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}
v3.5.6
 
   1/*
   2 * Firmware Assisted dump: A robust mechanism to get reliable kernel crash
   3 * dump with assistance from firmware. This approach does not use kexec,
   4 * instead firmware assists in booting the kdump kernel while preserving
   5 * memory contents. The most of the code implementation has been adapted
   6 * from phyp assisted dump implementation written by Linas Vepstas and
   7 * Manish Ahuja
   8 *
   9 * This program is free software; you can redistribute it and/or modify
  10 * it under the terms of the GNU General Public License as published by
  11 * the Free Software Foundation; either version 2 of the License, or
  12 * (at your option) any later version.
  13 *
  14 * This program is distributed in the hope that it will be useful,
  15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  17 * GNU General Public License for more details.
  18 *
  19 * You should have received a copy of the GNU General Public License
  20 * along with this program; if not, write to the Free Software
  21 * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
  22 *
  23 * Copyright 2011 IBM Corporation
  24 * Author: Mahesh Salgaonkar <mahesh@linux.vnet.ibm.com>
  25 */
  26
  27#undef DEBUG
  28#define pr_fmt(fmt) "fadump: " fmt
  29
  30#include <linux/string.h>
  31#include <linux/memblock.h>
  32#include <linux/delay.h>
  33#include <linux/debugfs.h>
  34#include <linux/seq_file.h>
  35#include <linux/crash_dump.h>
  36#include <linux/kobject.h>
  37#include <linux/sysfs.h>
 
 
 
  38
 
  39#include <asm/page.h>
  40#include <asm/prom.h>
  41#include <asm/rtas.h>
  42#include <asm/fadump.h>
  43#include <asm/debug.h>
  44#include <asm/setup.h>
  45
 
 
 
 
 
 
 
 
  46static struct fw_dump fw_dump;
  47static struct fadump_mem_struct fdm;
  48static const struct fadump_mem_struct *fdm_active;
  49
 
 
 
 
 
 
 
  50static DEFINE_MUTEX(fadump_mutex);
  51struct fad_crash_memory_ranges crash_memory_ranges[INIT_CRASHMEM_RANGES];
  52int crash_mem_ranges;
  53
  54/* Scan the Firmware Assisted dump configuration details. */
  55int __init early_init_dt_scan_fw_dump(unsigned long node,
  56			const char *uname, int depth, void *data)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  57{
  58	__be32 *sections;
  59	int i, num_sections;
  60	unsigned long size;
  61	const int *token;
  62
  63	if (depth != 1 || strcmp(uname, "rtas") != 0)
  64		return 0;
  65
  66	/*
  67	 * Check if Firmware Assisted dump is supported. if yes, check
  68	 * if dump has been initiated on last reboot.
  69	 */
  70	token = of_get_flat_dt_prop(node, "ibm,configure-kernel-dump", NULL);
  71	if (!token)
 
 
 
 
 
  72		return 0;
  73
  74	fw_dump.fadump_supported = 1;
  75	fw_dump.ibm_configure_kernel_dump = *token;
 
 
 
 
 
 
 
 
 
  76
  77	/*
  78	 * The 'ibm,kernel-dump' rtas node is present only if there is
  79	 * dump data waiting for us.
  80	 */
  81	fdm_active = of_get_flat_dt_prop(node, "ibm,kernel-dump", NULL);
  82	if (fdm_active)
  83		fw_dump.dump_active = 1;
  84
  85	/* Get the sizes required to store dump data for the firmware provided
  86	 * dump sections.
  87	 * For each dump section type supported, a 32bit cell which defines
  88	 * the ID of a supported section followed by two 32 bit cells which
  89	 * gives teh size of the section in bytes.
  90	 */
  91	sections = of_get_flat_dt_prop(node, "ibm,configure-kernel-dump-sizes",
  92					&size);
  93
  94	if (!sections)
 
 
 
 
 
  95		return 0;
 
  96
  97	num_sections = size / (3 * sizeof(u32));
 
  98
  99	for (i = 0; i < num_sections; i++, sections += 3) {
 100		u32 type = (u32)of_read_number(sections, 1);
 
 
 101
 102		switch (type) {
 103		case FADUMP_CPU_STATE_DATA:
 104			fw_dump.cpu_state_data_size =
 105					of_read_ulong(&sections[1], 2);
 106			break;
 107		case FADUMP_HPTE_REGION:
 108			fw_dump.hpte_region_size =
 109					of_read_ulong(&sections[1], 2);
 110			break;
 111		}
 112	}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 113	return 1;
 114}
 115
 116int is_fadump_active(void)
 117{
 118	return fw_dump.dump_active;
 119}
 120
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 121/* Print firmware assisted dump configurations for debugging purpose. */
 122static void fadump_show_config(void)
 123{
 
