2 * Firmware Assisted dump: A robust mechanism to get reliable kernel crash
3 * dump with assistance from firmware. This approach does not use kexec,
4 * instead firmware assists in booting the kdump kernel while preserving
5 * memory contents. The most of the code implementation has been adapted
6 * from phyp assisted dump implementation written by Linas Vepstas and
9 * This program is free software; you can redistribute it and/or modify
10 * it under the terms of the GNU General Public License as published by
11 * the Free Software Foundation; either version 2 of the License, or
12 * (at your option) any later version.
14 * This program is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
17 * GNU General Public License for more details.
19 * You should have received a copy of the GNU General Public License
20 * along with this program; if not, write to the Free Software
21 * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
23 * Copyright 2011 IBM Corporation
24 * Author: Mahesh Salgaonkar <mahesh@linux.vnet.ibm.com>
28 #define pr_fmt(fmt) "fadump: " fmt
30 #include <linux/string.h>
31 #include <linux/memblock.h>
32 #include <linux/delay.h>
33 #include <linux/debugfs.h>
34 #include <linux/seq_file.h>
35 #include <linux/crash_dump.h>
36 #include <linux/kobject.h>
37 #include <linux/sysfs.h>
42 #include <asm/fadump.h>
43 #include <asm/debug.h>
44 #include <asm/setup.h>
46 static struct fw_dump fw_dump
;
47 static struct fadump_mem_struct fdm
;
48 static const struct fadump_mem_struct
*fdm_active
;
50 static DEFINE_MUTEX(fadump_mutex
);
51 struct fad_crash_memory_ranges crash_memory_ranges
[INIT_CRASHMEM_RANGES
];
54 /* Scan the Firmware Assisted dump configuration details. */
55 int __init
early_init_dt_scan_fw_dump(unsigned long node
,
56 const char *uname
, int depth
, void *data
)
58 const __be32
*sections
;
63 if (depth
!= 1 || strcmp(uname
, "rtas") != 0)
67 * Check if Firmware Assisted dump is supported. if yes, check
68 * if dump has been initiated on last reboot.
70 token
= of_get_flat_dt_prop(node
, "ibm,configure-kernel-dump", NULL
);
74 fw_dump
.fadump_supported
= 1;
75 fw_dump
.ibm_configure_kernel_dump
= be32_to_cpu(*token
);
78 * The 'ibm,kernel-dump' rtas node is present only if there is
79 * dump data waiting for us.
81 fdm_active
= of_get_flat_dt_prop(node
, "ibm,kernel-dump", NULL
);
83 fw_dump
.dump_active
= 1;
85 /* Get the sizes required to store dump data for the firmware provided
87 * For each dump section type supported, a 32bit cell which defines
88 * the ID of a supported section followed by two 32 bit cells which
89 * gives teh size of the section in bytes.
91 sections
= of_get_flat_dt_prop(node
, "ibm,configure-kernel-dump-sizes",
97 num_sections
= size
/ (3 * sizeof(u32
));
99 for (i
= 0; i
< num_sections
; i
++, sections
+= 3) {
100 u32 type
= (u32
)of_read_number(sections
, 1);
103 case FADUMP_CPU_STATE_DATA
:
104 fw_dump
.cpu_state_data_size
=
105 of_read_ulong(§ions
[1], 2);
107 case FADUMP_HPTE_REGION
:
108 fw_dump
.hpte_region_size
=
109 of_read_ulong(§ions
[1], 2);
117 int is_fadump_active(void)
119 return fw_dump
.dump_active
;
122 /* Print firmware assisted dump configurations for debugging purpose. */
123 static void fadump_show_config(void)
125 pr_debug("Support for firmware-assisted dump (fadump): %s\n",
126 (fw_dump
.fadump_supported
? "present" : "no support"));
128 if (!fw_dump
.fadump_supported
)
131 pr_debug("Fadump enabled : %s\n",
132 (fw_dump
.fadump_enabled
? "yes" : "no"));
133 pr_debug("Dump Active : %s\n",
134 (fw_dump
.dump_active
? "yes" : "no"));
135 pr_debug("Dump section sizes:\n");
136 pr_debug(" CPU state data size: %lx\n", fw_dump
.cpu_state_data_size
);
137 pr_debug(" HPTE region size : %lx\n", fw_dump
.hpte_region_size
);
138 pr_debug("Boot memory size : %lx\n", fw_dump
.boot_memory_size
);
141 static unsigned long init_fadump_mem_struct(struct fadump_mem_struct
*fdm
,
147 memset(fdm
, 0, sizeof(struct fadump_mem_struct
));
148 addr
= addr
& PAGE_MASK
;
150 fdm
->header
.dump_format_version
= cpu_to_be32(0x00000001);
151 fdm
->header
.dump_num_sections
= cpu_to_be16(3);
152 fdm
->header
.dump_status_flag
= 0;
153 fdm
->header
.offset_first_dump_section
=
154 cpu_to_be32((u32
)offsetof(struct fadump_mem_struct
, cpu_state_data
));
157 * Fields for disk dump option.
158 * We are not using disk dump option, hence set these fields to 0.
