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/debugfs.h>
34 #include <linux/seq_file.h>
35 #include <linux/crash_dump.h>
36 #include <linux/kobject.h>
37 #include <linux/sysfs.h>
42 #include <asm/fadump.h>
43 #include <asm/debug.h>
44 #include <asm/setup.h>
46 static struct fw_dump fw_dump;
47 static struct fadump_mem_struct fdm;
48 static const struct fadump_mem_struct *fdm_active;
50 static DEFINE_MUTEX(fadump_mutex);
51 struct fad_crash_memory_ranges crash_memory_ranges[INIT_CRASHMEM_RANGES];
54 /* Scan the Firmware Assisted dump configuration details. */
55 int __init early_init_dt_scan_fw_dump(unsigned long node,
56 const char *uname, int depth, void *data)
63 if (depth != 1 || strcmp(uname, "rtas") != 0)
67 * Check if Firmware Assisted dump is supported. if yes, check
68 * if dump has been initiated on last reboot.
70 token = of_get_flat_dt_prop(node, "ibm,configure-kernel-dump", NULL);
74 fw_dump.fadump_supported = 1;
75 fw_dump.ibm_configure_kernel_dump = *token;
78 * The 'ibm,kernel-dump' rtas node is present only if there is
79 * dump data waiting for us.
81 fdm_active = of_get_flat_dt_prop(node, "ibm,kernel-dump", NULL);
83 fw_dump.dump_active = 1;
85 /* Get the sizes required to store dump data for the firmware provided
87 * For each dump section type supported, a 32bit cell which defines
88 * the ID of a supported section followed by two 32 bit cells which
89 * gives teh size of the section in bytes.
91 sections = of_get_flat_dt_prop(node, "ibm,configure-kernel-dump-sizes",
97 num_sections = size / (3 * sizeof(u32));
99 for (i = 0; i < num_sections; i++, sections += 3) {
100 u32 type = (u32)of_read_number(sections, 1);
103 case FADUMP_CPU_STATE_DATA:
104 fw_dump.cpu_state_data_size =
105 of_read_ulong(§ions[1], 2);
107 case FADUMP_HPTE_REGION:
108 fw_dump.hpte_region_size =
109 of_read_ulong(§ions[1], 2);
117 int is_fadump_active(void)
119 return fw_dump.dump_active;
122 /* Print firmware assisted dump configurations for debugging purpose. */
123 static void fadump_show_config(void)
125 pr_debug("Support for firmware-assisted dump (fadump): %s\n",
126 (fw_dump.fadump_supported ? "present" : "no support"));
128 if (!fw_dump.fadump_supported)
131 pr_debug("Fadump enabled : %s\n",
132 (fw_dump.fadump_enabled ? "yes" : "no"));
133 pr_debug("Dump Active : %s\n",
134 (fw_dump.dump_active ? "yes" : "no"));
135 pr_debug("Dump section sizes:\n");
136 pr_debug(" CPU state data size: %lx\n", fw_dump.cpu_state_data_size);
137 pr_debug(" HPTE region size : %lx\n", fw_dump.hpte_region_size);
138 pr_debug("Boot memory size : %lx\n", fw_dump.boot_memory_size);
141 static unsigned long init_fadump_mem_struct(struct fadump_mem_struct *fdm,
147 memset(fdm, 0, sizeof(struct fadump_mem_struct));
148 addr = addr & PAGE_MASK;
150 fdm->header.dump_format_version = 0x00000001;
151 fdm->header.dump_num_sections = 3;
152 fdm->header.dump_status_flag = 0;
153 fdm->header.offset_first_dump_section =
154 (u32)offsetof(struct fadump_mem_struct, cpu_state_data);
157 * Fields for disk dump option.
158 * We are not using disk dump option, hence set these fields to 0.
160 fdm->header.dd_block_size = 0;
161 fdm->header.dd_block_offset = 0;
162 fdm->header.dd_num_blocks = 0;
163 fdm->header.dd_offset_disk_path = 0;
165 /* set 0 to disable an automatic dump-reboot. */
166 fdm->header.max_time_auto = 0;
168 /* Kernel dump sections */
169 /* cpu state data section. */
170 fdm->cpu_state_data.request_flag = FADUMP_REQUEST_FLAG;
171 fdm->cpu_state_data.source_data_type = FADUMP_CPU_STATE_DATA;
172 fdm->cpu_state_data.source_address = 0;
173 fdm->cpu_state_data.source_len = fw_dump.cpu_state_data_size;
174 fdm->cpu_state_data.destination_address = addr;
175 addr += fw_dump.cpu_state_data_size;
177 /* hpte region section */
178 fdm->hpte_region.request_flag = FADUMP_REQUEST_FLAG;
179 fdm->hpte_region.source_data_type = FADUMP_HPTE_REGION;
180 fdm->hpte_region.source_address = 0;
181 fdm->hpte_region.source_len = fw_dump.hpte_region_size;
182 fdm->hpte_region.destination_address = addr;
183 addr += fw_dump.hpte_region_size;
185 /* RMA region section */
186 fdm->rmr_region.request_flag = FADUMP_REQUEST_FLAG;
187 fdm->rmr_region.source_data_type = FADUMP_REAL_MODE_REGION;
188 fdm->rmr_region.source_address = RMA_START;
189 fdm->rmr_region.source_len = fw_dump.boot_memory_size;
190 fdm->rmr_region.destination_address = addr;
191 addr += fw_dump.boot_memory_size;
197 * fadump_calculate_reserve_size(): reserve variable boot area 5% of System RAM
199 * Function to find the largest memory size we need to reserve during early
200 * boot process. This will be the size of the memory that is required for a
201 * kernel to boot successfully.
