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>
34 #include <asm/interrupt.h>
37 * The CPU who acquired the lock to trigger the fadump crash should
38 * wait for other CPUs to enter.
40 * The timeout is in milliseconds.
42 #define CRASH_TIMEOUT 500
44 static struct fw_dump fw_dump;
46 static void __init fadump_reserve_crash_area(u64 base);
48 struct kobject *fadump_kobj;
50 #ifndef CONFIG_PRESERVE_FA_DUMP
52 static atomic_t cpus_in_fadump;
53 static DEFINE_MUTEX(fadump_mutex);
55 struct fadump_mrange_info crash_mrange_info = { "crash", NULL, 0, 0, 0, false };
57 #define RESERVED_RNGS_SZ 16384 /* 16K - 128 entries */
58 #define RESERVED_RNGS_CNT (RESERVED_RNGS_SZ / \
59 sizeof(struct fadump_memory_range))
60 static struct fadump_memory_range rngs[RESERVED_RNGS_CNT];
61 struct fadump_mrange_info reserved_mrange_info = { "reserved", rngs,
63 RESERVED_RNGS_CNT, true };
65 static void __init early_init_dt_scan_reserved_ranges(unsigned long node);
68 static struct cma *fadump_cma;
71 * fadump_cma_init() - Initialize CMA area from a fadump reserved memory
73 * This function initializes CMA area from fadump reserved memory.
74 * The total size of fadump reserved memory covers for boot memory size
75 * + cpu data size + hpte size and metadata.
76 * Initialize only the area equivalent to boot memory size for CMA use.
77 * The reamining portion of fadump reserved memory will be not given
78 * to CMA and pages for thoes will stay reserved. boot memory size is
79 * aligned per CMA requirement to satisy cma_init_reserved_mem() call.
80 * But for some reason even if it fails we still have the memory reservation
81 * with us and we can still continue doing fadump.
83 int __init fadump_cma_init(void)
85 unsigned long long base, size;
88 if (!fw_dump.fadump_enabled)
92 * Do not use CMA if user has provided fadump=nocma kernel parameter.
93 * Return 1 to continue with fadump old behaviour.
98 base = fw_dump.reserve_dump_area_start;
99 size = fw_dump.boot_memory_size;
104 rc = cma_init_reserved_mem(base, size, 0, "fadump_cma", &fadump_cma);
106 pr_err("Failed to init cma area for firmware-assisted dump,%d\n", rc);
108 * Though the CMA init has failed we still have memory
109 * reservation with us. The reserved memory will be
110 * blocked from production system usage. Hence return 1,
111 * so that we can continue with fadump.
117 * So we now have successfully initialized cma area for fadump.
119 pr_info("Initialized 0x%lx bytes cma area at %ldMB from 0x%lx "
120 "bytes of memory reserved for firmware-assisted dump\n",
121 cma_get_size(fadump_cma),
122 (unsigned long)cma_get_base(fadump_cma) >> 20,
123 fw_dump.reserve_dump_area_size);
127 static int __init fadump_cma_init(void) { return 1; }
128 #endif /* CONFIG_CMA */
130 /* Scan the Firmware Assisted dump configuration details. */
131 int __init early_init_dt_scan_fw_dump(unsigned long node, const char *uname,
132 int depth, void *data)
135 early_init_dt_scan_reserved_ranges(node);
142 if (strcmp(uname, "rtas") == 0) {
143 rtas_fadump_dt_scan(&fw_dump, node);
147 if (strcmp(uname, "ibm,opal") == 0) {
148 opal_fadump_dt_scan(&fw_dump, node);
156 * If fadump is registered, check if the memory provided
157 * falls within boot memory area and reserved memory area.
