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}
v5.4
   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
 
 
 
 
 
 
 
 
  35static struct fw_dump fw_dump;
  36
  37static void __init fadump_reserve_crash_area(u64 base);
  38
 
 
  39#ifndef CONFIG_PRESERVE_FA_DUMP
 
 
  40static DEFINE_MUTEX(fadump_mutex);
  41struct fadump_mrange_info crash_mrange_info = { "crash", NULL, 0, 0, 0 };
  42struct fadump_mrange_info reserved_mrange_info = { "reserved", NULL, 0, 0, 0 };
 
 
 
 
 
 
 
 
 
 
  43
  44#ifdef CONFIG_CMA
  45static struct cma *fadump_cma;
  46
  47/*
  48 * fadump_cma_init() - Initialize CMA area from a fadump reserved memory
  49 *
  50 * This function initializes CMA area from fadump reserved memory.
  51 * The total size of fadump reserved memory covers for boot memory size
  52 * + cpu data size + hpte size and metadata.
  53 * Initialize only the area equivalent to boot memory size for CMA use.
  54 * The reamining portion of fadump reserved memory will be not given
  55 * to CMA and pages for thoes will stay reserved. boot memory size is
  56 * aligned per CMA requirement to satisy cma_init_reserved_mem() call.
  57 * But for some reason even if it fails we still have the memory reservation
  58 * with us and we can still continue doing fadump.
  59 */
  60int __init fadump_cma_init(void)
  61{
  62	unsigned long long base, size;
  63	int rc;
  64
  65	if (!fw_dump.fadump_enabled)
  66		return 0;
  67
  68	/*
  69	 * Do not use CMA if user has provided fadump=nocma kernel parameter.
  70	 * Return 1 to continue with fadump old behaviour.
  71	 */
  72	if (fw_dump.nocma)
  73		return 1;
  74
  75	base = fw_dump.reserve_dump_area_start;
  76	size = fw_dump.boot_memory_size;
  77
  78	if (!size)
  79		return 0;
  80
  81	rc = cma_init_reserved_mem(base, size, 0, "fadump_cma", &fadump_cma);
  82	if (rc) {
  83		pr_err("Failed to init cma area for firmware-assisted dump,%d\n", rc);
  84		/*
  85		 * Though the CMA init has failed we still have memory
  86		 * reservation with us. The reserved memory will be
  87		 * blocked from production system usage.  Hence return 1,
  88		 * so that we can continue with fadump.
  89		 */
  90		return 1;
  91	}
  92
  93	/*
  94	 * So we now have successfully initialized cma area for fadump.
  95	 */
  96	pr_info("Initialized 0x%lx bytes cma area at %ldMB from 0x%lx "
  97		"bytes of memory reserved for firmware-assisted dump\n",
  98		cma_get_size(fadump_cma),
  99		(unsigned long)cma_get_base(fadump_cma) >> 20,
 100		fw_dump.reserve_dump_area_size);
 101	return 1;
 102}
 103#else
 104static int __init fadump_cma_init(void) { return 1; }
 105#endif /* CONFIG_CMA */
 106
 107/* Scan the Firmware Assisted dump configuration details. */
 108int __init early_init_dt_scan_fw_dump(unsigned long node, const char *uname,
 109				      int depth, void *data)
 110{
 
 
 
 
 
