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
46static struct fw_dump fw_dump;
47static struct fadump_mem_struct fdm;
48static const struct fadump_mem_struct *fdm_active;
49
50static DEFINE_MUTEX(fadump_mutex);
51struct fad_crash_memory_ranges crash_memory_ranges[INIT_CRASHMEM_RANGES];
52int crash_mem_ranges;
53
54/* Scan the Firmware Assisted dump configuration details. */
55int __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
117int is_fadump_active(void)
118{
119	return fw_dump.dump_active;
120}
121
122/* Print firmware assisted dump configurations for debugging purpose. */
123static 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
141static 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 */
211static 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 */
239static 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
256int __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/* Look for fadump= cmdline option. */
337static int __init early_fadump_param(char *p)
338{
339	if (!p)
340		return 1;
341
342	if (strncmp(p, "on", 2) == 0)
343		fw_dump.fadump_enabled = 1;
344	else if (strncmp(p, "off", 3) == 0)
345		fw_dump.fadump_enabled = 0;
346
347	return 0;
348}
349early_param("fadump", early_fadump_param);
350
351/* Look for fadump_reserve_mem= cmdline option */
352static int __init early_fadump_reserve_mem(char *p)
353{
354	if (p)
355		fw_dump.reserve_bootvar = memparse(p, &p);
356	return 0;
357}
358early_param("fadump_reserve_mem", early_fadump_reserve_mem);
359
360static void register_fw_dump(struct fadump_mem_struct *fdm)
361{
362	int rc;
363	unsigned int wait_time;
364
365	pr_debug("Registering for firmware-assisted kernel dump...\n");
366
367	/* TODO: Add upper time limit for the delay */
368	do {
369		rc = rtas_call(fw_dump.ibm_configure_kernel_dump, 3, 1, NULL,
370			FADUMP_REGISTER, fdm,
371			sizeof(struct fadump_mem_struct));
372
373		wait_time = rtas_busy_delay_time(rc);
374		if (wait_time)
375			mdelay(wait_time);
376
377	} while (wait_time);
378
379	switch (rc) {
380	case -1:
381		printk(KERN_ERR "Failed to register firmware-assisted kernel"
382			" dump. Hardware Error(%d).\n", rc);
383		break;
384	case -3:
385		printk(KERN_ERR "Failed to register firmware-assisted kernel"
386			" dump. Parameter Error(%d).\n", rc);
387		break;
388	case -9:
389		printk(KERN_ERR "firmware-assisted kernel dump is already "
390			" registered.");
391		fw_dump.dump_registered = 1;
392		break;
393	case 0:
394		printk(KERN_INFO "firmware-assisted kernel dump registration"
395			" is successful\n");
396		fw_dump.dump_registered = 1;
397		break;
398	}
399}
400
401void crash_fadump(struct pt_regs *regs, const char *str)
402{
403	struct fadump_crash_info_header *fdh = NULL;
404
405	if (!fw_dump.dump_registered || !fw_dump.fadumphdr_addr)
406		return;
407
408	fdh = __va(fw_dump.fadumphdr_addr);
409	crashing_cpu = smp_processor_id();
410	fdh->crashing_cpu = crashing_cpu;
411	crash_save_vmcoreinfo();
412
413	if (regs)
414		fdh->regs = *regs;
415	else
416		ppc_save_regs(&fdh->regs);
417
418	fdh->cpu_online_mask = *cpu_online_mask;
419
420	/* Call ibm,os-term rtas call to trigger firmware assisted dump */
421	rtas_os_term((char *)str);
422}
423
424#define GPR_MASK	0xffffff0000000000
425static inline int fadump_gpr_index(u64 id)
426{
427	int i = -1;
428	char str[3];
429
430	if ((id & GPR_MASK) == REG_ID("GPR")) {
431		/* get the digits at the end */
432		id &= ~GPR_MASK;
433		id >>= 24;
434		str[2] = '\0';
435		str[1] = id & 0xff;
436		str[0] = (id >> 8) & 0xff;
437		sscanf(str, "%d", &i);
438		if (i > 31)
439			i = -1;
440	}
441	return i;
442}
443
444static inline void fadump_set_regval(struct pt_regs *regs, u64 reg_id,
445								u64 reg_val)
446{
447	int i;
448
449	i = fadump_gpr_index(reg_id);
450	if (i >= 0)
451		regs->gpr[i] = (unsigned long)reg_val;
452	else if (reg_id == REG_ID("NIA"))
453		regs->nip = (unsigned long)reg_val;
454	else if (reg_id == REG_ID("MSR"))
