xref: /openbmc/linux/arch/powerpc/kernel/fadump.c (revision 9a29f5fc)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * Firmware Assisted dump: A robust mechanism to get reliable kernel crash
4  * dump with assistance from firmware. This approach does not use kexec,
5  * instead firmware assists in booting the kdump kernel while preserving
6  * memory contents. The most of the code implementation has been adapted
7  * from phyp assisted dump implementation written by Linas Vepstas and
8  * Manish Ahuja
9  *
10  * Copyright 2011 IBM Corporation
11  * Author: Mahesh Salgaonkar <mahesh@linux.vnet.ibm.com>
12  */
13 
14 #undef DEBUG
15 #define pr_fmt(fmt) "fadump: " fmt
16 
17 #include <linux/string.h>
18 #include <linux/memblock.h>
19 #include <linux/delay.h>
20 #include <linux/seq_file.h>
21 #include <linux/crash_dump.h>
22 #include <linux/kobject.h>
23 #include <linux/sysfs.h>
24 #include <linux/slab.h>
25 #include <linux/cma.h>
26 #include <linux/hugetlb.h>
27 #include <linux/debugfs.h>
28 #include <linux/of.h>
29 #include <linux/of_fdt.h>
30 
31 #include <asm/page.h>
32 #include <asm/fadump.h>
33 #include <asm/fadump-internal.h>
34 #include <asm/setup.h>
35 #include <asm/interrupt.h>
36 
37 /*
38  * The CPU who acquired the lock to trigger the fadump crash should
39  * wait for other CPUs to enter.
40  *
41  * The timeout is in milliseconds.
42  */
43 #define CRASH_TIMEOUT		500
44 
45 static struct fw_dump fw_dump;
46 
47 static void __init fadump_reserve_crash_area(u64 base);
48 
49 #ifndef CONFIG_PRESERVE_FA_DUMP
50 
51 static struct kobject *fadump_kobj;
52 
53 static atomic_t cpus_in_fadump;
54 static DEFINE_MUTEX(fadump_mutex);
55 
56 static struct fadump_mrange_info crash_mrange_info = { "crash", NULL, 0, 0, 0, false };
57 
58 #define RESERVED_RNGS_SZ	16384 /* 16K - 128 entries */
59 #define RESERVED_RNGS_CNT	(RESERVED_RNGS_SZ / \
60 				 sizeof(struct fadump_memory_range))
61 static struct fadump_memory_range rngs[RESERVED_RNGS_CNT];
62 static struct fadump_mrange_info
63 reserved_mrange_info = { "reserved", rngs, RESERVED_RNGS_SZ, 0, RESERVED_RNGS_CNT, true };
64 
65 static void __init early_init_dt_scan_reserved_ranges(unsigned long node);
66 
67 #ifdef CONFIG_CMA
68 static struct cma *fadump_cma;
69 
70 /*
71  * fadump_cma_init() - Initialize CMA area from a fadump reserved memory
72  *
73  * This function initializes CMA area from fadump reserved memory.
74  * The total size of fadump reserved memory covers for boot memory size
75  * + cpu data size + hpte size and metadata.
76  * Initialize only the area equivalent to boot memory size for CMA use.
77  * The remaining portion of fadump reserved memory will be not given
78  * to CMA and pages for those will stay reserved. boot memory size is
79  * aligned per CMA requirement to satisy cma_init_reserved_mem() call.
80  * But for some reason even if it fails we still have the memory reservation
81  * with us and we can still continue doing fadump.
82  */
83 static int __init fadump_cma_init(void)
84 {
85 	unsigned long long base, size;
86 	int rc;
87 
88 	if (!fw_dump.fadump_enabled)
89 		return 0;
90 
91 	/*
92 	 * Do not use CMA if user has provided fadump=nocma kernel parameter.
93 	 * Return 1 to continue with fadump old behaviour.
