xref: /openbmc/linux/arch/powerpc/kernel/fadump.c (revision 7bcae826)
1 /*
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
7  * Manish Ahuja
8  *
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.
13  *
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.
18  *
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.
22  *
23  * Copyright 2011 IBM Corporation
24  * Author: Mahesh Salgaonkar <mahesh@linux.vnet.ibm.com>
25  */
26 
27 #undef DEBUG
28 #define pr_fmt(fmt) "fadump: " fmt
29 
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 
39 #include <asm/page.h>
40 #include <asm/prom.h>
41 #include <asm/rtas.h>
42 #include <asm/fadump.h>
43 #include <asm/debug.h>
44 #include <asm/setup.h>
45 
46 static struct fw_dump fw_dump;
47 static struct fadump_mem_struct fdm;
48 static const struct fadump_mem_struct *fdm_active;
49 
50 static DEFINE_MUTEX(fadump_mutex);
51 struct fad_crash_memory_ranges crash_memory_ranges[INIT_CRASHMEM_RANGES];
52 int crash_mem_ranges;
53 
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)
57 {
58 	const __be32 *sections;
59 	int i, num_sections;
60 	int size;
61 	const __be32 *token;
62 
63 	if (depth != 1 || strcmp(uname, "rtas") != 0)
64 		return 0;
65 
66 	/*
67 	 * Check if Firmware Assisted dump is supported. if yes, check
68 	 * if dump has been initiated on last reboot.
69 	 */
70 	token = of_get_flat_dt_prop(node, "ibm,configure-kernel-dump", NULL);
71 	if (!token)
72 		return 1;
73 
74 	fw_dump.fadump_supported = 1;
75 	fw_dump.ibm_configure_kernel_dump = be32_to_cpu(*token);
76 
77 	/*
78 	 * The 'ibm,kernel-dump' rtas node is present only if there is
79 	 * dump data waiting for us.
80 	 */
81 	fdm_active = of_get_flat_dt_prop(node, "ibm,kernel-dump", NULL);
82 	if (fdm_active)
83 		fw_dump.dump_active = 1;
84 
85 	/* Get the sizes required to store dump data for the firmware provided
86 	 * dump sections.
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.
90 	 */
91 	sections = of_get_flat_dt_prop(node, "ibm,configure-kernel-dump-sizes",
92 					&size);
93 
94 	if (!sections)
95 		return 1;
96 
97 	num_sections = size / (3 * sizeof(u32));
98 
99 	for (i = 0; i < num_sections; i++, sections += 3) {
100 		u32 type = (u32)of_read_number(sections, 1);
101 
102 		switch (type) {
103 		case FADUMP_CPU_STATE_DATA:
104 			fw_dump.cpu_state_data_size =
105 					of_read_ulong(&sections[1], 2);
106 			break;
107 		case FADUMP_HPTE_REGION:
108 			fw_dump.hpte_region_size =
109 					of_read_ulong(&sections[1], 2);
110 			break;
111 		}
112 	}
113 
114 	return 1;
115 }
116 
117 int is_fadump_active(void)
118 {
119 	return fw_dump.dump_active;
120 }
121 
122 /* Print firmware assisted dump configurations for debugging purpose. */
123 static void fadump_show_config(void)
124 {
125 	pr_debug("Support for firmware-assisted dump (fadump): %s\n",
126 			(fw_dump.fadump_supported ? "present" : "no support"));
127 
128 	if (!fw_dump.fadump_supported)
129 		return;
130 
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);
139 }
140 
141 static unsigned long init_fadump_mem_struct(struct fadump_mem_struct *fdm,
142 				unsigned long addr)
143 {
144 	if (!fdm)
145 		return 0;
146 
147 	memset(fdm, 0, sizeof(struct fadump_mem_struct));
148 	addr = addr & PAGE_MASK;
149 
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));
155 
156 	/*
157 	 * Fields for disk dump option.
158 	 * We are not using disk dump option, hence set these fields to 0.
159 	 */
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;
164 
165 	/* set 0 to disable an automatic dump-reboot. */
166 	fdm->header.max_time_auto = 0;
167 
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;
176 
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;
184 
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;
192 
193 	return addr;
194 }
195 
196 /**
197  * fadump_calculate_reserve_size(): reserve variable boot area 5% of System RAM
198  *
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.
202  *
203  * This function has been taken from phyp-assisted dump feature implementation.
204  *
205  * returns larger of 256MB or 5% rounded down to multiples of 256MB.
206  *
207  * TODO: Come up with better approach to find out more accurate memory size
208  * that is required for a kernel to boot successfully.
209  *
210  */
211 static inline unsigned long fadump_calculate_reserve_size(void)
212 {
213 	unsigned long size;
214 
215 	/*
216 	 * Check if the size is specified through fadump_reserve_mem= cmdline
217 	 * option. If yes, then use that.
218 	 */
219 	if (fw_dump.reserve_bootvar)
220 		return fw_dump.reserve_bootvar;
221 
222 	/* divide by 20 to get 5% of value */
223 	size = memblock_end_of_DRAM() / 20;
224 
225 	/* round it down in multiples of 256 */
226 	size = size & ~0x0FFFFFFFUL;
227 
228 	/* Truncate to memory_limit. We don't want to over reserve the memory.*/
229 	if (memory_limit && size > memory_limit)
230 		size = memory_limit;
231 
232 	return (size > MIN_BOOT_MEM ? size : MIN_BOOT_MEM);
233 }
234 
235 /*
236  * Calculate the total memory size required to be reserved for
237  * firmware-assisted dump registration.
238  */
239 static unsigned long get_fadump_area_size(void)
240 {
241 	unsigned long size = 0;
242 
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);
251 
252 	size = PAGE_ALIGN(size);
253 	return size;
254 }
255 
256 int __init fadump_reserve_mem(void)
257 {
258 	unsigned long base, size, memory_boundary;
259 
260 	if (!fw_dump.fadump_enabled)
261 		return 0;
262 
263 	if (!fw_dump.fadump_supported) {
264 		printk(KERN_INFO "Firmware-assisted dump is not supported on"
265 				" this hardware\n");
266 		fw_dump.fadump_enabled = 0;
267 		return 0;
268 	}
269 	/*
270 	 * Initialize boot memory size
271 	 * If dump is active then we have already calculated the size during
272 	 * first kernel.
