2 * Firmware Assisted dump: A robust mechanism to get reliable kernel crash
3 * dump with assistance from firmware. This approach does not use kexec,
4 * instead firmware assists in booting the kdump kernel while preserving
5 * memory contents. The most of the code implementation has been adapted
6 * from phyp assisted dump implementation written by Linas Vepstas and
9 * This program is free software; you can redistribute it and/or modify
10 * it under the terms of the GNU General Public License as published by
11 * the Free Software Foundation; either version 2 of the License, or
12 * (at your option) any later version.
14 * This program is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
17 * GNU General Public License for more details.
19 * You should have received a copy of the GNU General Public License
20 * along with this program; if not, write to the Free Software
21 * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
23 * Copyright 2011 IBM Corporation
24 * Author: Mahesh Salgaonkar <mahesh@linux.vnet.ibm.com>
28 #define pr_fmt(fmt) "fadump: " fmt
30 #include <linux/string.h>
31 #include <linux/memblock.h>
32 #include <linux/delay.h>
33 #include <linux/seq_file.h>
34 #include <linux/crash_dump.h>
35 #include <linux/kobject.h>
36 #include <linux/sysfs.h>
38 #include <asm/debugfs.h>
42 #include <asm/fadump.h>
43 #include <asm/setup.h>
45 static struct fw_dump fw_dump;
46 static struct fadump_mem_struct fdm;
47 static const struct fadump_mem_struct *fdm_active;
49 static DEFINE_MUTEX(fadump_mutex);
50 struct fad_crash_memory_ranges crash_memory_ranges[INIT_CRASHMEM_RANGES];
53 /* Scan the Firmware Assisted dump configuration details. */
54 int __init early_init_dt_scan_fw_dump(unsigned long node,
55 const char *uname, int depth, void *data)
57 const __be32 *sections;
62 if (depth != 1 || strcmp(uname, "rtas") != 0)
66 * Check if Firmware Assisted dump is supported. if yes, check
67 * if dump has been initiated on last reboot.
69 token = of_get_flat_dt_prop(node, "ibm,configure-kernel-dump", NULL);
73 fw_dump.fadump_supported = 1;
74 fw_dump.ibm_configure_kernel_dump = be32_to_cpu(*token);
77 * The 'ibm,kernel-dump' rtas node is present only if there is
78 * dump data waiting for us.
80 fdm_active = of_get_flat_dt_prop(node, "ibm,kernel-dump", NULL);
82 fw_dump.dump_active = 1;
84 /* Get the sizes required to store dump data for the firmware provided
86 * For each dump section type supported, a 32bit cell which defines
87 * the ID of a supported section followed by two 32 bit cells which
88 * gives teh size of the section in bytes.
90 sections = of_get_flat_dt_prop(node, "ibm,configure-kernel-dump-sizes",
96 num_sections = size / (3 * sizeof(u32));
98 for (i = 0; i < num_sections; i++, sections += 3) {
99 u32 type = (u32)of_read_number(sections, 1);
102 case FADUMP_CPU_STATE_DATA:
103 fw_dump.cpu_state_data_size =
104 of_read_ulong(§ions[1], 2);
106 case FADUMP_HPTE_REGION:
107 fw_dump.hpte_region_size =
108 of_read_ulong(§ions[1], 2);
117 * If fadump is registered, check if the memory provided
118 * falls within boot memory area.
120 int is_fadump_boot_memory_area(u64 addr, ulong size)
122 if (!fw_dump.dump_registered)
125 return (addr + size) > RMA_START && addr <= fw_dump.boot_memory_size;
128 int should_fadump_crash(void)
130 if (!fw_dump.dump_registered || !fw_dump.fadumphdr_addr)
135 int is_fadump_active(void)
137 return fw_dump.dump_active;
141 * Returns 1, if there are no holes in boot memory area,
144 static int is_boot_memory_area_contiguous(void)
146 struct memblock_region *reg;
147 unsigned long tstart, tend;
148 unsigned long start_pfn = PHYS_PFN(RMA_START);
149 unsigned long end_pfn = PHYS_PFN(RMA_START + fw_dump.boot_memory_size);
150 unsigned int ret = 0;
152 for_each_memblock(memory, reg) {
153 tstart = max(start_pfn, memblock_region_memory_base_pfn(reg));
154 tend = min(end_pfn, memblock_region_memory_end_pfn(reg));
156 /* Memory hole from start_pfn to tstart */
157 if (tstart > start_pfn)
160 if (tend == end_pfn) {
165 start_pfn = tend + 1;
172 /* Print firmware assisted dump configurations for debugging purpose. */
173 static void fadump_show_config(void)
175 pr_debug("Support for firmware-assisted dump (fadump): %s\n",
176 (fw_dump.fadump_supported ? "present" : "no support"));
178 if (!fw_dump.fadump_supported)
181 pr_debug("Fadump enabled : %s\n",
182 (fw_dump.fadump_enabled ? "yes" : "no"));
183 pr_debug("Dump Active : %s\n",
184 (fw_dump.dump_active ? "yes" : "no"));
185 pr_debug("Dump section sizes:\n");
186 pr_debug(" CPU state data size: %lx\n", fw_dump.cpu_state_data_size);
187 pr_debug(" HPTE region size : %lx\n", fw_dump.hpte_region_size);
188 pr_debug("Boot memory size : %lx\n", fw_dump.boot_memory_size);
191 static unsigned long init_fadump_mem_struct(struct fadump_mem_struct *fdm,
197 memset(fdm, 0, sizeof(struct fadump_mem_struct));
198 addr = addr & PAGE_MASK;
200 fdm->header.dump_format_version = cpu_to_be32(0x00000001);
201 fdm->header.dump_num_sections = cpu_to_be16(3);
202 fdm->header.dump_status_flag = 0;
203 fdm->header.offset_first_dump_section =
204 cpu_to_be32((u32)offsetof(struct fadump_mem_struct, cpu_state_data));
207 * Fields for disk dump option.
