1 // SPDX-License-Identifier: GPL-2.0-or-later
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
10 * Copyright 2011 IBM Corporation
11 * Author: Mahesh Salgaonkar <mahesh@linux.vnet.ibm.com>
15 #define pr_fmt(fmt) "fadump: " fmt
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>
28 #include <asm/debugfs.h>
31 #include <asm/fadump.h>
32 #include <asm/fadump-internal.h>
33 #include <asm/setup.h>
35 static struct fw_dump fw_dump;
37 static void __init fadump_reserve_crash_area(u64 base);
39 #ifndef CONFIG_PRESERVE_FA_DUMP
40 static DEFINE_MUTEX(fadump_mutex);
41 struct fadump_mrange_info crash_mrange_info = { "crash", NULL, 0, 0, 0, false };
43 #define RESERVED_RNGS_SZ 16384 /* 16K - 128 entries */
44 #define RESERVED_RNGS_CNT (RESERVED_RNGS_SZ / \
45 sizeof(struct fadump_memory_range))
46 static struct fadump_memory_range rngs[RESERVED_RNGS_CNT];
47 struct fadump_mrange_info reserved_mrange_info = { "reserved", rngs,
49 RESERVED_RNGS_CNT, true };
51 static void __init early_init_dt_scan_reserved_ranges(unsigned long node);
54 static struct cma *fadump_cma;
57 * fadump_cma_init() - Initialize CMA area from a fadump reserved memory
59 * This function initializes CMA area from fadump reserved memory.
60 * The total size of fadump reserved memory covers for boot memory size
61 * + cpu data size + hpte size and metadata.
62 * Initialize only the area equivalent to boot memory size for CMA use.
63 * The reamining portion of fadump reserved memory will be not given
64 * to CMA and pages for thoes will stay reserved. boot memory size is
65 * aligned per CMA requirement to satisy cma_init_reserved_mem() call.
66 * But for some reason even if it fails we still have the memory reservation
67 * with us and we can still continue doing fadump.
69 int __init fadump_cma_init(void)
71 unsigned long long base, size;
74 if (!fw_dump.fadump_enabled)
78 * Do not use CMA if user has provided fadump=nocma kernel parameter.
79 * Return 1 to continue with fadump old behaviour.
84 base = fw_dump.reserve_dump_area_start;
85 size = fw_dump.boot_memory_size;
90 rc = cma_init_reserved_mem(base, size, 0, "fadump_cma", &fadump_cma);
92 pr_err("Failed to init cma area for firmware-assisted dump,%d\n", rc);
94 * Though the CMA init has failed we still have memory
95 * reservation with us. The reserved memory will be
96 * blocked from production system usage. Hence return 1,
97 * so that we can continue with fadump.
103 * So we now have successfully initialized cma area for fadump.
105 pr_info("Initialized 0x%lx bytes cma area at %ldMB from 0x%lx "
106 "bytes of memory reserved for firmware-assisted dump\n",
107 cma_get_size(fadump_cma),
108 (unsigned long)cma_get_base(fadump_cma) >> 20,
109 fw_dump.reserve_dump_area_size);
113 static int __init fadump_cma_init(void) { return 1; }
114 #endif /* CONFIG_CMA */
116 /* Scan the Firmware Assisted dump configuration details. */
117 int __init early_init_dt_scan_fw_dump(unsigned long node, const char *uname,
118 int depth, void *data)
121 early_init_dt_scan_reserved_ranges(node);
128 if (strcmp(uname, "rtas") == 0) {
129 rtas_fadump_dt_scan(&fw_dump, node);
133 if (strcmp(uname, "ibm,opal") == 0) {
134 opal_fadump_dt_scan(&fw_dump, node);
142 * If fadump is registered, check if the memory provided
143 * falls within boot memory area and reserved memory area.
145 int is_fadump_memory_area(u64 addr, unsigned long size)
149 if (!fw_dump.dump_registered)
155 d_start = fw_dump.reserve_dump_area_start;
156 d_end = d_start + fw_dump.reserve_dump_area_size;
157 if (((addr + size) > d_start) && (addr <= d_end))
160 return (addr <= fw_dump.boot_mem_top);
163 int should_fadump_crash(void)
165 if (!fw_dump.dump_registered || !fw_dump.fadumphdr_addr)
170 int is_fadump_active(void)
172 return fw_dump.dump_active;
176 * Returns true, if there are no holes in memory area between d_start to d_end,
179 static bool is_fadump_mem_area_contiguous(u64 d_start, u64 d_end)
181 struct memblock_region *reg;
185 for_each_memblock(memory, reg) {
186 start = max_t(u64, d_start, reg->base);
187 end = min_t(u64, d_end, (reg->base + reg->size));
189 /* Memory hole from d_start to start */
206 * Returns true, if there are no holes in boot memory area,
209 bool is_fadump_boot_mem_contiguous(void)
211 unsigned long d_start, d_end;
215 for (i = 0; i < fw_dump.boot_mem_regs_cnt; i++) {
216 d_start = fw_dump.boot_mem_addr[i];
217 d_end = d_start + fw_dump.boot_mem_sz[i];
219 ret = is_fadump_mem_area_contiguous(d_start, d_end);
228 * Returns true, if there are no holes in reserved memory area,
231 bool is_fadump_reserved_mem_contiguous(void)
235 d_start = fw_dump.reserve_dump_area_start;
236 d_end = d_start + fw_dump.reserve_dump_area_size;
237 return is_fadump_mem_area_contiguous(d_start, d_end);
240 /* Print firmware assisted dump configurations for debugging purpose. */
241 static void fadump_show_config(void)
245 pr_debug("Support for firmware-assisted dump (fadump): %s\n",
246 (fw_dump.fadump_supported ? "present" : "no support"));
248 if (!fw_dump.fadump_supported)
251 pr_debug("Fadump enabled : %s\n",
252 (fw_dump.fadump_enabled ? "yes" : "no"));
253 pr_debug("Dump Active : %s\n",
254 (fw_dump.dump_active ? "yes" : "no"));
255 pr_debug("Dump section sizes:\n");
256 pr_debug(" CPU state data size: %lx\n", fw_dump.cpu_state_data_size);
257 pr_debug(" HPTE region size : %lx\n", fw_dump.hpte_region_size);
258 pr_debug(" Boot memory size : %lx\n", fw_dump.boot_memory_size);
259 pr_debug(" Boot memory top : %llx\n", fw_dump.boot_mem_top);
260 pr_debug("Boot memory regions cnt: %llx\n", fw_dump.boot_mem_regs_cnt);
261 for (i = 0; i < fw_dump.boot_mem_regs_cnt; i++) {
262 pr_debug("[%03d] base = %llx, size = %llx\n", i,
263 fw_dump.boot_mem_addr[i], fw_dump.boot_mem_sz[i]);
268 * fadump_calculate_reserve_size(): reserve variable boot area 5% of System RAM
270 * Function to find the largest memory size we need to reserve during early
271 * boot process. This will be the size of the memory that is required for a
272 * kernel to boot successfully.
