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>
36 * The CPU who acquired the lock to trigger the fadump crash should
37 * wait for other CPUs to enter.
39 * The timeout is in milliseconds.
41 #define CRASH_TIMEOUT 500
43 static struct fw_dump fw_dump;
45 static void __init fadump_reserve_crash_area(u64 base);
47 struct kobject *fadump_kobj;
49 #ifndef CONFIG_PRESERVE_FA_DUMP
51 static atomic_t cpus_in_fadump;
52 static DEFINE_MUTEX(fadump_mutex);
54 struct fadump_mrange_info crash_mrange_info = { "crash", NULL, 0, 0, 0, false };
56 #define RESERVED_RNGS_SZ 16384 /* 16K - 128 entries */
57 #define RESERVED_RNGS_CNT (RESERVED_RNGS_SZ / \
58 sizeof(struct fadump_memory_range))
59 static struct fadump_memory_range rngs[RESERVED_RNGS_CNT];
60 struct fadump_mrange_info reserved_mrange_info = { "reserved", rngs,
62 RESERVED_RNGS_CNT, true };
64 static void __init early_init_dt_scan_reserved_ranges(unsigned long node);
67 static struct cma *fadump_cma;
70 * fadump_cma_init() - Initialize CMA area from a fadump reserved memory
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.
82 int __init fadump_cma_init(void)
84 unsigned long long base, size;
87 if (!fw_dump.fadump_enabled)
91 * Do not use CMA if user has provided fadump=nocma kernel parameter.
92 * Return 1 to continue with fadump old behaviour.
97 base = fw_dump.reserve_dump_area_start;
98 size = fw_dump.boot_memory_size;
103 rc = cma_init_reserved_mem(base, size, 0, "fadump_cma", &fadump_cma);
105 pr_err("Failed to init cma area for firmware-assisted dump,%d\n", rc);
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.
116 * So we now have successfully initialized cma area for fadump.
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);
126 static int __init fadump_cma_init(void) { return 1; }
127 #endif /* CONFIG_CMA */
129 /* Scan the Firmware Assisted dump configuration details. */
130 int __init early_init_dt_scan_fw_dump(unsigned long node, const char *uname,
131 int depth, void *data)
134 early_init_dt_scan_reserved_ranges(node);
141 if (strcmp(uname, "rtas") == 0) {
142 rtas_fadump_dt_scan(&fw_dump, node);
146 if (strcmp(uname, "ibm,opal") == 0) {
147 opal_fadump_dt_scan(&fw_dump, node);
155 * If fadump is registered, check if the memory provided
156 * falls within boot memory area and reserved memory area.
158 int is_fadump_memory_area(u64 addr, unsigned long size)
162 if (!fw_dump.dump_registered)
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))
173 return (addr <= fw_dump.boot_mem_top);
176 int should_fadump_crash(void)
178 if (!fw_dump.dump_registered || !fw_dump.fadumphdr_addr)
183 int is_fadump_active(void)
185 return fw_dump.dump_active;
189 * Returns true, if there are no holes in memory area between d_start to d_end,
192 static bool is_fadump_mem_area_contiguous(u64 d_start, u64 d_end)
194 phys_addr_t reg_start, reg_end;
198 for_each_mem_range(i, ®_start, ®_end) {
199 start = max_t(u64, d_start, reg_start);
200 end = min_t(u64, d_end, reg_end);
202 /* Memory hole from d_start to start */
219 * Returns true, if there are no holes in boot memory area,
222 bool is_fadump_boot_mem_contiguous(void)
224 unsigned long d_start, d_end;
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];
232 ret = is_fadump_mem_area_contiguous(d_start, d_end);
241 * Returns true, if there are no holes in reserved memory area,
244 bool is_fadump_reserved_mem_contiguous(void)
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);
253 /* Print firmware assisted dump configurations for debugging purpose. */
254 static void fadump_show_config(void)
258 pr_debug("Support for firmware-assisted dump (fadump): %s\n",
259 (fw_dump.fadump_supported ? "present" : "no support"));
261 if (!fw_dump.fadump_supported)
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]);
281 * fadump_calculate_reserve_size(): reserve variable boot area 5% of System RAM
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.
287 * This function has been taken from phyp-assisted dump feature implementation.
289 * returns larger of 256MB or 5% rounded down to multiples of 256MB.
291 * TODO: Come up with better approach to find out more accurate memory size
292 * that is required for a kernel to boot successfully.
295 static __init u64 fadump_calculate_reserve_size(void)
297 u64 base, size, bootmem_min;
300 if (fw_dump.reserve_bootvar)
301 pr_warn("'fadump_reserve_mem=' parameter is deprecated in favor of 'crashkernel=' parameter.\n");
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.
308 ret = parse_crashkernel(boot_command_line, memblock_phys_mem_size(),
310 if (ret == 0 && size > 0) {
311 unsigned long max_size;
313 if (fw_dump.reserve_bootvar)
314 pr_info("Using 'crashkernel=' parameter for memory reservation.\n");
316 fw_dump.reserve_bootvar = (unsigned long)size;
319 * Adjust if the boot memory size specified is above
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));
329 return fw_dump.reserve_bootvar;
330 } else if (fw_dump.reserve_bootvar) {
332 * 'fadump_reserve_mem=' is being used to reserve memory
333 * for firmware-assisted dump.