 
 124	pr_debug("Support for firmware-assisted dump (fadump): %s\n",
 125			(fw_dump.fadump_supported ? "present" : "no support"));
 126
 127	if (!fw_dump.fadump_supported)
 128		return;
 129
 130	pr_debug("Fadump enabled    : %s\n",
 131				(fw_dump.fadump_enabled ? "yes" : "no"));
 132	pr_debug("Dump Active       : %s\n",
 133				(fw_dump.dump_active ? "yes" : "no"));
 134	pr_debug("Dump section sizes:\n");
 135	pr_debug("    CPU state data size: %lx\n", fw_dump.cpu_state_data_size);
 136	pr_debug("    HPTE region size   : %lx\n", fw_dump.hpte_region_size);
 137	pr_debug("Boot memory size  : %lx\n", fw_dump.boot_memory_size);
 138}
 139
 140static unsigned long init_fadump_mem_struct(struct fadump_mem_struct *fdm,
 141				unsigned long addr)
 142{
 143	if (!fdm)
 144		return 0;
 145
 146	memset(fdm, 0, sizeof(struct fadump_mem_struct));
 147	addr = addr & PAGE_MASK;
 148
 149	fdm->header.dump_format_version = 0x00000001;
 150	fdm->header.dump_num_sections = 3;
 151	fdm->header.dump_status_flag = 0;
 152	fdm->header.offset_first_dump_section =
 153		(u32)offsetof(struct fadump_mem_struct, cpu_state_data);
 154
 155	/*
 156	 * Fields for disk dump option.
 157	 * We are not using disk dump option, hence set these fields to 0.
 158	 */
 159	fdm->header.dd_block_size = 0;
 160	fdm->header.dd_block_offset = 0;
 161	fdm->header.dd_num_blocks = 0;
 162	fdm->header.dd_offset_disk_path = 0;
 163
 164	/* set 0 to disable an automatic dump-reboot. */
 165	fdm->header.max_time_auto = 0;
 166
 167	/* Kernel dump sections */
 168	/* cpu state data section. */
 169	fdm->cpu_state_data.request_flag = FADUMP_REQUEST_FLAG;
 170	fdm->cpu_state_data.source_data_type = FADUMP_CPU_STATE_DATA;
 171	fdm->cpu_state_data.source_address = 0;
 172	fdm->cpu_state_data.source_len = fw_dump.cpu_state_data_size;
 173	fdm->cpu_state_data.destination_address = addr;
 174	addr += fw_dump.cpu_state_data_size;
 175
 176	/* hpte region section */
 177	fdm->hpte_region.request_flag = FADUMP_REQUEST_FLAG;
 178	fdm->hpte_region.source_data_type = FADUMP_HPTE_REGION;
 179	fdm->hpte_region.source_address = 0;
 180	fdm->hpte_region.source_len = fw_dump.hpte_region_size;
 181	fdm->hpte_region.destination_address = addr;
 182	addr += fw_dump.hpte_region_size;
 183
 184	/* RMA region section */
 185	fdm->rmr_region.request_flag = FADUMP_REQUEST_FLAG;
 186	fdm->rmr_region.source_data_type = FADUMP_REAL_MODE_REGION;
 187	fdm->rmr_region.source_address = RMA_START;
 188	fdm->rmr_region.source_len = fw_dump.boot_memory_size;
 189	fdm->rmr_region.destination_address = addr;
 190	addr += fw_dump.boot_memory_size;
 191
 192	return addr;
 193}
 194
 195/**
 196 * fadump_calculate_reserve_size(): reserve variable boot area 5% of System RAM
 197 *
 198 * Function to find the largest memory size we need to reserve during early
 199 * boot process. This will be the size of the memory that is required for a
 200 * kernel to boot successfully.
 201 *
 202 * This function has been taken from phyp-assisted dump feature implementation.
 203 *
 204 * returns larger of 256MB or 5% rounded down to multiples of 256MB.
 205 *
 206 * TODO: Come up with better approach to find out more accurate memory size
 207 * that is required for a kernel to boot successfully.
 208 *
 209 */
 210static inline unsigned long fadump_calculate_reserve_size(void)
 211{
 212	unsigned long size;
 
 
 
 
 213
 214	/*
 215	 * Check if the size is specified through fadump_reserve_mem= cmdline
 216	 * option. If yes, then use that.
 
 217	 */
 218	if (fw_dump.reserve_bootvar)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 219		return fw_dump.reserve_bootvar;
 
 220
 221	/* divide by 20 to get 5% of value */
 222	size = memblock_end_of_DRAM() / 20;
 223
 224	/* round it down in multiples of 256 */
 225	size = size & ~0x0FFFFFFFUL;
 226
 227	/* Truncate to memory_limit. We don't want to over reserve the memory.*/
 228	if (memory_limit && size > memory_limit)
 229		size = memory_limit;
 230
 231	return (size > MIN_BOOT_MEM ? size : MIN_BOOT_MEM);
 
 232}
 233
 234/*
 235 * Calculate the total memory size required to be reserved for
 236 * firmware-assisted dump registration.
 237 */
 238static unsigned long get_fadump_area_size(void)
 239{
 240	unsigned long size = 0;
 241
 242	size += fw_dump.cpu_state_data_size;
 243	size += fw_dump.hpte_region_size;
 244	size += fw_dump.boot_memory_size;
 245	size += sizeof(struct fadump_crash_info_header);
 246	size += sizeof(struct elfhdr); /* ELF core header.*/
 247	size += sizeof(struct elf_phdr); /* place holder for cpu notes */
 248	/* Program headers for crash memory regions. */
 249	size += sizeof(struct elf_phdr) * (memblock_num_regions(memory) + 2);
 250
 251	size = PAGE_ALIGN(size);
 
 
 
 
 252	return size;
 253}
 254
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 255int __init fadump_reserve_mem(void)
 256{
 257	unsigned long base, size, memory_boundary;
 
 258
 259	if (!fw_dump.fadump_enabled)
 260		return 0;
 261
 262	if (!fw_dump.fadump_supported) {
 263		printk(KERN_INFO "Firmware-assisted dump is not supported on"
 264				" this hardware\n");
 265		fw_dump.fadump_enabled = 0;
 266		return 0;
 267	}
 
 268	/*
 269	 * Initialize boot memory size
 270	 * If dump is active then we have already calculated the size during
 271	 * first kernel.
 272	 */
 273	if (fdm_active)
 274		fw_dump.boot_memory_size = fdm_active->rmr_region.source_len;
 275	else
 276		fw_dump.boot_memory_size = fadump_calculate_reserve_size();
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 277
 278	/*
 279	 * Calculate the memory boundary.
 280	 * If memory_limit is less than actual memory boundary then reserve
 281	 * the memory for fadump beyond the memory_limit and adjust the
 282	 * memory_limit accordingly, so that the running kernel can run with
 283	 * specified memory_limit.
 284	 */
 285	if (memory_limit && memory_limit < memblock_end_of_DRAM()) {
 286		size = get_fadump_area_size();
 287		if ((memory_limit + size) < memblock_end_of_DRAM())
 288			memory_limit += size;
 289		else
 290			memory_limit = memblock_end_of_DRAM();
 291		printk(KERN_INFO "Adjusted memory_limit for firmware-assisted"
 292				" dump, now %#016llx\n",
 293				(unsigned long long)memory_limit);
 294	}
 295	if (memory_limit)
 296		memory_boundary = memory_limit;
 297	else
 298		memory_boundary = memblock_end_of_DRAM();
 299
 
 
 
 300	if (fw_dump.dump_active) {
 301		printk(KERN_INFO "Firmware-assisted dump is active.\n");
 
 
 
 
 
 
 
 
 
 302		/*
 303		 * If last boot has crashed then reserve all the memory
 304		 * above boot_memory_size so that we don't touch it until
 305		 * dump is written to disk by userspace tool. This memory
 306		 * will be released for general use once the dump is saved.
 307		 */
 308		base = fw_dump.boot_memory_size;
 309		size = memory_boundary - base;
 310		memblock_reserve(base, size);
 311		printk(KERN_INFO "Reserved %ldMB of memory at %ldMB "
 312				"for saving crash dump\n",
 313				(unsigned long)(size >> 20),
 314				(unsigned long)(base >> 20));
 315
 316		fw_dump.fadumphdr_addr =
 317				fdm_active->rmr_region.destination_address +
 318				fdm_active->rmr_region.source_len;
 319		pr_debug("fadumphdr_addr = %p\n",
 320				(void *) fw_dump.fadumphdr_addr);
 321	} else {
 322		/* Reserve the memory at the top of memory. */
 323		size = get_fadump_area_size();
 324		base = memory_boundary - size;
 325		memblock_reserve(base, size);
 326		printk(KERN_INFO "Reserved %ldMB of memory at %ldMB "
 327				"for firmware-assisted dump\n",
 328				(unsigned long)(size >> 20),
 329				(unsigned long)(base >> 20));
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 330	}
 331	fw_dump.reserve_dump_area_start = base;
 332	fw_dump.reserve_dump_area_size = size;
 333	return 1;
 