273 	 */
274 	if (fdm_active)
275 		fw_dump.boot_memory_size = be64_to_cpu(fdm_active->rmr_region.source_len);
276 	else
277 		fw_dump.boot_memory_size = fadump_calculate_reserve_size();
278 
279 	/*
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.
285 	 */
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;
290 		else
291 			memory_limit = memblock_end_of_DRAM();
292 		printk(KERN_INFO "Adjusted memory_limit for firmware-assisted"
293 				" dump, now %#016llx\n", memory_limit);
294 	}
295 	if (memory_limit)
296 		memory_boundary = memory_limit;
297 	else
298 		memory_boundary = memblock_end_of_DRAM();
299 
300 	if (fw_dump.dump_active) {
301 		printk(KERN_INFO "Firmware-assisted dump is active.\n");
302 		/*
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.
307 		 */
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));
315 
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);
321 	} else {
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));
330 	}
331 	fw_dump.reserve_dump_area_start = base;
332 	fw_dump.reserve_dump_area_size = size;
333 	return 1;
334 }
335 
336 unsigned long __init arch_reserved_kernel_pages(void)
337 {
338 	return memblock_reserved_size() / PAGE_SIZE;
339 }
340 
341 /* Look for fadump= cmdline option. */
342 static int __init early_fadump_param(char *p)
343 {
344 	if (!p)
345 		return 1;
346 
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;
351 
352 	return 0;
353 }
354 early_param("fadump", early_fadump_param);
355 
356 /* Look for fadump_reserve_mem= cmdline option */
357 static int __init early_fadump_reserve_mem(char *p)
358 {
359 	if (p)
360 		fw_dump.reserve_bootvar = memparse(p, &p);
361 	return 0;
362 }
363 early_param("fadump_reserve_mem", early_fadump_reserve_mem);
364 
365 static void register_fw_dump(struct fadump_mem_struct *fdm)
366 {
367 	int rc;
368 	unsigned int wait_time;
369 
370 	pr_debug("Registering for firmware-assisted kernel dump...\n");
371 
372 	/* TODO: Add upper time limit for the delay */
373 	do {
374 		rc = rtas_call(fw_dump.ibm_configure_kernel_dump, 3, 1, NULL,
375 			FADUMP_REGISTER, fdm,
376 			sizeof(struct fadump_mem_struct));
377 
378 		wait_time = rtas_busy_delay_time(rc);
379 		if (wait_time)
380 			mdelay(wait_time);
381 
382 	} while (wait_time);
383 
384 	switch (rc) {
385 	case -1:
386 		printk(KERN_ERR "Failed to register firmware-assisted kernel"
387 			" dump. Hardware Error(%d).\n", rc);
388 		break;
389 	case -3:
390 		printk(KERN_ERR "Failed to register firmware-assisted kernel"
391 			" dump. Parameter Error(%d).\n", rc);
392 		break;
393 	case -9:
394 		printk(KERN_ERR "firmware-assisted kernel dump is already "
395 			" registered.");
396 		fw_dump.dump_registered = 1;
397 		break;
398 	case 0:
399 		printk(KERN_INFO "firmware-assisted kernel dump registration"
400 			" is successful\n");
401 		fw_dump.dump_registered = 1;
402 		break;
403 	}
404 }
405 
406 void crash_fadump(struct pt_regs *regs, const char *str)
407 {
408 	struct fadump_crash_info_header *fdh = NULL;
409 	int old_cpu, this_cpu;
410 
411 	if (!fw_dump.dump_registered || !fw_dump.fadumphdr_addr)
412 		return;
413 
414 	/*
415 	 * old_cpu == -1 means this is the first CPU which has come here,
416 	 * go ahead and trigger fadump.
417 	 *
418 	 * old_cpu != -1 means some other CPU has already on it's way
419 	 * to trigger fadump, just keep looping here.
420 	 */
421 	this_cpu = smp_processor_id();
422 	old_cpu = cmpxchg(&crashing_cpu, -1, this_cpu);
423 
424 	if (old_cpu != -1) {
425 		/*
426 		 * We can't loop here indefinitely. Wait as long as fadump
427 		 * is in force. If we race with fadump un-registration this
428 		 * loop will break and then we go down to normal panic path
429 		 * and reboot. If fadump is in force the first crashing
430 		 * cpu will definitely trigger fadump.
