/crash_dump.h>
  22#include <linux/kobject.h>
  23#include <linux/sysfs.h>
  24#include <linux/slab.h>
  25#include <linux/cma.h>
  26#include <linux/hugetlb.h>
 
 
 
  27
  28#include <asm/debugfs.h>
  29#include <asm/page.h>
  30#include <asm/prom.h>
  31#include <asm/fadump.h>
  32#include <asm/fadump-internal.h>
  33#include <asm/setup.h>
 
  34
  35/*
  36 * The CPU who acquired the lock to trigger the fadump crash should
  37 * wait for other CPUs to enter.
  38 *
  39 * The timeout is in milliseconds.
  40 */
  41#define CRASH_TIMEOUT		500
  42
  43static struct fw_dump fw_dump;
  44
  45static void __init fadump_reserve_crash_area(u64 base);
  46
  47struct kobject *fadump_kobj;
  48
  49#ifndef CONFIG_PRESERVE_FA_DUMP
  50
 
 
  51static atomic_t cpus_in_fadump;
  52static DEFINE_MUTEX(fadump_mutex);
  53
  54struct fadump_mrange_info crash_mrange_info = { "crash", NULL, 0, 0, 0, false };
  55
  56#define RESERVED_RNGS_SZ	16384 /* 16K - 128 entries */
  57#define RESERVED_RNGS_CNT	(RESERVED_RNGS_SZ / \
  58				 sizeof(struct fadump_memory_range))
  59static struct fadump_memory_range rngs[RESERVED_RNGS_CNT];
  60struct fadump_mrange_info reserved_mrange_info = { "reserved", rngs,
  61						   RESERVED_RNGS_SZ, 0,
  62						   RESERVED_RNGS_CNT, true };
  63
  64static void __init early_init_dt_scan_reserved_ranges(unsigned long node);
  65
  66#ifdef CONFIG_CMA
  67static struct cma *fadump_cma;
  68
  69/*
  70 * fadump_cma_init() - Initialize CMA area from a fadump reserved memory
  71 *
  72 * This function initializes CMA area from fadump reserved memory.
  73 * The total size of fadump reserved memory covers for boot memory size
  74 * + cpu data size + hpte size and metadata.
  75 * Initialize only the area equivalent to boot memory size for CMA use.
  76 * The reamining portion of fadump reserved memory will be not given
  77 * to CMA and pages for thoes will stay reserved. boot memory size is
  78 * aligned per CMA requirement to satisy cma_init_reserved_mem() call.
  79 * But for some reason even if it fails we still have the memory reservation
  80 * with us and we can still continue doing fadump.
  81 */
  82int __init fadump_cma_init(void)
  83{
  84	unsigned long long base, size;
  85	int rc;
  86
  87	if (!fw_dump.fadump_enabled)
  88		return 0;
  89
  90	/*
  91	 * Do not use CMA if user has provided fadump=nocma kernel parameter.
  92	 * Return 1 to continue with fadump old behaviour.
  93	 */
  94	if (fw_dump.nocma)
  95		return 1;
  96
 
 
 
 
 
 
  97	base = fw_dump.reserve_dump_area_start;
  98	size = fw_dump.boot_memory_size;
 
  99
 100	if (!size)
 101		return 0;
 
 
 
 
 
 
 102
 103	rc = cma_init_reserved_mem(base, size, 0, "fadump_cma", &fadump_cma);
 104	if (rc) {
 105		pr_err("Failed to init cma area for firmware-assisted dump,%d\n", rc);
 106		/*
 107		 * Though the CMA init has failed we still have memory
 108		 * reservation with us. The reserved memory will be
 109		 * blocked from production system usage.  Hence return 1,
 110		 * so that we can continue with fadump.
 111		 */
 112		return 1;
 113	}
 114
 115	/*
 
 
 
 
 
