2 * Firmware Assisted dump: A robust mechanism to get reliable kernel crash
3 * dump with assistance from firmware. This approach does not use kexec,
4 * instead firmware assists in booting the kdump kernel while preserving
5 * memory contents. The most of the code implementation has been adapted
6 * from phyp assisted dump implementation written by Linas Vepstas and
9 * This program is free software; you can redistribute it and/or modify
10 * it under the terms of the GNU General Public License as published by
11 * the Free Software Foundation; either version 2 of the License, or
12 * (at your option) any later version.
14 * This program is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
17 * GNU General Public License for more details.
19 * You should have received a copy of the GNU General Public License
20 * along with this program; if not, write to the Free Software
21 * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
23 * Copyright 2011 IBM Corporation
24 * Author: Mahesh Salgaonkar <mahesh@linux.vnet.ibm.com>
28 #define pr_fmt(fmt) "fadump: " fmt
30 #include <linux/string.h>
31 #include <linux/memblock.h>
32 #include <linux/delay.h>
33 #include <linux/seq_file.h>
34 #include <linux/crash_dump.h>
35 #include <linux/kobject.h>
36 #include <linux/sysfs.h>
38 #include <asm/debugfs.h>
42 #include <asm/fadump.h>
43 #include <asm/setup.h>
45 static struct fw_dump fw_dump;
46 static struct fadump_mem_struct fdm;
47 static const struct fadump_mem_struct *fdm_active;
49 static DEFINE_MUTEX(fadump_mutex);
50 struct fad_crash_memory_ranges crash_memory_ranges[INIT_CRASHMEM_RANGES];
53 /* Scan the Firmware Assisted dump configuration details. */
54 int __init early_init_dt_scan_fw_dump(unsigned long node,
55 const char *uname, int depth, void *data)
57 const __be32 *sections;
62 if (depth != 1 || strcmp(uname, "rtas") != 0)
66 * Check if Firmware Assisted dump is supported. if yes, check
67 * if dump has been initiated on last reboot.
69 token = of_get_flat_dt_prop(node, "ibm,configure-kernel-dump", NULL);
73 fw_dump.fadump_supported = 1;
74 fw_dump.ibm_configure_kernel_dump = be32_to_cpu(*token);
77 * The 'ibm,kernel-dump' rtas node is present only if there is
78 * dump data waiting for us.
80 fdm_active = of_get_flat_dt_prop(node, "ibm,kernel-dump", NULL);
82 fw_dump.dump_active = 1;
84 /* Get the sizes required to store dump data for the firmware provided
86 * For each dump section type supported, a 32bit cell which defines
87 * the ID of a supported section followed by two 32 bit cells which
88 * gives teh size of the section in bytes.
90 sections = of_get_flat_dt_prop(node, "ibm,configure-kernel-dump-sizes",
96 num_sections = size / (3 * sizeof(u32));
98 for (i = 0; i < num_sections; i++, sections += 3) {
99 u32 type = (u32)of_read_number(sections, 1);
102 case FADUMP_CPU_STATE_DATA:
103 fw_dump.cpu_state_data_size =
104 of_read_ulong(§ions[1], 2);
106 case FADUMP_HPTE_REGION:
107 fw_dump.hpte_region_size =
108 of_read_ulong(§ions[1], 2);
117 * If fadump is registered, check if the memory provided
118 * falls within boot memory area.
120 int is_fadump_boot_memory_area(u64 addr, ulong size)
122 if (!fw_dump.dump_registered)
125 return (addr + size) > RMA_START && addr <= fw_dump.boot_memory_size;
128 int should_fadump_crash(void)
130 if (!fw_dump.dump_registered || !fw_dump.fadumphdr_addr)
135 int is_fadump_active(void)
137 return fw_dump.dump_active;
141 * Returns 1, if there are no holes in boot memory area,
144 static int is_boot_memory_area_contiguous(void)
146 struct memblock_region *reg;
147 unsigned long tstart, tend;
148 unsigned long start_pfn = PHYS_PFN(RMA_START);
149 unsigned long end_pfn = PHYS_PFN(RMA_START + fw_dump.boot_memory_size);
150 unsigned int ret = 0;
152 for_each_memblock(memory, reg) {
153 tstart = max(start_pfn, memblock_region_memory_base_pfn(reg));
154 tend = min(end_pfn, memblock_region_memory_end_pfn(reg));
156 /* Memory hole from start_pfn to tstart */
157 if (tstart > start_pfn)
160 if (tend == end_pfn) {
165 start_pfn = tend + 1;
172 /* Print firmware assisted dump configurations for debugging purpose. */
173 static void fadump_show_config(void)
175 pr_debug("Support for firmware-assisted dump (fadump): %s\n",
176 (fw_dump.fadump_supported ? "present" : "no support"));
178 if (!fw_dump.fadump_supported)
181 pr_debug("Fadump enabled : %s\n",
182 (fw_dump.fadump_enabled ? "yes" : "no"));
183 pr_debug("Dump Active : %s\n",
184 (fw_dump.dump_active ? "yes" : "no"));
185 pr_debug("Dump section sizes:\n");
186 pr_debug(" CPU state data size: %lx\n", fw_dump.cpu_state_data_size);
187 pr_debug(" HPTE region size : %lx\n", fw_dump.hpte_region_size);
188 pr_debug("Boot memory size : %lx\n", fw_dump.boot_memory_size);
191 static unsigned long init_fadump_mem_struct(struct fadump_mem_struct *fdm,
197 memset(fdm, 0, sizeof(struct fadump_mem_struct));
198 addr = addr & PAGE_MASK;
200 fdm->header.dump_format_version = cpu_to_be32(0x00000001);
201 fdm->header.dump_num_sections = cpu_to_be16(3);
202 fdm->header.dump_status_flag = 0;
203 fdm->header.offset_first_dump_section =
204 cpu_to_be32((u32)offsetof(struct fadump_mem_struct, cpu_state_data));
207 * Fields for disk dump option.