 
 334}
 335
 336/* Look for fadump= cmdline option. */
 337static int __init early_fadump_param(char *p)
 338{
 339	if (!p)
 340		return 1;
 341
 342	if (strncmp(p, "on", 2) == 0)
 343		fw_dump.fadump_enabled = 1;
 344	else if (strncmp(p, "off", 3) == 0)
 345		fw_dump.fadump_enabled = 0;
 
 
 
 
 346
 347	return 0;
 348}
 349early_param("fadump", early_fadump_param);
 350
 351/* Look for fadump_reserve_mem= cmdline option */
 
 
 
 
 352static int __init early_fadump_reserve_mem(char *p)
 353{
 354	if (p)
 355		fw_dump.reserve_bootvar = memparse(p, &p);
 356	return 0;
 357}
 358early_param("fadump_reserve_mem", early_fadump_reserve_mem);
 359
 360static void register_fw_dump(struct fadump_mem_struct *fdm)
 361{
 362	int rc;
 363	unsigned int wait_time;
 
 
 
 364
 365	pr_debug("Registering for firmware-assisted kernel dump...\n");
 
 366
 367	/* TODO: Add upper time limit for the delay */
 368	do {
 369		rc = rtas_call(fw_dump.ibm_configure_kernel_dump, 3, 1, NULL,
 370			FADUMP_REGISTER, fdm,
 371			sizeof(struct fadump_mem_struct));
 372
 373		wait_time = rtas_busy_delay_time(rc);
 374		if (wait_time)
 375			mdelay(wait_time);
 376
 377	} while (wait_time);
 378
 379	switch (rc) {
 380	case -1:
 381		printk(KERN_ERR "Failed to register firmware-assisted kernel"
 382			" dump. Hardware Error(%d).\n", rc);
 383		break;
 384	case -3:
 385		printk(KERN_ERR "Failed to register firmware-assisted kernel"
 386			" dump. Parameter Error(%d).\n", rc);
 387		break;
 388	case -9:
 389		printk(KERN_ERR "firmware-assisted kernel dump is already "
 390			" registered.");
 391		fw_dump.dump_registered = 1;
 392		break;
 393	case 0:
 394		printk(KERN_INFO "firmware-assisted kernel dump registration"
 395			" is successful\n");
 396		fw_dump.dump_registered = 1;
 397		break;
 398	}
 399}
 400
 401void crash_fadump(struct pt_regs *regs, const char *str)
 402{
 403	struct fadump_crash_info_header *fdh = NULL;
 404
 405	if (!fw_dump.dump_registered || !fw_dump.fadumphdr_addr)
 
 
 
 
 
 
 
 
 406		return;
 
 407
 408	fdh = __va(fw_dump.fadumphdr_addr);
 409	crashing_cpu = smp_processor_id();
 410	fdh->crashing_cpu = crashing_cpu;
 411	crash_save_vmcoreinfo();
 412
 413	if (regs)
 414		fdh->regs = *regs;
 415	else
 416		ppc_save_regs(&fdh->regs);
 417
 418	fdh->cpu_online_mask = *cpu_online_mask;
 419
 420	/* Call ibm,os-term rtas call to trigger firmware assisted dump */
 421	rtas_os_term((char *)str);
 422}
 423
 424#define GPR_MASK	0xffffff0000000000
 425static inline int fadump_gpr_index(u64 id)
 426{
 427	int i = -1;
 428	char str[3];
 429
 430	if ((id & GPR_MASK) == REG_ID("GPR")) {
 431		/* get the digits at the end */
 432		id &= ~GPR_MASK;
 433		id >>= 24;
 434		str[2] = '\0';
 435		str[1] = id & 0xff;
 436		str[0] = (id >> 8) & 0xff;
 437		sscanf(str, "%d", &i);
 438		if (i > 31)
 439			i = -1;
 440	}
 441	return i;
 442}
 443
 444static inline void fadump_set_regval(struct pt_regs *regs, u64 reg_id,
 445								u64 reg_val)
 446{
 447	int i;
 448
 449	i = fadump_gpr_index(reg_id);
 450	if (i >= 0)
 451		regs->gpr[i] = (unsigned long)reg_val;
 452	else if (reg_id == REG_ID("NIA"))
 453		regs->nip = (unsigned long)reg_val;
 454	else if (reg_id == REG_ID("MSR"))
 455		regs->msr = (unsigned long)reg_val;
 456	else if (reg_id == REG_ID("CTR"))
 457		regs->ctr = (unsigned long)reg_val;
 458	else if (reg_id == REG_ID("LR"))
 459		regs->link = (unsigned long)reg_val;
 460	else if (reg_id == REG_ID("XER"))
 461		regs->xer = (unsigned long)reg_val;
 462	else if (reg_id == REG_ID("CR"))
 463		regs->ccr = (unsigned long)reg_val;
 464	else if (reg_id == REG_ID("DAR"))
 465		regs->dar = (unsigned long)reg_val;
 466	else if (reg_id == REG_ID("DSISR"))
 467		regs->dsisr = (unsigned long)reg_val;
 468}
 469
 470static struct fadump_reg_entry*
 471fadump_read_registers(struct fadump_reg_entry *reg_entry, struct pt_regs *regs)
 472{
 473	memset(regs, 0, sizeof(struct pt_regs));
 474
 475	while (reg_entry->reg_id != REG_ID("CPUEND")) {
 476		fadump_set_regval(regs, reg_entry->reg_id,
 477					reg_entry->reg_value);
 478		reg_entry++;
 479	}
 480	reg_entry++;
 481	return reg_entry;
 482}
 483
 484static u32 *fadump_append_elf_note(u32 *buf, char *name, unsigned type,
 485						void *data, size_t data_len)
 486{
 487	struct elf_note note;
 488
 489	note.n_namesz = strlen(name) + 1;
 490	note.n_descsz = data_len;
 491	note.n_type   = type;
 492	memcpy(buf, &note, sizeof(note));
 493	buf += (sizeof(note) + 3)/4;
 494	memcpy(buf, name, note.n_namesz);
 495	buf += (note.n_namesz + 3)/4;
 496	memcpy(buf, data, note.n_descsz);
 497	buf += (note.n_descsz + 3)/4;
 498
 499	return buf;
 500}
 501
 502static void fadump_final_note(u32 *buf)
 503{
 504	struct elf_note note;
 505
 506	note.n_namesz = 0;
 507	note.n_descsz = 0;
 508	note.n_type   = 0;
 509	memcpy(buf, &note, sizeof(note));
 510}
 511
 512static u32 *fadump_regs_to_elf_notes(u32 *buf, struct pt_regs *regs)
 513{
 514	struct elf_prstatus prstatus;
 515
 516	memset(&prstatus, 0, sizeof(prstatus));
 517	/*
 518	 * FIXME: How do i get PID? Do I really need it?
 519	 * prstatus.pr_pid = ????
 520	 */
 521	elf_core_copy_kernel_regs(&prstatus.pr_reg, regs);
 522	buf = fadump_append_elf_note(buf, KEXEC_CORE_NOTE_NAME, NT_PRSTATUS,
 523				&prstatus, sizeof(prstatus));
 524	return buf;
 525}
 526
 527static void fadump_update_elfcore_header(char *bufp)
 528{
 529	struct elfhdr *elf;
 530	struct elf_phdr *phdr;
 531
 532	elf = (struct elfhdr *)bufp;
 533	bufp += sizeof(struct elfhdr);
 534
 535	/* First note is a place holder for cpu notes info. */
 536	phdr = (struct elf_phdr *)bufp;
 537
 538	if (phdr->p_type == PT_NOTE) {
 539		phdr->p_paddr = fw_dump.cpu_notes_buf;
 540		phdr->p_offset	= phdr->p_paddr;
 541		phdr->p_filesz	= fw_dump.cpu_notes_buf_size;
 542		phdr->p_memsz = fw_dump.cpu_notes_buf_size;
 543	}
 544	return;
 545}
 546
 547static void *fadump_cpu_notes_buf_alloc(unsigned long size)
 548{
 