203 * This function has been taken from phyp-assisted dump feature implementation.
205 * returns larger of 256MB or 5% rounded down to multiples of 256MB.
207 * TODO: Come up with better approach to find out more accurate memory size
208 * that is required for a kernel to boot successfully.
211 static inline unsigned long fadump_calculate_reserve_size(void)
216 * Check if the size is specified through fadump_reserve_mem= cmdline
217 * option. If yes, then use that.
219 if (fw_dump.reserve_bootvar)
220 return fw_dump.reserve_bootvar;
222 /* divide by 20 to get 5% of value */
223 size = memblock_end_of_DRAM() / 20;
225 /* round it down in multiples of 256 */
226 size = size & ~0x0FFFFFFFUL;
228 /* Truncate to memory_limit. We don't want to over reserve the memory.*/
229 if (memory_limit && size > memory_limit)
232 return (size > MIN_BOOT_MEM ? size : MIN_BOOT_MEM);
236 * Calculate the total memory size required to be reserved for
237 * firmware-assisted dump registration.
239 static unsigned long get_fadump_area_size(void)
241 unsigned long size = 0;
243 size += fw_dump.cpu_state_data_size;
244 size += fw_dump.hpte_region_size;
245 size += fw_dump.boot_memory_size;
246 size += sizeof(struct fadump_crash_info_header);
247 size += sizeof(struct elfhdr); /* ELF core header.*/
248 size += sizeof(struct elf_phdr); /* place holder for cpu notes */
249 /* Program headers for crash memory regions. */
250 size += sizeof(struct elf_phdr) * (memblock_num_regions(memory) + 2);
252 size = PAGE_ALIGN(size);
256 int __init fadump_reserve_mem(void)
258 unsigned long base, size, memory_boundary;
260 if (!fw_dump.fadump_enabled)
263 if (!fw_dump.fadump_supported) {
264 printk(KERN_INFO "Firmware-assisted dump is not supported on"
266 fw_dump.fadump_enabled = 0;
270 * Initialize boot memory size
271 * If dump is active then we have already calculated the size during
275 fw_dump.boot_memory_size = fdm_active->rmr_region.source_len;
277 fw_dump.boot_memory_size = fadump_calculate_reserve_size();
280 * Calculate the memory boundary.
281 * If memory_limit is less than actual memory boundary then reserve
282 * the memory for fadump beyond the memory_limit and adjust the
283 * memory_limit accordingly, so that the running kernel can run with
284 * specified memory_limit.
286 if (memory_limit && memory_limit < memblock_end_of_DRAM()) {
287 size = get_fadump_area_size();
288 if ((memory_limit + size) < memblock_end_of_DRAM())
289 memory_limit += size;
291 memory_limit = memblock_end_of_DRAM();
292 printk(KERN_INFO "Adjusted memory_limit for firmware-assisted"
293 " dump, now %#016llx\n", memory_limit);
296 memory_boundary = memory_limit;
298 memory_boundary = memblock_end_of_DRAM();
300 if (fw_dump.dump_active) {
301 printk(KERN_INFO "Firmware-assisted dump is active.\n");
303 * If last boot has crashed then reserve all the memory
304 * above boot_memory_size so that we don't touch it until
305 * dump is written to disk by userspace tool. This memory
306 * will be released for general use once the dump is saved.