431 		 */
432 		while (fw_dump.dump_registered)
433 			cpu_relax();
434 		return;
435 	}
436 
437 	fdh = __va(fw_dump.fadumphdr_addr);
438 	fdh->crashing_cpu = crashing_cpu;
439 	crash_save_vmcoreinfo();
440 
441 	if (regs)
442 		fdh->regs = *regs;
443 	else
444 		ppc_save_regs(&fdh->regs);
445 
446 	fdh->online_mask = *cpu_online_mask;
447 
448 	/* Call ibm,os-term rtas call to trigger firmware assisted dump */
449 	rtas_os_term((char *)str);
450 }
451 
452 #define GPR_MASK	0xffffff0000000000
453 static inline int fadump_gpr_index(u64 id)
454 {
455 	int i = -1;
456 	char str[3];
457 
458 	if ((id & GPR_MASK) == REG_ID("GPR")) {
459 		/* get the digits at the end */
460 		id &= ~GPR_MASK;
461 		id >>= 24;
462 		str[2] = '\0';
463 		str[1] = id & 0xff;
464 		str[0] = (id >> 8) & 0xff;
465 		sscanf(str, "%d", &i);
466 		if (i > 31)
467 			i = -1;
468 	}
469 	return i;
470 }
471 
472 static inline void fadump_set_regval(struct pt_regs *regs, u64 reg_id,
473 								u64 reg_val)
474 {
475 	int i;
476 
477 	i = fadump_gpr_index(reg_id);
478 	if (i >= 0)
479 		regs->gpr[i] = (unsigned long)reg_val;
480 	else if (reg_id == REG_ID("NIA"))
481 		regs->nip = (unsigned long)reg_val;
482 	else if (reg_id == REG_ID("MSR"))
483 		regs->msr = (unsigned long)reg_val;
484 	else if (reg_id == REG_ID("CTR"))
485 		regs->ctr = (unsigned long)reg_val;
486 	else if (reg_id == REG_ID("LR"))
487 		regs->link = (unsigned long)reg_val;
488 	else if (reg_id == REG_ID("XER"))
489 		regs->xer = (unsigned long)reg_val;
490 	else if (reg_id == REG_ID("CR"))
491 		regs->ccr = (unsigned long)reg_val;
492 	else if (reg_id == REG_ID("DAR"))
493 		regs->dar = (unsigned long)reg_val;
494 	else if (reg_id == REG_ID("DSISR"))
495 		regs->dsisr = (unsigned long)reg_val;
496 }
497 
498 static struct fadump_reg_entry*
499 fadump_read_registers(struct fadump_reg_entry *reg_entry, struct pt_regs *regs)
500 {
501 	memset(regs, 0, sizeof(struct pt_regs));
502 
503 	while (be64_to_cpu(reg_entry->reg_id) != REG_ID("CPUEND")) {
504 		fadump_set_regval(regs, be64_to_cpu(reg_entry->reg_id),
505 					be64_to_cpu(reg_entry->reg_value));
506 		reg_entry++;
507 	}
508 	reg_entry++;
509 	return reg_entry;
510 }
511 
512 static u32 *fadump_append_elf_note(u32 *buf, char *name, unsigned type,
513 						void *data, size_t data_len)
514 {
515 	struct elf_note note;
516 
517 	note.n_namesz = strlen(name) + 1;
518 	note.n_descsz = data_len;
519 	note.n_type   = type;
520 	memcpy(buf, &note, sizeof(note));
521 	buf += (sizeof(note) + 3)/4;
522 	memcpy(buf, name, note.n_namesz);
523 	buf += (note.n_namesz + 3)/4;
524 	memcpy(buf, data, note.n_descsz);
525 	buf += (note.n_descsz + 3)/4;
526 
527 	return buf;
528 }
529 
530 static void fadump_final_note(u32 *buf)
531 {
532 	struct elf_note note;
533 
534 	note.n_namesz = 0;
535 	note.n_descsz = 0;
536 	note.n_type   = 0;
537 	memcpy(buf, &note, sizeof(note));
538 }
539 
540 static u32 *fadump_regs_to_elf_notes(u32 *buf, struct pt_regs *regs)
541 {
542 	struct elf_prstatus prstatus;
543 
544 	memset(&prstatus, 0, sizeof(prstatus));
545 	/*
546 	 * FIXME: How do i get PID? Do I really need it?
547 	 * prstatus.pr_pid = ????
548 	 */
549 	elf_core_copy_kernel_regs(&prstatus.pr_reg, regs);
550 	buf = fadump_append_elf_note(buf, KEXEC_CORE_NOTE_NAME, NT_PRSTATUS,
551 				&prstatus, sizeof(prstatus));
552 	return buf;
553 }
554 
555 static void fadump_update_elfcore_header(char *bufp)
556 {
557 	struct elfhdr *elf;
558 	struct elf_phdr *phdr;
559 
560 	elf = (struct elfhdr *)bufp;
561 	bufp += sizeof(struct elfhdr);
562 
563 	/* First note is a place holder for cpu notes info. */
564 	phdr = (struct elf_phdr *)bufp;
565 
566 	if (phdr->p_type == PT_NOTE) {
567 		phdr->p_paddr = fw_dump.cpu_notes_buf;
568 		phdr->p_offset	= phdr->p_paddr;
569 		phdr->p_filesz	= fw_dump.cpu_notes_buf_size;
570 		phdr->p_memsz = fw_dump.cpu_notes_buf_size;
571 	}
572 	return;
573 }
574 
575 static void *fadump_cpu_notes_buf_alloc(unsigned long size)
576 {
577 	void *vaddr;
578 	struct page *page;
579 	unsigned long order, count, i;
580 
581 	order = get_order(size);
582 	vaddr = (void *)__get_free_pages(GFP_KERNEL|__GFP_ZERO, order);
583 	if (!vaddr)
584 		return NULL;
585 
586 	count = 1 << order;
587 	page = virt_to_page(vaddr);
588 	for (i = 0; i < count; i++)
589 		SetPageReserved(page + i);
590 	return vaddr;
591 }
592 
593 static void fadump_cpu_notes_buf_free(unsigned long vaddr, unsigned long size)
594 {
595 	struct page *page;
596 	unsigned long order, count, i;
597 
598 	order = get_order(size);
599 	count = 1 << order;
600 	page = virt_to_page(vaddr);
601 	for (i = 0; i < count; i++)
602 		ClearPageReserved(page + i);
603 	__free_pages(page, order);
604 }
605 
606 /*
607  * Read CPU state dump data and convert it into ELF notes.
608  * The CPU dump starts with magic number "REGSAVE". NumCpusOffset should be
609  * used to access the data to allow for additional fields to be added without
610  * affecting compatibility. Each list of registers for a CPU starts with
611  * "CPUSTRT" and ends with "CPUEND". Each register entry is of 16 bytes,
612  * 8 Byte ASCII identifier and 8 Byte register value. The register entry
613  * with identifier "CPUSTRT" and "CPUEND" contains 4 byte cpu id as part
614  * of register value. For more details refer to PAPR document.
615  *
616  * Only for the crashing cpu we ignore the CPU dump data and get exact
617  * state from fadump crash info structure populated by first kernel at the
618  * time of crash.
