xref: /openbmc/linux/arch/powerpc/kernel/fadump.c (revision 4fc4dca8)
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/seq_file.h>
34 #include <linux/crash_dump.h>
35 #include <linux/kobject.h>
36 #include <linux/sysfs.h>
37 #include <linux/slab.h>
38 #include <linux/cma.h>
39 #include <linux/hugetlb.h>
40 
41 #include <asm/debugfs.h>
42 #include <asm/page.h>
43 #include <asm/prom.h>
44 #include <asm/rtas.h>
45 #include <asm/fadump.h>
46 #include <asm/setup.h>
47 
48 static struct fw_dump fw_dump;
49 static struct fadump_mem_struct fdm;
50 static const struct fadump_mem_struct *fdm_active;
51 #ifdef CONFIG_CMA
52 static struct cma *fadump_cma;
53 #endif
54 
55 static DEFINE_MUTEX(fadump_mutex);
56 struct fad_crash_memory_ranges *crash_memory_ranges;
57 int crash_memory_ranges_size;
58 int crash_mem_ranges;
59 int max_crash_mem_ranges;
60 
61 #ifdef CONFIG_CMA
62 /*
63  * fadump_cma_init() - Initialize CMA area from a fadump reserved memory
64  *
65  * This function initializes CMA area from fadump reserved memory.
66  * The total size of fadump reserved memory covers for boot memory size
67  * + cpu data size + hpte size and metadata.
68  * Initialize only the area equivalent to boot memory size for CMA use.
69  * The reamining portion of fadump reserved memory will be not given
70  * to CMA and pages for thoes will stay reserved. boot memory size is
71  * aligned per CMA requirement to satisy cma_init_reserved_mem() call.
72  * But for some reason even if it fails we still have the memory reservation
73  * with us and we can still continue doing fadump.
74  */
75 int __init fadump_cma_init(void)
76 {
77 	unsigned long long base, size;
78 	int rc;
79 
80 	if (!fw_dump.fadump_enabled)
81 		return 0;
82 
83 	/*
84 	 * Do not use CMA if user has provided fadump=nocma kernel parameter.
85 	 * Return 1 to continue with fadump old behaviour.
86 	 */
87 	if (fw_dump.nocma)
88 		return 1;
89 
90 	base = fw_dump.reserve_dump_area_start;
91 	size = fw_dump.boot_memory_size;
92 
93 	if (!size)
94 		return 0;
95 
96 	rc = cma_init_reserved_mem(base, size, 0, "fadump_cma", &fadump_cma);
97 	if (rc) {
98 		pr_err("Failed to init cma area for firmware-assisted dump,%d\n", rc);
99 		/*
100 		 * Though the CMA init has failed we still have memory
101 		 * reservation with us. The reserved memory will be
102 		 * blocked from production system usage.  Hence return 1,
103 		 * so that we can continue with fadump.
104 		 */
105 		return 1;
106 	}
107 
108 	/*
109 	 * So we now have successfully initialized cma area for fadump.
110 	 */
111 	pr_info("Initialized 0x%lx bytes cma area at %ldMB from 0x%lx "
112 		"bytes of memory reserved for firmware-assisted dump\n",
113 		cma_get_size(fadump_cma),
114 		(unsigned long)cma_get_base(fadump_cma) >> 20,
115 		fw_dump.reserve_dump_area_size);
116 	return 1;
117 }
118 #else
119 static int __init fadump_cma_init(void) { return 1; }
120 #endif /* CONFIG_CMA */
121 
122 /* Scan the Firmware Assisted dump configuration details. */
123 int __init early_init_dt_scan_fw_dump(unsigned long node,
124 			const char *uname, int depth, void *data)
125 {
126 	const __be32 *sections;
127 	int i, num_sections;
128 	int size;
129 	const __be32 *token;
130 
131 	if (depth != 1 || strcmp(uname, "rtas") != 0)
132 		return 0;
133 
134 	/*
135 	 * Check if Firmware Assisted dump is supported. if yes, check
136 	 * if dump has been initiated on last reboot.
137 	 */
138 	token = of_get_flat_dt_prop(node, "ibm,configure-kernel-dump", NULL);
139 	if (!token)
140 		return 1;
141 
142 	fw_dump.fadump_supported = 1;
143 	fw_dump.ibm_configure_kernel_dump = be32_to_cpu(*token);
144 
145 	/*
146 	 * The 'ibm,kernel-dump' rtas node is present only if there is
147 	 * dump data waiting for us.
148 	 */
149 	fdm_active = of_get_flat_dt_prop(node, "ibm,kernel-dump", NULL);
150 	if (fdm_active)
151 		fw_dump.dump_active = 1;
152 
153 	/* Get the sizes required to store dump data for the firmware provided
154 	 * dump sections.
155 	 * For each dump section type supported, a 32bit cell which defines
156 	 * the ID of a supported section followed by two 32 bit cells which
157 	 * gives teh size of the section in bytes.
158 	 */
159 	sections = of_get_flat_dt_prop(node, "ibm,configure-kernel-dump-sizes",
160 					&size);
161 
162 	if (!sections)
163 		return 1;
164 
165 	num_sections = size / (3 * sizeof(u32));
166 
167 	for (i = 0; i < num_sections; i++, sections += 3) {
168 		u32 type = (u32)of_read_number(sections, 1);
169 
170 		switch (type) {
171 		case FADUMP_CPU_STATE_DATA:
172 			fw_dump.cpu_state_data_size =
173 					of_read_ulong(&sections[1], 2);
174 			break;
175 		case FADUMP_HPTE_REGION:
176 			fw_dump.hpte_region_size =
177 					of_read_ulong(&sections[1], 2);
178 			break;
179 		}
180 	}
181 
182 	return 1;
183 }
184 
185 /*
186  * If fadump is registered, check if the memory provided
187  * falls within boot memory area and reserved memory area.
188  */
189 int is_fadump_memory_area(u64 addr, ulong size)
190 {
191 	u64 d_start = fw_dump.reserve_dump_area_start;
192 	u64 d_end = d_start + fw_dump.reserve_dump_area_size;
193 
194 	if (!fw_dump.dump_registered)
195 		return 0;
196 
197 	if (((addr + size) > d_start) && (addr <= d_end))
198 		return 1;
199 
200 	return (addr + size) > RMA_START && addr <= fw_dump.boot_memory_size;
201 }
202 
203 int should_fadump_crash(void)
204 {
205 	if (!fw_dump.dump_registered || !fw_dump.fadumphdr_addr)
206 		return 0;
207 	return 1;
208 }
209 
210 int is_fadump_active(void)
211 {
212 	return fw_dump.dump_active;
213 }
214 
215 /*
216  * Returns 1, if there are no holes in boot memory area,
217  * 0 otherwise.
218  */
219 static int is_boot_memory_area_contiguous(void)
220 {
221 	struct memblock_region *reg;
222 	unsigned long tstart, tend;
223 	unsigned long start_pfn = PHYS_PFN(RMA_START);
224 	unsigned long end_pfn = PHYS_PFN(RMA_START + fw_dump.boot_memory_size);
225 	unsigned int ret = 0;
226 
227 	for_each_memblock(memory, reg) {
228 		tstart = max(start_pfn, memblock_region_memory_base_pfn(reg));
229 		tend = min(end_pfn, memblock_region_memory_end_pfn(reg));
230 		if (tstart < tend) {
231 			/* Memory hole from start_pfn to tstart */
232 			if (tstart > start_pfn)
233 				break;
234 
235 			if (tend == end_pfn) {
236 				ret = 1;
237 				break;
238 			}
239 
240 			start_pfn = tend + 1;
241 		}
242 	}
243 
244 	return ret;
245 }
246 
247 /*
248  * Returns true, if there are no holes in reserved memory area,
249  * false otherwise.