308 base = fw_dump.boot_memory_size;
309 size = memory_boundary - base;
310 memblock_reserve(base, size);
311 printk(KERN_INFO "Reserved %ldMB of memory at %ldMB "
312 "for saving crash dump\n",
313 (unsigned long)(size >> 20),
314 (unsigned long)(base >> 20));
316 fw_dump.fadumphdr_addr =
317 fdm_active->rmr_region.destination_address +
318 fdm_active->rmr_region.source_len;
319 pr_debug("fadumphdr_addr = %p\n",
320 (void *) fw_dump.fadumphdr_addr);
322 /* Reserve the memory at the top of memory. */
323 size = get_fadump_area_size();
324 base = memory_boundary - size;
325 memblock_reserve(base, size);
326 printk(KERN_INFO "Reserved %ldMB of memory at %ldMB "
327 "for firmware-assisted dump\n",
328 (unsigned long)(size >> 20),
329 (unsigned long)(base >> 20));
331 fw_dump.reserve_dump_area_start = base;
332 fw_dump.reserve_dump_area_size = size;
336 /* Look for fadump= cmdline option. */
337 static int __init early_fadump_param(char *p)
342 if (strncmp(p, "on", 2) == 0)
343 fw_dump.fadump_enabled = 1;
344 else if (strncmp(p, "off", 3) == 0)
345 fw_dump.fadump_enabled = 0;
349 early_param("fadump", early_fadump_param);
351 /* Look for fadump_reserve_mem= cmdline option */
352 static int __init early_fadump_reserve_mem(char *p)
355 fw_dump.reserve_bootvar = memparse(p, &p);
358 early_param("fadump_reserve_mem", early_fadump_reserve_mem);
360 static void register_fw_dump(struct fadump_mem_struct *fdm)
363 unsigned int wait_time;
365 pr_debug("Registering for firmware-assisted kernel dump...\n");
367 /* TODO: Add upper time limit for the delay */
369 rc = rtas_call(fw_dump.ibm_configure_kernel_dump, 3, 1, NULL,
370 FADUMP_REGISTER, fdm,
371 sizeof(struct fadump_mem_struct));
373 wait_time = rtas_busy_delay_time(rc);
381 printk(KERN_ERR "Failed to register firmware-assisted kernel"
382 " dump. Hardware Error(%d).\n", rc);
385 printk(KERN_ERR "Failed to register firmware-assisted kernel"
386 " dump. Parameter Error(%d).\n", rc);
389 printk(KERN_ERR "firmware-assisted kernel dump is already "
391 fw_dump.dump_registered = 1;
394 printk(KERN_INFO "firmware-assisted kernel dump registration"
396 fw_dump.dump_registered = 1;
401 void crash_fadump(struct pt_regs *regs, const char *str)
403 struct fadump_crash_info_header *fdh = NULL;
405 if (!fw_dump.dump_registered || !fw_dump.fadumphdr_addr)
408 fdh = __va(fw_dump.fadumphdr_addr);
409 crashing_cpu = smp_processor_id();
410 fdh->crashing_cpu = crashing_cpu;
411 crash_save_vmcoreinfo();
416 ppc_save_regs(&fdh->regs);
418 fdh->cpu_online_mask = *cpu_online_mask;
420 /* Call ibm,os-term rtas call to trigger firmware assisted dump */
421 rtas_os_term((char *)str);
424 #define GPR_MASK 0xffffff0000000000
425 static inline int fadump_gpr_index(u64 id)
430 if ((id & GPR_MASK) == REG_ID("GPR")) {
431 /* get the digits at the end */
436 str[0] = (id >> 8) & 0xff;
437 sscanf(str, "%d", &i);
444 static inline void fadump_set_regval(struct pt_regs *regs, u64 reg_id,
449 i = fadump_gpr_index(reg_id);
451 regs->gpr[i] = (unsigned long)reg_val;
452 else if (reg_id == REG_ID("NIA"))
453 regs->nip = (unsigned long)reg_val;
454 else if (reg_id == REG_ID("MSR"))
455 regs->msr = (unsigned long)reg_val;
456 else if (reg_id == REG_ID("CTR"))
457 regs->ctr = (unsigned long)reg_val;
458 else if (reg_id == REG_ID("LR"))
459 regs->link = (unsigned long)reg_val;
460 else if (reg_id == REG_ID("XER"))
461 regs->xer = (unsigned long)reg_val;
462 else if (reg_id == REG_ID("CR"))
463 regs->ccr = (unsigned long)reg_val;
464 else if (reg_id == REG_ID("DAR"))
465 regs->dar = (unsigned long)reg_val;
466 else if (reg_id == REG_ID("DSISR"))
467 regs->dsisr = (unsigned long)reg_val;
470 static struct fadump_reg_entry*
471 fadump_read_registers(struct fadump_reg_entry *reg_entry, struct pt_regs *regs)
473 memset(regs, 0, sizeof(struct pt_regs));
475 while (reg_entry->reg_id != REG_ID("CPUEND")) {
476 fadump_set_regval(regs, reg_entry->reg_id,
477 reg_entry->reg_value);
484 static u32 *fadump_append_elf_note(u32 *buf, char *name, unsigned type,
485 void *data, size_t data_len)
487 struct elf_note note;
489 note.n_namesz = strlen(name) + 1;
490 note.n_descsz = data_len;
492 memcpy(buf, ¬e, sizeof(note));
493 buf += (sizeof(note) + 3)/4;
494 memcpy(buf, name, note.n_namesz);
495 buf += (note.n_namesz + 3)/4;
496 memcpy(buf, data, note.n_descsz);
497 buf += (note.n_descsz + 3)/4;
502 static void fadump_final_note(u32 *buf)
504 struct elf_note note;
509 memcpy(buf, ¬e, sizeof(note));
512 static u32 *fadump_regs_to_elf_notes(u32 *buf, struct pt_regs *regs)
514 struct elf_prstatus prstatus;
516 memset(&prstatus, 0, sizeof(prstatus));
518 * FIXME: How do i get PID? Do I really need it?