159 int is_fadump_memory_area(u64 addr, unsigned long size)
163 if (!fw_dump.dump_registered)
169 d_start = fw_dump.reserve_dump_area_start;
170 d_end = d_start + fw_dump.reserve_dump_area_size;
171 if (((addr + size) > d_start) && (addr <= d_end))
174 return (addr <= fw_dump.boot_mem_top);
177 int should_fadump_crash(void)
179 if (!fw_dump.dump_registered || !fw_dump.fadumphdr_addr)
184 int is_fadump_active(void)
186 return fw_dump.dump_active;
190 * Returns true, if there are no holes in memory area between d_start to d_end,
193 static bool is_fadump_mem_area_contiguous(u64 d_start, u64 d_end)
195 phys_addr_t reg_start, reg_end;
199 for_each_mem_range(i, ®_start, ®_end) {
200 start = max_t(u64, d_start, reg_start);
201 end = min_t(u64, d_end, reg_end);
203 /* Memory hole from d_start to start */
220 * Returns true, if there are no holes in boot memory area,
223 bool is_fadump_boot_mem_contiguous(void)
225 unsigned long d_start, d_end;
229 for (i = 0; i < fw_dump.boot_mem_regs_cnt; i++) {
230 d_start = fw_dump.boot_mem_addr[i];
231 d_end = d_start + fw_dump.boot_mem_sz[i];
233 ret = is_fadump_mem_area_contiguous(d_start, d_end);
242 * Returns true, if there are no holes in reserved memory area,
245 bool is_fadump_reserved_mem_contiguous(void)
249 d_start = fw_dump.reserve_dump_area_start;
250 d_end = d_start + fw_dump.reserve_dump_area_size;
251 return is_fadump_mem_area_contiguous(d_start, d_end);
254 /* Print firmware assisted dump configurations for debugging purpose. */
255 static void fadump_show_config(void)
259 pr_debug("Support for firmware-assisted dump (fadump): %s\n",
260 (fw_dump.fadump_supported ? "present" : "no support"));
262 if (!fw_dump.fadump_supported)
265 pr_debug("Fadump enabled : %s\n",
266 (fw_dump.fadump_enabled ? "yes" : "no"));
267 pr_debug("Dump Active : %s\n",
268 (fw_dump.dump_active ? "yes" : "no"));
269 pr_debug("Dump section sizes:\n");
270 pr_debug(" CPU state data size: %lx\n", fw_dump.cpu_state_data_size);
271 pr_debug(" HPTE region size : %lx\n", fw_dump.hpte_region_size);
272 pr_debug(" Boot memory size : %lx\n", fw_dump.boot_memory_size);
273 pr_debug(" Boot memory top : %llx\n", fw_dump.boot_mem_top);
274 pr_debug("Boot memory regions cnt: %llx\n", fw_dump.boot_mem_regs_cnt);
275 for (i = 0; i < fw_dump.boot_mem_regs_cnt; i++) {
276 pr_debug("[%03d] base = %llx, size = %llx\n", i,
277 fw_dump.boot_mem_addr[i], fw_dump.boot_mem_sz[i]);
282 * fadump_calculate_reserve_size(): reserve variable boot area 5% of System RAM
284 * Function to find the largest memory size we need to reserve during early
285 * boot process. This will be the size of the memory that is required for a
286 * kernel to boot successfully.
288 * This function has been taken from phyp-assisted dump feature implementation.
290 * returns larger of 256MB or 5% rounded down to multiples of 256MB.
292 * TODO: Come up with better approach to find out more accurate memory size
293 * that is required for a kernel to boot successfully.
296 static __init u64 fadump_calculate_reserve_size(void)
298 u64 base, size, bootmem_min;
301 if (fw_dump.reserve_bootvar)
302 pr_warn("'fadump_reserve_mem=' parameter is deprecated in favor of 'crashkernel=' parameter.\n");
305 * Check if the size is specified through crashkernel= cmdline
306 * option. If yes, then use that but ignore base as fadump reserves
307 * memory at a predefined offset.
309 ret = parse_crashkernel(boot_command_line, memblock_phys_mem_size(),
311 if (ret == 0 && size > 0) {
312 unsigned long max_size;
314 if (fw_dump.reserve_bootvar)
315 pr_info("Using 'crashkernel=' parameter for memory reservation.\n");
317 fw_dump.reserve_bootvar = (unsigned long)size;
320 * Adjust if the boot memory size specified is above
323 max_size = memblock_phys_mem_size() / MAX_BOOT_MEM_RATIO;
324 if (fw_dump.reserve_bootvar > max_size) {
325 fw_dump.reserve_bootvar = max_size;
326 pr_info("Adjusted boot memory size to %luMB\n",
327 (fw_dump.reserve_bootvar >> 20));
330 return fw_dump.reserve_bootvar;
331 } else if (fw_dump.reserve_bootvar) {
333 * 'fadump_reserve_mem=' is being used to reserve memory
334 * for firmware-assisted dump.
336 return fw_dump.reserve_bootvar;
339 /* divide by 20 to get 5% of value */
340 size = memblock_phys_mem_size() / 20;
342 /* round it down in multiples of 256 */
343 size = size & ~0x0FFFFFFFUL;
345 /* Truncate to memory_limit. We don't want to over reserve the memory.*/
346 if (memory_limit && size > memory_limit)
349 bootmem_min = fw_dump.ops->fadump_get_bootmem_min();
350 return (size > bootmem_min ? size : bootmem_min);
354 * Calculate the total memory size required to be reserved for
355 * firmware-assisted dump registration.
357 static unsigned long get_fadump_area_size(void)
359 unsigned long size = 0;
361 size += fw_dump.cpu_state_data_size;
362 size += fw_dump.hpte_region_size;
363 size += fw_dump.boot_memory_size;
364 size += sizeof(struct fadump_crash_info_header);
365 size += sizeof(struct elfhdr); /* ELF core header.*/
366 size += sizeof(struct elf_phdr); /* place holder for cpu notes */
367 /* Program headers for crash memory regions. */
368 size += sizeof(struct elf_phdr) * (memblock_num_regions(memory) + 2);
370 size = PAGE_ALIGN(size);
372 /* This is to hold kernel metadata on platforms that support it */
373 size += (fw_dump.ops->fadump_get_metadata_size ?