160 fdm
->header
.dd_block_size
= 0;
161 fdm
->header
.dd_block_offset
= 0;
162 fdm
->header
.dd_num_blocks
= 0;
163 fdm
->header
.dd_offset_disk_path
= 0;
165 /* set 0 to disable an automatic dump-reboot. */
166 fdm
->header
.max_time_auto
= 0;
168 /* Kernel dump sections */
169 /* cpu state data section. */
170 fdm
->cpu_state_data
.request_flag
= cpu_to_be32(FADUMP_REQUEST_FLAG
);
171 fdm
->cpu_state_data
.source_data_type
= cpu_to_be16(FADUMP_CPU_STATE_DATA
);
172 fdm
->cpu_state_data
.source_address
= 0;
173 fdm
->cpu_state_data
.source_len
= cpu_to_be64(fw_dump
.cpu_state_data_size
);
174 fdm
->cpu_state_data
.destination_address
= cpu_to_be64(addr
);
175 addr
+= fw_dump
.cpu_state_data_size
;
177 /* hpte region section */
178 fdm
->hpte_region
.request_flag
= cpu_to_be32(FADUMP_REQUEST_FLAG
);
179 fdm
->hpte_region
.source_data_type
= cpu_to_be16(FADUMP_HPTE_REGION
);
180 fdm
->hpte_region
.source_address
= 0;
181 fdm
->hpte_region
.source_len
= cpu_to_be64(fw_dump
.hpte_region_size
);
182 fdm
->hpte_region
.destination_address
= cpu_to_be64(addr
);
183 addr
+= fw_dump
.hpte_region_size
;
185 /* RMA region section */
186 fdm
->rmr_region
.request_flag
= cpu_to_be32(FADUMP_REQUEST_FLAG
);
187 fdm
->rmr_region
.source_data_type
= cpu_to_be16(FADUMP_REAL_MODE_REGION
);
188 fdm
->rmr_region
.source_address
= cpu_to_be64(RMA_START
);
189 fdm
->rmr_region
.source_len
= cpu_to_be64(fw_dump
.boot_memory_size
);
190 fdm
->rmr_region
.destination_address
= cpu_to_be64(addr
);
191 addr
+= fw_dump
.boot_memory_size
;
197 * fadump_calculate_reserve_size(): reserve variable boot area 5% of System RAM
199 * Function to find the largest memory size we need to reserve during early
200 * boot process. This will be the size of the memory that is required for a
201 * kernel to boot successfully.
203 * This function has been taken from phyp-assisted dump feature implementation.
205 * returns larger of 256MB or 5% rounded down to multiples of 256MB.
207 * TODO: Come up with better approach to find out more accurate memory size
208 * that is required for a kernel to boot successfully.
211 static inline unsigned long fadump_calculate_reserve_size(void)
216 * Check if the size is specified through fadump_reserve_mem= cmdline
217 * option. If yes, then use that.
219 if (fw_dump
.reserve_bootvar
)
220 return fw_dump
.reserve_bootvar
;
222 /* divide by 20 to get 5% of value */
223 size
= memblock_end_of_DRAM() / 20;
225 /* round it down in multiples of 256 */
226 size
= size
& ~0x0FFFFFFFUL
;
228 /* Truncate to memory_limit. We don't want to over reserve the memory.*/
229 if (memory_limit
&& size
> memory_limit
)
232 return (size
> MIN_BOOT_MEM
? size
: MIN_BOOT_MEM
);
236 * Calculate the total memory size required to be reserved for
237 * firmware-assisted dump registration.
239 static unsigned long get_fadump_area_size(void)
241 unsigned long size
= 0;
243 size
+= fw_dump
.cpu_state_data_size
;
244 size
+= fw_dump
.hpte_region_size
;
245 size
+= fw_dump
.boot_memory_size
;
246 size
+= sizeof(struct fadump_crash_info_header
);
247 size
+= sizeof(struct elfhdr
); /* ELF core header.*/
248 size
+= sizeof(struct elf_phdr
); /* place holder for cpu notes */
249 /* Program headers for crash memory regions. */
250 size
+= sizeof(struct elf_phdr
) * (memblock_num_regions(memory
) + 2);
252 size
= PAGE_ALIGN(size
);
256 int __init
fadump_reserve_mem(void)
258 unsigned long base
, size
, memory_boundary
;
260 if (!fw_dump
.fadump_enabled
)
263 if (!fw_dump
.fadump_supported
) {
264 printk(KERN_INFO
"Firmware-assisted dump is not supported on"
266 fw_dump
.fadump_enabled
= 0;
270 * Initialize boot memory size
271 * If dump is active then we have already calculated the size during
275 fw_dump
.boot_memory_size
= be64_to_cpu(fdm_active
->rmr_region
.source_len
);
277 fw_dump
.boot_memory_size
= fadump_calculate_reserve_size();
280 * Calculate the memory boundary.
281 * If memory_limit is less than actual memory boundary then reserve
282 * the memory for fadump beyond the memory_limit and adjust the
283 * memory_limit accordingly, so that the running kernel can run with
284 * specified memory_limit.
286 if (memory_limit
&& memory_limit
< memblock_end_of_DRAM()) {
287 size
= get_fadump_area_size();
288 if ((memory_limit
+ size
) < memblock_end_of_DRAM())
289 memory_limit
+= size
;
291 memory_limit
= memblock_end_of_DRAM();
292 printk(KERN_INFO
"Adjusted memory_limit for firmware-assisted"
293 " dump, now %#016llx\n", memory_limit
);
296 memory_boundary
= memory_limit
;
298 memory_boundary
= memblock_end_of_DRAM();
300 if (fw_dump
.dump_active
) {
301 printk(KERN_INFO
"Firmware-assisted dump is active.\n");
303 * If last boot has crashed then reserve all the memory
304 * above boot_memory_size so that we don't touch it until
305 * dump is written to disk by userspace tool. This memory
306 * will be released for general use once the dump is saved.