519 * prstatus.pr_pid = ????
521 elf_core_copy_kernel_regs(&prstatus.pr_reg, regs);
522 buf = fadump_append_elf_note(buf, KEXEC_CORE_NOTE_NAME, NT_PRSTATUS,
523 &prstatus, sizeof(prstatus));
527 static void fadump_update_elfcore_header(char *bufp)
530 struct elf_phdr *phdr;
532 elf = (struct elfhdr *)bufp;
533 bufp += sizeof(struct elfhdr);
535 /* First note is a place holder for cpu notes info. */
536 phdr = (struct elf_phdr *)bufp;
538 if (phdr->p_type == PT_NOTE) {
539 phdr->p_paddr = fw_dump.cpu_notes_buf;
540 phdr->p_offset = phdr->p_paddr;
541 phdr->p_filesz = fw_dump.cpu_notes_buf_size;
542 phdr->p_memsz = fw_dump.cpu_notes_buf_size;
547 static void *fadump_cpu_notes_buf_alloc(unsigned long size)
551 unsigned long order, count, i;
553 order = get_order(size);
554 vaddr = (void *)__get_free_pages(GFP_KERNEL|__GFP_ZERO, order);
559 page = virt_to_page(vaddr);
560 for (i = 0; i < count; i++)
561 SetPageReserved(page + i);
565 static void fadump_cpu_notes_buf_free(unsigned long vaddr, unsigned long size)
568 unsigned long order, count, i;
570 order = get_order(size);
572 page = virt_to_page(vaddr);
573 for (i = 0; i < count; i++)
574 ClearPageReserved(page + i);
575 __free_pages(page, order);
579 * Read CPU state dump data and convert it into ELF notes.
580 * The CPU dump starts with magic number "REGSAVE". NumCpusOffset should be
581 * used to access the data to allow for additional fields to be added without
582 * affecting compatibility. Each list of registers for a CPU starts with
583 * "CPUSTRT" and ends with "CPUEND". Each register entry is of 16 bytes,
584 * 8 Byte ASCII identifier and 8 Byte register value. The register entry
585 * with identifier "CPUSTRT" and "CPUEND" contains 4 byte cpu id as part
586 * of register value. For more details refer to PAPR document.
588 * Only for the crashing cpu we ignore the CPU dump data and get exact
589 * state from fadump crash info structure populated by first kernel at the
592 static int __init fadump_build_cpu_notes(const struct fadump_mem_struct *fdm)
594 struct fadump_reg_save_area_header *reg_header;
595 struct fadump_reg_entry *reg_entry;
596 struct fadump_crash_info_header *fdh = NULL;
599 u32 num_cpus, *note_buf;
601 int i, rc = 0, cpu = 0;
603 if (!fdm->cpu_state_data.bytes_dumped)
606 addr = fdm->cpu_state_data.destination_address;
610 if (reg_header->magic_number != REGSAVE_AREA_MAGIC) {
611 printk(KERN_ERR "Unable to read register save area.\n");
614 pr_debug("--------CPU State Data------------\n");
615 pr_debug("Magic Number: %llx\n", reg_header->magic_number);
616 pr_debug("NumCpuOffset: %x\n", reg_header->num_cpu_offset);
618 vaddr += reg_header->num_cpu_offset;
619 num_cpus = *((u32 *)(vaddr));
620 pr_debug("NumCpus : %u\n", num_cpus);
621 vaddr += sizeof(u32);
622 reg_entry = (struct fadump_reg_entry *)vaddr;
624 /* Allocate buffer to hold cpu crash notes. */
625 fw_dump.cpu_notes_buf_size = num_cpus * sizeof(note_buf_t);
626 fw_dump.cpu_notes_buf_size = PAGE_ALIGN(fw_dump.cpu_notes_buf_size);
627 note_buf = fadump_cpu_notes_buf_alloc(fw_dump.cpu_notes_buf_size);
629 printk(KERN_ERR "Failed to allocate 0x%lx bytes for "
630 "cpu notes buffer\n", fw_dump.cpu_notes_buf_size);
633 fw_dump.cpu_notes_buf = __pa(note_buf);
635 pr_debug("Allocated buffer for cpu notes of size %ld at %p\n",
636 (num_cpus * sizeof(note_buf_t)), note_buf);
638 if (fw_dump.fadumphdr_addr)
639 fdh = __va(fw_dump.fadumphdr_addr);
641 for (i = 0; i < num_cpus; i++) {
642 if (reg_entry->reg_id != REG_ID("CPUSTRT")) {
643 printk(KERN_ERR "Unable to read CPU state data\n");
647 /* Lower 4 bytes of reg_value contains logical cpu id */
648 cpu = reg_entry->reg_value & FADUMP_CPU_ID_MASK;
649 if (!cpumask_test_cpu(cpu, &fdh->cpu_online_mask)) {
650 SKIP_TO_NEXT_CPU(reg_entry);
653 pr_debug("Reading register data for cpu %d...\n", cpu);
654 if (fdh && fdh->crashing_cpu == cpu) {
656 note_buf = fadump_regs_to_elf_notes(note_buf, ®s);
657 SKIP_TO_NEXT_CPU(reg_entry);
660 reg_entry = fadump_read_registers(reg_entry, ®s);
661 note_buf = fadump_regs_to_elf_notes(note_buf, ®s);
664 fadump_final_note(note_buf);
666 pr_debug("Updating elfcore header (%llx) with cpu notes\n",
667 fdh->elfcorehdr_addr);
668 fadump_update_elfcore_header((char *)__va(fdh->elfcorehdr_addr));
672 fadump_cpu_notes_buf_free((unsigned long)__va(fw_dump.cpu_notes_buf),
673 fw_dump.cpu_notes_buf_size);
674 fw_dump.cpu_notes_buf = 0;
675 fw_dump.cpu_notes_buf_size = 0;
681 * Validate and process the dump data stored by firmware before exporting
682 * it through '/proc/vmcore'.