374 fw_dump.ops->fadump_get_metadata_size() : 0);
378 static int __init add_boot_mem_region(unsigned long rstart,
381 int i = fw_dump.boot_mem_regs_cnt++;
383 if (fw_dump.boot_mem_regs_cnt > FADUMP_MAX_MEM_REGS) {
384 fw_dump.boot_mem_regs_cnt = FADUMP_MAX_MEM_REGS;
388 pr_debug("Added boot memory range[%d] [%#016lx-%#016lx)\n",
389 i, rstart, (rstart + rsize));
390 fw_dump.boot_mem_addr[i] = rstart;
391 fw_dump.boot_mem_sz[i] = rsize;
396 * Firmware usually has a hard limit on the data it can copy per region.
397 * Honour that by splitting a memory range into multiple regions.
399 static int __init add_boot_mem_regions(unsigned long mstart,
402 unsigned long rstart, rsize, max_size;
406 max_size = fw_dump.max_copy_size ? fw_dump.max_copy_size : msize;
408 if (msize > max_size)
413 ret = add_boot_mem_region(rstart, rsize);
424 static int __init fadump_get_boot_mem_regions(void)
426 unsigned long size, cur_size, hole_size, last_end;
427 unsigned long mem_size = fw_dump.boot_memory_size;
428 phys_addr_t reg_start, reg_end;
432 fw_dump.boot_mem_regs_cnt = 0;
437 for_each_mem_range(i, ®_start, ®_end) {
438 size = reg_end - reg_start;
439 hole_size += (reg_start - last_end);
441 if ((cur_size + size) >= mem_size) {
442 size = (mem_size - cur_size);
443 ret = add_boot_mem_regions(reg_start, size);
449 ret = add_boot_mem_regions(reg_start, size);
455 fw_dump.boot_mem_top = PAGE_ALIGN(fw_dump.boot_memory_size + hole_size);
461 * Returns true, if the given range overlaps with reserved memory ranges
462 * starting at idx. Also, updates idx to index of overlapping memory range
463 * with the given memory range.
466 static bool overlaps_reserved_ranges(u64 base, u64 end, int *idx)
471 for (i = *idx; i < reserved_mrange_info.mem_range_cnt; i++) {
472 u64 rbase = reserved_mrange_info.mem_ranges[i].base;
473 u64 rend = rbase + reserved_mrange_info.mem_ranges[i].size;
478 if ((end > rbase) && (base < rend)) {
489 * Locate a suitable memory area to reserve memory for FADump. While at it,
490 * lookup reserved-ranges & avoid overlap with them, as they are used by F/W.
492 static u64 __init fadump_locate_reserve_mem(u64 base, u64 size)
494 struct fadump_memory_range *mrngs;
495 phys_addr_t mstart, mend;
499 mrngs = reserved_mrange_info.mem_ranges;
500 for_each_free_mem_range(i, NUMA_NO_NODE, MEMBLOCK_NONE,
501 &mstart, &mend, NULL) {
502 pr_debug("%llu) mstart: %llx, mend: %llx, base: %llx\n",
503 i, mstart, mend, base);
506 base = PAGE_ALIGN(mstart);
508 while ((mend > base) && ((mend - base) >= size)) {
509 if (!overlaps_reserved_ranges(base, base+size, &idx)) {
514 base = mrngs[idx].base + mrngs[idx].size;
515 base = PAGE_ALIGN(base);
523 int __init fadump_reserve_mem(void)
525 u64 base, size, mem_boundary, bootmem_min;
528 if (!fw_dump.fadump_enabled)
531 if (!fw_dump.fadump_supported) {
532 pr_info("Firmware-Assisted Dump is not supported on this hardware\n");
537 * Initialize boot memory size
538 * If dump is active then we have already calculated the size during
541 if (!fw_dump.dump_active) {
542 fw_dump.boot_memory_size =
543 PAGE_ALIGN(fadump_calculate_reserve_size());
545 if (!fw_dump.nocma) {
546 fw_dump.boot_memory_size =
547 ALIGN(fw_dump.boot_memory_size,
548 FADUMP_CMA_ALIGNMENT);
552 bootmem_min = fw_dump.ops->fadump_get_bootmem_min();
553 if (fw_dump.boot_memory_size < bootmem_min) {
554 pr_err("Can't enable fadump with boot memory size (0x%lx) less than 0x%llx\n",
555 fw_dump.boot_memory_size, bootmem_min);
559 if (!fadump_get_boot_mem_regions()) {
560 pr_err("Too many holes in boot memory area to enable fadump\n");
566 * Calculate the memory boundary.
567 * If memory_limit is less than actual memory boundary then reserve
568 * the memory for fadump beyond the memory_limit and adjust the
569 * memory_limit accordingly, so that the running kernel can run with
570 * specified memory_limit.
572 if (memory_limit && memory_limit < memblock_end_of_DRAM()) {
573 size = get_fadump_area_size();
574 if ((memory_limit + size) < memblock_end_of_DRAM())
575 memory_limit += size;
577 memory_limit = memblock_end_of_DRAM();
578 printk(KERN_INFO "Adjusted memory_limit for firmware-assisted"
579 " dump, now %#016llx\n", memory_limit);
582 mem_boundary = memory_limit;
584 mem_boundary = memblock_end_of_DRAM();
586 base = fw_dump.boot_mem_top;
587 size = get_fadump_area_size();
588 fw_dump.reserve_dump_area_size = size;
589 if (fw_dump.dump_active) {
590 pr_info("Firmware-assisted dump is active.\n");
592 #ifdef CONFIG_HUGETLB_PAGE
594 * FADump capture kernel doesn't care much about hugepages.