308 base
= fw_dump
.boot_memory_size
;
309 size
= memory_boundary
- base
;
310 memblock_reserve(base
, size
);
311 printk(KERN_INFO
"Reserved %ldMB of memory at %ldMB "
312 "for saving crash dump\n",
313 (unsigned long)(size
>> 20),
314 (unsigned long)(base
>> 20));
316 fw_dump
.fadumphdr_addr
=
317 be64_to_cpu(fdm_active
->rmr_region
.destination_address
) +
318 be64_to_cpu(fdm_active
->rmr_region
.source_len
);
319 pr_debug("fadumphdr_addr = %p\n",
320 (void *) fw_dump
.fadumphdr_addr
);
322 /* Reserve the memory at the top of memory. */
323 size
= get_fadump_area_size();
324 base
= memory_boundary
- size
;
325 memblock_reserve(base
, size
);
326 printk(KERN_INFO
"Reserved %ldMB of memory at %ldMB "
327 "for firmware-assisted dump\n",
328 (unsigned long)(size
>> 20),
329 (unsigned long)(base
>> 20));
331 fw_dump
.reserve_dump_area_start
= base
;
332 fw_dump
.reserve_dump_area_size
= size
;
336 unsigned long __init
arch_reserved_kernel_pages(void)
338 return memblock_reserved_size() / PAGE_SIZE
;
341 /* Look for fadump= cmdline option. */
342 static int __init
early_fadump_param(char *p
)
347 if (strncmp(p
, "on", 2) == 0)
348 fw_dump
.fadump_enabled
= 1;
349 else if (strncmp(p
, "off", 3) == 0)
350 fw_dump
.fadump_enabled
= 0;
354 early_param("fadump", early_fadump_param
);
356 /* Look for fadump_reserve_mem= cmdline option */
357 static int __init
early_fadump_reserve_mem(char *p
)
360 fw_dump
.reserve_bootvar
= memparse(p
, &p
);
363 early_param("fadump_reserve_mem", early_fadump_reserve_mem
);
365 static void register_fw_dump(struct fadump_mem_struct
*fdm
)
368 unsigned int wait_time
;
370 pr_debug("Registering for firmware-assisted kernel dump...\n");
372 /* TODO: Add upper time limit for the delay */
374 rc
= rtas_call(fw_dump
.ibm_configure_kernel_dump
, 3, 1, NULL
,
375 FADUMP_REGISTER
, fdm
,
376 sizeof(struct fadump_mem_struct
));
378 wait_time
= rtas_busy_delay_time(rc
);
386 printk(KERN_ERR
"Failed to register firmware-assisted kernel"
387 " dump. Hardware Error(%d).\n", rc
);
390 printk(KERN_ERR
"Failed to register firmware-assisted kernel"
391 " dump. Parameter Error(%d).\n", rc
);
394 printk(KERN_ERR
"firmware-assisted kernel dump is already "
396 fw_dump
.dump_registered
= 1;
399 printk(KERN_INFO
"firmware-assisted kernel dump registration"
401 fw_dump
.dump_registered
= 1;
406 void crash_fadump(struct pt_regs
*regs
, const char *str
)
408 struct fadump_crash_info_header
*fdh
= NULL
;
410 if (!fw_dump
.dump_registered
|| !fw_dump
.fadumphdr_addr
)
413 fdh
= __va(fw_dump
.fadumphdr_addr
);
414 crashing_cpu
= smp_processor_id();
415 fdh
->crashing_cpu
= crashing_cpu
;
416 crash_save_vmcoreinfo();
421 ppc_save_regs(&fdh
->regs
);
423 fdh
->online_mask
= *cpu_online_mask
;
425 /* Call ibm,os-term rtas call to trigger firmware assisted dump */
426 rtas_os_term((char *)str
);
429 #define GPR_MASK 0xffffff0000000000
430 static inline int fadump_gpr_index(u64 id
)
435 if ((id
& GPR_MASK
) == REG_ID("GPR")) {
436 /* get the digits at the end */
441 str
[0] = (id
>> 8) & 0xff;
442 sscanf(str
, "%d", &i
);
449 static inline void fadump_set_regval(struct pt_regs
*regs
, u64 reg_id
,
454 i
= fadump_gpr_index(reg_id
);
456 regs
->gpr
[i
] = (unsigned long)reg_val
;
457 else if (reg_id
== REG_ID("NIA"))
458 regs
->nip
= (unsigned long)reg_val
;
459 else if (reg_id
== REG_ID("MSR"))
460 regs
->msr
= (unsigned long)reg_val
;
461 else if (reg_id
== REG_ID("CTR"))
462 regs
->ctr
= (unsigned long)reg_val
;
463 else if (reg_id
== REG_ID("LR"))
464 regs
->link
= (unsigned long)reg_val
;
465 else if (reg_id
== REG_ID("XER"))
466 regs
->xer
= (unsigned long)reg_val
;
467 else if (reg_id
== REG_ID("CR"))
468 regs
->ccr
= (unsigned long)reg_val
;
469 else if (reg_id
== REG_ID("DAR"))
470 regs
->dar
= (unsigned long)reg_val
;
471 else if (reg_id
== REG_ID("DSISR"))
472 regs
->dsisr
= (unsigned long)reg_val
;
475 static struct fadump_reg_entry
*
476 fadump_read_registers(struct fadump_reg_entry
*reg_entry
, struct pt_regs
*regs
)
478 memset(regs
, 0, sizeof(struct pt_regs
));
480 while (be64_to_cpu(reg_entry
->reg_id
) != REG_ID("CPUEND")) {
481 fadump_set_regval(regs
, be64_to_cpu(reg_entry
->reg_id
),
482 be64_to_cpu(reg_entry
->reg_value
));
489 static u32
*fadump_append_elf_note(u32
*buf
, char *name
, unsigned type
,
490 void *data
, size_t data_len
)
492 struct elf_note note
;
494 note
.n_namesz
= strlen(name
) + 1;
495 note
.n_descsz
= data_len
;
497 memcpy(buf
, ¬e
, sizeof(note
));
498 buf
+= (sizeof(note
) + 3)/4;
499 memcpy(buf
, name
, note
.n_namesz
);
500 buf
+= (note
.n_namesz
+ 3)/4;
501 memcpy(buf
, data
, note
.n_descsz
);
502 buf
+= (note
.n_descsz
+ 3)/4;
507 static void fadump_final_note(u32
*buf
)
509 struct elf_note note
;
514 memcpy(buf
, ¬e
, sizeof(note
));
517 static u32
*fadump_regs_to_elf_notes(u32
*buf
, struct pt_regs
*regs
)
519 struct elf_prstatus prstatus
;
521 memset(&prstatus
, 0, sizeof(prstatus
));
523 * FIXME: How do i get PID? Do I really need it?