335 return fw_dump.reserve_bootvar;
338 /* divide by 20 to get 5% of value */
339 size = memblock_phys_mem_size() / 20;
341 /* round it down in multiples of 256 */
342 size = size & ~0x0FFFFFFFUL;
344 /* Truncate to memory_limit. We don't want to over reserve the memory.*/
345 if (memory_limit && size > memory_limit)
348 bootmem_min = fw_dump.ops->fadump_get_bootmem_min();
349 return (size > bootmem_min ? size : bootmem_min);
353 * Calculate the total memory size required to be reserved for
354 * firmware-assisted dump registration.
356 static unsigned long get_fadump_area_size(void)
358 unsigned long size = 0;
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);
369 size = PAGE_ALIGN(size);
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);
377 static int __init add_boot_mem_region(unsigned long rstart,
380 int i = fw_dump.boot_mem_regs_cnt++;
382 if (fw_dump.boot_mem_regs_cnt > FADUMP_MAX_MEM_REGS) {
383 fw_dump.boot_mem_regs_cnt = FADUMP_MAX_MEM_REGS;
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;
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.
398 static int __init add_boot_mem_regions(unsigned long mstart,
401 unsigned long rstart, rsize, max_size;
405 max_size = fw_dump.max_copy_size ? fw_dump.max_copy_size : msize;
407 if (msize > max_size)
412 ret = add_boot_mem_region(rstart, rsize);
423 static int __init fadump_get_boot_mem_regions(void)
425 unsigned long size, cur_size, hole_size, last_end;
426 unsigned long mem_size = fw_dump.boot_memory_size;
427 phys_addr_t reg_start, reg_end;
431 fw_dump.boot_mem_regs_cnt = 0;
436 for_each_mem_range(i, ®_start, ®_end) {
437 size = reg_end - reg_start;
438 hole_size += (reg_start - last_end);
440 if ((cur_size + size) >= mem_size) {
441 size = (mem_size - cur_size);
442 ret = add_boot_mem_regions(reg_start, size);
448 ret = add_boot_mem_regions(reg_start, size);
454 fw_dump.boot_mem_top = PAGE_ALIGN(fw_dump.boot_memory_size + hole_size);
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.
465 static bool overlaps_reserved_ranges(u64 base, u64 end, int *idx)
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;
477 if ((end > rbase) && (base < rend)) {
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.
491 static u64 __init fadump_locate_reserve_mem(u64 base, u64 size)
493 struct fadump_memory_range *mrngs;
494 phys_addr_t mstart, mend;
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);
505 base = PAGE_ALIGN(mstart);
507 while ((mend > base) && ((mend - base) >= size)) {
508 if (!overlaps_reserved_ranges(base, base+size, &idx)) {
513 base = mrngs[idx].base + mrngs[idx].size;
514 base = PAGE_ALIGN(base);
522 int __init fadump_reserve_mem(void)
524 u64 base, size, mem_boundary, bootmem_min;
527 if (!fw_dump.fadump_enabled)
530 if (!fw_dump.fadump_supported) {
531 pr_info("Firmware-Assisted Dump is not supported on this hardware\n");
536 * Initialize boot memory size
537 * If dump is active then we have already calculated the size during
540 if (!fw_dump.dump_active) {
541 fw_dump.boot_memory_size =
542 PAGE_ALIGN(fadump_calculate_reserve_size());
544 if (!fw_dump.nocma) {
545 fw_dump.boot_memory_size =
546 ALIGN(fw_dump.boot_memory_size,
547 FADUMP_CMA_ALIGNMENT);
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);
558 if (!fadump_get_boot_mem_regions()) {
559 pr_err("Too many holes in boot memory area to enable fadump\n");
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.
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;
576 memory_limit = memblock_end_of_DRAM();
577 printk(KERN_INFO "Adjusted memory_limit for firmware-assisted"
578 " dump, now %#016llx\n", memory_limit);
581 mem_boundary = memory_limit;
583 mem_boundary = memblock_end_of_DRAM();
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");
591 #ifdef CONFIG_HUGETLB_PAGE
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.
597 hugetlb_disabled = true;
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.
605 fadump_reserve_crash_area(base);
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);
612 * Reserve memory at an offset closer to bottom of the RAM to
613 * minimize the impact of memory hot-remove operation.
615 base = fadump_locate_reserve_mem(base, size);
617 if (!base || (base + size > mem_boundary)) {
618 pr_err("Failed to find memory chunk for reservation!\n");
621 fw_dump.reserve_dump_area_start = base;
624 * Calculate the kernel metadata address and register it with
625 * f/w if the platform supports.