250  */
251 static bool is_reserved_memory_area_contiguous(void)
252 {
253 	struct memblock_region *reg;
254 	unsigned long start, end;
255 	unsigned long d_start = fw_dump.reserve_dump_area_start;
256 	unsigned long d_end = d_start + fw_dump.reserve_dump_area_size;
257 
258 	for_each_memblock(memory, reg) {
259 		start = max(d_start, (unsigned long)reg->base);
260 		end = min(d_end, (unsigned long)(reg->base + reg->size));
261 		if (d_start < end) {
262 			/* Memory hole from d_start to start */
263 			if (start > d_start)
264 				break;
265 
266 			if (end == d_end)
267 				return true;
268 
269 			d_start = end + 1;
270 		}
271 	}
272 
273 	return false;
274 }
275 
276 /* Print firmware assisted dump configurations for debugging purpose. */
277 static void fadump_show_config(void)
278 {
279 	pr_debug("Support for firmware-assisted dump (fadump): %s\n",
280 			(fw_dump.fadump_supported ? "present" : "no support"));
281 
282 	if (!fw_dump.fadump_supported)
283 		return;
284 
285 	pr_debug("Fadump enabled    : %s\n",
286 				(fw_dump.fadump_enabled ? "yes" : "no"));
287 	pr_debug("Dump Active       : %s\n",
288 				(fw_dump.dump_active ? "yes" : "no"));
289 	pr_debug("Dump section sizes:\n");
290 	pr_debug("    CPU state data size: %lx\n", fw_dump.cpu_state_data_size);
291 	pr_debug("    HPTE region size   : %lx\n", fw_dump.hpte_region_size);
292 	pr_debug("Boot memory size  : %lx\n", fw_dump.boot_memory_size);
293 }
294 
295 static unsigned long init_fadump_mem_struct(struct fadump_mem_struct *fdm,
296 				unsigned long addr)
297 {
298 	if (!fdm)
299 		return 0;
300 
301 	memset(fdm, 0, sizeof(struct fadump_mem_struct));
302 	addr = addr & PAGE_MASK;
303 
304 	fdm->header.dump_format_version = cpu_to_be32(0x00000001);
305 	fdm->header.dump_num_sections = cpu_to_be16(3);
306 	fdm->header.dump_status_flag = 0;
307 	fdm->header.offset_first_dump_section =
308 		cpu_to_be32((u32)offsetof(struct fadump_mem_struct, cpu_state_data));
309 
310 	/*
311 	 * Fields for disk dump option.
312 	 * We are not using disk dump option, hence set these fields to 0.
313 	 */
314 	fdm->header.dd_block_size = 0;
315 	fdm->header.dd_block_offset = 0;
316 	fdm->header.dd_num_blocks = 0;
317 	fdm->header.dd_offset_disk_path = 0;
318 
319 	/* set 0 to disable an automatic dump-reboot. */
320 	fdm->header.max_time_auto = 0;
321 
322 	/* Kernel dump sections */
323 	/* cpu state data section. */
324 	fdm->cpu_state_data.request_flag = cpu_to_be32(FADUMP_REQUEST_FLAG);
325 	fdm->cpu_state_data.source_data_type = cpu_to_be16(FADUMP_CPU_STATE_DATA);
326 	fdm->cpu_state_data.source_address = 0;
327 	fdm->cpu_state_data.source_len = cpu_to_be64(fw_dump.cpu_state_data_size);
328 	fdm->cpu_state_data.destination_address = cpu_to_be64(addr);
329 	addr += fw_dump.cpu_state_data_size;
330 
331 	/* hpte region section */
332 	fdm->hpte_region.request_flag = cpu_to_be32(FADUMP_REQUEST_FLAG);
333 	fdm->hpte_region.source_data_type = cpu_to_be16(FADUMP_HPTE_REGION);
334 	fdm->hpte_region.source_address = 0;
335 	fdm->hpte_region.source_len = cpu_to_be64(fw_dump.hpte_region_size);
336 	fdm->hpte_region.destination_address = cpu_to_be64(addr);
337 	addr += fw_dump.hpte_region_size;
338 
339 	/* RMA region section */
340 	fdm->rmr_region.request_flag = cpu_to_be32(FADUMP_REQUEST_FLAG);
341 	fdm->rmr_region.source_data_type = cpu_to_be16(FADUMP_REAL_MODE_REGION);
342 	fdm->rmr_region.source_address = cpu_to_be64(RMA_START);
343 	fdm->rmr_region.source_len = cpu_to_be64(fw_dump.boot_memory_size);
344 	fdm->rmr_region.destination_address = cpu_to_be64(addr);
345 	addr += fw_dump.boot_memory_size;
346 
347 	return addr;
348 }
349 
350 /**
351  * fadump_calculate_reserve_size(): reserve variable boot area 5% of System RAM
352  *
353  * Function to find the largest memory size we need to reserve during early
354  * boot process. This will be the size of the memory that is required for a
355  * kernel to boot successfully.
356  *
357  * This function has been taken from phyp-assisted dump feature implementation.
358  *
359  * returns larger of 256MB or 5% rounded down to multiples of 256MB.
360  *
361  * TODO: Come up with better approach to find out more accurate memory size
362  * that is required for a kernel to boot successfully.
363  *
364  */
365 static inline unsigned long fadump_calculate_reserve_size(void)
366 {
367 	int ret;
368 	unsigned long long base, size;
369 
370 	if (fw_dump.reserve_bootvar)
371 		pr_warn("'fadump_reserve_mem=' parameter is deprecated in favor of 'crashkernel=' parameter.\n");
372 
373 	/*
374 	 * Check if the size is specified through crashkernel= cmdline
375 	 * option. If yes, then use that but ignore base as fadump reserves
376 	 * memory at a predefined offset.
377 	 */
378 	ret = parse_crashkernel(boot_command_line, memblock_phys_mem_size(),
379 				&size, &base);
380 	if (ret == 0 && size > 0) {
381 		unsigned long max_size;
382 
383 		if (fw_dump.reserve_bootvar)
384 			pr_info("Using 'crashkernel=' parameter for memory reservation.\n");
385 
386 		fw_dump.reserve_bootvar = (unsigned long)size;
387 
388 		/*
389 		 * Adjust if the boot memory size specified is above
390 		 * the upper limit.
391 		 */
392 		max_size = memblock_phys_mem_size() / MAX_BOOT_MEM_RATIO;
393 		if (fw_dump.reserve_bootvar > max_size) {
394 			fw_dump.reserve_bootvar = max_size;
395 			pr_info("Adjusted boot memory size to %luMB\n",
396 				(fw_dump.reserve_bootvar >> 20));
397 		}
398 
399 		return fw_dump.reserve_bootvar;
400 	} else if (fw_dump.reserve_bootvar) {
401 		/*
402 		 * 'fadump_reserve_mem=' is being used to reserve memory
403 		 * for firmware-assisted dump.
404 		 */
405 		return fw_dump.reserve_bootvar;
406 	}
407 
408 	/* divide by 20 to get 5% of value */
409 	size = memblock_phys_mem_size() / 20;
410 
411 	/* round it down in multiples of 256 */
412 	size = size & ~0x0FFFFFFFUL;
413 
414 	/* Truncate to memory_limit. We don't want to over reserve the memory.*/
415 	if (memory_limit && size > memory_limit)
416 		size = memory_limit;
417 
418 	return (size > MIN_BOOT_MEM ? size : MIN_BOOT_MEM);
419 }
420 
421 /*
422  * Calculate the total memory size required to be reserved for
423  * firmware-assisted dump registration.
424  */
425 static unsigned long get_fadump_area_size(void)
426 {
427 	unsigned long size = 0;
428 
429 	size += fw_dump.cpu_state_data_size;
430 	size += fw_dump.hpte_region_size;
431 	size += fw_dump.boot_memory_size;
432 	size += sizeof(struct fadump_crash_info_header);
433 	size += sizeof(struct elfhdr); /* ELF core header.*/
434 	size += sizeof(struct elf_phdr); /* place holder for cpu notes */
435 	/* Program headers for crash memory regions. */
436 	size += sizeof(struct elf_phdr) * (memblock_num_regions(memory) + 2);
437 
438 	size = PAGE_ALIGN(size);
439 	return size;
440 }
441 
442 static void __init fadump_reserve_crash_area(unsigned long base,
443 					     unsigned long size)
444 {
445 	struct memblock_region *reg;
446 	unsigned long mstart, mend, msize;
447 
448 	for_each_memblock(memory, reg) {
449 		mstart = max_t(unsigned long, base, reg->base);
450 		mend = reg->base + reg->size;
451 		mend = min(base + size, mend);
452 
453 		if (mstart < mend) {
454 			msize = mend - mstart;
455 			memblock_reserve(mstart, msize);
456 			pr_info("Reserved %ldMB of memory at %#016lx for saving crash dump\n",
457 				(msize >> 20), mstart);
458 		}
459 	}
460 }
461 
462 int __init fadump_reserve_mem(void)
463 {
464 	unsigned long base, size, memory_boundary;
465 
466 	if (!fw_dump.fadump_enabled)
467 		return 0;
468 
469 	if (!fw_dump.fadump_supported) {
470 		printk(KERN_INFO "Firmware-assisted dump is not supported on"
471 				" this hardware\n");
472 		fw_dump.fadump_enabled = 0;
473 		return 0;
474 	}
475 	/*
476 	 * Initialize boot memory size
477 	 * If dump is active then we have already calculated the size during
478 	 * first kernel.