208 * We are not using disk dump option, hence set these fields to 0.
210 fdm->header.dd_block_size = 0;
211 fdm->header.dd_block_offset = 0;
212 fdm->header.dd_num_blocks = 0;
213 fdm->header.dd_offset_disk_path = 0;
215 /* set 0 to disable an automatic dump-reboot. */
216 fdm->header.max_time_auto = 0;
218 /* Kernel dump sections */
219 /* cpu state data section. */
220 fdm->cpu_state_data.request_flag = cpu_to_be32(FADUMP_REQUEST_FLAG);
221 fdm->cpu_state_data.source_data_type = cpu_to_be16(FADUMP_CPU_STATE_DATA);
222 fdm->cpu_state_data.source_address = 0;
223 fdm->cpu_state_data.source_len = cpu_to_be64(fw_dump.cpu_state_data_size);
224 fdm->cpu_state_data.destination_address = cpu_to_be64(addr);
225 addr += fw_dump.cpu_state_data_size;
227 /* hpte region section */
228 fdm->hpte_region.request_flag = cpu_to_be32(FADUMP_REQUEST_FLAG);
229 fdm->hpte_region.source_data_type = cpu_to_be16(FADUMP_HPTE_REGION);
230 fdm->hpte_region.source_address = 0;
231 fdm->hpte_region.source_len = cpu_to_be64(fw_dump.hpte_region_size);
232 fdm->hpte_region.destination_address = cpu_to_be64(addr);
233 addr += fw_dump.hpte_region_size;
235 /* RMA region section */
236 fdm->rmr_region.request_flag = cpu_to_be32(FADUMP_REQUEST_FLAG);
237 fdm->rmr_region.source_data_type = cpu_to_be16(FADUMP_REAL_MODE_REGION);
238 fdm->rmr_region.source_address = cpu_to_be64(RMA_START);
239 fdm->rmr_region.source_len = cpu_to_be64(fw_dump.boot_memory_size);
240 fdm->rmr_region.destination_address = cpu_to_be64(addr);
241 addr += fw_dump.boot_memory_size;
247 * fadump_calculate_reserve_size(): reserve variable boot area 5% of System RAM
249 * Function to find the largest memory size we need to reserve during early
250 * boot process. This will be the size of the memory that is required for a
251 * kernel to boot successfully.
253 * This function has been taken from phyp-assisted dump feature implementation.
255 * returns larger of 256MB or 5% rounded down to multiples of 256MB.
257 * TODO: Come up with better approach to find out more accurate memory size
258 * that is required for a kernel to boot successfully.
261 static inline unsigned long fadump_calculate_reserve_size(void)
264 unsigned long long base, size;
266 if (fw_dump.reserve_bootvar)
267 pr_warn("'fadump_reserve_mem=' parameter is deprecated in favor of 'crashkernel=' parameter.\n");
270 * Check if the size is specified through crashkernel= cmdline
271 * option. If yes, then use that but ignore base as fadump reserves
272 * memory at a predefined offset.
274 ret = parse_crashkernel(boot_command_line, memblock_phys_mem_size(),
276 if (ret == 0 && size > 0) {
277 unsigned long max_size;
279 if (fw_dump.reserve_bootvar)
280 pr_info("Using 'crashkernel=' parameter for memory reservation.\n");
282 fw_dump.reserve_bootvar = (unsigned long)size;
285 * Adjust if the boot memory size specified is above
288 max_size = memblock_phys_mem_size() / MAX_BOOT_MEM_RATIO;
289 if (fw_dump.reserve_bootvar > max_size) {
290 fw_dump.reserve_bootvar = max_size;
291 pr_info("Adjusted boot memory size to %luMB\n",
292 (fw_dump.reserve_bootvar >> 20));
295 return fw_dump.reserve_bootvar;
296 } else if (fw_dump.reserve_bootvar) {
298 * 'fadump_reserve_mem=' is being used to reserve memory
299 * for firmware-assisted dump.
301 return fw_dump.reserve_bootvar;
304 /* divide by 20 to get 5% of value */
305 size = memblock_phys_mem_size() / 20;
307 /* round it down in multiples of 256 */
308 size = size & ~0x0FFFFFFFUL;
310 /* Truncate to memory_limit. We don't want to over reserve the memory.*/
311 if (memory_limit && size > memory_limit)
314 return (size > MIN_BOOT_MEM ? size : MIN_BOOT_MEM);
318 * Calculate the total memory size required to be reserved for
319 * firmware-assisted dump registration.
321 static unsigned long get_fadump_area_size(void)
323 unsigned long size = 0;
325 size += fw_dump.cpu_state_data_size;
326 size += fw_dump.hpte_region_size;
327 size += fw_dump.boot_memory_size;
328 size += sizeof(struct fadump_crash_info_header);
329 size += sizeof(struct elfhdr); /* ELF core header.*/
330 size += sizeof(struct elf_phdr); /* place holder for cpu notes */
331 /* Program headers for crash memory regions. */
332 size += sizeof(struct elf_phdr) * (memblock_num_regions(memory) + 2);
334 size = PAGE_ALIGN(size);
338 static void __init fadump_reserve_crash_area(unsigned long base,
341 struct memblock_region *reg;
342 unsigned long mstart, mend, msize;
344 for_each_memblock(memory, reg) {
345 mstart = max_t(unsigned long, base, reg->base);
346 mend = reg->base + reg->size;
347 mend = min(base + size, mend);
350 msize = mend - mstart;
351 memblock_reserve(mstart, msize);
352 pr_info("Reserved %ldMB of memory at %#016lx for saving crash dump\n",
353 (msize >> 20), mstart);
358 int __init fadump_reserve_mem(void)
360 unsigned long base, size, memory_boundary;
362 if (!fw_dump.fadump_enabled)
365 if (!fw_dump.fadump_supported) {
366 printk(KERN_INFO "Firmware-assisted dump is not supported on"
368 fw_dump.fadump_enabled = 0;
372 * Initialize boot memory size
373 * If dump is active then we have already calculated the size during
377 fw_dump.boot_memory_size = be64_to_cpu(fdm_active->rmr_region.source_len);
379 fw_dump.boot_memory_size = fadump_calculate_reserve_size();
382 * Calculate the memory boundary.