94 	 */
95 	if (fw_dump.nocma)
96 		return 1;
97 
98 	base = fw_dump.reserve_dump_area_start;
99 	size = fw_dump.boot_memory_size;
100 
101 	if (!size)
102 		return 0;
103 
104 	rc = cma_init_reserved_mem(base, size, 0, "fadump_cma", &fadump_cma);
105 	if (rc) {
106 		pr_err("Failed to init cma area for firmware-assisted dump,%d\n", rc);
107 		/*
108 		 * Though the CMA init has failed we still have memory
109 		 * reservation with us. The reserved memory will be
110 		 * blocked from production system usage.  Hence return 1,
111 		 * so that we can continue with fadump.
112 		 */
113 		return 1;
114 	}
115 
116 	/*
117 	 *  If CMA activation fails, keep the pages reserved, instead of
118 	 *  exposing them to buddy allocator. Same as 'fadump=nocma' case.
119 	 */
120 	cma_reserve_pages_on_error(fadump_cma);
121 
122 	/*
123 	 * So we now have successfully initialized cma area for fadump.
124 	 */
125 	pr_info("Initialized 0x%lx bytes cma area at %ldMB from 0x%lx "
126 		"bytes of memory reserved for firmware-assisted dump\n",
127 		cma_get_size(fadump_cma),
128 		(unsigned long)cma_get_base(fadump_cma) >> 20,
129 		fw_dump.reserve_dump_area_size);
130 	return 1;
131 }
132 #else
133 static int __init fadump_cma_init(void) { return 1; }
134 #endif /* CONFIG_CMA */
135 
136 /* Scan the Firmware Assisted dump configuration details. */
137 int __init early_init_dt_scan_fw_dump(unsigned long node, const char *uname,
138 				      int depth, void *data)
139 {
140 	if (depth == 0) {
141 		early_init_dt_scan_reserved_ranges(node);
142 		return 0;
143 	}
144 
145 	if (depth != 1)
146 		return 0;
147 
148 	if (strcmp(uname, "rtas") == 0) {
149 		rtas_fadump_dt_scan(&fw_dump, node);
150 		return 1;
151 	}
152 
153 	if (strcmp(uname, "ibm,opal") == 0) {
154 		opal_fadump_dt_scan(&fw_dump, node);
155 		return 1;
156 	}
157 
158 	return 0;
159 }
160 
161 /*
162  * If fadump is registered, check if the memory provided
163  * falls within boot memory area and reserved memory area.
164  */
165 int is_fadump_memory_area(u64 addr, unsigned long size)
166 {
167 	u64 d_start, d_end;
168 
169 	if (!fw_dump.dump_registered)
170 		return 0;
171 
172 	if (!size)
173 		return 0;
174 
175 	d_start = fw_dump.reserve_dump_area_start;
176 	d_end = d_start + fw_dump.reserve_dump_area_size;
177 	if (((addr + size) > d_start) && (addr <= d_end))
178 		return 1;
179 
180 	return (addr <= fw_dump.boot_mem_top);
181 }
182 
183 int should_fadump_crash(void)
184 {
185 	if (!fw_dump.dump_registered || !fw_dump.fadumphdr_addr)
186 		return 0;
187 	return 1;
188 }
189 
190 int is_fadump_active(void)
191 {
192 	return fw_dump.dump_active;
193 }
194 
195 /*
196  * Returns true, if there are no holes in memory area between d_start to d_end,
197  * false otherwise.
198  */
199 static bool is_fadump_mem_area_contiguous(u64 d_start, u64 d_end)
200 {
201 	phys_addr_t reg_start, reg_end;
202 	bool ret = false;
203 	u64 i, start, end;
204 
205 	for_each_mem_range(i, &reg_start, &reg_end) {
206 		start = max_t(u64, d_start, reg_start);
207 		end = min_t(u64, d_end, reg_end);
208 		if (d_start < end) {
209 			/* Memory hole from d_start to start */
210 			if (start > d_start)
211 				break;
212 
213 			if (end == d_end) {
214 				ret = true;
215 				break;
216 			}
217 
218 			d_start = end + 1;
219 		}
220 	}
221 
222 	return ret;
223 }
224 
225 /*
226  * Returns true, if there are no holes in boot memory area,
227  * false otherwise.
