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 struct kobject *fadump_kobj;
41 #ifndef CONFIG_PRESERVE_FA_DUMP
42 static DEFINE_MUTEX(fadump_mutex);
43 struct fadump_mrange_info crash_mrange_info = { "crash", NULL, 0, 0, 0, false };
45 #define RESERVED_RNGS_SZ 16384 /* 16K - 128 entries */
46 #define RESERVED_RNGS_CNT (RESERVED_RNGS_SZ / \
47 sizeof(struct fadump_memory_range))
48 static struct fadump_memory_range rngs[RESERVED_RNGS_CNT];
49 struct fadump_mrange_info reserved_mrange_info = { "reserved", rngs,
51 RESERVED_RNGS_CNT, true };
53 static void __init early_init_dt_scan_reserved_ranges(unsigned long node);
56 static struct cma *fadump_cma;
59 * fadump_cma_init() - Initialize CMA area from a fadump reserved memory
61 * This function initializes CMA area from fadump reserved memory.
62 * The total size of fadump reserved memory covers for boot memory size
63 * + cpu data size + hpte size and metadata.
64 * Initialize only the area equivalent to boot memory size for CMA use.
65 * The reamining portion of fadump reserved memory will be not given
66 * to CMA and pages for thoes will stay reserved. boot memory size is
67 * aligned per CMA requirement to satisy cma_init_reserved_mem() call.
68 * But for some reason even if it fails we still have the memory reservation
69 * with us and we can still continue doing fadump.
71 int __init fadump_cma_init(void)
73 unsigned long long base, size;
76 if (!fw_dump.fadump_enabled)
80 * Do not use CMA if user has provided fadump=nocma kernel parameter.
81 * Return 1 to continue with fadump old behaviour.
86 base = fw_dump.reserve_dump_area_start;
87 size = fw_dump.boot_memory_size;
92 rc = cma_init_reserved_mem(base, size, 0, "fadump_cma", &fadump_cma);
94 pr_err("Failed to init cma area for firmware-assisted dump,%d\n", rc);
96 * Though the CMA init has failed we still have memory
97 * reservation with us. The reserved memory will be
98 * blocked from production system usage. Hence return 1,
99 * so that we can continue with fadump.
105 * So we now have successfully initialized cma area for fadump.
107 pr_info("Initialized 0x%lx bytes cma area at %ldMB from 0x%lx "
108 "bytes of memory reserved for firmware-assisted dump\n",
109 cma_get_size(fadump_cma),
110 (unsigned long)cma_get_base(fadump_cma) >> 20,
111 fw_dump.reserve_dump_area_size);
115 static int __init fadump_cma_init(void) { return 1; }
116 #endif /* CONFIG_CMA */
118 /* Scan the Firmware Assisted dump configuration details. */
119 int __init early_init_dt_scan_fw_dump(unsigned long node, const char *uname,
120 int depth, void *data)
123 early_init_dt_scan_reserved_ranges(node);
130 if (strcmp(uname, "rtas") == 0) {
131 rtas_fadump_dt_scan(&fw_dump, node);
135 if (strcmp(uname, "ibm,opal") == 0) {
136 opal_fadump_dt_scan(&fw_dump, node);
144 * If fadump is registered, check if the memory provided
145 * falls within boot memory area and reserved memory area.
147 int is_fadump_memory_area(u64 addr, unsigned long size)
151 if (!fw_dump.dump_registered)
157 d_start = fw_dump.reserve_dump_area_start;
158 d_end = d_start + fw_dump.reserve_dump_area_size;
159 if (((addr + size) > d_start) && (addr <= d_end))
162 return (addr <= fw_dump.boot_mem_top);
165 int should_fadump_crash(void)
167 if (!fw_dump.dump_registered || !fw_dump.fadumphdr_addr)
172 int is_fadump_active(void)
174 return fw_dump.dump_active;
178 * Returns true, if there are no holes in memory area between d_start to d_end,
181 static bool is_fadump_mem_area_contiguous(u64 d_start, u64 d_end)
183 struct memblock_region *reg;
187 for_each_memblock(memory, reg) {
188 start = max_t(u64, d_start, reg->base);
189 end = min_t(u64, d_end, (reg->base + reg->size));
191 /* Memory hole from d_start to start */
208 * Returns true, if there are no holes in boot memory area,
211 bool is_fadump_boot_mem_contiguous(void)
213 unsigned long d_start, d_end;
217 for (i = 0; i < fw_dump.boot_mem_regs_cnt; i++) {
218 d_start = fw_dump.boot_mem_addr[i];
219 d_end = d_start + fw_dump.boot_mem_sz[i];
221 ret = is_fadump_mem_area_contiguous(d_start, d_end);
230 * Returns true, if there are no holes in reserved memory area,
233 bool is_fadump_reserved_mem_contiguous(void)
237 d_start = fw_dump.reserve_dump_area_start;
238 d_end = d_start + fw_dump.reserve_dump_area_size;
239 return is_fadump_mem_area_contiguous(d_start, d_end);
242 /* Print firmware assisted dump configurations for debugging purpose. */
243 static void fadump_show_config(void)
247 pr_debug("Support for firmware-assisted dump (fadump): %s\n",
248 (fw_dump.fadump_supported ? "present" : "no support"));
250 if (!fw_dump.fadump_supported)
253 pr_debug("Fadump enabled : %s\n",
254 (fw_dump.fadump_enabled ? "yes" : "no"));
255 pr_debug("Dump Active : %s\n",
256 (fw_dump.dump_active ? "yes" : "no"));
257 pr_debug("Dump section sizes:\n");
258 pr_debug(" CPU state data size: %lx\n", fw_dump.cpu_state_data_size);
259 pr_debug(" HPTE region size : %lx\n", fw_dump.hpte_region_size);
260 pr_debug(" Boot memory size : %lx\n", fw_dump.boot_memory_size);
261 pr_debug(" Boot memory top : %llx\n", fw_dump.boot_mem_top);
262 pr_debug("Boot memory regions cnt: %llx\n", fw_dump.boot_mem_regs_cnt);
263 for (i = 0; i < fw_dump.boot_mem_regs_cnt; i++) {
264 pr_debug("[%03d] base = %llx, size = %llx\n", i,
265 fw_dump.boot_mem_addr[i], fw_dump.boot_mem_sz[i]);
270 * fadump_calculate_reserve_size(): reserve variable boot area 5% of System RAM
272 * Function to find the largest memory size we need to reserve during early
273 * boot process. This will be the size of the memory that is required for a
274 * kernel to boot successfully.