619  */
620 static int __init fadump_build_cpu_notes(const struct fadump_mem_struct *fdm)
621 {
622 	struct fadump_reg_save_area_header *reg_header;
623 	struct fadump_reg_entry *reg_entry;
624 	struct fadump_crash_info_header *fdh = NULL;
625 	void *vaddr;
626 	unsigned long addr;
627 	u32 num_cpus, *note_buf;
628 	struct pt_regs regs;
629 	int i, rc = 0, cpu = 0;
630 
631 	if (!fdm->cpu_state_data.bytes_dumped)
632 		return -EINVAL;
633 
634 	addr = be64_to_cpu(fdm->cpu_state_data.destination_address);
635 	vaddr = __va(addr);
636 
637 	reg_header = vaddr;
638 	if (be64_to_cpu(reg_header->magic_number) != REGSAVE_AREA_MAGIC) {
639 		printk(KERN_ERR "Unable to read register save area.\n");
640 		return -ENOENT;
641 	}
642 	pr_debug("--------CPU State Data------------\n");
643 	pr_debug("Magic Number: %llx\n", be64_to_cpu(reg_header->magic_number));
644 	pr_debug("NumCpuOffset: %x\n", be32_to_cpu(reg_header->num_cpu_offset));
645 
646 	vaddr += be32_to_cpu(reg_header->num_cpu_offset);
647 	num_cpus = be32_to_cpu(*((__be32 *)(vaddr)));
648 	pr_debug("NumCpus     : %u\n", num_cpus);
649 	vaddr += sizeof(u32);
650 	reg_entry = (struct fadump_reg_entry *)vaddr;
651 
652 	/* Allocate buffer to hold cpu crash notes. */
653 	fw_dump.cpu_notes_buf_size = num_cpus * sizeof(note_buf_t);
654 	fw_dump.cpu_notes_buf_size = PAGE_ALIGN(fw_dump.cpu_notes_buf_size);
655 	note_buf = fadump_cpu_notes_buf_alloc(fw_dump.cpu_notes_buf_size);
656 	if (!note_buf) {
657 		printk(KERN_ERR "Failed to allocate 0x%lx bytes for "
658 			"cpu notes buffer\n", fw_dump.cpu_notes_buf_size);
659 		return -ENOMEM;
660 	}
661 	fw_dump.cpu_notes_buf = __pa(note_buf);
662 
663 	pr_debug("Allocated buffer for cpu notes of size %ld at %p\n",
664 			(num_cpus * sizeof(note_buf_t)), note_buf);
665 
666 	if (fw_dump.fadumphdr_addr)
667 		fdh = __va(fw_dump.fadumphdr_addr);
668 
669 	for (i = 0; i < num_cpus; i++) {
670 		if (be64_to_cpu(reg_entry->reg_id) != REG_ID("CPUSTRT")) {
671 			printk(KERN_ERR "Unable to read CPU state data\n");
672 			rc = -ENOENT;
673 			goto error_out;
674 		}
675 		/* Lower 4 bytes of reg_value contains logical cpu id */
676 		cpu = be64_to_cpu(reg_entry->reg_value) & FADUMP_CPU_ID_MASK;
677 		if (fdh && !cpumask_test_cpu(cpu, &fdh->online_mask)) {
678 			SKIP_TO_NEXT_CPU(reg_entry);
679 			continue;
680 		}
681 		pr_debug("Reading register data for cpu %d...\n", cpu);
682 		if (fdh && fdh->crashing_cpu == cpu) {
683 			regs = fdh->regs;
684 			note_buf = fadump_regs_to_elf_notes(note_buf, &regs);
685 			SKIP_TO_NEXT_CPU(reg_entry);
686 		} else {
687 			reg_entry++;
688 			reg_entry = fadump_read_registers(reg_entry, &regs);
689 			note_buf = fadump_regs_to_elf_notes(note_buf, &regs);
690 		}
691 	}
692 	fadump_final_note(note_buf);
693 
694 	if (fdh) {
695 		pr_debug("Updating elfcore header (%llx) with cpu notes\n",
696 							fdh->elfcorehdr_addr);
697 		fadump_update_elfcore_header((char *)__va(fdh->elfcorehdr_addr));
698 	}
699 	return 0;
700 
701 error_out:
702 	fadump_cpu_notes_buf_free((unsigned long)__va(fw_dump.cpu_notes_buf),
703 					fw_dump.cpu_notes_buf_size);
704 	fw_dump.cpu_notes_buf = 0;
705 	fw_dump.cpu_notes_buf_size = 0;
706 	return rc;
707 
708 }
709 
710 /*
711  * Validate and process the dump data stored by firmware before exporting
712  * it through '/proc/vmcore'.
713  */
714 static int __init process_fadump(const struct fadump_mem_struct *fdm_active)
715 {
716 	struct fadump_crash_info_header *fdh;
717 	int rc = 0;
718 
719 	if (!fdm_active || !fw_dump.fadumphdr_addr)
720 		return -EINVAL;
721 
722 	/* Check if the dump data is valid. */
723 	if ((be16_to_cpu(fdm_active->header.dump_status_flag) == FADUMP_ERROR_FLAG) ||
724 			(fdm_active->cpu_state_data.error_flags != 0) ||
725 			(fdm_active->rmr_region.error_flags != 0)) {
726 		printk(KERN_ERR "Dump taken by platform is not valid\n");
727 		return -EINVAL;
728 	}
729 	if ((fdm_active->rmr_region.bytes_dumped !=
730 			fdm_active->rmr_region.source_len) ||
731 			!fdm_active->cpu_state_data.bytes_dumped) {
732 		printk(KERN_ERR "Dump taken by platform is incomplete\n");
733 		return -EINVAL;
734 	}
735 
736 	/* Validate the fadump crash info header */
737 	fdh = __va(fw_dump.fadumphdr_addr);
738 	if (fdh->magic_number != FADUMP_CRASH_INFO_MAGIC) {
739 		printk(KERN_ERR "Crash info header is not valid.\n");
740 		return -EINVAL;
741 	}
742 
743 	rc = fadump_build_cpu_notes(fdm_active);
744 	if (rc)
745 		return rc;
746 
747 	/*
748 	 * We are done validating dump info and elfcore header is now ready
749 	 * to be exported. set elfcorehdr_addr so that vmcore module will
750 	 * export the elfcore header through '/proc/vmcore'.