524 * prstatus.pr_pid = ????
526 elf_core_copy_kernel_regs(&prstatus
.pr_reg
, regs
);
527 buf
= fadump_append_elf_note(buf
, KEXEC_CORE_NOTE_NAME
, NT_PRSTATUS
,
528 &prstatus
, sizeof(prstatus
));
532 static void fadump_update_elfcore_header(char *bufp
)
535 struct elf_phdr
*phdr
;
537 elf
= (struct elfhdr
*)bufp
;
538 bufp
+= sizeof(struct elfhdr
);
540 /* First note is a place holder for cpu notes info. */
541 phdr
= (struct elf_phdr
*)bufp
;
543 if (phdr
->p_type
== PT_NOTE
) {
544 phdr
->p_paddr
= fw_dump
.cpu_notes_buf
;
545 phdr
->p_offset
= phdr
->p_paddr
;
546 phdr
->p_filesz
= fw_dump
.cpu_notes_buf_size
;
547 phdr
->p_memsz
= fw_dump
.cpu_notes_buf_size
;
552 static void *fadump_cpu_notes_buf_alloc(unsigned long size
)
556 unsigned long order
, count
, i
;
558 order
= get_order(size
);
559 vaddr
= (void *)__get_free_pages(GFP_KERNEL
|__GFP_ZERO
, order
);
564 page
= virt_to_page(vaddr
);
565 for (i
= 0; i
< count
; i
++)
566 SetPageReserved(page
+ i
);
570 static void fadump_cpu_notes_buf_free(unsigned long vaddr
, unsigned long size
)
573 unsigned long order
, count
, i
;
575 order
= get_order(size
);
577 page
= virt_to_page(vaddr
);
578 for (i
= 0; i
< count
; i
++)
579 ClearPageReserved(page
+ i
);
580 __free_pages(page
, order
);
584 * Read CPU state dump data and convert it into ELF notes.
585 * The CPU dump starts with magic number "REGSAVE". NumCpusOffset should be
586 * used to access the data to allow for additional fields to be added without
587 * affecting compatibility. Each list of registers for a CPU starts with
588 * "CPUSTRT" and ends with "CPUEND". Each register entry is of 16 bytes,
589 * 8 Byte ASCII identifier and 8 Byte register value. The register entry
590 * with identifier "CPUSTRT" and "CPUEND" contains 4 byte cpu id as part
591 * of register value. For more details refer to PAPR document.
593 * Only for the crashing cpu we ignore the CPU dump data and get exact
594 * state from fadump crash info structure populated by first kernel at the
597 static int __init
fadump_build_cpu_notes(const struct fadump_mem_struct
*fdm
)
599 struct fadump_reg_save_area_header
*reg_header
;
600 struct fadump_reg_entry
*reg_entry
;
601 struct fadump_crash_info_header
*fdh
= NULL
;
604 u32 num_cpus
, *note_buf
;
606 int i
, rc
= 0, cpu
= 0;
608 if (!fdm
->cpu_state_data
.bytes_dumped
)
611 addr
= be64_to_cpu(fdm
->cpu_state_data
.destination_address
);
615 if (be64_to_cpu(reg_header
->magic_number
) != REGSAVE_AREA_MAGIC
) {
616 printk(KERN_ERR
"Unable to read register save area.\n");
619 pr_debug("--------CPU State Data------------\n");
620 pr_debug("Magic Number: %llx\n", be64_to_cpu(reg_header
->magic_number
));
621 pr_debug("NumCpuOffset: %x\n", be32_to_cpu(reg_header
->num_cpu_offset
));
623 vaddr
+= be32_to_cpu(reg_header
->num_cpu_offset
);
624 num_cpus
= be32_to_cpu(*((__be32
*)(vaddr
)));
625 pr_debug("NumCpus : %u\n", num_cpus
);
626 vaddr
+= sizeof(u32
);
627 reg_entry
= (struct fadump_reg_entry
*)vaddr
;
629 /* Allocate buffer to hold cpu crash notes. */
630 fw_dump
.cpu_notes_buf_size
= num_cpus
* sizeof(note_buf_t
);
631 fw_dump
.cpu_notes_buf_size
= PAGE_ALIGN(fw_dump
.cpu_notes_buf_size
);
632 note_buf
= fadump_cpu_notes_buf_alloc(fw_dump
.cpu_notes_buf_size
);
634 printk(KERN_ERR
"Failed to allocate 0x%lx bytes for "
635 "cpu notes buffer\n", fw_dump
.cpu_notes_buf_size
);
638 fw_dump
.cpu_notes_buf
= __pa(note_buf
);
640 pr_debug("Allocated buffer for cpu notes of size %ld at %p\n",
641 (num_cpus
* sizeof(note_buf_t
)), note_buf
);
643 if (fw_dump
.fadumphdr_addr
)
644 fdh
= __va(fw_dump
.fadumphdr_addr
);
646 for (i
= 0; i
< num_cpus
; i
++) {
647 if (be64_to_cpu(reg_entry
->reg_id
) != REG_ID("CPUSTRT")) {
648 printk(KERN_ERR
"Unable to read CPU state data\n");
652 /* Lower 4 bytes of reg_value contains logical cpu id */
653 cpu
= be64_to_cpu(reg_entry
->reg_value
) & FADUMP_CPU_ID_MASK
;
654 if (fdh
&& !cpumask_test_cpu(cpu