276 * This function has been taken from phyp-assisted dump feature implementation.
278 * returns larger of 256MB or 5% rounded down to multiples of 256MB.
280 * TODO: Come up with better approach to find out more accurate memory size
281 * that is required for a kernel to boot successfully.
284 static inline u64 fadump_calculate_reserve_size(void)
286 u64 base, size, bootmem_min;
289 if (fw_dump.reserve_bootvar)
290 pr_warn("'fadump_reserve_mem=' parameter is deprecated in favor of 'crashkernel=' parameter.\n");
293 * Check if the size is specified through crashkernel= cmdline
294 * option. If yes, then use that but ignore base as fadump reserves
295 * memory at a predefined offset.
297 ret = parse_crashkernel(boot_command_line, memblock_phys_mem_size(),
299 if (ret == 0 && size > 0) {
300 unsigned long max_size;
302 if (fw_dump.reserve_bootvar)
303 pr_info("Using 'crashkernel=' parameter for memory reservation.\n");
305 fw_dump.reserve_bootvar = (unsigned long)size;
308 * Adjust if the boot memory size specified is above
311 max_size = memblock_phys_mem_size() / MAX_BOOT_MEM_RATIO;
312 if (fw_dump.reserve_bootvar > max_size) {
313 fw_dump.reserve_bootvar = max_size;
314 pr_info("Adjusted boot memory size to %luMB\n",
315 (fw_dump.reserve_bootvar >> 20));
318 return fw_dump.reserve_bootvar;
319 } else if (fw_dump.reserve_bootvar) {
321 * 'fadump_reserve_mem=' is being used to reserve memory
322 * for firmware-assisted dump.
324 return fw_dump.reserve_bootvar;
327 /* divide by 20 to get 5% of value */
328 size = memblock_phys_mem_size() / 20;
330 /* round it down in multiples of 256 */
331 size = size & ~0x0FFFFFFFUL;
333 /* Truncate to memory_limit. We don't want to over reserve the memory.*/
334 if (memory_limit && size > memory_limit)
337 bootmem_min = fw_dump.ops->fadump_get_bootmem_min();
338 return (size > bootmem_min ? size : bootmem_min);
342 * Calculate the total memory size required to be reserved for
343 * firmware-assisted dump registration.
345 static unsigned long get_fadump_area_size(void)
347 unsigned long size = 0;
349 size += fw_dump.cpu_state_data_size;
350 size += fw_dump.hpte_region_size;
351 size += fw_dump.boot_memory_size;
352 size += sizeof(struct fadump_crash_info_header);
353 size += sizeof(struct elfhdr); /* ELF core header.*/
354 size += sizeof(struct elf_phdr); /* place holder for cpu notes */
355 /* Program headers for crash memory regions. */
356 size += sizeof(struct elf_phdr) * (memblock_num_regions(memory) + 2);
358 size = PAGE_ALIGN(size);
360 /* This is to hold kernel metadata on platforms that support it */
361 size += (fw_dump.ops->fadump_get_metadata_size ?
362 fw_dump.ops->fadump_get_metadata_size() : 0);
366 static int __init add_boot_mem_region(unsigned long rstart,
369 int i = fw_dump.boot_mem_regs_cnt++;
371 if (fw_dump.boot_mem_regs_cnt > FADUMP_MAX_MEM_REGS) {
372 fw_dump.boot_mem_regs_cnt = FADUMP_MAX_MEM_REGS;
376 pr_debug("Added boot memory range[%d] [%#016lx-%#016lx)\n",
377 i, rstart, (rstart + rsize));
378 fw_dump.boot_mem_addr[i] = rstart;
379 fw_dump.boot_mem_sz[i] = rsize;
384 * Firmware usually has a hard limit on the data it can copy per region.
385 * Honour that by splitting a memory range into multiple regions.
387 static int __init add_boot_mem_regions(unsigned long mstart,
390 unsigned long rstart, rsize, max_size;
394 max_size = fw_dump.max_copy_size ? fw_dump.max_copy_size : msize;
396 if (msize > max_size)
401 ret = add_boot_mem_region(rstart, rsize);
412 static int __init fadump_get_boot_mem_regions(void)
414 unsigned long base, size, cur_size, hole_size, last_end;
415 unsigned long mem_size = fw_dump.boot_memory_size;
416 struct memblock_region *reg;
419 fw_dump.boot_mem_regs_cnt = 0;
424 for_each_memblock(memory, reg) {
427 hole_size += (base - last_end);
429 if ((cur_size + size) >= mem_size) {
430 size = (mem_size - cur_size);
431 ret = add_boot_mem_regions(base, size);
437 ret = add_boot_mem_regions(base, size);
441 last_end = base + size;
443 fw_dump.boot_mem_top = PAGE_ALIGN(fw_dump.boot_memory_size + hole_size);
449 * Returns true, if the given range overlaps with reserved memory ranges
450 * starting at idx. Also, updates idx to index of overlapping memory range
451 * with the given memory range.