 
 116	 * So we now have successfully initialized cma area for fadump.
 117	 */
 118	pr_info("Initialized 0x%lx bytes cma area at %ldMB from 0x%lx "
 119		"bytes of memory reserved for firmware-assisted dump\n",
 120		cma_get_size(fadump_cma),
 121		(unsigned long)cma_get_base(fadump_cma) >> 20,
 122		fw_dump.reserve_dump_area_size);
 123	return 1;
 124}
 125#else
 126static int __init fadump_cma_init(void) { return 1; }
 127#endif /* CONFIG_CMA */
 128
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 129/* Scan the Firmware Assisted dump configuration details. */
 130int __init early_init_dt_scan_fw_dump(unsigned long node, const char *uname,
 131				      int depth, void *data)
 132{
 133	if (depth == 0) {
 134		early_init_dt_scan_reserved_ranges(node);
 135		return 0;
 136	}
 137
 138	if (depth != 1)
 139		return 0;
 140
 141	if (strcmp(uname, "rtas") == 0) {
 142		rtas_fadump_dt_scan(&fw_dump, node);
 143		return 1;
 144	}
 145
 146	if (strcmp(uname, "ibm,opal") == 0) {
 147		opal_fadump_dt_scan(&fw_dump, node);
 148		return 1;
 149	}
 150
 151	return 0;
 152}
 153
 154/*
 155 * If fadump is registered, check if the memory provided
 156 * falls within boot memory area and reserved memory area.
 157 */
 158int is_fadump_memory_area(u64 addr, unsigned long size)
 159{
 160	u64 d_start, d_end;
 161
 162	if (!fw_dump.dump_registered)
 163		return 0;
 164
 165	if (!size)
 166		return 0;
 167
 168	d_start = fw_dump.reserve_dump_area_start;
 169	d_end = d_start + fw_dump.reserve_dump_area_size;
 170	if (((addr + size) > d_start) && (addr <= d_end))
 171		return 1;
 172
 173	return (addr <= fw_dump.boot_mem_top);
 174}
 175
 176int should_fadump_crash(void)
 177{
 178	if (!fw_dump.dump_registered || !fw_dump.fadumphdr_addr)
 179		return 0;
 180	return 1;
 181}
 182
 183int is_fadump_active(void)
 184{
 185	return fw_dump.dump_active;
 186}
 187
 188/*
 189 * Returns true, if there are no holes in memory area between d_start to d_end,
 190 * false otherwise.
 191 */
 192static bool is_fadump_mem_area_contiguous(u64 d_start, u64 d_end)
 193{
 194	struct memblock_region *reg;
 195	bool ret = false;
 196	u64 start, end;
 197
 198	for_each_memblock(memory, reg) {
 199		start = max_t(u64, d_start, reg->base);
 200		end = min_t(u64, d_end, (reg->base + reg->size));
 201		if (d_start < end) {
 202			/* Memory hole from d_start to start */
 203			if (start > d_start)
 204				break;
 205
 206			if (end == d_end) {
 207				ret = true;
 208				break;
 209			}
 210
 211			d_start = end + 1;
 212		}
 213	}
 214
 215	return ret;
 216}
 217
 218/*
 219 * Returns true, if there are no holes in boot memory area,
 220 * false otherwise.
 221 */
 222bool is_fadump_boot_mem_contiguous(void)
 223{
 224	unsigned long d_start, d_end;
 225	bool ret = false;
 226	int i;
 227
 228	for (i = 0; i < fw_dump.boot_mem_regs_cnt; i++) {
 229		d_start = fw_dump.boot_mem_addr[i];
 230		d_end   = d_start + fw_dump.boot_mem_sz[i];
 231
 232		ret = is_fadump_mem_area_contiguous(d_start, d_end);
 233		if (!ret)
 234			break;
 235	}
 236
 237	return ret;
 238}
 239
 240/*
 241 * Returns true, if there are no holes in reserved memory area,
 242 * false otherwise.
 243 */
 244bool is_fadump_reserved_mem_contiguous(void)
 245{
 246	u64 d_start, d_end;
 247
 248	d_start	= fw_dump.reserve_dump_area_start;
 249	d_end	= d_start + fw_dump.reserve_dump_area_size;
 250	return is_fadump_mem_area_contiguous(d_start, d_end);
 251}
 252
 253/* Print firmware assisted dump configurations for debugging purpose. */
 254static void fadump_show_config(void)
 255{
 256	int i;
 257
 258	pr_debug("Support for firmware-assisted dump (fadump): %s\n",
 259			(fw_dump.fadump_supported ? "present" : "no support"));
 260
 261	if (!fw_dump.fadump_supported)
 262		return;
 263
 264	pr_debug("Fadump enabled    : %s\n",
 265				(fw_dump.fadump_enabled ? "yes" : "no"));
 266	pr_debug("Dump Active       : %s\n",
 267				(fw_dump.dump_active ? "yes" : "no"));
 268	pr_debug("Dump section sizes:\n");
 269	pr_debug("    CPU state data size: %lx\n", fw_dump.cpu_state_data_size);
 270	pr_debug("    HPTE region size   : %lx\n", fw_dump.hpte_region_size);
 271	pr_debug("    Boot memory size   : %lx\n", fw_dump.boot_memory_size);
 272	pr_debug("    Boot memory top    : %llx\n", fw_dump.boot_mem_top);
 273	pr_debug("Boot memory regions cnt: %llx\n", fw_dump.boot_mem_regs_cnt);
 274	for (i = 0; i < fw_dump.boot_mem_regs_cnt; i++) {
 275		pr_debug("[%03d] base = %llx, size = %llx\n", i,
 276			 fw_dump.boot_mem_addr[i], fw_dump.boot_mem_sz[i]);
 277	}
 278}
 279
 280/**
 281 * fadump_calculate_reserve_size(): reserve variable boot area 5% of System RAM
 282 *
 283 * Function to find the largest memory size we need to reserve during early
 284 * boot process. This will be the size of the memory that is required for a
 285 * kernel to boot successfully.
 286 *
 287 * This function has been taken from phyp-assisted dump feature implementation.
 288 *
 289 * returns larger of 256MB or 5% rounded down to multiples of 256MB.
 290 *
 291 * TODO: Come up with better approach to find out more accurate memory size
 292 * that is required for a kernel to boot successfully.
 293 *
 294 */
 295static inline u64 fadump_calculate_reserve_size(void)
 296{
 297	u64 base, size, bootmem_min;
 298	int ret;
 299
 300	if (fw_dump.reserve_bootvar)
 301		pr_warn("'fadump_reserve_mem=' parameter is deprecated in favor of 'crashkernel=' parameter.\n");
 302
 303	/*
 304	 * Check if the size is specified through crashkernel= cmdline
 305	 * option. If yes, then use that but ignore base as fadump reserves
 306	 * memory at a predefined offset.
 307	 */
 308	ret = parse_crashkernel(boot_command_line, memblock_phys_mem_size(),
 309				&size, &base);
 310	if (ret == 0 && size > 0) {
 311		unsigned long max_size;
 312
 313		if (fw_dump.reserve_bootvar)
 314			pr_info("Using 'crashkernel=' parameter for memory reservation.\n");
 315
 316		fw_dump.reserve_bootvar = (unsigned long)size;
 317
 318		/*
 319		 * Adjust if the boot memory size specified is above
 320		 * the upper limit.
 321		 */
 322		max_size = memblock_phys_mem_size() / MAX_BOOT_MEM_RATIO;
 323		if (fw_dump.reserve_bootvar > max_size) {
 324			fw_dump.reserve_bootvar = max_size;
 325			pr_info("Adjusted boot memory size to %luMB\n",
 326				(fw_dump.reserve_bootvar >> 20));
 327		}
 328
 329		return fw_dump.reserve_bootvar;
 330	} else if (fw_dump.reserve_bootvar) {
 331		/*
 332		 * 'fadump_reserve_mem=' is being used to reserve memory
 333		 * for firmware-assisted dump.
 334		 */
 335		return fw_dump.reserve_bootvar;
 336	}
 337
 338	/* divide by 20 to get 5% of value */
 339	size = memblock_phys_mem_size() / 20;
 340
 341	/* round it down in multiples of 256 */
 342	size = size & ~0x0FFFFFFFUL;
 343
 344	/* Truncate to memory_limit. We don't want to over reserve the memory.*/
 345	if (memory_limit && size > memory_limit)
 346		size = memory_limit;
 347
 348	bootmem_min = fw_dump.ops->fadump_get_bootmem_min();
 349	return (size > bootmem_min ? size : bootmem_min);
 350}
 351
 352/*
 353 * Calculate the total memory size required to be reserved for
 354 * firmware-assisted dump registration.
 355 */
 356static unsigned long get_fadump_area_size(void)
 357{
 358	unsigned long size = 0;
 359
 360	size += fw_dump.cpu_state_data_size;
 361	size += fw_dump.hpte_region_size;
 
 
 
 
 
 362	size += fw_dump.boot_memory_size;
 363	size += sizeof(struct fadump_crash_info_header);
 364	size += sizeof(struct elfhdr); /* ELF core header.*/
 365	size += sizeof(struct elf_phdr); /* place holder for cpu notes */
 366	/* Program headers for crash memory regions. */
 367	size += sizeof(struct elf_phdr) * (memblock_num_regions(memory) + 2);
 368
 369	size = PAGE_ALIGN(size);
 370
 371	/* This is to hold kernel metadata on platforms that support it */
 372	size += (fw_dump.ops->fadump_get_metadata_size ?
 373		 fw_dump.ops->fadump_get_metadata_size() : 0);
 374	return size;
 375}
 376
 377static int __init add_boot_mem_region(unsigned long rstart,
 378				      unsigned long rsize)
 379{
 
 380	int i = fw_dump.boot_mem_regs_cnt++;
 381
 382	if (fw_dump.boot_mem_regs_cnt > FADUMP_MAX_MEM_REGS) {
 383		fw_dump.boot_mem_regs_cnt = FADUMP_MAX_MEM_REGS;
 384		return 0;
 385	}
 386
 387	pr_debug("Added boot memory range[%d] [%#016lx-%#016lx)\n",
 388		 i, rstart, (rstart + rsize));
 389	fw_dump.boot_mem_addr[i] = rstart;
 390	fw_dump.boot_mem_sz[i] = rsize;
 391	return 1;
 392}
 393
 394/*
 395 * Firmware usually has a hard limit on the data it can copy per region.
 396 * Honour that by splitting a memory range into multiple regions.
 397 */
 398static int __init add_boot_mem_regions(unsigned long mstart,
 399				       unsigned long msize)
 400{
 401	unsigned long rstart, rsize, max_size;
 402	int ret = 1;
 403
 404	rstart = mstart;
 405	max_size = fw_dump.max_copy_size ? fw_dump.max_copy_size : msize;
 406	while (msize) {
 407		if (msize > max_size)
 408			rsize = max_size;
 409		else
 410			rsize = msize;
 411
 412		ret = add_boot_mem_region(rstart, rsize);
 413		if (!ret)
 414			break;
 415
 416		msize -= rsize;
 417		rstart += rsize;
 418	}
 419
 420	return ret;
 421}
 422
 423static int __init fadump_get_boot_mem_regions(void)
 424{
 425	unsigned long base, size, cur_size, hole_size, last_end;
 426	unsigned long mem_size = fw_dump.boot_memory_size;
 427	struct memblock_region *reg;
 428	int ret = 1;
 
