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>
38 #include <linux/slab.h>
43 #include <asm/fadump.h>
44 #include <asm/debug.h>
45 #include <asm/setup.h>
47 static struct fw_dump fw_dump;
48 static struct fadump_mem_struct fdm;
49 static const struct fadump_mem_struct *fdm_active;
51 static DEFINE_MUTEX(fadump_mutex);
52 struct fad_crash_memory_ranges *crash_memory_ranges;
53 int crash_memory_ranges_size;
55 int max_crash_mem_ranges;
57 /* Scan the Firmware Assisted dump configuration details. */
58 int __init early_init_dt_scan_fw_dump(unsigned long node,
59 const char *uname, int depth, void *data)
61 const __be32 *sections;
66 if (depth != 1 || strcmp(uname, "rtas") != 0)
70 * Check if Firmware Assisted dump is supported. if yes, check
71 * if dump has been initiated on last reboot.
73 token = of_get_flat_dt_prop(node, "ibm,configure-kernel-dump", NULL);
77 fw_dump.fadump_supported = 1;
78 fw_dump.ibm_configure_kernel_dump = *token;
81 * The 'ibm,kernel-dump' rtas node is present only if there is
82 * dump data waiting for us.
84 fdm_active = of_get_flat_dt_prop(node, "ibm,kernel-dump", NULL);
86 fw_dump.dump_active = 1;
88 /* Get the sizes required to store dump data for the firmware provided
90 * For each dump section type supported, a 32bit cell which defines
91 * the ID of a supported section followed by two 32 bit cells which
92 * gives teh size of the section in bytes.
94 sections = of_get_flat_dt_prop(node, "ibm,configure-kernel-dump-sizes",
100 num_sections = size / (3 * sizeof(u32));
102 for (i = 0; i < num_sections; i++, sections += 3) {
103 u32 type = (u32)of_read_number(sections, 1);
106 case FADUMP_CPU_STATE_DATA:
107 fw_dump.cpu_state_data_size =
108 of_read_ulong(§ions[1], 2);
110 case FADUMP_HPTE_REGION:
111 fw_dump.hpte_region_size =
112 of_read_ulong(§ions[1], 2);
120 int is_fadump_active(void)
122 return fw_dump.dump_active;
125 /* Print firmware assisted dump configurations for debugging purpose. */
126 static void fadump_show_config(void)
128 pr_debug("Support for firmware-assisted dump (fadump): %s\n",
129 (fw_dump.fadump_supported ? "present" : "no support"));
131 if (!fw_dump.fadump_supported)
134 pr_debug("Fadump enabled : %s\n",
135 (fw_dump.fadump_enabled ? "yes" : "no"));
136 pr_debug("Dump Active : %s\n",
137 (fw_dump.dump_active ? "yes" : "no"));
138 pr_debug("Dump section sizes:\n");
139 pr_debug(" CPU state data size: %lx\n", fw_dump.cpu_state_data_size);
140 pr_debug(" HPTE region size : %lx\n", fw_dump.hpte_region_size);
141 pr_debug("Boot memory size : %lx\n", fw_dump.boot_memory_size);
144 static unsigned long init_fadump_mem_struct(struct fadump_mem_struct *fdm,
150 memset(fdm, 0, sizeof(struct fadump_mem_struct));
151 addr = addr & PAGE_MASK;
153 fdm->header.dump_format_version = 0x00000001;
154 fdm->header.dump_num_sections = 3;
155 fdm->header.dump_status_flag = 0;
156 fdm->header.offset_first_dump_section =
157 (u32)offsetof(struct fadump_mem_struct, cpu_state_data);
160 * Fields for disk dump option.
161 * We are not using disk dump option, hence set these fields to 0.
163 fdm->header.dd_block_size = 0;
164 fdm->header.dd_block_offset = 0;
165 fdm->header.dd_num_blocks = 0;
166 fdm->header.dd_offset_disk_path = 0;
168 /* set 0 to disable an automatic dump-reboot. */
169 fdm->header.max_time_auto = 0;
171 /* Kernel dump sections */
172 /* cpu state data section. */
173 fdm->cpu_state_data.request_flag = FADUMP_REQUEST_FLAG;
174 fdm->cpu_state_data.source_data_type = FADUMP_CPU_STATE_DATA;
175 fdm->cpu_state_data.source_address = 0;
176 fdm->cpu_state_data.source_len = fw_dump.cpu_state_data_size;
177 fdm->cpu_state_data.destination_address = addr;
178 addr += fw_dump.cpu_state_data_size;
180 /* hpte region section */
181 fdm->hpte_region.request_flag = FADUMP_REQUEST_FLAG;
182 fdm->hpte_region.source_data_type = FADUMP_HPTE_REGION;
183 fdm->hpte_region.source_address = 0;
184 fdm->hpte_region.source_len = fw_dump.hpte_region_size;
185 fdm->hpte_region.destination_address = addr;
186 addr += fw_dump.hpte_region_size;
188 /* RMA region section */
189 fdm->rmr_region.request_flag = FADUMP_REQUEST_FLAG;
190 fdm->rmr_region.source_data_type = FADUMP_REAL_MODE_REGION;
191 fdm->rmr_region.source_address = RMA_START;
192 fdm->rmr_region.source_len = fw_dump.boot_memory_size;
193 fdm->rmr_region.destination_address = addr;
194 addr += fw_dump.boot_memory_size;
200 * fadump_calculate_reserve_size(): reserve variable boot area 5% of System RAM
202 * Function to find the largest memory size we need to reserve during early
203 * boot process. This will be the size of the memory that is required for a
204 * kernel to boot successfully.