, &fdh
->online_mask
)) {
655 SKIP_TO_NEXT_CPU(reg_entry
);
658 pr_debug("Reading register data for cpu %d...\n", cpu
);
659 if (fdh
&& fdh
->crashing_cpu
== cpu
) {
661 note_buf
= fadump_regs_to_elf_notes(note_buf
, ®s
);
662 SKIP_TO_NEXT_CPU(reg_entry
);
665 reg_entry
= fadump_read_registers(reg_entry
, ®s
);
666 note_buf
= fadump_regs_to_elf_notes(note_buf
, ®s
);
669 fadump_final_note(note_buf
);
672 pr_debug("Updating elfcore header (%llx) with cpu notes\n",
673 fdh
->elfcorehdr_addr
);
674 fadump_update_elfcore_header((char *)__va(fdh
->elfcorehdr_addr
));
679 fadump_cpu_notes_buf_free((unsigned long)__va(fw_dump
.cpu_notes_buf
),
680 fw_dump
.cpu_notes_buf_size
);
681 fw_dump
.cpu_notes_buf
= 0;
682 fw_dump
.cpu_notes_buf_size
= 0;
688 * Validate and process the dump data stored by firmware before exporting
689 * it through '/proc/vmcore'.
691 static int __init
process_fadump(const struct fadump_mem_struct
*fdm_active
)
693 struct fadump_crash_info_header
*fdh
;
696 if (!fdm_active
|| !fw_dump
.fadumphdr_addr
)
699 /* Check if the dump data is valid. */
700 if ((be16_to_cpu(fdm_active
->header
.dump_status_flag
) == FADUMP_ERROR_FLAG
) ||
701 (fdm_active
->cpu_state_data
.error_flags
!= 0) ||
702 (fdm_active
->rmr_region
.error_flags
!= 0)) {
703 printk(KERN_ERR
"Dump taken by platform is not valid\n");
706 if ((fdm_active
->rmr_region
.bytes_dumped
!=
707 fdm_active
->rmr_region
.source_len
) ||
708 !fdm_active
->cpu_state_data
.bytes_dumped
) {
709 printk(KERN_ERR
"Dump taken by platform is incomplete\n");
713 /* Validate the fadump crash info header */
714 fdh
= __va(fw_dump
.fadumphdr_addr
);
715 if (fdh
->magic_number
!= FADUMP_CRASH_INFO_MAGIC
) {
716 printk(KERN_ERR
"Crash info header is not valid.\n");
720 rc
= fadump_build_cpu_notes(fdm_active
);
725 * We are done validating dump info and elfcore header is now ready
726 * to be exported. set elfcorehdr_addr so that vmcore module will
727 * export the elfcore header through '/proc/vmcore'.
729 elfcorehdr_addr
= fdh
->elfcorehdr_addr
;
734 static inline void fadump_add_crash_memory(unsigned long long base
,
735 unsigned long long end
)
740 pr_debug("crash_memory_range[%d] [%#016llx-%#016llx], %#llx bytes\n",
741 crash_mem_ranges
, base
, end
- 1, (end
- base
));
742 crash_memory_ranges
[crash_mem_ranges
].base
= base
;
743 crash_memory_ranges
[crash_mem_ranges
].size
= end
- base
;
747 static void fadump_exclude_reserved_area(unsigned long long start
,
748 unsigned long long end
)
750 unsigned long long ra_start
, ra_end
;
752 ra_start
= fw_dump
.reserve_dump_area_start
;
753 ra_end
= ra_start
+ fw_dump
.reserve_dump_area_size
;
755 if ((ra_start
< end
) && (ra_end
> start
)) {
756 if ((start
< ra_start
) && (end
> ra_end
)) {
757 fadump_add_crash_memory(start
, ra_start
);
758 fadump_add_crash_memory(ra_end
, end
);
759 } else if (start
< ra_start
) {
760 fadump_add_crash_memory(start
, ra_start
);
761 } else if (ra_end
< end
) {
762 fadump_add_crash_memory(ra_end
, end
);
765 fadump_add_crash_memory(start
, end
);
768 static int fadump_init_elfcore_header(char *bufp
)
772 elf
= (struct elfhdr
*) bufp
;
773 bufp
+= sizeof(struct elfhdr
);
774 memcpy(elf
->e_ident
, ELFMAG
, SELFMAG
);
775 elf
->e_ident
[EI_CLASS
] = ELF_CLASS
;
776 elf
->e_ident
[EI_DATA
] = ELF_DATA
;
777 elf
->e_ident
[EI_VERSION
] = EV_CURRENT
;
778 elf
->e_ident
[EI_OSABI
] = ELF_OSABI
;
779 memset(elf
->e_ident
+EI_PAD
, 0, EI_NIDENT
-EI_PAD
);
780 elf
->e_type
= ET_CORE
;
781 elf
->e_machine
= ELF_ARCH
;
782 elf
->e_version
= EV_CURRENT
;
784 elf
->e_phoff
= sizeof(struct elfhdr
);
786 #if defined(_CALL_ELF)
787 elf
->e_flags
= _CALL_ELF
;
791 elf
->e_ehsize
= sizeof(struct elfhdr
);
792 elf
->e_phentsize
= sizeof(struct elf_phdr
);
794 elf
->e_shentsize
= 0;
802 * Traverse through memblock structure and setup crash memory ranges. These
803 * ranges will be used create PT_LOAD program headers in elfcore header.