479 	 */
480 	if (fdm_active)
481 		fw_dump.boot_memory_size = be64_to_cpu(fdm_active->rmr_region.source_len);
482 	else {
483 		fw_dump.boot_memory_size = fadump_calculate_reserve_size();
484 #ifdef CONFIG_CMA
485 		if (!fw_dump.nocma)
486 			fw_dump.boot_memory_size =
487 				ALIGN(fw_dump.boot_memory_size,
488 							FADUMP_CMA_ALIGNMENT);
489 #endif
490 	}
491 
492 	/*
493 	 * Calculate the memory boundary.
494 	 * If memory_limit is less than actual memory boundary then reserve
495 	 * the memory for fadump beyond the memory_limit and adjust the
496 	 * memory_limit accordingly, so that the running kernel can run with
497 	 * specified memory_limit.
498 	 */
499 	if (memory_limit && memory_limit < memblock_end_of_DRAM()) {
500 		size = get_fadump_area_size();
501 		if ((memory_limit + size) < memblock_end_of_DRAM())
502 			memory_limit += size;
503 		else
504 			memory_limit = memblock_end_of_DRAM();
505 		printk(KERN_INFO "Adjusted memory_limit for firmware-assisted"
506 				" dump, now %#016llx\n", memory_limit);
507 	}
508 	if (memory_limit)
509 		memory_boundary = memory_limit;
510 	else
511 		memory_boundary = memblock_end_of_DRAM();
512 
513 	if (fw_dump.dump_active) {
514 		pr_info("Firmware-assisted dump is active.\n");
515 
516 #ifdef CONFIG_HUGETLB_PAGE
517 		/*
518 		 * FADump capture kernel doesn't care much about hugepages.
519 		 * In fact, handling hugepages in capture kernel is asking for
520 		 * trouble. So, disable HugeTLB support when fadump is active.
521 		 */
522 		hugetlb_disabled = true;
523 #endif
524 		/*
525 		 * If last boot has crashed then reserve all the memory
526 		 * above boot_memory_size so that we don't touch it until
527 		 * dump is written to disk by userspace tool. This memory
528 		 * will be released for general use once the dump is saved.
529 		 */
530 		base = fw_dump.boot_memory_size;
531 		size = memory_boundary - base;
532 		fadump_reserve_crash_area(base, size);
533 
534 		fw_dump.fadumphdr_addr =
535 				be64_to_cpu(fdm_active->rmr_region.destination_address) +
536 				be64_to_cpu(fdm_active->rmr_region.source_len);
537 		pr_debug("fadumphdr_addr = %pa\n", &fw_dump.fadumphdr_addr);
538 		fw_dump.reserve_dump_area_start = base;
539 		fw_dump.reserve_dump_area_size = size;
540 	} else {
541 		size = get_fadump_area_size();
542 
543 		/*
544 		 * Reserve memory at an offset closer to bottom of the RAM to
545 		 * minimize the impact of memory hot-remove operation. We can't
546 		 * use memblock_find_in_range() here since it doesn't allocate
547 		 * from bottom to top.
548 		 */
549 		for (base = fw_dump.boot_memory_size;
550 		     base <= (memory_boundary - size);
551 		     base += size) {
552 			if (memblock_is_region_memory(base, size) &&
553 			    !memblock_is_region_reserved(base, size))
554 				break;
555 		}
556 		if ((base > (memory_boundary - size)) ||
557 		    memblock_reserve(base, size)) {
558 			pr_err("Failed to reserve memory\n");
559 			return 0;
560 		}
561 
562 		pr_info("Reserved %ldMB of memory at %ldMB for firmware-"
563 			"assisted dump (System RAM: %ldMB)\n",
564 			(unsigned long)(size >> 20),
565 			(unsigned long)(base >> 20),
566 			(unsigned long)(memblock_phys_mem_size() >> 20));
567 
568 		fw_dump.reserve_dump_area_start = base;
569 		fw_dump.reserve_dump_area_size = size;
570 		return fadump_cma_init();
571 	}
572 	return 1;
573 }
574 
575 unsigned long __init arch_reserved_kernel_pages(void)
576 {
577 	return memblock_reserved_size() / PAGE_SIZE;
578 }
579 
580 /* Look for fadump= cmdline option. */
581 static int __init early_fadump_param(char *p)
582 {
583 	if (!p)
584 		return 1;
585 
586 	if (strncmp(p, "on", 2) == 0)
587 		fw_dump.fadump_enabled = 1;
588 	else if (strncmp(p, "off", 3) == 0)
589 		fw_dump.fadump_enabled = 0;
590 	else if (strncmp(p, "nocma", 5) == 0) {
591 		fw_dump.fadump_enabled = 1;
592 		fw_dump.nocma = 1;
593 	}
594 
595 	return 0;
596 }
597 early_param("fadump", early_fadump_param);
598 
599 /*
600  * Look for fadump_reserve_mem= cmdline option
601  * TODO: Remove references to 'fadump_reserve_mem=' parameter,
602  *       the sooner 'crashkernel=' parameter is accustomed to.
603  */
604 static int __init early_fadump_reserve_mem(char *p)
605 {
606 	if (p)
607 		fw_dump.reserve_bootvar = memparse(p, &p);
608 	return 0;
609 }
610 early_param("fadump_reserve_mem", early_fadump_reserve_mem);
611 
612 static int register_fw_dump(struct fadump_mem_struct *fdm)
613 {
614 	int rc, err;
615 	unsigned int wait_time;
616 
617 	pr_debug("Registering for firmware-assisted kernel dump...\n");
618 
619 	/* TODO: Add upper time limit for the delay */
620 	do {
621 		rc = rtas_call(fw_dump.ibm_configure_kernel_dump, 3, 1, NULL,
622 			FADUMP_REGISTER, fdm,
623 			sizeof(struct fadump_mem_struct));
624 
625 		wait_time = rtas_busy_delay_time(rc);
626 		if (wait_time)
627 			mdelay(wait_time);
628 
629 	} while (wait_time);
630 
631 	err = -EIO;
632 	switch (rc) {
633 	default:
634 		pr_err("Failed to register. Unknown Error(%d).\n", rc);
635 		break;
636 	case -1:
637 		printk(KERN_ERR "Failed to register firmware-assisted kernel"
638 			" dump. Hardware Error(%d).\n", rc);
639 		break;
640 	case -3:
641 		if (!is_boot_memory_area_contiguous())
642 			pr_err("Can't have holes in boot memory area while registering fadump\n");
643 		else if (!is_reserved_memory_area_contiguous())
644 			pr_err("Can't have holes in reserved memory area while"
645 			       " registering fadump\n");
646 
647 		printk(KERN_ERR "Failed to register firmware-assisted kernel"
648 			" dump. Parameter Error(%d).\n", rc);
649 		err = -EINVAL;
650 		break;
651 	case -9:
652 		printk(KERN_ERR "firmware-assisted kernel dump is already "
653 			" registered.");
654 		fw_dump.dump_registered = 1;
655 		err = -EEXIST;
656 		break;
657 	case 0:
658 		printk(KERN_INFO "firmware-assisted kernel dump registration"
659 			" is successful\n");
660 		fw_dump.dump_registered = 1;
661 		err = 0;
662 		break;
663 	}
664 	return err;
665 }
666 
667 void crash_fadump(struct pt_regs *regs, const char *str)
668 {
669 	struct fadump_crash_info_header *fdh = NULL;
670 	int old_cpu, this_cpu;
671 
672 	if (!should_fadump_crash())
673 		return;
674 
675 	/*
676 	 * old_cpu == -1 means this is the first CPU which has come here,
677 	 * go ahead and trigger fadump.
678 	 *
679 	 * old_cpu != -1 means some other CPU has already on it's way
680 	 * to trigger fadump, just keep looping here.
681 	 */
682 	this_cpu = smp_processor_id();
683 	old_cpu = cmpxchg(&crashing_cpu, -1, this_cpu);
684 
685 	if (old_cpu != -1) {
686 		/*
687 		 * We can't loop here indefinitely. Wait as long as fadump
688 		 * is in force. If we race with fadump un-registration this
689 		 * loop will break and then we go down to normal panic path
690 		 * and reboot. If fadump is in force the first crashing
691 		 * cpu will definitely trigger fadump.