383 * If memory_limit is less than actual memory boundary then reserve
384 * the memory for fadump beyond the memory_limit and adjust the
385 * memory_limit accordingly, so that the running kernel can run with
386 * specified memory_limit.
388 if (memory_limit && memory_limit < memblock_end_of_DRAM()) {
389 size = get_fadump_area_size();
390 if ((memory_limit + size) < memblock_end_of_DRAM())
391 memory_limit += size;
393 memory_limit = memblock_end_of_DRAM();
394 printk(KERN_INFO "Adjusted memory_limit for firmware-assisted"
395 " dump, now %#016llx\n", memory_limit);
398 memory_boundary = memory_limit;
400 memory_boundary = memblock_end_of_DRAM();
402 if (fw_dump.dump_active) {
403 pr_info("Firmware-assisted dump is active.\n");
405 #ifdef CONFIG_HUGETLB_PAGE
407 * FADump capture kernel doesn't care much about hugepages.
408 * In fact, handling hugepages in capture kernel is asking for
409 * trouble. So, disable HugeTLB support when fadump is active.
411 hugetlb_disabled = true;
414 * If last boot has crashed then reserve all the memory
415 * above boot_memory_size so that we don't touch it until
416 * dump is written to disk by userspace tool. This memory
417 * will be released for general use once the dump is saved.
419 base = fw_dump.boot_memory_size;
420 size = memory_boundary - base;
421 fadump_reserve_crash_area(base, size);
423 fw_dump.fadumphdr_addr =
424 be64_to_cpu(fdm_active->rmr_region.destination_address) +
425 be64_to_cpu(fdm_active->rmr_region.source_len);
426 pr_debug("fadumphdr_addr = %p\n",
427 (void *) fw_dump.fadumphdr_addr);
429 size = get_fadump_area_size();
432 * Reserve memory at an offset closer to bottom of the RAM to
433 * minimize the impact of memory hot-remove operation. We can't
434 * use memblock_find_in_range() here since it doesn't allocate
435 * from bottom to top.
437 for (base = fw_dump.boot_memory_size;
438 base <= (memory_boundary - size);
440 if (memblock_is_region_memory(base, size) &&
441 !memblock_is_region_reserved(base, size))
444 if ((base > (memory_boundary - size)) ||
445 memblock_reserve(base, size)) {
446 pr_err("Failed to reserve memory\n");
450 pr_info("Reserved %ldMB of memory at %ldMB for firmware-"
451 "assisted dump (System RAM: %ldMB)\n",
452 (unsigned long)(size >> 20),
453 (unsigned long)(base >> 20),
454 (unsigned long)(memblock_phys_mem_size() >> 20));
457 fw_dump.reserve_dump_area_start = base;
458 fw_dump.reserve_dump_area_size = size;
462 unsigned long __init arch_reserved_kernel_pages(void)
464 return memblock_reserved_size() / PAGE_SIZE;
467 /* Look for fadump= cmdline option. */
468 static int __init early_fadump_param(char *p)
473 if (strncmp(p, "on", 2) == 0)
474 fw_dump.fadump_enabled = 1;
475 else if (strncmp(p, "off", 3) == 0)
476 fw_dump.fadump_enabled = 0;
480 early_param("fadump", early_fadump_param);
483 * Look for fadump_reserve_mem= cmdline option
484 * TODO: Remove references to 'fadump_reserve_mem=' parameter,
485 * the sooner 'crashkernel=' parameter is accustomed to.
487 static int __init early_fadump_reserve_mem(char *p)
490 fw_dump.reserve_bootvar = memparse(p, &p);
493 early_param("fadump_reserve_mem", early_fadump_reserve_mem);
495 static int register_fw_dump(struct fadump_mem_struct *fdm)
498 unsigned int wait_time;
500 pr_debug("Registering for firmware-assisted kernel dump...\n");
502 /* TODO: Add upper time limit for the delay */
504 rc = rtas_call(fw_dump.ibm_configure_kernel_dump, 3, 1, NULL,
505 FADUMP_REGISTER, fdm,
506 sizeof(struct fadump_mem_struct));
508 wait_time = rtas_busy_delay_time(rc);
517 pr_err("Failed to register. Unknown Error(%d).\n", rc);
520 printk(KERN_ERR "Failed to register firmware-assisted kernel"
521 " dump. Hardware Error(%d).\n", rc);
524 if (!is_boot_memory_area_contiguous())
525 pr_err("Can't have holes in boot memory area while "
526 "registering fadump\n");
528 printk(KERN_ERR "Failed to register firmware-assisted kernel"
529 " dump. Parameter Error(%d).\n", rc);
533 printk(KERN_ERR "firmware-assisted kernel dump is already "
535 fw_dump.dump_registered = 1;
539 printk(KERN_INFO "firmware-assisted kernel dump registration"
541 fw_dump.dump_registered = 1;
548 void crash_fadump(struct pt_regs *regs, const char *str)
550 struct fadump_crash_info_header *fdh = NULL;
551 int old_cpu, this_cpu;
553 if (!should_fadump_crash())
557 * old_cpu == -1 means this is the first CPU which has come here,
558 * go ahead and trigger fadump.
560 * old_cpu != -1 means some other CPU has already on it's way
561 * to trigger fadump, just keep looping here.
563 this_cpu = smp_processor_id();
564 old_cpu = cmpxchg(&crashing_cpu, -1, this_cpu);
568 * We can't loop here indefinitely. Wait as long as fadump
569 * is in force. If we race with fadump un-registration this
570 * loop will break and then we go down to normal panic path
571 * and reboot. If fadump is in force the first crashing
572 * cpu will definitely trigger fadump.