 
 549	void *vaddr;
 550	struct page *page;
 551	unsigned long order, count, i;
 552
 553	order = get_order(size);
 554	vaddr = (void *)__get_free_pages(GFP_KERNEL|__GFP_ZERO, order);
 555	if (!vaddr)
 556		return NULL;
 557
 558	count = 1 << order;
 559	page = virt_to_page(vaddr);
 560	for (i = 0; i < count; i++)
 561		SetPageReserved(page + i);
 562	return vaddr;
 563}
 564
 565static void fadump_cpu_notes_buf_free(unsigned long vaddr, unsigned long size)
 566{
 567	struct page *page;
 568	unsigned long order, count, i;
 569
 570	order = get_order(size);
 571	count = 1 << order;
 572	page = virt_to_page(vaddr);
 573	for (i = 0; i < count; i++)
 574		ClearPageReserved(page + i);
 575	__free_pages(page, order);
 576}
 577
 578/*
 579 * Read CPU state dump data and convert it into ELF notes.
 580 * The CPU dump starts with magic number "REGSAVE". NumCpusOffset should be
 581 * used to access the data to allow for additional fields to be added without
 582 * affecting compatibility. Each list of registers for a CPU starts with
 583 * "CPUSTRT" and ends with "CPUEND". Each register entry is of 16 bytes,
 584 * 8 Byte ASCII identifier and 8 Byte register value. The register entry
 585 * with identifier "CPUSTRT" and "CPUEND" contains 4 byte cpu id as part
 586 * of register value. For more details refer to PAPR document.
 587 *
 588 * Only for the crashing cpu we ignore the CPU dump data and get exact
 589 * state from fadump crash info structure populated by first kernel at the
 590 * time of crash.
 591 */
 592static int __init fadump_build_cpu_notes(const struct fadump_mem_struct *fdm)
 593{
 594	struct fadump_reg_save_area_header *reg_header;
 595	struct fadump_reg_entry *reg_entry;
 596	struct fadump_crash_info_header *fdh = NULL;
 597	void *vaddr;
 598	unsigned long addr;
 599	u32 num_cpus, *note_buf;
 600	struct pt_regs regs;
 601	int i, rc = 0, cpu = 0;
 602
 603	if (!fdm->cpu_state_data.bytes_dumped)
 604		return -EINVAL;
 605
 606	addr = fdm->cpu_state_data.destination_address;
 607	vaddr = __va(addr);
 608
 609	reg_header = vaddr;
 610	if (reg_header->magic_number != REGSAVE_AREA_MAGIC) {
 611		printk(KERN_ERR "Unable to read register save area.\n");
 612		return -ENOENT;
 613	}
 614	pr_debug("--------CPU State Data------------\n");
 615	pr_debug("Magic Number: %llx\n", reg_header->magic_number);
 616	pr_debug("NumCpuOffset: %x\n", reg_header->num_cpu_offset);
 617
 618	vaddr += reg_header->num_cpu_offset;
 619	num_cpus = *((u32 *)(vaddr));
 620	pr_debug("NumCpus     : %u\n", num_cpus);
 621	vaddr += sizeof(u32);
 622	reg_entry = (struct fadump_reg_entry *)vaddr;
 623
 624	/* Allocate buffer to hold cpu crash notes. */
 625	fw_dump.cpu_notes_buf_size = num_cpus * sizeof(note_buf_t);
 626	fw_dump.cpu_notes_buf_size = PAGE_ALIGN(fw_dump.cpu_notes_buf_size);
 627	note_buf = fadump_cpu_notes_buf_alloc(fw_dump.cpu_notes_buf_size);
 628	if (!note_buf) {
 629		printk(KERN_ERR "Failed to allocate 0x%lx bytes for "
 630			"cpu notes buffer\n", fw_dump.cpu_notes_buf_size);
 
 631		return -ENOMEM;
 632	}
 633	fw_dump.cpu_notes_buf = __pa(note_buf);
 634
 635	pr_debug("Allocated buffer for cpu notes of size %ld at %p\n",
 636			(num_cpus * sizeof(note_buf_t)), note_buf);
 
 
 
 637
 638	if (fw_dump.fadumphdr_addr)
 639		fdh = __va(fw_dump.fadumphdr_addr);
 