206 * This function has been taken from phyp-assisted dump feature implementation.
208 * returns larger of 256MB or 5% rounded down to multiples of 256MB.
210 * TODO: Come up with better approach to find out more accurate memory size
211 * that is required for a kernel to boot successfully.
214 static inline unsigned long fadump_calculate_reserve_size(void)
219 * Check if the size is specified through fadump_reserve_mem= cmdline
220 * option. If yes, then use that.
222 if (fw_dump.reserve_bootvar)
223 return fw_dump.reserve_bootvar;
225 /* divide by 20 to get 5% of value */
226 size = memblock_end_of_DRAM() / 20;
228 /* round it down in multiples of 256 */
229 size = size & ~0x0FFFFFFFUL;
231 /* Truncate to memory_limit. We don't want to over reserve the memory.*/
232 if (memory_limit && size > memory_limit)
235 return (size > MIN_BOOT_MEM ? size : MIN_BOOT_MEM);
239 * Calculate the total memory size required to be reserved for
240 * firmware-assisted dump registration.
242 static unsigned long get_fadump_area_size(void)
244 unsigned long size = 0;
246 size += fw_dump.cpu_state_data_size;
247 size += fw_dump.hpte_region_size;
248 size += fw_dump.boot_memory_size;
249 size += sizeof(struct fadump_crash_info_header);
250 size += sizeof(struct elfhdr); /* ELF core header.*/
251 size += sizeof(struct elf_phdr); /* place holder for cpu notes */
252 /* Program headers for crash memory regions. */
253 size += sizeof(struct elf_phdr) * (memblock_num_regions(memory) + 2);
255 size = PAGE_ALIGN(size);
259 int __init fadump_reserve_mem(void)
261 unsigned long base, size, memory_boundary;
263 if (!fw_dump.fadump_enabled)
266 if (!fw_dump.fadump_supported) {
267 printk(KERN_INFO "Firmware-assisted dump is not supported on"
269 fw_dump.fadump_enabled = 0;
273 * Initialize boot memory size
274 * If dump is active then we have already calculated the size during
278 fw_dump.boot_memory_size = fdm_active->rmr_region.source_len;
280 fw_dump.boot_memory_size = fadump_calculate_reserve_size();
283 * Calculate the memory boundary.
284 * If memory_limit is less than actual memory boundary then reserve
285 * the memory for fadump beyond the memory_limit and adjust the
286 * memory_limit accordingly, so that the running kernel can run with
287 * specified memory_limit.
289 if (memory_limit && memory_limit < memblock_end_of_DRAM()) {
290 size = get_fadump_area_size();
291 if ((memory_limit + size) < memblock_end_of_DRAM())
292 memory_limit += size;
294 memory_limit = memblock_end_of_DRAM();
295 printk(KERN_INFO "Adjusted memory_limit for firmware-assisted"
296 " dump, now %#016llx\n", memory_limit);
299 memory_boundary = memory_limit;
301 memory_boundary = memblock_end_of_DRAM();
303 if (fw_dump.dump_active) {
304 printk(KERN_INFO "Firmware-assisted dump is active.\n");
306 * If last boot has crashed then reserve all the memory
307 * above boot_memory_size so that we don't touch it until
308 * dump is written to disk by userspace tool. This memory
309 * will be released for general use once the dump is saved.