274 * This function has been taken from phyp-assisted dump feature implementation.
276 * returns larger of 256MB or 5% rounded down to multiples of 256MB.
278 * TODO: Come up with better approach to find out more accurate memory size
279 * that is required for a kernel to boot successfully.
282 static __init u64 fadump_calculate_reserve_size(void)
284 u64 base, size, bootmem_min;
287 if (fw_dump.reserve_bootvar)
288 pr_warn("'fadump_reserve_mem=' parameter is deprecated in favor of 'crashkernel=' parameter.\n");
291 * Check if the size is specified through crashkernel= cmdline
292 * option. If yes, then use that but ignore base as fadump reserves
293 * memory at a predefined offset.
295 ret = parse_crashkernel(boot_command_line, memblock_phys_mem_size(),
297 if (ret == 0 && size > 0) {
298 unsigned long max_size;
300 if (fw_dump.reserve_bootvar)
301 pr_info("Using 'crashkernel=' parameter for memory reservation.\n");
303 fw_dump.reserve_bootvar = (unsigned long)size;
306 * Adjust if the boot memory size specified is above
309 max_size = memblock_phys_mem_size() / MAX_BOOT_MEM_RATIO;
310 if (fw_dump.reserve_bootvar > max_size) {
311 fw_dump.reserve_bootvar = max_size;
312 pr_info("Adjusted boot memory size to %luMB\n",
313 (fw_dump.reserve_bootvar >> 20));
316 return fw_dump.reserve_bootvar;
317 } else if (fw_dump.reserve_bootvar) {
319 * 'fadump_reserve_mem=' is being used to reserve memory
320 * for firmware-assisted dump.
322 return fw_dump.reserve_bootvar;
325 /* divide by 20 to get 5% of value */
326 size = memblock_phys_mem_size() / 20;
328 /* round it down in multiples of 256 */
329 size = size & ~0x0FFFFFFFUL;
331 /* Truncate to memory_limit. We don't want to over reserve the memory.*/
332 if (memory_limit && size > memory_limit)
335 bootmem_min = fw_dump.ops->fadump_get_bootmem_min();
336 return (size > bootmem_min ? size : bootmem_min);
340 * Calculate the total memory size required to be reserved for
341 * firmware-assisted dump registration.
343 static unsigned long get_fadump_area_size(void)
345 unsigned long size = 0;
347 size += fw_dump.cpu_state_data_size;
348 size += fw_dump.hpte_region_size;
349 size += fw_dump.boot_memory_size;
350 size += sizeof(struct fadump_crash_info_header);
351 size += sizeof(struct elfhdr); /* ELF core header.*/
352 size += sizeof(struct elf_phdr); /* place holder for cpu notes */
353 /* Program headers for crash memory regions. */
354 size += sizeof(struct elf_phdr) * (memblock_num_regions(memory) + 2);
356 size = PAGE_ALIGN(size);
358 /* This is to hold kernel metadata on platforms that support it */
359 size += (fw_dump.ops->fadump_get_metadata_size ?
360 fw_dump.ops->fadump_get_metadata_size() : 0);
364 static int __init add_boot_mem_region(unsigned long rstart,
367 int i = fw_dump.boot_mem_regs_cnt++;
369 if (fw_dump.boot_mem_regs_cnt > FADUMP_MAX_MEM_REGS) {
370 fw_dump.boot_mem_regs_cnt = FADUMP_MAX_MEM_REGS;
374 pr_debug("Added boot memory range[%d] [%#016lx-%#016lx)\n",
375 i, rstart, (rstart + rsize));
376 fw_dump.boot_mem_addr[i] = rstart;
377 fw_dump.boot_mem_sz[i] = rsize;
382 * Firmware usually has a hard limit on the data it can copy per region.
383 * Honour that by splitting a memory range into multiple regions.
385 static int __init add_boot_mem_regions(unsigned long mstart,
388 unsigned long rstart, rsize, max_size;
392 max_size = fw_dump.max_copy_size ? fw_dump.max_copy_size : msize;
394 if (msize > max_size)
399 ret = add_boot_mem_region(rstart, rsize);
410 static int __init fadump_get_boot_mem_regions(void)
412 unsigned long base, size, cur_size, hole_size, last_end;
413 unsigned long mem_size = fw_dump.boot_memory_size;
414 struct memblock_region *reg;
417 fw_dump.boot_mem_regs_cnt = 0;
422 for_each_memblock(memory, reg) {
425 hole_size += (base - last_end);
427 if ((cur_size + size) >= mem_size) {
428 size = (mem_size - cur_size);
429 ret = add_boot_mem_regions(base, size);
435 ret = add_boot_mem_regions(base, size);
439 last_end = base + size;
441 fw_dump.boot_mem_top = PAGE_ALIGN(fw_dump.boot_memory_size + hole_size);
447 * Returns true, if the given range overlaps with reserved memory ranges
448 * starting at idx. Also, updates idx to index of overlapping memory range
449 * with the given memory range.
