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>
44 static struct fw_dump fw_dump;
45 static struct fadump_mem_struct fdm;
46 static const struct fadump_mem_struct *fdm_active;
48 static DEFINE_MUTEX(fadump_mutex);
49 struct fad_crash_memory_ranges crash_memory_ranges[INIT_CRASHMEM_RANGES];
52 /* Scan the Firmware Assisted dump configuration details. */
53 int __init early_init_dt_scan_fw_dump(unsigned long node,
54 const char *uname, int depth, void *data)
61 if (depth != 1 || strcmp(uname, "rtas") != 0)
65 * Check if Firmware Assisted dump is supported. if yes, check
66 * if dump has been initiated on last reboot.
68 token = of_get_flat_dt_prop(node, "ibm,configure-kernel-dump", NULL);
72 fw_dump.fadump_supported = 1;
73 fw_dump.ibm_configure_kernel_dump = *token;
76 * The 'ibm,kernel-dump' rtas node is present only if there is
77 * dump data waiting for us.
79 fdm_active = of_get_flat_dt_prop(node, "ibm,kernel-dump", NULL);
81 fw_dump.dump_active = 1;
83 /* Get the sizes required to store dump data for the firmware provided
85 * For each dump section type supported, a 32bit cell which defines
86 * the ID of a supported section followed by two 32 bit cells which
87 * gives teh size of the section in bytes.
89 sections = of_get_flat_dt_prop(node, "ibm,configure-kernel-dump-sizes",
95 num_sections = size / (3 * sizeof(u32));
97 for (i = 0; i < num_sections; i++, sections += 3) {
98 u32 type = (u32)of_read_number(sections, 1);
101 case FADUMP_CPU_STATE_DATA:
102 fw_dump.cpu_state_data_size =
103 of_read_ulong(§ions[1], 2);
105 case FADUMP_HPTE_REGION:
106 fw_dump.hpte_region_size =
107 of_read_ulong(§ions[1], 2);
114 int is_fadump_active(void)
116 return fw_dump.dump_active;
119 /* Print firmware assisted dump configurations for debugging purpose. */
120 static void fadump_show_config(void)
122 pr_debug("Support for firmware-assisted dump (fadump): %s\n",
123 (fw_dump.fadump_supported ? "present" : "no support"));
125 if (!fw_dump.fadump_supported)
128 pr_debug("Fadump enabled : %s\n",
129 (fw_dump.fadump_enabled ? "yes" : "no"));
130 pr_debug("Dump Active : %s\n",
131 (fw_dump.dump_active ? "yes" : "no"));
132 pr_debug("Dump section sizes:\n");
133 pr_debug(" CPU state data size: %lx\n", fw_dump.cpu_state_data_size);
134 pr_debug(" HPTE region size : %lx\n", fw_dump.hpte_region_size);
135 pr_debug("Boot memory size : %lx\n", fw_dump.boot_memory_size);
138 static unsigned long init_fadump_mem_struct(struct fadump_mem_struct *fdm,
144 memset(fdm, 0, sizeof(struct fadump_mem_struct));
145 addr = addr & PAGE_MASK;
147 fdm->header.dump_format_version = 0x00000001;
148 fdm->header.dump_num_sections = 3;
149 fdm->header.dump_status_flag = 0;
150 fdm->header.offset_first_dump_section =
151 (u32)offsetof(struct fadump_mem_struct, cpu_state_data);
154 * Fields for disk dump option.
155 * We are not using disk dump option, hence set these fields to 0.
157 fdm->header.dd_block_size = 0;
158 fdm->header.dd_block_offset = 0;
159 fdm->header.dd_num_blocks = 0;
160 fdm->header.dd_offset_disk_path = 0;
162 /* set 0 to disable an automatic dump-reboot. */
163 fdm->header.max_time_auto = 0;
165 /* Kernel dump sections */
166 /* cpu state data section. */
167 fdm->cpu_state_data.request_flag = FADUMP_REQUEST_FLAG;
168 fdm->cpu_state_data.source_data_type = FADUMP_CPU_STATE_DATA;
169 fdm->cpu_state_data.source_address = 0;
170 fdm->cpu_state_data.source_len = fw_dump.cpu_state_data_size;
171 fdm->cpu_state_data.destination_address = addr;
172 addr += fw_dump.cpu_state_data_size;
174 /* hpte region section */
175 fdm->hpte_region.request_flag = FADUMP_REQUEST_FLAG;
176 fdm->hpte_region.source_data_type = FADUMP_HPTE_REGION;
177 fdm->hpte_region.source_address = 0;
178 fdm->hpte_region.source_len = fw_dump.hpte_region_size;
179 fdm->hpte_region.destination_address = addr;
180 addr += fw_dump.hpte_region_size;
182 /* RMA region section */
183 fdm->rmr_region.request_flag = FADUMP_REQUEST_FLAG;
184 fdm->rmr_region.source_data_type = FADUMP_REAL_MODE_REGION;
185 fdm->rmr_region.source_address = RMA_START;
186 fdm->rmr_region.source_len = fw_dump.boot_memory_size;
187 fdm->rmr_region.destination_address = addr;
188 addr += fw_dump.boot_memory_size;
194 * fadump_calculate_reserve_size(): reserve variable boot area 5% of System RAM
196 * Function to find the largest memory size we need to reserve during early
197 * boot process. This will be the size of the memory that is required for a
198 * kernel to boot successfully.