311 base = fw_dump.boot_memory_size;
312 size = memory_boundary - base;
313 memblock_reserve(base, size);
314 printk(KERN_INFO "Reserved %ldMB of memory at %ldMB "
315 "for saving crash dump\n",
316 (unsigned long)(size >> 20),
317 (unsigned long)(base >> 20));
319 fw_dump.fadumphdr_addr =
320 fdm_active->rmr_region.destination_address +
321 fdm_active->rmr_region.source_len;
322 pr_debug("fadumphdr_addr = %p\n",
323 (void *) fw_dump.fadumphdr_addr);
325 /* Reserve the memory at the top of memory. */
326 size = get_fadump_area_size();
327 base = memory_boundary - size;
328 memblock_reserve(base, size);
329 printk(KERN_INFO "Reserved %ldMB of memory at %ldMB "
330 "for firmware-assisted dump\n",
331 (unsigned long)(size >> 20),
332 (unsigned long)(base >> 20));
334 fw_dump.reserve_dump_area_start = base;
335 fw_dump.reserve_dump_area_size = size;
339 /* Look for fadump= cmdline option. */
340 static int __init early_fadump_param(char *p)
345 if (strncmp(p, "on", 2) == 0)
346 fw_dump.fadump_enabled = 1;
347 else if (strncmp(p, "off", 3) == 0)
348 fw_dump.fadump_enabled = 0;
352 early_param("fadump", early_fadump_param);
354 /* Look for fadump_reserve_mem= cmdline option */
355 static int __init early_fadump_reserve_mem(char *p)
358 fw_dump.reserve_bootvar = memparse(p, &p);
361 early_param("fadump_reserve_mem", early_fadump_reserve_mem);
363 static void register_fw_dump(struct fadump_mem_struct *fdm)
366 unsigned int wait_time;
368 pr_debug("Registering for firmware-assisted kernel dump...\n");
370 /* TODO: Add upper time limit for the delay */
372 rc = rtas_call(fw_dump.ibm_configure_kernel_dump, 3, 1, NULL,
373 FADUMP_REGISTER, fdm,
374 sizeof(struct fadump_mem_struct));
376 wait_time = rtas_busy_delay_time(rc);
384 printk(KERN_ERR "Failed to register firmware-assisted kernel"
385 " dump. Hardware Error(%d).\n", rc);
388 printk(KERN_ERR "Failed to register firmware-assisted kernel"
389 " dump. Parameter Error(%d).\n", rc);
392 printk(KERN_ERR "firmware-assisted kernel dump is already "
394 fw_dump.dump_registered = 1;
397 printk(KERN_INFO "firmware-assisted kernel dump registration"
399 fw_dump.dump_registered = 1;
404 void crash_fadump(struct pt_regs *regs, const char *str)
406 struct fadump_crash_info_header *fdh = NULL;
408 if (!fw_dump.dump_registered || !fw_dump.fadumphdr_addr)
411 fdh = __va(fw_dump.fadumphdr_addr);
412 crashing_cpu = smp_processor_id();
413 fdh->crashing_cpu = crashing_cpu;
414 crash_save_vmcoreinfo();
419 ppc_save_regs(&fdh->regs);
421 fdh->cpu_online_mask = *cpu_online_mask;
423 /* Call ibm,os-term rtas call to trigger firmware assisted dump */
424 rtas_os_term((char *)str);
427 #define GPR_MASK 0xffffff0000000000
428 static inline int fadump_gpr_index(u64 id)
433 if ((id & GPR_MASK) == REG_ID("GPR")) {
434 /* get the digits at the end */
439 str[0] = (id >> 8) & 0xff;
440 sscanf(str, "%d", &i);
447 static inline void fadump_set_regval(struct pt_regs *regs, u64 reg_id,
452 i = fadump_gpr_index(reg_id);
454 regs->gpr[i] = (unsigned long)reg_val;
455 else if (reg_id == REG_ID("NIA"))
456 regs->nip = (unsigned long)reg_val;
457 else if (reg_id == REG_ID("MSR"))
458 regs->msr = (unsigned long)reg_val;
459 else if (reg_id == REG_ID("CTR"))
460 regs->ctr = (unsigned long)reg_val;
461 else if (reg_id == REG_ID("LR"))
462 regs->link = (unsigned long)reg_val;
463 else if (reg_id == REG_ID("XER"))
464 regs->xer = (unsigned long)reg_val;
465 else if (reg_id == REG_ID("CR"))
466 regs->ccr = (unsigned long)reg_val;
467 else if (reg_id == REG_ID("DAR"))
468 regs->dar = (unsigned long)reg_val;
469 else if (reg_id == REG_ID("DSISR"))
470 regs->dsisr = (unsigned long)reg_val;
473 static struct fadump_reg_entry*
474 fadump_read_registers(struct fadump_reg_entry *reg_entry, struct pt_regs *regs)
476 memset(regs, 0, sizeof(struct pt_regs));
478 while (reg_entry->reg_id != REG_ID("CPUEND")) {
479 fadump_set_regval(regs, reg_entry->reg_id,
480 reg_entry->reg_value);
487 static u32 *fadump_append_elf_note(u32 *buf, char *name, unsigned type,
488 void *data, size_t data_len)
490 struct elf_note note;
492 note.n_namesz = strlen(name) + 1;
493 note.n_descsz = data_len;
495 memcpy(buf, ¬e, sizeof(note));
496 buf += (sizeof(note) + 3)/4;
497 memcpy(buf, name, note.n_namesz);
498 buf += (note.n_namesz + 3)/4;
499 memcpy(buf, data, note.n_descsz);
500 buf += (note.n_descsz + 3)/4;
505 static void fadump_final_note(u32 *buf)
507 struct elf_note note;
512 memcpy(buf, ¬e, sizeof(note));
515 static u32 *fadump_regs_to_elf_notes(u32 *buf, struct pt_regs *regs)
517 struct elf_prstatus prstatus;
519 memset(&prstatus, 0, sizeof(prstatus));
521 * FIXME: How do i get PID? Do I really need it?
522 * prstatus.pr_pid = ????