751 	 */
752 	elfcorehdr_addr = fdh->elfcorehdr_addr;
753 
754 	return 0;
755 }
756 
757 static inline void fadump_add_crash_memory(unsigned long long base,
758 					unsigned long long end)
759 {
760 	if (base == end)
761 		return;
762 
763 	pr_debug("crash_memory_range[%d] [%#016llx-%#016llx], %#llx bytes\n",
764 		crash_mem_ranges, base, end - 1, (end - base));
765 	crash_memory_ranges[crash_mem_ranges].base = base;
766 	crash_memory_ranges[crash_mem_ranges].size = end - base;
767 	crash_mem_ranges++;
768 }
769 
770 static void fadump_exclude_reserved_area(unsigned long long start,
771 					unsigned long long end)
772 {
773 	unsigned long long ra_start, ra_end;
774 
775 	ra_start = fw_dump.reserve_dump_area_start;
776 	ra_end = ra_start + fw_dump.reserve_dump_area_size;
777 
778 	if ((ra_start < end) && (ra_end > start)) {
779 		if ((start < ra_start) && (end > ra_end)) {
780 			fadump_add_crash_memory(start, ra_start);
781 			fadump_add_crash_memory(ra_end, end);
782 		} else if (start < ra_start) {
783 			fadump_add_crash_memory(start, ra_start);
784 		} else if (ra_end < end) {
785 			fadump_add_crash_memory(ra_end, end);
786 		}
787 	} else
788 		fadump_add_crash_memory(start, end);
789 }
790 
791 static int fadump_init_elfcore_header(char *bufp)
792 {
793 	struct elfhdr *elf;
794 
795 	elf = (struct elfhdr *) bufp;
796 	bufp += sizeof(struct elfhdr);
797 	memcpy(elf->e_ident, ELFMAG, SELFMAG);
798 	elf->e_ident[EI_CLASS] = ELF_CLASS;
799 	elf->e_ident[EI_DATA] = ELF_DATA;
800 	elf->e_ident[EI_VERSION] = EV_CURRENT;
801 	elf->e_ident[EI_OSABI] = ELF_OSABI;
802 	memset(elf->e_ident+EI_PAD, 0, EI_NIDENT-EI_PAD);
803 	elf->e_type = ET_CORE;
804 	elf->e_machine = ELF_ARCH;
805 	elf->e_version = EV_CURRENT;
806 	elf->e_entry = 0;
807 	elf->e_phoff = sizeof(struct elfhdr);
808 	elf->e_shoff = 0;
809 #if defined(_CALL_ELF)
810 	elf->e_flags = _CALL_ELF;
811 #else
812 	elf->e_flags = 0;
813 #endif
814 	elf->e_ehsize = sizeof(struct elfhdr);
815 	elf->e_phentsize = sizeof(struct elf_phdr);
816 	elf->e_phnum = 0;
817 	elf->e_shentsize = 0;
818 	elf->e_shnum = 0;
819 	elf->e_shstrndx = 0;
820 
821 	return 0;
822 }
823 
824 /*
825  * Traverse through memblock structure and setup crash memory ranges. These
826  * ranges will be used create PT_LOAD program headers in elfcore header.
827  */
828 static void fadump_setup_crash_memory_ranges(void)
829 {
830 	struct memblock_region *reg;
831 	unsigned long long start, end;
832 
833 	pr_debug("Setup crash memory ranges.\n");
834 	crash_mem_ranges = 0;
835 	/*
836 	 * add the first memory chunk (RMA_START through boot_memory_size) as
837 	 * a separate memory chunk. The reason is, at the time crash firmware
838 	 * will move the content of this memory chunk to different location
839 	 * specified during fadump registration. We need to create a separate
840 	 * program header for this chunk with the correct offset.
841 	 */
842 	fadump_add_crash_memory(RMA_START, fw_dump.boot_memory_size);
843 
844 	for_each_memblock(memory, reg) {
845 		start = (unsigned long long)reg->base;
846 		end = start + (unsigned long long)reg->size;
847 		if (start == RMA_START && end >= fw_dump.boot_memory_size)
848 			start = fw_dump.boot_memory_size;
849 
850 		/* add this range excluding the reserved dump area. */
851 		fadump_exclude_reserved_area(start, end);
852 	}
853 }
854 
855 /*
856  * If the given physical address falls within the boot memory region then
857  * return the relocated address that points to the dump region reserved
858  * for saving initial boot memory contents.
859  */
860 static inline unsigned long fadump_relocate(unsigned long paddr)
861 {
862 	if (paddr > RMA_START && paddr < fw_dump.boot_memory_size)
863 		return be64_to_cpu(fdm.rmr_region.destination_address) + paddr;
864 	else
865 		return paddr;
866 }
867 
868 static int fadump_create_elfcore_headers(char *bufp)
869 {
870 	struct elfhdr *elf;
871 	struct elf_phdr *phdr;
872 	int i;
873 
874 	fadump_init_elfcore_header(bufp);
875 	elf = (struct elfhdr *)bufp;
876 	bufp += sizeof(struct elfhdr);
877 
878 	/*
879 	 * setup ELF PT_NOTE, place holder for cpu notes info. The notes info
880 	 * will be populated during second kernel boot after crash. Hence
881 	 * this PT_NOTE will always be the first elf note.
882 	 *
883 	 * NOTE: Any new ELF note addition should be placed after this note.