208 * We are not using disk dump option, hence set these fields to 0.
210 fdm->header.dd_block_size = 0;
211 fdm->header.dd_block_offset = 0;
212 fdm->header.dd_num_blocks = 0;
213 fdm->header.dd_offset_disk_path = 0;
215 /* set 0 to disable an automatic dump-reboot. */
216 fdm->header.max_time_auto = 0;
218 /* Kernel dump sections */
219 /* cpu state data section. */
220 fdm->cpu_state_data.request_flag = cpu_to_be32(FADUMP_REQUEST_FLAG);
221 fdm->cpu_state_data.source_data_type = cpu_to_be16(FADUMP_CPU_STATE_DATA);
222 fdm->cpu_state_data.source_address = 0;
223 fdm->cpu_state_data.source_len = cpu_to_be64(fw_dump.cpu_state_data_size);
224 fdm->cpu_state_data.destination_address = cpu_to_be64(addr);
225 addr += fw_dump.cpu_state_data_size;
227 /* hpte region section */
228 fdm->hpte_region.request_flag = cpu_to_be32(FADUMP_REQUEST_FLAG);
229 fdm->hpte_region.source_data_type = cpu_to_be16(FADUMP_HPTE_REGION);
230 fdm->hpte_region.source_address = 0;
231 fdm->hpte_region.source_len = cpu_to_be64(fw_dump.hpte_region_size);
232 fdm->hpte_region.destination_address = cpu_to_be64(addr);
233 addr += fw_dump.hpte_region_size;
235 /* RMA region section */
236 fdm->rmr_region.request_flag = cpu_to_be32(FADUMP_REQUEST_FLAG);
237 fdm->rmr_region.source_data_type = cpu_to_be16(FADUMP_REAL_MODE_REGION);
238 fdm->rmr_region.source_address = cpu_to_be64(RMA_START);
239 fdm->rmr_region.source_len = cpu_to_be64(fw_dump.boot_memory_size);
240 fdm->rmr_region.destination_address = cpu_to_be64(addr);
241 addr += fw_dump.boot_memory_size;
247 * fadump_calculate_reserve_size(): reserve variable boot area 5% of System RAM
249 * Function to find the largest memory size we need to reserve during early
250 * boot process. This will be the size of the memory that is required for a
251 * kernel to boot successfully.
253 * This function has been taken from phyp-assisted dump feature implementation.
255 * returns larger of 256MB or 5% rounded down to multiples of 256MB.
257 * TODO: Come up with better approach to find out more accurate memory size
258 * that is required for a kernel to boot successfully.
261 static inline unsigned long fadump_calculate_reserve_size(void)
264 unsigned long long base, size;
266 if (fw_dump.reserve_bootvar)
267 pr_warn("'fadump_reserve_mem=' parameter is deprecated in favor of 'crashkernel=' parameter.\n");
270 * Check if the size is specified through crashkernel= cmdline
271 * option. If yes, then use that but ignore base as fadump reserves
272 * memory at a predefined offset.
274 ret = parse_crashkernel(boot_command_line, memblock_phys_mem_size(),
276 if (ret == 0 && size > 0) {
277 unsigned long max_size;
279 if (fw_dump.reserve_bootvar)
280 pr_info("Using 'crashkernel=' parameter for memory reservation.\n");
282 fw_dump.reserve_bootvar = (unsigned long)size;
285 * Adjust if the boot memory size specified is above
288 max_size = memblock_phys_mem_size() / MAX_BOOT_MEM_RATIO;
289 if (fw_dump.reserve_bootvar > max_size) {
290 fw_dump.reserve_bootvar = max_size;
291 pr_info("Adjusted boot memory size to %luMB\n",
292 (fw_dump.reserve_bootvar >> 20));
295 return fw_dump.reserve_bootvar;
296 } else if (fw_dump.reserve_bootvar) {
298 * 'fadump_reserve_mem=' is being used to reserve memory
299 * for firmware-assisted dump.
301 return fw_dump.reserve_bootvar;
304 /* divide by 20 to get 5% of value */
305 size = memblock_phys_mem_size() / 20;
307 /* round it down in multiples of 256 */
308 size = size & ~0x0FFFFFFFUL;
310 /* Truncate to memory_limit. We don't want to over reserve the memory.*/
311 if (memory_limit && size > memory_limit)
314 return (size > MIN_BOOT_MEM ? size : MIN_BOOT_MEM);
318 * Calculate the total memory size required to be reserved for
319 * firmware-assisted dump registration.
321 static unsigned long get_fadump_area_size(void)
323 unsigned long size = 0;
325 size += fw_dump.cpu_state_data_size;
326 size += fw_dump.hpte_region_size;
327 size += fw_dump.boot_memory_size;
328 size += sizeof(struct fadump_crash_info_header);
329 size += sizeof(struct elfhdr); /* ELF core header.*/
330 size += sizeof(struct elf_phdr); /* place holder for cpu notes */
331 /* Program headers for crash memory regions. */
332 size += sizeof(struct elf_phdr) * (memblock_num_regions(memory) + 2);
334 size = PAGE_ALIGN(size);
338 int __init fadump_reserve_mem(void)
340 unsigned long base, size, memory_boundary;
342 if (!fw_dump.fadump_enabled)
345 if (!fw_dump.fadump_supported) {
346 printk(KERN_INFO "Firmware-assisted dump is not supported on"
348 fw_dump.fadump_enabled = 0;
352 * Initialize boot memory size
353 * If dump is active then we have already calculated the size during
357 fw_dump.boot_memory_size = be64_to_cpu(fdm_active->rmr_region.source_len);
359 fw_dump.boot_memory_size = fadump_calculate_reserve_size();
362 * Calculate the memory boundary.
363 * If memory_limit is less than actual memory boundary then reserve
364 * the memory for fadump beyond the memory_limit and adjust the
365 * memory_limit accordingly, so that the running kernel can run with
366 * specified memory_limit.