200 * This function has been taken from phyp-assisted dump feature implementation.
202 * returns larger of 256MB or 5% rounded down to multiples of 256MB.
204 * TODO: Come up with better approach to find out more accurate memory size
205 * that is required for a kernel to boot successfully.
208 static inline unsigned long fadump_calculate_reserve_size(void)
213 * Check if the size is specified through fadump_reserve_mem= cmdline
214 * option. If yes, then use that.
216 if (fw_dump.reserve_bootvar)
217 return fw_dump.reserve_bootvar;
219 /* divide by 20 to get 5% of value */
220 size = memblock_end_of_DRAM() / 20;
222 /* round it down in multiples of 256 */
223 size = size & ~0x0FFFFFFFUL;
225 /* Truncate to memory_limit. We don't want to over reserve the memory.*/
226 if (memory_limit && size > memory_limit)
229 return (size > MIN_BOOT_MEM ? size : MIN_BOOT_MEM);
233 * Calculate the total memory size required to be reserved for
234 * firmware-assisted dump registration.
236 static unsigned long get_fadump_area_size(void)
238 unsigned long size = 0;
240 size += fw_dump.cpu_state_data_size;
241 size += fw_dump.hpte_region_size;
242 size += fw_dump.boot_memory_size;
243 size += sizeof(struct fadump_crash_info_header);
244 size += sizeof(struct elfhdr); /* ELF core header.*/
245 size += sizeof(struct elf_phdr); /* place holder for cpu notes */
246 /* Program headers for crash memory regions. */
247 size += sizeof(struct elf_phdr) * (memblock_num_regions(memory) + 2);
249 size = PAGE_ALIGN(size);
253 int __init fadump_reserve_mem(void)
255 unsigned long base, size, memory_boundary;
257 if (!fw_dump.fadump_enabled)
260 if (!fw_dump.fadump_supported) {
261 printk(KERN_INFO "Firmware-assisted dump is not supported on"
263 fw_dump.fadump_enabled = 0;
267 * Initialize boot memory size
268 * If dump is active then we have already calculated the size during
272 fw_dump.boot_memory_size = fdm_active->rmr_region.source_len;
274 fw_dump.boot_memory_size = fadump_calculate_reserve_size();
277 * Calculate the memory boundary.
278 * If memory_limit is less than actual memory boundary then reserve
279 * the memory for fadump beyond the memory_limit and adjust the
280 * memory_limit accordingly, so that the running kernel can run with
281 * specified memory_limit.
283 if (memory_limit && memory_limit < memblock_end_of_DRAM()) {
284 size = get_fadump_area_size();
285 if ((memory_limit + size) < memblock_end_of_DRAM())
286 memory_limit += size;
288 memory_limit = memblock_end_of_DRAM();
289 printk(KERN_INFO "Adjusted memory_limit for firmware-assisted"
290 " dump, now %#016llx\n",
291 (unsigned long long)memory_limit);
294 memory_boundary = memory_limit;
296 memory_boundary = memblock_end_of_DRAM();
298 if (fw_dump.dump_active) {
299 printk(KERN_INFO "Firmware-assisted dump is active.\n");
301 * If last boot has crashed then reserve all the memory
302 * above boot_memory_size so that we don't touch it until
303 * dump is written to disk by userspace tool. This memory
304 * will be released for general use once the dump is saved.