 
 640
 641	for (i = 0; i < num_cpus; i++) {
 642		if (reg_entry->reg_id != REG_ID("CPUSTRT")) {
 643			printk(KERN_ERR "Unable to read CPU state data\n");
 644			rc = -ENOENT;
 645			goto error_out;
 646		}
 647		/* Lower 4 bytes of reg_value contains logical cpu id */
 648		cpu = reg_entry->reg_value & FADUMP_CPU_ID_MASK;
 649		if (!cpumask_test_cpu(cpu, &fdh->cpu_online_mask)) {
 650			SKIP_TO_NEXT_CPU(reg_entry);
 651			continue;
 652		}
 653		pr_debug("Reading register data for cpu %d...\n", cpu);
 654		if (fdh && fdh->crashing_cpu == cpu) {
 655			regs = fdh->regs;
 656			note_buf = fadump_regs_to_elf_notes(note_buf, &regs);
 657			SKIP_TO_NEXT_CPU(reg_entry);
 658		} else {
 659			reg_entry++;
 660			reg_entry = fadump_read_registers(reg_entry, &regs);
 661			note_buf = fadump_regs_to_elf_notes(note_buf, &regs);
 662		}
 663	}
 664	fadump_final_note(note_buf);
 665
 666	pr_debug("Updating elfcore header (%llx) with cpu notes\n",
 667							fdh->elfcorehdr_addr);
 668	fadump_update_elfcore_header((char *)__va(fdh->elfcorehdr_addr));
 669	return 0;
 670
 671error_out:
 672	fadump_cpu_notes_buf_free((unsigned long)__va(fw_dump.cpu_notes_buf),
 673					fw_dump.cpu_notes_buf_size);
 674	fw_dump.cpu_notes_buf = 0;
 675	fw_dump.cpu_notes_buf_size = 0;
 676	return rc;
 677
 
 
 
 678}
 679
 680/*
 681 * Validate and process the dump data stored by firmware before exporting
 682 * it through '/proc/vmcore'.
 683 */
 684static int __init process_fadump(const struct fadump_mem_struct *fdm_active)
 685{
 686	struct fadump_crash_info_header *fdh;
 687	int rc = 0;
 688
 689	if (!fdm_active || !fw_dump.fadumphdr_addr)
 690		return -EINVAL;
 
 
 
 
 
 
 
 
 
 691
 692	/* Check if the dump data is valid. */
 693	if ((fdm_active->header.dump_status_flag == FADUMP_ERROR_FLAG) ||
 694			(fdm_active->cpu_state_data.error_flags != 0) ||
 695			(fdm_active->rmr_region.error_flags != 0)) {
 696		printk(KERN_ERR "Dump taken by platform is not valid\n");
 697		return -EINVAL;
 698	}
 699	if ((fdm_active->rmr_region.bytes_dumped !=
 700			fdm_active->rmr_region.source_len) ||
 701			!fdm_active->cpu_state_data.bytes_dumped) {
 702		printk(KERN_ERR "Dump taken by platform is incomplete\n");
 703		return -EINVAL;
 704	}
 705
 706	/* Validate the fadump crash info header */
 707	fdh = __va(fw_dump.fadumphdr_addr);
 708	if (fdh->magic_number != FADUMP_CRASH_INFO_MAGIC) {
 709		printk(KERN_ERR "Crash info header is not valid.\n");
 710		return -EINVAL;
 711	}
 712
 713	rc = fadump_build_cpu_notes(fdm_active);
 714	if (rc)
 715		return rc;
 716
 717	/*
 718	 * We are done validating dump info and elfcore header is now ready
 719	 * to be exported. set elfcorehdr_addr so that vmcore module will
 720	 * export the elfcore header through '/proc/vmcore'.
 721	 */
 722	elfcorehdr_addr = fdh->elfcorehdr_addr;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 723
 724	return 0;
 725}
 
 726
 727static inline void fadump_add_crash_memory(unsigned long long base,
 728					unsigned long long end)
 729{
 730	if (base == end)
 731		return;
 732
 733	pr_debug("crash_memory_range[%d] [%#016llx-%#016llx], %#llx bytes\n",
 734		crash_mem_ranges, base, end - 1, (end - base));
 735	crash_memory_ranges[crash_mem_ranges].base = base;
 736	crash_memory_ranges[crash_mem_ranges].size = end - base;
 737	crash_mem_ranges++;
 738}
 739
 740static void fadump_exclude_reserved_area(unsigned long long start,
 741					unsigned long long end)
 742{
 743	unsigned long long ra_start, ra_end;
 
 744
 745	ra_start = fw_dump.reserve_dump_area_start;
 746	ra_end = ra_start + fw_dump.reserve_dump_area_size;
 747
 748	if ((ra_start < end) && (ra_end > start)) {
 749		if ((start < ra_start) && (end > ra_end)) {
 750			fadump_add_crash_memory(start, ra_start);
 751			fadump_add_crash_memory(ra_end, end);
 
 
 
 
 
 752		} else if (start < ra_start) {
 753			fadump_add_crash_memory(start, ra_start);
 
 754		} else if (ra_end < end) {
 755			fadump_add_crash_memory(ra_end, end);
 
 756		}
 757	} else
 758		fadump_add_crash_memory(start, end);
 
 
 759}
 760
 761static int fadump_init_elfcore_header(char *bufp)
 762{
 763	struct elfhdr *elf;
 764
 765	elf = (struct elfhdr *) bufp;
 766	bufp += sizeof(struct elfhdr);
 767	memcpy(elf->e_ident, ELFMAG, SELFMAG);
 768	elf->e_ident[EI_CLASS] = ELF_CLASS;
 769	elf->e_ident[EI_DATA] = ELF_DATA;
 770	elf->e_ident[EI_VERSION] = EV_CURRENT;
 771	elf->e_ident[EI_OSABI] = ELF_OSABI;
 772	memset(elf->e_ident+EI_PAD, 0, EI_NIDENT-EI_PAD);
 773	elf->e_type = ET_CORE;
 774	elf->e_machine = ELF_ARCH;
 775	elf->e_version = EV_CURRENT;
 776	elf->e_entry = 0;
 777	elf->e_phoff = sizeof(struct elfhdr);
 778	elf->e_shoff = 0;
 779	elf->e_flags = ELF_CORE_EFLAGS;
 
 
 
 
 780	elf->e_ehsize = sizeof(struct elfhdr);
 781	elf->e_phentsize = sizeof(struct elf_phdr);
 782	elf->e_phnum = 0;
 783	elf->e_shentsize = 0;
 784	elf->e_shnum = 0;
 785	elf->e_shstrndx = 0;
 786
 787	return 0;
 788}
 789
 790/*
 791 * Traverse through memblock structure and setup crash memory ranges. These
 792 * ranges will be used create PT_LOAD program headers in elfcore header.
 793 */
 794static void fadump_setup_crash_memory_ranges(void)
 795{
 796	struct memblock_region *reg;
 797	unsigned long long start, end;
 
 798
 799	pr_debug("Setup crash memory ranges.\n");
 800	crash_mem_ranges = 0;
 