368 if (memory_limit && memory_limit < memblock_end_of_DRAM()) {
369 size = get_fadump_area_size();
370 if ((memory_limit + size) < memblock_end_of_DRAM())
371 memory_limit += size;
373 memory_limit = memblock_end_of_DRAM();
374 printk(KERN_INFO "Adjusted memory_limit for firmware-assisted"
375 " dump, now %#016llx\n", memory_limit);
378 memory_boundary = memory_limit;
380 memory_boundary = memblock_end_of_DRAM();
382 if (fw_dump.dump_active) {
383 printk(KERN_INFO "Firmware-assisted dump is active.\n");
385 * If last boot has crashed then reserve all the memory
386 * above boot_memory_size so that we don't touch it until
387 * dump is written to disk by userspace tool. This memory
388 * will be released for general use once the dump is saved.
390 base = fw_dump.boot_memory_size;
391 size = memory_boundary - base;
392 memblock_reserve(base, size);
393 printk(KERN_INFO "Reserved %ldMB of memory at %ldMB "
394 "for saving crash dump\n",
395 (unsigned long)(size >> 20),
396 (unsigned long)(base >> 20));
398 fw_dump.fadumphdr_addr =
399 be64_to_cpu(fdm_active->rmr_region.destination_address) +
400 be64_to_cpu(fdm_active->rmr_region.source_len);
401 pr_debug("fadumphdr_addr = %p\n",
402 (void *) fw_dump.fadumphdr_addr);
404 size = get_fadump_area_size();
407 * Reserve memory at an offset closer to bottom of the RAM to
408 * minimize the impact of memory hot-remove operation. We can't
409 * use memblock_find_in_range() here since it doesn't allocate
410 * from bottom to top.
412 for (base = fw_dump.boot_memory_size;
413 base <= (memory_boundary - size);
415 if (memblock_is_region_memory(base, size) &&
416 !memblock_is_region_reserved(base, size))
419 if ((base > (memory_boundary - size)) ||
420 memblock_reserve(base, size)) {
421 pr_err("Failed to reserve memory\n");
425 pr_info("Reserved %ldMB of memory at %ldMB for firmware-"
426 "assisted dump (System RAM: %ldMB)\n",
427 (unsigned long)(size >> 20),
428 (unsigned long)(base >> 20),
429 (unsigned long)(memblock_phys_mem_size() >> 20));
432 fw_dump.reserve_dump_area_start = base;
433 fw_dump.reserve_dump_area_size = size;
437 unsigned long __init arch_reserved_kernel_pages(void)
439 return memblock_reserved_size() / PAGE_SIZE;
442 /* Look for fadump= cmdline option. */
443 static int __init early_fadump_param(char *p)
448 if (strncmp(p, "on", 2) == 0)
449 fw_dump.fadump_enabled = 1;
450 else if (strncmp(p, "off", 3) == 0)
451 fw_dump.fadump_enabled = 0;
455 early_param("fadump", early_fadump_param);
458 * Look for fadump_reserve_mem= cmdline option
459 * TODO: Remove references to 'fadump_reserve_mem=' parameter,
460 * the sooner 'crashkernel=' parameter is accustomed to.
462 static int __init early_fadump_reserve_mem(char *p)
465 fw_dump.reserve_bootvar = memparse(p, &p);
468 early_param("fadump_reserve_mem", early_fadump_reserve_mem);
470 static int register_fw_dump(struct fadump_mem_struct *fdm)
473 unsigned int wait_time;
475 pr_debug("Registering for firmware-assisted kernel dump...\n");
477 /* TODO: Add upper time limit for the delay */
479 rc = rtas_call(fw_dump.ibm_configure_kernel_dump, 3, 1, NULL,
480 FADUMP_REGISTER, fdm,
481 sizeof(struct fadump_mem_struct));
483 wait_time = rtas_busy_delay_time(rc);
492 pr_err("Failed to register. Unknown Error(%d).\n", rc);
495 printk(KERN_ERR "Failed to register firmware-assisted kernel"
496 " dump. Hardware Error(%d).\n", rc);
499 if (!is_boot_memory_area_contiguous())
500 pr_err("Can't have holes in boot memory area while "
501 "registering fadump\n");
503 printk(KERN_ERR "Failed to register firmware-assisted kernel"
504 " dump. Parameter Error(%d).\n", rc);
508 printk(KERN_ERR "firmware-assisted kernel dump is already "
510 fw_dump.dump_registered = 1;
514 printk(KERN_INFO "firmware-assisted kernel dump registration"
516 fw_dump.dump_registered = 1;
523 void crash_fadump(struct pt_regs *regs, const char *str)
525 struct fadump_crash_info_header *fdh = NULL;
526 int old_cpu, this_cpu;
528 if (!should_fadump_crash())
532 * old_cpu == -1 means this is the first CPU which has come here,
533 * go ahead and trigger fadump.
535 * old_cpu != -1 means some other CPU has already on it's way
536 * to trigger fadump, just keep looping here.
538 this_cpu = smp_processor_id();
539 old_cpu = cmpxchg(&crashing_cpu, -1, this_cpu);
543 * We can't loop here indefinitely. Wait as long as fadump
544 * is in force. If we race with fadump un-registration this
545 * loop will break and then we go down to normal panic path
546 * and reboot. If fadump is in force the first crashing
547 * cpu will definitely trigger fadump.