627 if (fw_dump.ops->fadump_setup_metadata &&
628 (fw_dump.ops->fadump_setup_metadata(&fw_dump) < 0))
631 if (memblock_reserve(base, size)) {
632 pr_err("Failed to reserve memory!\n");
636 pr_info("Reserved %lldMB of memory at %#016llx (System RAM: %lldMB)\n",
637 (size >> 20), base, (memblock_phys_mem_size() >> 20));
639 ret = fadump_cma_init();
644 fw_dump.fadump_enabled = 0;
645 fw_dump.reserve_dump_area_size = 0;
649 /* Look for fadump= cmdline option. */
650 static int __init early_fadump_param(char *p)
655 if (strncmp(p, "on", 2) == 0)
656 fw_dump.fadump_enabled = 1;
657 else if (strncmp(p, "off", 3) == 0)
658 fw_dump.fadump_enabled = 0;
659 else if (strncmp(p, "nocma", 5) == 0) {
660 fw_dump.fadump_enabled = 1;
666 early_param("fadump", early_fadump_param);
669 * Look for fadump_reserve_mem= cmdline option
670 * TODO: Remove references to 'fadump_reserve_mem=' parameter,
671 * the sooner 'crashkernel=' parameter is accustomed to.
673 static int __init early_fadump_reserve_mem(char *p)
676 fw_dump.reserve_bootvar = memparse(p, &p);
679 early_param("fadump_reserve_mem", early_fadump_reserve_mem);
681 void crash_fadump(struct pt_regs *regs, const char *str)
684 struct fadump_crash_info_header *fdh = NULL;
685 int old_cpu, this_cpu;
686 /* Do not include first CPU */
687 unsigned int ncpus = num_online_cpus() - 1;
689 if (!should_fadump_crash())
693 * old_cpu == -1 means this is the first CPU which has come here,
694 * go ahead and trigger fadump.
696 * old_cpu != -1 means some other CPU has already on it's way
697 * to trigger fadump, just keep looping here.
699 this_cpu = smp_processor_id();
700 old_cpu = cmpxchg(&crashing_cpu, -1, this_cpu);
703 atomic_inc(&cpus_in_fadump);
706 * We can't loop here indefinitely. Wait as long as fadump
707 * is in force. If we race with fadump un-registration this
708 * loop will break and then we go down to normal panic path
709 * and reboot. If fadump is in force the first crashing
710 * cpu will definitely trigger fadump.
712 while (fw_dump.dump_registered)
717 fdh = __va(fw_dump.fadumphdr_addr);
718 fdh->crashing_cpu = crashing_cpu;
719 crash_save_vmcoreinfo();
724 ppc_save_regs(&fdh->regs);
726 fdh->online_mask = *cpu_online_mask;
729 * If we came in via system reset, wait a while for the secondary
732 if (TRAP(&(fdh->regs)) == 0x100) {
733 msecs = CRASH_TIMEOUT;
734 while ((atomic_read(&cpus_in_fadump) < ncpus) && (--msecs > 0))
738 fw_dump.ops->fadump_trigger(fdh, str);
741 u32 *fadump_regs_to_elf_notes(u32 *buf, struct pt_regs *regs)
743 struct elf_prstatus prstatus;
745 memset(&prstatus, 0, sizeof(prstatus));
747 * FIXME: How do i get PID? Do I really need it?
748 * prstatus.pr_pid = ????
750 elf_core_copy_kernel_regs(&prstatus.pr_reg, regs);
751 buf = append_elf_note(buf, CRASH_CORE_NOTE_NAME, NT_PRSTATUS,
752 &prstatus, sizeof(prstatus));
756 void fadump_update_elfcore_header(char *bufp)
758 struct elf_phdr *phdr;
760 bufp += sizeof(struct elfhdr);
762 /* First note is a place holder for cpu notes info. */
763 phdr = (struct elf_phdr *)bufp;
765 if (phdr->p_type == PT_NOTE) {
766 phdr->p_paddr = __pa(fw_dump.cpu_notes_buf_vaddr);
767 phdr->p_offset = phdr->p_paddr;
768 phdr->p_filesz = fw_dump.cpu_notes_buf_size;
769 phdr->p_memsz = fw_dump.cpu_notes_buf_size;
774 static void *fadump_alloc_buffer(unsigned long size)
776 unsigned long count, i;
780 vaddr = alloc_pages_exact(size, GFP_KERNEL | __GFP_ZERO);
784 count = PAGE_ALIGN(size) / PAGE_SIZE;
785 page = virt_to_page(vaddr);
786 for (i = 0; i < count; i++)
787 mark_page_reserved(page + i);
791 static void fadump_free_buffer(unsigned long vaddr, unsigned long size)
793 free_reserved_area((void *)vaddr, (void *)(vaddr + size), -1, NULL);
796 s32 fadump_setup_cpu_notes_buf(u32 num_cpus)
798 /* Allocate buffer to hold cpu crash notes. */
799 fw_dump.cpu_notes_buf_size = num_cpus * sizeof(note_buf_t);
800 fw_dump.cpu_notes_buf_size = PAGE_ALIGN(fw_dump.cpu_notes_buf_size);
801 fw_dump.cpu_notes_buf_vaddr =
802 (unsigned long)fadump_alloc_buffer(fw_dump.cpu_notes_buf_size);
803 if (!fw_dump.cpu_notes_buf_vaddr) {
804 pr_err("Failed to allocate %ld bytes for CPU notes buffer\n",
805 fw_dump.cpu_notes_buf_size);