454 static bool overlaps_reserved_ranges(u64 base, u64 end, int *idx)
459 for (i = *idx; i < reserved_mrange_info.mem_range_cnt; i++) {
460 u64 rbase = reserved_mrange_info.mem_ranges[i].base;
461 u64 rend = rbase + reserved_mrange_info.mem_ranges[i].size;
466 if ((end > rbase) && (base < rend)) {
477 * Locate a suitable memory area to reserve memory for FADump. While at it,
478 * lookup reserved-ranges & avoid overlap with them, as they are used by F/W.
480 static u64 __init fadump_locate_reserve_mem(u64 base, u64 size)
482 struct fadump_memory_range *mrngs;
483 phys_addr_t mstart, mend;
487 mrngs = reserved_mrange_info.mem_ranges;
488 for_each_free_mem_range(i, NUMA_NO_NODE, MEMBLOCK_NONE,
489 &mstart, &mend, NULL) {
490 pr_debug("%llu) mstart: %llx, mend: %llx, base: %llx\n",
491 i, mstart, mend, base);
494 base = PAGE_ALIGN(mstart);
496 while ((mend > base) && ((mend - base) >= size)) {
497 if (!overlaps_reserved_ranges(base, base+size, &idx)) {
502 base = mrngs[idx].base + mrngs[idx].size;
503 base = PAGE_ALIGN(base);
511 int __init fadump_reserve_mem(void)
513 u64 base, size, mem_boundary, bootmem_min;
516 if (!fw_dump.fadump_enabled)
519 if (!fw_dump.fadump_supported) {
520 pr_info("Firmware-Assisted Dump is not supported on this hardware\n");
525 * Initialize boot memory size
526 * If dump is active then we have already calculated the size during
529 if (!fw_dump.dump_active) {
530 fw_dump.boot_memory_size =
531 PAGE_ALIGN(fadump_calculate_reserve_size());
533 if (!fw_dump.nocma) {
534 fw_dump.boot_memory_size =
535 ALIGN(fw_dump.boot_memory_size,
536 FADUMP_CMA_ALIGNMENT);
540 bootmem_min = fw_dump.ops->fadump_get_bootmem_min();
541 if (fw_dump.boot_memory_size < bootmem_min) {
542 pr_err("Can't enable fadump with boot memory size (0x%lx) less than 0x%llx\n",
543 fw_dump.boot_memory_size, bootmem_min);
547 if (!fadump_get_boot_mem_regions()) {
548 pr_err("Too many holes in boot memory area to enable fadump\n");
554 * Calculate the memory boundary.
555 * If memory_limit is less than actual memory boundary then reserve
556 * the memory for fadump beyond the memory_limit and adjust the
557 * memory_limit accordingly, so that the running kernel can run with
558 * specified memory_limit.
560 if (memory_limit && memory_limit < memblock_end_of_DRAM()) {
561 size = get_fadump_area_size();
562 if ((memory_limit + size) < memblock_end_of_DRAM())
563 memory_limit += size;
565 memory_limit = memblock_end_of_DRAM();
566 printk(KERN_INFO "Adjusted memory_limit for firmware-assisted"
567 " dump, now %#016llx\n", memory_limit);
570 mem_boundary = memory_limit;
572 mem_boundary = memblock_end_of_DRAM();
574 base = fw_dump.boot_mem_top;
575 size = get_fadump_area_size();
576 fw_dump.reserve_dump_area_size = size;
577 if (fw_dump.dump_active) {
578 pr_info("Firmware-assisted dump is active.\n");
580 #ifdef CONFIG_HUGETLB_PAGE
582 * FADump capture kernel doesn't care much about hugepages.
583 * In fact, handling hugepages in capture kernel is asking for
584 * trouble. So, disable HugeTLB support when fadump is active.
586 hugetlb_disabled = true;
589 * If last boot has crashed then reserve all the memory
590 * above boot memory size so that we don't touch it until
591 * dump is written to disk by userspace tool. This memory
592 * can be released for general use by invalidating fadump.
594 fadump_reserve_crash_area(base);
596 pr_debug("fadumphdr_addr = %#016lx\n", fw_dump.fadumphdr_addr);
597 pr_debug("Reserve dump area start address: 0x%lx\n",
598 fw_dump.reserve_dump_area_start);
601 * Reserve memory at an offset closer to bottom of the RAM to
602 * minimize the impact of memory hot-remove operation.
604 base = fadump_locate_reserve_mem(base, size);
606 if (!base || (base + size > mem_boundary)) {
607 pr_err("Failed to find memory chunk for reservation!\n");
610 fw_dump.reserve_dump_area_start = base;
613 * Calculate the kernel metadata address and register it with
614 * f/w if the platform supports.
616 if (fw_dump.ops->fadump_setup_metadata &&
617 (fw_dump.ops->fadump_setup_metadata(&fw_dump) < 0))
620 if (memblock_reserve(base, size)) {
621 pr_err("Failed to reserve memory!\n");
625 pr_info("Reserved %lldMB of memory at %#016llx (System RAM: %lldMB)\n",
626 (size >> 20), base, (memblock_phys_mem_size() >> 20));
628 ret = fadump_cma_init();
633 fw_dump.fadump_enabled = 0;
637 /* Look for fadump= cmdline option. */
638 static int __init early_fadump_param(char *p)
643 if (strncmp(p, "on", 2) == 0)
644 fw_dump.fadump_enabled = 1;
645 else if (strncmp(p, "off", 3) == 0)
646 fw_dump.fadump_enabled = 0;
647 else if (strncmp(p, "nocma", 5) == 0) {
648 fw_dump.fadump_enabled = 1;
654 early_param("fadump", early_fadump_param);
657 * Look for fadump_reserve_mem= cmdline option
658 * TODO: Remove references to 'fadump_reserve_mem=' parameter,
659 * the sooner 'crashkernel=' parameter is accustomed to.