306 base = fw_dump.boot_memory_size;
307 size = memory_boundary - base;
308 memblock_reserve(base, size);
309 printk(KERN_INFO "Reserved %ldMB of memory at %ldMB "
310 "for saving crash dump\n",
311 (unsigned long)(size >> 20),
312 (unsigned long)(base >> 20));
314 fw_dump.fadumphdr_addr =
315 fdm_active->rmr_region.destination_address +
316 fdm_active->rmr_region.source_len;
317 pr_debug("fadumphdr_addr = %p\n",
318 (void *) fw_dump.fadumphdr_addr);
320 /* Reserve the memory at the top of memory. */
321 size = get_fadump_area_size();
322 base = memory_boundary - size;
323 memblock_reserve(base, size);
324 printk(KERN_INFO "Reserved %ldMB of memory at %ldMB "
325 "for firmware-assisted dump\n",
326 (unsigned long)(size >> 20),
327 (unsigned long)(base >> 20));
329 fw_dump.reserve_dump_area_start = base;
330 fw_dump.reserve_dump_area_size = size;
334 /* Look for fadump= cmdline option. */
335 static int __init early_fadump_param(char *p)
340 if (strncmp(p, "on", 2) == 0)
341 fw_dump.fadump_enabled = 1;
342 else if (strncmp(p, "off", 3) == 0)
343 fw_dump.fadump_enabled = 0;
347 early_param("fadump", early_fadump_param);
349 /* Look for fadump_reserve_mem= cmdline option */
350 static int __init early_fadump_reserve_mem(char *p)
353 fw_dump.reserve_bootvar = memparse(p, &p);
356 early_param("fadump_reserve_mem", early_fadump_reserve_mem);
358 static void register_fw_dump(struct fadump_mem_struct *fdm)
361 unsigned int wait_time;
363 pr_debug("Registering for firmware-assisted kernel dump...\n");
365 /* TODO: Add upper time limit for the delay */
367 rc = rtas_call(fw_dump.ibm_configure_kernel_dump, 3, 1, NULL,
368 FADUMP_REGISTER, fdm,
369 sizeof(struct fadump_mem_struct));
371 wait_time = rtas_busy_delay_time(rc);
379 printk(KERN_ERR "Failed to register firmware-assisted kernel"
380 " dump. Hardware Error(%d).\n", rc);
383 printk(KERN_ERR "Failed to register firmware-assisted kernel"
384 " dump. Parameter Error(%d).\n", rc);
387 printk(KERN_ERR "firmware-assisted kernel dump is already "
389 fw_dump.dump_registered = 1;
392 printk(KERN_INFO "firmware-assisted kernel dump registration"
394 fw_dump.dump_registered = 1;
399 void crash_fadump(struct pt_regs *regs, const char *str)
401 struct fadump_crash_info_header *fdh = NULL;
403 if (!fw_dump.dump_registered || !fw_dump.fadumphdr_addr)
406 fdh = __va(fw_dump.fadumphdr_addr);
407 crashing_cpu = smp_processor_id();
408 fdh->crashing_cpu = crashing_cpu;
409 crash_save_vmcoreinfo();
414 ppc_save_regs(&fdh->regs);
416 fdh->cpu_online_mask = *cpu_online_mask;
418 /* Call ibm,os-term rtas call to trigger firmware assisted dump */
419 rtas_os_term((char *)str);
422 #define GPR_MASK 0xffffff0000000000
423 static inline int fadump_gpr_index(u64 id)
428 if ((id & GPR_MASK) == REG_ID("GPR")) {
429 /* get the digits at the end */
434 str[0] = (id >> 8) & 0xff;
435 sscanf(str, "%d", &i);
442 static inline void fadump_set_regval(struct pt_regs *regs, u64 reg_id,
447 i = fadump_gpr_index(reg_id);
449 regs->gpr[i] = (unsigned long)reg_val;
450 else if (reg_id == REG_ID("NIA"))
451 regs->nip = (unsigned long)reg_val;
452 else if (reg_id == REG_ID("MSR"))
453 regs->msr = (unsigned long)reg_val;
454 else if (reg_id == REG_ID("CTR"))
455 regs->ctr = (unsigned long)reg_val;
456 else if (reg_id == REG_ID("LR"))
457 regs->link = (unsigned long)reg_val;
458 else if (reg_id == REG_ID("XER"))
459 regs->xer = (unsigned long)reg_val;
460 else if (reg_id == REG_ID("CR"))
461 regs->ccr = (unsigned long)reg_val;
462 else if (reg_id == REG_ID("DAR"))
463 regs->dar = (unsigned long)reg_val;
464 else if (reg_id == REG_ID("DSISR"))
465 regs->dsisr = (unsigned long)reg_val;
468 static struct fadump_reg_entry*
469 fadump_read_registers(struct fadump_reg_entry *reg_entry, struct pt_regs *regs)
471 memset(regs, 0, sizeof(struct pt_regs));
473 while (reg_entry->reg_id != REG_ID("CPUEND")) {
474 fadump_set_regval(regs, reg_entry->reg_id,
475 reg_entry->reg_value);
482 static u32 *fadump_append_elf_note(u32 *buf, char *name, unsigned type,
483 void *data, size_t data_len)
485 struct elf_note note;
487 note.n_namesz = strlen(name) + 1;
488 note.n_descsz = data_len;
490 memcpy(buf, ¬e, sizeof(note));
491 buf += (sizeof(note) + 3)/4;
492 memcpy(buf, name, note.n_namesz);
493 buf += (note.n_namesz + 3)/4;
494 memcpy(buf, data, note.n_descsz);
495 buf += (note.n_descsz + 3)/4;
500 static void fadump_final_note(u32 *buf)
502 struct elf_note note;
507 memcpy(buf, ¬e, sizeof(note));
510 static u32 *fadump_regs_to_elf_notes(u32 *buf, struct pt_regs *regs)
512 struct elf_prstatus prstatus;
514 memset(&prstatus, 0, sizeof(prstatus));
516 * FIXME: How do i get PID? Do I really need it?
517 * prstatus.pr_pid = ????