452 static bool overlaps_reserved_ranges(u64 base, u64 end, int *idx)
457 for (i = *idx; i < reserved_mrange_info.mem_range_cnt; i++) {
458 u64 rbase = reserved_mrange_info.mem_ranges[i].base;
459 u64 rend = rbase + reserved_mrange_info.mem_ranges[i].size;
464 if ((end > rbase) && (base < rend)) {
475 * Locate a suitable memory area to reserve memory for FADump. While at it,
476 * lookup reserved-ranges & avoid overlap with them, as they are used by F/W.
478 static u64 __init fadump_locate_reserve_mem(u64 base, u64 size)
480 struct fadump_memory_range *mrngs;
481 phys_addr_t mstart, mend;
485 mrngs = reserved_mrange_info.mem_ranges;
486 for_each_free_mem_range(i, NUMA_NO_NODE, MEMBLOCK_NONE,
487 &mstart, &mend, NULL) {
488 pr_debug("%llu) mstart: %llx, mend: %llx, base: %llx\n",
489 i, mstart, mend, base);
492 base = PAGE_ALIGN(mstart);
494 while ((mend > base) && ((mend - base) >= size)) {
495 if (!overlaps_reserved_ranges(base, base+size, &idx)) {
500 base = mrngs[idx].base + mrngs[idx].size;
501 base = PAGE_ALIGN(base);
509 int __init fadump_reserve_mem(void)
511 u64 base, size, mem_boundary, bootmem_min;
514 if (!fw_dump.fadump_enabled)
517 if (!fw_dump.fadump_supported) {
518 pr_info("Firmware-Assisted Dump is not supported on this hardware\n");
523 * Initialize boot memory size
524 * If dump is active then we have already calculated the size during
527 if (!fw_dump.dump_active) {
528 fw_dump.boot_memory_size =
529 PAGE_ALIGN(fadump_calculate_reserve_size());
531 if (!fw_dump.nocma) {
532 fw_dump.boot_memory_size =
533 ALIGN(fw_dump.boot_memory_size,
534 FADUMP_CMA_ALIGNMENT);
538 bootmem_min = fw_dump.ops->fadump_get_bootmem_min();
539 if (fw_dump.boot_memory_size < bootmem_min) {
540 pr_err("Can't enable fadump with boot memory size (0x%lx) less than 0x%llx\n",
541 fw_dump.boot_memory_size, bootmem_min);
545 if (!fadump_get_boot_mem_regions()) {
546 pr_err("Too many holes in boot memory area to enable fadump\n");
552 * Calculate the memory boundary.
553 * If memory_limit is less than actual memory boundary then reserve
554 * the memory for fadump beyond the memory_limit and adjust the
555 * memory_limit accordingly, so that the running kernel can run with
556 * specified memory_limit.
558 if (memory_limit && memory_limit < memblock_end_of_DRAM()) {
559 size = get_fadump_area_size();
560 if ((memory_limit + size) < memblock_end_of_DRAM())
561 memory_limit += size;
563 memory_limit = memblock_end_of_DRAM();
564 printk(KERN_INFO "Adjusted memory_limit for firmware-assisted"
565 " dump, now %#016llx\n", memory_limit);
568 mem_boundary = memory_limit;
570 mem_boundary = memblock_end_of_DRAM();
572 base = fw_dump.boot_mem_top;
573 size = get_fadump_area_size();
574 fw_dump.reserve_dump_area_size = size;
575 if (fw_dump.dump_active) {
576 pr_info("Firmware-assisted dump is active.\n");
578 #ifdef CONFIG_HUGETLB_PAGE
580 * FADump capture kernel doesn't care much about hugepages.
581 * In fact, handling hugepages in capture kernel is asking for
582 * trouble. So, disable HugeTLB support when fadump is active.
584 hugetlb_disabled = true;
587 * If last boot has crashed then reserve all the memory
588 * above boot memory size so that we don't touch it until
589 * dump is written to disk by userspace tool. This memory
590 * can be released for general use by invalidating fadump.
592 fadump_reserve_crash_area(base);
594 pr_debug("fadumphdr_addr = %#016lx\n", fw_dump.fadumphdr_addr);
595 pr_debug("Reserve dump area start address: 0x%lx\n",
596 fw_dump.reserve_dump_area_start);
599 * Reserve memory at an offset closer to bottom of the RAM to
600 * minimize the impact of memory hot-remove operation.
602 base = fadump_locate_reserve_mem(base, size);
604 if (!base || (base + size > mem_boundary)) {
605 pr_err("Failed to find memory chunk for reservation!\n");
608 fw_dump.reserve_dump_area_start = base;
611 * Calculate the kernel metadata address and register it with
612 * f/w if the platform supports.