661 static int __init early_fadump_reserve_mem(char *p)
664 fw_dump.reserve_bootvar = memparse(p, &p);
667 early_param("fadump_reserve_mem", early_fadump_reserve_mem);
669 void crash_fadump(struct pt_regs *regs, const char *str)
671 struct fadump_crash_info_header *fdh = NULL;
672 int old_cpu, this_cpu;
674 if (!should_fadump_crash())
678 * old_cpu == -1 means this is the first CPU which has come here,
679 * go ahead and trigger fadump.
681 * old_cpu != -1 means some other CPU has already on it's way
682 * to trigger fadump, just keep looping here.
684 this_cpu = smp_processor_id();
685 old_cpu = cmpxchg(&crashing_cpu, -1, this_cpu);
689 * We can't loop here indefinitely. Wait as long as fadump
690 * is in force. If we race with fadump un-registration this
691 * loop will break and then we go down to normal panic path
692 * and reboot. If fadump is in force the first crashing
693 * cpu will definitely trigger fadump.
695 while (fw_dump.dump_registered)
700 fdh = __va(fw_dump.fadumphdr_addr);
701 fdh->crashing_cpu = crashing_cpu;
702 crash_save_vmcoreinfo();
707 ppc_save_regs(&fdh->regs);
709 fdh->online_mask = *cpu_online_mask;
711 fw_dump.ops->fadump_trigger(fdh, str);
714 u32 *fadump_regs_to_elf_notes(u32 *buf, struct pt_regs *regs)
716 struct elf_prstatus prstatus;
718 memset(&prstatus, 0, sizeof(prstatus));
720 * FIXME: How do i get PID? Do I really need it?
721 * prstatus.pr_pid = ????
723 elf_core_copy_kernel_regs(&prstatus.pr_reg, regs);
724 buf = append_elf_note(buf, CRASH_CORE_NOTE_NAME, NT_PRSTATUS,
725 &prstatus, sizeof(prstatus));
729 void fadump_update_elfcore_header(char *bufp)
732 struct elf_phdr *phdr;
734 elf = (struct elfhdr *)bufp;
735 bufp += sizeof(struct elfhdr);
737 /* First note is a place holder for cpu notes info. */
738 phdr = (struct elf_phdr *)bufp;
740 if (phdr->p_type == PT_NOTE) {
741 phdr->p_paddr = __pa(fw_dump.cpu_notes_buf_vaddr);
742 phdr->p_offset = phdr->p_paddr;
743 phdr->p_filesz = fw_dump.cpu_notes_buf_size;
744 phdr->p_memsz = fw_dump.cpu_notes_buf_size;
749 static void *fadump_alloc_buffer(unsigned long size)
751 unsigned long count, i;
755 vaddr = alloc_pages_exact(size, GFP_KERNEL | __GFP_ZERO);
759 count = PAGE_ALIGN(size) / PAGE_SIZE;
760 page = virt_to_page(vaddr);
761 for (i = 0; i < count; i++)
762 mark_page_reserved(page + i);
766 static void fadump_free_buffer(unsigned long vaddr, unsigned long size)
768 free_reserved_area((void *)vaddr, (void *)(vaddr + size), -1, NULL);
771 s32 fadump_setup_cpu_notes_buf(u32 num_cpus)
773 /* Allocate buffer to hold cpu crash notes. */
774 fw_dump.cpu_notes_buf_size = num_cpus * sizeof(note_buf_t);
775 fw_dump.cpu_notes_buf_size = PAGE_ALIGN(fw_dump.cpu_notes_buf_size);
776 fw_dump.cpu_notes_buf_vaddr =
777 (unsigned long)fadump_alloc_buffer(fw_dump.cpu_notes_buf_size);
778 if (!fw_dump.cpu_notes_buf_vaddr) {
779 pr_err("Failed to allocate %ld bytes for CPU notes buffer\n",
780 fw_dump.cpu_notes_buf_size);
784 pr_debug("Allocated buffer for cpu notes of size %ld at 0x%lx\n",
785 fw_dump.cpu_notes_buf_size,
786 fw_dump.cpu_notes_buf_vaddr);
790 void fadump_free_cpu_notes_buf(void)
792 if (!fw_dump.cpu_notes_buf_vaddr)
795 fadump_free_buffer(fw_dump.cpu_notes_buf_vaddr,
796 fw_dump.cpu_notes_buf_size);
797 fw_dump.cpu_notes_buf_vaddr = 0;
798 fw_dump.cpu_notes_buf_size = 0;
801 static void fadump_free_mem_ranges(struct fadump_mrange_info *mrange_info)
803 if (mrange_info->is_static) {
804 mrange_info->mem_range_cnt = 0;
808 kfree(mrange_info->mem_ranges);
809 memset((void *)((u64)mrange_info + RNG_NAME_SZ), 0,
810 (sizeof(struct fadump_mrange_info) - RNG_NAME_SZ));
814 * Allocate or reallocate mem_ranges array in incremental units
817 static int fadump_alloc_mem_ranges(struct fadump_mrange_info *mrange_info)
819 struct fadump_memory_range *new_array;
822 new_size = mrange_info->mem_ranges_sz + PAGE_SIZE;
823 pr_debug("Allocating %llu bytes of memory for %s memory ranges\n",
824 new_size, mrange_info->name);
826 new_array = krealloc(mrange_info->mem_ranges, new_size, GFP_KERNEL);
827 if (new_array == NULL) {
828 pr_err("Insufficient memory for setting up %s memory ranges\n",
830 fadump_free_mem_ranges(mrange_info);
834 mrange_info->mem_ranges = new_array;
835 mrange_info->mem_ranges_sz = new_size;
836 mrange_info->max_mem_ranges = (new_size /
837 sizeof(struct fadump_memory_range));
841 static inline int fadump_add_mem_range(struct fadump_mrange_info *mrange_info,
844 struct fadump_memory_range *mem_ranges = mrange_info->mem_ranges;
845 bool is_adjacent = false;
852 * Fold adjacent memory ranges to bring down the memory ranges/
853 * PT_LOAD segments count.