805 static void fadump_setup_crash_memory_ranges(void)
807 struct memblock_region
*reg
;
808 unsigned long long start
, end
;
810 pr_debug("Setup crash memory ranges.\n");
811 crash_mem_ranges
= 0;
813 * add the first memory chunk (RMA_START through boot_memory_size) as
814 * a separate memory chunk. The reason is, at the time crash firmware
815 * will move the content of this memory chunk to different location
816 * specified during fadump registration. We need to create a separate
817 * program header for this chunk with the correct offset.
819 fadump_add_crash_memory(RMA_START
, fw_dump
.boot_memory_size
);
821 for_each_memblock(memory
, reg
) {
822 start
= (unsigned long long)reg
->base
;
823 end
= start
+ (unsigned long long)reg
->size
;
824 if (start
== RMA_START
&& end
>= fw_dump
.boot_memory_size
)
825 start
= fw_dump
.boot_memory_size
;
827 /* add this range excluding the reserved dump area. */
828 fadump_exclude_reserved_area(start
, end
);
833 * If the given physical address falls within the boot memory region then
834 * return the relocated address that points to the dump region reserved
835 * for saving initial boot memory contents.
837 static inline unsigned long fadump_relocate(unsigned long paddr
)
839 if (paddr
> RMA_START
&& paddr
< fw_dump
.boot_memory_size
)
840 return be64_to_cpu(fdm
.rmr_region
.destination_address
) + paddr
;
845 static int fadump_create_elfcore_headers(char *bufp
)
848 struct elf_phdr
*phdr
;
851 fadump_init_elfcore_header(bufp
);
852 elf
= (struct elfhdr
*)bufp
;
853 bufp
+= sizeof(struct elfhdr
);
856 * setup ELF PT_NOTE, place holder for cpu notes info. The notes info
857 * will be populated during second kernel boot after crash. Hence
858 * this PT_NOTE will always be the first elf note.
860 * NOTE: Any new ELF note addition should be placed after this note.
862 phdr
= (struct elf_phdr
*)bufp
;
863 bufp
+= sizeof(struct elf_phdr
);
864 phdr
->p_type
= PT_NOTE
;
876 /* setup ELF PT_NOTE for vmcoreinfo */
877 phdr
= (struct elf_phdr
*)bufp
;
878 bufp
+= sizeof(struct elf_phdr
);
879 phdr
->p_type
= PT_NOTE
;
884 phdr
->p_paddr
= fadump_relocate(paddr_vmcoreinfo_note());
885 phdr
->p_offset
= phdr
->p_paddr
;
886 phdr
->p_memsz
= vmcoreinfo_max_size
;
887 phdr
->p_filesz
= vmcoreinfo_max_size
;
889 /* Increment number of program headers. */
892 /* setup PT_LOAD sections. */
894 for (i
= 0; i
< crash_mem_ranges
; i
++) {
895 unsigned long long mbase
, msize
;
896 mbase
= crash_memory_ranges
[i
].base
;
897 msize
= crash_memory_ranges
[i
].size
;
902 phdr
= (struct elf_phdr
*)bufp
;
903 bufp
+= sizeof(struct elf_phdr
);
904 phdr
->p_type
= PT_LOAD
;
905 phdr
->p_flags
= PF_R
|PF_W
|PF_X
;
906 phdr
->p_offset
= mbase
;
908 if (mbase
== RMA_START
) {
910 * The entire RMA region will be moved by firmware
911 * to the specified destination_address. Hence set
912 * the correct offset.
914 phdr
->p_offset
= be64_to_cpu(fdm
.rmr_region
.destination_address
);
917 phdr
->p_paddr
= mbase
;
918 phdr
->p_vaddr
= (unsigned long)__va(mbase
);
919 phdr
->p_filesz
= msize
;
920 phdr
->p_memsz
= msize
;
923 /* Increment number of program headers. */
929 static unsigned long init_fadump_header(unsigned long addr
)
931 struct fadump_crash_info_header
*fdh
;
936 fw_dump
.fadumphdr_addr
= addr
;
938 addr
+= sizeof(struct fadump_crash_info_header
);
940 memset(fdh
, 0, sizeof(struct fadump_crash_info_header
));
941 fdh
->magic_number
= FADUMP_CRASH_INFO_MAGIC
;
942 fdh
->elfcorehdr_addr
= addr
;
943 /* We will set the crashing cpu id in crash_fadump() during crash. */
944 fdh
->crashing_cpu
= CPU_UNKNOWN
;
949 static void register_fadump(void)
955 * If no memory is reserved then we can not register for firmware-
958 if (!fw_dump
.reserve_dump_area_size
)
961 fadump_setup_crash_memory_ranges();
963 addr
= be64_to_cpu(fdm
.rmr_region
.destination_address
) + be64_to_cpu(fdm
.rmr_region
.source_len
);
964 /* Initialize fadump crash info header. */
965 addr
= init_fadump_header(addr
);
968 pr_debug("Creating ELF core headers at %#016lx\n", addr
);
969 fadump_create_elfcore_headers(vaddr
);
971 /* register the future kernel dump with firmware. */
972 register_fw_dump(&fdm
);
975 static int fadump_unregister_dump(struct fadump_mem_struct
*fdm
)
978 unsigned int wait_time
;
980 pr_debug("Un-register firmware-assisted dump\n");
982 /* TODO: Add upper time limit for the delay */
984 rc
= rtas_call(fw_dump
.ibm_configure_kernel_dump
, 3, 1, NULL
,
985 FADUMP_UNREGISTER
, fdm
,
986 sizeof(struct fadump_mem_struct
));
988 wait_time
= rtas_busy_delay_time(rc
);
994 printk(KERN_ERR
"Failed to un-register firmware-assisted dump."