 429
 430	fw_dump.boot_mem_regs_cnt = 0;
 431
 432	last_end = 0;
 433	hole_size = 0;
 434	cur_size = 0;
 435	for_each_memblock(memory, reg) {
 436		base = reg->base;
 437		size = reg->size;
 438		hole_size += (base - last_end);
 439
 440		if ((cur_size + size) >= mem_size) {
 441			size = (mem_size - cur_size);
 442			ret = add_boot_mem_regions(base, size);
 443			break;
 444		}
 445
 446		mem_size -= size;
 447		cur_size += size;
 448		ret = add_boot_mem_regions(base, size);
 449		if (!ret)
 450			break;
 451
 452		last_end = base + size;
 453	}
 454	fw_dump.boot_mem_top = PAGE_ALIGN(fw_dump.boot_memory_size + hole_size);
 455
 456	return ret;
 457}
 458
 459/*
 460 * Returns true, if the given range overlaps with reserved memory ranges
 461 * starting at idx. Also, updates idx to index of overlapping memory range
 462 * with the given memory range.
 463 * False, otherwise.
 464 */
 465static bool overlaps_reserved_ranges(u64 base, u64 end, int *idx)
 466{
 467	bool ret = false;
 468	int i;
 469
 470	for (i = *idx; i < reserved_mrange_info.mem_range_cnt; i++) {
 471		u64 rbase = reserved_mrange_info.mem_ranges[i].base;
 472		u64 rend = rbase + reserved_mrange_info.mem_ranges[i].size;
 473
 474		if (end <= rbase)
 475			break;
 476
 477		if ((end > rbase) &&  (base < rend)) {
 478			*idx = i;
 479			ret = true;
 480			break;
 481		}
 482	}
 483
 484	return ret;
 485}
 486
 487/*
 488 * Locate a suitable memory area to reserve memory for FADump. While at it,
 489 * lookup reserved-ranges & avoid overlap with them, as they are used by F/W.
 490 */
 491static u64 __init fadump_locate_reserve_mem(u64 base, u64 size)
 492{
 493	struct fadump_memory_range *mrngs;
 494	phys_addr_t mstart, mend;
 495	int idx = 0;
 496	u64 i, ret = 0;
 497
 498	mrngs = reserved_mrange_info.mem_ranges;
 499	for_each_free_mem_range(i, NUMA_NO_NODE, MEMBLOCK_NONE,
 500				&mstart, &mend, NULL) {
 501		pr_debug("%llu) mstart: %llx, mend: %llx, base: %llx\n",
 502			 i, mstart, mend, base);
 503
 504		if (mstart > base)
 505			base = PAGE_ALIGN(mstart);
 506
 507		while ((mend > base) && ((mend - base) >= size)) {
 508			if (!overlaps_reserved_ranges(base, base+size, &idx)) {
 509				ret = base;
 510				goto out;
 511			}
 512
 513			base = mrngs[idx].base + mrngs[idx].size;
 514			base = PAGE_ALIGN(base);
 515		}
 516	}
 517
 518out:
 519	return ret;
 520}
 521
 522int __init fadump_reserve_mem(void)
 523{
 524	u64 base, size, mem_boundary, bootmem_min;
 525	int ret = 1;
 526
 527	if (!fw_dump.fadump_enabled)
 528		return 0;
 529
 530	if (!fw_dump.fadump_supported) {
 531		pr_info("Firmware-Assisted Dump is not supported on this hardware\n");
 532		goto error_out;
 533	}
 534
 535	/*
 536	 * Initialize boot memory size
 537	 * If dump is active then we have already calculated the size during
 538	 * first kernel.
 539	 */
 540	if (!fw_dump.dump_active) {
 541		fw_dump.boot_memory_size =
 542			PAGE_ALIGN(fadump_calculate_reserve_size());
 543#ifdef CONFIG_CMA
 544		if (!fw_dump.nocma) {
 545			fw_dump.boot_memory_size =
 546				ALIGN(fw_dump.boot_memory_size,
 547				      FADUMP_CMA_ALIGNMENT);
 548		}
 549#endif
 550
 551		bootmem_min = fw_dump.ops->fadump_get_bootmem_min();
 552		if (fw_dump.boot_memory_size < bootmem_min) {
 553			pr_err("Can't enable fadump with boot memory size (0x%lx) less than 0x%llx\n",
 554			       fw_dump.boot_memory_size, bootmem_min);
 555			goto error_out;
 556		}
 557
 558		if (!fadump_get_boot_mem_regions()) {
 559			pr_err("Too many holes in boot memory area to enable fadump\n");
 560			goto error_out;
 561		}
 562	}
 563
 564	/*
 565	 * Calculate the memory boundary.
 566	 * If memory_limit is less than actual memory boundary then reserve
 567	 * the memory for fadump beyond the memory_limit and adjust the
 568	 * memory_limit accordingly, so that the running kernel can run with
 569	 * specified memory_limit.
 570	 */
 571	if (memory_limit && memory_limit < memblock_end_of_DRAM()) {
 572		size = get_fadump_area_size();
 573		if ((memory_limit + size) < memblock_end_of_DRAM())
 574			memory_limit += size;
 575		else
 576			memory_limit = memblock_end_of_DRAM();
 577		printk(KERN_INFO "Adjusted memory_limit for firmware-assisted"
 578				" dump, now %#016llx\n", memory_limit);
 579	}
 580	if (memory_limit)
 581		mem_boundary = memory_limit;
 582	else
 583		mem_boundary = memblock_end_of_DRAM();
 584
 585	base = fw_dump.boot_mem_top;
 586	size = get_fadump_area_size();
 587	fw_dump.reserve_dump_area_size = size;
 588	if (fw_dump.dump_active) {
 589		pr_info("Firmware-assisted dump is active.\n");
 590
 591#ifdef CONFIG_HUGETLB_PAGE
 592		/*
 593		 * FADump capture kernel doesn't care much about hugepages.
 594		 * In fact, handling hugepages in capture kernel is asking for
 595		 * trouble. So, disable HugeTLB support when fadump is active.
 596		 */
 597		hugetlb_disabled = true;
 598#endif
 599		/*
 600		 * If last boot has crashed then reserve all the memory
 601		 * above boot memory size so that we don't touch it until
 602		 * dump is written to disk by userspace tool. This memory
 603		 * can be released for general use by invalidating fadump.
 604		 */
 605		fadump_reserve_crash_area(base);
 606
 607		pr_debug("fadumphdr_addr = %#016lx\n", fw_dump.fadumphdr_addr);
 608		pr_debug("Reserve dump area start address: 0x%lx\n",
 609			 fw_dump.reserve_dump_area_start);
 610	} else {
 611		/*
 612		 * Reserve memory at an offset closer to bottom of the RAM to
 613		 * minimize the impact of memory hot-remove operation.
 614		 */
 615		base = fadump_locate_reserve_mem(base, size);
 616
 617		if (!base || (base + size > mem_boundary)) {
 618			pr_err("Failed to find memory chunk for reservation!\n");
 619			goto error_out;
 620		}
 621		fw_dump.reserve_dump_area_start = base;
 622
 623		/*
 624		 * Calculate the kernel metadata address and register it with
 625		 * f/w if the platform supports.
 626		 */
 627		if (fw_dump.ops->fadump_setup_metadata &&
 628		    (fw_dump.ops->fadump_setup_metadata(&fw_dump) < 0))
 629			goto error_out;
 630
 631		if (memblock_reserve(base, size)) {
 632			pr_err("Failed to reserve memory!\n");
 633			goto error_out;
 634		}
 635
 636		pr_info("Reserved %lldMB of memory at %#016llx (System RAM: %lldMB)\n",
 637			(size >> 20), base, (memblock_phys_mem_size() >> 20));
 638
 639		ret = fadump_cma_init();
 640	}
 641
 642	return ret;
 643error_out:
 644	fw_dump.fadump_enabled = 0;
 