549 while (fw_dump.dump_registered)
554 fdh = __va(fw_dump.fadumphdr_addr);
555 fdh->crashing_cpu = crashing_cpu;
556 crash_save_vmcoreinfo();
561 ppc_save_regs(&fdh->regs);
563 fdh->online_mask = *cpu_online_mask;
565 /* Call ibm,os-term rtas call to trigger firmware assisted dump */
566 rtas_os_term((char *)str);
569 #define GPR_MASK 0xffffff0000000000
570 static inline int fadump_gpr_index(u64 id)
575 if ((id & GPR_MASK) == REG_ID("GPR")) {
576 /* get the digits at the end */
581 str[0] = (id >> 8) & 0xff;
582 sscanf(str, "%d", &i);
589 static inline void fadump_set_regval(struct pt_regs *regs, u64 reg_id,
594 i = fadump_gpr_index(reg_id);
596 regs->gpr[i] = (unsigned long)reg_val;
597 else if (reg_id == REG_ID("NIA"))
598 regs->nip = (unsigned long)reg_val;
599 else if (reg_id == REG_ID("MSR"))
600 regs->msr = (unsigned long)reg_val;
601 else if (reg_id == REG_ID("CTR"))
602 regs->ctr = (unsigned long)reg_val;
603 else if (reg_id == REG_ID("LR"))
604 regs->link = (unsigned long)reg_val;
605 else if (reg_id == REG_ID("XER"))
606 regs->xer = (unsigned long)reg_val;
607 else if (reg_id == REG_ID("CR"))
608 regs->ccr = (unsigned long)reg_val;
609 else if (reg_id == REG_ID("DAR"))
610 regs->dar = (unsigned long)reg_val;
611 else if (reg_id == REG_ID("DSISR"))
612 regs->dsisr = (unsigned long)reg_val;
615 static struct fadump_reg_entry*
616 fadump_read_registers(struct fadump_reg_entry *reg_entry, struct pt_regs *regs)
618 memset(regs, 0, sizeof(struct pt_regs));
620 while (be64_to_cpu(reg_entry->reg_id) != REG_ID("CPUEND")) {
621 fadump_set_regval(regs, be64_to_cpu(reg_entry->reg_id),
622 be64_to_cpu(reg_entry->reg_value));
629 static u32 *fadump_regs_to_elf_notes(u32 *buf, struct pt_regs *regs)
631 struct elf_prstatus prstatus;
633 memset(&prstatus, 0, sizeof(prstatus));
635 * FIXME: How do i get PID? Do I really need it?
636 * prstatus.pr_pid = ????
638 elf_core_copy_kernel_regs(&prstatus.pr_reg, regs);
639 buf = append_elf_note(buf, CRASH_CORE_NOTE_NAME, NT_PRSTATUS,
640 &prstatus, sizeof(prstatus));
644 static void fadump_update_elfcore_header(char *bufp)
647 struct elf_phdr *phdr;
649 elf = (struct elfhdr *)bufp;
650 bufp += sizeof(struct elfhdr);
652 /* First note is a place holder for cpu notes info. */
653 phdr = (struct elf_phdr *)bufp;
655 if (phdr->p_type == PT_NOTE) {
656 phdr->p_paddr = fw_dump.cpu_notes_buf;
657 phdr->p_offset = phdr->p_paddr;
658 phdr->p_filesz = fw_dump.cpu_notes_buf_size;
659 phdr->p_memsz = fw_dump.cpu_notes_buf_size;
664 static void *fadump_cpu_notes_buf_alloc(unsigned long size)
668 unsigned long order, count, i;
670 order = get_order(size);
671 vaddr = (void *)__get_free_pages(GFP_KERNEL|__GFP_ZERO, order);
676 page = virt_to_page(vaddr);
677 for (i = 0; i < count; i++)
678 SetPageReserved(page + i);
682 static void fadump_cpu_notes_buf_free(unsigned long vaddr, unsigned long size)
685 unsigned long order, count, i;
687 order = get_order(size);
689 page = virt_to_page(vaddr);
690 for (i = 0; i < count; i++)
691 ClearPageReserved(page + i);
692 __free_pages(page, order);
696 * Read CPU state dump data and convert it into ELF notes.
697 * The CPU dump starts with magic number "REGSAVE". NumCpusOffset should be
698 * used to access the data to allow for additional fields to be added without
699 * affecting compatibility. Each list of registers for a CPU starts with
700 * "CPUSTRT" and ends with "CPUEND". Each register entry is of 16 bytes,
701 * 8 Byte ASCII identifier and 8 Byte register value. The register entry
702 * with identifier "CPUSTRT" and "CPUEND" contains 4 byte cpu id as part
703 * of register value. For more details refer to PAPR document.
705 * Only for the crashing cpu we ignore the CPU dump data and get exact
706 * state from fadump crash info structure populated by first kernel at the
709 static int __init fadump_build_cpu_notes(const struct fadump_mem_struct *fdm)
711 struct fadump_reg_save_area_header *reg_header;
712 struct fadump_reg_entry *reg_entry;
713 struct fadump_crash_info_header *fdh = NULL;
716 u32 num_cpus, *note_buf;
718 int i, rc = 0, cpu = 0;
720 if (!fdm->cpu_state_data.bytes_dumped)
723 addr = be64_to_cpu(fdm->cpu_state_data.destination_address);
727 if (be64_to_cpu(reg_header->magic_number) != REGSAVE_AREA_MAGIC) {
728 printk(KERN_ERR "Unable to read register save area.\n");
731 pr_debug("--------CPU State Data------------\n");
732 pr_debug("Magic Number: %llx\n", be64_to_cpu(reg_header->magic_number));
733 pr_debug("NumCpuOffset: %x\n", be32_to_cpu(reg_header->num_cpu_offset));
735 vaddr += be32_to_cpu(reg_header->num_cpu_offset);
736 num_cpus = be32_to_cpu(*((__be32 *)(vaddr)));
737 pr_debug("NumCpus : %u\n", num_cpus);
738 vaddr += sizeof(u32);
739 reg_entry = (struct fadump_reg_entry *)vaddr;