809 pr_debug("Allocated buffer for cpu notes of size %ld at 0x%lx\n",
810 fw_dump.cpu_notes_buf_size,
811 fw_dump.cpu_notes_buf_vaddr);
815 void fadump_free_cpu_notes_buf(void)
817 if (!fw_dump.cpu_notes_buf_vaddr)
820 fadump_free_buffer(fw_dump.cpu_notes_buf_vaddr,
821 fw_dump.cpu_notes_buf_size);
822 fw_dump.cpu_notes_buf_vaddr = 0;
823 fw_dump.cpu_notes_buf_size = 0;
826 static void fadump_free_mem_ranges(struct fadump_mrange_info *mrange_info)
828 if (mrange_info->is_static) {
829 mrange_info->mem_range_cnt = 0;
833 kfree(mrange_info->mem_ranges);
834 memset((void *)((u64)mrange_info + RNG_NAME_SZ), 0,
835 (sizeof(struct fadump_mrange_info) - RNG_NAME_SZ));
839 * Allocate or reallocate mem_ranges array in incremental units
842 static int fadump_alloc_mem_ranges(struct fadump_mrange_info *mrange_info)
844 struct fadump_memory_range *new_array;
847 new_size = mrange_info->mem_ranges_sz + PAGE_SIZE;
848 pr_debug("Allocating %llu bytes of memory for %s memory ranges\n",
849 new_size, mrange_info->name);
851 new_array = krealloc(mrange_info->mem_ranges, new_size, GFP_KERNEL);
852 if (new_array == NULL) {
853 pr_err("Insufficient memory for setting up %s memory ranges\n",
855 fadump_free_mem_ranges(mrange_info);
859 mrange_info->mem_ranges = new_array;
860 mrange_info->mem_ranges_sz = new_size;
861 mrange_info->max_mem_ranges = (new_size /
862 sizeof(struct fadump_memory_range));
865 static inline int fadump_add_mem_range(struct fadump_mrange_info *mrange_info,
868 struct fadump_memory_range *mem_ranges = mrange_info->mem_ranges;
869 bool is_adjacent = false;
876 * Fold adjacent memory ranges to bring down the memory ranges/
877 * PT_LOAD segments count.
879 if (mrange_info->mem_range_cnt) {
880 start = mem_ranges[mrange_info->mem_range_cnt - 1].base;
881 size = mem_ranges[mrange_info->mem_range_cnt - 1].size;
884 * Boot memory area needs separate PT_LOAD segment(s) as it
885 * is moved to a different location at the time of crash.
886 * So, fold only if the region is not boot memory area.
888 if ((start + size) == base && start >= fw_dump.boot_mem_top)
892 /* resize the array on reaching the limit */
893 if (mrange_info->mem_range_cnt == mrange_info->max_mem_ranges) {
896 if (mrange_info->is_static) {
897 pr_err("Reached array size limit for %s memory ranges\n",
902 ret = fadump_alloc_mem_ranges(mrange_info);
906 /* Update to the new resized array */
907 mem_ranges = mrange_info->mem_ranges;
911 mem_ranges[mrange_info->mem_range_cnt].base = start;
912 mrange_info->mem_range_cnt++;
915 mem_ranges[mrange_info->mem_range_cnt - 1].size = (end - start);
916 pr_debug("%s_memory_range[%d] [%#016llx-%#016llx], %#llx bytes\n",
917 mrange_info->name, (mrange_info->mem_range_cnt - 1),
918 start, end - 1, (end - start));
922 static int fadump_exclude_reserved_area(u64 start, u64 end)
924 u64 ra_start, ra_end;
927 ra_start = fw_dump.reserve_dump_area_start;
928 ra_end = ra_start + fw_dump.reserve_dump_area_size;
930 if ((ra_start < end) && (ra_end > start)) {
931 if ((start < ra_start) && (end > ra_end)) {
932 ret = fadump_add_mem_range(&crash_mrange_info,
937 ret = fadump_add_mem_range(&crash_mrange_info,
939 } else if (start < ra_start) {
940 ret = fadump_add_mem_range(&crash_mrange_info,
942 } else if (ra_end < end) {
943 ret = fadump_add_mem_range(&crash_mrange_info,
947 ret = fadump_add_mem_range(&crash_mrange_info, start, end);
952 static int fadump_init_elfcore_header(char *bufp)
956 elf = (struct elfhdr *) bufp;
957 bufp += sizeof(struct elfhdr);
958 memcpy(elf->e_ident, ELFMAG, SELFMAG);
959 elf->e_ident[EI_CLASS] = ELF_CLASS;
960 elf->e_ident[EI_DATA] = ELF_DATA;
961 elf->e_ident[EI_VERSION] = EV_CURRENT;
962 elf->e_ident[EI_OSABI] = ELF_OSABI;
963 memset(elf->e_ident+EI_PAD, 0, EI_NIDENT-EI_PAD);
964 elf->e_type = ET_CORE;
965 elf->e_machine = ELF_ARCH;
966 elf->e_version = EV_CURRENT;
968 elf->e_phoff = sizeof(struct elfhdr);
970 #if defined(_CALL_ELF)
971 elf->e_flags = _CALL_ELF;
975 elf->e_ehsize = sizeof(struct elfhdr);
976 elf->e_phentsize = sizeof(struct elf_phdr);
978 elf->e_shentsize = 0;
986 * Traverse through memblock structure and setup crash memory ranges. These
987 * ranges will be used create PT_LOAD program headers in elfcore header.