884 	 */
885 	phdr = (struct elf_phdr *)bufp;
886 	bufp += sizeof(struct elf_phdr);
887 	phdr->p_type = PT_NOTE;
888 	phdr->p_flags = 0;
889 	phdr->p_vaddr = 0;
890 	phdr->p_align = 0;
891 
892 	phdr->p_offset = 0;
893 	phdr->p_paddr = 0;
894 	phdr->p_filesz = 0;
895 	phdr->p_memsz = 0;
896 
897 	(elf->e_phnum)++;
898 
899 	/* setup ELF PT_NOTE for vmcoreinfo */
900 	phdr = (struct elf_phdr *)bufp;
901 	bufp += sizeof(struct elf_phdr);
902 	phdr->p_type	= PT_NOTE;
903 	phdr->p_flags	= 0;
904 	phdr->p_vaddr	= 0;
905 	phdr->p_align	= 0;
906 
907 	phdr->p_paddr	= fadump_relocate(paddr_vmcoreinfo_note());
908 	phdr->p_offset	= phdr->p_paddr;
909 	phdr->p_memsz	= vmcoreinfo_max_size;
910 	phdr->p_filesz	= vmcoreinfo_max_size;
911 
912 	/* Increment number of program headers. */
913 	(elf->e_phnum)++;
914 
915 	/* setup PT_LOAD sections. */
916 
917 	for (i = 0; i < crash_mem_ranges; i++) {
918 		unsigned long long mbase, msize;
919 		mbase = crash_memory_ranges[i].base;
920 		msize = crash_memory_ranges[i].size;
921 
922 		if (!msize)
923 			continue;
924 
925 		phdr = (struct elf_phdr *)bufp;
926 		bufp += sizeof(struct elf_phdr);
927 		phdr->p_type	= PT_LOAD;
928 		phdr->p_flags	= PF_R|PF_W|PF_X;
929 		phdr->p_offset	= mbase;
930 
931 		if (mbase == RMA_START) {
932 			/*
933 			 * The entire RMA region will be moved by firmware
934 			 * to the specified destination_address. Hence set
935 			 * the correct offset.
936 			 */
937 			phdr->p_offset = be64_to_cpu(fdm.rmr_region.destination_address);
938 		}
939 
940 		phdr->p_paddr = mbase;
941 		phdr->p_vaddr = (unsigned long)__va(mbase);
942 		phdr->p_filesz = msize;
943 		phdr->p_memsz = msize;
944 		phdr->p_align = 0;
945 
946 		/* Increment number of program headers. */
947 		(elf->e_phnum)++;
948 	}
949 	return 0;
950 }
951 
952 static unsigned long init_fadump_header(unsigned long addr)
953 {
954 	struct fadump_crash_info_header *fdh;
955 
956 	if (!addr)
957 		return 0;
958 
959 	fw_dump.fadumphdr_addr = addr;
960 	fdh = __va(addr);
961 	addr += sizeof(struct fadump_crash_info_header);
962 
963 	memset(fdh, 0, sizeof(struct fadump_crash_info_header));
964 	fdh->magic_number = FADUMP_CRASH_INFO_MAGIC;
965 	fdh->elfcorehdr_addr = addr;
966 	/* We will set the crashing cpu id in crash_fadump() during crash. */
967 	fdh->crashing_cpu = CPU_UNKNOWN;
968 
969 	return addr;
970 }
971 
972 static void register_fadump(void)
973 {
974 	unsigned long addr;
975 	void *vaddr;
976 
977 	/*
978 	 * If no memory is reserved then we can not register for firmware-
979 	 * assisted dump.
980 	 */
981 	if (!fw_dump.reserve_dump_area_size)
982 		return;
983 
984 	fadump_setup_crash_memory_ranges();
985 
986 	addr = be64_to_cpu(fdm.rmr_region.destination_address) + be64_to_cpu(fdm.rmr_region.source_len);
987 	/* Initialize fadump crash info header. */
988 	addr = init_fadump_header(addr);
989 	vaddr = __va(addr);
990 
991 	pr_debug("Creating ELF core headers at %#016lx\n", addr);
992 	fadump_create_elfcore_headers(vaddr);
993 
994 	/* register the future kernel dump with firmware. */
995 	register_fw_dump(&fdm);
996 }
997 
998 static int fadump_unregister_dump(struct fadump_mem_struct *fdm)
999 {
1000 	int rc = 0;
1001 	unsigned int wait_time;
1002 
1003 	pr_debug("Un-register firmware-assisted dump\n");
1004 
1005 	/* TODO: Add upper time limit for the delay */
1006 	do {
1007 		rc = rtas_call(fw_dump.ibm_configure_kernel_dump, 3, 1, NULL,
1008 			FADUMP_UNREGISTER, fdm,
1009 			sizeof(struct fadump_mem_struct));
1010 
1011 		wait_time = rtas_busy_delay_time(rc);
1012 		if (wait_time)
1013 			mdelay(wait_time);
1014 	} while (wait_time);
1015 
1016 	if (rc) {
1017 		printk(KERN_ERR "Failed to un-register firmware-assisted dump."
1018 			" unexpected error(%d).\n", rc);
1019 		return rc;
1020 	}
1021 	fw_dump.dump_registered = 0;
1022 	return 0;
1023 }
1024 
1025 static int fadump_invalidate_dump(struct fadump_mem_struct *fdm)
1026 {
1027 	int rc = 0;
1028 	unsigned int wait_time;
1029 
1030 	pr_debug("Invalidating firmware-assisted dump registration\n");
1031 
1032 	/* TODO: Add upper time limit for the delay */
1033 	do {
1034 		rc = rtas_call(fw_dump.ibm_configure_kernel_dump, 3, 1, NULL,
1035 			FADUMP_INVALIDATE, fdm,
1036 			sizeof(struct fadump_mem_struct));
1037 
1038 		wait_time = rtas_busy_delay_time(rc);
1039 		if (wait_time)
1040 			mdelay(wait_time);
1041 	} while (wait_time);
1042 
1043 	if (rc) {
1044 		pr_err("Failed to invalidate firmware-assisted dump registration. Unexpected error (%d).\n", rc);
1045 		return rc;
1046 	}
1047 	fw_dump.dump_active = 0;
1048 	fdm_active = NULL;
1049 	return 0;
1050 }
1051 
1052 void fadump_cleanup(void)
1053 {
1054 	/* Invalidate the registration only if dump is active. */
1055 	if (fw_dump.dump_active) {
1056 		init_fadump_mem_struct(&fdm,
1057 			be64_to_cpu(fdm_active->cpu_state_data.destination_address));
1058 		fadump_invalidate_dump(&fdm);
1059 	}
1060 }
1061 
1062 /*
1063  * Release the memory that was reserved in early boot to preserve the memory
1064  * contents. The released memory will be available for general use.