 111	if (depth != 1)
 112		return 0;
 113
 114	if (strcmp(uname, "rtas") == 0) {
 115		rtas_fadump_dt_scan(&fw_dump, node);
 116		return 1;
 117	}
 118
 119	if (strcmp(uname, "ibm,opal") == 0) {
 120		opal_fadump_dt_scan(&fw_dump, node);
 121		return 1;
 122	}
 123
 124	return 0;
 125}
 126
 127/*
 128 * If fadump is registered, check if the memory provided
 129 * falls within boot memory area and reserved memory area.
 130 */
 131int is_fadump_memory_area(u64 addr, unsigned long size)
 132{
 133	u64 d_start, d_end;
 134
 135	if (!fw_dump.dump_registered)
 136		return 0;
 137
 138	if (!size)
 139		return 0;
 140
 141	d_start = fw_dump.reserve_dump_area_start;
 142	d_end = d_start + fw_dump.reserve_dump_area_size;
 143	if (((addr + size) > d_start) && (addr <= d_end))
 144		return 1;
 145
 146	return (addr <= fw_dump.boot_mem_top);
 147}
 148
 149int should_fadump_crash(void)
 150{
 151	if (!fw_dump.dump_registered || !fw_dump.fadumphdr_addr)
 152		return 0;
 153	return 1;
 154}
 155
 156int is_fadump_active(void)
 157{
 158	return fw_dump.dump_active;
 159}
 160
 161/*
 162 * Returns true, if there are no holes in memory area between d_start to d_end,
 163 * false otherwise.
 164 */
 165static bool is_fadump_mem_area_contiguous(u64 d_start, u64 d_end)
 166{
 167	struct memblock_region *reg;
 168	bool ret = false;
 169	u64 start, end;
 170
 171	for_each_memblock(memory, reg) {
 172		start = max_t(u64, d_start, reg->base);
 173		end = min_t(u64, d_end, (reg->base + reg->size));
 174		if (d_start < end) {
 175			/* Memory hole from d_start to start */
 176			if (start > d_start)
 177				break;
 178
 179			if (end == d_end) {
 180				ret = true;
 181				break;
 182			}
 183
 184			d_start = end + 1;
 185		}
 186	}
 187
 188	return ret;
 189}
 190
 191/*
 192 * Returns true, if there are no holes in boot memory area,
 193 * false otherwise.
 194 */
 195bool is_fadump_boot_mem_contiguous(void)
 196{
 197	unsigned long d_start, d_end;
 198	bool ret = false;
 199	int i;
 200
 201	for (i = 0; i < fw_dump.boot_mem_regs_cnt; i++) {
 202		d_start = fw_dump.boot_mem_addr[i];
 203		d_end   = d_start + fw_dump.boot_mem_sz[i];
 204
 205		ret = is_fadump_mem_area_contiguous(d_start, d_end);
 206		if (!ret)
 207			break;
 208	}
 209
 210	return ret;
 211}
 212
 213/*
 214 * Returns true, if there are no holes in reserved memory area,
 215 * false otherwise.
 216 */
 217bool is_fadump_reserved_mem_contiguous(void)
 218{
 219	u64 d_start, d_end;
 220
 221	d_start	= fw_dump.reserve_dump_area_start;
 222	d_end	= d_start + fw_dump.reserve_dump_area_size;
 223	return is_fadump_mem_area_contiguous(d_start, d_end);
 224}
 225
 226/* Print firmware assisted dump configurations for debugging purpose. */
 227static void fadump_show_config(void)
 228{
 229	int i;
 230
 231	pr_debug("Support for firmware-assisted dump (fadump): %s\n",
 232			(fw_dump.fadump_supported ? "present" : "no support"));
 233
 234	if (!fw_dump.fadump_supported)
 235		return;
 236
 237	pr_debug("Fadump enabled    : %s\n",
 238				(fw_dump.fadump_enabled ? "yes" : "no"));
 239	pr_debug("Dump Active       : %s\n",
 240				(fw_dump.dump_active ? "yes" : "no"));
 241	pr_debug("Dump section sizes:\n");
 242	pr_debug("    CPU state data size: %lx\n", fw_dump.cpu_state_data_size);
 243	pr_debug("    HPTE region size   : %lx\n", fw_dump.hpte_region_size);
 244	pr_debug("    Boot memory size   : %lx\n", fw_dump.boot_memory_size);
 245	pr_debug("    Boot memory top    : %llx\n", fw_dump.boot_mem_top);
 246	pr_debug("Boot memory regions cnt: %llx\n", fw_dump.boot_mem_regs_cnt);
 247	for (i = 0; i < fw_dump.boot_mem_regs_cnt; i++) {
 248		pr_debug("[%03d] base = %llx, size = %llx\n", i,
 249			 fw_dump.boot_mem_addr[i], fw_dump.boot_mem_sz[i]);
 250	}
 251}
 252
 253/**
 254 * fadump_calculate_reserve_size(): reserve variable boot area 5% of System RAM
 255 *
 256 * Function to find the largest memory size we need to reserve during early
 257 * boot process. This will be the size of the memory that is required for a
 258 * kernel to boot successfully.
 259 *
 260 * This function has been taken from phyp-assisted dump feature implementation.
 261 *
 262 * returns larger of 256MB or 5% rounded down to multiples of 256MB.
 263 *
 264 * TODO: Come up with better approach to find out more accurate memory size
 265 * that is required for a kernel to boot successfully.
 266 *
 267 */
 268static inline u64 fadump_calculate_reserve_size(void)
 269{
 270	u64 base, size, bootmem_min;
 271	int ret;
 272
 273	if (fw_dump.reserve_bootvar)
 274		pr_warn("'fadump_reserve_mem=' parameter is deprecated in favor of 'crashkernel=' parameter.\n");
 275
 276	/*
 277	 * Check if the size is specified through crashkernel= cmdline
 278	 * option. If yes, then use that but ignore base as fadump reserves
 279	 * memory at a predefined offset.
 280	 */
 281	ret = parse_crashkernel(boot_command_line, memblock_phys_mem_size(),
 282				&size, &base);
 283	if (ret == 0 && size > 0) {
 284		unsigned long max_size;
 285
 286		if (fw_dump.reserve_bootvar)
 287			pr_info("Using 'crashkernel=' parameter for memory reservation.\n");
 288
 289		fw_dump.reserve_bootvar = (unsigned long)size;
 290
 291		/*
 292		 * Adjust if the boot memory size specified is above
 293		 * the upper limit.
 294		 */
 295		max_size = memblock_phys_mem_size() / MAX_BOOT_MEM_RATIO;
 296		if (fw_dump.reserve_bootvar > max_size) {
 297			fw_dump.reserve_bootvar = max_size;
 298			pr_info("Adjusted boot memory size to %luMB\n",
 299				(fw_dump.reserve_bootvar >> 20));
 300		}
 301
 302		return fw_dump.reserve_bootvar;
 303	} else if (fw_dump.reserve_bootvar) {
 304		/*
 305		 * 'fadump_reserve_mem=' is being used to reserve memory
 306		 * for firmware-assisted dump.
 307		 */
 308		return fw_dump.reserve_bootvar;
 309	}
 310
 311	/* divide by 20 to get 5% of value */
 312	size = memblock_phys_mem_size() / 20;
 313
 314	/* round it down in multiples of 256 */
 315	size = size & ~0x0FFFFFFFUL;
 316
 317	/* Truncate to memory_limit. We don't want to over reserve the memory.*/
 318	if (memory_limit && size > memory_limit)
 319		size = memory_limit;
 320
 321	bootmem_min = fw_dump.ops->fadump_get_bootmem_min();
 322	return (size > bootmem_min ? size : bootmem_min);
 323}
 324
 325/*
 326 * Calculate the total memory size required to be reserved for
 327 * firmware-assisted dump registration.
 328 */
 329static unsigned long get_fadump_area_size(void)
 330{
 331	unsigned long size = 0;
 332
 333	size += fw_dump.cpu_state_data_size;
 334	size += fw_dump.hpte_region_size;
 335	size += fw_dump.boot_memory_size;
 336	size += sizeof(struct fadump_crash_info_header);
 337	size += sizeof(struct elfhdr); /* ELF core header.*/
 338	size += sizeof(struct elf_phdr); /* place holder for cpu notes */
 339	/* Program headers for crash memory regions. */
 340	size += sizeof(struct elf_phdr) * (memblock_num_regions(memory) + 2);
 341
 342	size = PAGE_ALIGN(size);
 343
 344	/* This is to hold kernel metadata on platforms that support it */
 345	size += (fw_dump.ops->fadump_get_metadata_size ?
 346		 fw_dump.ops->fadump_get_metadata_size() : 0);
 347	return size;
 348}
 349
 350static int __init add_boot_mem_region(unsigned long rstart,
 351				      unsigned long rsize)
 352{
 353	int i = fw_dump.boot_mem_regs_cnt++;
 354
 355	if (fw_dump.boot_mem_regs_cnt > FADUMP_MAX_MEM_REGS) {
 356		fw_dump.boot_mem_regs_cnt = FADUMP_MAX_MEM_REGS;
 357		return 0;
 358	}
 359
 360	pr_debug("Added boot memory range[%d] [%#016lx-%#016lx)\n",
 361		 i, rstart, (rstart + rsize));
 362	fw_dump.boot_mem_addr[i] = rstart;
 363	fw_dump.boot_mem_sz[i] = rsize;
 364	return 1;
 365}
 366
 367/*
 368 * Firmware usually has a hard limit on the data it can copy per region.
 369 * Honour that by splitting a memory range into multiple regions.
 370 */
 371static int __init add_boot_mem_regions(unsigned long mstart,
 372				       unsigned long msize)
 373{
 374	unsigned long rstart, rsize, max_size;
 375	int ret = 1;
 376
 377	rstart = mstart;
 378	max_size = fw_dump.max_copy_size ? fw_dump.max_copy_size : msize;
 379	while (msize) {
 380		if (msize > max_size)
 381			rsize = max_size;
 382		else
 383			rsize = msize;
 384
 385		ret = add_boot_mem_region(rstart, rsize);
 386		if (!ret)
 387			break;
 388
 389		msize -= rsize;
 390		rstart += rsize;
 391	}
 392
 393	return ret;
 394}
 395
 396static int __init fadump_get_boot_mem_regions(void)
 397{
 398	unsigned long base, size, cur_size, hole_size, last_end;
 399	unsigned long mem_size = fw_dump.boot_memory_size;
 400	struct memblock_region *reg;
 401	int ret = 1;
 402
 403	fw_dump.boot_mem_regs_cnt = 0;
 404
 405	last_end = 0;
 406	hole_size = 0;
 407	cur_size = 0;
 408	for_each_memblock(memory, reg) {
 409		base = reg->base;
 410		size = reg->size;
 411		hole_size += (base - last_end);
 412
 413		if ((cur_size + size) >= mem_size) {
 414			size = (mem_size - cur_size);
 415			ret = add_boot_mem_regions(base, size);
 416			break;
 417		}
 418
 419		mem_size -= size;
 420		cur_size += size;
 421		ret = add_boot_mem_regions(base, size);
 422		if (!ret)
 423			break;
 424
 425		last_end = base + size;
 426	}
 427	fw_dump.boot_mem_top = PAGE_ALIGN(fw_dump.boot_memory_size + hole_size);
 428
 429	return ret;
 430}
 431
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 432int __init fadump_reserve_mem(void)
 433{
 434	u64 base, size, mem_boundary, bootmem_min, align = PAGE_SIZE;
 435	bool is_memblock_bottom_up = memblock_bottom_up();
 436	int ret = 1;
 437
 438	if (!fw_dump.fadump_enabled)
 439		return 0;
 440
 441	if (!fw_dump.fadump_supported) {
 442		pr_info("Firmware-Assisted Dump is not supported on this hardware\n");
 443		goto error_out;
 444	}
 445
 446	/*
 447	 * Initialize boot memory size
 448	 * If dump is active then we have already calculated the size during
 449	 * first kernel.
 450	 */
 451	if (!fw_dump.dump_active) {
 452		fw_dump.boot_memory_size =
 453			PAGE_ALIGN(fadump_calculate_reserve_size());
 454#ifdef CONFIG_CMA
 455		if (!fw_dump.nocma) {
 456			align = FADUMP_CMA_ALIGNMENT;
 457			fw_dump.boot_memory_size =
 458				ALIGN(fw_dump.boot_memory_size, align);
 