855 if (mrange_info->mem_range_cnt) {
856 start = mem_ranges[mrange_info->mem_range_cnt - 1].base;
857 size = mem_ranges[mrange_info->mem_range_cnt - 1].size;
859 if ((start + size) == base)
863 /* resize the array on reaching the limit */
864 if (mrange_info->mem_range_cnt == mrange_info->max_mem_ranges) {
867 if (mrange_info->is_static) {
868 pr_err("Reached array size limit for %s memory ranges\n",
873 ret = fadump_alloc_mem_ranges(mrange_info);
877 /* Update to the new resized array */
878 mem_ranges = mrange_info->mem_ranges;
882 mem_ranges[mrange_info->mem_range_cnt].base = start;
883 mrange_info->mem_range_cnt++;
886 mem_ranges[mrange_info->mem_range_cnt - 1].size = (end - start);
887 pr_debug("%s_memory_range[%d] [%#016llx-%#016llx], %#llx bytes\n",
888 mrange_info->name, (mrange_info->mem_range_cnt - 1),
889 start, end - 1, (end - start));
893 static int fadump_exclude_reserved_area(u64 start, u64 end)
895 u64 ra_start, ra_end;
898 ra_start = fw_dump.reserve_dump_area_start;
899 ra_end = ra_start + fw_dump.reserve_dump_area_size;
901 if ((ra_start < end) && (ra_end > start)) {
902 if ((start < ra_start) && (end > ra_end)) {
903 ret = fadump_add_mem_range(&crash_mrange_info,
908 ret = fadump_add_mem_range(&crash_mrange_info,
910 } else if (start < ra_start) {
911 ret = fadump_add_mem_range(&crash_mrange_info,
913 } else if (ra_end < end) {
914 ret = fadump_add_mem_range(&crash_mrange_info,
918 ret = fadump_add_mem_range(&crash_mrange_info, start, end);
923 static int fadump_init_elfcore_header(char *bufp)
927 elf = (struct elfhdr *) bufp;
928 bufp += sizeof(struct elfhdr);
929 memcpy(elf->e_ident, ELFMAG, SELFMAG);
930 elf->e_ident[EI_CLASS] = ELF_CLASS;
931 elf->e_ident[EI_DATA] = ELF_DATA;
932 elf->e_ident[EI_VERSION] = EV_CURRENT;
933 elf->e_ident[EI_OSABI] = ELF_OSABI;
934 memset(elf->e_ident+EI_PAD, 0, EI_NIDENT-EI_PAD);
935 elf->e_type = ET_CORE;
936 elf->e_machine = ELF_ARCH;
937 elf->e_version = EV_CURRENT;
939 elf->e_phoff = sizeof(struct elfhdr);
941 #if defined(_CALL_ELF)
942 elf->e_flags = _CALL_ELF;
946 elf->e_ehsize = sizeof(struct elfhdr);
947 elf->e_phentsize = sizeof(struct elf_phdr);
949 elf->e_shentsize = 0;
957 * Traverse through memblock structure and setup crash memory ranges. These
958 * ranges will be used create PT_LOAD program headers in elfcore header.
960 static int fadump_setup_crash_memory_ranges(void)
962 struct memblock_region *reg;
966 pr_debug("Setup crash memory ranges.\n");
967 crash_mrange_info.mem_range_cnt = 0;
970 * Boot memory region(s) registered with firmware are moved to
971 * different location at the time of crash. Create separate program
972 * header(s) for this memory chunk(s) with the correct offset.
974 for (i = 0; i < fw_dump.boot_mem_regs_cnt; i++) {
975 start = fw_dump.boot_mem_addr[i];
976 end = start + fw_dump.boot_mem_sz[i];
977 ret = fadump_add_mem_range(&crash_mrange_info, start, end);
982 for_each_memblock(memory, reg) {
983 start = (u64)reg->base;
984 end = start + (u64)reg->size;
987 * skip the memory chunk that is already added
988 * (0 through boot_memory_top).
990 if (start < fw_dump.boot_mem_top) {
991 if (end > fw_dump.boot_mem_top)
992 start = fw_dump.boot_mem_top;
997 /* add this range excluding the reserved dump area. */
998 ret = fadump_exclude_reserved_area(start, end);
1007 * If the given physical address falls within the boot memory region then
1008 * return the relocated address that points to the dump region reserved
1009 * for saving initial boot memory contents.
1011 static inline unsigned long fadump_relocate(unsigned long paddr)
1013 unsigned long raddr, rstart, rend, rlast, hole_size;
1019 for (i = 0; i < fw_dump.boot_mem_regs_cnt; i++) {
1020 rstart = fw_dump.boot_mem_addr[i];
1021 rend = rstart + fw_dump.boot_mem_sz[i];
1022 hole_size += (rstart - rlast);
1024 if (paddr >= rstart && paddr < rend) {
1025 raddr += fw_dump.boot_mem_dest_addr - hole_size;
1032 pr_debug("vmcoreinfo: paddr = 0x%lx, raddr = 0x%lx\n", paddr, raddr);
1036 static int fadump_create_elfcore_headers(char *bufp)
1038 unsigned long long raddr, offset;
1039 struct elf_phdr *phdr;
1043 fadump_init_elfcore_header(bufp);
1044 elf = (struct elfhdr *)bufp;
1045 bufp += sizeof(struct elfhdr);
1048 * setup ELF PT_NOTE, place holder for cpu notes info. The notes info
1049 * will be populated during second kernel boot after crash. Hence
1050 * this PT_NOTE will always be the first elf note.