614 if (fw_dump.ops->fadump_setup_metadata &&
615 (fw_dump.ops->fadump_setup_metadata(&fw_dump) < 0))
618 if (memblock_reserve(base, size)) {
619 pr_err("Failed to reserve memory!\n");
623 pr_info("Reserved %lldMB of memory at %#016llx (System RAM: %lldMB)\n",
624 (size >> 20), base, (memblock_phys_mem_size() >> 20));
626 ret = fadump_cma_init();
631 fw_dump.fadump_enabled = 0;
632 fw_dump.reserve_dump_area_size = 0;
636 /* Look for fadump= cmdline option. */
637 static int __init early_fadump_param(char *p)
642 if (strncmp(p, "on", 2) == 0)
643 fw_dump.fadump_enabled = 1;
644 else if (strncmp(p, "off", 3) == 0)
645 fw_dump.fadump_enabled = 0;
646 else if (strncmp(p, "nocma", 5) == 0) {
647 fw_dump.fadump_enabled = 1;
653 early_param("fadump", early_fadump_param);
656 * Look for fadump_reserve_mem= cmdline option
657 * TODO: Remove references to 'fadump_reserve_mem=' parameter,
658 * the sooner 'crashkernel=' parameter is accustomed to.
660 static int __init early_fadump_reserve_mem(char *p)
663 fw_dump.reserve_bootvar = memparse(p, &p);
666 early_param("fadump_reserve_mem", early_fadump_reserve_mem);
668 void crash_fadump(struct pt_regs *regs, const char *str)
670 struct fadump_crash_info_header *fdh = NULL;
671 int old_cpu, this_cpu;
673 if (!should_fadump_crash())
677 * old_cpu == -1 means this is the first CPU which has come here,
678 * go ahead and trigger fadump.
680 * old_cpu != -1 means some other CPU has already on it's way
681 * to trigger fadump, just keep looping here.
683 this_cpu = smp_processor_id();
684 old_cpu = cmpxchg(&crashing_cpu, -1, this_cpu);
688 * We can't loop here indefinitely. Wait as long as fadump
689 * is in force. If we race with fadump un-registration this
690 * loop will break and then we go down to normal panic path
691 * and reboot. If fadump is in force the first crashing
692 * cpu will definitely trigger fadump.
694 while (fw_dump.dump_registered)
699 fdh = __va(fw_dump.fadumphdr_addr);
700 fdh->crashing_cpu = crashing_cpu;
701 crash_save_vmcoreinfo();
706 ppc_save_regs(&fdh->regs);
708 fdh->online_mask = *cpu_online_mask;
710 fw_dump.ops->fadump_trigger(fdh, str);
713 u32 *fadump_regs_to_elf_notes(u32 *buf, struct pt_regs *regs)
715 struct elf_prstatus prstatus;
717 memset(&prstatus, 0, sizeof(prstatus));
719 * FIXME: How do i get PID? Do I really need it?
720 * prstatus.pr_pid = ????
722 elf_core_copy_kernel_regs(&prstatus.pr_reg, regs);
723 buf = append_elf_note(buf, CRASH_CORE_NOTE_NAME, NT_PRSTATUS,
724 &prstatus, sizeof(prstatus));
728 void fadump_update_elfcore_header(char *bufp)
731 struct elf_phdr *phdr;
733 elf = (struct elfhdr *)bufp;
734 bufp += sizeof(struct elfhdr);
736 /* First note is a place holder for cpu notes info. */
737 phdr = (struct elf_phdr *)bufp;
739 if (phdr->p_type == PT_NOTE) {
740 phdr->p_paddr = __pa(fw_dump.cpu_notes_buf_vaddr);
741 phdr->p_offset = phdr->p_paddr;
742 phdr->p_filesz = fw_dump.cpu_notes_buf_size;
743 phdr->p_memsz = fw_dump.cpu_notes_buf_size;
748 static void *fadump_alloc_buffer(unsigned long size)
750 unsigned long count, i;
754 vaddr = alloc_pages_exact(size, GFP_KERNEL | __GFP_ZERO);
758 count = PAGE_ALIGN(size) / PAGE_SIZE;
759 page = virt_to_page(vaddr);
760 for (i = 0; i < count; i++)
761 mark_page_reserved(page + i);
765 static void fadump_free_buffer(unsigned long vaddr, unsigned long size)
767 free_reserved_area((void *)vaddr, (void *)(vaddr + size), -1, NULL);
770 s32 fadump_setup_cpu_notes_buf(u32 num_cpus)
772 /* Allocate buffer to hold cpu crash notes. */
773 fw_dump.cpu_notes_buf_size = num_cpus * sizeof(note_buf_t);
774 fw_dump.cpu_notes_buf_size = PAGE_ALIGN(fw_dump.cpu_notes_buf_size);
775 fw_dump.cpu_notes_buf_vaddr =
776 (unsigned long)fadump_alloc_buffer(fw_dump.cpu_notes_buf_size);
777 if (!fw_dump.cpu_notes_buf_vaddr) {
778 pr_err("Failed to allocate %ld bytes for CPU notes buffer\n",
779 fw_dump.cpu_notes_buf_size);
783 pr_debug("Allocated buffer for cpu notes of size %ld at 0x%lx\n",
784 fw_dump.cpu_notes_buf_size,
785 fw_dump.cpu_notes_buf_vaddr);
789 void fadump_free_cpu_notes_buf(void)
791 if (!fw_dump.cpu_notes_buf_vaddr)
794 fadump_free_buffer(fw_dump.cpu_notes_buf_vaddr,
795 fw_dump.cpu_notes_buf_size);
796 fw_dump.cpu_notes_buf_vaddr = 0;
797 fw_dump.cpu_notes_buf_size = 0;
800 static void fadump_free_mem_ranges(struct fadump_mrange_info *mrange_info)
802 if (mrange_info->is_static) {
803 mrange_info->mem_range_cnt = 0;
807 kfree(mrange_info->mem_ranges);
808 memset((void *)((u64)mrange_info + RNG_NAME_SZ), 0,
809 (sizeof(struct fadump_mrange_info) - RNG_NAME_SZ));
813 * Allocate or reallocate mem_ranges array in incremental units
816 static int fadump_alloc_mem_ranges(struct fadump_mrange_info *mrange_info)
818 struct fadump_memory_range *new_array;
821 new_size = mrange_info->mem_ranges_sz + PAGE_SIZE;
822 pr_debug("Allocating %llu bytes of memory for %s memory ranges\n",
823 new_size, mrange_info->name);
825 new_array = krealloc(mrange_info->mem_ranges, new_size, GFP_KERNEL);
826 if (new_array == NULL) {
827 pr_err("Insufficient memory for setting up %s memory ranges\n",
829 fadump_free_mem_ranges(mrange_info);
833 mrange_info->mem_ranges = new_array;
834 mrange_info->mem_ranges_sz = new_size;
835 mrange_info->max_mem_ranges = (new_size /
836 sizeof(struct fadump_memory_range));
839 static inline int fadump_add_mem_range(struct fadump_mrange_info *mrange_info,
842 struct fadump_memory_range *mem_ranges = mrange_info->mem_ranges;
843 bool is_adjacent = false;
850 * Fold adjacent memory ranges to bring down the memory ranges/
851 * PT_LOAD segments count.