692 		 */
693 		while (fw_dump.dump_registered)
694 			cpu_relax();
695 		return;
696 	}
697 
698 	fdh = __va(fw_dump.fadumphdr_addr);
699 	fdh->crashing_cpu = crashing_cpu;
700 	crash_save_vmcoreinfo();
701 
702 	if (regs)
703 		fdh->regs = *regs;
704 	else
705 		ppc_save_regs(&fdh->regs);
706 
707 	fdh->online_mask = *cpu_online_mask;
708 
709 	/* Call ibm,os-term rtas call to trigger firmware assisted dump */
710 	rtas_os_term((char *)str);
711 }
712 
713 #define GPR_MASK	0xffffff0000000000
714 static inline int fadump_gpr_index(u64 id)
715 {
716 	int i = -1;
717 	char str[3];
718 
719 	if ((id & GPR_MASK) == REG_ID("GPR")) {
720 		/* get the digits at the end */
721 		id &= ~GPR_MASK;
722 		id >>= 24;
723 		str[2] = '\0';
724 		str[1] = id & 0xff;
725 		str[0] = (id >> 8) & 0xff;
726 		sscanf(str, "%d", &i);
727 		if (i > 31)
728 			i = -1;
729 	}
730 	return i;
731 }
732 
733 static inline void fadump_set_regval(struct pt_regs *regs, u64 reg_id,
734 								u64 reg_val)
735 {
736 	int i;
737 
738 	i = fadump_gpr_index(reg_id);
739 	if (i >= 0)
740 		regs->gpr[i] = (unsigned long)reg_val;
741 	else if (reg_id == REG_ID("NIA"))
742 		regs->nip = (unsigned long)reg_val;
743 	else if (reg_id == REG_ID("MSR"))
744 		regs->msr = (unsigned long)reg_val;
745 	else if (reg_id == REG_ID("CTR"))
746 		regs->ctr = (unsigned long)reg_val;
747 	else if (reg_id == REG_ID("LR"))
748 		regs->link = (unsigned long)reg_val;
749 	else if (reg_id == REG_ID("XER"))
750 		regs->xer = (unsigned long)reg_val;
751 	else if (reg_id == REG_ID("CR"))
752 		regs->ccr = (unsigned long)reg_val;
753 	else if (reg_id == REG_ID("DAR"))
754 		regs->dar = (unsigned long)reg_val;
755 	else if (reg_id == REG_ID("DSISR"))
756 		regs->dsisr = (unsigned long)reg_val;
757 }
758 
759 static struct fadump_reg_entry*
760 fadump_read_registers(struct fadump_reg_entry *reg_entry, struct pt_regs *regs)
761 {
762 	memset(regs, 0, sizeof(struct pt_regs));
763 
764 	while (be64_to_cpu(reg_entry->reg_id) != REG_ID("CPUEND")) {
765 		fadump_set_regval(regs, be64_to_cpu(reg_entry->reg_id),
766 					be64_to_cpu(reg_entry->reg_value));
767 		reg_entry++;
768 	}
769 	reg_entry++;
770 	return reg_entry;
771 }
772 
773 static u32 *fadump_regs_to_elf_notes(u32 *buf, struct pt_regs *regs)
774 {
775 	struct elf_prstatus prstatus;
776 
777 	memset(&prstatus, 0, sizeof(prstatus));
778 	/*
779 	 * FIXME: How do i get PID? Do I really need it?
780 	 * prstatus.pr_pid = ????
781 	 */
782 	elf_core_copy_kernel_regs(&prstatus.pr_reg, regs);
783 	buf = append_elf_note(buf, CRASH_CORE_NOTE_NAME, NT_PRSTATUS,
784 			      &prstatus, sizeof(prstatus));
785 	return buf;
786 }
787 
788 static void fadump_update_elfcore_header(char *bufp)
789 {
790 	struct elfhdr *elf;
791 	struct elf_phdr *phdr;
792 
793 	elf = (struct elfhdr *)bufp;
794 	bufp += sizeof(struct elfhdr);
795 
796 	/* First note is a place holder for cpu notes info. */
797 	phdr = (struct elf_phdr *)bufp;
798 
799 	if (phdr->p_type == PT_NOTE) {
800 		phdr->p_paddr = fw_dump.cpu_notes_buf;
801 		phdr->p_offset	= phdr->p_paddr;
802 		phdr->p_filesz	= fw_dump.cpu_notes_buf_size;
803 		phdr->p_memsz = fw_dump.cpu_notes_buf_size;
804 	}
805 	return;
806 }
807 
808 static void *fadump_cpu_notes_buf_alloc(unsigned long size)
809 {
810 	void *vaddr;
811 	struct page *page;
812 	unsigned long order, count, i;
813 
814 	order = get_order(size);
815 	vaddr = (void *)__get_free_pages(GFP_KERNEL|__GFP_ZERO, order);
816 	if (!vaddr)
817 		return NULL;
818 
819 	count = 1 << order;
820 	page = virt_to_page(vaddr);
821 	for (i = 0; i < count; i++)
822 		SetPageReserved(page + i);
823 	return vaddr;
824 }
825 
826 static void fadump_cpu_notes_buf_free(unsigned long vaddr, unsigned long size)
827 {
828 	struct page *page;
829 	unsigned long order, count, i;
830 
831 	order = get_order(size);
832 	count = 1 << order;
833 	page = virt_to_page(vaddr);
834 	for (i = 0; i < count; i++)
835 		ClearPageReserved(page + i);
836 	__free_pages(page, order);
837 }
838 
839 /*
840  * Read CPU state dump data and convert it into ELF notes.
841  * The CPU dump starts with magic number "REGSAVE". NumCpusOffset should be
842  * used to access the data to allow for additional fields to be added without
843  * affecting compatibility. Each list of registers for a CPU starts with
844  * "CPUSTRT" and ends with "CPUEND". Each register entry is of 16 bytes,
845  * 8 Byte ASCII identifier and 8 Byte register value. The register entry
846  * with identifier "CPUSTRT" and "CPUEND" contains 4 byte cpu id as part
847  * of register value. For more details refer to PAPR document.
848  *
849  * Only for the crashing cpu we ignore the CPU dump data and get exact
850  * state from fadump crash info structure populated by first kernel at the
851  * time of crash.