 801	/*
 802	 * add the first memory chunk (RMA_START through boot_memory_size) as
 803	 * a separate memory chunk. The reason is, at the time crash firmware
 804	 * will move the content of this memory chunk to different location
 805	 * specified during fadump registration. We need to create a separate
 806	 * program header for this chunk with the correct offset.
 807	 */
 808	fadump_add_crash_memory(RMA_START, fw_dump.boot_memory_size);
 
 
 
 
 
 
 809
 810	for_each_memblock(memory, reg) {
 811		start = (unsigned long long)reg->base;
 812		end = start + (unsigned long long)reg->size;
 813		if (start == RMA_START && end >= fw_dump.boot_memory_size)
 814			start = fw_dump.boot_memory_size;
 
 
 
 
 
 
 
 
 
 815
 816		/* add this range excluding the reserved dump area. */
 817		fadump_exclude_reserved_area(start, end);
 
 
 818	}
 
 
 819}
 820
 821/*
 822 * If the given physical address falls within the boot memory region then
 823 * return the relocated address that points to the dump region reserved
 824 * for saving initial boot memory contents.
 825 */
 826static inline unsigned long fadump_relocate(unsigned long paddr)
 827{
 828	if (paddr > RMA_START && paddr < fw_dump.boot_memory_size)
 829		return fdm.rmr_region.destination_address + paddr;
 830	else
 831		return paddr;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 832}
 833
 834static int fadump_create_elfcore_headers(char *bufp)
 835{
 
 
 836	struct elfhdr *elf;
 837	struct elf_phdr *phdr;
 838	int i;
 839
 840	fadump_init_elfcore_header(bufp);
 841	elf = (struct elfhdr *)bufp;
 842	bufp += sizeof(struct elfhdr);
 843
 844	/*
 845	 * setup ELF PT_NOTE, place holder for cpu notes info. The notes info
 846	 * will be populated during second kernel boot after crash. Hence
 847	 * this PT_NOTE will always be the first elf note.
 848	 *
 849	 * NOTE: Any new ELF note addition should be placed after this note.
 850	 */
 851	phdr = (struct elf_phdr *)bufp;
 852	bufp += sizeof(struct elf_phdr);
 853	phdr->p_type = PT_NOTE;
 854	phdr->p_flags = 0;
 855	phdr->p_vaddr = 0;
 856	phdr->p_align = 0;
 857
 858	phdr->p_offset = 0;
 859	phdr->p_paddr = 0;
 860	phdr->p_filesz = 0;
 861	phdr->p_memsz = 0;
 862
 863	(elf->e_phnum)++;
 864
 865	/* setup ELF PT_NOTE for vmcoreinfo */
 866	phdr = (struct elf_phdr *)bufp;
 867	bufp += sizeof(struct elf_phdr);
 868	phdr->p_type	= PT_NOTE;
 869	phdr->p_flags	= 0;
 870	phdr->p_vaddr	= 0;
 871	phdr->p_align	= 0;
 872
 873	phdr->p_paddr	= fadump_relocate(paddr_vmcoreinfo_note());
 874	phdr->p_offset	= phdr->p_paddr;
 875	phdr->p_memsz	= vmcoreinfo_max_size;
 876	phdr->p_filesz	= vmcoreinfo_max_size;
 877
 878	/* Increment number of program headers. */
 879	(elf->e_phnum)++;
 880
 881	/* setup PT_LOAD sections. */
 
 
 
 
 
 882
 883	for (i = 0; i < crash_mem_ranges; i++) {
 884		unsigned long long mbase, msize;
 885		mbase = crash_memory_ranges[i].base;
 886		msize = crash_memory_ranges[i].size;
 887
 888		if (!msize)
 889			continue;
 890
 891		phdr = (struct elf_phdr *)bufp;
 892		bufp += sizeof(struct elf_phdr);
 893		phdr->p_type	= PT_LOAD;
 894		phdr->p_flags	= PF_R|PF_W|PF_X;
 895		phdr->p_offset	= mbase;
 896
 897		if (mbase == RMA_START) {
 898			/*
 899			 * The entire RMA region will be moved by firmware
 900			 * to the specified destination_address. Hence set
 901			 * the correct offset.
 902			 */
 903			phdr->p_offset = fdm.rmr_region.destination_address;
 
 
 
 
 904		}
 905
 906		phdr->p_paddr = mbase;
 907		phdr->p_vaddr = (unsigned long)__va(mbase);
 908		phdr->p_filesz = msize;
 909		phdr->p_memsz = msize;
 910		phdr->p_align = 0;
 911
 912		/* Increment number of program headers. */
 913		(elf->e_phnum)++;
 914	}
 915	return 0;
 916}
 917
 918static unsigned long init_fadump_header(unsigned long addr)
 919{
 920	struct fadump_crash_info_header *fdh;
 921
 922	if (!addr)
 923		return 0;
 924
 925	fw_dump.fadumphdr_addr = addr;
 926	fdh = __va(addr);
 927	addr += sizeof(struct fadump_crash_info_header);
 928
 929	memset(fdh, 0, sizeof(struct fadump_crash_info_header));
 930	fdh->magic_number = FADUMP_CRASH_INFO_MAGIC;
 931	fdh->elfcorehdr_addr = addr;
 932	/* We will set the crashing cpu id in crash_fadump() during crash. */
 933	fdh->crashing_cpu = CPU_UNKNOWN;
 934
 935	return addr;
 936}
 937
 938static void register_fadump(void)
 939{
 940	unsigned long addr;
 941	void *vaddr;
 
 942
 943	/*
 944	 * If no memory is reserved then we can not register for firmware-
 945	 * assisted dump.
 946	 */
 947	if (!fw_dump.reserve_dump_area_size)
 948		return;
 
 
 
 
 949
 950	fadump_setup_crash_memory_ranges();
 951
 952	addr = fdm.rmr_region.destination_address + fdm.rmr_region.source_len;
 953	/* Initialize fadump crash info header. */
 954	addr = init_fadump_header(addr);
 955	vaddr = __va(addr);
 956
 957	pr_debug("Creating ELF core headers at %#016lx\n", addr);
 958	fadump_create_elfcore_headers(vaddr);
 959
 960	/* register the future kernel dump with firmware. */
 961	register_fw_dump(&fdm);
 
 962}
 963
 964static int fadump_unregister_dump(struct fadump_mem_struct *fdm)
 965{
 966	int rc = 0;
 967	unsigned int wait_time;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 968
 969	pr_debug("Un-register firmware-assisted dump\n");
 
 
 
 
 