989 static int fadump_setup_crash_memory_ranges(void)
994 pr_debug("Setup crash memory ranges.\n");
995 crash_mrange_info.mem_range_cnt = 0;
998 * Boot memory region(s) registered with firmware are moved to
999 * different location at the time of crash. Create separate program
1000 * header(s) for this memory chunk(s) with the correct offset.
1002 for (i = 0; i < fw_dump.boot_mem_regs_cnt; i++) {
1003 start = fw_dump.boot_mem_addr[i];
1004 end = start + fw_dump.boot_mem_sz[i];
1005 ret = fadump_add_mem_range(&crash_mrange_info, start, end);
1010 for_each_mem_range(i, &start, &end) {
1012 * skip the memory chunk that is already added
1013 * (0 through boot_memory_top).
1015 if (start < fw_dump.boot_mem_top) {
1016 if (end > fw_dump.boot_mem_top)
1017 start = fw_dump.boot_mem_top;
1022 /* add this range excluding the reserved dump area. */
1023 ret = fadump_exclude_reserved_area(start, end);
1032 * If the given physical address falls within the boot memory region then
1033 * return the relocated address that points to the dump region reserved
1034 * for saving initial boot memory contents.
1036 static inline unsigned long fadump_relocate(unsigned long paddr)
1038 unsigned long raddr, rstart, rend, rlast, hole_size;
1044 for (i = 0; i < fw_dump.boot_mem_regs_cnt; i++) {
1045 rstart = fw_dump.boot_mem_addr[i];
1046 rend = rstart + fw_dump.boot_mem_sz[i];
1047 hole_size += (rstart - rlast);
1049 if (paddr >= rstart && paddr < rend) {
1050 raddr += fw_dump.boot_mem_dest_addr - hole_size;
1057 pr_debug("vmcoreinfo: paddr = 0x%lx, raddr = 0x%lx\n", paddr, raddr);
1061 static int fadump_create_elfcore_headers(char *bufp)
1063 unsigned long long raddr, offset;
1064 struct elf_phdr *phdr;
1068 fadump_init_elfcore_header(bufp);
1069 elf = (struct elfhdr *)bufp;
1070 bufp += sizeof(struct elfhdr);
1073 * setup ELF PT_NOTE, place holder for cpu notes info. The notes info
1074 * will be populated during second kernel boot after crash. Hence
1075 * this PT_NOTE will always be the first elf note.
1077 * NOTE: Any new ELF note addition should be placed after this note.
1079 phdr = (struct elf_phdr *)bufp;
1080 bufp += sizeof(struct elf_phdr);
1081 phdr->p_type = PT_NOTE;
1093 /* setup ELF PT_NOTE for vmcoreinfo */
1094 phdr = (struct elf_phdr *)bufp;
1095 bufp += sizeof(struct elf_phdr);
1096 phdr->p_type = PT_NOTE;
1101 phdr->p_paddr = fadump_relocate(paddr_vmcoreinfo_note());
1102 phdr->p_offset = phdr->p_paddr;
1103 phdr->p_memsz = phdr->p_filesz = VMCOREINFO_NOTE_SIZE;
1105 /* Increment number of program headers. */
1108 /* setup PT_LOAD sections. */
1111 raddr = fw_dump.boot_mem_addr[0];
1112 for (i = 0; i < crash_mrange_info.mem_range_cnt; i++) {
1115 mbase = crash_mrange_info.mem_ranges[i].base;
1116 msize = crash_mrange_info.mem_ranges[i].size;
1120 phdr = (struct elf_phdr *)bufp;
1121 bufp += sizeof(struct elf_phdr);
1122 phdr->p_type = PT_LOAD;
1123 phdr->p_flags = PF_R|PF_W|PF_X;
1124 phdr->p_offset = mbase;
1126 if (mbase == raddr) {
1128 * The entire real memory region will be moved by
1129 * firmware to the specified destination_address.
1130 * Hence set the correct offset.