455		regs->msr = (unsigned long)reg_val;
456	else if (reg_id == REG_ID("CTR"))
457		regs->ctr = (unsigned long)reg_val;
458	else if (reg_id == REG_ID("LR"))
459		regs->link = (unsigned long)reg_val;
460	else if (reg_id == REG_ID("XER"))
461		regs->xer = (unsigned long)reg_val;
462	else if (reg_id == REG_ID("CR"))
463		regs->ccr = (unsigned long)reg_val;
464	else if (reg_id == REG_ID("DAR"))
465		regs->dar = (unsigned long)reg_val;
466	else if (reg_id == REG_ID("DSISR"))
467		regs->dsisr = (unsigned long)reg_val;
468}
469
470static struct fadump_reg_entry*
471fadump_read_registers(struct fadump_reg_entry *reg_entry, struct pt_regs *regs)
472{
473	memset(regs, 0, sizeof(struct pt_regs));
474
475	while (be64_to_cpu(reg_entry->reg_id) != REG_ID("CPUEND")) {
476		fadump_set_regval(regs, be64_to_cpu(reg_entry->reg_id),
477					be64_to_cpu(reg_entry->reg_value));
478		reg_entry++;
479	}
480	reg_entry++;
481	return reg_entry;
482}
483
484static u32 *fadump_append_elf_note(u32 *buf, char *name, unsigned type,
485						void *data, size_t data_len)
486{
487	struct elf_note note;
488
489	note.n_namesz = strlen(name) + 1;
490	note.n_descsz = data_len;
491	note.n_type   = type;
492	memcpy(buf, &note, sizeof(note));
493	buf += (sizeof(note) + 3)/4;
494	memcpy(buf, name, note.n_namesz);
495	buf += (note.n_namesz + 3)/4;
496	memcpy(buf, data, note.n_descsz);
497	buf += (note.n_descsz + 3)/4;
498
499	return buf;
500}
501
502static void fadump_final_note(u32 *buf)
503{
504	struct elf_note note;
505
506	note.n_namesz = 0;
507	note.n_descsz = 0;
508	note.n_type   = 0;
509	memcpy(buf, &note, sizeof(note));
510}
511
512static u32 *fadump_regs_to_elf_notes(u32 *buf, struct pt_regs *regs)
513{
514	struct elf_prstatus prstatus;
515
516	memset(&prstatus, 0, sizeof(prstatus));
517	/*
518	 * FIXME: How do i get PID? Do I really need it?
519	 * prstatus.pr_pid = ????
520	 */
521	elf_core_copy_kernel_regs(&prstatus.pr_reg, regs);
522	buf = fadump_append_elf_note(buf, KEXEC_CORE_NOTE_NAME, NT_PRSTATUS,
523				&prstatus, sizeof(prstatus));
524	return buf;
525}
526
527static void fadump_update_elfcore_header(char *bufp)
528{
529	struct elfhdr *elf;
530	struct elf_phdr *phdr;
531
532	elf = (struct elfhdr *)bufp;
533	bufp += sizeof(struct elfhdr);
534
535	/* First note is a place holder for cpu notes info. */
536	phdr = (struct elf_phdr *)bufp;
537
538	if (phdr->p_type == PT_NOTE) {
539		phdr->p_paddr = fw_dump.cpu_notes_buf;
540		phdr->p_offset	= phdr->p_paddr;
541		phdr->p_filesz	= fw_dump.cpu_notes_buf_size;
542		phdr->p_memsz = fw_dump.cpu_notes_buf_size;
543	}
544	return;
545}
546
547static void *fadump_cpu_notes_buf_alloc(unsigned long size)
548{
549	void *vaddr;
550	struct page *page;
551	unsigned long order, count, i;
552
553	order = get_order(size);
554	vaddr = (void *)__get_free_pages(GFP_KERNEL|__GFP_ZERO, order);
555	if (!vaddr)
556		return NULL;
557
558	count = 1 << order;
559	page = virt_to_page(vaddr);
560	for (i = 0; i < count; i++)
561		SetPageReserved(page + i);
562	return vaddr;
563}
564
565static void fadump_cpu_notes_buf_free(unsigned long vaddr, unsigned long size)
566{
567	struct page *page;
568	unsigned long order, count, i;
569
570	order = get_order(size);
571	count = 1 << order;
572	page = virt_to_page(vaddr);
573	for (i = 0; i < count; i++)
574		ClearPageReserved(page + i);
575	__free_pages(page, order);
576}
577
578/*
579 * Read CPU state dump data and convert it into ELF notes.
580 * The CPU dump starts with magic number "REGSAVE". NumCpusOffset should be
581 * used to access the data to allow for additional fields to be added without
582 * affecting compatibility. Each list of registers for a CPU starts with
583 * "CPUSTRT" and ends with "CPUEND". Each register entry is of 16 bytes,
584 * 8 Byte ASCII identifier and 8 Byte register value. The register entry
585 * with identifier "CPUSTRT" and "CPUEND" contains 4 byte cpu id as part
586 * of register value. For more details refer to PAPR document.