1052 * NOTE: Any new ELF note addition should be placed after this note.
1054 phdr = (struct elf_phdr *)bufp;
1055 bufp += sizeof(struct elf_phdr);
1056 phdr->p_type = PT_NOTE;
1068 /* setup ELF PT_NOTE for vmcoreinfo */
1069 phdr = (struct elf_phdr *)bufp;
1070 bufp += sizeof(struct elf_phdr);
1071 phdr->p_type = PT_NOTE;
1076 phdr->p_paddr = fadump_relocate(paddr_vmcoreinfo_note());
1077 phdr->p_offset = phdr->p_paddr;
1078 phdr->p_memsz = phdr->p_filesz = VMCOREINFO_NOTE_SIZE;
1080 /* Increment number of program headers. */
1083 /* setup PT_LOAD sections. */
1086 raddr = fw_dump.boot_mem_addr[0];
1087 for (i = 0; i < crash_mrange_info.mem_range_cnt; i++) {
1090 mbase = crash_mrange_info.mem_ranges[i].base;
1091 msize = crash_mrange_info.mem_ranges[i].size;
1095 phdr = (struct elf_phdr *)bufp;
1096 bufp += sizeof(struct elf_phdr);
1097 phdr->p_type = PT_LOAD;
1098 phdr->p_flags = PF_R|PF_W|PF_X;
1099 phdr->p_offset = mbase;
1101 if (mbase == raddr) {
1103 * The entire real memory region will be moved by
1104 * firmware to the specified destination_address.
1105 * Hence set the correct offset.
1107 phdr->p_offset = fw_dump.boot_mem_dest_addr + offset;
1108 if (j < (fw_dump.boot_mem_regs_cnt - 1)) {
1109 offset += fw_dump.boot_mem_sz[j];
1110 raddr = fw_dump.boot_mem_addr[++j];
1114 phdr->p_paddr = mbase;
1115 phdr->p_vaddr = (unsigned long)__va(mbase);
1116 phdr->p_filesz = msize;
1117 phdr->p_memsz = msize;
1120 /* Increment number of program headers. */
1126 static unsigned long init_fadump_header(unsigned long addr)
1128 struct fadump_crash_info_header *fdh;
1134 addr += sizeof(struct fadump_crash_info_header);
1136 memset(fdh, 0, sizeof(struct fadump_crash_info_header));
1137 fdh->magic_number = FADUMP_CRASH_INFO_MAGIC;
1138 fdh->elfcorehdr_addr = addr;
1139 /* We will set the crashing cpu id in crash_fadump() during crash. */
1140 fdh->crashing_cpu = FADUMP_CPU_UNKNOWN;
1145 static int register_fadump(void)
1152 * If no memory is reserved then we can not register for firmware-
1155 if (!fw_dump.reserve_dump_area_size)
1158 ret = fadump_setup_crash_memory_ranges();
1162 addr = fw_dump.fadumphdr_addr;
1164 /* Initialize fadump crash info header. */
1165 addr = init_fadump_header(addr);
1168 pr_debug("Creating ELF core headers at %#016lx\n", addr);
1169 fadump_create_elfcore_headers(vaddr);
1171 /* register the future kernel dump with firmware. */
1172 pr_debug("Registering for firmware-assisted kernel dump...\n");
1173 return fw_dump.ops->fadump_register(&fw_dump);
1176 void fadump_cleanup(void)
1178 if (!fw_dump.fadump_supported)
1181 /* Invalidate the registration only if dump is active. */
1182 if (fw_dump.dump_active) {
1183 pr_debug("Invalidating firmware-assisted dump registration\n");
1184 fw_dump.ops->fadump_invalidate(&fw_dump);
1185 } else if (fw_dump.dump_registered) {
1186 /* Un-register Firmware-assisted dump if it was registered. */
1187 fw_dump.ops->fadump_unregister(&fw_dump);
1188 fadump_free_mem_ranges(&crash_mrange_info);
1191 if (fw_dump.ops->fadump_cleanup)
1192 fw_dump.ops->fadump_cleanup(&fw_dump);
1195 static void fadump_free_reserved_memory(unsigned long start_pfn,
1196 unsigned long end_pfn)
1199 unsigned long time_limit = jiffies + HZ;
1201 pr_info("freeing reserved memory (0x%llx - 0x%llx)\n",
1202 PFN_PHYS(start_pfn), PFN_PHYS(end_pfn));
1204 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
1205 free_reserved_page(pfn_to_page(pfn));
1207 if (time_after(jiffies, time_limit)) {
1209 time_limit = jiffies + HZ;
1215 * Skip memory holes and free memory that was actually reserved.
1217 static void fadump_release_reserved_area(u64 start, u64 end)
1219 u64 tstart, tend, spfn, epfn;
1220 struct memblock_region *reg;
1222 spfn = PHYS_PFN(start);
1223 epfn = PHYS_PFN(end);
1224 for_each_memblock(memory, reg) {
1225 tstart = max_t(u64, spfn, memblock_region_memory_base_pfn(reg));
1226 tend = min_t(u64, epfn, memblock_region_memory_end_pfn(reg));
1227 if (tstart < tend) {
1228 fadump_free_reserved_memory(tstart, tend);
1239 * Sort the mem ranges in-place and merge adjacent ranges
1240 * to minimize the memory ranges count.