 459		}
 460#endif
 461
 462		bootmem_min = fw_dump.ops->fadump_get_bootmem_min();
 463		if (fw_dump.boot_memory_size < bootmem_min) {
 464			pr_err("Can't enable fadump with boot memory size (0x%lx) less than 0x%llx\n",
 465			       fw_dump.boot_memory_size, bootmem_min);
 466			goto error_out;
 467		}
 468
 469		if (!fadump_get_boot_mem_regions()) {
 470			pr_err("Too many holes in boot memory area to enable fadump\n");
 471			goto error_out;
 472		}
 473	}
 474
 475	/*
 476	 * Calculate the memory boundary.
 477	 * If memory_limit is less than actual memory boundary then reserve
 478	 * the memory for fadump beyond the memory_limit and adjust the
 479	 * memory_limit accordingly, so that the running kernel can run with
 480	 * specified memory_limit.
 481	 */
 482	if (memory_limit && memory_limit < memblock_end_of_DRAM()) {
 483		size = get_fadump_area_size();
 484		if ((memory_limit + size) < memblock_end_of_DRAM())
 485			memory_limit += size;
 486		else
 487			memory_limit = memblock_end_of_DRAM();
 488		printk(KERN_INFO "Adjusted memory_limit for firmware-assisted"
 489				" dump, now %#016llx\n", memory_limit);
 490	}
 491	if (memory_limit)
 492		mem_boundary = memory_limit;
 493	else
 494		mem_boundary = memblock_end_of_DRAM();
 495
 496	base = fw_dump.boot_mem_top;
 497	size = get_fadump_area_size();
 498	fw_dump.reserve_dump_area_size = size;
 499	if (fw_dump.dump_active) {
 500		pr_info("Firmware-assisted dump is active.\n");
 501
 502#ifdef CONFIG_HUGETLB_PAGE
 503		/*
 504		 * FADump capture kernel doesn't care much about hugepages.
 505		 * In fact, handling hugepages in capture kernel is asking for
 506		 * trouble. So, disable HugeTLB support when fadump is active.
 507		 */
 508		hugetlb_disabled = true;
 509#endif
 510		/*
 511		 * If last boot has crashed then reserve all the memory
 512		 * above boot memory size so that we don't touch it until
 513		 * dump is written to disk by userspace tool. This memory
 514		 * can be released for general use by invalidating fadump.
 515		 */
 516		fadump_reserve_crash_area(base);
 517
 518		pr_debug("fadumphdr_addr = %#016lx\n", fw_dump.fadumphdr_addr);
 519		pr_debug("Reserve dump area start address: 0x%lx\n",
 520			 fw_dump.reserve_dump_area_start);
 521	} else {
 522		/*
 523		 * Reserve memory at an offset closer to bottom of the RAM to
 524		 * minimize the impact of memory hot-remove operation.
 525		 */
 526		memblock_set_bottom_up(true);
 527		base = memblock_find_in_range(base, mem_boundary, size, align);
 528
 529		/* Restore the previous allocation mode */
 530		memblock_set_bottom_up(is_memblock_bottom_up);
 531
 532		if (!base) {
 533			pr_err("Failed to find memory chunk for reservation!\n");
 534			goto error_out;
 535		}
 536		fw_dump.reserve_dump_area_start = base;
 537
 538		/*
 539		 * Calculate the kernel metadata address and register it with
 540		 * f/w if the platform supports.
 541		 */
 542		if (fw_dump.ops->fadump_setup_metadata &&
 543		    (fw_dump.ops->fadump_setup_metadata(&fw_dump) < 0))
 544			goto error_out;
 545
 546		if (memblock_reserve(base, size)) {
 547			pr_err("Failed to reserve memory!\n");
 548			goto error_out;
 549		}
 550
 551		pr_info("Reserved %lldMB of memory at %#016llx (System RAM: %lldMB)\n",
 552			(size >> 20), base, (memblock_phys_mem_size() >> 20));
 553
 554		ret = fadump_cma_init();
 555	}
 556
 557	return ret;
 558error_out:
 559	fw_dump.fadump_enabled = 0;
 560	return 0;
 561}
 562
 563/* Look for fadump= cmdline option. */
 564static int __init early_fadump_param(char *p)
 565{
 566	if (!p)
 567		return 1;
 568
 569	if (strncmp(p, "on", 2) == 0)
 570		fw_dump.fadump_enabled = 1;
 571	else if (strncmp(p, "off", 3) == 0)
 572		fw_dump.fadump_enabled = 0;
 573	else if (strncmp(p, "nocma", 5) == 0) {
 574		fw_dump.fadump_enabled = 1;
 575		fw_dump.nocma = 1;
 576	}
 577
 578	return 0;
 579}
 580early_param("fadump", early_fadump_param);
 581
 582/*
 583 * Look for fadump_reserve_mem= cmdline option
 584 * TODO: Remove references to 'fadump_reserve_mem=' parameter,
 585 *       the sooner 'crashkernel=' parameter is accustomed to.
 586 */
 587static int __init early_fadump_reserve_mem(char *p)
 588{
 589	if (p)
 590		fw_dump.reserve_bootvar = memparse(p, &p);
 591	return 0;
 592}
 593early_param("fadump_reserve_mem", early_fadump_reserve_mem);
 594
 595void crash_fadump(struct pt_regs *regs, const char *str)
 596{
 
 597	struct fadump_crash_info_header *fdh = NULL;
 598	int old_cpu, this_cpu;
 