524 elf_core_copy_kernel_regs(&prstatus.pr_reg, regs);
525 buf = fadump_append_elf_note(buf, KEXEC_CORE_NOTE_NAME, NT_PRSTATUS,
526 &prstatus, sizeof(prstatus));
530 static void fadump_update_elfcore_header(char *bufp)
533 struct elf_phdr *phdr;
535 elf = (struct elfhdr *)bufp;
536 bufp += sizeof(struct elfhdr);
538 /* First note is a place holder for cpu notes info. */
539 phdr = (struct elf_phdr *)bufp;
541 if (phdr->p_type == PT_NOTE) {
542 phdr->p_paddr = fw_dump.cpu_notes_buf;
543 phdr->p_offset = phdr->p_paddr;
544 phdr->p_filesz = fw_dump.cpu_notes_buf_size;
545 phdr->p_memsz = fw_dump.cpu_notes_buf_size;
550 static void *fadump_cpu_notes_buf_alloc(unsigned long size)
554 unsigned long order, count, i;
556 order = get_order(size);
557 vaddr = (void *)__get_free_pages(GFP_KERNEL|__GFP_ZERO, order);
562 page = virt_to_page(vaddr);
563 for (i = 0; i < count; i++)
564 SetPageReserved(page + i);
568 static void fadump_cpu_notes_buf_free(unsigned long vaddr, unsigned long size)
571 unsigned long order, count, i;
573 order = get_order(size);
575 page = virt_to_page(vaddr);
576 for (i = 0; i < count; i++)
577 ClearPageReserved(page + i);
578 __free_pages(page, order);
582 * Read CPU state dump data and convert it into ELF notes.
583 * The CPU dump starts with magic number "REGSAVE". NumCpusOffset should be
584 * used to access the data to allow for additional fields to be added without
585 * affecting compatibility. Each list of registers for a CPU starts with
586 * "CPUSTRT" and ends with "CPUEND". Each register entry is of 16 bytes,
587 * 8 Byte ASCII identifier and 8 Byte register value. The register entry
588 * with identifier "CPUSTRT" and "CPUEND" contains 4 byte cpu id as part
589 * of register value. For more details refer to PAPR document.
591 * Only for the crashing cpu we ignore the CPU dump data and get exact
592 * state from fadump crash info structure populated by first kernel at the
595 static int __init fadump_build_cpu_notes(const struct fadump_mem_struct *fdm)
597 struct fadump_reg_save_area_header *reg_header;
598 struct fadump_reg_entry *reg_entry;
599 struct fadump_crash_info_header *fdh = NULL;
602 u32 num_cpus, *note_buf;
604 int i, rc = 0, cpu = 0;
606 if (!fdm->cpu_state_data.bytes_dumped)
609 addr = fdm->cpu_state_data.destination_address;
613 if (reg_header->magic_number != REGSAVE_AREA_MAGIC) {
614 printk(KERN_ERR "Unable to read register save area.\n");
617 pr_debug("--------CPU State Data------------\n");
618 pr_debug("Magic Number: %llx\n", reg_header->magic_number);
619 pr_debug("NumCpuOffset: %x\n", reg_header->num_cpu_offset);
621 vaddr += reg_header->num_cpu_offset;
622 num_cpus = *((u32 *)(vaddr));
623 pr_debug("NumCpus : %u\n", num_cpus);
624 vaddr += sizeof(u32);
625 reg_entry = (struct fadump_reg_entry *)vaddr;
627 /* Allocate buffer to hold cpu crash notes. */
628 fw_dump.cpu_notes_buf_size = num_cpus * sizeof(note_buf_t);
629 fw_dump.cpu_notes_buf_size = PAGE_ALIGN(fw_dump.cpu_notes_buf_size);
630 note_buf = fadump_cpu_notes_buf_alloc(fw_dump.cpu_notes_buf_size);
632 printk(KERN_ERR "Failed to allocate 0x%lx bytes for "
633 "cpu notes buffer\n", fw_dump.cpu_notes_buf_size);
636 fw_dump.cpu_notes_buf = __pa(note_buf);
638 pr_debug("Allocated buffer for cpu notes of size %ld at %p\n",
639 (num_cpus * sizeof(note_buf_t)), note_buf);
641 if (fw_dump.fadumphdr_addr)
642 fdh = __va(fw_dump.fadumphdr_addr);
644 for (i = 0; i < num_cpus; i++) {
645 if (reg_entry->reg_id != REG_ID("CPUSTRT")) {
646 printk(KERN_ERR "Unable to read CPU state data\n");
650 /* Lower 4 bytes of reg_value contains logical cpu id */
651 cpu = reg_entry->reg_value & FADUMP_CPU_ID_MASK;
652 if (fdh && !cpumask_test_cpu(cpu, &fdh->cpu_online_mask)) {
653 SKIP_TO_NEXT_CPU(reg_entry);
656 pr_debug("Reading register data for cpu %d...\n", cpu);
657 if (fdh && fdh->crashing_cpu == cpu) {
659 note_buf = fadump_regs_to_elf_notes(note_buf, ®s);
660 SKIP_TO_NEXT_CPU(reg_entry);
663 reg_entry = fadump_read_registers(reg_entry, ®s);
664 note_buf = fadump_regs_to_elf_notes(note_buf, ®s);
667 fadump_final_note(note_buf);
670 pr_debug("Updating elfcore header (%llx) with cpu notes\n",
671 fdh->elfcorehdr_addr);
672 fadump_update_elfcore_header((char *)__va(fdh->elfcorehdr_addr));
677 fadump_cpu_notes_buf_free((unsigned long)__va(fw_dump.cpu_notes_buf),
678 fw_dump.cpu_notes_buf_size);
679 fw_dump.cpu_notes_buf = 0;
680 fw_dump.cpu_notes_buf_size = 0;
686 * Validate and process the dump data stored by firmware before exporting
687 * it through '/proc/vmcore'.