852  */
853 static int __init fadump_build_cpu_notes(const struct fadump_mem_struct *fdm)
854 {
855 	struct fadump_reg_save_area_header *reg_header;
856 	struct fadump_reg_entry *reg_entry;
857 	struct fadump_crash_info_header *fdh = NULL;
858 	void *vaddr;
859 	unsigned long addr;
860 	u32 num_cpus, *note_buf;
861 	struct pt_regs regs;
862 	int i, rc = 0, cpu = 0;
863 
864 	if (!fdm->cpu_state_data.bytes_dumped)
865 		return -EINVAL;
866 
867 	addr = be64_to_cpu(fdm->cpu_state_data.destination_address);
868 	vaddr = __va(addr);
869 
870 	reg_header = vaddr;
871 	if (be64_to_cpu(reg_header->magic_number) != REGSAVE_AREA_MAGIC) {
872 		printk(KERN_ERR "Unable to read register save area.\n");
873 		return -ENOENT;
874 	}
875 	pr_debug("--------CPU State Data------------\n");
876 	pr_debug("Magic Number: %llx\n", be64_to_cpu(reg_header->magic_number));
877 	pr_debug("NumCpuOffset: %x\n", be32_to_cpu(reg_header->num_cpu_offset));
878 
879 	vaddr += be32_to_cpu(reg_header->num_cpu_offset);
880 	num_cpus = be32_to_cpu(*((__be32 *)(vaddr)));
881 	pr_debug("NumCpus     : %u\n", num_cpus);
882 	vaddr += sizeof(u32);
883 	reg_entry = (struct fadump_reg_entry *)vaddr;
884 
885 	/* Allocate buffer to hold cpu crash notes. */
886 	fw_dump.cpu_notes_buf_size = num_cpus * sizeof(note_buf_t);
887 	fw_dump.cpu_notes_buf_size = PAGE_ALIGN(fw_dump.cpu_notes_buf_size);
888 	note_buf = fadump_cpu_notes_buf_alloc(fw_dump.cpu_notes_buf_size);
889 	if (!note_buf) {
890 		printk(KERN_ERR "Failed to allocate 0x%lx bytes for "
891 			"cpu notes buffer\n", fw_dump.cpu_notes_buf_size);
892 		return -ENOMEM;
893 	}
894 	fw_dump.cpu_notes_buf = __pa(note_buf);
895 
896 	pr_debug("Allocated buffer for cpu notes of size %ld at %p\n",
897 			(num_cpus * sizeof(note_buf_t)), note_buf);
898 
899 	if (fw_dump.fadumphdr_addr)
900 		fdh = __va(fw_dump.fadumphdr_addr);
901 
902 	for (i = 0; i < num_cpus; i++) {
903 		if (be64_to_cpu(reg_entry->reg_id) != REG_ID("CPUSTRT")) {
904 			printk(KERN_ERR "Unable to read CPU state data\n");
905 			rc = -ENOENT;
906 			goto error_out;
907 		}
908 		/* Lower 4 bytes of reg_value contains logical cpu id */
909 		cpu = be64_to_cpu(reg_entry->reg_value) & FADUMP_CPU_ID_MASK;
910 		if (fdh && !cpumask_test_cpu(cpu, &fdh->online_mask)) {
911 			SKIP_TO_NEXT_CPU(reg_entry);
912 			continue;
913 		}
914 		pr_debug("Reading register data for cpu %d...\n", cpu);
915 		if (fdh && fdh->crashing_cpu == cpu) {
916 			regs = fdh->regs;
917 			note_buf = fadump_regs_to_elf_notes(note_buf, &regs);
918 			SKIP_TO_NEXT_CPU(reg_entry);
919 		} else {
920 			reg_entry++;
921 			reg_entry = fadump_read_registers(reg_entry, &regs);
922 			note_buf = fadump_regs_to_elf_notes(note_buf, &regs);
923 		}
924 	}
925 	final_note(note_buf);
926 
927 	if (fdh) {
928 		pr_debug("Updating elfcore header (%llx) with cpu notes\n",
929 							fdh->elfcorehdr_addr);
930 		fadump_update_elfcore_header((char *)__va(fdh->elfcorehdr_addr));
931 	}
932 	return 0;
933 
934 error_out:
935 	fadump_cpu_notes_buf_free((unsigned long)__va(fw_dump.cpu_notes_buf),
936 					fw_dump.cpu_notes_buf_size);
937 	fw_dump.cpu_notes_buf = 0;
938 	fw_dump.cpu_notes_buf_size = 0;
939 	return rc;
940 
941 }
942 
943 /*
944  * Validate and process the dump data stored by firmware before exporting
945  * it through '/proc/vmcore'.
946  */
947 static int __init process_fadump(const struct fadump_mem_struct *fdm_active)
948 {
949 	struct fadump_crash_info_header *fdh;
950 	int rc = 0;
951 
952 	if (!fdm_active || !fw_dump.fadumphdr_addr)
953 		return -EINVAL;
954 
955 	/* Check if the dump data is valid. */
956 	if ((be16_to_cpu(fdm_active->header.dump_status_flag) == FADUMP_ERROR_FLAG) ||
957 			(fdm_active->cpu_state_data.error_flags != 0) ||
958 			(fdm_active->rmr_region.error_flags != 0)) {
959 		printk(KERN_ERR "Dump taken by platform is not valid\n");
960 		return -EINVAL;
961 	}
962 	if ((fdm_active->rmr_region.bytes_dumped !=
963 			fdm_active->rmr_region.source_len) ||
964 			!fdm_active->cpu_state_data.bytes_dumped) {
965 		printk(KERN_ERR "Dump taken by platform is incomplete\n");
966 		return -EINVAL;
967 	}
968 
969 	/* Validate the fadump crash info header */
970 	fdh = __va(fw_dump.fadumphdr_addr);
971 	if (fdh->magic_number != FADUMP_CRASH_INFO_MAGIC) {
972 		printk(KERN_ERR "Crash info header is not valid.\n");
973 		return -EINVAL;
974 	}
975 
976 	rc = fadump_build_cpu_notes(fdm_active);
977 	if (rc)
978 		return rc;
979 
980 	/*
981 	 * We are done validating dump info and elfcore header is now ready
982 	 * to be exported. set elfcorehdr_addr so that vmcore module will
983 	 * export the elfcore header through '/proc/vmcore'.
984 	 */
985 	elfcorehdr_addr = fdh->elfcorehdr_addr;
986 
987 	return 0;
988 }
989 
990 static void free_crash_memory_ranges(void)
991 {
992 	kfree(crash_memory_ranges);
993 	crash_memory_ranges = NULL;
994 	crash_memory_ranges_size = 0;
995 	max_crash_mem_ranges = 0;
996 }
997 
998 /*
999  * Allocate or reallocate crash memory ranges array in incremental units
1000  * of PAGE_SIZE.
1001  */
1002 static int allocate_crash_memory_ranges(void)
1003 {
1004 	struct fad_crash_memory_ranges *new_array;
1005 	u64 new_size;
1006 
1007 	new_size = crash_memory_ranges_size + PAGE_SIZE;
1008 	pr_debug("Allocating %llu bytes of memory for crash memory ranges\n",
1009 		 new_size);
1010 
1011 	new_array = krealloc(crash_memory_ranges, new_size, GFP_KERNEL);
1012 	if (new_array == NULL) {
1013 		pr_err("Insufficient memory for setting up crash memory ranges\n");
1014 		free_crash_memory_ranges();
1015 		return -ENOMEM;
1016 	}
1017 
1018 	crash_memory_ranges = new_array;
1019 	crash_memory_ranges_size = new_size;
1020 	max_crash_mem_ranges = (new_size /
1021 				sizeof(struct fad_crash_memory_ranges));
1022 	return 0;
1023 }
1024 
1025 static inline int fadump_add_crash_memory(unsigned long long base,
1026 					  unsigned long long end)
1027 {
1028 	u64  start, size;
1029 	bool is_adjacent = false;
1030 
1031 	if (base == end)
1032 		return 0;
1033 
1034 	/*
1035 	 * Fold adjacent memory ranges to bring down the memory ranges/
1036 	 * PT_LOAD segments count.
1037 	 */
1038 	if (crash_mem_ranges) {
1039 		start = crash_memory_ranges[crash_mem_ranges - 1].base;
1040 		size = crash_memory_ranges[crash_mem_ranges - 1].size;
1041 
1042 		if ((start + size) == base)
1043 			is_adjacent = true;
1044 	}
1045 	if (!is_adjacent) {
1046 		/* resize the array on reaching the limit */
1047 		if (crash_mem_ranges == max_crash_mem_ranges) {
1048 			int ret;
1049 
1050 			ret = allocate_crash_memory_ranges();
1051 			if (ret)
1052 				return ret;
1053 		}
1054 
1055 		start = base;
1056 		crash_memory_ranges[crash_mem_ranges].base = start;
1057 		crash_mem_ranges++;
1058 	}
1059 
1060 	crash_memory_ranges[crash_mem_ranges - 1].size = (end - start);
1061 	pr_debug("crash_memory_range[%d] [%#016llx-%#016llx], %#llx bytes\n",
1062 		(crash_mem_ranges - 1), start, end - 1, (end - start));
1063 	return 0;
1064 }
1065 
1066 static int fadump_exclude_reserved_area(unsigned long long start,
1067 					unsigned long long end)
1068 {
1069 	unsigned long long ra_start, ra_end;
1070 	int ret = 0;
1071 
1072 	ra_start = fw_dump.reserve_dump_area_start;
1073 	ra_end = ra_start + fw_dump.reserve_dump_area_size;
1074 
1075 	if ((ra_start < end) && (ra_end > start)) {
1076 		if ((start < ra_start) && (end > ra_end)) {
1077 			ret = fadump_add_crash_memory(start, ra_start);
1078 			if (ret)
1079 				return ret;
1080 
1081 			ret = fadump_add_crash_memory(ra_end, end);
1082 		} else if (start < ra_start) {
1083 			ret = fadump_add_crash_memory(start, ra_start);
1084 		} else if (ra_end < end) {
1085 			ret = fadump_add_crash_memory(ra_end, end);
1086 		}
1087 	} else
1088 		ret = fadump_add_crash_memory(start, end);
1089 
1090 	return ret;
1091 }
1092 
1093 static int fadump_init_elfcore_header(char *bufp)
1094 {
1095 	struct elfhdr *elf;
1096 
1097 	elf = (struct elfhdr *) bufp;
1098 	bufp += sizeof(struct elfhdr);
1099 	memcpy(elf->e_ident, ELFMAG, SELFMAG);
1100 	elf->e_ident[EI_CLASS] = ELF_CLASS;
1101 	elf->e_ident[EI_DATA] = ELF_DATA;
1102 	elf->e_ident[EI_VERSION] = EV_CURRENT;
1103 	elf->e_ident[EI_OSABI] = ELF_OSABI;
1104 	memset(elf->e_ident+EI_PAD, 0, EI_NIDENT-EI_PAD);
1105 	elf->e_type = ET_CORE;
1106 	elf->e_machine = ELF_ARCH;
1107 	elf->e_version = EV_CURRENT;
1108 	elf->e_entry = 0;
1109 	elf->e_phoff = sizeof(struct elfhdr);
1110 	elf->e_shoff = 0;
1111 #if defined(_CALL_ELF)
1112 	elf->e_flags = _CALL_ELF;
1113 #else
1114 	elf->e_flags = 0;
1115 #endif
1116 	elf->e_ehsize = sizeof(struct elfhdr);
1117 	elf->e_phentsize = sizeof(struct elf_phdr);
1118 	elf->e_phnum = 0;
1119 	elf->e_shentsize = 0;
1120 	elf->e_shnum = 0;
1121 	elf->e_shstrndx = 0;
1122 
1123 	return 0;
1124 }
1125 
1126 /*
1127  * Traverse through memblock structure and setup crash memory ranges. These
1128  * ranges will be used create PT_LOAD program headers in elfcore header.