519 elf_core_copy_kernel_regs(&prstatus.pr_reg, regs);
520 buf = fadump_append_elf_note(buf, KEXEC_CORE_NOTE_NAME, NT_PRSTATUS,
521 &prstatus, sizeof(prstatus));
525 static void fadump_update_elfcore_header(char *bufp)
528 struct elf_phdr *phdr;
530 elf = (struct elfhdr *)bufp;
531 bufp += sizeof(struct elfhdr);
533 /* First note is a place holder for cpu notes info. */
534 phdr = (struct elf_phdr *)bufp;
536 if (phdr->p_type == PT_NOTE) {
537 phdr->p_paddr = fw_dump.cpu_notes_buf;
538 phdr->p_offset = phdr->p_paddr;
539 phdr->p_filesz = fw_dump.cpu_notes_buf_size;
540 phdr->p_memsz = fw_dump.cpu_notes_buf_size;
545 static void *fadump_cpu_notes_buf_alloc(unsigned long size)
549 unsigned long order, count, i;
551 order = get_order(size);
552 vaddr = (void *)__get_free_pages(GFP_KERNEL|__GFP_ZERO, order);
557 page = virt_to_page(vaddr);
558 for (i = 0; i < count; i++)
559 SetPageReserved(page + i);
563 static void fadump_cpu_notes_buf_free(unsigned long vaddr, unsigned long size)
566 unsigned long order, count, i;
568 order = get_order(size);
570 page = virt_to_page(vaddr);
571 for (i = 0; i < count; i++)
572 ClearPageReserved(page + i);
573 __free_pages(page, order);
577 * Read CPU state dump data and convert it into ELF notes.
578 * The CPU dump starts with magic number "REGSAVE". NumCpusOffset should be
579 * used to access the data to allow for additional fields to be added without
580 * affecting compatibility. Each list of registers for a CPU starts with
581 * "CPUSTRT" and ends with "CPUEND". Each register entry is of 16 bytes,
582 * 8 Byte ASCII identifier and 8 Byte register value. The register entry
583 * with identifier "CPUSTRT" and "CPUEND" contains 4 byte cpu id as part
584 * of register value. For more details refer to PAPR document.
586 * Only for the crashing cpu we ignore the CPU dump data and get exact
587 * state from fadump crash info structure populated by first kernel at the
590 static int __init fadump_build_cpu_notes(const struct fadump_mem_struct *fdm)
592 struct fadump_reg_save_area_header *reg_header;
593 struct fadump_reg_entry *reg_entry;
594 struct fadump_crash_info_header *fdh = NULL;
597 u32 num_cpus, *note_buf;
599 int i, rc = 0, cpu = 0;
601 if (!fdm->cpu_state_data.bytes_dumped)
604 addr = fdm->cpu_state_data.destination_address;
608 if (reg_header->magic_number != REGSAVE_AREA_MAGIC) {
609 printk(KERN_ERR "Unable to read register save area.\n");
612 pr_debug("--------CPU State Data------------\n");
613 pr_debug("Magic Number: %llx\n", reg_header->magic_number);
614 pr_debug("NumCpuOffset: %x\n", reg_header->num_cpu_offset);
616 vaddr += reg_header->num_cpu_offset;
617 num_cpus = *((u32 *)(vaddr));
618 pr_debug("NumCpus : %u\n", num_cpus);
619 vaddr += sizeof(u32);
620 reg_entry = (struct fadump_reg_entry *)vaddr;
622 /* Allocate buffer to hold cpu crash notes. */
623 fw_dump.cpu_notes_buf_size = num_cpus * sizeof(note_buf_t);
624 fw_dump.cpu_notes_buf_size = PAGE_ALIGN(fw_dump.cpu_notes_buf_size);
625 note_buf = fadump_cpu_notes_buf_alloc(fw_dump.cpu_notes_buf_size);
627 printk(KERN_ERR "Failed to allocate 0x%lx bytes for "
628 "cpu notes buffer\n", fw_dump.cpu_notes_buf_size);
631 fw_dump.cpu_notes_buf = __pa(note_buf);
633 pr_debug("Allocated buffer for cpu notes of size %ld at %p\n",
634 (num_cpus * sizeof(note_buf_t)), note_buf);
636 if (fw_dump.fadumphdr_addr)
637 fdh = __va(fw_dump.fadumphdr_addr);
639 for (i = 0; i < num_cpus; i++) {
640 if (reg_entry->reg_id != REG_ID("CPUSTRT")) {
641 printk(KERN_ERR "Unable to read CPU state data\n");
645 /* Lower 4 bytes of reg_value contains logical cpu id */
646 cpu = reg_entry->reg_value & FADUMP_CPU_ID_MASK;
647 if (!cpumask_test_cpu(cpu, &fdh->cpu_online_mask)) {
648 SKIP_TO_NEXT_CPU(reg_entry);
651 pr_debug("Reading register data for cpu %d...\n", cpu);
652 if (fdh && fdh->crashing_cpu == cpu) {
654 note_buf = fadump_regs_to_elf_notes(note_buf, ®s);
655 SKIP_TO_NEXT_CPU(reg_entry);
658 reg_entry = fadump_read_registers(reg_entry, ®s);
659 note_buf = fadump_regs_to_elf_notes(note_buf, ®s);
662 fadump_final_note(note_buf);
664 pr_debug("Updating elfcore header (%llx) with cpu notes\n",
665 fdh->elfcorehdr_addr);
666 fadump_update_elfcore_header((char *)__va(fdh->elfcorehdr_addr));
670 fadump_cpu_notes_buf_free((unsigned long)__va(fw_dump.cpu_notes_buf),
671 fw_dump.cpu_notes_buf_size);
672 fw_dump.cpu_notes_buf = 0;
673 fw_dump.cpu_notes_buf_size = 0;
679 * Validate and process the dump data stored by firmware before exporting
680 * it through '/proc/vmcore'.