1132 phdr->p_offset = fw_dump.boot_mem_dest_addr + offset;
1133 if (j < (fw_dump.boot_mem_regs_cnt - 1)) {
1134 offset += fw_dump.boot_mem_sz[j];
1135 raddr = fw_dump.boot_mem_addr[++j];
1139 phdr->p_paddr = mbase;
1140 phdr->p_vaddr = (unsigned long)__va(mbase);
1141 phdr->p_filesz = msize;
1142 phdr->p_memsz = msize;
1145 /* Increment number of program headers. */
1151 static unsigned long init_fadump_header(unsigned long addr)
1153 struct fadump_crash_info_header *fdh;
1159 addr += sizeof(struct fadump_crash_info_header);
1161 memset(fdh, 0, sizeof(struct fadump_crash_info_header));
1162 fdh->magic_number = FADUMP_CRASH_INFO_MAGIC;
1163 fdh->elfcorehdr_addr = addr;
1164 /* We will set the crashing cpu id in crash_fadump() during crash. */
1165 fdh->crashing_cpu = FADUMP_CPU_UNKNOWN;
1170 static int register_fadump(void)
1177 * If no memory is reserved then we can not register for firmware-
1180 if (!fw_dump.reserve_dump_area_size)
1183 ret = fadump_setup_crash_memory_ranges();
1187 addr = fw_dump.fadumphdr_addr;
1189 /* Initialize fadump crash info header. */
1190 addr = init_fadump_header(addr);
1193 pr_debug("Creating ELF core headers at %#016lx\n", addr);
1194 fadump_create_elfcore_headers(vaddr);
1196 /* register the future kernel dump with firmware. */
1197 pr_debug("Registering for firmware-assisted kernel dump...\n");
1198 return fw_dump.ops->fadump_register(&fw_dump);
1201 void fadump_cleanup(void)
1203 if (!fw_dump.fadump_supported)
1206 /* Invalidate the registration only if dump is active. */
1207 if (fw_dump.dump_active) {
1208 pr_debug("Invalidating firmware-assisted dump registration\n");
1209 fw_dump.ops->fadump_invalidate(&fw_dump);
1210 } else if (fw_dump.dump_registered) {
1211 /* Un-register Firmware-assisted dump if it was registered. */
1212 fw_dump.ops->fadump_unregister(&fw_dump);
1213 fadump_free_mem_ranges(&crash_mrange_info);
1216 if (fw_dump.ops->fadump_cleanup)
1217 fw_dump.ops->fadump_cleanup(&fw_dump);
1220 static void fadump_free_reserved_memory(unsigned long start_pfn,
1221 unsigned long end_pfn)
1224 unsigned long time_limit = jiffies + HZ;
1226 pr_info("freeing reserved memory (0x%llx - 0x%llx)\n",
1227 PFN_PHYS(start_pfn), PFN_PHYS(end_pfn));
1229 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
1230 free_reserved_page(pfn_to_page(pfn));
1232 if (time_after(jiffies, time_limit)) {
1234 time_limit = jiffies + HZ;
1240 * Skip memory holes and free memory that was actually reserved.
1242 static void fadump_release_reserved_area(u64 start, u64 end)
1244 unsigned long reg_spfn, reg_epfn;
1245 u64 tstart, tend, spfn, epfn;
1248 spfn = PHYS_PFN(start);
1249 epfn = PHYS_PFN(end);
1251 for_each_mem_pfn_range(i, MAX_NUMNODES, ®_spfn, ®_epfn, NULL) {
1252 tstart = max_t(u64, spfn, reg_spfn);
1253 tend = min_t(u64, epfn, reg_epfn);
1255 if (tstart < tend) {
1256 fadump_free_reserved_memory(tstart, tend);
1267 * Sort the mem ranges in-place and merge adjacent ranges
1268 * to minimize the memory ranges count.
1270 static void sort_and_merge_mem_ranges(struct fadump_mrange_info *mrange_info)
1272 struct fadump_memory_range *mem_ranges;
1273 struct fadump_memory_range tmp_range;
1277 if (!reserved_mrange_info.mem_range_cnt)
1280 /* Sort the memory ranges */
1281 mem_ranges = mrange_info->mem_ranges;
1282 for (i = 0; i < mrange_info->mem_range_cnt; i++) {
1284 for (j = (i + 1); j < mrange_info->mem_range_cnt; j++) {
1285 if (mem_ranges[idx].base > mem_ranges[j].base)
1289 tmp_range = mem_ranges[idx];
1290 mem_ranges[idx] = mem_ranges[i];
1291 mem_ranges[i] = tmp_range;
1295 /* Merge adjacent reserved ranges */
1297 for (i = 1; i < mrange_info->mem_range_cnt; i++) {
1298 base = mem_ranges[i-1].base;
1299 size = mem_ranges[i-1].size;
1300 if (mem_ranges[i].base == (base + size))
1301 mem_ranges[idx].size += mem_ranges[i].size;
1307 mem_ranges[idx] = mem_ranges[i];
1310 mrange_info->mem_range_cnt = idx + 1;
1314 * Scan reserved-ranges to consider them while reserving/releasing
1315 * memory for FADump.
1317 static void __init early_init_dt_scan_reserved_ranges(unsigned long node)
1323 /* reserved-ranges already scanned */
1324 if (reserved_mrange_info.mem_range_cnt != 0)
1327 prop = of_get_flat_dt_prop(node, "reserved-ranges", &len);
1332 * Each reserved range is an (address,size) pair, 2 cells each,
1333 * totalling 4 cells per range.
1335 for (i = 0; i < len / (sizeof(*prop) * 4); i++) {
1338 base = of_read_number(prop + (i * 4) + 0, 2);
1339 size = of_read_number(prop + (i * 4) + 2, 2);
1342 ret = fadump_add_mem_range(&reserved_mrange_info,
1345 pr_warn("some reserved ranges are ignored!\n");
1351 /* Compact reserved ranges */
1352 sort_and_merge_mem_ranges(&reserved_mrange_info);
1356 * Release the memory that was reserved during early boot to preserve the
1357 * crash'ed kernel's memory contents except reserved dump area (permanent
1358 * reservation) and reserved ranges used by F/W. The released memory will
1359 * be available for general use.