689 static int __init process_fadump(const struct fadump_mem_struct *fdm_active)
691 struct fadump_crash_info_header *fdh;
694 if (!fdm_active || !fw_dump.fadumphdr_addr)
697 /* Check if the dump data is valid. */
698 if ((fdm_active->header.dump_status_flag == FADUMP_ERROR_FLAG) ||
699 (fdm_active->cpu_state_data.error_flags != 0) ||
700 (fdm_active->rmr_region.error_flags != 0)) {
701 printk(KERN_ERR "Dump taken by platform is not valid\n");
704 if ((fdm_active->rmr_region.bytes_dumped !=
705 fdm_active->rmr_region.source_len) ||
706 !fdm_active->cpu_state_data.bytes_dumped) {
707 printk(KERN_ERR "Dump taken by platform is incomplete\n");
711 /* Validate the fadump crash info header */
712 fdh = __va(fw_dump.fadumphdr_addr);
713 if (fdh->magic_number != FADUMP_CRASH_INFO_MAGIC) {
714 printk(KERN_ERR "Crash info header is not valid.\n");
718 rc = fadump_build_cpu_notes(fdm_active);
723 * We are done validating dump info and elfcore header is now ready
724 * to be exported. set elfcorehdr_addr so that vmcore module will
725 * export the elfcore header through '/proc/vmcore'.
727 elfcorehdr_addr = fdh->elfcorehdr_addr;
732 static void free_crash_memory_ranges(void)
734 kfree(crash_memory_ranges);
735 crash_memory_ranges = NULL;
736 crash_memory_ranges_size = 0;
737 max_crash_mem_ranges = 0;
741 * Allocate or reallocate crash memory ranges array in incremental units
744 static int allocate_crash_memory_ranges(void)
746 struct fad_crash_memory_ranges *new_array;
749 new_size = crash_memory_ranges_size + PAGE_SIZE;
750 pr_debug("Allocating %llu bytes of memory for crash memory ranges\n",
753 new_array = krealloc(crash_memory_ranges, new_size, GFP_KERNEL);
754 if (new_array == NULL) {
755 pr_err("Insufficient memory for setting up crash memory ranges\n");
756 free_crash_memory_ranges();
760 crash_memory_ranges = new_array;
761 crash_memory_ranges_size = new_size;
762 max_crash_mem_ranges = (new_size /
763 sizeof(struct fad_crash_memory_ranges));
767 static inline int fadump_add_crash_memory(unsigned long long base,
768 unsigned long long end)
773 if (crash_mem_ranges == max_crash_mem_ranges) {
776 ret = allocate_crash_memory_ranges();
781 pr_debug("crash_memory_range[%d] [%#016llx-%#016llx], %#llx bytes\n",
782 crash_mem_ranges, base, end - 1, (end - base));
783 crash_memory_ranges[crash_mem_ranges].base = base;
784 crash_memory_ranges[crash_mem_ranges].size = end - base;
789 static int fadump_exclude_reserved_area(unsigned long long start,
790 unsigned long long end)
792 unsigned long long ra_start, ra_end;
795 ra_start = fw_dump.reserve_dump_area_start;
796 ra_end = ra_start + fw_dump.reserve_dump_area_size;
798 if ((ra_start < end) && (ra_end > start)) {
799 if ((start < ra_start) && (end > ra_end)) {
800 ret = fadump_add_crash_memory(start, ra_start);
804 ret = fadump_add_crash_memory(ra_end, end);
805 } else if (start < ra_start) {
806 ret = fadump_add_crash_memory(start, ra_start);
807 } else if (ra_end < end) {
808 ret = fadump_add_crash_memory(ra_end, end);
811 ret = fadump_add_crash_memory(start, end);
816 static int fadump_init_elfcore_header(char *bufp)
820 elf = (struct elfhdr *) bufp;
821 bufp += sizeof(struct elfhdr);
822 memcpy(elf->e_ident, ELFMAG, SELFMAG);
823 elf->e_ident[EI_CLASS] = ELF_CLASS;
824 elf->e_ident[EI_DATA] = ELF_DATA;
825 elf->e_ident[EI_VERSION] = EV_CURRENT;
826 elf->e_ident[EI_OSABI] = ELF_OSABI;
827 memset(elf->e_ident+EI_PAD, 0, EI_NIDENT-EI_PAD);
828 elf->e_type = ET_CORE;
829 elf->e_machine = ELF_ARCH;
830 elf->e_version = EV_CURRENT;
832 elf->e_phoff = sizeof(struct elfhdr);
834 elf->e_flags = ELF_CORE_EFLAGS;
835 elf->e_ehsize = sizeof(struct elfhdr);
836 elf->e_phentsize = sizeof(struct elf_phdr);
838 elf->e_shentsize = 0;
846 * Traverse through memblock structure and setup crash memory ranges. These
847 * ranges will be used create PT_LOAD program headers in elfcore header.