682 static int __init process_fadump(const struct fadump_mem_struct *fdm_active)
684 struct fadump_crash_info_header *fdh;
687 if (!fdm_active || !fw_dump.fadumphdr_addr)
690 /* Check if the dump data is valid. */
691 if ((fdm_active->header.dump_status_flag == FADUMP_ERROR_FLAG) ||
692 (fdm_active->cpu_state_data.error_flags != 0) ||
693 (fdm_active->rmr_region.error_flags != 0)) {
694 printk(KERN_ERR "Dump taken by platform is not valid\n");
697 if ((fdm_active->rmr_region.bytes_dumped !=
698 fdm_active->rmr_region.source_len) ||
699 !fdm_active->cpu_state_data.bytes_dumped) {
700 printk(KERN_ERR "Dump taken by platform is incomplete\n");
704 /* Validate the fadump crash info header */
705 fdh = __va(fw_dump.fadumphdr_addr);
706 if (fdh->magic_number != FADUMP_CRASH_INFO_MAGIC) {
707 printk(KERN_ERR "Crash info header is not valid.\n");
711 rc = fadump_build_cpu_notes(fdm_active);
716 * We are done validating dump info and elfcore header is now ready
717 * to be exported. set elfcorehdr_addr so that vmcore module will
718 * export the elfcore header through '/proc/vmcore'.
720 elfcorehdr_addr = fdh->elfcorehdr_addr;
725 static inline void fadump_add_crash_memory(unsigned long long base,
726 unsigned long long end)
731 pr_debug("crash_memory_range[%d] [%#016llx-%#016llx], %#llx bytes\n",
732 crash_mem_ranges, base, end - 1, (end - base));
733 crash_memory_ranges[crash_mem_ranges].base = base;
734 crash_memory_ranges[crash_mem_ranges].size = end - base;
738 static void fadump_exclude_reserved_area(unsigned long long start,
739 unsigned long long end)
741 unsigned long long ra_start, ra_end;
743 ra_start = fw_dump.reserve_dump_area_start;
744 ra_end = ra_start + fw_dump.reserve_dump_area_size;
746 if ((ra_start < end) && (ra_end > start)) {
747 if ((start < ra_start) && (end > ra_end)) {
748 fadump_add_crash_memory(start, ra_start);
749 fadump_add_crash_memory(ra_end, end);
750 } else if (start < ra_start) {
751 fadump_add_crash_memory(start, ra_start);
752 } else if (ra_end < end) {
753 fadump_add_crash_memory(ra_end, end);
756 fadump_add_crash_memory(start, end);
759 static int fadump_init_elfcore_header(char *bufp)
763 elf = (struct elfhdr *) bufp;
764 bufp += sizeof(struct elfhdr);
765 memcpy(elf->e_ident, ELFMAG, SELFMAG);
766 elf->e_ident[EI_CLASS] = ELF_CLASS;
767 elf->e_ident[EI_DATA] = ELF_DATA;
768 elf->e_ident[EI_VERSION] = EV_CURRENT;
769 elf->e_ident[EI_OSABI] = ELF_OSABI;
770 memset(elf->e_ident+EI_PAD, 0, EI_NIDENT-EI_PAD);
771 elf->e_type = ET_CORE;
772 elf->e_machine = ELF_ARCH;
773 elf->e_version = EV_CURRENT;
775 elf->e_phoff = sizeof(struct elfhdr);
777 elf->e_flags = ELF_CORE_EFLAGS;
778 elf->e_ehsize = sizeof(struct elfhdr);
779 elf->e_phentsize = sizeof(struct elf_phdr);
781 elf->e_shentsize = 0;
789 * Traverse through memblock structure and setup crash memory ranges. These
790 * ranges will be used create PT_LOAD program headers in elfcore header.