853 if (mrange_info->mem_range_cnt) {
854 start = mem_ranges[mrange_info->mem_range_cnt - 1].base;
855 size = mem_ranges[mrange_info->mem_range_cnt - 1].size;
858 * Boot memory area needs separate PT_LOAD segment(s) as it
859 * is moved to a different location at the time of crash.
860 * So, fold only if the region is not boot memory area.
862 if ((start + size) == base && start >= fw_dump.boot_mem_top)
866 /* resize the array on reaching the limit */
867 if (mrange_info->mem_range_cnt == mrange_info->max_mem_ranges) {
870 if (mrange_info->is_static) {
871 pr_err("Reached array size limit for %s memory ranges\n",
876 ret = fadump_alloc_mem_ranges(mrange_info);
880 /* Update to the new resized array */
881 mem_ranges = mrange_info->mem_ranges;
885 mem_ranges[mrange_info->mem_range_cnt].base = start;
886 mrange_info->mem_range_cnt++;
889 mem_ranges[mrange_info->mem_range_cnt - 1].size = (end - start);
890 pr_debug("%s_memory_range[%d] [%#016llx-%#016llx], %#llx bytes\n",
891 mrange_info->name, (mrange_info->mem_range_cnt - 1),
892 start, end - 1, (end - start));
896 static int fadump_exclude_reserved_area(u64 start, u64 end)
898 u64 ra_start, ra_end;
901 ra_start = fw_dump.reserve_dump_area_start;
902 ra_end = ra_start + fw_dump.reserve_dump_area_size;
904 if ((ra_start < end) && (ra_end > start)) {
905 if ((start < ra_start) && (end > ra_end)) {
906 ret = fadump_add_mem_range(&crash_mrange_info,
911 ret = fadump_add_mem_range(&crash_mrange_info,
913 } else if (start < ra_start) {
914 ret = fadump_add_mem_range(&crash_mrange_info,
916 } else if (ra_end < end) {
917 ret = fadump_add_mem_range(&crash_mrange_info,
921 ret = fadump_add_mem_range(&crash_mrange_info, start, end);
926 static int fadump_init_elfcore_header(char *bufp)
930 elf = (struct elfhdr *) bufp;
931 bufp += sizeof(struct elfhdr);
932 memcpy(elf->e_ident, ELFMAG, SELFMAG);
933 elf->e_ident[EI_CLASS] = ELF_CLASS;
934 elf->e_ident[EI_DATA] = ELF_DATA;
935 elf->e_ident[EI_VERSION] = EV_CURRENT;
936 elf->e_ident[EI_OSABI] = ELF_OSABI;
937 memset(elf->e_ident+EI_PAD, 0, EI_NIDENT-EI_PAD);
938 elf->e_type = ET_CORE;
939 elf->e_machine = ELF_ARCH;
940 elf->e_version = EV_CURRENT;
942 elf->e_phoff = sizeof(struct elfhdr);
944 #if defined(_CALL_ELF)
945 elf->e_flags = _CALL_ELF;
949 elf->e_ehsize = sizeof(struct elfhdr);
950 elf->e_phentsize = sizeof(struct elf_phdr);
952 elf->e_shentsize = 0;
960 * Traverse through memblock structure and setup crash memory ranges. These
961 * ranges will be used create PT_LOAD program headers in elfcore header.
963 static int fadump_setup_crash_memory_ranges(void)
965 struct memblock_region *reg;
969 pr_debug("Setup crash memory ranges.\n");
970 crash_mrange_info.mem_range_cnt = 0;
973 * Boot memory region(s) registered with firmware are moved to
974 * different location at the time of crash. Create separate program
975 * header(s) for this memory chunk(s) with the correct offset.
977 for (i = 0; i < fw_dump.boot_mem_regs_cnt; i++) {
978 start = fw_dump.boot_mem_addr[i];
979 end = start + fw_dump.boot_mem_sz[i];
980 ret = fadump_add_mem_range(&crash_mrange_info, start, end);
985 for_each_memblock(memory, reg) {
986 start = (u64)reg->base;
987 end = start + (u64)reg->size;
990 * skip the memory chunk that is already added
991 * (0 through boot_memory_top).
993 if (start < fw_dump.boot_mem_top) {
994 if (end > fw_dump.boot_mem_top)
995 start = fw_dump.boot_mem_top;
1000 /* add this range excluding the reserved dump area. */
1001 ret = fadump_exclude_reserved_area(start, end);
1010 * If the given physical address falls within the boot memory region then
1011 * return the relocated address that points to the dump region reserved
1012 * for saving initial boot memory contents.