228  */
229 bool is_fadump_boot_mem_contiguous(void)
230 {
231 	unsigned long d_start, d_end;
232 	bool ret = false;
233 	int i;
234 
235 	for (i = 0; i < fw_dump.boot_mem_regs_cnt; i++) {
236 		d_start = fw_dump.boot_mem_addr[i];
237 		d_end   = d_start + fw_dump.boot_mem_sz[i];
238 
239 		ret = is_fadump_mem_area_contiguous(d_start, d_end);
240 		if (!ret)
241 			break;
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 bool is_fadump_reserved_mem_contiguous(void)
252 {
253 	u64 d_start, d_end;
254 
255 	d_start	= fw_dump.reserve_dump_area_start;
256 	d_end	= d_start + fw_dump.reserve_dump_area_size;
257 	return is_fadump_mem_area_contiguous(d_start, d_end);
258 }
259 
260 /* Print firmware assisted dump configurations for debugging purpose. */
261 static void __init fadump_show_config(void)
262 {
263 	int i;
264 
265 	pr_debug("Support for firmware-assisted dump (fadump): %s\n",
266 			(fw_dump.fadump_supported ? "present" : "no support"));
267 
268 	if (!fw_dump.fadump_supported)
269 		return;
270 
271 	pr_debug("Fadump enabled    : %s\n",
272 				(fw_dump.fadump_enabled ? "yes" : "no"));
273 	pr_debug("Dump Active       : %s\n",
274 				(fw_dump.dump_active ? "yes" : "no"));
275 	pr_debug("Dump section sizes:\n");
276 	pr_debug("    CPU state data size: %lx\n", fw_dump.cpu_state_data_size);
277 	pr_debug("    HPTE region size   : %lx\n", fw_dump.hpte_region_size);
278 	pr_debug("    Boot memory size   : %lx\n", fw_dump.boot_memory_size);
279 	pr_debug("    Boot memory top    : %llx\n", fw_dump.boot_mem_top);
280 	pr_debug("Boot memory regions cnt: %llx\n", fw_dump.boot_mem_regs_cnt);
281 	for (i = 0; i < fw_dump.boot_mem_regs_cnt; i++) {
282 		pr_debug("[%03d] base = %llx, size = %llx\n", i,
283 			 fw_dump.boot_mem_addr[i], fw_dump.boot_mem_sz[i]);
284 	}
285 }
286 
287 /**
288  * fadump_calculate_reserve_size(): reserve variable boot area 5% of System RAM
289  *
290  * Function to find the largest memory size we need to reserve during early
291  * boot process. This will be the size of the memory that is required for a
292  * kernel to boot successfully.
293  *
294  * This function has been taken from phyp-assisted dump feature implementation.
295  *
296  * returns larger of 256MB or 5% rounded down to multiples of 256MB.
297  *
298  * TODO: Come up with better approach to find out more accurate memory size
299  * that is required for a kernel to boot successfully.
300  *
301  */
302 static __init u64 fadump_calculate_reserve_size(void)
303 {
304 	u64 base, size, bootmem_min;
305 	int ret;
306 
307 	if (fw_dump.reserve_bootvar)
308 		pr_warn("'fadump_reserve_mem=' parameter is deprecated in favor of 'crashkernel=' parameter.\n");
309 
310 	/*
311 	 * Check if the size is specified through crashkernel= cmdline
312 	 * option. If yes, then use that but ignore base as fadump reserves
313 	 * memory at a predefined offset.
314 	 */
315 	ret = parse_crashkernel(boot_command_line, memblock_phys_mem_size(),
316 				&size, &base);
317 	if (ret == 0 && size > 0) {
318 		unsigned long max_size;
319 
320 		if (fw_dump.reserve_bootvar)
321 			pr_info("Using 'crashkernel=' parameter for memory reservation.\n");
322 
323 		fw_dump.reserve_bootvar = (unsigned long)size;
324 
325 		/*
326 		 * Adjust if the boot memory size specified is above
327 		 * the upper limit.