595 * In fact, handling hugepages in capture kernel is asking for
596 * trouble. So, disable HugeTLB support when fadump is active.
598 hugetlb_disabled = true;
601 * If last boot has crashed then reserve all the memory
602 * above boot memory size so that we don't touch it until
603 * dump is written to disk by userspace tool. This memory
604 * can be released for general use by invalidating fadump.
606 fadump_reserve_crash_area(base);
608 pr_debug("fadumphdr_addr = %#016lx\n", fw_dump.fadumphdr_addr);
609 pr_debug("Reserve dump area start address: 0x%lx\n",
610 fw_dump.reserve_dump_area_start);
613 * Reserve memory at an offset closer to bottom of the RAM to
614 * minimize the impact of memory hot-remove operation.
616 base = fadump_locate_reserve_mem(base, size);
618 if (!base || (base + size > mem_boundary)) {
619 pr_err("Failed to find memory chunk for reservation!\n");
622 fw_dump.reserve_dump_area_start = base;
625 * Calculate the kernel metadata address and register it with
626 * f/w if the platform supports.
628 if (fw_dump.ops->fadump_setup_metadata &&
629 (fw_dump.ops->fadump_setup_metadata(&fw_dump) < 0))
632 if (memblock_reserve(base, size)) {
633 pr_err("Failed to reserve memory!\n");
637 pr_info("Reserved %lldMB of memory at %#016llx (System RAM: %lldMB)\n",
638 (size >> 20), base, (memblock_phys_mem_size() >> 20));
640 ret = fadump_cma_init();
645 fw_dump.fadump_enabled = 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)) == INTERRUPT_SYSTEM_RESET) {
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));
866 static inline int fadump_add_mem_range(struct fadump_mrange_info *mrange_info,
869 struct fadump_memory_range *mem_ranges = mrange_info->mem_ranges;
870 bool is_adjacent = false;
877 * Fold adjacent memory ranges to bring down the memory ranges/
878 * PT_LOAD segments count.
880 if (mrange_info->mem_range_cnt) {
881 start = mem_ranges[mrange_info->mem_range_cnt - 1].base;
882 size = mem_ranges[mrange_info->mem_range_cnt - 1].size;
884 if ((start + size) == base)
888 /* resize the array on reaching the limit */
889 if (mrange_info->mem_range_cnt == mrange_info->max_mem_ranges) {
892 if (mrange_info->is_static) {
893 pr_err("Reached array size limit for %s memory ranges\n",
898 ret = fadump_alloc_mem_ranges(mrange_info);
902 /* Update to the new resized array */
903 mem_ranges = mrange_info->mem_ranges;
907 mem_ranges[mrange_info->mem_range_cnt].base = start;
908 mrange_info->mem_range_cnt++;
911 mem_ranges[mrange_info->mem_range_cnt - 1].size = (end - start);
912 pr_debug("%s_memory_range[%d] [%#016llx-%#016llx], %#llx bytes\n",
913 mrange_info->name, (mrange_info->mem_range_cnt - 1),
914 start, end - 1, (end - start));
918 static int fadump_exclude_reserved_area(u64 start, u64 end)
920 u64 ra_start, ra_end;
923 ra_start = fw_dump.reserve_dump_area_start;
924 ra_end = ra_start + fw_dump.reserve_dump_area_size;
926 if ((ra_start < end) && (ra_end > start)) {
927 if ((start < ra_start) && (end > ra_end)) {
928 ret = fadump_add_mem_range(&crash_mrange_info,
933 ret = fadump_add_mem_range(&crash_mrange_info,
935 } else if (start < ra_start) {
936 ret = fadump_add_mem_range(&crash_mrange_info,
938 } else if (ra_end < end) {
939 ret = fadump_add_mem_range(&crash_mrange_info,
943 ret = fadump_add_mem_range(&crash_mrange_info, start, end);
948 static int fadump_init_elfcore_header(char *bufp)
952 elf = (struct elfhdr *) bufp;
953 bufp += sizeof(struct elfhdr);
954 memcpy(elf->e_ident, ELFMAG, SELFMAG);
955 elf->e_ident[EI_CLASS] = ELF_CLASS;
956 elf->e_ident[EI_DATA] = ELF_DATA;
957 elf->e_ident[EI_VERSION] = EV_CURRENT;
958 elf->e_ident[EI_OSABI] = ELF_OSABI;
959 memset(elf->e_ident+EI_PAD, 0, EI_NIDENT-EI_PAD);
960 elf->e_type = ET_CORE;
961 elf->e_machine = ELF_ARCH;
962 elf->e_version = EV_CURRENT;
964 elf->e_phoff = sizeof(struct elfhdr);
966 #if defined(_CALL_ELF)
967 elf->e_flags = _CALL_ELF;
971 elf->e_ehsize = sizeof(struct elfhdr);
972 elf->e_phentsize = sizeof(struct elf_phdr);
974 elf->e_shentsize = 0;
982 * Traverse through memblock structure and setup crash memory ranges. These
983 * ranges will be used create PT_LOAD program headers in elfcore header.