1014 static inline unsigned long fadump_relocate(unsigned long paddr)
1016 unsigned long raddr, rstart, rend, rlast, hole_size;
1022 for (i = 0; i < fw_dump.boot_mem_regs_cnt; i++) {
1023 rstart = fw_dump.boot_mem_addr[i];
1024 rend = rstart + fw_dump.boot_mem_sz[i];
1025 hole_size += (rstart - rlast);
1027 if (paddr >= rstart && paddr < rend) {
1028 raddr += fw_dump.boot_mem_dest_addr - hole_size;
1035 pr_debug("vmcoreinfo: paddr = 0x%lx, raddr = 0x%lx\n", paddr, raddr);
1039 static int fadump_create_elfcore_headers(char *bufp)
1041 unsigned long long raddr, offset;
1042 struct elf_phdr *phdr;
1046 fadump_init_elfcore_header(bufp);
1047 elf = (struct elfhdr *)bufp;
1048 bufp += sizeof(struct elfhdr);
1051 * setup ELF PT_NOTE, place holder for cpu notes info. The notes info
1052 * will be populated during second kernel boot after crash. Hence
1053 * this PT_NOTE will always be the first elf note.
1055 * NOTE: Any new ELF note addition should be placed after this note.
1057 phdr = (struct elf_phdr *)bufp;
1058 bufp += sizeof(struct elf_phdr);
1059 phdr->p_type = PT_NOTE;
1071 /* setup ELF PT_NOTE for vmcoreinfo */
1072 phdr = (struct elf_phdr *)bufp;
1073 bufp += sizeof(struct elf_phdr);
1074 phdr->p_type = PT_NOTE;
1079 phdr->p_paddr = fadump_relocate(paddr_vmcoreinfo_note());
1080 phdr->p_offset = phdr->p_paddr;
1081 phdr->p_memsz = phdr->p_filesz = VMCOREINFO_NOTE_SIZE;
1083 /* Increment number of program headers. */
1086 /* setup PT_LOAD sections. */
1089 raddr = fw_dump.boot_mem_addr[0];
1090 for (i = 0; i < crash_mrange_info.mem_range_cnt; i++) {
1093 mbase = crash_mrange_info.mem_ranges[i].base;
1094 msize = crash_mrange_info.mem_ranges[i].size;
1098 phdr = (struct elf_phdr *)bufp;
1099 bufp += sizeof(struct elf_phdr);
1100 phdr->p_type = PT_LOAD;
1101 phdr->p_flags = PF_R|PF_W|PF_X;
1102 phdr->p_offset = mbase;
1104 if (mbase == raddr) {
1106 * The entire real memory region will be moved by
1107 * firmware to the specified destination_address.
1108 * Hence set the correct offset.
1110 phdr->p_offset = fw_dump.boot_mem_dest_addr + offset;
1111 if (j < (fw_dump.boot_mem_regs_cnt - 1)) {
1112 offset += fw_dump.boot_mem_sz[j];
1113 raddr = fw_dump.boot_mem_addr[++j];
1117 phdr->p_paddr = mbase;
1118 phdr->p_vaddr = (unsigned long)__va(mbase);
1119 phdr->p_filesz = msize;
1120 phdr->p_memsz = msize;
1123 /* Increment number of program headers. */
1129 static unsigned long init_fadump_header(unsigned long addr)
1131 struct fadump_crash_info_header *fdh;
1137 addr += sizeof(struct fadump_crash_info_header);
1139 memset(fdh, 0, sizeof(struct fadump_crash_info_header));
1140 fdh->magic_number = FADUMP_CRASH_INFO_MAGIC;
1141 fdh->elfcorehdr_addr = addr;
1142 /* We will set the crashing cpu id in crash_fadump() during crash. */
1143 fdh->crashing_cpu = FADUMP_CPU_UNKNOWN;
1148 static int register_fadump(void)
1155 * If no memory is reserved then we can not register for firmware-
1158 if (!fw_dump.reserve_dump_area_size)
1161 ret = fadump_setup_crash_memory_ranges();
1165 addr = fw_dump.fadumphdr_addr;
1167 /* Initialize fadump crash info header. */
1168 addr = init_fadump_header(addr);
1171 pr_debug("Creating ELF core headers at %#016lx\n", addr);
1172 fadump_create_elfcore_headers(vaddr);
1174 /* register the future kernel dump with firmware. */
1175 pr_debug("Registering for firmware-assisted kernel dump...\n");
1176 return fw_dump.ops->fadump_register(&fw_dump);
1179 void fadump_cleanup(void)
1181 if (!fw_dump.fadump_supported)
1184 /* Invalidate the registration only if dump is active. */
1185 if (fw_dump.dump_active) {
1186 pr_debug("Invalidating firmware-assisted dump registration\n");
1187 fw_dump.ops->fadump_invalidate(&fw_dump);
1188 } else if (fw_dump.dump_registered) {
1189 /* Un-register Firmware-assisted dump if it was registered. */
1190 fw_dump.ops->fadump_unregister(&fw_dump);
1191 fadump_free_mem_ranges(&crash_mrange_info);
1194 if (fw_dump.ops->fadump_cleanup)
1195 fw_dump.ops->fadump_cleanup(&fw_dump);
1198 static void fadump_free_reserved_memory(unsigned long start_pfn,
1199 unsigned long end_pfn)
1202 unsigned long time_limit = jiffies + HZ;
1204 pr_info("freeing reserved memory (0x%llx - 0x%llx)\n",
1205 PFN_PHYS(start_pfn), PFN_PHYS(end_pfn));
1207 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
1208 free_reserved_page(pfn_to_page(pfn));
1210 if (time_after(jiffies, time_limit)) {
1212 time_limit = jiffies + HZ;
1218 * Skip memory holes and free memory that was actually reserved.