574 while (fw_dump.dump_registered)
579 fdh = __va(fw_dump.fadumphdr_addr);
580 fdh->crashing_cpu = crashing_cpu;
581 crash_save_vmcoreinfo();
586 ppc_save_regs(&fdh->regs);
588 fdh->online_mask = *cpu_online_mask;
590 /* Call ibm,os-term rtas call to trigger firmware assisted dump */
591 rtas_os_term((char *)str);
594 #define GPR_MASK 0xffffff0000000000
595 static inline int fadump_gpr_index(u64 id)
600 if ((id & GPR_MASK) == REG_ID("GPR")) {
601 /* get the digits at the end */
606 str[0] = (id >> 8) & 0xff;
607 sscanf(str, "%d", &i);
614 static inline void fadump_set_regval(struct pt_regs *regs, u64 reg_id,
619 i = fadump_gpr_index(reg_id);
621 regs->gpr[i] = (unsigned long)reg_val;
622 else if (reg_id == REG_ID("NIA"))
623 regs->nip = (unsigned long)reg_val;
624 else if (reg_id == REG_ID("MSR"))
625 regs->msr = (unsigned long)reg_val;
626 else if (reg_id == REG_ID("CTR"))
627 regs->ctr = (unsigned long)reg_val;
628 else if (reg_id == REG_ID("LR"))
629 regs->link = (unsigned long)reg_val;
630 else if (reg_id == REG_ID("XER"))
631 regs->xer = (unsigned long)reg_val;
632 else if (reg_id == REG_ID("CR"))
633 regs->ccr = (unsigned long)reg_val;
634 else if (reg_id == REG_ID("DAR"))
635 regs->dar = (unsigned long)reg_val;
636 else if (reg_id == REG_ID("DSISR"))
637 regs->dsisr = (unsigned long)reg_val;
640 static struct fadump_reg_entry*
641 fadump_read_registers(struct fadump_reg_entry *reg_entry, struct pt_regs *regs)
643 memset(regs, 0, sizeof(struct pt_regs));
645 while (be64_to_cpu(reg_entry->reg_id) != REG_ID("CPUEND")) {
646 fadump_set_regval(regs, be64_to_cpu(reg_entry->reg_id),
647 be64_to_cpu(reg_entry->reg_value));
654 static u32 *fadump_regs_to_elf_notes(u32 *buf, struct pt_regs *regs)
656 struct elf_prstatus prstatus;
658 memset(&prstatus, 0, sizeof(prstatus));
660 * FIXME: How do i get PID? Do I really need it?
661 * prstatus.pr_pid = ????
663 elf_core_copy_kernel_regs(&prstatus.pr_reg, regs);
664 buf = append_elf_note(buf, CRASH_CORE_NOTE_NAME, NT_PRSTATUS,
665 &prstatus, sizeof(prstatus));
669 static void fadump_update_elfcore_header(char *bufp)
672 struct elf_phdr *phdr;
674 elf = (struct elfhdr *)bufp;
675 bufp += sizeof(struct elfhdr);
677 /* First note is a place holder for cpu notes info. */
678 phdr = (struct elf_phdr *)bufp;
680 if (phdr->p_type == PT_NOTE) {
681 phdr->p_paddr = fw_dump.cpu_notes_buf;
682 phdr->p_offset = phdr->p_paddr;
683 phdr->p_filesz = fw_dump.cpu_notes_buf_size;
684 phdr->p_memsz = fw_dump.cpu_notes_buf_size;
689 static void *fadump_cpu_notes_buf_alloc(unsigned long size)
693 unsigned long order, count, i;
695 order = get_order(size);
696 vaddr = (void *)__get_free_pages(GFP_KERNEL|__GFP_ZERO, order);
701 page = virt_to_page(vaddr);
702 for (i = 0; i < count; i++)
703 SetPageReserved(page + i);
707 static void fadump_cpu_notes_buf_free(unsigned long vaddr, unsigned long size)
710 unsigned long order, count, i;
712 order = get_order(size);
714 page = virt_to_page(vaddr);
715 for (i = 0; i < count; i++)
716 ClearPageReserved(page + i);
717 __free_pages(page, order);
721 * Read CPU state dump data and convert it into ELF notes.
722 * The CPU dump starts with magic number "REGSAVE". NumCpusOffset should be
723 * used to access the data to allow for additional fields to be added without
724 * affecting compatibility. Each list of registers for a CPU starts with
725 * "CPUSTRT" and ends with "CPUEND". Each register entry is of 16 bytes,
726 * 8 Byte ASCII identifier and 8 Byte register value. The register entry
727 * with identifier "CPUSTRT" and "CPUEND" contains 4 byte cpu id as part
728 * of register value. For more details refer to PAPR document.