 645	return 0;
 646}
 647
 648/* Look for fadump= cmdline option. */
 649static int __init early_fadump_param(char *p)
 650{
 651	if (!p)
 652		return 1;
 653
 654	if (strncmp(p, "on", 2) == 0)
 655		fw_dump.fadump_enabled = 1;
 656	else if (strncmp(p, "off", 3) == 0)
 657		fw_dump.fadump_enabled = 0;
 658	else if (strncmp(p, "nocma", 5) == 0) {
 659		fw_dump.fadump_enabled = 1;
 660		fw_dump.nocma = 1;
 661	}
 662
 663	return 0;
 664}
 665early_param("fadump", early_fadump_param);
 666
 667/*
 668 * Look for fadump_reserve_mem= cmdline option
 669 * TODO: Remove references to 'fadump_reserve_mem=' parameter,
 670 *       the sooner 'crashkernel=' parameter is accustomed to.
 671 */
 672static int __init early_fadump_reserve_mem(char *p)
 673{
 674	if (p)
 675		fw_dump.reserve_bootvar = memparse(p, &p);
 676	return 0;
 677}
 678early_param("fadump_reserve_mem", early_fadump_reserve_mem);
 679
 680void crash_fadump(struct pt_regs *regs, const char *str)
 681{
 682	unsigned int msecs;
 683	struct fadump_crash_info_header *fdh = NULL;
 684	int old_cpu, this_cpu;
 685	/* Do not include first CPU */
 686	unsigned int ncpus = num_online_cpus() - 1;
 687
 688	if (!should_fadump_crash())
 689		return;
 690
 691	/*
 692	 * old_cpu == -1 means this is the first CPU which has come here,
 693	 * go ahead and trigger fadump.
 694	 *
 695	 * old_cpu != -1 means some other CPU has already on it's way
 696	 * to trigger fadump, just keep looping here.
 697	 */
 698	this_cpu = smp_processor_id();
 699	old_cpu = cmpxchg(&crashing_cpu, -1, this_cpu);
 700
 701	if (old_cpu != -1) {
 702		atomic_inc(&cpus_in_fadump);
 703
 704		/*
 705		 * We can't loop here indefinitely. Wait as long as fadump
 706		 * is in force. If we race with fadump un-registration this
 707		 * loop will break and then we go down to normal panic path
 708		 * and reboot. If fadump is in force the first crashing
 709		 * cpu will definitely trigger fadump.
 710		 */
 711		while (fw_dump.dump_registered)
 712			cpu_relax();
 713		return;
 714	}
 715
 716	fdh = __va(fw_dump.fadumphdr_addr);
 717	fdh->crashing_cpu = crashing_cpu;
 718	crash_save_vmcoreinfo();
 719
 720	if (regs)
 721		fdh->regs = *regs;
 722	else
 723		ppc_save_regs(&fdh->regs);
 724
 725	fdh->online_mask = *cpu_online_mask;
 726
 727	/*
 728	 * If we came in via system reset, wait a while for the secondary
 729	 * CPUs to enter.
 730	 */
 731	if (TRAP(&(fdh->regs)) == 0x100) {
 732		msecs = CRASH_TIMEOUT;
 733		while ((atomic_read(&cpus_in_fadump) < ncpus) && (--msecs > 0))
 734			mdelay(1);
 735	}
 736
 737	fw_dump.ops->fadump_trigger(fdh, str);
 738}
 739
 740u32 *fadump_regs_to_elf_notes(u32 *buf, struct pt_regs *regs)
 741{
 742	struct elf_prstatus prstatus;
 743
 744	memset(&prstatus, 0, sizeof(prstatus));
 745	/*
 746	 * FIXME: How do i get PID? Do I really need it?
 747	 * prstatus.pr_pid = ????
 748	 */
 749	elf_core_copy_kernel_regs(&prstatus.pr_reg, regs);
 750	buf = append_elf_note(buf, CRASH_CORE_NOTE_NAME, NT_PRSTATUS,
 751			      &prstatus, sizeof(prstatus));
 752	return buf;
 753}
 754
 755void fadump_update_elfcore_header(char *bufp)
 756{
 757	struct elfhdr *elf;
 758	struct elf_phdr *phdr;
 759
 760	elf = (struct elfhdr *)bufp;
 761	bufp += sizeof(struct elfhdr);
 762
 763	/* First note is a place holder for cpu notes info. */
 764	phdr = (struct elf_phdr *)bufp;
 765
 766	if (phdr->p_type == PT_NOTE) {
 767		phdr->p_paddr	= __pa(fw_dump.cpu_notes_buf_vaddr);
 768		phdr->p_offset	= phdr->p_paddr;
 769		phdr->p_filesz	= fw_dump.cpu_notes_buf_size;
 770		phdr->p_memsz = fw_dump.cpu_notes_buf_size;
 771	}
 772	return;
 773}
 774
 775static void *fadump_alloc_buffer(unsigned long size)
 776{
 777	unsigned long count, i;
 778	struct page *page;
 779	void *vaddr;
 780
 781	vaddr = alloc_pages_exact(size, GFP_KERNEL | __GFP_ZERO);
 782	if (!vaddr)
 783		return NULL;
 784
 785	count = PAGE_ALIGN(size) / PAGE_SIZE;
 786	page = virt_to_page(vaddr);
 787	for (i = 0; i < count; i++)
 788		mark_page_reserved(page + i);
 789	return vaddr;
 790}
 791
 792static void fadump_free_buffer(unsigned long vaddr, unsigned long size)
 793{
 794	free_reserved_area((void *)vaddr, (void *)(vaddr + size), -1, NULL);
 795}
 796
 797s32 fadump_setup_cpu_notes_buf(u32 num_cpus)
 798{
 799	/* Allocate buffer to hold cpu crash notes. */
 800	fw_dump.cpu_notes_buf_size = num_cpus * sizeof(note_buf_t);
 801	fw_dump.cpu_notes_buf_size = PAGE_ALIGN(fw_dump.cpu_notes_buf_size);
 802	fw_dump.cpu_notes_buf_vaddr =
 803		(unsigned long)fadump_alloc_buffer(fw_dump.cpu_notes_buf_size);
 804	if (!fw_dump.cpu_notes_buf_vaddr) {
 805		pr_err("Failed to allocate %ld bytes for CPU notes buffer\n",
 806		       fw_dump.cpu_notes_buf_size);
 807		return -ENOMEM;
 808	}
 809
 810	pr_debug("Allocated buffer for cpu notes of size %ld at 0x%lx\n",
 811		 fw_dump.cpu_notes_buf_size,
 812		 fw_dump.cpu_notes_buf_vaddr);
 813	return 0;
 814}
 815
 816void fadump_free_cpu_notes_buf(void)
 817{
 818	if (!fw_dump.cpu_notes_buf_vaddr)
 819		return;
 820
 821	fadump_free_buffer(fw_dump.cpu_notes_buf_vaddr,
 822			   fw_dump.cpu_notes_buf_size);
 823	fw_dump.cpu_notes_buf_vaddr = 0;
 824	fw_dump.cpu_notes_buf_size = 0;
 825}
 826
 827static void fadump_free_mem_ranges(struct fadump_mrange_info *mrange_info)
 828{
 829	if (mrange_info->is_static) {
 830		mrange_info->mem_range_cnt = 0;
 831		return;
 832	}
 833
 834	kfree(mrange_info->mem_ranges);
 835	memset((void *)((u64)mrange_info + RNG_NAME_SZ), 0,
 836	       (sizeof(struct fadump_mrange_info) - RNG_NAME_SZ));
 837}
 838
 839/*
 840 * Allocate or reallocate mem_ranges array in incremental units
 841 * of PAGE_SIZE.
 842 */
 843static int fadump_alloc_mem_ranges(struct fadump_mrange_info *mrange_info)
 844{
 845	struct fadump_memory_range *new_array;
 846	u64 new_size;
 847
 848	new_size = mrange_info->mem_ranges_sz + PAGE_SIZE;
 849	pr_debug("Allocating %llu bytes of memory for %s memory ranges\n",
 850		 new_size, mrange_info->name);
 851
 852	new_array = krealloc(mrange_info->mem_ranges, new_size, GFP_KERNEL);
 853	if (new_array == NULL) {
 854		pr_err("Insufficient memory for setting up %s memory ranges\n",
 855		       mrange_info->name);
 856		fadump_free_mem_ranges(mrange_info);
 857		return -ENOMEM;
 858	}
 859
 860	mrange_info->mem_ranges = new_array;
 861	mrange_info->mem_ranges_sz = new_size;
 862	mrange_info->max_mem_ranges = (new_size /
 863				       sizeof(struct fadump_memory_range));
 864	return 0;
 865}
 866
 867static inline int fadump_add_mem_range(struct fadump_mrange_info *mrange_info,
 868				       u64 base, u64 end)
 869{
 870	struct fadump_memory_range *mem_ranges = mrange_info->mem_ranges;
 871	bool is_adjacent = false;
 872	u64 start, size;
 873
 874	if (base == end)
 875		return 0;
 876
 877	/*
 878	 * Fold adjacent memory ranges to bring down the memory ranges/
 879	 * PT_LOAD segments count.
 880	 */
 881	if (mrange_info->mem_range_cnt) {
 882		start = mem_ranges[mrange_info->mem_range_cnt - 1].base;
 883		size  = mem_ranges[mrange_info->mem_range_cnt - 1].size;
 884
 885		if ((start + size) == base)
 