 
 599
 600	if (!should_fadump_crash())
 601		return;
 602
 603	/*
 604	 * old_cpu == -1 means this is the first CPU which has come here,
 605	 * go ahead and trigger fadump.
 606	 *
 607	 * old_cpu != -1 means some other CPU has already on it's way
 608	 * to trigger fadump, just keep looping here.
 609	 */
 610	this_cpu = smp_processor_id();
 611	old_cpu = cmpxchg(&crashing_cpu, -1, this_cpu);
 612
 613	if (old_cpu != -1) {
 
 
 614		/*
 615		 * We can't loop here indefinitely. Wait as long as fadump
 616		 * is in force. If we race with fadump un-registration this
 617		 * loop will break and then we go down to normal panic path
 618		 * and reboot. If fadump is in force the first crashing
 619		 * cpu will definitely trigger fadump.
 620		 */
 621		while (fw_dump.dump_registered)
 622			cpu_relax();
 623		return;
 624	}
 625
 626	fdh = __va(fw_dump.fadumphdr_addr);
 627	fdh->crashing_cpu = crashing_cpu;
 628	crash_save_vmcoreinfo();
 629
 630	if (regs)
 631		fdh->regs = *regs;
 632	else
 633		ppc_save_regs(&fdh->regs);
 634
 635	fdh->online_mask = *cpu_online_mask;
 636
 
 
 
 
 
 
 
 
 
 
 637	fw_dump.ops->fadump_trigger(fdh, str);
 638}
 639
 640u32 *fadump_regs_to_elf_notes(u32 *buf, struct pt_regs *regs)
 641{
 642	struct elf_prstatus prstatus;
 643
 644	memset(&prstatus, 0, sizeof(prstatus));
 645	/*
 646	 * FIXME: How do i get PID? Do I really need it?
 647	 * prstatus.pr_pid = ????
 648	 */
 649	elf_core_copy_kernel_regs(&prstatus.pr_reg, regs);
 650	buf = append_elf_note(buf, CRASH_CORE_NOTE_NAME, NT_PRSTATUS,
 651			      &prstatus, sizeof(prstatus));
 652	return buf;
 653}
 654
 655void fadump_update_elfcore_header(char *bufp)
 656{
 657	struct elfhdr *elf;
 658	struct elf_phdr *phdr;
 659
 660	elf = (struct elfhdr *)bufp;
 661	bufp += sizeof(struct elfhdr);
 662
 663	/* First note is a place holder for cpu notes info. */
 664	phdr = (struct elf_phdr *)bufp;
 665
 666	if (phdr->p_type == PT_NOTE) {
 667		phdr->p_paddr	= __pa(fw_dump.cpu_notes_buf_vaddr);
 668		phdr->p_offset	= phdr->p_paddr;
 669		phdr->p_filesz	= fw_dump.cpu_notes_buf_size;
 670		phdr->p_memsz = fw_dump.cpu_notes_buf_size;
 671	}
 672	return;
 673}
 674
 675static void *fadump_alloc_buffer(unsigned long size)
 676{
 677	unsigned long count, i;
 678	struct page *page;
 679	void *vaddr;
 680
 681	vaddr = alloc_pages_exact(size, GFP_KERNEL | __GFP_ZERO);
 682	if (!vaddr)
 683		return NULL;
 684
 685	count = PAGE_ALIGN(size) / PAGE_SIZE;
 686	page = virt_to_page(vaddr);
 687	for (i = 0; i < count; i++)
 688		mark_page_reserved(page + i);
 689	return vaddr;
 690}
 691
 692static void fadump_free_buffer(unsigned long vaddr, unsigned long size)
 693{
 694	free_reserved_area((void *)vaddr, (void *)(vaddr + size), -1, NULL);
 695}
 696
 697s32 fadump_setup_cpu_notes_buf(u32 num_cpus)
 698{
 699	/* Allocate buffer to hold cpu crash notes. */
 700	fw_dump.cpu_notes_buf_size = num_cpus * sizeof(note_buf_t);
 701	fw_dump.cpu_notes_buf_size = PAGE_ALIGN(fw_dump.cpu_notes_buf_size);
 702	fw_dump.cpu_notes_buf_vaddr =
 703		(unsigned long)fadump_alloc_buffer(fw_dump.cpu_notes_buf_size);
 704	if (!fw_dump.cpu_notes_buf_vaddr) {
 705		pr_err("Failed to allocate %ld bytes for CPU notes buffer\n",
 706		       fw_dump.cpu_notes_buf_size);
 707		return -ENOMEM;
 708	}
 709
 710	pr_debug("Allocated buffer for cpu notes of size %ld at 0x%lx\n",
 711		 fw_dump.cpu_notes_buf_size,
 712		 fw_dump.cpu_notes_buf_vaddr);
 713	return 0;
 714}
 715
 716void fadump_free_cpu_notes_buf(void)
 717{
 718	if (!fw_dump.cpu_notes_buf_vaddr)
 719		return;
 720
 721	fadump_free_buffer(fw_dump.cpu_notes_buf_vaddr,
 722			   fw_dump.cpu_notes_buf_size);
 723	fw_dump.cpu_notes_buf_vaddr = 0;
 724	fw_dump.cpu_notes_buf_size = 0;
 725}
 726
 727static void fadump_free_mem_ranges(struct fadump_mrange_info *mrange_info)
 728{
 
 
 
 
 
 729	kfree(mrange_info->mem_ranges);
 730	mrange_info->mem_ranges = NULL;
 731	mrange_info->mem_ranges_sz = 0;
 732	mrange_info->max_mem_ranges = 0;
 733}
 734
 735/*
 736 * Allocate or reallocate mem_ranges array in incremental units
 737 * of PAGE_SIZE.
 738 */
 739static int fadump_alloc_mem_ranges(struct fadump_mrange_info *mrange_info)
 740{
 741	struct fadump_memory_range *new_array;
 742	u64 new_size;
 743
 744	new_size = mrange_info->mem_ranges_sz + PAGE_SIZE;
 745	pr_debug("Allocating %llu bytes of memory for %s memory ranges\n",
 746		 new_size, mrange_info->name);
 747
 748	new_array = krealloc(mrange_info->mem_ranges, new_size, GFP_KERNEL);
 749	if (new_array == NULL) {
 750		pr_err("Insufficient memory for setting up %s memory ranges\n",
 751		       mrange_info->name);
 752		fadump_free_mem_ranges(mrange_info);
 753		return -ENOMEM;
 754	}
 755
 756	mrange_info->mem_ranges = new_array;
 757	mrange_info->mem_ranges_sz = new_size;
 758	mrange_info->max_mem_ranges = (new_size /
 759				       sizeof(struct fadump_memory_range));
 760	return 0;
 761}
 762
 763static inline int fadump_add_mem_range(struct fadump_mrange_info *mrange_info,
 764				       u64 base, u64 end)
 765{
 766	struct fadump_memory_range *mem_ranges = mrange_info->mem_ranges;
 767	bool is_adjacent = false;
 768	u64 start, size;
 769
 770	if (base == end)
 771		return 0;
 772
 773	/*
 774	 * Fold adjacent memory ranges to bring down the memory ranges/
 775	 * PT_LOAD segments count.
 776	 */
 777	if (mrange_info->mem_range_cnt) {
 778		start = mem_ranges[mrange_info->mem_range_cnt - 1].base;
 779		size  = mem_ranges[mrange_info->mem_range_cnt - 1].size;
 780
 781		if ((start + size) == base)
 782			is_adjacent = true;
 783	}
 784	if (!is_adjacent) {
 785		/* resize the array on reaching the limit */
 786		if (mrange_info->mem_range_cnt == mrange_info->max_mem_ranges) {
 787			int ret;
 788
 