1242 static void sort_and_merge_mem_ranges(struct fadump_mrange_info *mrange_info)
1244 struct fadump_memory_range *mem_ranges;
1245 struct fadump_memory_range tmp_range;
1249 if (!reserved_mrange_info.mem_range_cnt)
1252 /* Sort the memory ranges */
1253 mem_ranges = mrange_info->mem_ranges;
1254 for (i = 0; i < mrange_info->mem_range_cnt; i++) {
1256 for (j = (i + 1); j < mrange_info->mem_range_cnt; j++) {
1257 if (mem_ranges[idx].base > mem_ranges[j].base)
1261 tmp_range = mem_ranges[idx];
1262 mem_ranges[idx] = mem_ranges[i];
1263 mem_ranges[i] = tmp_range;
1267 /* Merge adjacent reserved ranges */
1269 for (i = 1; i < mrange_info->mem_range_cnt; i++) {
1270 base = mem_ranges[i-1].base;
1271 size = mem_ranges[i-1].size;
1272 if (mem_ranges[i].base == (base + size))
1273 mem_ranges[idx].size += mem_ranges[i].size;
1279 mem_ranges[idx] = mem_ranges[i];
1282 mrange_info->mem_range_cnt = idx + 1;
1286 * Scan reserved-ranges to consider them while reserving/releasing
1287 * memory for FADump.
1289 static void __init early_init_dt_scan_reserved_ranges(unsigned long node)
1295 /* reserved-ranges already scanned */
1296 if (reserved_mrange_info.mem_range_cnt != 0)
1299 prop = of_get_flat_dt_prop(node, "reserved-ranges", &len);
1304 * Each reserved range is an (address,size) pair, 2 cells each,
1305 * totalling 4 cells per range.
1307 for (i = 0; i < len / (sizeof(*prop) * 4); i++) {
1310 base = of_read_number(prop + (i * 4) + 0, 2);
1311 size = of_read_number(prop + (i * 4) + 2, 2);
1314 ret = fadump_add_mem_range(&reserved_mrange_info,
1317 pr_warn("some reserved ranges are ignored!\n");
1323 /* Compact reserved ranges */
1324 sort_and_merge_mem_ranges(&reserved_mrange_info);
1328 * Release the memory that was reserved during early boot to preserve the
1329 * crash'ed kernel's memory contents except reserved dump area (permanent
1330 * reservation) and reserved ranges used by F/W. The released memory will
1331 * be available for general use.
1333 static void fadump_release_memory(u64 begin, u64 end)
1335 u64 ra_start, ra_end, tstart;
1338 ra_start = fw_dump.reserve_dump_area_start;
1339 ra_end = ra_start + fw_dump.reserve_dump_area_size;
1342 * If reserved ranges array limit is hit, overwrite the last reserved
1343 * memory range with reserved dump area to ensure it is excluded from
1344 * the memory being released (reused for next FADump registration).
1346 if (reserved_mrange_info.mem_range_cnt ==
1347 reserved_mrange_info.max_mem_ranges)
1348 reserved_mrange_info.mem_range_cnt--;
1350 ret = fadump_add_mem_range(&reserved_mrange_info, ra_start, ra_end);
1354 /* Get the reserved ranges list in order first. */
1355 sort_and_merge_mem_ranges(&reserved_mrange_info);
1357 /* Exclude reserved ranges and release remaining memory */
1359 for (i = 0; i < reserved_mrange_info.mem_range_cnt; i++) {
1360 ra_start = reserved_mrange_info.mem_ranges[i].base;
1361 ra_end = ra_start + reserved_mrange_info.mem_ranges[i].size;
1363 if (tstart >= ra_end)
1366 if (tstart < ra_start)
1367 fadump_release_reserved_area(tstart, ra_start);
1372 fadump_release_reserved_area(tstart, end);
1375 static void fadump_invalidate_release_mem(void)
1377 mutex_lock(&fadump_mutex);
1378 if (!fw_dump.dump_active) {
1379 mutex_unlock(&fadump_mutex);
1384 mutex_unlock(&fadump_mutex);
1386 fadump_release_memory(fw_dump.boot_mem_top, memblock_end_of_DRAM());
1387 fadump_free_cpu_notes_buf();
1390 * Setup kernel metadata and initialize the kernel dump
1391 * memory structure for FADump re-registration.
1393 if (fw_dump.ops->fadump_setup_metadata &&
1394 (fw_dump.ops->fadump_setup_metadata(&fw_dump) < 0))
1395 pr_warn("Failed to setup kernel metadata!\n");
1396 fw_dump.ops->fadump_init_mem_struct(&fw_dump);
1399 static ssize_t release_mem_store(struct kobject *kobj,
1400 struct kobj_attribute *attr,
1401 const char *buf, size_t count)
1405 if (!fw_dump.dump_active)
1408 if (kstrtoint(buf, 0, &input))
1413 * Take away the '/proc/vmcore'. We are releasing the dump
1414 * memory, hence it will not be valid anymore.