792 static void fadump_setup_crash_memory_ranges(void)
794 struct memblock_region *reg;
795 unsigned long long start, end;
797 pr_debug("Setup crash memory ranges.\n");
798 crash_mem_ranges = 0;
800 * add the first memory chunk (RMA_START through boot_memory_size) as
801 * a separate memory chunk. The reason is, at the time crash firmware
802 * will move the content of this memory chunk to different location
803 * specified during fadump registration. We need to create a separate
804 * program header for this chunk with the correct offset.
806 fadump_add_crash_memory(RMA_START, fw_dump.boot_memory_size);
808 for_each_memblock(memory, reg) {
809 start = (unsigned long long)reg->base;
810 end = start + (unsigned long long)reg->size;
811 if (start == RMA_START && end >= fw_dump.boot_memory_size)
812 start = fw_dump.boot_memory_size;
814 /* add this range excluding the reserved dump area. */
815 fadump_exclude_reserved_area(start, end);
820 * If the given physical address falls within the boot memory region then
821 * return the relocated address that points to the dump region reserved
822 * for saving initial boot memory contents.
824 static inline unsigned long fadump_relocate(unsigned long paddr)
826 if (paddr > RMA_START && paddr < fw_dump.boot_memory_size)
827 return fdm.rmr_region.destination_address + paddr;
832 static int fadump_create_elfcore_headers(char *bufp)
835 struct elf_phdr *phdr;
838 fadump_init_elfcore_header(bufp);
839 elf = (struct elfhdr *)bufp;
840 bufp += sizeof(struct elfhdr);
843 * setup ELF PT_NOTE, place holder for cpu notes info. The notes info
844 * will be populated during second kernel boot after crash. Hence
845 * this PT_NOTE will always be the first elf note.
847 * NOTE: Any new ELF note addition should be placed after this note.
849 phdr = (struct elf_phdr *)bufp;
850 bufp += sizeof(struct elf_phdr);
851 phdr->p_type = PT_NOTE;
863 /* setup ELF PT_NOTE for vmcoreinfo */
864 phdr = (struct elf_phdr *)bufp;
865 bufp += sizeof(struct elf_phdr);
866 phdr->p_type = PT_NOTE;
871 phdr->p_paddr = fadump_relocate(paddr_vmcoreinfo_note());
872 phdr->p_offset = phdr->p_paddr;
873 phdr->p_memsz = vmcoreinfo_max_size;
874 phdr->p_filesz = vmcoreinfo_max_size;
876 /* Increment number of program headers. */
879 /* setup PT_LOAD sections. */
881 for (i = 0; i < crash_mem_ranges; i++) {
882 unsigned long long mbase, msize;
883 mbase = crash_memory_ranges[i].base;
884 msize = crash_memory_ranges[i].size;
889 phdr = (struct elf_phdr *)bufp;
890 bufp += sizeof(struct elf_phdr);
891 phdr->p_type = PT_LOAD;
892 phdr->p_flags = PF_R|PF_W|PF_X;
893 phdr->p_offset = mbase;
895 if (mbase == RMA_START) {
897 * The entire RMA region will be moved by firmware
898 * to the specified destination_address. Hence set
899 * the correct offset.
901 phdr->p_offset = fdm.rmr_region.destination_address;
904 phdr->p_paddr = mbase;
905 phdr->p_vaddr = (unsigned long)__va(mbase);
906 phdr->p_filesz = msize;
907 phdr->p_memsz = msize;
910 /* Increment number of program headers. */
916 static unsigned long init_fadump_header(unsigned long addr)
918 struct fadump_crash_info_header *fdh;
923 fw_dump.fadumphdr_addr = addr;
925 addr += sizeof(struct fadump_crash_info_header);
927 memset(fdh, 0, sizeof(struct fadump_crash_info_header));
928 fdh->magic_number = FADUMP_CRASH_INFO_MAGIC;
929 fdh->elfcorehdr_addr = addr;
930 /* We will set the crashing cpu id in crash_fadump() during crash. */
931 fdh->crashing_cpu = CPU_UNKNOWN;
936 static void register_fadump(void)
942 * If no memory is reserved then we can not register for firmware-
945 if (!fw_dump.reserve_dump_area_size)
948 fadump_setup_crash_memory_ranges();
950 addr = fdm.rmr_region.destination_address + fdm.rmr_region.source_len;
951 /* Initialize fadump crash info header. */
952 addr = init_fadump_header(addr);
955 pr_debug("Creating ELF core headers at %#016lx\n", addr);
956 fadump_create_elfcore_headers(vaddr);
958 /* register the future kernel dump with firmware. */
959 register_fw_dump(&fdm);
962 static int fadump_unregister_dump(struct fadump_mem_struct *fdm)
965 unsigned int wait_time;
967 pr_debug("Un-register firmware-assisted dump\n");
969 /* TODO: Add upper time limit for the delay */
971 rc = rtas_call(fw_dump.ibm_configure_kernel_dump, 3, 1, NULL,
972 FADUMP_UNREGISTER, fdm,
973 sizeof(struct fadump_mem_struct));
975 wait_time = rtas_busy_delay_time(rc);
981 printk(KERN_ERR "Failed to un-register firmware-assisted dump."