741 /* Allocate buffer to hold cpu crash notes. */
742 fw_dump.cpu_notes_buf_size = num_cpus * sizeof(note_buf_t);
743 fw_dump.cpu_notes_buf_size = PAGE_ALIGN(fw_dump.cpu_notes_buf_size);
744 note_buf = fadump_cpu_notes_buf_alloc(fw_dump.cpu_notes_buf_size);
746 printk(KERN_ERR "Failed to allocate 0x%lx bytes for "
747 "cpu notes buffer\n", fw_dump.cpu_notes_buf_size);
750 fw_dump.cpu_notes_buf = __pa(note_buf);
752 pr_debug("Allocated buffer for cpu notes of size %ld at %p\n",
753 (num_cpus * sizeof(note_buf_t)), note_buf);
755 if (fw_dump.fadumphdr_addr)
756 fdh = __va(fw_dump.fadumphdr_addr);
758 for (i = 0; i < num_cpus; i++) {
759 if (be64_to_cpu(reg_entry->reg_id) != REG_ID("CPUSTRT")) {
760 printk(KERN_ERR "Unable to read CPU state data\n");
764 /* Lower 4 bytes of reg_value contains logical cpu id */
765 cpu = be64_to_cpu(reg_entry->reg_value) & FADUMP_CPU_ID_MASK;
766 if (fdh && !cpumask_test_cpu(cpu, &fdh->online_mask)) {
767 SKIP_TO_NEXT_CPU(reg_entry);
770 pr_debug("Reading register data for cpu %d...\n", cpu);
771 if (fdh && fdh->crashing_cpu == cpu) {
773 note_buf = fadump_regs_to_elf_notes(note_buf, ®s);
774 SKIP_TO_NEXT_CPU(reg_entry);
777 reg_entry = fadump_read_registers(reg_entry, ®s);
778 note_buf = fadump_regs_to_elf_notes(note_buf, ®s);
781 final_note(note_buf);
784 pr_debug("Updating elfcore header (%llx) with cpu notes\n",
785 fdh->elfcorehdr_addr);
786 fadump_update_elfcore_header((char *)__va(fdh->elfcorehdr_addr));
791 fadump_cpu_notes_buf_free((unsigned long)__va(fw_dump.cpu_notes_buf),
792 fw_dump.cpu_notes_buf_size);
793 fw_dump.cpu_notes_buf = 0;
794 fw_dump.cpu_notes_buf_size = 0;
800 * Validate and process the dump data stored by firmware before exporting
801 * it through '/proc/vmcore'.
803 static int __init process_fadump(const struct fadump_mem_struct *fdm_active)
805 struct fadump_crash_info_header *fdh;
808 if (!fdm_active || !fw_dump.fadumphdr_addr)
811 /* Check if the dump data is valid. */
812 if ((be16_to_cpu(fdm_active->header.dump_status_flag) == FADUMP_ERROR_FLAG) ||
813 (fdm_active->cpu_state_data.error_flags != 0) ||
814 (fdm_active->rmr_region.error_flags != 0)) {
815 printk(KERN_ERR "Dump taken by platform is not valid\n");
818 if ((fdm_active->rmr_region.bytes_dumped !=
819 fdm_active->rmr_region.source_len) ||
820 !fdm_active->cpu_state_data.bytes_dumped) {
821 printk(KERN_ERR "Dump taken by platform is incomplete\n");
825 /* Validate the fadump crash info header */
826 fdh = __va(fw_dump.fadumphdr_addr);
827 if (fdh->magic_number != FADUMP_CRASH_INFO_MAGIC) {
828 printk(KERN_ERR "Crash info header is not valid.\n");
832 rc = fadump_build_cpu_notes(fdm_active);
837 * We are done validating dump info and elfcore header is now ready
838 * to be exported. set elfcorehdr_addr so that vmcore module will
839 * export the elfcore header through '/proc/vmcore'.
841 elfcorehdr_addr = fdh->elfcorehdr_addr;
846 static inline void fadump_add_crash_memory(unsigned long long base,
847 unsigned long long end)
852 pr_debug("crash_memory_range[%d] [%#016llx-%#016llx], %#llx bytes\n",
853 crash_mem_ranges, base, end - 1, (end - base));
854 crash_memory_ranges[crash_mem_ranges].base = base;
855 crash_memory_ranges[crash_mem_ranges].size = end - base;
859 static void fadump_exclude_reserved_area(unsigned long long start,
860 unsigned long long end)
862 unsigned long long ra_start, ra_end;
864 ra_start = fw_dump.reserve_dump_area_start;
865 ra_end = ra_start + fw_dump.reserve_dump_area_size;
867 if ((ra_start < end) && (ra_end > start)) {
868 if ((start < ra_start) && (end > ra_end)) {
869 fadump_add_crash_memory(start, ra_start);
870 fadump_add_crash_memory(ra_end, end);
871 } else if (start < ra_start) {
872 fadump_add_crash_memory(start, ra_start);
873 } else if (ra_end < end) {
874 fadump_add_crash_memory(ra_end, end);
877 fadump_add_crash_memory(start, end);
880 static int fadump_init_elfcore_header(char *bufp)
884 elf = (struct elfhdr *) bufp;
885 bufp += sizeof(struct elfhdr);
886 memcpy(elf->e_ident, ELFMAG, SELFMAG);
887 elf->e_ident[EI_CLASS] = ELF_CLASS;
888 elf->e_ident[EI_DATA] = ELF_DATA;
889 elf->e_ident[EI_VERSION] = EV_CURRENT;
890 elf->e_ident[EI_OSABI] = ELF_OSABI;
891 memset(elf->e_ident+EI_PAD, 0, EI_NIDENT-EI_PAD);
892 elf->e_type = ET_CORE;
893 elf->e_machine = ELF_ARCH;
894 elf->e_version = EV_CURRENT;
896 elf->e_phoff = sizeof(struct elfhdr);
898 #if defined(_CALL_ELF)
899 elf->e_flags = _CALL_ELF;
903 elf->e_ehsize = sizeof(struct elfhdr);
904 elf->e_phentsize = sizeof(struct elf_phdr);
906 elf->e_shentsize = 0;
914 * Traverse through memblock structure and setup crash memory ranges. These
915 * ranges will be used create PT_LOAD program headers in elfcore header.