 
 
 
 
 
 789			ret = fadump_alloc_mem_ranges(mrange_info);
 790			if (ret)
 791				return ret;
 792
 793			/* Update to the new resized array */
 794			mem_ranges = mrange_info->mem_ranges;
 795		}
 796
 797		start = base;
 798		mem_ranges[mrange_info->mem_range_cnt].base = start;
 799		mrange_info->mem_range_cnt++;
 800	}
 801
 802	mem_ranges[mrange_info->mem_range_cnt - 1].size = (end - start);
 803	pr_debug("%s_memory_range[%d] [%#016llx-%#016llx], %#llx bytes\n",
 804		 mrange_info->name, (mrange_info->mem_range_cnt - 1),
 805		 start, end - 1, (end - start));
 806	return 0;
 807}
 808
 809static int fadump_exclude_reserved_area(u64 start, u64 end)
 810{
 811	u64 ra_start, ra_end;
 812	int ret = 0;
 813
 814	ra_start = fw_dump.reserve_dump_area_start;
 815	ra_end = ra_start + fw_dump.reserve_dump_area_size;
 816
 817	if ((ra_start < end) && (ra_end > start)) {
 818		if ((start < ra_start) && (end > ra_end)) {
 819			ret = fadump_add_mem_range(&crash_mrange_info,
 820						   start, ra_start);
 821			if (ret)
 822				return ret;
 823
 824			ret = fadump_add_mem_range(&crash_mrange_info,
 825						   ra_end, end);
 826		} else if (start < ra_start) {
 827			ret = fadump_add_mem_range(&crash_mrange_info,
 828						   start, ra_start);
 829		} else if (ra_end < end) {
 830			ret = fadump_add_mem_range(&crash_mrange_info,
 831						   ra_end, end);
 832		}
 833	} else
 834		ret = fadump_add_mem_range(&crash_mrange_info, start, end);
 835
 836	return ret;
 837}
 838
 839static int fadump_init_elfcore_header(char *bufp)
 840{
 841	struct elfhdr *elf;
 842
 843	elf = (struct elfhdr *) bufp;
 844	bufp += sizeof(struct elfhdr);
 845	memcpy(elf->e_ident, ELFMAG, SELFMAG);
 846	elf->e_ident[EI_CLASS] = ELF_CLASS;
 847	elf->e_ident[EI_DATA] = ELF_DATA;
 848	elf->e_ident[EI_VERSION] = EV_CURRENT;
 849	elf->e_ident[EI_OSABI] = ELF_OSABI;
 850	memset(elf->e_ident+EI_PAD, 0, EI_NIDENT-EI_PAD);
 851	elf->e_type = ET_CORE;
 852	elf->e_machine = ELF_ARCH;
 853	elf->e_version = EV_CURRENT;
 854	elf->e_entry = 0;
 855	elf->e_phoff = sizeof(struct elfhdr);
 856	elf->e_shoff = 0;
 857#if defined(_CALL_ELF)
 858	elf->e_flags = _CALL_ELF;
 859#else
 860	elf->e_flags = 0;
 861#endif
 862	elf->e_ehsize = sizeof(struct elfhdr);
 863	elf->e_phentsize = sizeof(struct elf_phdr);
 864	elf->e_phnum = 0;
 865	elf->e_shentsize = 0;
 866	elf->e_shnum = 0;
 867	elf->e_shstrndx = 0;
 868
 869	return 0;
 870}
 871
 872/*
 873 * Traverse through memblock structure and setup crash memory ranges. These
 874 * ranges will be used create PT_LOAD program headers in elfcore header.
 875 */
 876static int fadump_setup_crash_memory_ranges(void)
 877{
 878	struct memblock_region *reg;
 879	u64 start, end;
 880	int i, ret;
 881
 882	pr_debug("Setup crash memory ranges.\n");
 883	crash_mrange_info.mem_range_cnt = 0;
 884
 885	/*
 886	 * Boot memory region(s) registered with firmware are moved to
 887	 * different location at the time of crash. Create separate program
 888	 * header(s) for this memory chunk(s) with the correct offset.
 889	 */
 890	for (i = 0; i < fw_dump.boot_mem_regs_cnt; i++) {
 891		start = fw_dump.boot_mem_addr[i];
 892		end = start + fw_dump.boot_mem_sz[i];
 893		ret = fadump_add_mem_range(&crash_mrange_info, start, end);
 894		if (ret)
 895			return ret;
 896	}
 897
 898	for_each_memblock(memory, reg) {
 899		start = (u64)reg->base;
 900		end = start + (u64)reg->size;
 901
 902		/*
 903		 * skip the memory chunk that is already added
 904		 * (0 through boot_memory_top).
 905		 */
 906		if (start < fw_dump.boot_mem_top) {
 907			if (end > fw_dump.boot_mem_top)
 908				start = fw_dump.boot_mem_top;
 909			else
 910				continue;
 911		}
 912
 913		/* add this range excluding the reserved dump area. */
 914		ret = fadump_exclude_reserved_area(start, end);
 915		if (ret)
 916			return ret;
 917	}
 918
 919	return 0;
 920}
 921
 922/*
 923 * If the given physical address falls within the boot memory region then
 924 * return the relocated address that points to the dump region reserved
 925 * for saving initial boot memory contents.
 926 */
 927static inline unsigned long fadump_relocate(unsigned long paddr)
 928{
 929	unsigned long raddr, rstart, rend, rlast, hole_size;
 930	int i;
 931
 932	hole_size = 0;
 933	rlast = 0;
 934	raddr = paddr;
 935	for (i = 0; i < fw_dump.boot_mem_regs_cnt; i++) {
 936		rstart = fw_dump.boot_mem_addr[i];
 937		rend = rstart + fw_dump.boot_mem_sz[i];
 938		hole_size += (rstart - rlast);
 939
 940		if (paddr >= rstart && paddr < rend) {
 941			raddr += fw_dump.boot_mem_dest_addr - hole_size;
 942			break;
 943		}
 944
 945		rlast = rend;
 946	}
 947
 948	pr_debug("vmcoreinfo: paddr = 0x%lx, raddr = 0x%lx\n", paddr, raddr);
 949	return raddr;
 950}
 951
 952static int fadump_create_elfcore_headers(char *bufp)
 953{
 954	unsigned long long raddr, offset;
 955	struct elf_phdr *phdr;
 956	struct elfhdr *elf;
 957	int i, j;
 958
 959	fadump_init_elfcore_header(bufp);
 960	elf = (struct elfhdr *)bufp;
 961	bufp += sizeof(struct elfhdr);
 962
 963	/*
 964	 * setup ELF PT_NOTE, place holder for cpu notes info. The notes info
 965	 * will be populated during second kernel boot after crash. Hence
 966	 * this PT_NOTE will always be the first elf note.
 967	 *
 968	 * NOTE: Any new ELF note addition should be placed after this note.
 969	 */
 970	phdr = (struct elf_phdr *)bufp;
 971	bufp += sizeof(struct elf_phdr);
 972	phdr->p_type = PT_NOTE;
 973	phdr->p_flags = 0;
 974	phdr->p_vaddr = 0;
 975	phdr->p_align = 0;
 976
 977	phdr->p_offset = 0;
 978	phdr->p_paddr = 0;
 979	phdr->p_filesz = 0;
 980	phdr->p_memsz = 0;
 981
 982	(elf->e_phnum)++;
 983
 984	/* setup ELF PT_NOTE for vmcoreinfo */
 985	phdr = (struct elf_phdr *)bufp;
 986	bufp += sizeof(struct elf_phdr);
 987	phdr->p_type	= PT_NOTE;
 988	phdr->p_flags	= 0;
 989	phdr->p_vaddr	= 0;
 990	phdr->p_align	= 0;
 991
 992	phdr->p_paddr	= fadump_relocate(paddr_vmcoreinfo_note());
 993	phdr->p_offset	= phdr->p_paddr;
 994	phdr->p_memsz	= phdr->p_filesz = VMCOREINFO_NOTE_SIZE;
 995
 996	/* Increment number of program headers. */
 997	(elf->e_phnum)++;
 998
 999	/* setup PT_LOAD sections. */
1000	j = 0;
1001	offset = 0;
1002	raddr = fw_dump.boot_mem_addr[0];
1003	for (i = 0; i < crash_mrange_info.mem_range_cnt; i++) {
1004		u64 mbase, msize;
1005
1006		mbase = crash_mrange_info.mem_ranges[i].base;
1007		msize = crash_mrange_info.mem_ranges[i].size;