849 static int fadump_setup_crash_memory_ranges(void)
851 struct memblock_region *reg;
852 unsigned long long start, end;
855 pr_debug("Setup crash memory ranges.\n");
856 crash_mem_ranges = 0;
858 * add the first memory chunk (RMA_START through boot_memory_size) as
859 * a separate memory chunk. The reason is, at the time crash firmware
860 * will move the content of this memory chunk to different location
861 * specified during fadump registration. We need to create a separate
862 * program header for this chunk with the correct offset.
864 ret = fadump_add_crash_memory(RMA_START, fw_dump.boot_memory_size);
868 for_each_memblock(memory, reg) {
869 start = (unsigned long long)reg->base;
870 end = start + (unsigned long long)reg->size;
871 if (start == RMA_START && end >= fw_dump.boot_memory_size)
872 start = fw_dump.boot_memory_size;
874 /* add this range excluding the reserved dump area. */
875 ret = fadump_exclude_reserved_area(start, end);
884 * If the given physical address falls within the boot memory region then
885 * return the relocated address that points to the dump region reserved
886 * for saving initial boot memory contents.
888 static inline unsigned long fadump_relocate(unsigned long paddr)
890 if (paddr > RMA_START && paddr < fw_dump.boot_memory_size)
891 return fdm.rmr_region.destination_address + paddr;
896 static int fadump_create_elfcore_headers(char *bufp)
899 struct elf_phdr *phdr;
902 fadump_init_elfcore_header(bufp);
903 elf = (struct elfhdr *)bufp;
904 bufp += sizeof(struct elfhdr);
907 * setup ELF PT_NOTE, place holder for cpu notes info. The notes info
908 * will be populated during second kernel boot after crash. Hence
909 * this PT_NOTE will always be the first elf note.
911 * NOTE: Any new ELF note addition should be placed after this note.
913 phdr = (struct elf_phdr *)bufp;
914 bufp += sizeof(struct elf_phdr);
915 phdr->p_type = PT_NOTE;
927 /* setup ELF PT_NOTE for vmcoreinfo */
928 phdr = (struct elf_phdr *)bufp;
929 bufp += sizeof(struct elf_phdr);
930 phdr->p_type = PT_NOTE;
935 phdr->p_paddr = fadump_relocate(paddr_vmcoreinfo_note());
936 phdr->p_offset = phdr->p_paddr;
937 phdr->p_memsz = vmcoreinfo_max_size;
938 phdr->p_filesz = vmcoreinfo_max_size;
940 /* Increment number of program headers. */
943 /* setup PT_LOAD sections. */
945 for (i = 0; i < crash_mem_ranges; i++) {
946 unsigned long long mbase, msize;
947 mbase = crash_memory_ranges[i].base;
948 msize = crash_memory_ranges[i].size;
953 phdr = (struct elf_phdr *)bufp;
954 bufp += sizeof(struct elf_phdr);
955 phdr->p_type = PT_LOAD;
956 phdr->p_flags = PF_R|PF_W|PF_X;
957 phdr->p_offset = mbase;
959 if (mbase == RMA_START) {
961 * The entire RMA region will be moved by firmware
962 * to the specified destination_address. Hence set
963 * the correct offset.
965 phdr->p_offset = fdm.rmr_region.destination_address;
968 phdr->p_paddr = mbase;
969 phdr->p_vaddr = (unsigned long)__va(mbase);
970 phdr->p_filesz = msize;
971 phdr->p_memsz = msize;
974 /* Increment number of program headers. */
980 static unsigned long init_fadump_header(unsigned long addr)
982 struct fadump_crash_info_header *fdh;
987 fw_dump.fadumphdr_addr = addr;
989 addr += sizeof(struct fadump_crash_info_header);
991 memset(fdh, 0, sizeof(struct fadump_crash_info_header));
992 fdh->magic_number = FADUMP_CRASH_INFO_MAGIC;
993 fdh->elfcorehdr_addr = addr;
994 /* We will set the crashing cpu id in crash_fadump() during crash. */
995 fdh->crashing_cpu = CPU_UNKNOWN;
1000 static void register_fadump(void)
1007 * If no memory is reserved then we can not register for firmware-
1010 if (!fw_dump.reserve_dump_area_size)
1013 ret = fadump_setup_crash_memory_ranges();
1017 addr = fdm.rmr_region.destination_address + fdm.rmr_region.source_len;
1018 /* Initialize fadump crash info header. */
1019 addr = init_fadump_header(addr);
1022 pr_debug("Creating ELF core headers at %#016lx\n", addr);
1023 fadump_create_elfcore_headers(vaddr);
1025 /* register the future kernel dump with firmware. */
1026 register_fw_dump(&fdm);
1029 static int fadump_unregister_dump(struct fadump_mem_struct *fdm)
1032 unsigned int wait_time;
1034 pr_debug("Un-register firmware-assisted dump\n");
1036 /* TODO: Add upper time limit for the delay */
1038 rc = rtas_call(fw_dump.ibm_configure_kernel_dump, 3, 1, NULL,
1039 FADUMP_UNREGISTER, fdm,
1040 sizeof(struct fadump_mem_struct));
1042 wait_time = rtas_busy_delay_time(rc);
1045 } while (wait_time);
1048 printk(KERN_ERR "Failed to un-register firmware-assisted dump."