917 static void fadump_setup_crash_memory_ranges(void)
919 struct memblock_region *reg;
920 unsigned long long start, end;
922 pr_debug("Setup crash memory ranges.\n");
923 crash_mem_ranges = 0;
925 * add the first memory chunk (RMA_START through boot_memory_size) as
926 * a separate memory chunk. The reason is, at the time crash firmware
927 * will move the content of this memory chunk to different location
928 * specified during fadump registration. We need to create a separate
929 * program header for this chunk with the correct offset.
931 fadump_add_crash_memory(RMA_START, fw_dump.boot_memory_size);
933 for_each_memblock(memory, reg) {
934 start = (unsigned long long)reg->base;
935 end = start + (unsigned long long)reg->size;
938 * skip the first memory chunk that is already added (RMA_START
939 * through boot_memory_size). This logic needs a relook if and
940 * when RMA_START changes to a non-zero value.
942 BUILD_BUG_ON(RMA_START != 0);
943 if (start < fw_dump.boot_memory_size) {
944 if (end > fw_dump.boot_memory_size)
945 start = fw_dump.boot_memory_size;
950 /* add this range excluding the reserved dump area. */
951 fadump_exclude_reserved_area(start, end);
956 * If the given physical address falls within the boot memory region then
957 * return the relocated address that points to the dump region reserved
958 * for saving initial boot memory contents.
960 static inline unsigned long fadump_relocate(unsigned long paddr)
962 if (paddr > RMA_START && paddr < fw_dump.boot_memory_size)
963 return be64_to_cpu(fdm.rmr_region.destination_address) + paddr;
968 static int fadump_create_elfcore_headers(char *bufp)
971 struct elf_phdr *phdr;
974 fadump_init_elfcore_header(bufp);
975 elf = (struct elfhdr *)bufp;
976 bufp += sizeof(struct elfhdr);
979 * setup ELF PT_NOTE, place holder for cpu notes info. The notes info
980 * will be populated during second kernel boot after crash. Hence
981 * this PT_NOTE will always be the first elf note.
983 * NOTE: Any new ELF note addition should be placed after this note.
985 phdr = (struct elf_phdr *)bufp;
986 bufp += sizeof(struct elf_phdr);
987 phdr->p_type = PT_NOTE;
999 /* setup ELF PT_NOTE for vmcoreinfo */
1000 phdr = (struct elf_phdr *)bufp;
1001 bufp += sizeof(struct elf_phdr);
1002 phdr->p_type = PT_NOTE;
1007 phdr->p_paddr = fadump_relocate(paddr_vmcoreinfo_note());
1008 phdr->p_offset = phdr->p_paddr;
1009 phdr->p_memsz = phdr->p_filesz = VMCOREINFO_NOTE_SIZE;
1011 /* Increment number of program headers. */
1014 /* setup PT_LOAD sections. */
1016 for (i = 0; i < crash_mem_ranges; i++) {
1017 unsigned long long mbase, msize;
1018 mbase = crash_memory_ranges[i].base;
1019 msize = crash_memory_ranges[i].size;
1024 phdr = (struct elf_phdr *)bufp;
1025 bufp += sizeof(struct elf_phdr);
1026 phdr->p_type = PT_LOAD;
1027 phdr->p_flags = PF_R|PF_W|PF_X;
1028 phdr->p_offset = mbase;
1030 if (mbase == RMA_START) {
1032 * The entire RMA region will be moved by firmware
1033 * to the specified destination_address. Hence set
1034 * the correct offset.
1036 phdr->p_offset = be64_to_cpu(fdm.rmr_region.destination_address);
1039 phdr->p_paddr = mbase;
1040 phdr->p_vaddr = (unsigned long)__va(mbase);
1041 phdr->p_filesz = msize;
1042 phdr->p_memsz = msize;
1045 /* Increment number of program headers. */
1051 static unsigned long init_fadump_header(unsigned long addr)
1053 struct fadump_crash_info_header *fdh;
1058 fw_dump.fadumphdr_addr = addr;
1060 addr += sizeof(struct fadump_crash_info_header);
1062 memset(fdh, 0, sizeof(struct fadump_crash_info_header));
1063 fdh->magic_number = FADUMP_CRASH_INFO_MAGIC;
1064 fdh->elfcorehdr_addr = addr;
1065 /* We will set the crashing cpu id in crash_fadump() during crash. */
1066 fdh->crashing_cpu = CPU_UNKNOWN;
1071 static int register_fadump(void)
1077 * If no memory is reserved then we can not register for firmware-
1080 if (!fw_dump.reserve_dump_area_size)
1083 fadump_setup_crash_memory_ranges();
1085 addr = be64_to_cpu(fdm.rmr_region.destination_address) + be64_to_cpu(fdm.rmr_region.source_len);
1086 /* Initialize fadump crash info header. */
1087 addr = init_fadump_header(addr);
1090 pr_debug("Creating ELF core headers at %#016lx\n", addr);
1091 fadump_create_elfcore_headers(vaddr);
1093 /* register the future kernel dump with firmware. */
1094 return register_fw_dump(&fdm);
1097 static int fadump_unregister_dump(struct fadump_mem_struct *fdm)
1100 unsigned int wait_time;
1102 pr_debug("Un-register firmware-assisted dump\n");
1104 /* TODO: Add upper time limit for the delay */
1106 rc = rtas_call(fw_dump.ibm_configure_kernel_dump, 3, 1, NULL,
1107 FADUMP_UNREGISTER, fdm,
1108 sizeof(struct fadump_mem_struct));
1110 wait_time = rtas_busy_delay_time(rc);
1113 } while (wait_time);
1116 printk(KERN_ERR "Failed to un-register firmware-assisted dump."