 
 
 
 
 886			is_adjacent = true;
 887	}
 888	if (!is_adjacent) {
 889		/* resize the array on reaching the limit */
 890		if (mrange_info->mem_range_cnt == mrange_info->max_mem_ranges) {
 891			int ret;
 892
 893			if (mrange_info->is_static) {
 894				pr_err("Reached array size limit for %s memory ranges\n",
 895				       mrange_info->name);
 896				return -ENOSPC;
 897			}
 898
 899			ret = fadump_alloc_mem_ranges(mrange_info);
 900			if (ret)
 901				return ret;
 902
 903			/* Update to the new resized array */
 904			mem_ranges = mrange_info->mem_ranges;
 905		}
 906
 907		start = base;
 908		mem_ranges[mrange_info->mem_range_cnt].base = start;
 909		mrange_info->mem_range_cnt++;
 910	}
 911
 912	mem_ranges[mrange_info->mem_range_cnt - 1].size = (end - start);
 913	pr_debug("%s_memory_range[%d] [%#016llx-%#016llx], %#llx bytes\n",
 914		 mrange_info->name, (mrange_info->mem_range_cnt - 1),
 915		 start, end - 1, (end - start));
 916	return 0;
 917}
 918
 919static int fadump_exclude_reserved_area(u64 start, u64 end)
 920{
 921	u64 ra_start, ra_end;
 922	int ret = 0;
 923
 924	ra_start = fw_dump.reserve_dump_area_start;
 925	ra_end = ra_start + fw_dump.reserve_dump_area_size;
 926
 927	if ((ra_start < end) && (ra_end > start)) {
 928		if ((start < ra_start) && (end > ra_end)) {
 929			ret = fadump_add_mem_range(&crash_mrange_info,
 930						   start, ra_start);
 931			if (ret)
 932				return ret;
 933
 934			ret = fadump_add_mem_range(&crash_mrange_info,
 935						   ra_end, end);
 936		} else if (start < ra_start) {
 937			ret = fadump_add_mem_range(&crash_mrange_info,
 938						   start, ra_start);
 939		} else if (ra_end < end) {
 940			ret = fadump_add_mem_range(&crash_mrange_info,
 941						   ra_end, end);
 942		}
 943	} else
 944		ret = fadump_add_mem_range(&crash_mrange_info, start, end);
 945
 946	return ret;
 947}
 948
 949static int fadump_init_elfcore_header(char *bufp)
 950{
 951	struct elfhdr *elf;
 952
 953	elf = (struct elfhdr *) bufp;
 954	bufp += sizeof(struct elfhdr);
 955	memcpy(elf->e_ident, ELFMAG, SELFMAG);
 956	elf->e_ident[EI_CLASS] = ELF_CLASS;
 957	elf->e_ident[EI_DATA] = ELF_DATA;
 958	elf->e_ident[EI_VERSION] = EV_CURRENT;
 959	elf->e_ident[EI_OSABI] = ELF_OSABI;
 960	memset(elf->e_ident+EI_PAD, 0, EI_NIDENT-EI_PAD);
 961	elf->e_type = ET_CORE;
 962	elf->e_machine = ELF_ARCH;
 963	elf->e_version = EV_CURRENT;
 964	elf->e_entry = 0;
 965	elf->e_phoff = sizeof(struct elfhdr);
 966	elf->e_shoff = 0;
 967#if defined(_CALL_ELF)
 968	elf->e_flags = _CALL_ELF;
 969#else
 970	elf->e_flags = 0;
 971#endif
 
 
 
 972	elf->e_ehsize = sizeof(struct elfhdr);
 973	elf->e_phentsize = sizeof(struct elf_phdr);
 974	elf->e_phnum = 0;
 975	elf->e_shentsize = 0;
 976	elf->e_shnum = 0;
 977	elf->e_shstrndx = 0;
 978
 979	return 0;
 980}
 981
 982/*
 983 * Traverse through memblock structure and setup crash memory ranges. These
 984 * ranges will be used create PT_LOAD program headers in elfcore header.
 985 */
 986static int fadump_setup_crash_memory_ranges(void)
 987{
 988	struct memblock_region *reg;
 989	u64 start, end;
 990	int i, ret;
 991
 992	pr_debug("Setup crash memory ranges.\n");
 993	crash_mrange_info.mem_range_cnt = 0;
 994
 995	/*
 996	 * Boot memory region(s) registered with firmware are moved to
 997	 * different location at the time of crash. Create separate program
 998	 * header(s) for this memory chunk(s) with the correct offset.
 999	 */
1000	for (i = 0; i < fw_dump.boot_mem_regs_cnt; i++) {
1001		start = fw_dump.boot_mem_addr[i];
1002		end = start + fw_dump.boot_mem_sz[i];
1003		ret = fadump_add_mem_range(&crash_mrange_info, start, end);
1004		if (ret)
1005			return ret;
1006	}
1007
1008	for_each_memblock(memory, reg) {
1009		start = (u64)reg->base;
1010		end = start + (u64)reg->size;
1011
1012		/*
1013		 * skip the memory chunk that is already added
1014		 * (0 through boot_memory_top).
1015		 */
1016		if (start < fw_dump.boot_mem_top) {
1017			if (end > fw_dump.boot_mem_top)
1018				start = fw_dump.boot_mem_top;
1019			else
1020				continue;
1021		}
1022
1023		/* add this range excluding the reserved dump area. */
1024		ret = fadump_exclude_reserved_area(start, end);
1025		if (ret)
1026			return ret;
1027	}
1028
1029	return 0;
1030}
1031
1032/*
1033 * If the given physical address falls within the boot memory region then
1034 * return the relocated address that points to the dump region reserved
1035 * for saving initial boot memory contents.
1036 */
1037static inline unsigned long fadump_relocate(unsigned long paddr)
1038{
1039	unsigned long raddr, rstart, rend, rlast, hole_size;
1040	int i;
1041
1042	hole_size = 0;
1043	rlast = 0;
1044	raddr = paddr;
1045	for (i = 0; i < fw_dump.boot_mem_regs_cnt; i++) {
1046		rstart = fw_dump.boot_mem_addr[i];
1047		rend = rstart + fw_dump.boot_mem_sz[i];
1048		hole_size += (rstart - rlast);
1049
1050		if (paddr >= rstart && paddr < rend) {
1051			raddr += fw_dump.boot_mem_dest_addr - hole_size;
1052			break;
1053		}
1054
1055		rlast = rend;
1056	}
1057
1058	pr_debug("vmcoreinfo: paddr = 0x%lx, raddr = 0x%lx\n", paddr, raddr);
1059	return raddr;
1060}
1061
1062static int fadump_create_elfcore_headers(char *bufp)
 
1063{
1064	unsigned long long raddr, offset;
1065	struct elf_phdr *phdr;
 
 
 
 
 
 
 
 
 
 
 