995 " unexpected error(%d).\n", rc
);
998 fw_dump
.dump_registered
= 0;
1002 static int fadump_invalidate_dump(struct fadump_mem_struct
*fdm
)
1005 unsigned int wait_time
;
1007 pr_debug("Invalidating firmware-assisted dump registration\n");
1009 /* TODO: Add upper time limit for the delay */
1011 rc
= rtas_call(fw_dump
.ibm_configure_kernel_dump
, 3, 1, NULL
,
1012 FADUMP_INVALIDATE
, fdm
,
1013 sizeof(struct fadump_mem_struct
));
1015 wait_time
= rtas_busy_delay_time(rc
);
1018 } while (wait_time
);
1021 pr_err("Failed to invalidate firmware-assisted dump registration. Unexpected error (%d).\n", rc
);
1024 fw_dump
.dump_active
= 0;
1029 void fadump_cleanup(void)
1031 /* Invalidate the registration only if dump is active. */
1032 if (fw_dump
.dump_active
) {
1033 init_fadump_mem_struct(&fdm
,
1034 be64_to_cpu(fdm_active
->cpu_state_data
.destination_address
));
1035 fadump_invalidate_dump(&fdm
);
1040 * Release the memory that was reserved in early boot to preserve the memory
1041 * contents. The released memory will be available for general use.
1043 static void fadump_release_memory(unsigned long begin
, unsigned long end
)
1046 unsigned long ra_start
, ra_end
;
1048 ra_start
= fw_dump
.reserve_dump_area_start
;
1049 ra_end
= ra_start
+ fw_dump
.reserve_dump_area_size
;
1051 for (addr
= begin
; addr
< end
; addr
+= PAGE_SIZE
) {
1053 * exclude the dump reserve area. Will reuse it for next
1054 * fadump registration.
1056 if (addr
<= ra_end
&& ((addr
+ PAGE_SIZE
) > ra_start
))
1059 free_reserved_page(pfn_to_page(addr
>> PAGE_SHIFT
));
1063 static void fadump_invalidate_release_mem(void)
1065 unsigned long reserved_area_start
, reserved_area_end
;
1066 unsigned long destination_address
;
1068 mutex_lock(&fadump_mutex
);
1069 if (!fw_dump
.dump_active
) {
1070 mutex_unlock(&fadump_mutex
);
1074 destination_address
= be64_to_cpu(fdm_active
->cpu_state_data
.destination_address
);
1076 mutex_unlock(&fadump_mutex
);
1079 * Save the current reserved memory bounds we will require them
1080 * later for releasing the memory for general use.
1082 reserved_area_start
= fw_dump
.reserve_dump_area_start
;
1083 reserved_area_end
= reserved_area_start
+
1084 fw_dump
.reserve_dump_area_size
;
1086 * Setup reserve_dump_area_start and its size so that we can
1087 * reuse this reserved memory for Re-registration.
1089 fw_dump
.reserve_dump_area_start
= destination_address
;
1090 fw_dump
.reserve_dump_area_size
= get_fadump_area_size();
1092 fadump_release_memory(reserved_area_start
, reserved_area_end
);
1093 if (fw_dump
.cpu_notes_buf
) {
1094 fadump_cpu_notes_buf_free(
1095 (unsigned long)__va(fw_dump
.cpu_notes_buf
),
1096 fw_dump
.cpu_notes_buf_size
);
1097 fw_dump
.cpu_notes_buf
= 0;
1098 fw_dump
.cpu_notes_buf_size
= 0;
1100 /* Initialize the kernel dump memory structure for FAD registration. */
1101 init_fadump_mem_struct(&fdm
, fw_dump
.reserve_dump_area_start
);
1104 static ssize_t
fadump_release_memory_store(struct kobject
*kobj
,
1105 struct kobj_attribute
*attr
,
1106 const char *buf
, size_t count
)
1108 if (!fw_dump
.dump_active
)
1111 if (buf
[0] == '1') {
1113 * Take away the '/proc/vmcore'. We are releasing the dump
1114 * memory, hence it will not be valid anymore.