1008		if (!msize)
1009			continue;
1010
1011		phdr = (struct elf_phdr *)bufp;
1012		bufp += sizeof(struct elf_phdr);
1013		phdr->p_type	= PT_LOAD;
1014		phdr->p_flags	= PF_R|PF_W|PF_X;
1015		phdr->p_offset	= mbase;
1016
1017		if (mbase == raddr) {
1018			/*
1019			 * The entire real memory region will be moved by
1020			 * firmware to the specified destination_address.
1021			 * Hence set the correct offset.
1022			 */
1023			phdr->p_offset = fw_dump.boot_mem_dest_addr + offset;
1024			if (j < (fw_dump.boot_mem_regs_cnt - 1)) {
1025				offset += fw_dump.boot_mem_sz[j];
1026				raddr = fw_dump.boot_mem_addr[++j];
1027			}
1028		}
1029
1030		phdr->p_paddr = mbase;
1031		phdr->p_vaddr = (unsigned long)__va(mbase);
1032		phdr->p_filesz = msize;
1033		phdr->p_memsz = msize;
1034		phdr->p_align = 0;
1035
1036		/* Increment number of program headers. */
1037		(elf->e_phnum)++;
1038	}
1039	return 0;
1040}
1041
1042static unsigned long init_fadump_header(unsigned long addr)
1043{
1044	struct fadump_crash_info_header *fdh;
1045
1046	if (!addr)
1047		return 0;
1048
1049	fdh = __va(addr);
1050	addr += sizeof(struct fadump_crash_info_header);
1051
1052	memset(fdh, 0, sizeof(struct fadump_crash_info_header));
1053	fdh->magic_number = FADUMP_CRASH_INFO_MAGIC;
1054	fdh->elfcorehdr_addr = addr;
1055	/* We will set the crashing cpu id in crash_fadump() during crash. */
1056	fdh->crashing_cpu = FADUMP_CPU_UNKNOWN;
1057
1058	return addr;
1059}
1060
1061static int register_fadump(void)
1062{
1063	unsigned long addr;
1064	void *vaddr;
1065	int ret;
1066
1067	/*
1068	 * If no memory is reserved then we can not register for firmware-
1069	 * assisted dump.
1070	 */
1071	if (!fw_dump.reserve_dump_area_size)
1072		return -ENODEV;
1073
1074	ret = fadump_setup_crash_memory_ranges();
1075	if (ret)
1076		return ret;
1077
1078	addr = fw_dump.fadumphdr_addr;
1079
1080	/* Initialize fadump crash info header. */
1081	addr = init_fadump_header(addr);
1082	vaddr = __va(addr);
1083
1084	pr_debug("Creating ELF core headers at %#016lx\n", addr);
1085	fadump_create_elfcore_headers(vaddr);
1086
1087	/* register the future kernel dump with firmware. */
1088	pr_debug("Registering for firmware-assisted kernel dump...\n");
1089	return fw_dump.ops->fadump_register(&fw_dump);
1090}
1091
1092void fadump_cleanup(void)
1093{
1094	if (!fw_dump.fadump_supported)
1095		return;
1096
1097	/* Invalidate the registration only if dump is active. */
1098	if (fw_dump.dump_active) {
1099		pr_debug("Invalidating firmware-assisted dump registration\n");
1100		fw_dump.ops->fadump_invalidate(&fw_dump);
1101	} else if (fw_dump.dump_registered) {
1102		/* Un-register Firmware-assisted dump if it was registered. */
1103		fw_dump.ops->fadump_unregister(&fw_dump);
1104		fadump_free_mem_ranges(&crash_mrange_info);
1105	}
1106
1107	if (fw_dump.ops->fadump_cleanup)
1108		fw_dump.ops->fadump_cleanup(&fw_dump);
1109}
1110
1111static void fadump_free_reserved_memory(unsigned long start_pfn,
1112					unsigned long end_pfn)
1113{
1114	unsigned long pfn;
1115	unsigned long time_limit = jiffies + HZ;
1116
1117	pr_info("freeing reserved memory (0x%llx - 0x%llx)\n",
1118		PFN_PHYS(start_pfn), PFN_PHYS(end_pfn));
1119
1120	for (pfn = start_pfn; pfn < end_pfn; pfn++) {
1121		free_reserved_page(pfn_to_page(pfn));
1122
1123		if (time_after(jiffies, time_limit)) {
1124			cond_resched();
1125			time_limit = jiffies + HZ;
1126		}
1127	}
1128}
1129
1130/*
1131 * Skip memory holes and free memory that was actually reserved.
1132 */
1133static void fadump_release_reserved_area(u64 start, u64 end)
1134{
1135	u64 tstart, tend, spfn, epfn;
1136	struct memblock_region *reg;
1137
1138	spfn = PHYS_PFN(start);
1139	epfn = PHYS_PFN(end);
1140	for_each_memblock(memory, reg) {
1141		tstart = max_t(u64, spfn, memblock_region_memory_base_pfn(reg));
1142		tend   = min_t(u64, epfn, memblock_region_memory_end_pfn(reg));
1143		if (tstart < tend) {
1144			fadump_free_reserved_memory(tstart, tend);
1145
1146			if (tend == epfn)
1147				break;
1148
1149			spfn = tend;
1150		}
1151	}
1152}
1153
1154/*
1155 * Sort the mem ranges in-place and merge adjacent ranges
1156 * to minimize the memory ranges count.
1157 */
1158static void sort_and_merge_mem_ranges(struct fadump_mrange_info *mrange_info)
1159{
1160	struct fadump_memory_range *mem_ranges;
1161	struct fadump_memory_range tmp_range;
1162	u64 base, size;
1163	int i, j, idx;
1164
1165	if (!reserved_mrange_info.mem_range_cnt)
1166		return;
1167
1168	/* Sort the memory ranges */
1169	mem_ranges = mrange_info->mem_ranges;
1170	for (i = 0; i < mrange_info->mem_range_cnt; i++) {
1171		idx = i;
1172		for (j = (i + 1); j < mrange_info->mem_range_cnt; j++) {
1173			if (mem_ranges[idx].base > mem_ranges[j].base)
1174				idx = j;
1175		}
1176		if (idx != i) {
1177			tmp_range = mem_ranges[idx];
1178			mem_ranges[idx] = mem_ranges[i];
1179			mem_ranges[i] = tmp_range;
1180		}
1181	}
1182
1183	/* Merge adjacent reserved ranges */
1184	idx = 0;
1185	for (i = 1; i < mrange_info->mem_range_cnt; i++) {
1186		base = mem_ranges[i-1].base;
1187		size = mem_ranges[i-1].size;
1188		if (mem_ranges[i].base == (base + size))
1189			mem_ranges[idx].size += mem_ranges[i].size;
1190		else {
1191			idx++;
1192			if (i == idx)
1193				continue;
1194
1195			mem_ranges[idx] = mem_ranges[i];
1196		}
1197	}
1198	mrange_info->mem_range_cnt = idx + 1;
1199}
1200
1201/*
1202 * Scan reserved-ranges to consider them while reserving/releasing
1203 * memory for FADump.
1204 */
1205static inline int fadump_scan_reserved_mem_ranges(void)
1206{
1207	struct device_node *root;
1208	const __be32 *prop;
1209	int len, ret = -1;
1210	unsigned long i;
1211
1212	root = of_find_node_by_path("/");
1213	if (!root)
1214		return ret;
1215
1216	prop = of_get_property(root, "reserved-ranges", &len);
1217	if (!prop)
1218		return ret;
1219
1220	/*
1221	 * Each reserved range is an (address,size) pair, 2 cells each,
1222	 * totalling 4 cells per range.
1223	 */
1224	for (i = 0; i < len / (sizeof(*prop) * 4); i++) {
1225		u64 base, size;
1226
1227		base = of_read_number(prop + (i * 4) + 0, 2);
1228		size = of_read_number(prop + (i * 4) + 2, 2);
1229
1230		if (size) {
1231			ret = fadump_add_mem_range(&reserved_mrange_info,
1232						   base, base + size);
1233			if (ret < 0) {
1234				pr_warn("some reserved ranges are ignored!\n");
1235				break;
1236			}
1237		}
1238	}
1239
1240	return ret;
 