1129  */
1130 static int fadump_setup_crash_memory_ranges(void)
1131 {
1132 	struct memblock_region *reg;
1133 	unsigned long long start, end;
1134 	int ret;
1135 
1136 	pr_debug("Setup crash memory ranges.\n");
1137 	crash_mem_ranges = 0;
1138 
1139 	/*
1140 	 * add the first memory chunk (RMA_START through boot_memory_size) as
1141 	 * a separate memory chunk. The reason is, at the time crash firmware
1142 	 * will move the content of this memory chunk to different location
1143 	 * specified during fadump registration. We need to create a separate
1144 	 * program header for this chunk with the correct offset.
1145 	 */
1146 	ret = fadump_add_crash_memory(RMA_START, fw_dump.boot_memory_size);
1147 	if (ret)
1148 		return ret;
1149 
1150 	for_each_memblock(memory, reg) {
1151 		start = (unsigned long long)reg->base;
1152 		end = start + (unsigned long long)reg->size;
1153 
1154 		/*
1155 		 * skip the first memory chunk that is already added (RMA_START
1156 		 * through boot_memory_size). This logic needs a relook if and
1157 		 * when RMA_START changes to a non-zero value.
1158 		 */
1159 		BUILD_BUG_ON(RMA_START != 0);
1160 		if (start < fw_dump.boot_memory_size) {
1161 			if (end > fw_dump.boot_memory_size)
1162 				start = fw_dump.boot_memory_size;
1163 			else
1164 				continue;
1165 		}
1166 
1167 		/* add this range excluding the reserved dump area. */
1168 		ret = fadump_exclude_reserved_area(start, end);
1169 		if (ret)
1170 			return ret;
1171 	}
1172 
1173 	return 0;
1174 }
1175 
1176 /*
1177  * If the given physical address falls within the boot memory region then
1178  * return the relocated address that points to the dump region reserved
1179  * for saving initial boot memory contents.
1180  */
1181 static inline unsigned long fadump_relocate(unsigned long paddr)
1182 {
1183 	if (paddr > RMA_START && paddr < fw_dump.boot_memory_size)
1184 		return be64_to_cpu(fdm.rmr_region.destination_address) + paddr;
1185 	else
1186 		return paddr;
1187 }
1188 
1189 static int fadump_create_elfcore_headers(char *bufp)
1190 {
1191 	struct elfhdr *elf;
1192 	struct elf_phdr *phdr;
1193 	int i;
1194 
1195 	fadump_init_elfcore_header(bufp);
1196 	elf = (struct elfhdr *)bufp;
1197 	bufp += sizeof(struct elfhdr);
1198 
1199 	/*
1200 	 * setup ELF PT_NOTE, place holder for cpu notes info. The notes info
1201 	 * will be populated during second kernel boot after crash. Hence
1202 	 * this PT_NOTE will always be the first elf note.
1203 	 *
1204 	 * NOTE: Any new ELF note addition should be placed after this note.
1205 	 */
1206 	phdr = (struct elf_phdr *)bufp;
1207 	bufp += sizeof(struct elf_phdr);
1208 	phdr->p_type = PT_NOTE;
1209 	phdr->p_flags = 0;
1210 	phdr->p_vaddr = 0;
1211 	phdr->p_align = 0;
1212 
1213 	phdr->p_offset = 0;
1214 	phdr->p_paddr = 0;
1215 	phdr->p_filesz = 0;
1216 	phdr->p_memsz = 0;
1217 
1218 	(elf->e_phnum)++;
1219 
1220 	/* setup ELF PT_NOTE for vmcoreinfo */
1221 	phdr = (struct elf_phdr *)bufp;
1222 	bufp += sizeof(struct elf_phdr);
1223 	phdr->p_type	= PT_NOTE;
1224 	phdr->p_flags	= 0;
1225 	phdr->p_vaddr	= 0;
1226 	phdr->p_align	= 0;
1227 
1228 	phdr->p_paddr	= fadump_relocate(paddr_vmcoreinfo_note());
1229 	phdr->p_offset	= phdr->p_paddr;
1230 	phdr->p_memsz	= phdr->p_filesz = VMCOREINFO_NOTE_SIZE;
1231 
1232 	/* Increment number of program headers. */
1233 	(elf->e_phnum)++;
1234 
1235 	/* setup PT_LOAD sections. */
1236 
1237 	for (i = 0; i < crash_mem_ranges; i++) {
1238 		unsigned long long mbase, msize;
1239 		mbase = crash_memory_ranges[i].base;
1240 		msize = crash_memory_ranges[i].size;
1241 
1242 		if (!msize)
1243 			continue;
1244 
1245 		phdr = (struct elf_phdr *)bufp;
1246 		bufp += sizeof(struct elf_phdr);
1247 		phdr->p_type	= PT_LOAD;
1248 		phdr->p_flags	= PF_R|PF_W|PF_X;
1249 		phdr->p_offset	= mbase;
1250 
1251 		if (mbase == RMA_START) {
1252 			/*
1253 			 * The entire RMA region will be moved by firmware
1254 			 * to the specified destination_address. Hence set
1255 			 * the correct offset.
1256 			 */
1257 			phdr->p_offset = be64_to_cpu(fdm.rmr_region.destination_address);
1258 		}
1259 
1260 		phdr->p_paddr = mbase;
1261 		phdr->p_vaddr = (unsigned long)__va(mbase);
1262 		phdr->p_filesz = msize;
1263 		phdr->p_memsz = msize;
1264 		phdr->p_align = 0;
1265 
1266 		/* Increment number of program headers. */
1267 		(elf->e_phnum)++;
1268 	}
1269 	return 0;
1270 }
1271 
1272 static unsigned long init_fadump_header(unsigned long addr)
1273 {
1274 	struct fadump_crash_info_header *fdh;
1275 
1276 	if (!addr)
1277 		return 0;
1278 
1279 	fw_dump.fadumphdr_addr = addr;
1280 	fdh = __va(addr);
1281 	addr += sizeof(struct fadump_crash_info_header);
1282 
1283 	memset(fdh, 0, sizeof(struct fadump_crash_info_header));
1284 	fdh->magic_number = FADUMP_CRASH_INFO_MAGIC;
1285 	fdh->elfcorehdr_addr = addr;
1286 	/* We will set the crashing cpu id in crash_fadump() during crash. */
1287 	fdh->crashing_cpu = CPU_UNKNOWN;
1288 
1289 	return addr;
1290 }
1291 
1292 static int register_fadump(void)
1293 {
1294 	unsigned long addr;
1295 	void *vaddr;
1296 	int ret;
1297 
1298 	/*
1299 	 * If no memory is reserved then we can not register for firmware-
1300 	 * assisted dump.