1220 static void fadump_release_reserved_area(u64 start, u64 end)
1222 u64 tstart, tend, spfn, epfn;
1223 struct memblock_region *reg;
1225 spfn = PHYS_PFN(start);
1226 epfn = PHYS_PFN(end);
1227 for_each_memblock(memory, reg) {
1228 tstart = max_t(u64, spfn, memblock_region_memory_base_pfn(reg));
1229 tend = min_t(u64, epfn, memblock_region_memory_end_pfn(reg));
1230 if (tstart < tend) {
1231 fadump_free_reserved_memory(tstart, tend);
1242 * Sort the mem ranges in-place and merge adjacent ranges
1243 * to minimize the memory ranges count.
1245 static void sort_and_merge_mem_ranges(struct fadump_mrange_info *mrange_info)
1247 struct fadump_memory_range *mem_ranges;
1248 struct fadump_memory_range tmp_range;
1252 if (!reserved_mrange_info.mem_range_cnt)
1255 /* Sort the memory ranges */
1256 mem_ranges = mrange_info->mem_ranges;
1257 for (i = 0; i < mrange_info->mem_range_cnt; i++) {
1259 for (j = (i + 1); j < mrange_info->mem_range_cnt; j++) {
1260 if (mem_ranges[idx].base > mem_ranges[j].base)
1264 tmp_range = mem_ranges[idx];
1265 mem_ranges[idx] = mem_ranges[i];
1266 mem_ranges[i] = tmp_range;
1270 /* Merge adjacent reserved ranges */
1272 for (i = 1; i < mrange_info->mem_range_cnt; i++) {
1273 base = mem_ranges[i-1].base;
1274 size = mem_ranges[i-1].size;
1275 if (mem_ranges[i].base == (base + size))
1276 mem_ranges[idx].size += mem_ranges[i].size;
1282 mem_ranges[idx] = mem_ranges[i];
1285 mrange_info->mem_range_cnt = idx + 1;
1289 * Scan reserved-ranges to consider them while reserving/releasing
1290 * memory for FADump.
1292 static void __init early_init_dt_scan_reserved_ranges(unsigned long node)
1298 /* reserved-ranges already scanned */
1299 if (reserved_mrange_info.mem_range_cnt != 0)
1302 prop = of_get_flat_dt_prop(node, "reserved-ranges", &len);
1307 * Each reserved range is an (address,size) pair, 2 cells each,
1308 * totalling 4 cells per range.
1310 for (i = 0; i < len / (sizeof(*prop) * 4); i++) {
1313 base = of_read_number(prop + (i * 4) + 0, 2);
1314 size = of_read_number(prop + (i * 4) + 2, 2);
1317 ret = fadump_add_mem_range(&reserved_mrange_info,
1320 pr_warn("some reserved ranges are ignored!\n");
1326 /* Compact reserved ranges */
1327 sort_and_merge_mem_ranges(&reserved_mrange_info);
1331 * Release the memory that was reserved during early boot to preserve the
1332 * crash'ed kernel's memory contents except reserved dump area (permanent
1333 * reservation) and reserved ranges used by F/W. The released memory will
1334 * be available for general use.
1336 static void fadump_release_memory(u64 begin, u64 end)
1338 u64 ra_start, ra_end, tstart;
1341 ra_start = fw_dump.reserve_dump_area_start;
1342 ra_end = ra_start + fw_dump.reserve_dump_area_size;
1345 * If reserved ranges array limit is hit, overwrite the last reserved
1346 * memory range with reserved dump area to ensure it is excluded from
1347 * the memory being released (reused for next FADump registration).
1349 if (reserved_mrange_info.mem_range_cnt ==
1350 reserved_mrange_info.max_mem_ranges)
1351 reserved_mrange_info.mem_range_cnt--;
1353 ret = fadump_add_mem_range(&reserved_mrange_info, ra_start, ra_end);
1357 /* Get the reserved ranges list in order first. */
1358 sort_and_merge_mem_ranges(&reserved_mrange_info);
1360 /* Exclude reserved ranges and release remaining memory */
1362 for (i = 0; i < reserved_mrange_info.mem_range_cnt; i++) {
1363 ra_start = reserved_mrange_info.mem_ranges[i].base;
1364 ra_end = ra_start + reserved_mrange_info.mem_ranges[i].size;
1366 if (tstart >= ra_end)
1369 if (tstart < ra_start)
1370 fadump_release_reserved_area(tstart, ra_start);
1375 fadump_release_reserved_area(tstart, end);
1378 static void fadump_invalidate_release_mem(void)
1380 mutex_lock(&fadump_mutex);
1381 if (!fw_dump.dump_active) {
1382 mutex_unlock(&fadump_mutex);
1387 mutex_unlock(&fadump_mutex);
1389 fadump_release_memory(fw_dump.boot_mem_top, memblock_end_of_DRAM());
1390 fadump_free_cpu_notes_buf();
1393 * Setup kernel metadata and initialize the kernel dump
1394 * memory structure for FADump re-registration.
1396 if (fw_dump.ops->fadump_setup_metadata &&
1397 (fw_dump.ops->fadump_setup_metadata(&fw_dump) < 0))
1398 pr_warn("Failed to setup kernel metadata!\n");
1399 fw_dump.ops->fadump_init_mem_struct(&fw_dump);
1402 static ssize_t fadump_release_memory_store(struct kobject *kobj,
1403 struct kobj_attribute *attr,
1404 const char *buf, size_t count)
1408 if (!fw_dump.dump_active)
1411 if (kstrtoint(buf, 0, &input))
1416 * Take away the '/proc/vmcore'. We are releasing the dump
1417 * memory, hence it will not be valid anymore.