1241}
1242
1243/*
1244 * Release the memory that was reserved during early boot to preserve the
1245 * crash'ed kernel's memory contents except reserved dump area (permanent
1246 * reservation) and reserved ranges used by F/W. The released memory will
1247 * be available for general use.
1248 */
1249static void fadump_release_memory(u64 begin, u64 end)
1250{
1251	u64 ra_start, ra_end, tstart;
1252	int i, ret;
1253
1254	fadump_scan_reserved_mem_ranges();
1255
1256	ra_start = fw_dump.reserve_dump_area_start;
1257	ra_end = ra_start + fw_dump.reserve_dump_area_size;
1258
1259	/*
1260	 * Add reserved dump area to reserved ranges list
1261	 * and exclude all these ranges while releasing memory.
 
1262	 */
 
 
 
 
1263	ret = fadump_add_mem_range(&reserved_mrange_info, ra_start, ra_end);
1264	if (ret != 0) {
1265		/*
1266		 * Not enough memory to setup reserved ranges but the system is
1267		 * running shortage of memory. So, release all the memory except
1268		 * Reserved dump area (reused for next fadump registration).
1269		 */
1270		if (begin < ra_end && end > ra_start) {
1271			if (begin < ra_start)
1272				fadump_release_reserved_area(begin, ra_start);
1273			if (end > ra_end)
1274				fadump_release_reserved_area(ra_end, end);
1275		} else
1276			fadump_release_reserved_area(begin, end);
1277
1278		return;
1279	}
1280
1281	/* Get the reserved ranges list in order first. */
1282	sort_and_merge_mem_ranges(&reserved_mrange_info);
1283
1284	/* Exclude reserved ranges and release remaining memory */
1285	tstart = begin;
1286	for (i = 0; i < reserved_mrange_info.mem_range_cnt; i++) {
1287		ra_start = reserved_mrange_info.mem_ranges[i].base;
1288		ra_end = ra_start + reserved_mrange_info.mem_ranges[i].size;
1289
1290		if (tstart >= ra_end)
1291			continue;
1292
1293		if (tstart < ra_start)
1294			fadump_release_reserved_area(tstart, ra_start);
1295		tstart = ra_end;
1296	}
1297
1298	if (tstart < end)
1299		fadump_release_reserved_area(tstart, end);
1300}
1301
1302static void fadump_invalidate_release_mem(void)
1303{
1304	mutex_lock(&fadump_mutex);
1305	if (!fw_dump.dump_active) {
1306		mutex_unlock(&fadump_mutex);
1307		return;
1308	}
1309
1310	fadump_cleanup();
1311	mutex_unlock(&fadump_mutex);
1312
1313	fadump_release_memory(fw_dump.boot_mem_top, memblock_end_of_DRAM());
1314	fadump_free_cpu_notes_buf();
1315
1316	/*
1317	 * Setup kernel metadata and initialize the kernel dump
1318	 * memory structure for FADump re-registration.
1319	 */
1320	if (fw_dump.ops->fadump_setup_metadata &&
1321	    (fw_dump.ops->fadump_setup_metadata(&fw_dump) < 0))
1322		pr_warn("Failed to setup kernel metadata!\n");
1323	fw_dump.ops->fadump_init_mem_struct(&fw_dump);
1324}
1325
1326static ssize_t fadump_release_memory_store(struct kobject *kobj,
1327					struct kobj_attribute *attr,
1328					const char *buf, size_t count)
1329{
1330	int input = -1;
1331
1332	if (!fw_dump.dump_active)
1333		return -EPERM;
1334
1335	if (kstrtoint(buf, 0, &input))
1336		return -EINVAL;
1337
1338	if (input == 1) {
1339		/*
1340		 * Take away the '/proc/vmcore'. We are releasing the dump
1341		 * memory, hence it will not be valid anymore.
1342		 */
1343#ifdef CONFIG_PROC_VMCORE
1344		vmcore_cleanup();
1345#endif
1346		fadump_invalidate_release_mem();
1347
1348	} else
1349		return -EINVAL;
1350	return count;
1351}
1352
1353static ssize_t fadump_enabled_show(struct kobject *kobj,
1354					struct kobj_attribute *attr,
1355					char *buf)
 
 
 
 
 
 
 
 
 