328 		 */
329 		max_size = memblock_phys_mem_size() / MAX_BOOT_MEM_RATIO;
330 		if (fw_dump.reserve_bootvar > max_size) {
331 			fw_dump.reserve_bootvar = max_size;
332 			pr_info("Adjusted boot memory size to %luMB\n",
333 				(fw_dump.reserve_bootvar >> 20));
334 		}
335 
336 		return fw_dump.reserve_bootvar;
337 	} else if (fw_dump.reserve_bootvar) {
338 		/*
339 		 * 'fadump_reserve_mem=' is being used to reserve memory
340 		 * for firmware-assisted dump.
341 		 */
342 		return fw_dump.reserve_bootvar;
343 	}
344 
345 	/* divide by 20 to get 5% of value */
346 	size = memblock_phys_mem_size() / 20;
347 
348 	/* round it down in multiples of 256 */
349 	size = size & ~0x0FFFFFFFUL;
350 
351 	/* Truncate to memory_limit. We don't want to over reserve the memory.*/
352 	if (memory_limit && size > memory_limit)
353 		size = memory_limit;
354 
355 	bootmem_min = fw_dump.ops->fadump_get_bootmem_min();
356 	return (size > bootmem_min ? size : bootmem_min);
357 }
358 
359 /*
360  * Calculate the total memory size required to be reserved for
361  * firmware-assisted dump registration.
362  */
363 static unsigned long __init get_fadump_area_size(void)
364 {
365 	unsigned long size = 0;
366 
367 	size += fw_dump.cpu_state_data_size;
368 	size += fw_dump.hpte_region_size;
369 	/*
370 	 * Account for pagesize alignment of boot memory area destination address.
371 	 * This faciliates in mmap reading of first kernel's memory.
372 	 */
373 	size = PAGE_ALIGN(size);
374 	size += fw_dump.boot_memory_size;
375 	size += sizeof(struct fadump_crash_info_header);
376 	size += sizeof(struct elfhdr); /* ELF core header.*/
377 	size += sizeof(struct elf_phdr); /* place holder for cpu notes */
378 	/* Program headers for crash memory regions. */
379 	size += sizeof(struct elf_phdr) * (memblock_num_regions(memory) + 2);
380 
381 	size = PAGE_ALIGN(size);
382 
383 	/* This is to hold kernel metadata on platforms that support it */
384 	size += (fw_dump.ops->fadump_get_metadata_size ?
385 		 fw_dump.ops->fadump_get_metadata_size() : 0);
386 	return size;
387 }
388 
389 static int __init add_boot_mem_region(unsigned long rstart,
390 				      unsigned long rsize)
391 {
392 	int i = fw_dump.boot_mem_regs_cnt++;
393 
394 	if (fw_dump.boot_mem_regs_cnt > FADUMP_MAX_MEM_REGS) {
395 		fw_dump.boot_mem_regs_cnt = FADUMP_MAX_MEM_REGS;
396 		return 0;
397 	}
398 
399 	pr_debug("Added boot memory range[%d] [%#016lx-%#016lx)\n",
400 		 i, rstart, (rstart + rsize));
401 	fw_dump.boot_mem_addr[i] = rstart;
402 	fw_dump.boot_mem_sz[i] = rsize;
403 	return 1;
404 }
405 
406 /*
407  * Firmware usually has a hard limit on the data it can copy per region.
408  * Honour that by splitting a memory range into multiple regions.