982 " unexpected error(%d).\n", rc);
985 fw_dump.dump_registered = 0;
989 static int fadump_invalidate_dump(struct fadump_mem_struct *fdm)
992 unsigned int wait_time;
994 pr_debug("Invalidating firmware-assisted dump registration\n");
996 /* TODO: Add upper time limit for the delay */
998 rc = rtas_call(fw_dump.ibm_configure_kernel_dump, 3, 1, NULL,
999 FADUMP_INVALIDATE, fdm,
1000 sizeof(struct fadump_mem_struct));
1002 wait_time = rtas_busy_delay_time(rc);
1005 } while (wait_time);
1008 printk(KERN_ERR "Failed to invalidate firmware-assisted dump "
1009 "rgistration. unexpected error(%d).\n", rc);
1012 fw_dump.dump_active = 0;
1017 void fadump_cleanup(void)
1019 /* Invalidate the registration only if dump is active. */
1020 if (fw_dump.dump_active) {
1021 init_fadump_mem_struct(&fdm,
1022 fdm_active->cpu_state_data.destination_address);
1023 fadump_invalidate_dump(&fdm);
1028 * Release the memory that was reserved in early boot to preserve the memory
1029 * contents. The released memory will be available for general use.
1031 static void fadump_release_memory(unsigned long begin, unsigned long end)
1034 unsigned long ra_start, ra_end;
1036 ra_start = fw_dump.reserve_dump_area_start;
1037 ra_end = ra_start + fw_dump.reserve_dump_area_size;
1039 for (addr = begin; addr < end; addr += PAGE_SIZE) {
1041 * exclude the dump reserve area. Will reuse it for next
1042 * fadump registration.
1044 if (addr <= ra_end && ((addr + PAGE_SIZE) > ra_start))
1047 ClearPageReserved(pfn_to_page(addr >> PAGE_SHIFT));
1048 init_page_count(pfn_to_page(addr >> PAGE_SHIFT));
1049 free_page((unsigned long)__va(addr));
1054 static void fadump_invalidate_release_mem(void)
1056 unsigned long reserved_area_start, reserved_area_end;
1057 unsigned long destination_address;
1059 mutex_lock(&fadump_mutex);
1060 if (!fw_dump.dump_active) {
1061 mutex_unlock(&fadump_mutex);
1065 destination_address = fdm_active->cpu_state_data.destination_address;
1067 mutex_unlock(&fadump_mutex);
1070 * Save the current reserved memory bounds we will require them
1071 * later for releasing the memory for general use.
1073 reserved_area_start = fw_dump.reserve_dump_area_start;
1074 reserved_area_end = reserved_area_start +
1075 fw_dump.reserve_dump_area_size;
1077 * Setup reserve_dump_area_start and its size so that we can
1078 * reuse this reserved memory for Re-registration.
1080 fw_dump.reserve_dump_area_start = destination_address;
1081 fw_dump.reserve_dump_area_size = get_fadump_area_size();
1083 fadump_release_memory(reserved_area_start, reserved_area_end);
1084 if (fw_dump.cpu_notes_buf) {
1085 fadump_cpu_notes_buf_free(
1086 (unsigned long)__va(fw_dump.cpu_notes_buf),
1087 fw_dump.cpu_notes_buf_size);
1088 fw_dump.cpu_notes_buf = 0;
1089 fw_dump.cpu_notes_buf_size = 0;
1091 /* Initialize the kernel dump memory structure for FAD registration. */
1092 init_fadump_mem_struct(&fdm, fw_dump.reserve_dump_area_start);
1095 static ssize_t fadump_release_memory_store(struct kobject *kobj,
1096 struct kobj_attribute *attr,
1097 const char *buf, size_t count)
1099 if (!fw_dump.dump_active)
1102 if (buf[0] == '1') {
1104 * Take away the '/proc/vmcore'. We are releasing the dump
1105 * memory, hence it will not be valid anymore.