1117 " unexpected error(%d).\n", rc);
1120 fw_dump.dump_registered = 0;
1124 static int fadump_invalidate_dump(struct fadump_mem_struct *fdm)
1127 unsigned int wait_time;
1129 pr_debug("Invalidating firmware-assisted dump registration\n");
1131 /* TODO: Add upper time limit for the delay */
1133 rc = rtas_call(fw_dump.ibm_configure_kernel_dump, 3, 1, NULL,
1134 FADUMP_INVALIDATE, fdm,
1135 sizeof(struct fadump_mem_struct));
1137 wait_time = rtas_busy_delay_time(rc);
1140 } while (wait_time);
1143 pr_err("Failed to invalidate firmware-assisted dump registration. Unexpected error (%d).\n", rc);
1146 fw_dump.dump_active = 0;
1151 void fadump_cleanup(void)
1153 /* Invalidate the registration only if dump is active. */
1154 if (fw_dump.dump_active) {
1155 init_fadump_mem_struct(&fdm,
1156 be64_to_cpu(fdm_active->cpu_state_data.destination_address));
1157 fadump_invalidate_dump(&fdm);
1161 static void fadump_free_reserved_memory(unsigned long start_pfn,
1162 unsigned long end_pfn)
1165 unsigned long time_limit = jiffies + HZ;
1167 pr_info("freeing reserved memory (0x%llx - 0x%llx)\n",
1168 PFN_PHYS(start_pfn), PFN_PHYS(end_pfn));
1170 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
1171 free_reserved_page(pfn_to_page(pfn));
1173 if (time_after(jiffies, time_limit)) {
1175 time_limit = jiffies + HZ;
1181 * Skip memory holes and free memory that was actually reserved.
1183 static void fadump_release_reserved_area(unsigned long start, unsigned long end)
1185 struct memblock_region *reg;
1186 unsigned long tstart, tend;
1187 unsigned long start_pfn = PHYS_PFN(start);
1188 unsigned long end_pfn = PHYS_PFN(end);
1190 for_each_memblock(memory, reg) {
1191 tstart = max(start_pfn, memblock_region_memory_base_pfn(reg));
1192 tend = min(end_pfn, memblock_region_memory_end_pfn(reg));
1193 if (tstart < tend) {
1194 fadump_free_reserved_memory(tstart, tend);
1196 if (tend == end_pfn)
1199 start_pfn = tend + 1;
1205 * Release the memory that was reserved in early boot to preserve the memory
1206 * contents. The released memory will be available for general use.
1208 static void fadump_release_memory(unsigned long begin, unsigned long end)
1210 unsigned long ra_start, ra_end;
1212 ra_start = fw_dump.reserve_dump_area_start;
1213 ra_end = ra_start + fw_dump.reserve_dump_area_size;
1216 * exclude the dump reserve area. Will reuse it for next
1217 * fadump registration.
1219 if (begin < ra_end && end > ra_start) {
1220 if (begin < ra_start)
1221 fadump_release_reserved_area(begin, ra_start);
1223 fadump_release_reserved_area(ra_end, end);
1225 fadump_release_reserved_area(begin, end);
1228 static void fadump_invalidate_release_mem(void)
1230 unsigned long reserved_area_start, reserved_area_end;
1231 unsigned long destination_address;
1233 mutex_lock(&fadump_mutex);
1234 if (!fw_dump.dump_active) {
1235 mutex_unlock(&fadump_mutex);
1239 destination_address = be64_to_cpu(fdm_active->cpu_state_data.destination_address);
1241 mutex_unlock(&fadump_mutex);
1244 * Save the current reserved memory bounds we will require them
1245 * later for releasing the memory for general use.
1247 reserved_area_start = fw_dump.reserve_dump_area_start;
1248 reserved_area_end = reserved_area_start +
1249 fw_dump.reserve_dump_area_size;
1251 * Setup reserve_dump_area_start and its size so that we can
1252 * reuse this reserved memory for Re-registration.
1254 fw_dump.reserve_dump_area_start = destination_address;
1255 fw_dump.reserve_dump_area_size = get_fadump_area_size();
1257 fadump_release_memory(reserved_area_start, reserved_area_end);
1258 if (fw_dump.cpu_notes_buf) {
1259 fadump_cpu_notes_buf_free(
1260 (unsigned long)__va(fw_dump.cpu_notes_buf),
1261 fw_dump.cpu_notes_buf_size);
1262 fw_dump.cpu_notes_buf = 0;
1263 fw_dump.cpu_notes_buf_size = 0;
1265 /* Initialize the kernel dump memory structure for FAD registration. */
1266 init_fadump_mem_struct(&fdm, fw_dump.reserve_dump_area_start);
1269 static ssize_t fadump_release_memory_store(struct kobject *kobj,
1270 struct kobj_attribute *attr,
1271 const char *buf, size_t count)
1275 if (!fw_dump.dump_active)
1278 if (kstrtoint(buf, 0, &input))
1283 * Take away the '/proc/vmcore'. We are releasing the dump
1284 * memory, hence it will not be valid anymore.