587 *
588 * Only for the crashing cpu we ignore the CPU dump data and get exact
589 * state from fadump crash info structure populated by first kernel at the
590 * time of crash.
591 */
592static int __init fadump_build_cpu_notes(const struct fadump_mem_struct *fdm)
593{
594	struct fadump_reg_save_area_header *reg_header;
595	struct fadump_reg_entry *reg_entry;
596	struct fadump_crash_info_header *fdh = NULL;
597	void *vaddr;
598	unsigned long addr;
599	u32 num_cpus, *note_buf;
600	struct pt_regs regs;
601	int i, rc = 0, cpu = 0;
602
603	if (!fdm->cpu_state_data.bytes_dumped)
604		return -EINVAL;
605
606	addr = be64_to_cpu(fdm->cpu_state_data.destination_address);
607	vaddr = __va(addr);
608
609	reg_header = vaddr;
610	if (be64_to_cpu(reg_header->magic_number) != REGSAVE_AREA_MAGIC) {
611		printk(KERN_ERR "Unable to read register save area.\n");
612		return -ENOENT;
613	}
614	pr_debug("--------CPU State Data------------\n");
615	pr_debug("Magic Number: %llx\n", be64_to_cpu(reg_header->magic_number));
616	pr_debug("NumCpuOffset: %x\n", be32_to_cpu(reg_header->num_cpu_offset));
617
618	vaddr += be32_to_cpu(reg_header->num_cpu_offset);
619	num_cpus = be32_to_cpu(*((__be32 *)(vaddr)));
620	pr_debug("NumCpus     : %u\n", num_cpus);
621	vaddr += sizeof(u32);
622	reg_entry = (struct fadump_reg_entry *)vaddr;
623
624	/* Allocate buffer to hold cpu crash notes. */
625	fw_dump.cpu_notes_buf_size = num_cpus * sizeof(note_buf_t);
626	fw_dump.cpu_notes_buf_size = PAGE_ALIGN(fw_dump.cpu_notes_buf_size);
627	note_buf = fadump_cpu_notes_buf_alloc(fw_dump.cpu_notes_buf_size);
628	if (!note_buf) {
629		printk(KERN_ERR "Failed to allocate 0x%lx bytes for "
630			"cpu notes buffer\n", fw_dump.cpu_notes_buf_size);
631		return -ENOMEM;
632	}
633	fw_dump.cpu_notes_buf = __pa(note_buf);
634
635	pr_debug("Allocated buffer for cpu notes of size %ld at %p\n",
636			(num_cpus * sizeof(note_buf_t)), note_buf);
637
638	if (fw_dump.fadumphdr_addr)
639		fdh = __va(fw_dump.fadumphdr_addr);
640
641	for (i = 0; i < num_cpus; i++) {
642		if (be64_to_cpu(reg_entry->reg_id) != REG_ID("CPUSTRT")) {
643			printk(KERN_ERR "Unable to read CPU state data\n");
644			rc = -ENOENT;
645			goto error_out;
646		}
647		/* Lower 4 bytes of reg_value contains logical cpu id */
648		cpu = be64_to_cpu(reg_entry->reg_value) & FADUMP_CPU_ID_MASK;
649		if (fdh && !cpumask_test_cpu(cpu, &fdh->cpu_online_mask)) {
650			SKIP_TO_NEXT_CPU(reg_entry);
651			continue;
652		}
653		pr_debug("Reading register data for cpu %d...\n", cpu);
654		if (fdh && fdh->crashing_cpu == cpu) {
655			regs = fdh->regs;
656			note_buf = fadump_regs_to_elf_notes(note_buf, &regs);
657			SKIP_TO_NEXT_CPU(reg_entry);
658		} else {
659			reg_entry++;
660			reg_entry = fadump_read_registers(reg_entry, &regs);
661			note_buf = fadump_regs_to_elf_notes(note_buf, &regs);
662		}
663	}
664	fadump_final_note(note_buf);
665
666	if (fdh) {
667		pr_debug("Updating elfcore header (%llx) with cpu notes\n",
668							fdh->elfcorehdr_addr);
669		fadump_update_elfcore_header((char *)__va(fdh->elfcorehdr_addr));
670	}
671	return 0;
672
673error_out:
674	fadump_cpu_notes_buf_free((unsigned long)__va(fw_dump.cpu_notes_buf),
675					fw_dump.cpu_notes_buf_size);
676	fw_dump.cpu_notes_buf = 0;
677	fw_dump.cpu_notes_buf_size = 0;
678	return rc;
679
680}
681
682/*
683 * Validate and process the dump data stored by firmware before exporting
684 * it through '/proc/vmcore'.