985 static int fadump_setup_crash_memory_ranges(void)
990 pr_debug("Setup crash memory ranges.\n");
991 crash_mrange_info.mem_range_cnt = 0;
994 * Boot memory region(s) registered with firmware are moved to
995 * different location at the time of crash. Create separate program
996 * header(s) for this memory chunk(s) with the correct offset.
998 for (i = 0; i < fw_dump.boot_mem_regs_cnt; i++) {
999 start = fw_dump.boot_mem_addr[i];
1000 end = start + fw_dump.boot_mem_sz[i];
1001 ret = fadump_add_mem_range(&crash_mrange_info, start, end);
1006 for_each_mem_range(i, &start, &end) {
1008 * skip the memory chunk that is already added
1009 * (0 through boot_memory_top).
1011 if (start < fw_dump.boot_mem_top) {
1012 if (end > fw_dump.boot_mem_top)
1013 start = fw_dump.boot_mem_top;
1018 /* add this range excluding the reserved dump area. */
1019 ret = fadump_exclude_reserved_area(start, end);
1028 * If the given physical address falls within the boot memory region then
1029 * return the relocated address that points to the dump region reserved
1030 * for saving initial boot memory contents.
1032 static inline unsigned long fadump_relocate(unsigned long paddr)
1034 unsigned long raddr, rstart, rend, rlast, hole_size;
1040 for (i = 0; i < fw_dump.boot_mem_regs_cnt; i++) {
1041 rstart = fw_dump.boot_mem_addr[i];
1042 rend = rstart + fw_dump.boot_mem_sz[i];
1043 hole_size += (rstart - rlast);
1045 if (paddr >= rstart && paddr < rend) {
1046 raddr += fw_dump.boot_mem_dest_addr - hole_size;
1053 pr_debug("vmcoreinfo: paddr = 0x%lx, raddr = 0x%lx\n", paddr, raddr);
1057 static int fadump_create_elfcore_headers(char *bufp)
1059 unsigned long long raddr, offset;
1060 struct elf_phdr *phdr;
1064 fadump_init_elfcore_header(bufp);
1065 elf = (struct elfhdr *)bufp;
1066 bufp += sizeof(struct elfhdr);
1069 * setup ELF PT_NOTE, place holder for cpu notes info. The notes info
1070 * will be populated during second kernel boot after crash. Hence
1071 * this PT_NOTE will always be the first elf note.
1073 * NOTE: Any new ELF note addition should be placed after this note.
1075 phdr = (struct elf_phdr *)bufp;
1076 bufp += sizeof(struct elf_phdr);
1077 phdr->p_type = PT_NOTE;
1089 /* setup ELF PT_NOTE for vmcoreinfo */
1090 phdr = (struct elf_phdr *)bufp;
1091 bufp += sizeof(struct elf_phdr);
1092 phdr->p_type = PT_NOTE;
1097 phdr->p_paddr = fadump_relocate(paddr_vmcoreinfo_note());
1098 phdr->p_offset = phdr->p_paddr;
1099 phdr->p_memsz = phdr->p_filesz = VMCOREINFO_NOTE_SIZE;
1101 /* Increment number of program headers. */
1104 /* setup PT_LOAD sections. */
1107 raddr = fw_dump.boot_mem_addr[0];
1108 for (i = 0; i < crash_mrange_info.mem_range_cnt; i++) {
1111 mbase = crash_mrange_info.mem_ranges[i].base;
1112 msize = crash_mrange_info.mem_ranges[i].size;
1116 phdr = (struct elf_phdr *)bufp;
1117 bufp += sizeof(struct elf_phdr);
1118 phdr->p_type = PT_LOAD;
1119 phdr->p_flags = PF_R|PF_W|PF_X;
1120 phdr->p_offset = mbase;
1122 if (mbase == raddr) {
1124 * The entire real memory region will be moved by
1125 * firmware to the specified destination_address.
1126 * Hence set the correct offset.