684 static int __init process_fadump(const struct fadump_mem_struct *fdm_active)
686 struct fadump_crash_info_header *fdh;
689 if (!fdm_active || !fw_dump.fadumphdr_addr)
692 /* Check if the dump data is valid. */
693 if ((fdm_active->header.dump_status_flag == FADUMP_ERROR_FLAG) ||
694 (fdm_active->cpu_state_data.error_flags != 0) ||
695 (fdm_active->rmr_region.error_flags != 0)) {
696 printk(KERN_ERR "Dump taken by platform is not valid\n");
699 if ((fdm_active->rmr_region.bytes_dumped !=
700 fdm_active->rmr_region.source_len) ||
701 !fdm_active->cpu_state_data.bytes_dumped) {
702 printk(KERN_ERR "Dump taken by platform is incomplete\n");
706 /* Validate the fadump crash info header */
707 fdh = __va(fw_dump.fadumphdr_addr);
708 if (fdh->magic_number != FADUMP_CRASH_INFO_MAGIC) {
709 printk(KERN_ERR "Crash info header is not valid.\n");
713 rc = fadump_build_cpu_notes(fdm_active);
718 * We are done validating dump info and elfcore header is now ready
719 * to be exported. set elfcorehdr_addr so that vmcore module will
720 * export the elfcore header through '/proc/vmcore'.
722 elfcorehdr_addr = fdh->elfcorehdr_addr;
727 static inline void fadump_add_crash_memory(unsigned long long base,
728 unsigned long long end)
733 pr_debug("crash_memory_range[%d] [%#016llx-%#016llx], %#llx bytes\n",
734 crash_mem_ranges, base, end - 1, (end - base));
735 crash_memory_ranges[crash_mem_ranges].base = base;
736 crash_memory_ranges[crash_mem_ranges].size = end - base;
740 static void fadump_exclude_reserved_area(unsigned long long start,
741 unsigned long long end)
743 unsigned long long ra_start, ra_end;
745 ra_start = fw_dump.reserve_dump_area_start;
746 ra_end = ra_start + fw_dump.reserve_dump_area_size;
748 if ((ra_start < end) && (ra_end > start)) {
749 if ((start < ra_start) && (end > ra_end)) {
750 fadump_add_crash_memory(start, ra_start);
751 fadump_add_crash_memory(ra_end, end);
752 } else if (start < ra_start) {
753 fadump_add_crash_memory(start, ra_start);
754 } else if (ra_end < end) {
755 fadump_add_crash_memory(ra_end, end);
758 fadump_add_crash_memory(start, end);
761 static int fadump_init_elfcore_header(char *bufp)
765 elf = (struct elfhdr *) bufp;
766 bufp += sizeof(struct elfhdr);
767 memcpy(elf->e_ident, ELFMAG, SELFMAG);
768 elf->e_ident[EI_CLASS] = ELF_CLASS;
769 elf->e_ident[EI_DATA] = ELF_DATA;
770 elf->e_ident[EI_VERSION] = EV_CURRENT;
771 elf->e_ident[EI_OSABI] = ELF_OSABI;
772 memset(elf->e_ident+EI_PAD, 0, EI_NIDENT-EI_PAD);
773 elf->e_type = ET_CORE;
774 elf->e_machine = ELF_ARCH;
775 elf->e_version = EV_CURRENT;
777 elf->e_phoff = sizeof(struct elfhdr);
779 elf->e_flags = ELF_CORE_EFLAGS;
780 elf->e_ehsize = sizeof(struct elfhdr);
781 elf->e_phentsize = sizeof(struct elf_phdr);
783 elf->e_shentsize = 0;
791 * Traverse through memblock structure and setup crash memory ranges. These
792 * ranges will be used create PT_LOAD program headers in elfcore header.