1301 	 */
1302 	if (!fw_dump.reserve_dump_area_size)
1303 		return -ENODEV;
1304 
1305 	ret = fadump_setup_crash_memory_ranges();
1306 	if (ret)
1307 		return ret;
1308 
1309 	addr = be64_to_cpu(fdm.rmr_region.destination_address) + be64_to_cpu(fdm.rmr_region.source_len);
1310 	/* Initialize fadump crash info header. */
1311 	addr = init_fadump_header(addr);
1312 	vaddr = __va(addr);
1313 
1314 	pr_debug("Creating ELF core headers at %#016lx\n", addr);
1315 	fadump_create_elfcore_headers(vaddr);
1316 
1317 	/* register the future kernel dump with firmware. */
1318 	return register_fw_dump(&fdm);
1319 }
1320 
1321 static int fadump_unregister_dump(struct fadump_mem_struct *fdm)
1322 {
1323 	int rc = 0;
1324 	unsigned int wait_time;
1325 
1326 	pr_debug("Un-register firmware-assisted dump\n");
1327 
1328 	/* TODO: Add upper time limit for the delay */
1329 	do {
1330 		rc = rtas_call(fw_dump.ibm_configure_kernel_dump, 3, 1, NULL,
1331 			FADUMP_UNREGISTER, fdm,
1332 			sizeof(struct fadump_mem_struct));
1333 
1334 		wait_time = rtas_busy_delay_time(rc);
1335 		if (wait_time)
1336 			mdelay(wait_time);
1337 	} while (wait_time);
1338 
1339 	if (rc) {
1340 		printk(KERN_ERR "Failed to un-register firmware-assisted dump."
1341 			" unexpected error(%d).\n", rc);
1342 		return rc;
1343 	}
1344 	fw_dump.dump_registered = 0;
1345 	return 0;
1346 }
1347 
1348 static int fadump_invalidate_dump(const struct fadump_mem_struct *fdm)
1349 {
1350 	int rc = 0;
1351 	unsigned int wait_time;
1352 
1353 	pr_debug("Invalidating firmware-assisted dump registration\n");
1354 
1355 	/* TODO: Add upper time limit for the delay */
1356 	do {
1357 		rc = rtas_call(fw_dump.ibm_configure_kernel_dump, 3, 1, NULL,
1358 			FADUMP_INVALIDATE, fdm,
1359 			sizeof(struct fadump_mem_struct));
1360 
1361 		wait_time = rtas_busy_delay_time(rc);
1362 		if (wait_time)
1363 			mdelay(wait_time);
1364 	} while (wait_time);
1365 
1366 	if (rc) {
1367 		pr_err("Failed to invalidate firmware-assisted dump registration. Unexpected error (%d).\n", rc);
1368 		return rc;
1369 	}
1370 	fw_dump.dump_active = 0;
1371 	fdm_active = NULL;
1372 	return 0;
1373 }
1374 
1375 void fadump_cleanup(void)
1376 {
1377 	/* Invalidate the registration only if dump is active. */
1378 	if (fw_dump.dump_active) {
1379 		/* pass the same memory dump structure provided by platform */
1380 		fadump_invalidate_dump(fdm_active);
1381 	} else if (fw_dump.dump_registered) {
1382 		/* Un-register Firmware-assisted dump if it was registered. */
1383 		fadump_unregister_dump(&fdm);
1384 		free_crash_memory_ranges();
1385 	}
1386 }
1387 
1388 static void fadump_free_reserved_memory(unsigned long start_pfn,
1389 					unsigned long end_pfn)
1390 {
1391 	unsigned long pfn;
1392 	unsigned long time_limit = jiffies + HZ;
1393 
1394 	pr_info("freeing reserved memory (0x%llx - 0x%llx)\n",
1395 		PFN_PHYS(start_pfn), PFN_PHYS(end_pfn));
1396 
1397 	for (pfn = start_pfn; pfn < end_pfn; pfn++) {
1398 		free_reserved_page(pfn_to_page(pfn));
1399 
1400 		if (time_after(jiffies, time_limit)) {
1401 			cond_resched();
1402 			time_limit = jiffies + HZ;
1403 		}
1404 	}
1405 }
1406 
1407 /*
1408  * Skip memory holes and free memory that was actually reserved.
1409  */
1410 static void fadump_release_reserved_area(unsigned long start, unsigned long end)
1411 {
1412 	struct memblock_region *reg;
1413 	unsigned long tstart, tend;
1414 	unsigned long start_pfn = PHYS_PFN(start);
1415 	unsigned long end_pfn = PHYS_PFN(end);
1416 
1417 	for_each_memblock(memory, reg) {
1418 		tstart = max(start_pfn, memblock_region_memory_base_pfn(reg));
1419 		tend = min(end_pfn, memblock_region_memory_end_pfn(reg));
1420 		if (tstart < tend) {
1421 			fadump_free_reserved_memory(tstart, tend);
1422 
1423 			if (tend == end_pfn)
1424 				break;
1425 
1426 			start_pfn = tend + 1;
1427 		}
1428 	}
1429 }
1430 
1431 /*
1432  * Release the memory that was reserved in early boot to preserve the memory
1433  * contents. The released memory will be available for general use.
1434  */
1435 static void fadump_release_memory(unsigned long begin, unsigned long end)
1436 {
1437 	unsigned long ra_start, ra_end;
1438 
1439 	ra_start = fw_dump.reserve_dump_area_start;
1440 	ra_end = ra_start + fw_dump.reserve_dump_area_size;
1441 
1442 	/*
1443 	 * exclude the dump reserve area. Will reuse it for next
1444 	 * fadump registration.
1445 	 */
1446 	if (begin < ra_end && end > ra_start) {
1447 		if (begin < ra_start)
1448 			fadump_release_reserved_area(begin, ra_start);
1449 		if (end > ra_end)
1450 			fadump_release_reserved_area(ra_end, end);
1451 	} else
1452 		fadump_release_reserved_area(begin, end);
1453 }
1454 
1455 static void fadump_invalidate_release_mem(void)
1456 {
1457 	unsigned long reserved_area_start, reserved_area_end;
1458 	unsigned long destination_address;
1459 
1460 	mutex_lock(&fadump_mutex);
1461 	if (!fw_dump.dump_active) {
1462 		mutex_unlock(&fadump_mutex);
1463 		return;
1464 	}
1465 
1466 	destination_address = be64_to_cpu(fdm_active->cpu_state_data.destination_address);
1467 	fadump_cleanup();
1468 	mutex_unlock(&fadump_mutex);
1469 
1470 	/*
1471 	 * Save the current reserved memory bounds we will require them
1472 	 * later for releasing the memory for general use.
1473 	 */
1474 	reserved_area_start = fw_dump.reserve_dump_area_start;
1475 	reserved_area_end = reserved_area_start +
1476 			fw_dump.reserve_dump_area_size;
1477 	/*
1478 	 * Setup reserve_dump_area_start and its size so that we can
1479 	 * reuse this reserved memory for Re-registration.
1480 	 */
1481 	fw_dump.reserve_dump_area_start = destination_address;
1482 	fw_dump.reserve_dump_area_size = get_fadump_area_size();
1483 
1484 	fadump_release_memory(reserved_area_start, reserved_area_end);
1485 	if (fw_dump.cpu_notes_buf) {
1486 		fadump_cpu_notes_buf_free(
1487 				(unsigned long)__va(fw_dump.cpu_notes_buf),
1488 				fw_dump.cpu_notes_buf_size);
1489 		fw_dump.cpu_notes_buf = 0;
1490 		fw_dump.cpu_notes_buf_size = 0;
1491 	}
1492 	/* Initialize the kernel dump memory structure for FAD registration. */
1493 	init_fadump_mem_struct(&fdm, fw_dump.reserve_dump_area_start);
1494 }
1495 
1496 static ssize_t fadump_release_memory_store(struct kobject *kobj,
1497 					struct kobj_attribute *attr,
1498 					const char *buf, size_t count)
1499 {
1500 	int input = -1;
1501 
1502 	if (!fw_dump.dump_active)
1503 		return -EPERM;
1504 
1505 	if (kstrtoint(buf, 0, &input))
1506 		return -EINVAL;
1507 
1508 	if (input == 1) {
1509 		/*
1510 		 * Take away the '/proc/vmcore'. We are releasing the dump
1511 		 * memory, hence it will not be valid anymore.