1128 phdr->p_offset = fw_dump.boot_mem_dest_addr + offset;
1129 if (j < (fw_dump.boot_mem_regs_cnt - 1)) {
1130 offset += fw_dump.boot_mem_sz[j];
1131 raddr = fw_dump.boot_mem_addr[++j];
1135 phdr->p_paddr = mbase;
1136 phdr->p_vaddr = (unsigned long)__va(mbase);
1137 phdr->p_filesz = msize;
1138 phdr->p_memsz = msize;
1141 /* Increment number of program headers. */
1147 static unsigned long init_fadump_header(unsigned long addr)
1149 struct fadump_crash_info_header *fdh;
1155 addr += sizeof(struct fadump_crash_info_header);
1157 memset(fdh, 0, sizeof(struct fadump_crash_info_header));
1158 fdh->magic_number = FADUMP_CRASH_INFO_MAGIC;
1159 fdh->elfcorehdr_addr = addr;
1160 /* We will set the crashing cpu id in crash_fadump() during crash. */
1161 fdh->crashing_cpu = FADUMP_CPU_UNKNOWN;
1166 static int register_fadump(void)
1173 * If no memory is reserved then we can not register for firmware-
1176 if (!fw_dump.reserve_dump_area_size)
1179 ret = fadump_setup_crash_memory_ranges();
1183 addr = fw_dump.fadumphdr_addr;
1185 /* Initialize fadump crash info header. */
1186 addr = init_fadump_header(addr);
1189 pr_debug("Creating ELF core headers at %#016lx\n", addr);
1190 fadump_create_elfcore_headers(vaddr);
1192 /* register the future kernel dump with firmware. */
1193 pr_debug("Registering for firmware-assisted kernel dump...\n");
1194 return fw_dump.ops->fadump_register(&fw_dump);
1197 void fadump_cleanup(void)
1199 if (!fw_dump.fadump_supported)
1202 /* Invalidate the registration only if dump is active. */
1203 if (fw_dump.dump_active) {
1204 pr_debug("Invalidating firmware-assisted dump registration\n");
1205 fw_dump.ops->fadump_invalidate(&fw_dump);
1206 } else if (fw_dump.dump_registered) {
1207 /* Un-register Firmware-assisted dump if it was registered. */
1208 fw_dump.ops->fadump_unregister(&fw_dump);
1209 fadump_free_mem_ranges(&crash_mrange_info);
1212 if (fw_dump.ops->fadump_cleanup)
1213 fw_dump.ops->fadump_cleanup(&fw_dump);
1216 static void fadump_free_reserved_memory(unsigned long start_pfn,
1217 unsigned long end_pfn)
1220 unsigned long time_limit = jiffies + HZ;
1222 pr_info("freeing reserved memory (0x%llx - 0x%llx)\n",
1223 PFN_PHYS(start_pfn), PFN_PHYS(end_pfn));
1225 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
1226 free_reserved_page(pfn_to_page(pfn));
1228 if (time_after(jiffies, time_limit)) {
1230 time_limit = jiffies + HZ;
1236 * Skip memory holes and free memory that was actually reserved.
1238 static void fadump_release_reserved_area(u64 start, u64 end)
1240 unsigned long reg_spfn, reg_epfn;
1241 u64 tstart, tend, spfn, epfn;
1244 spfn = PHYS_PFN(start);
1245 epfn = PHYS_PFN(end);
1247 for_each_mem_pfn_range(i, MAX_NUMNODES, ®_spfn, ®_epfn, NULL) {
1248 tstart = max_t(u64, spfn, reg_spfn);
1249 tend = min_t(u64, epfn, reg_epfn);
1251 if (tstart < tend) {
1252 fadump_free_reserved_memory(tstart, tend);
1263 * Sort the mem ranges in-place and merge adjacent ranges
1264 * to minimize the memory ranges count.
1266 static void sort_and_merge_mem_ranges(struct fadump_mrange_info *mrange_info)
1268 struct fadump_memory_range *mem_ranges;
1269 struct fadump_memory_range tmp_range;
1273 if (!reserved_mrange_info.mem_range_cnt)
1276 /* Sort the memory ranges */
1277 mem_ranges = mrange_info->mem_ranges;
1278 for (i = 0; i < mrange_info->mem_range_cnt; i++) {
1280 for (j = (i + 1); j < mrange_info->mem_range_cnt; j++) {
1281 if (mem_ranges[idx].base > mem_ranges[j].base)
1285 tmp_range = mem_ranges[idx];
1286 mem_ranges[idx] = mem_ranges[i];
1287 mem_ranges[i] = tmp_range;
1291 /* Merge adjacent reserved ranges */
1293 for (i = 1; i < mrange_info->mem_range_cnt; i++) {
1294 base = mem_ranges[i-1].base;
1295 size = mem_ranges[i-1].size;
1296 if (mem_ranges[i].base == (base + size))
1297 mem_ranges[idx].size += mem_ranges[i].size;
1303 mem_ranges[idx] = mem_ranges[i];
1306 mrange_info->mem_range_cnt = idx + 1;
1310 * Scan reserved-ranges to consider them while reserving/releasing
1311 * memory for FADump.
1313 static void __init early_init_dt_scan_reserved_ranges(unsigned long node)
1319 /* reserved-ranges already scanned */
1320 if (reserved_mrange_info.mem_range_cnt != 0)
1323 prop = of_get_flat_dt_prop(node, "reserved-ranges", &len);
1328 * Each reserved range is an (address,size) pair, 2 cells each,
1329 * totalling 4 cells per range.