409  */
410 static int __init add_boot_mem_regions(unsigned long mstart,
411 				       unsigned long msize)
412 {
413 	unsigned long rstart, rsize, max_size;
414 	int ret = 1;
415 
416 	rstart = mstart;
417 	max_size = fw_dump.max_copy_size ? fw_dump.max_copy_size : msize;
418 	while (msize) {
419 		if (msize > max_size)
420 			rsize = max_size;
421 		else
422 			rsize = msize;
423 
424 		ret = add_boot_mem_region(rstart, rsize);
425 		if (!ret)
426 			break;
427 
428 		msize -= rsize;
429 		rstart += rsize;
430 	}
431 
432 	return ret;
433 }
434 
435 static int __init fadump_get_boot_mem_regions(void)
436 {
437 	unsigned long size, cur_size, hole_size, last_end;
438 	unsigned long mem_size = fw_dump.boot_memory_size;
439 	phys_addr_t reg_start, reg_end;
440 	int ret = 1;
441 	u64 i;
442 
443 	fw_dump.boot_mem_regs_cnt = 0;
444 
445 	last_end = 0;
446 	hole_size = 0;
447 	cur_size = 0;
448 	for_each_mem_range(i, &reg_start, &reg_end) {
449 		size = reg_end - reg_start;
450 		hole_size += (reg_start - last_end);
451 
452 		if ((cur_size + size) >= mem_size) {
453 			size = (mem_size - cur_size);
454 			ret = add_boot_mem_regions(reg_start, size);
455 			break;
456 		}
457 
458 		mem_size -= size;
459 		cur_size += size;
460 		ret = add_boot_mem_regions(reg_start, size);
461 		if (!ret)
462 			break;
463 
464 		last_end = reg_end;
465 	}
466 	fw_dump.boot_mem_top = PAGE_ALIGN(fw_dump.boot_memory_size + hole_size);
467 
468 	return ret;
469 }
470 
471 /*
472  * Returns true, if the given range overlaps with reserved memory ranges
473  * starting at idx. Also, updates idx to index of overlapping memory range
474  * with the given memory range.
475  * False, otherwise.
476  */
477 static bool __init overlaps_reserved_ranges(u64 base, u64 end, int *idx)
478 {
479 	bool ret = false;
480 	int i;
481 
482 	for (i = *idx; i < reserved_mrange_info.mem_range_cnt; i++) {
483 		u64 rbase = reserved_mrange_info.mem_ranges[i].base;
484 		u64 rend = rbase + reserved_mrange_info.mem_ranges[i].size;
485 
486 		if (end <= rbase)
487 			break;
488 
489 		if ((end > rbase) &&  (base < rend)) {
490 			*idx = i;
491 			ret = true;
492 			break;
493 		}
494 	}
495 
496 	return ret;
497 }
498 
499 /*
500  * Locate a suitable memory area to reserve memory for FADump. While at it,
501  * lookup reserved-ranges & avoid overlap with them, as they are used by F/W.
502  */
503 static u64 __init fadump_locate_reserve_mem(u64 base, u64 size)
504 {
505 	struct fadump_memory_range *mrngs;
506 	phys_addr_t mstart, mend;
507 	int idx = 0;
508 	u64 i, ret = 0;
509 
510 	mrngs = reserved_mrange_info.mem_ranges;
511 	for_each_free_mem_range(i, NUMA_NO_NODE, MEMBLOCK_NONE,
512 				&mstart, &mend, NULL) {
513 		pr_debug("%llu) mstart: %llx, mend: %llx, base: %llx\n",
514 			 i, mstart, mend, base);
515 
516 		if (mstart > base)
517 			base = PAGE_ALIGN(mstart);
518 
519 		while ((mend > base) && ((mend - base) >= size)) {
520 			if (!overlaps_reserved_ranges(base, base+size, &idx)) {
521 				ret = base;
522 				goto out;
523 			}
524 
525 			base = mrngs[idx].base + mrngs[idx].size;
526 			base = PAGE_ALIGN(base);
527 		}
528 	}
529 
530 out:
531 	return ret;
532 }
533 
534 int __init fadump_reserve_mem(void)
535 {
536 	u64 base, size, mem_boundary, bootmem_min;
537 	int ret = 1;
538 
539 	if (!fw_dump.fadump_enabled)
540 		return 0;
541 
542 	if (!fw_dump.fadump_supported) {
543 		pr_info("Firmware-Assisted Dump is not supported on this hardware\n");
544 		goto error_out;
545 	}
546 
547 	/*
548 	 * Initialize boot memory size
549 	 * If dump is active then we have already calculated the size during
550 	 * first kernel.