685 */
686static int __init process_fadump(const struct fadump_mem_struct *fdm_active)
687{
688	struct fadump_crash_info_header *fdh;
689	int rc = 0;
690
691	if (!fdm_active || !fw_dump.fadumphdr_addr)
692		return -EINVAL;
693
694	/* Check if the dump data is valid. */
695	if ((be16_to_cpu(fdm_active->header.dump_status_flag) == FADUMP_ERROR_FLAG) ||
696			(fdm_active->cpu_state_data.error_flags != 0) ||
697			(fdm_active->rmr_region.error_flags != 0)) {
698		printk(KERN_ERR "Dump taken by platform is not valid\n");
699		return -EINVAL;
700	}
701	if ((fdm_active->rmr_region.bytes_dumped !=
702			fdm_active->rmr_region.source_len) ||
703			!fdm_active->cpu_state_data.bytes_dumped) {
704		printk(KERN_ERR "Dump taken by platform is incomplete\n");
705		return -EINVAL;
706	}
707
708	/* Validate the fadump crash info header */
709	fdh = __va(fw_dump.fadumphdr_addr);
710	if (fdh->magic_number != FADUMP_CRASH_INFO_MAGIC) {
711		printk(KERN_ERR "Crash info header is not valid.\n");
712		return -EINVAL;
713	}
714
715	rc = fadump_build_cpu_notes(fdm_active);
716	if (rc)
717		return rc;
718
719	/*
720	 * We are done validating dump info and elfcore header is now ready
721	 * to be exported. set elfcorehdr_addr so that vmcore module will
722	 * export the elfcore header through '/proc/vmcore'.
723	 */
724	elfcorehdr_addr = fdh->elfcorehdr_addr;
725
726	return 0;
727}
728
729static inline void fadump_add_crash_memory(unsigned long long base,
730					unsigned long long end)
731{
732	if (base == end)
733		return;
734
735	pr_debug("crash_memory_range[%d] [%#016llx-%#016llx], %#llx bytes\n",
736		crash_mem_ranges, base, end - 1, (end - base));
737	crash_memory_ranges[crash_mem_ranges].base = base;
738	crash_memory_ranges[crash_mem_ranges].size = end - base;
739	crash_mem_ranges++;
740}
741
742static void fadump_exclude_reserved_area(unsigned long long start,
743					unsigned long long end)
744{
745	unsigned long long ra_start, ra_end;
746
747	ra_start = fw_dump.reserve_dump_area_start;
748	ra_end = ra_start + fw_dump.reserve_dump_area_size;
749
750	if ((ra_start < end) && (ra_end > start)) {
751		if ((start < ra_start) && (end > ra_end)) {
752			fadump_add_crash_memory(start, ra_start);
753			fadump_add_crash_memory(ra_end, end);
754		} else if (start < ra_start) {
755			fadump_add_crash_memory(start, ra_start);
756		} else if (ra_end < end) {
757			fadump_add_crash_memory(ra_end, end);
758		}
759	} else
760		fadump_add_crash_memory(start, end);
761}
762
763static int fadump_init_elfcore_header(char *bufp)
764{
765	struct elfhdr *elf;
766
767	elf = (struct elfhdr *) bufp;
768	bufp += sizeof(struct elfhdr);
769	memcpy(elf->e_ident, ELFMAG, SELFMAG);
770	elf->e_ident[EI_CLASS] = ELF_CLASS;
771	elf->e_ident[EI_DATA] = ELF_DATA;
772	elf->e_ident[EI_VERSION] = EV_CURRENT;
773	elf->e_ident[EI_OSABI] = ELF_OSABI;
774	memset(elf->e_ident+EI_PAD, 0, EI_NIDENT-EI_PAD);
775	elf->e_type = ET_CORE;
776	elf->e_machine = ELF_ARCH;
777	elf->e_version = EV_CURRENT;
778	elf->e_entry = 0;
779	elf->e_phoff = sizeof(struct elfhdr);
780	elf->e_shoff = 0;
781	elf->e_flags = ELF_CORE_EFLAGS;
782	elf->e_ehsize = sizeof(struct elfhdr);
783	elf->e_phentsize = sizeof(struct elf_phdr);
784	elf->e_phnum = 0;
785	elf->e_shentsize = 0;
786	elf->e_shnum = 0;
787	elf->e_shstrndx = 0;
788
789	return 0;
790}
791
792/*
793 * Traverse through memblock structure and setup crash memory ranges. These
794 * ranges will be used create PT_LOAD program headers in elfcore header.
795 */
796static void fadump_setup_crash_memory_ranges(void)
797{
798	struct memblock_region *reg;
799	unsigned long long start, end;
800
801	pr_debug("Setup crash memory ranges.\n");
802	crash_mem_ranges = 0;
803	/*
804	 * add the first memory chunk (RMA_START through boot_memory_size) as
805	 * a separate memory chunk. The reason is, at the time crash firmware
806	 * will move the content of this memory chunk to different location
807	 * specified during fadump registration. We need to create a separate
808	 * program header for this chunk with the correct offset.