1416 #ifdef CONFIG_PROC_VMCORE
1419 fadump_invalidate_release_mem();
1426 /* Release the reserved memory and disable the FADump */
1427 static void unregister_fadump(void)
1430 fadump_release_memory(fw_dump.reserve_dump_area_start,
1431 fw_dump.reserve_dump_area_size);
1432 fw_dump.fadump_enabled = 0;
1433 kobject_put(fadump_kobj);
1436 static ssize_t enabled_show(struct kobject *kobj,
1437 struct kobj_attribute *attr,
1440 return sprintf(buf, "%d\n", fw_dump.fadump_enabled);
1443 static ssize_t mem_reserved_show(struct kobject *kobj,
1444 struct kobj_attribute *attr,
1447 return sprintf(buf, "%ld\n", fw_dump.reserve_dump_area_size);
1450 static ssize_t registered_show(struct kobject *kobj,
1451 struct kobj_attribute *attr,
1454 return sprintf(buf, "%d\n", fw_dump.dump_registered);
1457 static ssize_t registered_store(struct kobject *kobj,
1458 struct kobj_attribute *attr,
1459 const char *buf, size_t count)
1464 if (!fw_dump.fadump_enabled || fw_dump.dump_active)
1467 if (kstrtoint(buf, 0, &input))
1470 mutex_lock(&fadump_mutex);
1474 if (fw_dump.dump_registered == 0) {
1478 /* Un-register Firmware-assisted dump */
1479 pr_debug("Un-register firmware-assisted dump\n");
1480 fw_dump.ops->fadump_unregister(&fw_dump);
1483 if (fw_dump.dump_registered == 1) {
1484 /* Un-register Firmware-assisted dump */
1485 fw_dump.ops->fadump_unregister(&fw_dump);
1487 /* Register Firmware-assisted dump */
1488 ret = register_fadump();
1496 mutex_unlock(&fadump_mutex);
1497 return ret < 0 ? ret : count;
1500 static int fadump_region_show(struct seq_file *m, void *private)
1502 if (!fw_dump.fadump_enabled)
1505 mutex_lock(&fadump_mutex);
1506 fw_dump.ops->fadump_region_show(&fw_dump, m);
1507 mutex_unlock(&fadump_mutex);
1511 static struct kobj_attribute release_attr = __ATTR_WO(release_mem);
1512 static struct kobj_attribute enable_attr = __ATTR_RO(enabled);
1513 static struct kobj_attribute register_attr = __ATTR_RW(registered);
1514 static struct kobj_attribute mem_reserved_attr = __ATTR_RO(mem_reserved);
1516 static struct attribute *fadump_attrs[] = {
1518 ®ister_attr.attr,
1519 &mem_reserved_attr.attr,
1523 ATTRIBUTE_GROUPS(fadump);
1525 DEFINE_SHOW_ATTRIBUTE(fadump_region);
1527 static void fadump_init_files(void)
1531 fadump_kobj = kobject_create_and_add("fadump", kernel_kobj);
1533 pr_err("failed to create fadump kobject\n");
1537 debugfs_create_file("fadump_region", 0444, powerpc_debugfs_root, NULL,
1538 &fadump_region_fops);
1540 if (fw_dump.dump_active) {
1541 rc = sysfs_create_file(fadump_kobj, &release_attr.attr);
1543 pr_err("unable to create release_mem sysfs file (%d)\n",
1547 rc = sysfs_create_groups(fadump_kobj, fadump_groups);
1549 pr_err("sysfs group creation failed (%d), unregistering FADump",
1551 unregister_fadump();
1556 * The FADump sysfs are moved from kernel_kobj to fadump_kobj need to
1557 * create symlink at old location to maintain backward compatibility.
1559 * - fadump_enabled -> fadump/enabled
1560 * - fadump_registered -> fadump/registered
1561 * - fadump_release_mem -> fadump/release_mem
1563 rc = compat_only_sysfs_link_entry_to_kobj(kernel_kobj, fadump_kobj,
1564 "enabled", "fadump_enabled");
1566 pr_err("unable to create fadump_enabled symlink (%d)", rc);
1570 rc = compat_only_sysfs_link_entry_to_kobj(kernel_kobj, fadump_kobj,
1572 "fadump_registered");
1574 pr_err("unable to create fadump_registered symlink (%d)", rc);
1575 sysfs_remove_link(kernel_kobj, "fadump_enabled");
1579 if (fw_dump.dump_active) {
1580 rc = compat_only_sysfs_link_entry_to_kobj(kernel_kobj,
1583 "fadump_release_mem");
1585 pr_err("unable to create fadump_release_mem symlink (%d)",
1592 * Prepare for firmware-assisted dump.
1594 int __init setup_fadump(void)
1596 if (!fw_dump.fadump_supported)
1599 fadump_init_files();
1600 fadump_show_config();
1602 if (!fw_dump.fadump_enabled)
1606 * If dump data is available then see if it is valid and prepare for
1607 * saving it to the disk.
1609 if (fw_dump.dump_active) {
1611 * if dump process fails then invalidate the registration
1612 * and release memory before proceeding for re-registration.
1614 if (fw_dump.ops->fadump_process(&fw_dump) < 0)
1615 fadump_invalidate_release_mem();
1617 /* Initialize the kernel dump memory structure for FAD registration. */
1618 else if (fw_dump.reserve_dump_area_size)
1619 fw_dump.ops->fadump_init_mem_struct(&fw_dump);
1623 subsys_initcall(setup_fadump);
1624 #else /* !CONFIG_PRESERVE_FA_DUMP */
1626 /* Scan the Firmware Assisted dump configuration details. */
1627 int __init early_init_dt_scan_fw_dump(unsigned long node, const char *uname,
1628 int depth, void *data)
1630 if ((depth != 1) || (strcmp(uname, "ibm,opal") != 0))
1633 opal_fadump_dt_scan(&fw_dump, node);
1638 * When dump is active but PRESERVE_FA_DUMP is enabled on the kernel,
1639 * preserve crash data. The subsequent memory preserving kernel boot
1640 * is likely to process this crash data.
1642 int __init fadump_reserve_mem(void)
1644 if (fw_dump.dump_active) {
1646 * If last boot has crashed then reserve all the memory
1647 * above boot memory to preserve crash data.
1649 pr_info("Preserving crash data for processing in next boot.\n");
1650 fadump_reserve_crash_area(fw_dump.boot_mem_top);
1652 pr_debug("FADump-aware kernel..\n");
1656 #endif /* CONFIG_PRESERVE_FA_DUMP */
1658 /* Preserve everything above the base address */
1659 static void __init fadump_reserve_crash_area(u64 base)
1661 struct memblock_region *reg;
1664 for_each_memblock(memory, reg) {
1668 if ((mstart + msize) < base)
1671 if (mstart < base) {
1672 msize -= (base - mstart);
1676 pr_info("Reserving %lluMB of memory at %#016llx for preserving crash data",
1677 (msize >> 20), mstart);
1678 memblock_reserve(mstart, msize);
1682 unsigned long __init arch_reserved_kernel_pages(void)
1684 return memblock_reserved_size() / PAGE_SIZE;