1331 for (i = 0; i < len / (sizeof(*prop) * 4); i++) {
1334 base = of_read_number(prop + (i * 4) + 0, 2);
1335 size = of_read_number(prop + (i * 4) + 2, 2);
1338 ret = fadump_add_mem_range(&reserved_mrange_info,
1341 pr_warn("some reserved ranges are ignored!\n");
1347 /* Compact reserved ranges */
1348 sort_and_merge_mem_ranges(&reserved_mrange_info);
1352 * Release the memory that was reserved during early boot to preserve the
1353 * crash'ed kernel's memory contents except reserved dump area (permanent
1354 * reservation) and reserved ranges used by F/W. The released memory will
1355 * be available for general use.
1357 static void fadump_release_memory(u64 begin, u64 end)
1359 u64 ra_start, ra_end, tstart;
1362 ra_start = fw_dump.reserve_dump_area_start;
1363 ra_end = ra_start + fw_dump.reserve_dump_area_size;
1366 * If reserved ranges array limit is hit, overwrite the last reserved
1367 * memory range with reserved dump area to ensure it is excluded from
1368 * the memory being released (reused for next FADump registration).
1370 if (reserved_mrange_info.mem_range_cnt ==
1371 reserved_mrange_info.max_mem_ranges)
1372 reserved_mrange_info.mem_range_cnt--;
1374 ret = fadump_add_mem_range(&reserved_mrange_info, ra_start, ra_end);
1378 /* Get the reserved ranges list in order first. */
1379 sort_and_merge_mem_ranges(&reserved_mrange_info);
1381 /* Exclude reserved ranges and release remaining memory */
1383 for (i = 0; i < reserved_mrange_info.mem_range_cnt; i++) {
1384 ra_start = reserved_mrange_info.mem_ranges[i].base;
1385 ra_end = ra_start + reserved_mrange_info.mem_ranges[i].size;
1387 if (tstart >= ra_end)
1390 if (tstart < ra_start)
1391 fadump_release_reserved_area(tstart, ra_start);
1396 fadump_release_reserved_area(tstart, end);
1399 static void fadump_invalidate_release_mem(void)
1401 mutex_lock(&fadump_mutex);
1402 if (!fw_dump.dump_active) {
1403 mutex_unlock(&fadump_mutex);
1408 mutex_unlock(&fadump_mutex);
1410 fadump_release_memory(fw_dump.boot_mem_top, memblock_end_of_DRAM());
1411 fadump_free_cpu_notes_buf();
1414 * Setup kernel metadata and initialize the kernel dump
1415 * memory structure for FADump re-registration.
1417 if (fw_dump.ops->fadump_setup_metadata &&
1418 (fw_dump.ops->fadump_setup_metadata(&fw_dump) < 0))
1419 pr_warn("Failed to setup kernel metadata!\n");
1420 fw_dump.ops->fadump_init_mem_struct(&fw_dump);
1423 static ssize_t release_mem_store(struct kobject *kobj,
1424 struct kobj_attribute *attr,
1425 const char *buf, size_t count)
1429 if (!fw_dump.dump_active)
1432 if (kstrtoint(buf, 0, &input))
1437 * Take away the '/proc/vmcore'. We are releasing the dump
1438 * memory, hence it will not be valid anymore.
1440 #ifdef CONFIG_PROC_VMCORE
1443 fadump_invalidate_release_mem();
1450 /* Release the reserved memory and disable the FADump */
1451 static void unregister_fadump(void)
1454 fadump_release_memory(fw_dump.reserve_dump_area_start,
1455 fw_dump.reserve_dump_area_size);
1456 fw_dump.fadump_enabled = 0;
1457 kobject_put(fadump_kobj);
1460 static ssize_t enabled_show(struct kobject *kobj,
1461 struct kobj_attribute *attr,
1464 return sprintf(buf, "%d\n", fw_dump.fadump_enabled);
1467 static ssize_t mem_reserved_show(struct kobject *kobj,
1468 struct kobj_attribute *attr,
1471 return sprintf(buf, "%ld\n", fw_dump.reserve_dump_area_size);
1474 static ssize_t registered_show(struct kobject *kobj,
1475 struct kobj_attribute *attr,
1478 return sprintf(buf, "%d\n", fw_dump.dump_registered);
1481 static ssize_t registered_store(struct kobject *kobj,
1482 struct kobj_attribute *attr,
1483 const char *buf, size_t count)
1488 if (!fw_dump.fadump_enabled || fw_dump.dump_active)
1491 if (kstrtoint(buf, 0, &input))
1494 mutex_lock(&fadump_mutex);
1498 if (fw_dump.dump_registered == 0) {
1502 /* Un-register Firmware-assisted dump */
1503 pr_debug("Un-register firmware-assisted dump\n");
1504 fw_dump.ops->fadump_unregister(&fw_dump);
1507 if (fw_dump.dump_registered == 1) {
1508 /* Un-register Firmware-assisted dump */
1509 fw_dump.ops->fadump_unregister(&fw_dump);
1511 /* Register Firmware-assisted dump */
1512 ret = register_fadump();
1520 mutex_unlock(&fadump_mutex);
1521 return ret < 0 ? ret : count;
1524 static int fadump_region_show(struct seq_file *m, void *private)