 
1356{
1357	return sprintf(buf, "%d\n", fw_dump.fadump_enabled);
1358}
1359
1360static ssize_t fadump_register_show(struct kobject *kobj,
1361					struct kobj_attribute *attr,
1362					char *buf)
 
 
 
 
 
 
 
1363{
1364	return sprintf(buf, "%d\n", fw_dump.dump_registered);
1365}
1366
1367static ssize_t fadump_register_store(struct kobject *kobj,
1368					struct kobj_attribute *attr,
1369					const char *buf, size_t count)
1370{
1371	int ret = 0;
1372	int input = -1;
1373
1374	if (!fw_dump.fadump_enabled || fw_dump.dump_active)
1375		return -EPERM;
1376
1377	if (kstrtoint(buf, 0, &input))
1378		return -EINVAL;
1379
1380	mutex_lock(&fadump_mutex);
1381
1382	switch (input) {
1383	case 0:
1384		if (fw_dump.dump_registered == 0) {
1385			goto unlock_out;
1386		}
1387
1388		/* Un-register Firmware-assisted dump */
1389		pr_debug("Un-register firmware-assisted dump\n");
1390		fw_dump.ops->fadump_unregister(&fw_dump);
1391		break;
1392	case 1:
1393		if (fw_dump.dump_registered == 1) {
1394			/* Un-register Firmware-assisted dump */
1395			fw_dump.ops->fadump_unregister(&fw_dump);
1396		}
1397		/* Register Firmware-assisted dump */
1398		ret = register_fadump();
1399		break;
1400	default:
1401		ret = -EINVAL;
1402		break;
1403	}
1404
1405unlock_out:
1406	mutex_unlock(&fadump_mutex);
1407	return ret < 0 ? ret : count;
1408}
1409
1410static int fadump_region_show(struct seq_file *m, void *private)
1411{
1412	if (!fw_dump.fadump_enabled)
1413		return 0;
1414
1415	mutex_lock(&fadump_mutex);
1416	fw_dump.ops->fadump_region_show(&fw_dump, m);
1417	mutex_unlock(&fadump_mutex);
1418	return 0;
1419}
1420
1421static struct kobj_attribute fadump_release_attr = __ATTR(fadump_release_mem,
1422						0200, NULL,
1423						fadump_release_memory_store);
1424static struct kobj_attribute fadump_attr = __ATTR(fadump_enabled,
1425						0444, fadump_enabled_show,
1426						NULL);
1427static struct kobj_attribute fadump_register_attr = __ATTR(fadump_registered,
1428						0644, fadump_register_show,
1429						fadump_register_store);
 
 
 
 
1430
1431DEFINE_SHOW_ATTRIBUTE(fadump_region);
1432
1433static void fadump_init_files(void)
1434{
1435	struct dentry *debugfs_file;
1436	int rc = 0;
1437
1438	rc = sysfs_create_file(kernel_kobj, &fadump_attr.attr);
1439	if (rc)
1440		printk(KERN_ERR "fadump: unable to create sysfs file"
1441			" fadump_enabled (%d)\n", rc);
1442
1443	rc = sysfs_create_file(kernel_kobj, &fadump_register_attr.attr);
1444	if (rc)
1445		printk(KERN_ERR "fadump: unable to create sysfs file"
1446			" fadump_registered (%d)\n", rc);
1447
1448	debugfs_file = debugfs_create_file("fadump_region", 0444,
1449					powerpc_debugfs_root, NULL,
1450					&fadump_region_fops);
1451	if (!debugfs_file)
1452		printk(KERN_ERR "fadump: unable to create debugfs file"
1453				" fadump_region\n");
1454
1455	if (fw_dump.dump_active) {
1456		rc = sysfs_create_file(kernel_kobj, &fadump_release_attr.attr);
1457		if (rc)
1458			printk(KERN_ERR "fadump: unable to create sysfs file"
1459				" fadump_release_mem (%d)\n", rc);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1460	}
1461	return;
1462}
1463
1464/*
1465 * Prepare for firmware-assisted dump.
1466 */
1467int __init setup_fadump(void)
1468{
1469	if (!fw_dump.fadump_enabled)
1470		return 0;
1471
1472	if (!fw_dump.fadump_supported) {
1473		printk(KERN_ERR "Firmware-assisted dump is not supported on"
1474			" this hardware\n");
1475		return 0;
1476	}
1477
1478	fadump_show_config();
1479	/*
1480	 * If dump data is available then see if it is valid and prepare for
1481	 * saving it to the disk.
1482	 */
1483	if (fw_dump.dump_active) {
1484		/*
1485		 * if dump process fails then invalidate the registration
1486		 * and release memory before proceeding for re-registration.
1487		 */
1488		if (fw_dump.ops->fadump_process(&fw_dump) < 0)
1489			fadump_invalidate_release_mem();
1490	}
1491	/* Initialize the kernel dump memory structure for FAD registration. */
1492	else if (fw_dump.reserve_dump_area_size)
1493		fw_dump.ops->fadump_init_mem_struct(&fw_dump);
1494
1495	fadump_init_files();
1496
1497	return 1;
1498}
1499subsys_initcall(setup_fadump);
1500#else /* !CONFIG_PRESERVE_FA_DUMP */
1501
1502/* Scan the Firmware Assisted dump configuration details. */
1503int __init early_init_dt_scan_fw_dump(unsigned long node, const char *uname,
1504				      int depth, void *data)
1505{
1506	if ((depth != 1) || (strcmp(uname, "ibm,opal") != 0))
1507		return 0;
1508
1509	opal_fadump_dt_scan(&fw_dump, node);
1510	return 1;
1511}
1512
1513/*
1514 * When dump is active but PRESERVE_FA_DUMP is enabled on the kernel,
1515 * preserve crash data. The subsequent memory preserving kernel boot
1516 * is likely to process this crash data.
1517 */
1518int __init fadump_reserve_mem(void)
1519{
1520	if (fw_dump.dump_active) {
1521		/*
1522		 * If last boot has crashed then reserve all the memory
1523		 * above boot memory to preserve crash data.
1524		 */
1525		pr_info("Preserving crash data for processing in next boot.\n");
1526		fadump_reserve_crash_area(fw_dump.boot_mem_top);
1527	} else
1528		pr_debug("FADump-aware kernel..\n");
1529
1530	return 1;
1531}
1532#endif /* CONFIG_PRESERVE_FA_DUMP */
1533
1534/* Preserve everything above the base address */
1535static void __init fadump_reserve_crash_area(u64 base)
1536{
1537	struct memblock_region *reg;
1538	u64 mstart, msize;
1539
1540	for_each_memblock(memory, reg) {
1541		mstart = reg->base;
1542		msize  = reg->size;
1543
1544		if ((mstart + msize) < base)
1545			continue;
1546
1547		if (mstart < base) {
1548			msize -= (base - mstart);
1549			mstart = base;
1550		}
1551
1552		pr_info("Reserving %lluMB of memory at %#016llx for preserving crash data",
1553			(msize >> 20), mstart);
1554		memblock_reserve(mstart, msize);
1555	}
1556}
1557
1558unsigned long __init arch_reserved_kernel_pages(void)
1559{
1560	return memblock_reserved_size() / PAGE_SIZE;
1561}