1286 #ifdef CONFIG_PROC_VMCORE
1289 fadump_invalidate_release_mem();
1296 static ssize_t fadump_enabled_show(struct kobject *kobj,
1297 struct kobj_attribute *attr,
1300 return sprintf(buf, "%d\n", fw_dump.fadump_enabled);
1303 static ssize_t fadump_register_show(struct kobject *kobj,
1304 struct kobj_attribute *attr,
1307 return sprintf(buf, "%d\n", fw_dump.dump_registered);
1310 static ssize_t fadump_register_store(struct kobject *kobj,
1311 struct kobj_attribute *attr,
1312 const char *buf, size_t count)
1317 if (!fw_dump.fadump_enabled || fdm_active)
1320 if (kstrtoint(buf, 0, &input))
1323 mutex_lock(&fadump_mutex);
1327 if (fw_dump.dump_registered == 0) {
1330 /* Un-register Firmware-assisted dump */
1331 fadump_unregister_dump(&fdm);
1334 if (fw_dump.dump_registered == 1) {
1338 /* Register Firmware-assisted dump */
1339 ret = register_fadump();
1347 mutex_unlock(&fadump_mutex);
1348 return ret < 0 ? ret : count;
1351 static int fadump_region_show(struct seq_file *m, void *private)
1353 const struct fadump_mem_struct *fdm_ptr;
1355 if (!fw_dump.fadump_enabled)
1358 mutex_lock(&fadump_mutex);
1360 fdm_ptr = fdm_active;
1362 mutex_unlock(&fadump_mutex);
1367 "CPU : [%#016llx-%#016llx] %#llx bytes, "
1369 be64_to_cpu(fdm_ptr->cpu_state_data.destination_address),
1370 be64_to_cpu(fdm_ptr->cpu_state_data.destination_address) +
1371 be64_to_cpu(fdm_ptr->cpu_state_data.source_len) - 1,
1372 be64_to_cpu(fdm_ptr->cpu_state_data.source_len),
1373 be64_to_cpu(fdm_ptr->cpu_state_data.bytes_dumped));
1375 "HPTE: [%#016llx-%#016llx] %#llx bytes, "
1377 be64_to_cpu(fdm_ptr->hpte_region.destination_address),
1378 be64_to_cpu(fdm_ptr->hpte_region.destination_address) +
1379 be64_to_cpu(fdm_ptr->hpte_region.source_len) - 1,
1380 be64_to_cpu(fdm_ptr->hpte_region.source_len),
1381 be64_to_cpu(fdm_ptr->hpte_region.bytes_dumped));
1383 "DUMP: [%#016llx-%#016llx] %#llx bytes, "
1385 be64_to_cpu(fdm_ptr->rmr_region.destination_address),
1386 be64_to_cpu(fdm_ptr->rmr_region.destination_address) +
1387 be64_to_cpu(fdm_ptr->rmr_region.source_len) - 1,
1388 be64_to_cpu(fdm_ptr->rmr_region.source_len),
1389 be64_to_cpu(fdm_ptr->rmr_region.bytes_dumped));
1392 (fw_dump.reserve_dump_area_start ==
1393 be64_to_cpu(fdm_ptr->cpu_state_data.destination_address)))
1396 /* Dump is active. Show reserved memory region. */
1398 " : [%#016llx-%#016llx] %#llx bytes, "
1400 (unsigned long long)fw_dump.reserve_dump_area_start,
1401 be64_to_cpu(fdm_ptr->cpu_state_data.destination_address) - 1,
1402 be64_to_cpu(fdm_ptr->cpu_state_data.destination_address) -
1403 fw_dump.reserve_dump_area_start,
1404 be64_to_cpu(fdm_ptr->cpu_state_data.destination_address) -
1405 fw_dump.reserve_dump_area_start);
1408 mutex_unlock(&fadump_mutex);
1412 static struct kobj_attribute fadump_release_attr = __ATTR(fadump_release_mem,
1414 fadump_release_memory_store);
1415 static struct kobj_attribute fadump_attr = __ATTR(fadump_enabled,
1416 0444, fadump_enabled_show,
1418 static struct kobj_attribute fadump_register_attr = __ATTR(fadump_registered,
1419 0644, fadump_register_show,
1420 fadump_register_store);
1422 static int fadump_region_open(struct inode *inode, struct file *file)
1424 return single_open(file, fadump_region_show, inode->i_private);
1427 static const struct file_operations fadump_region_fops = {
1428 .open = fadump_region_open,
1430 .llseek = seq_lseek,
1431 .release = single_release,
1434 static void fadump_init_files(void)
1436 struct dentry *debugfs_file;
1439 rc = sysfs_create_file(kernel_kobj, &fadump_attr.attr);
1441 printk(KERN_ERR "fadump: unable to create sysfs file"
1442 " fadump_enabled (%d)\n", rc);
1444 rc = sysfs_create_file(kernel_kobj, &fadump_register_attr.attr);
1446 printk(KERN_ERR "fadump: unable to create sysfs file"
1447 " fadump_registered (%d)\n", rc);
1449 debugfs_file = debugfs_create_file("fadump_region", 0444,
1450 powerpc_debugfs_root, NULL,
1451 &fadump_region_fops);
1453 printk(KERN_ERR "fadump: unable to create debugfs file"
1454 " fadump_region\n");
1456 if (fw_dump.dump_active) {
1457 rc = sysfs_create_file(kernel_kobj, &fadump_release_attr.attr);
1459 printk(KERN_ERR "fadump: unable to create sysfs file"
1460 " fadump_release_mem (%d)\n", rc);
1465 static int fadump_panic_event(struct notifier_block *this,
1466 unsigned long event, void *ptr)
1469 * If firmware-assisted dump has been registered then trigger
1470 * firmware-assisted dump and let firmware handle everything
1471 * else. If this returns, then fadump was not registered, so
1472 * go through the rest of the panic path.
1474 crash_fadump(NULL, ptr);
1479 static struct notifier_block fadump_panic_block = {
1480 .notifier_call = fadump_panic_event,
1481 .priority = INT_MIN /* may not return; must be done last */
1485 * Prepare for firmware-assisted dump.
1487 int __init setup_fadump(void)
1489 if (!fw_dump.fadump_enabled)
1492 if (!fw_dump.fadump_supported) {
1493 printk(KERN_ERR "Firmware-assisted dump is not supported on"
1494 " this hardware\n");
1498 fadump_show_config();
1500 * If dump data is available then see if it is valid and prepare for
1501 * saving it to the disk.
1503 if (fw_dump.dump_active) {
1505 * if dump process fails then invalidate the registration
1506 * and release memory before proceeding for re-registration.
1508 if (process_fadump(fdm_active) < 0)
1509 fadump_invalidate_release_mem();
1511 /* Initialize the kernel dump memory structure for FAD registration. */
1512 else if (fw_dump.reserve_dump_area_size)
1513 init_fadump_mem_struct(&fdm, fw_dump.reserve_dump_area_start);
1514 fadump_init_files();
1516 atomic_notifier_chain_register(&panic_notifier_list,
1517 &fadump_panic_block);
1521 subsys_initcall(setup_fadump);