809	 */
810	fadump_add_crash_memory(RMA_START, fw_dump.boot_memory_size);
811
812	for_each_memblock(memory, reg) {
813		start = (unsigned long long)reg->base;
814		end = start + (unsigned long long)reg->size;
815		if (start == RMA_START && end >= fw_dump.boot_memory_size)
816			start = fw_dump.boot_memory_size;
817
818		/* add this range excluding the reserved dump area. */
819		fadump_exclude_reserved_area(start, end);
820	}
821}
822
823/*
824 * If the given physical address falls within the boot memory region then
825 * return the relocated address that points to the dump region reserved
826 * for saving initial boot memory contents.
827 */
828static inline unsigned long fadump_relocate(unsigned long paddr)
829{
830	if (paddr > RMA_START && paddr < fw_dump.boot_memory_size)
831		return be64_to_cpu(fdm.rmr_region.destination_address) + paddr;
832	else
833		return paddr;
834}
835
836static int fadump_create_elfcore_headers(char *bufp)
837{
838	struct elfhdr *elf;
839	struct elf_phdr *phdr;
840	int i;
841
842	fadump_init_elfcore_header(bufp);
843	elf = (struct elfhdr *)bufp;
844	bufp += sizeof(struct elfhdr);
845
846	/*
847	 * setup ELF PT_NOTE, place holder for cpu notes info. The notes info
848	 * will be populated during second kernel boot after crash. Hence
849	 * this PT_NOTE will always be the first elf note.
850	 *
851	 * NOTE: Any new ELF note addition should be placed after this note.
852	 */
853	phdr = (struct elf_phdr *)bufp;
854	bufp += sizeof(struct elf_phdr);
855	phdr->p_type = PT_NOTE;
856	phdr->p_flags = 0;
857	phdr->p_vaddr = 0;
858	phdr->p_align = 0;
859
860	phdr->p_offset = 0;
861	phdr->p_paddr = 0;
862	phdr->p_filesz = 0;
863	phdr->p_memsz = 0;
864
865	(elf->e_phnum)++;
866
867	/* setup ELF PT_NOTE for vmcoreinfo */
868	phdr = (struct elf_phdr *)bufp;
869	bufp += sizeof(struct elf_phdr);
870	phdr->p_type	= PT_NOTE;
871	phdr->p_flags	= 0;
872	phdr->p_vaddr	= 0;
873	phdr->p_align	= 0;
874
875	phdr->p_paddr	= fadump_relocate(paddr_vmcoreinfo_note());
876	phdr->p_offset	= phdr->p_paddr;
877	phdr->p_memsz	= vmcoreinfo_max_size;
878	phdr->p_filesz	= vmcoreinfo_max_size;
879
880	/* Increment number of program headers. */
881	(elf->e_phnum)++;
882
883	/* setup PT_LOAD sections. */
884
885	for (i = 0; i < crash_mem_ranges; i++) {
886		unsigned long long mbase, msize;
887		mbase = crash_memory_ranges[i].base;
888		msize = crash_memory_ranges[i].size;
889
890		if (!msize)
891			continue;
892
893		phdr = (struct elf_phdr *)bufp;
894		bufp += sizeof(struct elf_phdr);
895		phdr->p_type	= PT_LOAD;
896		phdr->p_flags	= PF_R|PF_W|PF_X;
897		phdr->p_offset	= mbase;
898
899		if (mbase == RMA_START) {
900			/*
901			 * The entire RMA region will be moved by firmware
902			 * to the specified destination_address. Hence set
903			 * the correct offset.
904			 */
905			phdr->p_offset = be64_to_cpu(fdm.rmr_region.destination_address);
906		}
907
908		phdr->p_paddr = mbase;
909		phdr->p_vaddr = (unsigned long)__va(mbase);
910		phdr->p_filesz = msize;
911		phdr->p_memsz = msize;
912		phdr->p_align = 0;
913
914		/* Increment number of program headers. */
915		(elf->e_phnum)++;
916	}
917	return 0;
918}
919
920static unsigned long init_fadump_header(unsigned long addr)
921{
922	struct fadump_crash_info_header *fdh;
923
924	if (!addr)
925		return 0;
926
927	fw_dump.fadumphdr_addr = addr;
928	fdh = __va(addr);
929	addr += sizeof(struct fadump_crash_info_header);
930
931	memset(fdh, 0, sizeof(struct fadump_crash_info_header));
932	fdh->magic_number = FADUMP_CRASH_INFO_MAGIC;
933	fdh->elfcorehdr_addr = addr;
934	/* We will set the crashing cpu id in crash_fadump() during crash. */
935	fdh->crashing_cpu = CPU_UNKNOWN;
936
937	return addr;
938}
939
940static void register_fadump(void)
941{
942	unsigned long addr;
943	void *vaddr;
944
945	/*
946	 * If no memory is reserved then we can not register for firmware-
947	 * assisted dump.