1066	struct elfhdr *elf;
1067	int i, j;
 
 
1068
 
1069	fadump_init_elfcore_header(bufp);
1070	elf = (struct elfhdr *)bufp;
1071	bufp += sizeof(struct elfhdr);
1072
1073	/*
1074	 * setup ELF PT_NOTE, place holder for cpu notes info. The notes info
1075	 * will be populated during second kernel boot after crash. Hence
1076	 * this PT_NOTE will always be the first elf note.
1077	 *
1078	 * NOTE: Any new ELF note addition should be placed after this note.
1079	 */
1080	phdr = (struct elf_phdr *)bufp;
1081	bufp += sizeof(struct elf_phdr);
1082	phdr->p_type = PT_NOTE;
1083	phdr->p_flags = 0;
1084	phdr->p_vaddr = 0;
1085	phdr->p_align = 0;
1086
1087	phdr->p_offset = 0;
1088	phdr->p_paddr = 0;
1089	phdr->p_filesz = 0;
1090	phdr->p_memsz = 0;
1091
1092	(elf->e_phnum)++;
1093
1094	/* setup ELF PT_NOTE for vmcoreinfo */
1095	phdr = (struct elf_phdr *)bufp;
1096	bufp += sizeof(struct elf_phdr);
1097	phdr->p_type	= PT_NOTE;
1098	phdr->p_flags	= 0;
1099	phdr->p_vaddr	= 0;
1100	phdr->p_align	= 0;
1101
1102	phdr->p_paddr	= fadump_relocate(paddr_vmcoreinfo_note());
1103	phdr->p_offset	= phdr->p_paddr;
1104	phdr->p_memsz	= phdr->p_filesz = VMCOREINFO_NOTE_SIZE;
1105
1106	/* Increment number of program headers. */
1107	(elf->e_phnum)++;
1108
1109	/* setup PT_LOAD sections. */
1110	j = 0;
1111	offset = 0;
1112	raddr = fw_dump.boot_mem_addr[0];
1113	for (i = 0; i < crash_mrange_info.mem_range_cnt; i++) {
1114		u64 mbase, msize;
1115
1116		mbase = crash_mrange_info.mem_ranges[i].base;
1117		msize = crash_mrange_info.mem_ranges[i].size;
1118		if (!msize)
1119			continue;
1120
1121		phdr = (struct elf_phdr *)bufp;
1122		bufp += sizeof(struct elf_phdr);
1123		phdr->p_type	= PT_LOAD;
1124		phdr->p_flags	= PF_R|PF_W|PF_X;
1125		phdr->p_offset	= mbase;
1126
1127		if (mbase == raddr) {
1128			/*
1129			 * The entire real memory region will be moved by
1130			 * firmware to the specified destination_address.
1131			 * Hence set the correct offset.
1132			 */
1133			phdr->p_offset = fw_dump.boot_mem_dest_addr + offset;
1134			if (j < (fw_dump.boot_mem_regs_cnt - 1)) {
1135				offset += fw_dump.boot_mem_sz[j];
1136				raddr = fw_dump.boot_mem_addr[++j];
1137			}
 
 
 
 
 
 
1138		}
1139
1140		phdr->p_paddr = mbase;
1141		phdr->p_vaddr = (unsigned long)__va(mbase);
1142		phdr->p_filesz = msize;
1143		phdr->p_memsz = msize;
1144		phdr->p_align = 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
1145
1146		/* Increment number of program headers. */
1147		(elf->e_phnum)++;
 
 
1148	}
1149	return 0;
1150}
1151
1152static unsigned long init_fadump_header(unsigned long addr)
1153{
1154	struct fadump_crash_info_header *fdh;
1155
1156	if (!addr)
1157		return 0;
1158
1159	fdh = __va(addr);
1160	addr += sizeof(struct fadump_crash_info_header);
1161
1162	memset(fdh, 0, sizeof(struct fadump_crash_info_header));
1163	fdh->magic_number = FADUMP_CRASH_INFO_MAGIC;
1164	fdh->elfcorehdr_addr = addr;
1165	/* We will set the crashing cpu id in crash_fadump() during crash. */
1166	fdh->crashing_cpu = FADUMP_CPU_UNKNOWN;
1167
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1168	return addr;
1169}
1170
1171static int register_fadump(void)
1172{
1173	unsigned long addr;
1174	void *vaddr;
1175	int ret;
1176
1177	/*
1178	 * If no memory is reserved then we can not register for firmware-
1179	 * assisted dump.
1180	 */
1181	if (!fw_dump.reserve_dump_area_size)
1182		return -ENODEV;
1183
1184	ret = fadump_setup_crash_memory_ranges();
1185	if (ret)
1186		return ret;
1187
1188	addr = fw_dump.fadumphdr_addr;
1189
1190	/* Initialize fadump crash info header. */
1191	addr = init_fadump_header(addr);
1192	vaddr = __va(addr);
1193
1194	pr_debug("Creating ELF core headers at %#016lx\n", addr);
1195	fadump_create_elfcore_headers(vaddr);
1196
1197	/* register the future kernel dump with firmware. */
1198	pr_debug("Registering for firmware-assisted kernel dump...\n");
1199	return fw_dump.ops->fadump_register(&fw_dump);
1200}
1201
1202void fadump_cleanup(void)
1203{
1204	if (!fw_dump.fadump_supported)
1205		return;
1206
1207	/* Invalidate the registration only if dump is active. */
1208	if (fw_dump.dump_active) {
1209		pr_debug("Invalidating firmware-assisted dump registration\n");
1210		fw_dump.ops->fadump_invalidate(&fw_dump);
1211	} else if (fw_dump.dump_registered) {
1212		/* Un-register Firmware-assisted dump if it was registered. */
1213		fw_dump.ops->fadump_unregister(&fw_dump);
1214		fadump_free_mem_ranges(&crash_mrange_info);
1215	}
1216
1217	if (fw_dump.ops->fadump_cleanup)
1218		fw_dump.ops->fadump_cleanup(&fw_dump);
1219}
1220
1221static void fadump_free_reserved_memory(unsigned long start_pfn,
1222					unsigned long end_pfn)
1223{
1224	unsigned long pfn;
1225	unsigned long time_limit = jiffies + HZ;
1226
1227	pr_info("freeing reserved memory (0x%llx - 0x%llx)\n",
1228		PFN_PHYS(start_pfn), PFN_PHYS(end_pfn));
1229
1230	for (pfn = start_pfn; pfn < end_pfn; pfn++) {
1231		free_reserved_page(pfn_to_page(pfn));
1232
1233		if (time_after(jiffies, time_limit)) {
1234			cond_resched();
1235			time_limit = jiffies + HZ;
1236		}
1237	}
1238}
1239
1240/*
1241 * Skip memory holes and free memory that was actually reserved.
1242 */
1243static void fadump_release_reserved_area(u64 start, u64 end)
1244{
 
1245	u64 tstart, tend, spfn, epfn;
1246	struct memblock_region *reg;
1247
1248	spfn = PHYS_PFN(start);
1249	epfn = PHYS_PFN(end);
1250	for_each_memblock(memory, reg) {
1251		tstart = max_t(u64, spfn, memblock_region_memory_base_pfn(reg));
1252		tend   = min_t(u64, epfn, memblock_region_memory_end_pfn(reg));
 