551 	 */
552 	if (!fw_dump.dump_active) {
553 		fw_dump.boot_memory_size =
554 			PAGE_ALIGN(fadump_calculate_reserve_size());
555 #ifdef CONFIG_CMA
556 		if (!fw_dump.nocma) {
557 			fw_dump.boot_memory_size =
558 				ALIGN(fw_dump.boot_memory_size,
559 				      CMA_MIN_ALIGNMENT_BYTES);
560 		}
561 #endif
562 
563 		bootmem_min = fw_dump.ops->fadump_get_bootmem_min();
564 		if (fw_dump.boot_memory_size < bootmem_min) {
565 			pr_err("Can't enable fadump with boot memory size (0x%lx) less than 0x%llx\n",
566 			       fw_dump.boot_memory_size, bootmem_min);
567 			goto error_out;
568 		}
569 
570 		if (!fadump_get_boot_mem_regions()) {
571 			pr_err("Too many holes in boot memory area to enable fadump\n");
572 			goto error_out;
573 		}
574 	}
575 
576 	/*
577 	 * Calculate the memory boundary.
578 	 * If memory_limit is less than actual memory boundary then reserve
579 	 * the memory for fadump beyond the memory_limit and adjust the
580 	 * memory_limit accordingly, so that the running kernel can run with
581 	 * specified memory_limit.
582 	 */
583 	if (memory_limit && memory_limit < memblock_end_of_DRAM()) {
584 		size = get_fadump_area_size();
585 		if ((memory_limit + size) < memblock_end_of_DRAM())
586 			memory_limit += size;
587 		else
588 			memory_limit = memblock_end_of_DRAM();
589 		printk(KERN_INFO "Adjusted memory_limit for firmware-assisted"
590 				" dump, now %#016llx\n", memory_limit);
591 	}
592 	if (memory_limit)
593 		mem_boundary = memory_limit;
594 	else
595 		mem_boundary = memblock_end_of_DRAM();
596 
597 	base = fw_dump.boot_mem_top;
598 	size = get_fadump_area_size();
599 	fw_dump.reserve_dump_area_size = size;
600 	if (fw_dump.dump_active) {
601 		pr_info("Firmware-assisted dump is active.\n");
602 
603 #ifdef CONFIG_HUGETLB_PAGE
604 		/*
605 		 * FADump capture kernel doesn't care much about hugepages.
606 		 * In fact, handling hugepages in capture kernel is asking for
607 		 * trouble. So, disable HugeTLB support when fadump is active.
608 		 */
609 		hugetlb_disabled = true;
610 #endif
611 		/*
612 		 * If last boot has crashed then reserve all the memory
613 		 * above boot memory size so that we don't touch it until
614 		 * dump is written to disk by userspace tool. This memory
615 		 * can be released for general use by invalidating fadump.
616 		 */
617 		fadump_reserve_crash_area(base);
618 
619 		pr_debug("fadumphdr_addr = %#016lx\n", fw_dump.fadumphdr_addr);
620 		pr_debug("Reserve dump area start address: 0x%lx\n",
621 			 fw_dump.reserve_dump_area_start);
622 	} else {
623 		/*
624 		 * Reserve memory at an offset closer to bottom of the RAM to
625 		 * minimize the impact of memory hot-remove operation.
626 		 */
627 		base = fadump_locate_reserve_mem(base, size);
628 
629 		if (!base || (base + size > mem_boundary)) {
630 			pr_err("Failed to find memory chunk for reservation!\n");
631 			goto error_out;
632 		}
633 		fw_dump.reserve_dump_area_start = base;
634 
635 		/*
636 		 * Calculate the kernel metadata address and register it with
637 		 * f/w if the platform supports.