1361 static void fadump_release_memory(u64 begin, u64 end)
1363 u64 ra_start, ra_end, tstart;
1366 ra_start = fw_dump.reserve_dump_area_start;
1367 ra_end = ra_start + fw_dump.reserve_dump_area_size;
1370 * If reserved ranges array limit is hit, overwrite the last reserved
1371 * memory range with reserved dump area to ensure it is excluded from
1372 * the memory being released (reused for next FADump registration).
1374 if (reserved_mrange_info.mem_range_cnt ==
1375 reserved_mrange_info.max_mem_ranges)
1376 reserved_mrange_info.mem_range_cnt--;
1378 ret = fadump_add_mem_range(&reserved_mrange_info, ra_start, ra_end);
1382 /* Get the reserved ranges list in order first. */
1383 sort_and_merge_mem_ranges(&reserved_mrange_info);
1385 /* Exclude reserved ranges and release remaining memory */
1387 for (i = 0; i < reserved_mrange_info.mem_range_cnt; i++) {
1388 ra_start = reserved_mrange_info.mem_ranges[i].base;
1389 ra_end = ra_start + reserved_mrange_info.mem_ranges[i].size;
1391 if (tstart >= ra_end)
1394 if (tstart < ra_start)
1395 fadump_release_reserved_area(tstart, ra_start);
1400 fadump_release_reserved_area(tstart, end);
1403 static void fadump_invalidate_release_mem(void)
1405 mutex_lock(&fadump_mutex);
1406 if (!fw_dump.dump_active) {
1407 mutex_unlock(&fadump_mutex);
1412 mutex_unlock(&fadump_mutex);
1414 fadump_release_memory(fw_dump.boot_mem_top, memblock_end_of_DRAM());
1415 fadump_free_cpu_notes_buf();
1418 * Setup kernel metadata and initialize the kernel dump
1419 * memory structure for FADump re-registration.
1421 if (fw_dump.ops->fadump_setup_metadata &&
1422 (fw_dump.ops->fadump_setup_metadata(&fw_dump) < 0))
1423 pr_warn("Failed to setup kernel metadata!\n");
1424 fw_dump.ops->fadump_init_mem_struct(&fw_dump);
1427 static ssize_t release_mem_store(struct kobject *kobj,
1428 struct kobj_attribute *attr,
1429 const char *buf, size_t count)
1433 if (!fw_dump.dump_active)
1436 if (kstrtoint(buf, 0, &input))
1441 * Take away the '/proc/vmcore'. We are releasing the dump
1442 * memory, hence it will not be valid anymore.
1444 #ifdef CONFIG_PROC_VMCORE
1447 fadump_invalidate_release_mem();
1454 /* Release the reserved memory and disable the FADump */
1455 static void unregister_fadump(void)
1458 fadump_release_memory(fw_dump.reserve_dump_area_start,
1459 fw_dump.reserve_dump_area_size);
1460 fw_dump.fadump_enabled = 0;
1461 kobject_put(fadump_kobj);
1464 static ssize_t enabled_show(struct kobject *kobj,
1465 struct kobj_attribute *attr,
1468 return sprintf(buf, "%d\n", fw_dump.fadump_enabled);
1471 static ssize_t mem_reserved_show(struct kobject *kobj,
1472 struct kobj_attribute *attr,
1475 return sprintf(buf, "%ld\n", fw_dump.reserve_dump_area_size);
1478 static ssize_t registered_show(struct kobject *kobj,
1479 struct kobj_attribute *attr,
1482 return sprintf(buf, "%d\n", fw_dump.dump_registered);
1485 static ssize_t registered_store(struct kobject *kobj,
1486 struct kobj_attribute *attr,
1487 const char *buf, size_t count)
1492 if (!fw_dump.fadump_enabled || fw_dump.dump_active)
1495 if (kstrtoint(buf, 0, &input))
1498 mutex_lock(&fadump_mutex);
1502 if (fw_dump.dump_registered == 0) {
1506 /* Un-register Firmware-assisted dump */
1507 pr_debug("Un-register firmware-assisted dump\n");
1508 fw_dump.ops->fadump_unregister(&fw_dump);
1511 if (fw_dump.dump_registered == 1) {
1512 /* Un-register Firmware-assisted dump */
1513 fw_dump.ops->fadump_unregister(&fw_dump);
1515 /* Register Firmware-assisted dump */
1516 ret = register_fadump();
1524 mutex_unlock(&fadump_mutex);
1525 return ret < 0 ? ret : count;
1528 static int fadump_region_show(struct seq_file *m, void *private)
1530 if (!fw_dump.fadump_enabled)
1533 mutex_lock(&fadump_mutex);
1534 fw_dump.ops->fadump_region_show(&fw_dump, m);
1535 mutex_unlock(&fadump_mutex);
1539 static struct kobj_attribute release_attr = __ATTR_WO(release_mem);
1540 static struct kobj_attribute enable_attr = __ATTR_RO(enabled);
1541 static struct kobj_attribute register_attr = __ATTR_RW(registered);
1542 static struct kobj_attribute mem_reserved_attr = __ATTR_RO(mem_reserved);
1544 static struct attribute *fadump_attrs[] = {
1546 ®ister_attr.attr,
1547 &mem_reserved_attr.attr,
1551 ATTRIBUTE_GROUPS(fadump);
1553 DEFINE_SHOW_ATTRIBUTE(fadump_region);
1555 static void fadump_init_files(void)
1559 fadump_kobj = kobject_create_and_add("fadump", kernel_kobj);
1561 pr_err("failed to create fadump kobject\n");
1565 debugfs_create_file("fadump_region", 0444, powerpc_debugfs_root, NULL,
1566 &fadump_region_fops);
1568 if (fw_dump.dump_active) {
1569 rc = sysfs_create_file(fadump_kobj, &release_attr.attr);
1571 pr_err("unable to create release_mem sysfs file (%d)\n",
1575 rc = sysfs_create_groups(fadump_kobj, fadump_groups);
1577 pr_err("sysfs group creation failed (%d), unregistering FADump",
1579 unregister_fadump();
1584 * The FADump sysfs are moved from kernel_kobj to fadump_kobj need to
1585 * create symlink at old location to maintain backward compatibility.