1049 " unexpected error(%d).\n", rc);
1052 fw_dump.dump_registered = 0;
1056 static int fadump_invalidate_dump(struct fadump_mem_struct *fdm)
1059 unsigned int wait_time;
1061 pr_debug("Invalidating firmware-assisted dump registration\n");
1063 /* TODO: Add upper time limit for the delay */
1065 rc = rtas_call(fw_dump.ibm_configure_kernel_dump, 3, 1, NULL,
1066 FADUMP_INVALIDATE, fdm,
1067 sizeof(struct fadump_mem_struct));
1069 wait_time = rtas_busy_delay_time(rc);
1072 } while (wait_time);
1075 printk(KERN_ERR "Failed to invalidate firmware-assisted dump "
1076 "rgistration. unexpected error(%d).\n", rc);
1079 fw_dump.dump_active = 0;
1084 void fadump_cleanup(void)
1086 /* Invalidate the registration only if dump is active. */
1087 if (fw_dump.dump_active) {
1088 init_fadump_mem_struct(&fdm,
1089 fdm_active->cpu_state_data.destination_address);
1090 fadump_invalidate_dump(&fdm);
1091 } else if (fw_dump.dump_registered) {
1092 /* Un-register Firmware-assisted dump if it was registered. */
1093 fadump_unregister_dump(&fdm);
1094 free_crash_memory_ranges();
1099 * Release the memory that was reserved in early boot to preserve the memory
1100 * contents. The released memory will be available for general use.
1102 static void fadump_release_memory(unsigned long begin, unsigned long end)
1105 unsigned long ra_start, ra_end;
1107 ra_start = fw_dump.reserve_dump_area_start;
1108 ra_end = ra_start + fw_dump.reserve_dump_area_size;
1110 for (addr = begin; addr < end; addr += PAGE_SIZE) {
1112 * exclude the dump reserve area. Will reuse it for next
1113 * fadump registration.
1115 if (addr <= ra_end && ((addr + PAGE_SIZE) > ra_start))
1118 free_reserved_page(pfn_to_page(addr >> PAGE_SHIFT));
1122 static void fadump_invalidate_release_mem(void)
1124 unsigned long reserved_area_start, reserved_area_end;
1125 unsigned long destination_address;
1127 mutex_lock(&fadump_mutex);
1128 if (!fw_dump.dump_active) {
1129 mutex_unlock(&fadump_mutex);
1133 destination_address = fdm_active->cpu_state_data.destination_address;
1135 mutex_unlock(&fadump_mutex);
1138 * Save the current reserved memory bounds we will require them
1139 * later for releasing the memory for general use.
1141 reserved_area_start = fw_dump.reserve_dump_area_start;
1142 reserved_area_end = reserved_area_start +
1143 fw_dump.reserve_dump_area_size;
1145 * Setup reserve_dump_area_start and its size so that we can
1146 * reuse this reserved memory for Re-registration.
1148 fw_dump.reserve_dump_area_start = destination_address;
1149 fw_dump.reserve_dump_area_size = get_fadump_area_size();
1151 fadump_release_memory(reserved_area_start, reserved_area_end);
1152 if (fw_dump.cpu_notes_buf) {
1153 fadump_cpu_notes_buf_free(
1154 (unsigned long)__va(fw_dump.cpu_notes_buf),
1155 fw_dump.cpu_notes_buf_size);
1156 fw_dump.cpu_notes_buf = 0;
1157 fw_dump.cpu_notes_buf_size = 0;
1159 /* Initialize the kernel dump memory structure for FAD registration. */
1160 init_fadump_mem_struct(&fdm, fw_dump.reserve_dump_area_start);
1163 static ssize_t fadump_release_memory_store(struct kobject *kobj,
1164 struct kobj_attribute *attr,
1165 const char *buf, size_t count)
1167 if (!fw_dump.dump_active)
1170 if (buf[0] == '1') {
1172 * Take away the '/proc/vmcore'. We are releasing the dump
1173 * memory, hence it will not be valid anymore.