638 		 */
639 		if (fw_dump.ops->fadump_setup_metadata &&
640 		    (fw_dump.ops->fadump_setup_metadata(&fw_dump) < 0))
641 			goto error_out;
642 
643 		if (memblock_reserve(base, size)) {
644 			pr_err("Failed to reserve memory!\n");
645 			goto error_out;
646 		}
647 
648 		pr_info("Reserved %lldMB of memory at %#016llx (System RAM: %lldMB)\n",
649 			(size >> 20), base, (memblock_phys_mem_size() >> 20));
650 
651 		ret = fadump_cma_init();
652 	}
653 
654 	return ret;
655 error_out:
656 	fw_dump.fadump_enabled = 0;
657 	return 0;
658 }
659 
660 /* Look for fadump= cmdline option. */
661 static int __init early_fadump_param(char *p)
662 {
663 	if (!p)
664 		return 1;
665 
666 	if (strncmp(p, "on", 2) == 0)
667 		fw_dump.fadump_enabled = 1;
668 	else if (strncmp(p, "off", 3) == 0)
669 		fw_dump.fadump_enabled = 0;
670 	else if (strncmp(p, "nocma", 5) == 0) {
671 		fw_dump.fadump_enabled = 1;
672 		fw_dump.nocma = 1;
673 	}
674 
675 	return 0;
676 }
677 early_param("fadump", early_fadump_param);
678 
679 /*
680  * Look for fadump_reserve_mem= cmdline option
681  * TODO: Remove references to 'fadump_reserve_mem=' parameter,
682  *       the sooner 'crashkernel=' parameter is accustomed to.
683  */
684 static int __init early_fadump_reserve_mem(char *p)
685 {
686 	if (p)
687 		fw_dump.reserve_bootvar = memparse(p, &p);
688 	return 0;
689 }
690 early_param("fadump_reserve_mem", early_fadump_reserve_mem);
691 
692 void crash_fadump(struct pt_regs *regs, const char *str)
693 {
694 	unsigned int msecs;
695 	struct fadump_crash_info_header *fdh = NULL;
696 	int old_cpu, this_cpu;
697 	/* Do not include first CPU */
698 	unsigned int ncpus = num_online_cpus() - 1;
699 
700 	if (!should_fadump_crash())
701 		return;
702 
703 	/*
704 	 * old_cpu == -1 means this is the first CPU which has come here,
705 	 * go ahead and trigger fadump.
706 	 *
707 	 * old_cpu != -1 means some other CPU has already on it's way
708 	 * to trigger fadump, just keep looping here.
709 	 */
710 	this_cpu = smp_processor_id();
711 	old_cpu = cmpxchg(&crashing_cpu, -1, this_cpu);
712 
713 	if (old_cpu != -1) {
714 		atomic_inc(&cpus_in_fadump);
715 
716 		/*
717 		 * We can't loop here indefinitely. Wait as long as fadump
718 		 * is in force. If we race with fadump un-registration this
719 		 * loop will break and then we go down to normal panic path
720 		 * and reboot. If fadump is in force the first crashing
721 		 * cpu will definitely trigger fadump.
722 		 */
723 		while (fw_dump.dump_registered)
724 			cpu_relax();
725 		return;
726 	}
727 
728 	fdh = __va(fw_dump.fadumphdr_addr);
729 	fdh->crashing_cpu = crashing_cpu;
730 	crash_save_vmcoreinfo();
731 
732 	if (regs)
733 		fdh->regs = *regs;
734 	else
735 		ppc_save_regs(&fdh->regs);
736 
737 	fdh->cpu_mask = *cpu_online_mask;
738 
739 	/*
740 	 * If we came in via system reset, wait a while for the secondary
741 	 * CPUs to enter.
742 	 */
743 	if (TRAP(&(fdh->regs)) == INTERRUPT_SYSTEM_RESET) {
744 		msecs = CRASH_TIMEOUT;
745 		while ((atomic_read(&cpus_in_fadump) < ncpus) && (--msecs > 0))
746 			mdelay(1);
747 	}
748 
749 	fw_dump.ops->fadump_trigger(fdh, str);
750 }
751 
752 u32 *__init fadump_regs_to_elf_notes(u32 *buf, struct pt_regs *regs)
753 {
754 	struct elf_prstatus prstatus;
755 
756 	memset(&prstatus, 0, sizeof(prstatus));
757 	/*
758 	 * FIXME: How do i get PID? Do I really need it?