1419 #ifdef CONFIG_PROC_VMCORE
1422 fadump_invalidate_release_mem();
1429 static ssize_t fadump_enabled_show(struct kobject *kobj,
1430 struct kobj_attribute *attr,
1433 return sprintf(buf, "%d\n", fw_dump.fadump_enabled);
1436 static ssize_t fadump_register_show(struct kobject *kobj,
1437 struct kobj_attribute *attr,
1440 return sprintf(buf, "%d\n", fw_dump.dump_registered);
1443 static ssize_t fadump_register_store(struct kobject *kobj,
1444 struct kobj_attribute *attr,
1445 const char *buf, size_t count)
1450 if (!fw_dump.fadump_enabled || fw_dump.dump_active)
1453 if (kstrtoint(buf, 0, &input))
1456 mutex_lock(&fadump_mutex);
1460 if (fw_dump.dump_registered == 0) {
1464 /* Un-register Firmware-assisted dump */
1465 pr_debug("Un-register firmware-assisted dump\n");
1466 fw_dump.ops->fadump_unregister(&fw_dump);
1469 if (fw_dump.dump_registered == 1) {
1470 /* Un-register Firmware-assisted dump */
1471 fw_dump.ops->fadump_unregister(&fw_dump);
1473 /* Register Firmware-assisted dump */
1474 ret = register_fadump();
1482 mutex_unlock(&fadump_mutex);
1483 return ret < 0 ? ret : count;
1486 static int fadump_region_show(struct seq_file *m, void *private)
1488 if (!fw_dump.fadump_enabled)
1491 mutex_lock(&fadump_mutex);
1492 fw_dump.ops->fadump_region_show(&fw_dump, m);
1493 mutex_unlock(&fadump_mutex);
1497 static struct kobj_attribute fadump_release_attr = __ATTR(fadump_release_mem,
1499 fadump_release_memory_store);
1500 static struct kobj_attribute fadump_attr = __ATTR(fadump_enabled,
1501 0444, fadump_enabled_show,
1503 static struct kobj_attribute fadump_register_attr = __ATTR(fadump_registered,
1504 0644, fadump_register_show,
1505 fadump_register_store);
1507 DEFINE_SHOW_ATTRIBUTE(fadump_region);
1509 static void fadump_init_files(void)
1511 struct dentry *debugfs_file;
1514 rc = sysfs_create_file(kernel_kobj, &fadump_attr.attr);
1516 printk(KERN_ERR "fadump: unable to create sysfs file"
1517 " fadump_enabled (%d)\n", rc);
1519 rc = sysfs_create_file(kernel_kobj, &fadump_register_attr.attr);
1521 printk(KERN_ERR "fadump: unable to create sysfs file"
1522 " fadump_registered (%d)\n", rc);
1524 debugfs_file = debugfs_create_file("fadump_region", 0444,
1525 powerpc_debugfs_root, NULL,
1526 &fadump_region_fops);
1528 printk(KERN_ERR "fadump: unable to create debugfs file"
1529 " fadump_region\n");
1531 if (fw_dump.dump_active) {
1532 rc = sysfs_create_file(kernel_kobj, &fadump_release_attr.attr);
1534 printk(KERN_ERR "fadump: unable to create sysfs file"
1535 " fadump_release_mem (%d)\n", rc);
1541 * Prepare for firmware-assisted dump.
1543 int __init setup_fadump(void)
1545 if (!fw_dump.fadump_enabled)
1548 if (!fw_dump.fadump_supported) {
1549 printk(KERN_ERR "Firmware-assisted dump is not supported on"
1550 " this hardware\n");
1554 fadump_show_config();
1556 * If dump data is available then see if it is valid and prepare for
1557 * saving it to the disk.
1559 if (fw_dump.dump_active) {
1561 * if dump process fails then invalidate the registration
1562 * and release memory before proceeding for re-registration.
1564 if (fw_dump.ops->fadump_process(&fw_dump) < 0)
1565 fadump_invalidate_release_mem();
1567 /* Initialize the kernel dump memory structure for FAD registration. */
1568 else if (fw_dump.reserve_dump_area_size)
1569 fw_dump.ops->fadump_init_mem_struct(&fw_dump);
1571 fadump_init_files();
1575 subsys_initcall(setup_fadump);
1576 #else /* !CONFIG_PRESERVE_FA_DUMP */
1578 /* Scan the Firmware Assisted dump configuration details. */
1579 int __init early_init_dt_scan_fw_dump(unsigned long node, const char *uname,
1580 int depth, void *data)
1582 if ((depth != 1) || (strcmp(uname, "ibm,opal") != 0))
1585 opal_fadump_dt_scan(&fw_dump, node);
1590 * When dump is active but PRESERVE_FA_DUMP is enabled on the kernel,
1591 * preserve crash data. The subsequent memory preserving kernel boot
1592 * is likely to process this crash data.
1594 int __init fadump_reserve_mem(void)
1596 if (fw_dump.dump_active) {
1598 * If last boot has crashed then reserve all the memory
1599 * above boot memory to preserve crash data.
1601 pr_info("Preserving crash data for processing in next boot.\n");
1602 fadump_reserve_crash_area(fw_dump.boot_mem_top);
1604 pr_debug("FADump-aware kernel..\n");
1608 #endif /* CONFIG_PRESERVE_FA_DUMP */
1610 /* Preserve everything above the base address */
1611 static void __init fadump_reserve_crash_area(u64 base)
1613 struct memblock_region *reg;
1616 for_each_memblock(memory, reg) {
1620 if ((mstart + msize) < base)
1623 if (mstart < base) {
1624 msize -= (base - mstart);
1628 pr_info("Reserving %lluMB of memory at %#016llx for preserving crash data",
1629 (msize >> 20), mstart);
1630 memblock_reserve(mstart, msize);
1634 unsigned long __init arch_reserved_kernel_pages(void)
1636 return memblock_reserved_size() / PAGE_SIZE;