730 * Only for the crashing cpu we ignore the CPU dump data and get exact
731 * state from fadump crash info structure populated by first kernel at the
734 static int __init fadump_build_cpu_notes(const struct fadump_mem_struct *fdm)
736 struct fadump_reg_save_area_header *reg_header;
737 struct fadump_reg_entry *reg_entry;
738 struct fadump_crash_info_header *fdh = NULL;
741 u32 num_cpus, *note_buf;
743 int i, rc = 0, cpu = 0;
745 if (!fdm->cpu_state_data.bytes_dumped)
748 addr = be64_to_cpu(fdm->cpu_state_data.destination_address);
752 if (be64_to_cpu(reg_header->magic_number) != REGSAVE_AREA_MAGIC) {
753 printk(KERN_ERR "Unable to read register save area.\n");
756 pr_debug("--------CPU State Data------------\n");
757 pr_debug("Magic Number: %llx\n", be64_to_cpu(reg_header->magic_number));
758 pr_debug("NumCpuOffset: %x\n", be32_to_cpu(reg_header->num_cpu_offset));
760 vaddr += be32_to_cpu(reg_header->num_cpu_offset);
761 num_cpus = be32_to_cpu(*((__be32 *)(vaddr)));
762 pr_debug("NumCpus : %u\n", num_cpus);
763 vaddr += sizeof(u32);
764 reg_entry = (struct fadump_reg_entry *)vaddr;
766 /* Allocate buffer to hold cpu crash notes. */
767 fw_dump.cpu_notes_buf_size = num_cpus * sizeof(note_buf_t);
768 fw_dump.cpu_notes_buf_size = PAGE_ALIGN(fw_dump.cpu_notes_buf_size);
769 note_buf = fadump_cpu_notes_buf_alloc(fw_dump.cpu_notes_buf_size);
771 printk(KERN_ERR "Failed to allocate 0x%lx bytes for "
772 "cpu notes buffer\n", fw_dump.cpu_notes_buf_size);
775 fw_dump.cpu_notes_buf = __pa(note_buf);
777 pr_debug("Allocated buffer for cpu notes of size %ld at %p\n",
778 (num_cpus * sizeof(note_buf_t)), note_buf);
780 if (fw_dump.fadumphdr_addr)
781 fdh = __va(fw_dump.fadumphdr_addr);
783 for (i = 0; i < num_cpus; i++) {
784 if (be64_to_cpu(reg_entry->reg_id) != REG_ID("CPUSTRT")) {
785 printk(KERN_ERR "Unable to read CPU state data\n");
789 /* Lower 4 bytes of reg_value contains logical cpu id */
790 cpu = be64_to_cpu(reg_entry->reg_value) & FADUMP_CPU_ID_MASK;
791 if (fdh && !cpumask_test_cpu(cpu, &fdh->online_mask)) {
792 SKIP_TO_NEXT_CPU(reg_entry);
795 pr_debug("Reading register data for cpu %d...\n", cpu);
796 if (fdh && fdh->crashing_cpu == cpu) {
798 note_buf = fadump_regs_to_elf_notes(note_buf, ®s);
799 SKIP_TO_NEXT_CPU(reg_entry);
802 reg_entry = fadump_read_registers(reg_entry, ®s);
803 note_buf = fadump_regs_to_elf_notes(note_buf, ®s);
806 final_note(note_buf);
809 pr_debug("Updating elfcore header (%llx) with cpu notes\n",
810 fdh->elfcorehdr_addr);
811 fadump_update_elfcore_header((char *)__va(fdh->elfcorehdr_addr));
816 fadump_cpu_notes_buf_free((unsigned long)__va(fw_dump.cpu_notes_buf),
817 fw_dump.cpu_notes_buf_size);
818 fw_dump.cpu_notes_buf = 0;
819 fw_dump.cpu_notes_buf_size = 0;
825 * Validate and process the dump data stored by firmware before exporting
826 * it through '/proc/vmcore'.
828 static int __init process_fadump(const struct fadump_mem_struct *fdm_active)
830 struct fadump_crash_info_header *fdh;
833 if (!fdm_active || !fw_dump.fadumphdr_addr)
836 /* Check if the dump data is valid. */
837 if ((be16_to_cpu(fdm_active->header.dump_status_flag) == FADUMP_ERROR_FLAG) ||
838 (fdm_active->cpu_state_data.error_flags != 0) ||
839 (fdm_active->rmr_region.error_flags != 0)) {
840 printk(KERN_ERR "Dump taken by platform is not valid\n");
843 if ((fdm_active->rmr_region.bytes_dumped !=
844 fdm_active->rmr_region.source_len) ||
845 !fdm_active->cpu_state_data.bytes_dumped) {
846 printk(KERN_ERR "Dump taken by platform is incomplete\n");
850 /* Validate the fadump crash info header */
851 fdh = __va(fw_dump.fadumphdr_addr);
852 if (fdh->magic_number != FADUMP_CRASH_INFO_MAGIC) {
853 printk(KERN_ERR "Crash info header is not valid.\n");
857 rc = fadump_build_cpu_notes(fdm_active);
862 * We are done validating dump info and elfcore header is now ready
863 * to be exported. set elfcorehdr_addr so that vmcore module will
864 * export the elfcore header through '/proc/vmcore'.
866 elfcorehdr_addr = fdh->elfcorehdr_addr;
871 static inline void fadump_add_crash_memory(unsigned long long base,
872 unsigned long long end)
877 pr_debug("crash_memory_range[%d] [%#016llx-%#016llx], %#llx bytes\n",
878 crash_mem_ranges, base, end - 1, (end - base));
879 crash_memory_ranges[crash_mem_ranges].base = base;
880 crash_memory_ranges[crash_mem_ranges].size = end - base;
884 static void fadump_exclude_reserved_area(unsigned long long start,
885 unsigned long long end)
887 unsigned long long ra_start, ra_end;
889 ra_start = fw_dump.reserve_dump_area_start;
890 ra_end = ra_start + fw_dump.reserve_dump_area_size;
892 if ((ra_start < end) && (ra_end > start)) {
893 if ((start < ra_start) && (end > ra_end)) {
894 fadump_add_crash_memory(start, ra_start);
895 fadump_add_crash_memory(ra_end, end);
896 } else if (start < ra_start) {
897 fadump_add_crash_memory(start, ra_start);
898 } else if (ra_end < end) {
899 fadump_add_crash_memory(ra_end, end);
902 fadump_add_crash_memory(start, end);
905 static int fadump_init_elfcore_header(char *bufp)
909 elf = (struct elfhdr *) bufp;
910 bufp += sizeof(struct elfhdr);
911 memcpy(elf->e_ident, ELFMAG, SELFMAG);
912 elf->e_ident[EI_CLASS] = ELF_CLASS;
913 elf->e_ident[EI_DATA] = ELF_DATA;
914 elf->e_ident[EI_VERSION] = EV_CURRENT;
915 elf->e_ident[EI_OSABI] = ELF_OSABI;
916 memset(elf->e_ident+EI_PAD, 0, EI_NIDENT-EI_PAD);
917 elf->e_type = ET_CORE;
918 elf->e_machine = ELF_ARCH;
919 elf->e_version = EV_CURRENT;
921 elf->e_phoff = sizeof(struct elfhdr);
923 #if defined(_CALL_ELF)
924 elf->e_flags = _CALL_ELF;
928 elf->e_ehsize = sizeof(struct elfhdr);
929 elf->e_phentsize = sizeof(struct elf_phdr);
931 elf->e_shentsize = 0;
939 * Traverse through memblock structure and setup crash memory ranges. These
940 * ranges will be used create PT_LOAD program headers in elfcore header.