794 static void fadump_setup_crash_memory_ranges(void)
796 struct memblock_region *reg;
797 unsigned long long start, end;
799 pr_debug("Setup crash memory ranges.\n");
800 crash_mem_ranges = 0;
802 * add the first memory chunk (RMA_START through boot_memory_size) as
803 * a separate memory chunk. The reason is, at the time crash firmware
804 * will move the content of this memory chunk to different location
805 * specified during fadump registration. We need to create a separate
806 * program header for this chunk with the correct offset.
808 fadump_add_crash_memory(RMA_START, fw_dump.boot_memory_size);
810 for_each_memblock(memory, reg) {
811 start = (unsigned long long)reg->base;
812 end = start + (unsigned long long)reg->size;
813 if (start == RMA_START && end >= fw_dump.boot_memory_size)
814 start = fw_dump.boot_memory_size;
816 /* add this range excluding the reserved dump area. */
817 fadump_exclude_reserved_area(start, end);
822 * If the given physical address falls within the boot memory region then
823 * return the relocated address that points to the dump region reserved
824 * for saving initial boot memory contents.
826 static inline unsigned long fadump_relocate(unsigned long paddr)
828 if (paddr > RMA_START && paddr < fw_dump.boot_memory_size)
829 return fdm.rmr_region.destination_address + paddr;
834 static int fadump_create_elfcore_headers(char *bufp)
837 struct elf_phdr *phdr;
840 fadump_init_elfcore_header(bufp);
841 elf = (struct elfhdr *)bufp;
842 bufp += sizeof(struct elfhdr);
845 * setup ELF PT_NOTE, place holder for cpu notes info. The notes info
846 * will be populated during second kernel boot after crash. Hence
847 * this PT_NOTE will always be the first elf note.
849 * NOTE: Any new ELF note addition should be placed after this note.
851 phdr = (struct elf_phdr *)bufp;
852 bufp += sizeof(struct elf_phdr);
853 phdr->p_type = PT_NOTE;
865 /* setup ELF PT_NOTE for vmcoreinfo */
866 phdr = (struct elf_phdr *)bufp;
867 bufp += sizeof(struct elf_phdr);
868 phdr->p_type = PT_NOTE;
873 phdr->p_paddr = fadump_relocate(paddr_vmcoreinfo_note());
874 phdr->p_offset = phdr->p_paddr;
875 phdr->p_memsz = vmcoreinfo_max_size;
876 phdr->p_filesz = vmcoreinfo_max_size;
878 /* Increment number of program headers. */
881 /* setup PT_LOAD sections. */
883 for (i = 0; i < crash_mem_ranges; i++) {
884 unsigned long long mbase, msize;
885 mbase = crash_memory_ranges[i].base;
886 msize = crash_memory_ranges[i].size;
891 phdr = (struct elf_phdr *)bufp;
892 bufp += sizeof(struct elf_phdr);
893 phdr->p_type = PT_LOAD;
894 phdr->p_flags = PF_R|PF_W|PF_X;
895 phdr->p_offset = mbase;
897 if (mbase == RMA_START) {
899 * The entire RMA region will be moved by firmware
900 * to the specified destination_address. Hence set
901 * the correct offset.
903 phdr->p_offset = fdm.rmr_region.destination_address;
906 phdr->p_paddr = mbase;
907 phdr->p_vaddr = (unsigned long)__va(mbase);
908 phdr->p_filesz = msize;
909 phdr->p_memsz = msize;
912 /* Increment number of program headers. */
918 static unsigned long init_fadump_header(unsigned long addr)
920 struct fadump_crash_info_header *fdh;
925 fw_dump.fadumphdr_addr = addr;
927 addr += sizeof(struct fadump_crash_info_header);
929 memset(fdh, 0, sizeof(struct fadump_crash_info_header));
930 fdh->magic_number = FADUMP_CRASH_INFO_MAGIC;
931 fdh->elfcorehdr_addr = addr;
932 /* We will set the crashing cpu id in crash_fadump() during crash. */
933 fdh->crashing_cpu = CPU_UNKNOWN;
938 static void register_fadump(void)
944 * If no memory is reserved then we can not register for firmware-
947 if (!fw_dump.reserve_dump_area_size)
950 fadump_setup_crash_memory_ranges();
952 addr = fdm.rmr_region.destination_address + fdm.rmr_region.source_len;
953 /* Initialize fadump crash info header. */
954 addr = init_fadump_header(addr);
957 pr_debug("Creating ELF core headers at %#016lx\n", addr);
958 fadump_create_elfcore_headers(vaddr);
960 /* register the future kernel dump with firmware. */
961 register_fw_dump(&fdm);
964 static int fadump_unregister_dump(struct fadump_mem_struct *fdm)
967 unsigned int wait_time;
969 pr_debug("Un-register firmware-assisted dump\n");
971 /* TODO: Add upper time limit for the delay */
973 rc = rtas_call(fw_dump.ibm_configure_kernel_dump, 3, 1, NULL,
974 FADUMP_UNREGISTER, fdm,
975 sizeof(struct fadump_mem_struct));
977 wait_time = rtas_busy_delay_time(rc);
983 printk(KERN_ERR "Failed to un-register firmware-assisted dump."