1526 if (!fw_dump.fadump_enabled)
1529 mutex_lock(&fadump_mutex);
1530 fw_dump.ops->fadump_region_show(&fw_dump, m);
1531 mutex_unlock(&fadump_mutex);
1535 static struct kobj_attribute release_attr = __ATTR_WO(release_mem);
1536 static struct kobj_attribute enable_attr = __ATTR_RO(enabled);
1537 static struct kobj_attribute register_attr = __ATTR_RW(registered);
1538 static struct kobj_attribute mem_reserved_attr = __ATTR_RO(mem_reserved);
1540 static struct attribute *fadump_attrs[] = {
1542 ®ister_attr.attr,
1543 &mem_reserved_attr.attr,
1547 ATTRIBUTE_GROUPS(fadump);
1549 DEFINE_SHOW_ATTRIBUTE(fadump_region);
1551 static void fadump_init_files(void)
1555 fadump_kobj = kobject_create_and_add("fadump", kernel_kobj);
1557 pr_err("failed to create fadump kobject\n");
1561 debugfs_create_file("fadump_region", 0444, powerpc_debugfs_root, NULL,
1562 &fadump_region_fops);
1564 if (fw_dump.dump_active) {
1565 rc = sysfs_create_file(fadump_kobj, &release_attr.attr);
1567 pr_err("unable to create release_mem sysfs file (%d)\n",
1571 rc = sysfs_create_groups(fadump_kobj, fadump_groups);
1573 pr_err("sysfs group creation failed (%d), unregistering FADump",
1575 unregister_fadump();
1580 * The FADump sysfs are moved from kernel_kobj to fadump_kobj need to
1581 * create symlink at old location to maintain backward compatibility.
1583 * - fadump_enabled -> fadump/enabled
1584 * - fadump_registered -> fadump/registered
1585 * - fadump_release_mem -> fadump/release_mem
1587 rc = compat_only_sysfs_link_entry_to_kobj(kernel_kobj, fadump_kobj,
1588 "enabled", "fadump_enabled");
1590 pr_err("unable to create fadump_enabled symlink (%d)", rc);
1594 rc = compat_only_sysfs_link_entry_to_kobj(kernel_kobj, fadump_kobj,
1596 "fadump_registered");
1598 pr_err("unable to create fadump_registered symlink (%d)", rc);
1599 sysfs_remove_link(kernel_kobj, "fadump_enabled");
1603 if (fw_dump.dump_active) {
1604 rc = compat_only_sysfs_link_entry_to_kobj(kernel_kobj,
1607 "fadump_release_mem");
1609 pr_err("unable to create fadump_release_mem symlink (%d)",
1616 * Prepare for firmware-assisted dump.
1618 int __init setup_fadump(void)
1620 if (!fw_dump.fadump_supported)
1623 fadump_init_files();
1624 fadump_show_config();
1626 if (!fw_dump.fadump_enabled)
1630 * If dump data is available then see if it is valid and prepare for
1631 * saving it to the disk.
1633 if (fw_dump.dump_active) {
1635 * if dump process fails then invalidate the registration
1636 * and release memory before proceeding for re-registration.
1638 if (fw_dump.ops->fadump_process(&fw_dump) < 0)
1639 fadump_invalidate_release_mem();
1641 /* Initialize the kernel dump memory structure for FAD registration. */
1642 else if (fw_dump.reserve_dump_area_size)
1643 fw_dump.ops->fadump_init_mem_struct(&fw_dump);
1647 subsys_initcall(setup_fadump);
1648 #else /* !CONFIG_PRESERVE_FA_DUMP */
1650 /* Scan the Firmware Assisted dump configuration details. */
1651 int __init early_init_dt_scan_fw_dump(unsigned long node, const char *uname,
1652 int depth, void *data)
1654 if ((depth != 1) || (strcmp(uname, "ibm,opal") != 0))
1657 opal_fadump_dt_scan(&fw_dump, node);
1662 * When dump is active but PRESERVE_FA_DUMP is enabled on the kernel,
1663 * preserve crash data. The subsequent memory preserving kernel boot
1664 * is likely to process this crash data.
1666 int __init fadump_reserve_mem(void)
1668 if (fw_dump.dump_active) {
1670 * If last boot has crashed then reserve all the memory
1671 * above boot memory to preserve crash data.
1673 pr_info("Preserving crash data for processing in next boot.\n");
1674 fadump_reserve_crash_area(fw_dump.boot_mem_top);
1676 pr_debug("FADump-aware kernel..\n");
1680 #endif /* CONFIG_PRESERVE_FA_DUMP */
1682 /* Preserve everything above the base address */
1683 static void __init fadump_reserve_crash_area(u64 base)
1685 u64 i, mstart, mend, msize;
1687 for_each_mem_range(i, &mstart, &mend) {
1688 msize = mend - mstart;
1690 if ((mstart + msize) < base)
1693 if (mstart < base) {
1694 msize -= (base - mstart);
1698 pr_info("Reserving %lluMB of memory at %#016llx for preserving crash data",
1699 (msize >> 20), mstart);
1700 memblock_reserve(mstart, msize);
1704 unsigned long __init arch_reserved_kernel_pages(void)
1706 return memblock_reserved_size() / PAGE_SIZE;