942 static void fadump_setup_crash_memory_ranges(void)
944 struct memblock_region *reg;
945 unsigned long long start, end;
947 pr_debug("Setup crash memory ranges.\n");
948 crash_mem_ranges = 0;
950 * add the first memory chunk (RMA_START through boot_memory_size) as
951 * a separate memory chunk. The reason is, at the time crash firmware
952 * will move the content of this memory chunk to different location
953 * specified during fadump registration. We need to create a separate
954 * program header for this chunk with the correct offset.
956 fadump_add_crash_memory(RMA_START, fw_dump.boot_memory_size);
958 for_each_memblock(memory, reg) {
959 start = (unsigned long long)reg->base;
960 end = start + (unsigned long long)reg->size;
963 * skip the first memory chunk that is already added (RMA_START
964 * through boot_memory_size). This logic needs a relook if and
965 * when RMA_START changes to a non-zero value.
967 BUILD_BUG_ON(RMA_START != 0);
968 if (start < fw_dump.boot_memory_size) {
969 if (end > fw_dump.boot_memory_size)
970 start = fw_dump.boot_memory_size;
975 /* add this range excluding the reserved dump area. */
976 fadump_exclude_reserved_area(start, end);
981 * If the given physical address falls within the boot memory region then
982 * return the relocated address that points to the dump region reserved
983 * for saving initial boot memory contents.
985 static inline unsigned long fadump_relocate(unsigned long paddr)
987 if (paddr > RMA_START && paddr < fw_dump.boot_memory_size)
988 return be64_to_cpu(fdm.rmr_region.destination_address) + paddr;
993 static int fadump_create_elfcore_headers(char *bufp)
996 struct elf_phdr *phdr;
999 fadump_init_elfcore_header(bufp);
1000 elf = (struct elfhdr *)bufp;
1001 bufp += sizeof(struct elfhdr);
1004 * setup ELF PT_NOTE, place holder for cpu notes info. The notes info
1005 * will be populated during second kernel boot after crash. Hence
1006 * this PT_NOTE will always be the first elf note.
1008 * NOTE: Any new ELF note addition should be placed after this note.
1010 phdr = (struct elf_phdr *)bufp;
1011 bufp += sizeof(struct elf_phdr);
1012 phdr->p_type = PT_NOTE;
1024 /* setup ELF PT_NOTE for vmcoreinfo */
1025 phdr = (struct elf_phdr *)bufp;
1026 bufp += sizeof(struct elf_phdr);
1027 phdr->p_type = PT_NOTE;
1032 phdr->p_paddr = fadump_relocate(paddr_vmcoreinfo_note());
1033 phdr->p_offset = phdr->p_paddr;
1034 phdr->p_memsz = phdr->p_filesz = VMCOREINFO_NOTE_SIZE;
1036 /* Increment number of program headers. */
1039 /* setup PT_LOAD sections. */
1041 for (i = 0; i < crash_mem_ranges; i++) {
1042 unsigned long long mbase, msize;
1043 mbase = crash_memory_ranges[i].base;
1044 msize = crash_memory_ranges[i].size;
1049 phdr = (struct elf_phdr *)bufp;
1050 bufp += sizeof(struct elf_phdr);
1051 phdr->p_type = PT_LOAD;
1052 phdr->p_flags = PF_R|PF_W|PF_X;
1053 phdr->p_offset = mbase;
1055 if (mbase == RMA_START) {
1057 * The entire RMA region will be moved by firmware
1058 * to the specified destination_address. Hence set
1059 * the correct offset.
1061 phdr->p_offset = be64_to_cpu(fdm.rmr_region.destination_address);
1064 phdr->p_paddr = mbase;
1065 phdr->p_vaddr = (unsigned long)__va(mbase);
1066 phdr->p_filesz = msize;
1067 phdr->p_memsz = msize;
1070 /* Increment number of program headers. */
1076 static unsigned long init_fadump_header(unsigned long addr)
1078 struct fadump_crash_info_header *fdh;
1083 fw_dump.fadumphdr_addr = addr;
1085 addr += sizeof(struct fadump_crash_info_header);
1087 memset(fdh, 0, sizeof(struct fadump_crash_info_header));
1088 fdh->magic_number = FADUMP_CRASH_INFO_MAGIC;
1089 fdh->elfcorehdr_addr = addr;
1090 /* We will set the crashing cpu id in crash_fadump() during crash. */
1091 fdh->crashing_cpu = CPU_UNKNOWN;
1096 static int register_fadump(void)
1102 * If no memory is reserved then we can not register for firmware-
1105 if (!fw_dump.reserve_dump_area_size)
1108 fadump_setup_crash_memory_ranges();
1110 addr = be64_to_cpu(fdm.rmr_region.destination_address) + be64_to_cpu(fdm.rmr_region.source_len);
1111 /* Initialize fadump crash info header. */
1112 addr = init_fadump_header(addr);
1115 pr_debug("Creating ELF core headers at %#016lx\n", addr);
1116 fadump_create_elfcore_headers(vaddr);
1118 /* register the future kernel dump with firmware. */
1119 return register_fw_dump(&fdm);
1122 static int fadump_unregister_dump(struct fadump_mem_struct *fdm)
1125 unsigned int wait_time;
1127 pr_debug("Un-register firmware-assisted dump\n");
1129 /* TODO: Add upper time limit for the delay */
1131 rc = rtas_call(fw_dump.ibm_configure_kernel_dump, 3, 1, NULL,
1132 FADUMP_UNREGISTER, fdm,
1133 sizeof(struct fadump_mem_struct));
1135 wait_time = rtas_busy_delay_time(rc);
1138 } while (wait_time);
1141 printk(KERN_ERR "Failed to un-register firmware-assisted dump."