984 " unexpected error(%d).\n", rc);
987 fw_dump.dump_registered = 0;
991 static int fadump_invalidate_dump(struct fadump_mem_struct *fdm)
994 unsigned int wait_time;
996 pr_debug("Invalidating firmware-assisted dump registration\n");
998 /* TODO: Add upper time limit for the delay */
1000 rc = rtas_call(fw_dump.ibm_configure_kernel_dump, 3, 1, NULL,
1001 FADUMP_INVALIDATE, fdm,
1002 sizeof(struct fadump_mem_struct));
1004 wait_time = rtas_busy_delay_time(rc);
1007 } while (wait_time);
1010 printk(KERN_ERR "Failed to invalidate firmware-assisted dump "
1011 "rgistration. unexpected error(%d).\n", rc);
1014 fw_dump.dump_active = 0;
1019 void fadump_cleanup(void)
1021 /* Invalidate the registration only if dump is active. */
1022 if (fw_dump.dump_active) {
1023 init_fadump_mem_struct(&fdm,
1024 fdm_active->cpu_state_data.destination_address);
1025 fadump_invalidate_dump(&fdm);
1030 * Release the memory that was reserved in early boot to preserve the memory
1031 * contents. The released memory will be available for general use.
1033 static void fadump_release_memory(unsigned long begin, unsigned long end)
1036 unsigned long ra_start, ra_end;
1038 ra_start = fw_dump.reserve_dump_area_start;
1039 ra_end = ra_start + fw_dump.reserve_dump_area_size;
1041 for (addr = begin; addr < end; addr += PAGE_SIZE) {
1043 * exclude the dump reserve area. Will reuse it for next
1044 * fadump registration.
1046 if (addr <= ra_end && ((addr + PAGE_SIZE) > ra_start))
1049 free_reserved_page(pfn_to_page(addr >> PAGE_SHIFT));
1053 static void fadump_invalidate_release_mem(void)
1055 unsigned long reserved_area_start, reserved_area_end;
1056 unsigned long destination_address;
1058 mutex_lock(&fadump_mutex);
1059 if (!fw_dump.dump_active) {
1060 mutex_unlock(&fadump_mutex);
1064 destination_address = fdm_active->cpu_state_data.destination_address;
1066 mutex_unlock(&fadump_mutex);
1069 * Save the current reserved memory bounds we will require them
1070 * later for releasing the memory for general use.
1072 reserved_area_start = fw_dump.reserve_dump_area_start;
1073 reserved_area_end = reserved_area_start +
1074 fw_dump.reserve_dump_area_size;
1076 * Setup reserve_dump_area_start and its size so that we can
1077 * reuse this reserved memory for Re-registration.
1079 fw_dump.reserve_dump_area_start = destination_address;
1080 fw_dump.reserve_dump_area_size = get_fadump_area_size();
1082 fadump_release_memory(reserved_area_start, reserved_area_end);
1083 if (fw_dump.cpu_notes_buf) {
1084 fadump_cpu_notes_buf_free(
1085 (unsigned long)__va(fw_dump.cpu_notes_buf),
1086 fw_dump.cpu_notes_buf_size);
1087 fw_dump.cpu_notes_buf = 0;
1088 fw_dump.cpu_notes_buf_size = 0;
1090 /* Initialize the kernel dump memory structure for FAD registration. */
1091 init_fadump_mem_struct(&fdm, fw_dump.reserve_dump_area_start);
1094 static ssize_t fadump_release_memory_store(struct kobject *kobj,
1095 struct kobj_attribute *attr,
1096 const char *buf, size_t count)
1098 if (!fw_dump.dump_active)
1101 if (buf[0] == '1') {
1103 * Take away the '/proc/vmcore'. We are releasing the dump
1104 * memory, hence it will not be valid anymore.