1065  */
1066 static void fadump_release_memory(unsigned long begin, unsigned long end)
1067 {
1068 	unsigned long addr;
1069 	unsigned long ra_start, ra_end;
1070 
1071 	ra_start = fw_dump.reserve_dump_area_start;
1072 	ra_end = ra_start + fw_dump.reserve_dump_area_size;
1073 
1074 	for (addr = begin; addr < end; addr += PAGE_SIZE) {
1075 		/*
1076 		 * exclude the dump reserve area. Will reuse it for next
1077 		 * fadump registration.
1078 		 */
1079 		if (addr <= ra_end && ((addr + PAGE_SIZE) > ra_start))
1080 			continue;
1081 
1082 		free_reserved_page(pfn_to_page(addr >> PAGE_SHIFT));
1083 	}
1084 }
1085 
1086 static void fadump_invalidate_release_mem(void)
1087 {
1088 	unsigned long reserved_area_start, reserved_area_end;
1089 	unsigned long destination_address;
1090 
1091 	mutex_lock(&fadump_mutex);
1092 	if (!fw_dump.dump_active) {
1093 		mutex_unlock(&fadump_mutex);
1094 		return;
1095 	}
1096 
1097 	destination_address = be64_to_cpu(fdm_active->cpu_state_data.destination_address);
1098 	fadump_cleanup();
1099 	mutex_unlock(&fadump_mutex);
1100 
1101 	/*
1102 	 * Save the current reserved memory bounds we will require them
1103 	 * later for releasing the memory for general use.
1104 	 */
1105 	reserved_area_start = fw_dump.reserve_dump_area_start;
1106 	reserved_area_end = reserved_area_start +
1107 			fw_dump.reserve_dump_area_size;
1108 	/*
1109 	 * Setup reserve_dump_area_start and its size so that we can
1110 	 * reuse this reserved memory for Re-registration.
1111 	 */
1112 	fw_dump.reserve_dump_area_start = destination_address;
1113 	fw_dump.reserve_dump_area_size = get_fadump_area_size();
1114 
1115 	fadump_release_memory(reserved_area_start, reserved_area_end);
1116 	if (fw_dump.cpu_notes_buf) {
1117 		fadump_cpu_notes_buf_free(
1118 				(unsigned long)__va(fw_dump.cpu_notes_buf),
1119 				fw_dump.cpu_notes_buf_size);
1120 		fw_dump.cpu_notes_buf = 0;
1121 		fw_dump.cpu_notes_buf_size = 0;
1122 	}
1123 	/* Initialize the kernel dump memory structure for FAD registration. */
1124 	init_fadump_mem_struct(&fdm, fw_dump.reserve_dump_area_start);
1125 }
1126 
1127 static ssize_t fadump_release_memory_store(struct kobject *kobj,
1128 					struct kobj_attribute *attr,
1129 					const char *buf, size_t count)
1130 {
1131 	if (!fw_dump.dump_active)
1132 		return -EPERM;
1133 
1134 	if (buf[0] == '1') {
1135 		/*
1136 		 * Take away the '/proc/vmcore'. We are releasing the dump
1137 		 * memory, hence it will not be valid anymore.
1138 		 */
1139 #ifdef CONFIG_PROC_VMCORE
1140 		vmcore_cleanup();
1141 #endif
1142 		fadump_invalidate_release_mem();
1143 
1144 	} else
1145 		return -EINVAL;
1146 	return count;
1147 }
1148 
1149 static ssize_t fadump_enabled_show(struct kobject *kobj,
1150 					struct kobj_attribute *attr,
1151 					char *buf)
1152 {
1153 	return sprintf(buf, "%d\n", fw_dump.fadump_enabled);
1154 }
1155 
1156 static ssize_t fadump_register_show(struct kobject *kobj,
1157 					struct kobj_attribute *attr,
1158 					char *buf)
1159 {
1160 	return sprintf(buf, "%d\n", fw_dump.dump_registered);
1161 }
1162 
1163 static ssize_t fadump_register_store(struct kobject *kobj,
1164 					struct kobj_attribute *attr,
1165 					const char *buf, size_t count)
1166 {
1167 	int ret = 0;
1168 
1169 	if (!fw_dump.fadump_enabled || fdm_active)
1170 		return -EPERM;
1171 
1172 	mutex_lock(&fadump_mutex);
1173 
1174 	switch (buf[0]) {
1175 	case '0':
1176 		if (fw_dump.dump_registered == 0) {
1177 			ret = -EINVAL;
1178 			goto unlock_out;
1179 		}
1180 		/* Un-register Firmware-assisted dump */
1181 		fadump_unregister_dump(&fdm);
1182 		break;
1183 	case '1':
1184 		if (fw_dump.dump_registered == 1) {
1185 			ret = -EINVAL;
1186 			goto unlock_out;
1187 		}
1188 		/* Register Firmware-assisted dump */
1189 		register_fadump();
1190 		break;
1191 	default:
1192 		ret = -EINVAL;
1193 		break;
1194 	}
1195 
1196 unlock_out:
1197 	mutex_unlock(&fadump_mutex);
1198 	return ret < 0 ? ret : count;
1199 }
1200 
1201 static int fadump_region_show(struct seq_file *m, void *private)
1202 {
1203 	const struct fadump_mem_struct *fdm_ptr;
1204 
1205 	if (!fw_dump.fadump_enabled)
1206 		return 0;
1207 
1208 	mutex_lock(&fadump_mutex);
1209 	if (fdm_active)
1210 		fdm_ptr = fdm_active;
1211 	else {
1212 		mutex_unlock(&fadump_mutex);
1213 		fdm_ptr = &fdm;
1214 	}
1215 
1216 	seq_printf(m,
1217 			"CPU : [%#016llx-%#016llx] %#llx bytes, "
1218 			"Dumped: %#llx\n",
1219 			be64_to_cpu(fdm_ptr->cpu_state_data.destination_address),
1220 			be64_to_cpu(fdm_ptr->cpu_state_data.destination_address) +