1108 fadump_invalidate_release_mem();
1115 static ssize_t fadump_enabled_show(struct kobject *kobj,
1116 struct kobj_attribute *attr,
1119 return sprintf(buf, "%d\n", fw_dump.fadump_enabled);
1122 static ssize_t fadump_register_show(struct kobject *kobj,
1123 struct kobj_attribute *attr,
1126 return sprintf(buf, "%d\n", fw_dump.dump_registered);
1129 static ssize_t fadump_register_store(struct kobject *kobj,
1130 struct kobj_attribute *attr,
1131 const char *buf, size_t count)
1135 if (!fw_dump.fadump_enabled || fdm_active)
1138 mutex_lock(&fadump_mutex);
1142 if (fw_dump.dump_registered == 0) {
1146 /* Un-register Firmware-assisted dump */
1147 fadump_unregister_dump(&fdm);
1150 if (fw_dump.dump_registered == 1) {
1154 /* Register Firmware-assisted dump */
1163 mutex_unlock(&fadump_mutex);
1164 return ret < 0 ? ret : count;
1167 static int fadump_region_show(struct seq_file *m, void *private)
1169 const struct fadump_mem_struct *fdm_ptr;
1171 if (!fw_dump.fadump_enabled)
1174 mutex_lock(&fadump_mutex);
1176 fdm_ptr = fdm_active;
1178 mutex_unlock(&fadump_mutex);
1183 "CPU : [%#016llx-%#016llx] %#llx bytes, "
1185 fdm_ptr->cpu_state_data.destination_address,
1186 fdm_ptr->cpu_state_data.destination_address +
1187 fdm_ptr->cpu_state_data.source_len - 1,
1188 fdm_ptr->cpu_state_data.source_len,
1189 fdm_ptr->cpu_state_data.bytes_dumped);
1191 "HPTE: [%#016llx-%#016llx] %#llx bytes, "
1193 fdm_ptr->hpte_region.destination_address,
1194 fdm_ptr->hpte_region.destination_address +
1195 fdm_ptr->hpte_region.source_len - 1,
1196 fdm_ptr->hpte_region.source_len,
1197 fdm_ptr->hpte_region.bytes_dumped);
1199 "DUMP: [%#016llx-%#016llx] %#llx bytes, "
1201 fdm_ptr->rmr_region.destination_address,
1202 fdm_ptr->rmr_region.destination_address +
1203 fdm_ptr->rmr_region.source_len - 1,
1204 fdm_ptr->rmr_region.source_len,
1205 fdm_ptr->rmr_region.bytes_dumped);
1208 (fw_dump.reserve_dump_area_start ==
1209 fdm_ptr->cpu_state_data.destination_address))
1212 /* Dump is active. Show reserved memory region. */
1214 " : [%#016llx-%#016llx] %#llx bytes, "
1216 (unsigned long long)fw_dump.reserve_dump_area_start,
1217 fdm_ptr->cpu_state_data.destination_address - 1,
1218 fdm_ptr->cpu_state_data.destination_address -
1219 fw_dump.reserve_dump_area_start,
1220 fdm_ptr->cpu_state_data.destination_address -
1221 fw_dump.reserve_dump_area_start);
1224 mutex_unlock(&fadump_mutex);
1228 static struct kobj_attribute fadump_release_attr = __ATTR(fadump_release_mem,
1230 fadump_release_memory_store);
1231 static struct kobj_attribute fadump_attr = __ATTR(fadump_enabled,
1232 0444, fadump_enabled_show,
1234 static struct kobj_attribute fadump_register_attr = __ATTR(fadump_registered,
1235 0644, fadump_register_show,
1236 fadump_register_store);
1238 static int fadump_region_open(struct inode *inode, struct file *file)
1240 return single_open(file, fadump_region_show, inode->i_private);
1243 static const struct file_operations fadump_region_fops = {
1244 .open = fadump_region_open,
1246 .llseek = seq_lseek,
1247 .release = single_release,
1250 static void fadump_init_files(void)
1252 struct dentry *debugfs_file;
1255 rc = sysfs_create_file(kernel_kobj, &fadump_attr.attr);
1257 printk(KERN_ERR "fadump: unable to create sysfs file"
1258 " fadump_enabled (%d)\n", rc);
1260 rc = sysfs_create_file(kernel_kobj, &fadump_register_attr.attr);
1262 printk(KERN_ERR "fadump: unable to create sysfs file"
1263 " fadump_registered (%d)\n", rc);
1265 debugfs_file = debugfs_create_file("fadump_region", 0444,
1266 powerpc_debugfs_root, NULL,
1267 &fadump_region_fops);
1269 printk(KERN_ERR "fadump: unable to create debugfs file"
1270 " fadump_region\n");
1272 if (fw_dump.dump_active) {
1273 rc = sysfs_create_file(kernel_kobj, &fadump_release_attr.attr);
1275 printk(KERN_ERR "fadump: unable to create sysfs file"
1276 " fadump_release_mem (%d)\n", rc);
1282 * Prepare for firmware-assisted dump.
1284 int __init setup_fadump(void)
1286 if (!fw_dump.fadump_enabled)
1289 if (!fw_dump.fadump_supported) {
1290 printk(KERN_ERR "Firmware-assisted dump is not supported on"
1291 " this hardware\n");
1295 fadump_show_config();
1297 * If dump data is available then see if it is valid and prepare for
1298 * saving it to the disk.
1300 if (fw_dump.dump_active) {
1302 * if dump process fails then invalidate the registration
1303 * and release memory before proceeding for re-registration.
1305 if (process_fadump(fdm_active) < 0)
1306 fadump_invalidate_release_mem();
1308 /* Initialize the kernel dump memory structure for FAD registration. */
1309 else if (fw_dump.reserve_dump_area_size)
1310 init_fadump_mem_struct(&fdm, fw_dump.reserve_dump_area_start);
1311 fadump_init_files();
1315 subsys_initcall(setup_fadump);