 
1253		if (tstart < tend) {
1254			fadump_free_reserved_memory(tstart, tend);
1255
1256			if (tend == epfn)
1257				break;
1258
1259			spfn = tend;
1260		}
1261	}
1262}
1263
1264/*
1265 * Sort the mem ranges in-place and merge adjacent ranges
1266 * to minimize the memory ranges count.
1267 */
1268static void sort_and_merge_mem_ranges(struct fadump_mrange_info *mrange_info)
1269{
1270	struct fadump_memory_range *mem_ranges;
1271	struct fadump_memory_range tmp_range;
1272	u64 base, size;
1273	int i, j, idx;
1274
1275	if (!reserved_mrange_info.mem_range_cnt)
1276		return;
1277
1278	/* Sort the memory ranges */
1279	mem_ranges = mrange_info->mem_ranges;
1280	for (i = 0; i < mrange_info->mem_range_cnt; i++) {
1281		idx = i;
1282		for (j = (i + 1); j < mrange_info->mem_range_cnt; j++) {
1283			if (mem_ranges[idx].base > mem_ranges[j].base)
1284				idx = j;
1285		}
1286		if (idx != i) {
1287			tmp_range = mem_ranges[idx];
1288			mem_ranges[idx] = mem_ranges[i];
1289			mem_ranges[i] = tmp_range;
1290		}
1291	}
1292
1293	/* Merge adjacent reserved ranges */
1294	idx = 0;
1295	for (i = 1; i < mrange_info->mem_range_cnt; i++) {
1296		base = mem_ranges[i-1].base;
1297		size = mem_ranges[i-1].size;
1298		if (mem_ranges[i].base == (base + size))
1299			mem_ranges[idx].size += mem_ranges[i].size;
1300		else {
1301			idx++;
1302			if (i == idx)
1303				continue;
1304
1305			mem_ranges[idx] = mem_ranges[i];
1306		}
1307	}
1308	mrange_info->mem_range_cnt = idx + 1;
1309}
1310
1311/*
1312 * Scan reserved-ranges to consider them while reserving/releasing
1313 * memory for FADump.
1314 */
1315static void __init early_init_dt_scan_reserved_ranges(unsigned long node)
1316{
1317	const __be32 *prop;
1318	int len, ret = -1;
1319	unsigned long i;
1320
1321	/* reserved-ranges already scanned */
1322	if (reserved_mrange_info.mem_range_cnt != 0)
1323		return;
1324
1325	prop = of_get_flat_dt_prop(node, "reserved-ranges", &len);
1326	if (!prop)
1327		return;
1328
1329	/*
1330	 * Each reserved range is an (address,size) pair, 2 cells each,
1331	 * totalling 4 cells per range.
1332	 */
1333	for (i = 0; i < len / (sizeof(*prop) * 4); i++) {
1334		u64 base, size;
1335
1336		base = of_read_number(prop + (i * 4) + 0, 2);
1337		size = of_read_number(prop + (i * 4) + 2, 2);
1338
1339		if (size) {
1340			ret = fadump_add_mem_range(&reserved_mrange_info,
1341						   base, base + size);
1342			if (ret < 0) {
1343				pr_warn("some reserved ranges are ignored!\n");
1344				break;
1345			}
1346		}
1347	}
1348
1349	/* Compact reserved ranges */
1350	sort_and_merge_mem_ranges(&reserved_mrange_info);
1351}
1352
1353/*
1354 * Release the memory that was reserved during early boot to preserve the
1355 * crash'ed kernel's memory contents except reserved dump area (permanent
1356 * reservation) and reserved ranges used by F/W. The released memory will
1357 * be available for general use.
1358 */
1359static void fadump_release_memory(u64 begin, u64 end)
1360{
1361	u64 ra_start, ra_end, tstart;
1362	int i, ret;
1363
1364	ra_start = fw_dump.reserve_dump_area_start;
1365	ra_end = ra_start + fw_dump.reserve_dump_area_size;
1366
1367	/*
1368	 * If reserved ranges array limit is hit, overwrite the last reserved
1369	 * memory range with reserved dump area to ensure it is excluded from
1370	 * the memory being released (reused for next FADump registration).
1371	 */
1372	if (reserved_mrange_info.mem_range_cnt ==
1373	    reserved_mrange_info.max_mem_ranges)
1374		reserved_mrange_info.mem_range_cnt--;
1375
1376	ret = fadump_add_mem_range(&reserved_mrange_info, ra_start, ra_end);
1377	if (ret != 0)
1378		return;
1379
1380	/* Get the reserved ranges list in order first. */
1381	sort_and_merge_mem_ranges(&reserved_mrange_info);
1382
1383	/* Exclude reserved ranges and release remaining memory */
1384	tstart = begin;
1385	for (i = 0; i < reserved_mrange_info.mem_range_cnt; i++) {
1386		ra_start = reserved_mrange_info.mem_ranges[i].base;
1387		ra_end = ra_start + reserved_mrange_info.mem_ranges[i].size;
1388
1389		if (tstart >= ra_end)
1390			continue;
1391
1392		if (tstart < ra_start)
1393			fadump_release_reserved_area(tstart, ra_start);
1394		tstart = ra_end;
1395	}
1396
1397	if (tstart < end)
1398		fadump_release_reserved_area(tstart, end);
1399}
1400
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1401static void fadump_invalidate_release_mem(void)
1402{
1403	mutex_lock(&fadump_mutex);
1404	if (!fw_dump.dump_active) {
1405		mutex_unlock(&fadump_mutex);
1406		return;
1407	}
1408
1409	fadump_cleanup();
1410	mutex_unlock(&fadump_mutex);
1411
 
1412	fadump_release_memory(fw_dump.boot_mem_top, memblock_end_of_DRAM());
1413	fadump_free_cpu_notes_buf();
1414
1415	/*
1416	 * Setup kernel metadata and initialize the kernel dump
1417	 * memory structure for FADump re-registration.
1418	 */
1419	if (fw_dump.ops->fadump_setup_metadata &&
1420	    (fw_dump.ops->fadump_setup_metadata(&fw_dump) < 0))
1421		pr_warn("Failed to setup kernel metadata!\n");
1422	fw_dump.ops->fadump_init_mem_struct(&fw_dump);
1423}
1424
1425static ssize_t release_mem_store(struct kobject *kobj,
1426				 struct kobj_attribute *attr,
1427				 const char *buf, size_t count)
1428{
1429	int input = -1;
1430
1431	if (!fw_dump.dump_active)
1432		return -EPERM;
1433
1434	if (kstrtoint(buf, 0, &input))
1435		return -EINVAL;
1436
1437	if (input == 1) {
1438		/*
1439		 * Take away the '/proc/vmcore'. We are releasing the dump
1440		 * memory, hence it will not be valid anymore.
1441		 */
1442#ifdef CONFIG_PROC_VMCORE
1443		vmcore_cleanup();
1444#endif
1445		fadump_invalidate_release_mem();
1446
1447	} else
1448		return -EINVAL;
1449	return count;
1450}
1451
1452/* Release the reserved memory and disable the FADump */
1453static void unregister_fadump(void)
1454{
1455	fadump_cleanup();
1456	fadump_release_memory(fw_dump.reserve_dump_area_start,
1457			      fw_dump.reserve_dump_area_size);
1458	fw_dump.fadump_enabled = 0;
1459	kobject_put(fadump_kobj);
1460}
1461
1462static ssize_t enabled_show(struct kobject *kobj,
1463			    struct kobj_attribute *attr,
1464			    char *buf)
1465{
1466	return sprintf(buf, "%d\n", fw_dump.fadump_enabled);
1467}
1468
 
 
 
 
 
 
 
 
 
 
 