1221 			be64_to_cpu(fdm_ptr->cpu_state_data.source_len) - 1,
1222 			be64_to_cpu(fdm_ptr->cpu_state_data.source_len),
1223 			be64_to_cpu(fdm_ptr->cpu_state_data.bytes_dumped));
1224 	seq_printf(m,
1225 			"HPTE: [%#016llx-%#016llx] %#llx bytes, "
1226 			"Dumped: %#llx\n",
1227 			be64_to_cpu(fdm_ptr->hpte_region.destination_address),
1228 			be64_to_cpu(fdm_ptr->hpte_region.destination_address) +
1229 			be64_to_cpu(fdm_ptr->hpte_region.source_len) - 1,
1230 			be64_to_cpu(fdm_ptr->hpte_region.source_len),
1231 			be64_to_cpu(fdm_ptr->hpte_region.bytes_dumped));
1232 	seq_printf(m,
1233 			"DUMP: [%#016llx-%#016llx] %#llx bytes, "
1234 			"Dumped: %#llx\n",
1235 			be64_to_cpu(fdm_ptr->rmr_region.destination_address),
1236 			be64_to_cpu(fdm_ptr->rmr_region.destination_address) +
1237 			be64_to_cpu(fdm_ptr->rmr_region.source_len) - 1,
1238 			be64_to_cpu(fdm_ptr->rmr_region.source_len),
1239 			be64_to_cpu(fdm_ptr->rmr_region.bytes_dumped));
1240 
1241 	if (!fdm_active ||
1242 		(fw_dump.reserve_dump_area_start ==
1243 		be64_to_cpu(fdm_ptr->cpu_state_data.destination_address)))
1244 		goto out;
1245 
1246 	/* Dump is active. Show reserved memory region. */
1247 	seq_printf(m,
1248 			"    : [%#016llx-%#016llx] %#llx bytes, "
1249 			"Dumped: %#llx\n",
1250 			(unsigned long long)fw_dump.reserve_dump_area_start,
1251 			be64_to_cpu(fdm_ptr->cpu_state_data.destination_address) - 1,
1252 			be64_to_cpu(fdm_ptr->cpu_state_data.destination_address) -
1253 			fw_dump.reserve_dump_area_start,
1254 			be64_to_cpu(fdm_ptr->cpu_state_data.destination_address) -
1255 			fw_dump.reserve_dump_area_start);
1256 out:
1257 	if (fdm_active)
1258 		mutex_unlock(&fadump_mutex);
1259 	return 0;
1260 }
1261 
1262 static struct kobj_attribute fadump_release_attr = __ATTR(fadump_release_mem,
1263 						0200, NULL,
1264 						fadump_release_memory_store);
1265 static struct kobj_attribute fadump_attr = __ATTR(fadump_enabled,
1266 						0444, fadump_enabled_show,
1267 						NULL);
1268 static struct kobj_attribute fadump_register_attr = __ATTR(fadump_registered,
1269 						0644, fadump_register_show,
1270 						fadump_register_store);
1271 
1272 static int fadump_region_open(struct inode *inode, struct file *file)
1273 {
1274 	return single_open(file, fadump_region_show, inode->i_private);
1275 }
1276 
1277 static const struct file_operations fadump_region_fops = {
1278 	.open    = fadump_region_open,
1279 	.read    = seq_read,
1280 	.llseek  = seq_lseek,
1281 	.release = single_release,
1282 };
1283 
1284 static void fadump_init_files(void)
1285 {
1286 	struct dentry *debugfs_file;
1287 	int rc = 0;
1288 
1289 	rc = sysfs_create_file(kernel_kobj, &fadump_attr.attr);
1290 	if (rc)
1291 		printk(KERN_ERR "fadump: unable to create sysfs file"
1292 			" fadump_enabled (%d)\n", rc);
1293 
1294 	rc = sysfs_create_file(kernel_kobj, &fadump_register_attr.attr);
1295 	if (rc)
1296 		printk(KERN_ERR "fadump: unable to create sysfs file"
1297 			" fadump_registered (%d)\n", rc);
1298 
1299 	debugfs_file = debugfs_create_file("fadump_region", 0444,
1300 					powerpc_debugfs_root, NULL,
1301 					&fadump_region_fops);
1302 	if (!debugfs_file)
1303 		printk(KERN_ERR "fadump: unable to create debugfs file"
1304 				" fadump_region\n");
1305 
1306 	if (fw_dump.dump_active) {
1307 		rc = sysfs_create_file(kernel_kobj, &fadump_release_attr.attr);
1308 		if (rc)
1309 			printk(KERN_ERR "fadump: unable to create sysfs file"
1310 				" fadump_release_mem (%d)\n", rc);
1311 	}
1312 	return;
1313 }
1314 
1315 /*
1316  * Prepare for firmware-assisted dump.
1317  */
1318 int __init setup_fadump(void)
1319 {
1320 	if (!fw_dump.fadump_enabled)
1321 		return 0;
1322 
1323 	if (!fw_dump.fadump_supported) {
1324 		printk(KERN_ERR "Firmware-assisted dump is not supported on"
1325 			" this hardware\n");
1326 		return 0;
1327 	}
1328 
1329 	fadump_show_config();
1330 	/*
1331 	 * If dump data is available then see if it is valid and prepare for
1332 	 * saving it to the disk.
1333 	 */
1334 	if (fw_dump.dump_active) {
1335 		/*
1336 		 * if dump process fails then invalidate the registration
1337 		 * and release memory before proceeding for re-registration.
1338 		 */
1339 		if (process_fadump(fdm_active) < 0)
1340 			fadump_invalidate_release_mem();
1341 	}
1342 	/* Initialize the kernel dump memory structure for FAD registration. */
1343 	else if (fw_dump.reserve_dump_area_size)
1344 		init_fadump_mem_struct(&fdm, fw_dump.reserve_dump_area_start);
1345 	fadump_init_files();
1346 
1347 	return 1;
1348 }
1349 subsys_initcall(setup_fadump);
1350