1142 " unexpected error(%d).\n", rc);
1145 fw_dump.dump_registered = 0;
1149 static int fadump_invalidate_dump(struct fadump_mem_struct *fdm)
1152 unsigned int wait_time;
1154 pr_debug("Invalidating firmware-assisted dump registration\n");
1156 /* TODO: Add upper time limit for the delay */
1158 rc = rtas_call(fw_dump.ibm_configure_kernel_dump, 3, 1, NULL,
1159 FADUMP_INVALIDATE, fdm,
1160 sizeof(struct fadump_mem_struct));
1162 wait_time = rtas_busy_delay_time(rc);
1165 } while (wait_time);
1168 pr_err("Failed to invalidate firmware-assisted dump registration. Unexpected error (%d).\n", rc);
1171 fw_dump.dump_active = 0;
1176 void fadump_cleanup(void)
1178 /* Invalidate the registration only if dump is active. */
1179 if (fw_dump.dump_active) {
1180 init_fadump_mem_struct(&fdm,
1181 be64_to_cpu(fdm_active->cpu_state_data.destination_address));
1182 fadump_invalidate_dump(&fdm);
1183 } else if (fw_dump.dump_registered) {
1184 /* Un-register Firmware-assisted dump if it was registered. */
1185 fadump_unregister_dump(&fdm);
1189 static void fadump_free_reserved_memory(unsigned long start_pfn,
1190 unsigned long end_pfn)
1193 unsigned long time_limit = jiffies + HZ;
1195 pr_info("freeing reserved memory (0x%llx - 0x%llx)\n",
1196 PFN_PHYS(start_pfn), PFN_PHYS(end_pfn));
1198 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
1199 free_reserved_page(pfn_to_page(pfn));
1201 if (time_after(jiffies, time_limit)) {
1203 time_limit = jiffies + HZ;
1209 * Skip memory holes and free memory that was actually reserved.
1211 static void fadump_release_reserved_area(unsigned long start, unsigned long end)
1213 struct memblock_region *reg;
1214 unsigned long tstart, tend;
1215 unsigned long start_pfn = PHYS_PFN(start);
1216 unsigned long end_pfn = PHYS_PFN(end);
1218 for_each_memblock(memory, reg) {
1219 tstart = max(start_pfn, memblock_region_memory_base_pfn(reg));
1220 tend = min(end_pfn, memblock_region_memory_end_pfn(reg));
1221 if (tstart < tend) {
1222 fadump_free_reserved_memory(tstart, tend);
1224 if (tend == end_pfn)
1227 start_pfn = tend + 1;
1233 * Release the memory that was reserved in early boot to preserve the memory
1234 * contents. The released memory will be available for general use.
1236 static void fadump_release_memory(unsigned long begin, unsigned long end)
1238 unsigned long ra_start, ra_end;
1240 ra_start = fw_dump.reserve_dump_area_start;
1241 ra_end = ra_start + fw_dump.reserve_dump_area_size;
1244 * exclude the dump reserve area. Will reuse it for next
1245 * fadump registration.
1247 if (begin < ra_end && end > ra_start) {
1248 if (begin < ra_start)
1249 fadump_release_reserved_area(begin, ra_start);
1251 fadump_release_reserved_area(ra_end, end);
1253 fadump_release_reserved_area(begin, end);
1256 static void fadump_invalidate_release_mem(void)
1258 unsigned long reserved_area_start, reserved_area_end;
1259 unsigned long destination_address;
1261 mutex_lock(&fadump_mutex);
1262 if (!fw_dump.dump_active) {
1263 mutex_unlock(&fadump_mutex);
1267 destination_address = be64_to_cpu(fdm_active->cpu_state_data.destination_address);
1269 mutex_unlock(&fadump_mutex);
1272 * Save the current reserved memory bounds we will require them
1273 * later for releasing the memory for general use.
1275 reserved_area_start = fw_dump.reserve_dump_area_start;
1276 reserved_area_end = reserved_area_start +
1277 fw_dump.reserve_dump_area_size;
1279 * Setup reserve_dump_area_start and its size so that we can
1280 * reuse this reserved memory for Re-registration.
1282 fw_dump.reserve_dump_area_start = destination_address;
1283 fw_dump.reserve_dump_area_size = get_fadump_area_size();
1285 fadump_release_memory(reserved_area_start, reserved_area_end);
1286 if (fw_dump.cpu_notes_buf) {
1287 fadump_cpu_notes_buf_free(
1288 (unsigned long)__va(fw_dump.cpu_notes_buf),
1289 fw_dump.cpu_notes_buf_size);
1290 fw_dump.cpu_notes_buf = 0;
1291 fw_dump.cpu_notes_buf_size = 0;
1293 /* Initialize the kernel dump memory structure for FAD registration. */
1294 init_fadump_mem_struct(&fdm, fw_dump.reserve_dump_area_start);
1297 static ssize_t fadump_release_memory_store(struct kobject *kobj,
1298 struct kobj_attribute *attr,
1299 const char *buf, size_t count)
1303 if (!fw_dump.dump_active)
1306 if (kstrtoint(buf, 0, &input))
1311 * Take away the '/proc/vmcore'. We are releasing the dump
1312 * memory, hence it will not be valid anymore.