948	 */
949	if (!fw_dump.reserve_dump_area_size)
950		return;
951
952	fadump_setup_crash_memory_ranges();
953
954	addr = be64_to_cpu(fdm.rmr_region.destination_address) + be64_to_cpu(fdm.rmr_region.source_len);
955	/* Initialize fadump crash info header. */
956	addr = init_fadump_header(addr);
957	vaddr = __va(addr);
958
959	pr_debug("Creating ELF core headers at %#016lx\n", addr);
960	fadump_create_elfcore_headers(vaddr);
961
962	/* register the future kernel dump with firmware. */
963	register_fw_dump(&fdm);
964}
965
966static int fadump_unregister_dump(struct fadump_mem_struct *fdm)
967{
968	int rc = 0;
969	unsigned int wait_time;
970
971	pr_debug("Un-register firmware-assisted dump\n");
972
973	/* TODO: Add upper time limit for the delay */
974	do {
975		rc = rtas_call(fw_dump.ibm_configure_kernel_dump, 3, 1, NULL,
976			FADUMP_UNREGISTER, fdm,
977			sizeof(struct fadump_mem_struct));
978
979		wait_time = rtas_busy_delay_time(rc);
980		if (wait_time)
981			mdelay(wait_time);
982	} while (wait_time);
983
984	if (rc) {
985		printk(KERN_ERR "Failed to un-register firmware-assisted dump."
986			" unexpected error(%d).\n", rc);
987		return rc;
988	}
989	fw_dump.dump_registered = 0;
990	return 0;
991}
992
993static int fadump_invalidate_dump(struct fadump_mem_struct *fdm)
994{
995	int rc = 0;
996	unsigned int wait_time;
997
998	pr_debug("Invalidating firmware-assisted dump registration\n");
999
1000	/* TODO: Add upper time limit for the delay */
1001	do {
1002		rc = rtas_call(fw_dump.ibm_configure_kernel_dump, 3, 1, NULL,
1003			FADUMP_INVALIDATE, fdm,
1004			sizeof(struct fadump_mem_struct));
1005
1006		wait_time = rtas_busy_delay_time(rc);
1007		if (wait_time)
1008			mdelay(wait_time);
1009	} while (wait_time);
1010
1011	if (rc) {
1012		printk(KERN_ERR "Failed to invalidate firmware-assisted dump "
1013			"rgistration. unexpected error(%d).\n", rc);
1014		return rc;
1015	}
1016	fw_dump.dump_active = 0;
1017	fdm_active = NULL;
1018	return 0;
1019}
1020
1021void fadump_cleanup(void)
1022{
1023	/* Invalidate the registration only if dump is active. */
1024	if (fw_dump.dump_active) {
1025		init_fadump_mem_struct(&fdm,
1026			be64_to_cpu(fdm_active->cpu_state_data.destination_address));
1027		fadump_invalidate_dump(&fdm);
1028	}
1029}
1030
1031/*
1032 * Release the memory that was reserved in early boot to preserve the memory
1033 * contents. The released memory will be available for general use.
1034 */
1035static void fadump_release_memory(unsigned long begin, unsigned long end)
1036{
1037	unsigned long addr;
1038	unsigned long ra_start, ra_end;
1039
1040	ra_start = fw_dump.reserve_dump_area_start;
1041	ra_end = ra_start + fw_dump.reserve_dump_area_size;
1042
1043	for (addr = begin; addr < end; addr += PAGE_SIZE) {
1044		/*
1045		 * exclude the dump reserve area. Will reuse it for next
1046		 * fadump registration.
1047		 */
1048		if (addr <= ra_end && ((addr + PAGE_SIZE) > ra_start))
1049			continue;
1050
1051		free_reserved_page(pfn_to_page(addr >> PAGE_SHIFT));
1052	}
1053}
1054
1055static void fadump_invalidate_release_mem(void)
1056{
1057	unsigned long reserved_area_start, reserved_area_end;
1058	unsigned long destination_address;
1059
1060	mutex_lock(&fadump_mutex);
1061	if (!fw_dump.dump_active) {
1062		mutex_unlock(&fadump_mutex);
1063		return;
1064	}
1065
1066	destination_address = be64_to_cpu(fdm_active->cpu_state_data.destination_address);
1067	fadump_cleanup();
1068	mutex_unlock(&fadump_mutex);
1069
1070	/*
1071	 * Save the current reserved memory bounds we will require them
1072	 * later for releasing the memory for general use.
1073	 */
1074	reserved_area_start = fw_dump.reserve_dump_area_start;
1075	reserved_area_end = reserved_area_start +
1076			fw_dump.reserve_dump_area_size;
1077	/*
1078	 * Setup reserve_dump_area_start and its size so that we can
1079	 * reuse this reserved memory for Re-registration.