 
1469static ssize_t mem_reserved_show(struct kobject *kobj,
1470				 struct kobj_attribute *attr,
1471				 char *buf)
1472{
1473	return sprintf(buf, "%ld\n", fw_dump.reserve_dump_area_size);
1474}
1475
1476static ssize_t registered_show(struct kobject *kobj,
1477			       struct kobj_attribute *attr,
1478			       char *buf)
1479{
1480	return sprintf(buf, "%d\n", fw_dump.dump_registered);
1481}
1482
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1483static ssize_t registered_store(struct kobject *kobj,
1484				struct kobj_attribute *attr,
1485				const char *buf, size_t count)
1486{
1487	int ret = 0;
1488	int input = -1;
1489
1490	if (!fw_dump.fadump_enabled || fw_dump.dump_active)
1491		return -EPERM;
1492
1493	if (kstrtoint(buf, 0, &input))
1494		return -EINVAL;
1495
1496	mutex_lock(&fadump_mutex);
1497
1498	switch (input) {
1499	case 0:
1500		if (fw_dump.dump_registered == 0) {
1501			goto unlock_out;
1502		}
1503
1504		/* Un-register Firmware-assisted dump */
1505		pr_debug("Un-register firmware-assisted dump\n");
1506		fw_dump.ops->fadump_unregister(&fw_dump);
1507		break;
1508	case 1:
1509		if (fw_dump.dump_registered == 1) {
1510			/* Un-register Firmware-assisted dump */
1511			fw_dump.ops->fadump_unregister(&fw_dump);
1512		}
1513		/* Register Firmware-assisted dump */
1514		ret = register_fadump();
1515		break;
1516	default:
1517		ret = -EINVAL;
1518		break;
1519	}
1520
1521unlock_out:
1522	mutex_unlock(&fadump_mutex);
1523	return ret < 0 ? ret : count;
1524}
1525
1526static int fadump_region_show(struct seq_file *m, void *private)
1527{
1528	if (!fw_dump.fadump_enabled)
1529		return 0;
1530
1531	mutex_lock(&fadump_mutex);
1532	fw_dump.ops->fadump_region_show(&fw_dump, m);
1533	mutex_unlock(&fadump_mutex);
1534	return 0;
1535}
1536
1537static struct kobj_attribute release_attr = __ATTR_WO(release_mem);
1538static struct kobj_attribute enable_attr = __ATTR_RO(enabled);
1539static struct kobj_attribute register_attr = __ATTR_RW(registered);
1540static struct kobj_attribute mem_reserved_attr = __ATTR_RO(mem_reserved);
 
 
1541
1542static struct attribute *fadump_attrs[] = {
1543	&enable_attr.attr,
1544	&register_attr.attr,
1545	&mem_reserved_attr.attr,
 
1546	NULL,
1547};
1548
1549ATTRIBUTE_GROUPS(fadump);
1550
1551DEFINE_SHOW_ATTRIBUTE(fadump_region);
1552
1553static void fadump_init_files(void)
1554{
1555	int rc = 0;
1556
1557	fadump_kobj = kobject_create_and_add("fadump", kernel_kobj);
1558	if (!fadump_kobj) {
1559		pr_err("failed to create fadump kobject\n");
1560		return;
1561	}
1562
1563	debugfs_create_file("fadump_region", 0444, powerpc_debugfs_root, NULL,
 
 
 
 
 
 
1564			    &fadump_region_fops);
1565
1566	if (fw_dump.dump_active) {
1567		rc = sysfs_create_file(fadump_kobj, &release_attr.attr);
1568		if (rc)
1569			pr_err("unable to create release_mem sysfs file (%d)\n",
1570			       rc);
1571	}
1572
1573	rc = sysfs_create_groups(fadump_kobj, fadump_groups);
1574	if (rc) {
1575		pr_err("sysfs group creation failed (%d), unregistering FADump",
1576		       rc);
1577		unregister_fadump();
1578		return;
1579	}
1580
1581	/*
1582	 * The FADump sysfs are moved from kernel_kobj to fadump_kobj need to
1583	 * create symlink at old location to maintain backward compatibility.
1584	 *
1585	 *      - fadump_enabled -> fadump/enabled
1586	 *      - fadump_registered -> fadump/registered
1587	 *      - fadump_release_mem -> fadump/release_mem
1588	 */
1589	rc = compat_only_sysfs_link_entry_to_kobj(kernel_kobj, fadump_kobj,
1590						  "enabled", "fadump_enabled");
1591	if (rc) {
1592		pr_err("unable to create fadump_enabled symlink (%d)", rc);
1593		return;
1594	}
1595
1596	rc = compat_only_sysfs_link_entry_to_kobj(kernel_kobj, fadump_kobj,
1597						  "registered",
1598						  "fadump_registered");
1599	if (rc) {
1600		pr_err("unable to create fadump_registered symlink (%d)", rc);
1601		sysfs_remove_link(kernel_kobj, "fadump_enabled");
1602		return;
1603	}
1604
1605	if (fw_dump.dump_active) {
1606		rc = compat_only_sysfs_link_entry_to_kobj(kernel_kobj,
1607							  fadump_kobj,
1608							  "release_mem",
1609							  "fadump_release_mem");
1610		if (rc)
1611			pr_err("unable to create fadump_release_mem symlink (%d)",
1612			       rc);
1613	}
1614	return;
1615}
1616
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1617/*
1618 * Prepare for firmware-assisted dump.
1619 */
1620int __init setup_fadump(void)
1621{
1622	if (!fw_dump.fadump_supported)
1623		return 0;
1624
1625	fadump_init_files();
1626	fadump_show_config();
1627
1628	if (!fw_dump.fadump_enabled)
1629		return 1;
1630
1631	/*
1632	 * If dump data is available then see if it is valid and prepare for
1633	 * saving it to the disk.
1634	 */
1635	if (fw_dump.dump_active) {
1636		/*
1637		 * if dump process fails then invalidate the registration
1638		 * and release memory before proceeding for re-registration.
1639		 */
1640		if (fw_dump.ops->fadump_process(&fw_dump) < 0)
1641			fadump_invalidate_release_mem();
1642	}
1643	/* Initialize the kernel dump memory structure for FAD registration. */
1644	else if (fw_dump.reserve_dump_area_size)
1645		fw_dump.ops->fadump_init_mem_struct(&fw_dump);
 
 
 
 
 
 
 
 
 
 
1646
1647	return 1;
1648}
1649subsys_initcall(setup_fadump);
 
 
 
 
 
1650#else /* !CONFIG_PRESERVE_FA_DUMP */
1651
1652/* Scan the Firmware Assisted dump configuration details. */
1653int __init early_init_dt_scan_fw_dump(unsigned long node, const char *uname,
1654				      int depth, void *data)
1655{
1656	if ((depth != 1) || (strcmp(uname, "ibm,opal") != 0))
1657		return 0;
1658
1659	opal_fadump_dt_scan(&fw_dump, node);
1660	return 1;
1661}
1662
1663/*
1664 * When dump is active but PRESERVE_FA_DUMP is enabled on the kernel,
1665 * preserve crash data. The subsequent memory preserving kernel boot
1666 * is likely to process this crash data.
1667 */
1668int __init fadump_reserve_mem(void)
1669{
1670	if (fw_dump.dump_active) {
1671		/*
1672		 * If last boot has crashed then reserve all the memory
1673		 * above boot memory to preserve crash data.
1674		 */
1675		pr_info("Preserving crash data for processing in next boot.\n");
1676		fadump_reserve_crash_area(fw_dump.boot_mem_top);
1677	} else
1678		pr_debug("FADump-aware kernel..\n");
1679
1680	return 1;
1681}
1682#endif /* CONFIG_PRESERVE_FA_DUMP */
1683
1684/* Preserve everything above the base address */
1685static void __init fadump_reserve_crash_area(u64 base)
1686{
1687	struct memblock_region *reg;
1688	u64 mstart, msize;
1689
1690	for_each_memblock(memory, reg) {
1691		mstart = reg->base;
1692		msize  = reg->size;
1693
1694		if ((mstart + msize) < base)
1695			continue;
1696
1697		if (mstart < base) {
1698			msize -= (base - mstart);
1699			mstart = base;
1700		}
1701
1702		pr_info("Reserving %lluMB of memory at %#016llx for preserving crash data",
1703			(msize >> 20), mstart);
1704		memblock_reserve(mstart, msize);
1705	}
1706}
1707
1708unsigned long __init arch_reserved_kernel_pages(void)
1709{
1710	return memblock_reserved_size() / PAGE_SIZE;
1711}
v6.13.7
   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, &reg_start, &reg_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, &reg_start, &reg_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, &reg_spfn, &reg_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	&register_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}