1176 fadump_invalidate_release_mem();
1183 static ssize_t fadump_enabled_show(struct kobject *kobj,
1184 struct kobj_attribute *attr,
1187 return sprintf(buf, "%d\n", fw_dump.fadump_enabled);
1190 static ssize_t fadump_register_show(struct kobject *kobj,
1191 struct kobj_attribute *attr,
1194 return sprintf(buf, "%d\n", fw_dump.dump_registered);
1197 static ssize_t fadump_register_store(struct kobject *kobj,
1198 struct kobj_attribute *attr,
1199 const char *buf, size_t count)
1203 if (!fw_dump.fadump_enabled || fdm_active)
1206 mutex_lock(&fadump_mutex);
1210 if (fw_dump.dump_registered == 0) {
1214 /* Un-register Firmware-assisted dump */
1215 fadump_unregister_dump(&fdm);
1218 if (fw_dump.dump_registered == 1) {
1222 /* Register Firmware-assisted dump */
1231 mutex_unlock(&fadump_mutex);
1232 return ret < 0 ? ret : count;
1235 static int fadump_region_show(struct seq_file *m, void *private)
1237 const struct fadump_mem_struct *fdm_ptr;
1239 if (!fw_dump.fadump_enabled)
1242 mutex_lock(&fadump_mutex);
1244 fdm_ptr = fdm_active;
1246 mutex_unlock(&fadump_mutex);
1251 "CPU : [%#016llx-%#016llx] %#llx bytes, "
1253 fdm_ptr->cpu_state_data.destination_address,
1254 fdm_ptr->cpu_state_data.destination_address +
1255 fdm_ptr->cpu_state_data.source_len - 1,
1256 fdm_ptr->cpu_state_data.source_len,
1257 fdm_ptr->cpu_state_data.bytes_dumped);
1259 "HPTE: [%#016llx-%#016llx] %#llx bytes, "
1261 fdm_ptr->hpte_region.destination_address,
1262 fdm_ptr->hpte_region.destination_address +
1263 fdm_ptr->hpte_region.source_len - 1,
1264 fdm_ptr->hpte_region.source_len,
1265 fdm_ptr->hpte_region.bytes_dumped);
1267 "DUMP: [%#016llx-%#016llx] %#llx bytes, "
1269 fdm_ptr->rmr_region.destination_address,
1270 fdm_ptr->rmr_region.destination_address +
1271 fdm_ptr->rmr_region.source_len - 1,
1272 fdm_ptr->rmr_region.source_len,
1273 fdm_ptr->rmr_region.bytes_dumped);
1276 (fw_dump.reserve_dump_area_start ==
1277 fdm_ptr->cpu_state_data.destination_address))
1280 /* Dump is active. Show reserved memory region. */
1282 " : [%#016llx-%#016llx] %#llx bytes, "
1284 (unsigned long long)fw_dump.reserve_dump_area_start,
1285 fdm_ptr->cpu_state_data.destination_address - 1,
1286 fdm_ptr->cpu_state_data.destination_address -
1287 fw_dump.reserve_dump_area_start,
1288 fdm_ptr->cpu_state_data.destination_address -
1289 fw_dump.reserve_dump_area_start);
1292 mutex_unlock(&fadump_mutex);
1296 static struct kobj_attribute fadump_release_attr = __ATTR(fadump_release_mem,
1298 fadump_release_memory_store);
1299 static struct kobj_attribute fadump_attr = __ATTR(fadump_enabled,
1300 0444, fadump_enabled_show,
1302 static struct kobj_attribute fadump_register_attr = __ATTR(fadump_registered,
1303 0644, fadump_register_show,
1304 fadump_register_store);
1306 static int fadump_region_open(struct inode *inode, struct file *file)
1308 return single_open(file, fadump_region_show, inode->i_private);
1311 static const struct file_operations fadump_region_fops = {
1312 .open = fadump_region_open,
1314 .llseek = seq_lseek,
1315 .release = single_release,
1318 static void fadump_init_files(void)
1320 struct dentry *debugfs_file;
1323 rc = sysfs_create_file(kernel_kobj, &fadump_attr.attr);
1325 printk(KERN_ERR "fadump: unable to create sysfs file"
1326 " fadump_enabled (%d)\n", rc);
1328 rc = sysfs_create_file(kernel_kobj, &fadump_register_attr.attr);
1330 printk(KERN_ERR "fadump: unable to create sysfs file"
1331 " fadump_registered (%d)\n", rc);
1333 debugfs_file = debugfs_create_file("fadump_region", 0444,
1334 powerpc_debugfs_root, NULL,
1335 &fadump_region_fops);
1337 printk(KERN_ERR "fadump: unable to create debugfs file"
1338 " fadump_region\n");
1340 if (fw_dump.dump_active) {
1341 rc = sysfs_create_file(kernel_kobj, &fadump_release_attr.attr);
1343 printk(KERN_ERR "fadump: unable to create sysfs file"
1344 " fadump_release_mem (%d)\n", rc);
1350 * Prepare for firmware-assisted dump.
1352 int __init setup_fadump(void)
1354 if (!fw_dump.fadump_enabled)
1357 if (!fw_dump.fadump_supported) {
1358 printk(KERN_ERR "Firmware-assisted dump is not supported on"
1359 " this hardware\n");
1363 fadump_show_config();
1365 * If dump data is available then see if it is valid and prepare for
1366 * saving it to the disk.
1368 if (fw_dump.dump_active) {
1370 * if dump process fails then invalidate the registration
1371 * and release memory before proceeding for re-registration.
1373 if (process_fadump(fdm_active) < 0)
1374 fadump_invalidate_release_mem();
1376 /* Initialize the kernel dump memory structure for FAD registration. */
1377 else if (fw_dump.reserve_dump_area_size)
1378 init_fadump_mem_struct(&fdm, fw_dump.reserve_dump_area_start);
1379 fadump_init_files();
1383 subsys_initcall(setup_fadump);