1080	 */
1081	fw_dump.reserve_dump_area_start = destination_address;
1082	fw_dump.reserve_dump_area_size = get_fadump_area_size();
1083
1084	fadump_release_memory(reserved_area_start, reserved_area_end);
1085	if (fw_dump.cpu_notes_buf) {
1086		fadump_cpu_notes_buf_free(
1087				(unsigned long)__va(fw_dump.cpu_notes_buf),
1088				fw_dump.cpu_notes_buf_size);
1089		fw_dump.cpu_notes_buf = 0;
1090		fw_dump.cpu_notes_buf_size = 0;
1091	}
1092	/* Initialize the kernel dump memory structure for FAD registration. */
1093	init_fadump_mem_struct(&fdm, fw_dump.reserve_dump_area_start);
1094}
1095
1096static ssize_t fadump_release_memory_store(struct kobject *kobj,
1097					struct kobj_attribute *attr,
1098					const char *buf, size_t count)
1099{
1100	if (!fw_dump.dump_active)
1101		return -EPERM;
1102
1103	if (buf[0] == '1') {
1104		/*
1105		 * Take away the '/proc/vmcore'. We are releasing the dump
1106		 * memory, hence it will not be valid anymore.
1107		 */
1108		vmcore_cleanup();
1109		fadump_invalidate_release_mem();
1110
1111	} else
1112		return -EINVAL;
1113	return count;
1114}
1115
1116static ssize_t fadump_enabled_show(struct kobject *kobj,
1117					struct kobj_attribute *attr,
1118					char *buf)
1119{
1120	return sprintf(buf, "%d\n", fw_dump.fadump_enabled);
1121}
1122
1123static ssize_t fadump_register_show(struct kobject *kobj,
1124					struct kobj_attribute *attr,
1125					char *buf)
1126{
1127	return sprintf(buf, "%d\n", fw_dump.dump_registered);
1128}
1129
1130static ssize_t fadump_register_store(struct kobject *kobj,
1131					struct kobj_attribute *attr,
1132					const char *buf, size_t count)
1133{
1134	int ret = 0;
1135
1136	if (!fw_dump.fadump_enabled || fdm_active)
1137		return -EPERM;
1138
1139	mutex_lock(&fadump_mutex);
1140
1141	switch (buf[0]) {
1142	case '0':
1143		if (fw_dump.dump_registered == 0) {
1144			ret = -EINVAL;
1145			goto unlock_out;
1146		}
1147		/* Un-register Firmware-assisted dump */
1148		fadump_unregister_dump(&fdm);
1149		break;
1150	case '1':
1151		if (fw_dump.dump_registered == 1) {
1152			ret = -EINVAL;
1153			goto unlock_out;
1154		}
1155		/* Register Firmware-assisted dump */
1156		register_fadump();
1157		break;
1158	default:
1159		ret = -EINVAL;
1160		break;
1161	}
1162
1163unlock_out:
1164	mutex_unlock(&fadump_mutex);
1165	return ret < 0 ? ret : count;
1166}
1167
1168static int fadump_region_show(struct seq_file *m, void *private)
1169{
1170	const struct fadump_mem_struct *fdm_ptr;
1171
1172	if (!fw_dump.fadump_enabled)
1173		return 0;
1174
1175	mutex_lock(&fadump_mutex);
1176	if (fdm_active)
1177		fdm_ptr = fdm_active;
1178	else {
1179		mutex_unlock(&fadump_mutex);
1180		fdm_ptr = &fdm;
1181	}
1182
1183	seq_printf(m,
1184			"CPU : [%#016llx-%#016llx] %#llx bytes, "
1185			"Dumped: %#llx\n",
1186			be64_to_cpu(fdm_ptr->cpu_state_data.destination_address),
1187			be64_to_cpu(fdm_ptr->cpu_state_data.destination_address) +
1188			be64_to_cpu(fdm_ptr->cpu_state_data.source_len) - 1,
1189			be64_to_cpu(fdm_ptr->cpu_state_data.source_len),
1190			be64_to_cpu(fdm_ptr->cpu_state_data.bytes_dumped));
1191	seq_printf(m,
1192			"HPTE: [%#016llx-%#016llx] %#llx bytes, "
1193			"Dumped: %#llx\n",
1194			be64_to_cpu(fdm_ptr->hpte_region.destination_address),
1195			be64_to_cpu(fdm_ptr->hpte_region.destination_address) +
1196			be64_to_cpu(fdm_ptr->hpte_region.source_len) - 1,
1197			be64_to_cpu(fdm_ptr->hpte_region.source_len),
1198			be64_to_cpu(fdm_ptr->hpte_region.bytes_dumped));
1199	seq_printf(m,
1200			"DUMP: [%#016llx-%#016llx] %#llx bytes, "
1201			"Dumped: %#llx\n",
1202			be64_to_cpu(fdm_ptr->rmr_region.destination_address),
1203			be64_to_cpu(fdm_ptr->rmr_region.destination_address) +