1107 fadump_invalidate_release_mem();
1114 static ssize_t fadump_enabled_show(struct kobject *kobj,
1115 struct kobj_attribute *attr,
1118 return sprintf(buf, "%d\n", fw_dump.fadump_enabled);
1121 static ssize_t fadump_register_show(struct kobject *kobj,
1122 struct kobj_attribute *attr,
1125 return sprintf(buf, "%d\n", fw_dump.dump_registered);
1128 static ssize_t fadump_register_store(struct kobject *kobj,
1129 struct kobj_attribute *attr,
1130 const char *buf, size_t count)
1134 if (!fw_dump.fadump_enabled || fdm_active)
1137 mutex_lock(&fadump_mutex);
1141 if (fw_dump.dump_registered == 0) {
1145 /* Un-register Firmware-assisted dump */
1146 fadump_unregister_dump(&fdm);
1149 if (fw_dump.dump_registered == 1) {
1153 /* Register Firmware-assisted dump */
1162 mutex_unlock(&fadump_mutex);
1163 return ret < 0 ? ret : count;
1166 static int fadump_region_show(struct seq_file *m, void *private)
1168 const struct fadump_mem_struct *fdm_ptr;
1170 if (!fw_dump.fadump_enabled)
1173 mutex_lock(&fadump_mutex);
1175 fdm_ptr = fdm_active;
1177 mutex_unlock(&fadump_mutex);
1182 "CPU : [%#016llx-%#016llx] %#llx bytes, "
1184 fdm_ptr->cpu_state_data.destination_address,
1185 fdm_ptr->cpu_state_data.destination_address +
1186 fdm_ptr->cpu_state_data.source_len - 1,
1187 fdm_ptr->cpu_state_data.source_len,
1188 fdm_ptr->cpu_state_data.bytes_dumped);
1190 "HPTE: [%#016llx-%#016llx] %#llx bytes, "
1192 fdm_ptr->hpte_region.destination_address,
1193 fdm_ptr->hpte_region.destination_address +
1194 fdm_ptr->hpte_region.source_len - 1,
1195 fdm_ptr->hpte_region.source_len,
1196 fdm_ptr->hpte_region.bytes_dumped);
1198 "DUMP: [%#016llx-%#016llx] %#llx bytes, "
1200 fdm_ptr->rmr_region.destination_address,
1201 fdm_ptr->rmr_region.destination_address +
1202 fdm_ptr->rmr_region.source_len - 1,
1203 fdm_ptr->rmr_region.source_len,
1204 fdm_ptr->rmr_region.bytes_dumped);
1207 (fw_dump.reserve_dump_area_start ==
1208 fdm_ptr->cpu_state_data.destination_address))
1211 /* Dump is active. Show reserved memory region. */
1213 " : [%#016llx-%#016llx] %#llx bytes, "
1215 (unsigned long long)fw_dump.reserve_dump_area_start,
1216 fdm_ptr->cpu_state_data.destination_address - 1,
1217 fdm_ptr->cpu_state_data.destination_address -
1218 fw_dump.reserve_dump_area_start,
1219 fdm_ptr->cpu_state_data.destination_address -
1220 fw_dump.reserve_dump_area_start);
1223 mutex_unlock(&fadump_mutex);
1227 static struct kobj_attribute fadump_release_attr = __ATTR(fadump_release_mem,
1229 fadump_release_memory_store);
1230 static struct kobj_attribute fadump_attr = __ATTR(fadump_enabled,
1231 0444, fadump_enabled_show,
1233 static struct kobj_attribute fadump_register_attr = __ATTR(fadump_registered,
1234 0644, fadump_register_show,
1235 fadump_register_store);
1237 static int fadump_region_open(struct inode *inode, struct file *file)
1239 return single_open(file, fadump_region_show, inode->i_private);
1242 static const struct file_operations fadump_region_fops = {
1243 .open = fadump_region_open,
1245 .llseek = seq_lseek,
1246 .release = single_release,
1249 static void fadump_init_files(void)
1251 struct dentry *debugfs_file;
1254 rc = sysfs_create_file(kernel_kobj, &fadump_attr.attr);
1256 printk(KERN_ERR "fadump: unable to create sysfs file"
1257 " fadump_enabled (%d)\n", rc);
1259 rc = sysfs_create_file(kernel_kobj, &fadump_register_attr.attr);
1261 printk(KERN_ERR "fadump: unable to create sysfs file"
1262 " fadump_registered (%d)\n", rc);
1264 debugfs_file = debugfs_create_file("fadump_region", 0444,
1265 powerpc_debugfs_root, NULL,
1266 &fadump_region_fops);
1268 printk(KERN_ERR "fadump: unable to create debugfs file"
1269 " fadump_region\n");
1271 if (fw_dump.dump_active) {
1272 rc = sysfs_create_file(kernel_kobj, &fadump_release_attr.attr);
1274 printk(KERN_ERR "fadump: unable to create sysfs file"
1275 " fadump_release_mem (%d)\n", rc);
1281 * Prepare for firmware-assisted dump.
1283 int __init setup_fadump(void)
1285 if (!fw_dump.fadump_enabled)
1288 if (!fw_dump.fadump_supported) {
1289 printk(KERN_ERR "Firmware-assisted dump is not supported on"
1290 " this hardware\n");
1294 fadump_show_config();
1296 * If dump data is available then see if it is valid and prepare for
1297 * saving it to the disk.
1299 if (fw_dump.dump_active) {
1301 * if dump process fails then invalidate the registration
1302 * and release memory before proceeding for re-registration.
1304 if (process_fadump(fdm_active) < 0)
1305 fadump_invalidate_release_mem();
1307 /* Initialize the kernel dump memory structure for FAD registration. */
1308 else if (fw_dump.reserve_dump_area_size)
1309 init_fadump_mem_struct(&fdm, fw_dump.reserve_dump_area_start);
1310 fadump_init_files();
1314 subsys_initcall(setup_fadump);