 
 970
 971	/* TODO: Add upper time limit for the delay */
 972	do {
 973		rc = rtas_call(fw_dump.ibm_configure_kernel_dump, 3, 1, NULL,
 974			FADUMP_UNREGISTER, fdm,
 975			sizeof(struct fadump_mem_struct));
 976
 977		wait_time = rtas_busy_delay_time(rc);
 978		if (wait_time)
 979			mdelay(wait_time);
 980	} while (wait_time);
 981
 982	if (rc) {
 983		printk(KERN_ERR "Failed to un-register firmware-assisted dump."
 984			" unexpected error(%d).\n", rc);
 985		return rc;
 986	}
 987	fw_dump.dump_registered = 0;
 988	return 0;
 989}
 990
 991static int fadump_invalidate_dump(struct fadump_mem_struct *fdm)
 
 
 
 
 992{
 993	int rc = 0;
 994	unsigned int wait_time;
 
 
 995
 996	pr_debug("Invalidating firmware-assisted dump registration\n");
 
 997
 998	/* TODO: Add upper time limit for the delay */
 999	do {
1000		rc = rtas_call(fw_dump.ibm_configure_kernel_dump, 3, 1, NULL,
1001			FADUMP_INVALIDATE, fdm,
1002			sizeof(struct fadump_mem_struct));
1003
1004		wait_time = rtas_busy_delay_time(rc);
1005		if (wait_time)
1006			mdelay(wait_time);
1007	} while (wait_time);
 
 
 
 
1008
1009	if (rc) {
1010		printk(KERN_ERR "Failed to invalidate firmware-assisted dump "
1011			"rgistration. unexpected error(%d).\n", rc);
1012		return rc;
 
 
 
 
 
 
 
 
 
 
1013	}
1014	fw_dump.dump_active = 0;
1015	fdm_active = NULL;
1016	return 0;
1017}
1018
1019void fadump_cleanup(void)
 
 
 
 
1020{
1021	/* Invalidate the registration only if dump is active. */
1022	if (fw_dump.dump_active) {
1023		init_fadump_mem_struct(&fdm,
1024			fdm_active->cpu_state_data.destination_address);
1025		fadump_invalidate_dump(&fdm);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1026	}
 
 
 
1027}
1028
1029/*
1030 * Release the memory that was reserved in early boot to preserve the memory
1031 * contents. The released memory will be available for general use.
 
 
1032 */
1033static void fadump_release_memory(unsigned long begin, unsigned long end)
1034{
1035	unsigned long addr;
1036	unsigned long ra_start, ra_end;
1037
1038	ra_start = fw_dump.reserve_dump_area_start;
1039	ra_end = ra_start + fw_dump.reserve_dump_area_size;
1040
1041	for (addr = begin; addr < end; addr += PAGE_SIZE) {
1042		/*
1043		 * exclude the dump reserve area. Will reuse it for next
1044		 * fadump registration.
1045		 */
1046		if (addr <= ra_end && ((addr + PAGE_SIZE) > ra_start))
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1047			continue;
1048
1049		ClearPageReserved(pfn_to_page(addr >> PAGE_SHIFT));
1050		init_page_count(pfn_to_page(addr >> PAGE_SHIFT));
1051		free_page((unsigned long)__va(addr));
1052		totalram_pages++;
1053	}
 
 
 
1054}
1055
1056static void fadump_invalidate_release_mem(void)
1057{
1058	unsigned long reserved_area_start, reserved_area_end;
1059	unsigned long destination_address;
1060
1061	mutex_lock(&fadump_mutex);
1062	if (!fw_dump.dump_active) {
1063		mutex_unlock(&fadump_mutex);
1064		return;
1065	}
1066
1067	destination_address = fdm_active->cpu_state_data.destination_address;
1068	fadump_cleanup();
1069	mutex_unlock(&fadump_mutex);
1070
 
 
 
1071	/*
1072	 * Save the current reserved memory bounds we will require them
1073	 * later for releasing the memory for general use.
1074	 */
1075	reserved_area_start = fw_dump.reserve_dump_area_start;
1076	reserved_area_end = reserved_area_start +
1077			fw_dump.reserve_dump_area_size;
1078	/*
1079	 * Setup reserve_dump_area_start and its size so that we can
1080	 * reuse this reserved memory for Re-registration.
1081	 */
1082	fw_dump.reserve_dump_area_start = destination_address;
1083	fw_dump.reserve_dump_area_size = get_fadump_area_size();
1084
1085	fadump_release_memory(reserved_area_start, reserved_area_end);
1086	if (fw_dump.cpu_notes_buf) {
1087		fadump_cpu_notes_buf_free(
1088				(unsigned long)__va(fw_dump.cpu_notes_buf),
1089				fw_dump.cpu_notes_buf_size);
1090		fw_dump.cpu_notes_buf = 0;
1091		fw_dump.cpu_notes_buf_size = 0;
1092	}
1093	/* Initialize the kernel dump memory structure for FAD registration. */
1094	init_fadump_mem_struct(&fdm, fw_dump.reserve_dump_area_start);
1095}
1096
1097static ssize_t fadump_release_memory_store(struct kobject *kobj,
1098					struct kobj_attribute *attr,
1099					const char *buf, size_t count)
1100{
 
 
1101	if (!fw_dump.dump_active)
1102		return -EPERM;
1103
1104	if (buf[0] == '1') {
 
 
 
1105		/*
1106		 * Take away the '/proc/vmcore'. We are releasing the dump
1107		 * memory, hence it will not be valid anymore.
1108		 */
 
1109		vmcore_cleanup();
 
1110		fadump_invalidate_release_mem();
1111
1112	} else
1113		return -EINVAL;
1114	return count;
1115}
1116
1117static ssize_t fadump_enabled_show(struct kobject *kobj,
1118					struct kobj_attribute *attr,
1119					char *buf)
 
 
 
 
 
 
 
 
 
 
1120{
1121	return sprintf(buf, "%d\n", fw_dump.fadump_enabled);
1122}
1123
1124static ssize_t fadump_register_show(struct kobject *kobj,
1125					struct kobj_attribute *attr,
1126					char *buf)
 
 
 
 
 
 
 
1127{
1128	return sprintf(buf, "%d\n", fw_dump.dump_registered);
1129}
1130
1131static ssize_t fadump_register_store(struct kobject *kobj,
1132					struct kobj_attribute *attr,
1133					const char *buf, size_t count)
1134{
1135	int ret = 0;
 
1136
1137	if (!fw_dump.fadump_enabled || fdm_active)
1138		return -EPERM;
1139
 
 
 
1140	mutex_lock(&fadump_mutex);
1141
1142	switch (buf[0]) {
1143	case '0':
1144		if (fw_dump.dump_registered == 0) {
1145			ret = -EINVAL;
1146			goto unlock_out;
1147		}
 