1116 #ifdef CONFIG_PROC_VMCORE
1119 fadump_invalidate_release_mem();
1126 static ssize_t
fadump_enabled_show(struct kobject
*kobj
,
1127 struct kobj_attribute
*attr
,
1130 return sprintf(buf
, "%d\n", fw_dump
.fadump_enabled
);
1133 static ssize_t
fadump_register_show(struct kobject
*kobj
,
1134 struct kobj_attribute
*attr
,
1137 return sprintf(buf
, "%d\n", fw_dump
.dump_registered
);
1140 static ssize_t
fadump_register_store(struct kobject
*kobj
,
1141 struct kobj_attribute
*attr
,
1142 const char *buf
, size_t count
)
1146 if (!fw_dump
.fadump_enabled
|| fdm_active
)
1149 mutex_lock(&fadump_mutex
);
1153 if (fw_dump
.dump_registered
== 0) {
1157 /* Un-register Firmware-assisted dump */
1158 fadump_unregister_dump(&fdm
);
1161 if (fw_dump
.dump_registered
== 1) {
1165 /* Register Firmware-assisted dump */
1174 mutex_unlock(&fadump_mutex
);
1175 return ret
< 0 ? ret
: count
;
1178 static int fadump_region_show(struct seq_file
*m
, void *private)
1180 const struct fadump_mem_struct
*fdm_ptr
;
1182 if (!fw_dump
.fadump_enabled
)
1185 mutex_lock(&fadump_mutex
);
1187 fdm_ptr
= fdm_active
;
1189 mutex_unlock(&fadump_mutex
);
1194 "CPU : [%#016llx-%#016llx] %#llx bytes, "
1196 be64_to_cpu(fdm_ptr
->cpu_state_data
.destination_address
),
1197 be64_to_cpu(fdm_ptr
->cpu_state_data
.destination_address
) +
1198 be64_to_cpu(fdm_ptr
->cpu_state_data
.source_len
) - 1,
1199 be64_to_cpu(fdm_ptr
->cpu_state_data
.source_len
),
1200 be64_to_cpu(fdm_ptr
->cpu_state_data
.bytes_dumped
));
1202 "HPTE: [%#016llx-%#016llx] %#llx bytes, "
1204 be64_to_cpu(fdm_ptr
->hpte_region
.destination_address
),
1205 be64_to_cpu(fdm_ptr
->hpte_region
.destination_address
) +
1206 be64_to_cpu(fdm_ptr
->hpte_region
.source_len
) - 1,
1207 be64_to_cpu(fdm_ptr
->hpte_region
.source_len
),
1208 be64_to_cpu(fdm_ptr
->hpte_region
.bytes_dumped
));
1210 "DUMP: [%#016llx-%#016llx] %#llx bytes, "
1212 be64_to_cpu(fdm_ptr
->rmr_region
.destination_address
),
1213 be64_to_cpu(fdm_ptr
->rmr_region
.destination_address
) +
1214 be64_to_cpu(fdm_ptr
->rmr_region
.source_len
) - 1,
1215 be64_to_cpu(fdm_ptr
->rmr_region
.source_len
),
1216 be64_to_cpu(fdm_ptr
->rmr_region
.bytes_dumped
));
1219 (fw_dump
.reserve_dump_area_start
==
1220 be64_to_cpu(fdm_ptr
->cpu_state_data
.destination_address
)))
1223 /* Dump is active. Show reserved memory region. */
1225 " : [%#016llx-%#016llx] %#llx bytes, "
1227 (unsigned long long)fw_dump
.reserve_dump_area_start
,
1228 be64_to_cpu(fdm_ptr
->cpu_state_data
.destination_address
) - 1,
1229 be64_to_cpu(fdm_ptr
->cpu_state_data
.destination_address
) -
1230 fw_dump
.reserve_dump_area_start
,
1231 be64_to_cpu(fdm_ptr
->cpu_state_data
.destination_address
) -
1232 fw_dump
.reserve_dump_area_start
);
1235 mutex_unlock(&fadump_mutex
);
1239 static struct kobj_attribute fadump_release_attr
= __ATTR(fadump_release_mem
,
1241 fadump_release_memory_store
);
1242 static struct kobj_attribute fadump_attr
= __ATTR(fadump_enabled
,
1243 0444, fadump_enabled_show
,
1245 static struct kobj_attribute fadump_register_attr
= __ATTR(fadump_registered
,
1246 0644, fadump_register_show
,
1247 fadump_register_store
);
1249 static int fadump_region_open(struct inode
*inode
, struct file
*file
)
1251 return single_open(file
, fadump_region_show
, inode
->i_private
);
1254 static const struct file_operations fadump_region_fops
= {
1255 .open
= fadump_region_open
,
1257 .llseek
= seq_lseek
,
1258 .release
= single_release
,
1261 static void fadump_init_files(void)
1263 struct dentry
*debugfs_file
;
1266 rc
= sysfs_create_file(kernel_kobj
, &fadump_attr
.attr
);
1268 printk(KERN_ERR
"fadump: unable to create sysfs file"
1269 " fadump_enabled (%d)\n", rc
);
1271 rc
= sysfs_create_file(kernel_kobj
, &fadump_register_attr
.attr
);
1273 printk(KERN_ERR
"fadump: unable to create sysfs file"
1274 " fadump_registered (%d)\n", rc
);
1276 debugfs_file
= debugfs_create_file("fadump_region", 0444,
1277 powerpc_debugfs_root
, NULL
,
1278 &fadump_region_fops
);
1280 printk(KERN_ERR
"fadump: unable to create debugfs file"
1281 " fadump_region\n");
1283 if (fw_dump
.dump_active
) {
1284 rc
= sysfs_create_file(kernel_kobj
, &fadump_release_attr
.attr
);
1286 printk(KERN_ERR
"fadump: unable to create sysfs file"
1287 " fadump_release_mem (%d)\n", rc
);
1293 * Prepare for firmware-assisted dump.
1295 int __init
setup_fadump(void)
1297 if (!fw_dump
.fadump_enabled
)
1300 if (!fw_dump
.fadump_supported
) {
1301 printk(KERN_ERR
"Firmware-assisted dump is not supported on"
1302 " this hardware\n");
1306 fadump_show_config();
1308 * If dump data is available then see if it is valid and prepare for
1309 * saving it to the disk.
1311 if (fw_dump
.dump_active
) {
1313 * if dump process fails then invalidate the registration
1314 * and release memory before proceeding for re-registration.
1316 if (process_fadump(fdm_active
) < 0)
1317 fadump_invalidate_release_mem();
1319 /* Initialize the kernel dump memory structure for FAD registration. */
1320 else if (fw_dump
.reserve_dump_area_size
)
1321 init_fadump_mem_struct(&fdm
, fw_dump
.reserve_dump_area_start
);
1322 fadump_init_files();
1326 subsys_initcall(setup_fadump
);