1314 #ifdef CONFIG_PROC_VMCORE
1317 fadump_invalidate_release_mem();
1324 static ssize_t fadump_enabled_show(struct kobject *kobj,
1325 struct kobj_attribute *attr,
1328 return sprintf(buf, "%d\n", fw_dump.fadump_enabled);
1331 static ssize_t fadump_register_show(struct kobject *kobj,
1332 struct kobj_attribute *attr,
1335 return sprintf(buf, "%d\n", fw_dump.dump_registered);
1338 static ssize_t fadump_register_store(struct kobject *kobj,
1339 struct kobj_attribute *attr,
1340 const char *buf, size_t count)
1345 if (!fw_dump.fadump_enabled || fdm_active)
1348 if (kstrtoint(buf, 0, &input))
1351 mutex_lock(&fadump_mutex);
1355 if (fw_dump.dump_registered == 0) {
1358 /* Un-register Firmware-assisted dump */
1359 fadump_unregister_dump(&fdm);
1362 if (fw_dump.dump_registered == 1) {
1366 /* Register Firmware-assisted dump */
1367 ret = register_fadump();
1375 mutex_unlock(&fadump_mutex);
1376 return ret < 0 ? ret : count;
1379 static int fadump_region_show(struct seq_file *m, void *private)
1381 const struct fadump_mem_struct *fdm_ptr;
1383 if (!fw_dump.fadump_enabled)
1386 mutex_lock(&fadump_mutex);
1388 fdm_ptr = fdm_active;
1390 mutex_unlock(&fadump_mutex);
1395 "CPU : [%#016llx-%#016llx] %#llx bytes, "
1397 be64_to_cpu(fdm_ptr->cpu_state_data.destination_address),
1398 be64_to_cpu(fdm_ptr->cpu_state_data.destination_address) +
1399 be64_to_cpu(fdm_ptr->cpu_state_data.source_len) - 1,
1400 be64_to_cpu(fdm_ptr->cpu_state_data.source_len),
1401 be64_to_cpu(fdm_ptr->cpu_state_data.bytes_dumped));
1403 "HPTE: [%#016llx-%#016llx] %#llx bytes, "
1405 be64_to_cpu(fdm_ptr->hpte_region.destination_address),
1406 be64_to_cpu(fdm_ptr->hpte_region.destination_address) +
1407 be64_to_cpu(fdm_ptr->hpte_region.source_len) - 1,
1408 be64_to_cpu(fdm_ptr->hpte_region.source_len),
1409 be64_to_cpu(fdm_ptr->hpte_region.bytes_dumped));
1411 "DUMP: [%#016llx-%#016llx] %#llx bytes, "
1413 be64_to_cpu(fdm_ptr->rmr_region.destination_address),
1414 be64_to_cpu(fdm_ptr->rmr_region.destination_address) +
1415 be64_to_cpu(fdm_ptr->rmr_region.source_len) - 1,
1416 be64_to_cpu(fdm_ptr->rmr_region.source_len),
1417 be64_to_cpu(fdm_ptr->rmr_region.bytes_dumped));
1420 (fw_dump.reserve_dump_area_start ==
1421 be64_to_cpu(fdm_ptr->cpu_state_data.destination_address)))
1424 /* Dump is active. Show reserved memory region. */
1426 " : [%#016llx-%#016llx] %#llx bytes, "
1428 (unsigned long long)fw_dump.reserve_dump_area_start,
1429 be64_to_cpu(fdm_ptr->cpu_state_data.destination_address) - 1,
1430 be64_to_cpu(fdm_ptr->cpu_state_data.destination_address) -
1431 fw_dump.reserve_dump_area_start,
1432 be64_to_cpu(fdm_ptr->cpu_state_data.destination_address) -
1433 fw_dump.reserve_dump_area_start);
1436 mutex_unlock(&fadump_mutex);
1440 static struct kobj_attribute fadump_release_attr = __ATTR(fadump_release_mem,
1442 fadump_release_memory_store);
1443 static struct kobj_attribute fadump_attr = __ATTR(fadump_enabled,
1444 0444, fadump_enabled_show,
1446 static struct kobj_attribute fadump_register_attr = __ATTR(fadump_registered,
1447 0644, fadump_register_show,
1448 fadump_register_store);
1450 static int fadump_region_open(struct inode *inode, struct file *file)
1452 return single_open(file, fadump_region_show, inode->i_private);
1455 static const struct file_operations fadump_region_fops = {
1456 .open = fadump_region_open,
1458 .llseek = seq_lseek,
1459 .release = single_release,
1462 static void fadump_init_files(void)
1464 struct dentry *debugfs_file;
1467 rc = sysfs_create_file(kernel_kobj, &fadump_attr.attr);
1469 printk(KERN_ERR "fadump: unable to create sysfs file"
1470 " fadump_enabled (%d)\n", rc);
1472 rc = sysfs_create_file(kernel_kobj, &fadump_register_attr.attr);
1474 printk(KERN_ERR "fadump: unable to create sysfs file"
1475 " fadump_registered (%d)\n", rc);
1477 debugfs_file = debugfs_create_file("fadump_region", 0444,
1478 powerpc_debugfs_root, NULL,
1479 &fadump_region_fops);
1481 printk(KERN_ERR "fadump: unable to create debugfs file"
1482 " fadump_region\n");
1484 if (fw_dump.dump_active) {
1485 rc = sysfs_create_file(kernel_kobj, &fadump_release_attr.attr);
1487 printk(KERN_ERR "fadump: unable to create sysfs file"
1488 " fadump_release_mem (%d)\n", rc);
1494 * Prepare for firmware-assisted dump.
1496 int __init setup_fadump(void)
1498 if (!fw_dump.fadump_enabled)
1501 if (!fw_dump.fadump_supported) {
1502 printk(KERN_ERR "Firmware-assisted dump is not supported on"
1503 " this hardware\n");
1507 fadump_show_config();
1509 * If dump data is available then see if it is valid and prepare for
1510 * saving it to the disk.
1512 if (fw_dump.dump_active) {
1514 * if dump process fails then invalidate the registration
1515 * and release memory before proceeding for re-registration.
1517 if (process_fadump(fdm_active) < 0)
1518 fadump_invalidate_release_mem();
1520 /* Initialize the kernel dump memory structure for FAD registration. */
1521 else if (fw_dump.reserve_dump_area_size)
1522 init_fadump_mem_struct(&fdm, fw_dump.reserve_dump_area_start);
1523 fadump_init_files();
1527 subsys_initcall(setup_fadump);