1148		/* Un-register Firmware-assisted dump */
1149		fadump_unregister_dump(&fdm);
 
1150		break;
1151	case '1':
1152		if (fw_dump.dump_registered == 1) {
1153			ret = -EINVAL;
1154			goto unlock_out;
1155		}
1156		/* Register Firmware-assisted dump */
1157		register_fadump();
1158		break;
1159	default:
1160		ret = -EINVAL;
1161		break;
1162	}
1163
1164unlock_out:
1165	mutex_unlock(&fadump_mutex);
1166	return ret < 0 ? ret : count;
1167}
1168
1169static int fadump_region_show(struct seq_file *m, void *private)
1170{
1171	const struct fadump_mem_struct *fdm_ptr;
1172
1173	if (!fw_dump.fadump_enabled)
1174		return 0;
1175
1176	mutex_lock(&fadump_mutex);
1177	if (fdm_active)
1178		fdm_ptr = fdm_active;
1179	else {
1180		mutex_unlock(&fadump_mutex);
1181		fdm_ptr = &fdm;
1182	}
1183
1184	seq_printf(m,
1185			"CPU : [%#016llx-%#016llx] %#llx bytes, "
1186			"Dumped: %#llx\n",
1187			fdm_ptr->cpu_state_data.destination_address,
1188			fdm_ptr->cpu_state_data.destination_address +
1189			fdm_ptr->cpu_state_data.source_len - 1,
1190			fdm_ptr->cpu_state_data.source_len,
1191			fdm_ptr->cpu_state_data.bytes_dumped);
1192	seq_printf(m,
1193			"HPTE: [%#016llx-%#016llx] %#llx bytes, "
1194			"Dumped: %#llx\n",
1195			fdm_ptr->hpte_region.destination_address,
1196			fdm_ptr->hpte_region.destination_address +
1197			fdm_ptr->hpte_region.source_len - 1,
1198			fdm_ptr->hpte_region.source_len,
1199			fdm_ptr->hpte_region.bytes_dumped);
1200	seq_printf(m,
1201			"DUMP: [%#016llx-%#016llx] %#llx bytes, "
1202			"Dumped: %#llx\n",
1203			fdm_ptr->rmr_region.destination_address,
1204			fdm_ptr->rmr_region.destination_address +
1205			fdm_ptr->rmr_region.source_len - 1,
1206			fdm_ptr->rmr_region.source_len,
1207			fdm_ptr->rmr_region.bytes_dumped);
1208
1209	if (!fdm_active ||
1210		(fw_dump.reserve_dump_area_start ==
1211		fdm_ptr->cpu_state_data.destination_address))
1212		goto out;
1213
1214	/* Dump is active. Show reserved memory region. */
1215	seq_printf(m,
1216			"    : [%#016llx-%#016llx] %#llx bytes, "
1217			"Dumped: %#llx\n",
1218			(unsigned long long)fw_dump.reserve_dump_area_start,
1219			fdm_ptr->cpu_state_data.destination_address - 1,
1220			fdm_ptr->cpu_state_data.destination_address -
1221			fw_dump.reserve_dump_area_start,
1222			fdm_ptr->cpu_state_data.destination_address -
1223			fw_dump.reserve_dump_area_start);
1224out:
1225	if (fdm_active)
1226		mutex_unlock(&fadump_mutex);
1227	return 0;
1228}
1229
1230static struct kobj_attribute fadump_release_attr = __ATTR(fadump_release_mem,
1231						0200, NULL,
1232						fadump_release_memory_store);
1233static struct kobj_attribute fadump_attr = __ATTR(fadump_enabled,
1234						0444, fadump_enabled_show,
1235						NULL);
1236static struct kobj_attribute fadump_register_attr = __ATTR(fadump_registered,
1237						0644, fadump_register_show,
1238						fadump_register_store);
1239
1240static int fadump_region_open(struct inode *inode, struct file *file)
1241{
1242	return single_open(file, fadump_region_show, inode->i_private);
1243}
1244
1245static const struct file_operations fadump_region_fops = {
1246	.open    = fadump_region_open,
1247	.read    = seq_read,
1248	.llseek  = seq_lseek,
1249	.release = single_release,
1250};
1251
 
 
 
 
1252static void fadump_init_files(void)
1253{
1254	struct dentry *debugfs_file;
1255	int rc = 0;
1256
1257	rc = sysfs_create_file(kernel_kobj, &fadump_attr.attr);
1258	if (rc)
1259		printk(KERN_ERR "fadump: unable to create sysfs file"
1260			" fadump_enabled (%d)\n", rc);
1261
1262	rc = sysfs_create_file(kernel_kobj, &fadump_register_attr.attr);
1263	if (rc)
1264		printk(KERN_ERR "fadump: unable to create sysfs file"
1265			" fadump_registered (%d)\n", rc);
1266
1267	debugfs_file = debugfs_create_file("fadump_region", 0444,
1268					powerpc_debugfs_root, NULL,
1269					&fadump_region_fops);
1270	if (!debugfs_file)
1271		printk(KERN_ERR "fadump: unable to create debugfs file"
1272				" fadump_region\n");
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1273
1274	if (fw_dump.dump_active) {
1275		rc = sysfs_create_file(kernel_kobj, &fadump_release_attr.attr);
 
 
 
1276		if (rc)
1277			printk(KERN_ERR "fadump: unable to create sysfs file"
1278				" fadump_release_mem (%d)\n", rc);
1279	}
1280	return;
1281}
1282
1283/*
1284 * Prepare for firmware-assisted dump.
1285 */
1286int __init setup_fadump(void)
1287{
1288	if (!fw_dump.fadump_enabled)
1289		return 0;
1290
1291	if (!fw_dump.fadump_supported) {
1292		printk(KERN_ERR "Firmware-assisted dump is not supported on"
1293			" this hardware\n");
1294		return 0;
1295	}
1296
1297	fadump_show_config();
1298	/*
1299	 * If dump data is available then see if it is valid and prepare for
1300	 * saving it to the disk.
1301	 */
1302	if (fw_dump.dump_active) {
1303		/*
1304		 * if dump process fails then invalidate the registration
1305		 * and release memory before proceeding for re-registration.
1306		 */
1307		if (process_fadump(fdm_active) < 0)
1308			fadump_invalidate_release_mem();
1309	}
1310	/* Initialize the kernel dump memory structure for FAD registration. */
1311	else if (fw_dump.reserve_dump_area_size)
1312		init_fadump_mem_struct(&fdm, fw_dump.reserve_dump_area_start);
1313	fadump_init_files();
1314
1315	return 1;
1316}
1317subsys_initcall(setup_fadump);