1587 * - fadump_enabled -> fadump/enabled
1588 * - fadump_registered -> fadump/registered
1589 * - fadump_release_mem -> fadump/release_mem
1591 rc = compat_only_sysfs_link_entry_to_kobj(kernel_kobj, fadump_kobj,
1592 "enabled", "fadump_enabled");
1594 pr_err("unable to create fadump_enabled symlink (%d)", rc);
1598 rc = compat_only_sysfs_link_entry_to_kobj(kernel_kobj, fadump_kobj,
1600 "fadump_registered");
1602 pr_err("unable to create fadump_registered symlink (%d)", rc);
1603 sysfs_remove_link(kernel_kobj, "fadump_enabled");
1607 if (fw_dump.dump_active) {
1608 rc = compat_only_sysfs_link_entry_to_kobj(kernel_kobj,
1611 "fadump_release_mem");
1613 pr_err("unable to create fadump_release_mem symlink (%d)",
1620 * Prepare for firmware-assisted dump.
1622 int __init setup_fadump(void)
1624 if (!fw_dump.fadump_supported)
1627 fadump_init_files();
1628 fadump_show_config();
1630 if (!fw_dump.fadump_enabled)
1634 * If dump data is available then see if it is valid and prepare for
1635 * saving it to the disk.
1637 if (fw_dump.dump_active) {
1639 * if dump process fails then invalidate the registration
1640 * and release memory before proceeding for re-registration.
1642 if (fw_dump.ops->fadump_process(&fw_dump) < 0)
1643 fadump_invalidate_release_mem();
1645 /* Initialize the kernel dump memory structure for FAD registration. */
1646 else if (fw_dump.reserve_dump_area_size)
1647 fw_dump.ops->fadump_init_mem_struct(&fw_dump);
1650 * In case of panic, fadump is triggered via ppc_panic_event()
1651 * panic notifier. Setting crash_kexec_post_notifiers to 'true'
1652 * lets panic() function take crash friendly path before panic
1653 * notifiers are invoked.
1655 crash_kexec_post_notifiers = true;
1659 subsys_initcall(setup_fadump);
1660 #else /* !CONFIG_PRESERVE_FA_DUMP */
1662 /* Scan the Firmware Assisted dump configuration details. */
1663 int __init early_init_dt_scan_fw_dump(unsigned long node, const char *uname,
1664 int depth, void *data)
1666 if ((depth != 1) || (strcmp(uname, "ibm,opal") != 0))
1669 opal_fadump_dt_scan(&fw_dump, node);
1674 * When dump is active but PRESERVE_FA_DUMP is enabled on the kernel,
1675 * preserve crash data. The subsequent memory preserving kernel boot
1676 * is likely to process this crash data.
1678 int __init fadump_reserve_mem(void)
1680 if (fw_dump.dump_active) {
1682 * If last boot has crashed then reserve all the memory
1683 * above boot memory to preserve crash data.
1685 pr_info("Preserving crash data for processing in next boot.\n");
1686 fadump_reserve_crash_area(fw_dump.boot_mem_top);
1688 pr_debug("FADump-aware kernel..\n");
1692 #endif /* CONFIG_PRESERVE_FA_DUMP */
1694 /* Preserve everything above the base address */
1695 static void __init fadump_reserve_crash_area(u64 base)
1697 u64 i, mstart, mend, msize;
1699 for_each_mem_range(i, &mstart, &mend) {
1700 msize = mend - mstart;
1702 if ((mstart + msize) < base)
1705 if (mstart < base) {
1706 msize -= (base - mstart);
1710 pr_info("Reserving %lluMB of memory at %#016llx for preserving crash data",
1711 (msize >> 20), mstart);
1712 memblock_reserve(mstart, msize);
1716 unsigned long __init arch_reserved_kernel_pages(void)
1718 return memblock_reserved_size() / PAGE_SIZE;