1204			be64_to_cpu(fdm_ptr->rmr_region.source_len) - 1,
1205			be64_to_cpu(fdm_ptr->rmr_region.source_len),
1206			be64_to_cpu(fdm_ptr->rmr_region.bytes_dumped));
1207
1208	if (!fdm_active ||
1209		(fw_dump.reserve_dump_area_start ==
1210		be64_to_cpu(fdm_ptr->cpu_state_data.destination_address)))
1211		goto out;
1212
1213	/* Dump is active. Show reserved memory region. */
1214	seq_printf(m,
1215			"    : [%#016llx-%#016llx] %#llx bytes, "
1216			"Dumped: %#llx\n",
1217			(unsigned long long)fw_dump.reserve_dump_area_start,
1218			be64_to_cpu(fdm_ptr->cpu_state_data.destination_address) - 1,
1219			be64_to_cpu(fdm_ptr->cpu_state_data.destination_address) -
1220			fw_dump.reserve_dump_area_start,
1221			be64_to_cpu(fdm_ptr->cpu_state_data.destination_address) -
1222			fw_dump.reserve_dump_area_start);
1223out:
1224	if (fdm_active)
1225		mutex_unlock(&fadump_mutex);
1226	return 0;
1227}
1228
1229static struct kobj_attribute fadump_release_attr = __ATTR(fadump_release_mem,
1230						0200, NULL,
1231						fadump_release_memory_store);
1232static struct kobj_attribute fadump_attr = __ATTR(fadump_enabled,
1233						0444, fadump_enabled_show,
1234						NULL);
1235static struct kobj_attribute fadump_register_attr = __ATTR(fadump_registered,
1236						0644, fadump_register_show,
1237						fadump_register_store);
1238
1239static int fadump_region_open(struct inode *inode, struct file *file)
1240{
1241	return single_open(file, fadump_region_show, inode->i_private);
1242}
1243
1244static const struct file_operations fadump_region_fops = {
1245	.open    = fadump_region_open,
1246	.read    = seq_read,
1247	.llseek  = seq_lseek,
1248	.release = single_release,
1249};
1250
1251static void fadump_init_files(void)
1252{
1253	struct dentry *debugfs_file;
1254	int rc = 0;
1255
1256	rc = sysfs_create_file(kernel_kobj, &fadump_attr.attr);
1257	if (rc)
1258		printk(KERN_ERR "fadump: unable to create sysfs file"
1259			" fadump_enabled (%d)\n", rc);
1260
1261	rc = sysfs_create_file(kernel_kobj, &fadump_register_attr.attr);
1262	if (rc)
1263		printk(KERN_ERR "fadump: unable to create sysfs file"
1264			" fadump_registered (%d)\n", rc);
1265
1266	debugfs_file = debugfs_create_file("fadump_region", 0444,
1267					powerpc_debugfs_root, NULL,
1268					&fadump_region_fops);
1269	if (!debugfs_file)
1270		printk(KERN_ERR "fadump: unable to create debugfs file"
1271				" fadump_region\n");
1272
1273	if (fw_dump.dump_active) {
1274		rc = sysfs_create_file(kernel_kobj, &fadump_release_attr.attr);
1275		if (rc)
1276			printk(KERN_ERR "fadump: unable to create sysfs file"
1277				" fadump_release_mem (%d)\n", rc);
1278	}
1279	return;
1280}
1281
1282/*
1283 * Prepare for firmware-assisted dump.
1284 */
1285int __init setup_fadump(void)
1286{
1287	if (!fw_dump.fadump_enabled)
1288		return 0;
1289
1290	if (!fw_dump.fadump_supported) {
1291		printk(KERN_ERR "Firmware-assisted dump is not supported on"
1292			" this hardware\n");
1293		return 0;
1294	}
1295
1296	fadump_show_config();
1297	/*
1298	 * If dump data is available then see if it is valid and prepare for
1299	 * saving it to the disk.
1300	 */
1301	if (fw_dump.dump_active) {
1302		/*
1303		 * if dump process fails then invalidate the registration
1304		 * and release memory before proceeding for re-registration.
1305		 */
1306		if (process_fadump(fdm_active) < 0)
1307			fadump_invalidate_release_mem();
1308	}
1309	/* Initialize the kernel dump memory structure for FAD registration. */
1310	else if (fw_dump.reserve_dump_area_size)
1311		init_fadump_mem_struct(&fdm, fw_dump.reserve_dump_area_start);
1312	fadump_init_files();
1313
1314	return 1;
1315}
1316subsys_initcall(setup_fadump);
1317