1512 		 */
1513 #ifdef CONFIG_PROC_VMCORE
1514 		vmcore_cleanup();
1515 #endif
1516 		fadump_invalidate_release_mem();
1517 
1518 	} else
1519 		return -EINVAL;
1520 	return count;
1521 }
1522 
1523 static ssize_t fadump_enabled_show(struct kobject *kobj,
1524 					struct kobj_attribute *attr,
1525 					char *buf)
1526 {
1527 	return sprintf(buf, "%d\n", fw_dump.fadump_enabled);
1528 }
1529 
1530 static ssize_t fadump_register_show(struct kobject *kobj,
1531 					struct kobj_attribute *attr,
1532 					char *buf)
1533 {
1534 	return sprintf(buf, "%d\n", fw_dump.dump_registered);
1535 }
1536 
1537 static ssize_t fadump_register_store(struct kobject *kobj,
1538 					struct kobj_attribute *attr,
1539 					const char *buf, size_t count)
1540 {
1541 	int ret = 0;
1542 	int input = -1;
1543 
1544 	if (!fw_dump.fadump_enabled || fdm_active)
1545 		return -EPERM;
1546 
1547 	if (kstrtoint(buf, 0, &input))
1548 		return -EINVAL;
1549 
1550 	mutex_lock(&fadump_mutex);
1551 
1552 	switch (input) {
1553 	case 0:
1554 		if (fw_dump.dump_registered == 0) {
1555 			goto unlock_out;
1556 		}
1557 		/* Un-register Firmware-assisted dump */
1558 		fadump_unregister_dump(&fdm);
1559 		break;
1560 	case 1:
1561 		if (fw_dump.dump_registered == 1) {
1562 			/* Un-register Firmware-assisted dump */
1563 			fadump_unregister_dump(&fdm);
1564 		}
1565 		/* Register Firmware-assisted dump */
1566 		ret = register_fadump();
1567 		break;
1568 	default:
1569 		ret = -EINVAL;
1570 		break;
1571 	}
1572 
1573 unlock_out:
1574 	mutex_unlock(&fadump_mutex);
1575 	return ret < 0 ? ret : count;
1576 }
1577 
1578 static int fadump_region_show(struct seq_file *m, void *private)
1579 {
1580 	const struct fadump_mem_struct *fdm_ptr;
1581 
1582 	if (!fw_dump.fadump_enabled)
1583 		return 0;
1584 
1585 	mutex_lock(&fadump_mutex);
1586 	if (fdm_active)
1587 		fdm_ptr = fdm_active;
1588 	else {
1589 		mutex_unlock(&fadump_mutex);
1590 		fdm_ptr = &fdm;
1591 	}
1592 
1593 	seq_printf(m,
1594 			"CPU : [%#016llx-%#016llx] %#llx bytes, "
1595 			"Dumped: %#llx\n",
1596 			be64_to_cpu(fdm_ptr->cpu_state_data.destination_address),
1597 			be64_to_cpu(fdm_ptr->cpu_state_data.destination_address) +
1598 			be64_to_cpu(fdm_ptr->cpu_state_data.source_len) - 1,
1599 			be64_to_cpu(fdm_ptr->cpu_state_data.source_len),
1600 			be64_to_cpu(fdm_ptr->cpu_state_data.bytes_dumped));
1601 	seq_printf(m,
1602 			"HPTE: [%#016llx-%#016llx] %#llx bytes, "
1603 			"Dumped: %#llx\n",
1604 			be64_to_cpu(fdm_ptr->hpte_region.destination_address),
1605 			be64_to_cpu(fdm_ptr->hpte_region.destination_address) +
1606 			be64_to_cpu(fdm_ptr->hpte_region.source_len) - 1,
1607 			be64_to_cpu(fdm_ptr->hpte_region.source_len),
1608 			be64_to_cpu(fdm_ptr->hpte_region.bytes_dumped));
1609 	seq_printf(m,
1610 			"DUMP: [%#016llx-%#016llx] %#llx bytes, "
1611 			"Dumped: %#llx\n",
1612 			be64_to_cpu(fdm_ptr->rmr_region.destination_address),
1613 			be64_to_cpu(fdm_ptr->rmr_region.destination_address) +
1614 			be64_to_cpu(fdm_ptr->rmr_region.source_len) - 1,
1615 			be64_to_cpu(fdm_ptr->rmr_region.source_len),
1616 			be64_to_cpu(fdm_ptr->rmr_region.bytes_dumped));
1617 
1618 	if (!fdm_active ||
1619 		(fw_dump.reserve_dump_area_start ==
1620 		be64_to_cpu(fdm_ptr->cpu_state_data.destination_address)))
1621 		goto out;
1622 
1623 	/* Dump is active. Show reserved memory region. */
1624 	seq_printf(m,
1625 			"    : [%#016llx-%#016llx] %#llx bytes, "
1626 			"Dumped: %#llx\n",
1627 			(unsigned long long)fw_dump.reserve_dump_area_start,
1628 			be64_to_cpu(fdm_ptr->cpu_state_data.destination_address) - 1,
1629 			be64_to_cpu(fdm_ptr->cpu_state_data.destination_address) -
1630 			fw_dump.reserve_dump_area_start,
1631 			be64_to_cpu(fdm_ptr->cpu_state_data.destination_address) -
1632 			fw_dump.reserve_dump_area_start);
1633 out:
1634 	if (fdm_active)
1635 		mutex_unlock(&fadump_mutex);
1636 	return 0;
1637 }
1638 
1639 static struct kobj_attribute fadump_release_attr = __ATTR(fadump_release_mem,
1640 						0200, NULL,
1641 						fadump_release_memory_store);
1642 static struct kobj_attribute fadump_attr = __ATTR(fadump_enabled,
1643 						0444, fadump_enabled_show,
1644 						NULL);
1645 static struct kobj_attribute fadump_register_attr = __ATTR(fadump_registered,
1646 						0644, fadump_register_show,
1647 						fadump_register_store);
1648 
1649 DEFINE_SHOW_ATTRIBUTE(fadump_region);
1650 
1651 static void fadump_init_files(void)
1652 {
1653 	struct dentry *debugfs_file;
1654 	int rc = 0;
1655 
1656 	rc = sysfs_create_file(kernel_kobj, &fadump_attr.attr);
1657 	if (rc)
1658 		printk(KERN_ERR "fadump: unable to create sysfs file"
1659 			" fadump_enabled (%d)\n", rc);
1660 
1661 	rc = sysfs_create_file(kernel_kobj, &fadump_register_attr.attr);
1662 	if (rc)
1663 		printk(KERN_ERR "fadump: unable to create sysfs file"
1664 			" fadump_registered (%d)\n", rc);
1665 
1666 	debugfs_file = debugfs_create_file("fadump_region", 0444,
1667 					powerpc_debugfs_root, NULL,
1668 					&fadump_region_fops);
1669 	if (!debugfs_file)
1670 		printk(KERN_ERR "fadump: unable to create debugfs file"
1671 				" fadump_region\n");
1672 
1673 	if (fw_dump.dump_active) {
1674 		rc = sysfs_create_file(kernel_kobj, &fadump_release_attr.attr);
1675 		if (rc)
1676 			printk(KERN_ERR "fadump: unable to create sysfs file"
1677 				" fadump_release_mem (%d)\n", rc);
1678 	}
1679 	return;
1680 }
1681 
1682 /*
1683  * Prepare for firmware-assisted dump.
1684  */
1685 int __init setup_fadump(void)
1686 {
1687 	if (!fw_dump.fadump_enabled)
1688 		return 0;
1689 
1690 	if (!fw_dump.fadump_supported) {
1691 		printk(KERN_ERR "Firmware-assisted dump is not supported on"
1692 			" this hardware\n");
1693 		return 0;
1694 	}
1695 
1696 	fadump_show_config();
1697 	/*
1698 	 * If dump data is available then see if it is valid and prepare for
1699 	 * saving it to the disk.
1700 	 */
1701 	if (fw_dump.dump_active) {
1702 		/*
1703 		 * if dump process fails then invalidate the registration
1704 		 * and release memory before proceeding for re-registration.
1705 		 */
1706 		if (process_fadump(fdm_active) < 0)
1707 			fadump_invalidate_release_mem();
1708 	}
1709 	/* Initialize the kernel dump memory structure for FAD registration. */
1710 	else if (fw_dump.reserve_dump_area_size)
1711 		init_fadump_mem_struct(&fdm, fw_dump.reserve_dump_area_start);
1712 	fadump_init_files();
1713 
1714 	return 1;
1715 }
1716 subsys_initcall(setup_fadump);
1717