759 	 * prstatus.pr_pid = ????
760 	 */
761 	elf_core_copy_regs(&prstatus.pr_reg, regs);
762 	buf = append_elf_note(buf, CRASH_CORE_NOTE_NAME, NT_PRSTATUS,
763 			      &prstatus, sizeof(prstatus));
764 	return buf;
765 }
766 
767 void __init fadump_update_elfcore_header(char *bufp)
768 {
769 	struct elf_phdr *phdr;
770 
771 	bufp += sizeof(struct elfhdr);
772 
773 	/* First note is a place holder for cpu notes info. */
774 	phdr = (struct elf_phdr *)bufp;
775 
776 	if (phdr->p_type == PT_NOTE) {
777 		phdr->p_paddr	= __pa(fw_dump.cpu_notes_buf_vaddr);
778 		phdr->p_offset	= phdr->p_paddr;
779 		phdr->p_filesz	= fw_dump.cpu_notes_buf_size;
780 		phdr->p_memsz = fw_dump.cpu_notes_buf_size;
781 	}
782 	return;
783 }
784 
785 static void *__init fadump_alloc_buffer(unsigned long size)
786 {
787 	unsigned long count, i;
788 	struct page *page;
789 	void *vaddr;
790 
791 	vaddr = alloc_pages_exact(size, GFP_KERNEL | __GFP_ZERO);
792 	if (!vaddr)
793 		return NULL;
794 
795 	count = PAGE_ALIGN(size) / PAGE_SIZE;
796 	page = virt_to_page(vaddr);
797 	for (i = 0; i < count; i++)
798 		mark_page_reserved(page + i);
799 	return vaddr;
800 }
801 
802 static void fadump_free_buffer(unsigned long vaddr, unsigned long size)
803 {
804 	free_reserved_area((void *)vaddr, (void *)(vaddr + size), -1, NULL);
805 }
806 
807 s32 __init fadump_setup_cpu_notes_buf(u32 num_cpus)
808 {
809 	/* Allocate buffer to hold cpu crash notes. */
810 	fw_dump.cpu_notes_buf_size = num_cpus * sizeof(note_buf_t);
811 	fw_dump.cpu_notes_buf_size = PAGE_ALIGN(fw_dump.cpu_notes_buf_size);
812 	fw_dump.cpu_notes_buf_vaddr =
813 		(unsigned long)fadump_alloc_buffer(fw_dump.cpu_notes_buf_size);
814 	if (!fw_dump.cpu_notes_buf_vaddr) {
815 		pr_err("Failed to allocate %ld bytes for CPU notes buffer\n",
816 		       fw_dump.cpu_notes_buf_size);
817 		return -ENOMEM;
818 	}
819 
820 	pr_debug("Allocated buffer for cpu notes of size %ld at 0x%lx\n",
821 		 fw_dump.cpu_notes_buf_size,
822 		 fw_dump.cpu_notes_buf_vaddr);
823 	return 0;
824 }
825 
826 void fadump_free_cpu_notes_buf(void)
827 {
828 	if (!fw_dump.cpu_notes_buf_vaddr)
829 		return;
830 
831 	fadump_free_buffer(fw_dump.cpu_notes_buf_vaddr,
832 			   fw_dump.cpu_notes_buf_size);
833 	fw_dump.cpu_notes_buf_vaddr = 0;
834 	fw_dump.cpu_notes_buf_size = 0;
835 }
836 
837 static void fadump_free_mem_ranges(struct fadump_mrange_info *mrange_info)
838 {
839 	if (mrange_info->is_static) {
840 		mrange_info->mem_range_cnt = 0;
841 		return;
842 	}
843 
844 	kfree(mrange_info->mem_ranges);
845 	memset((void *)((u64)mrange_info + RNG_NAME_SZ), 0,
846 	       (sizeof(struct fadump_mrange_info) - RNG_NAME_SZ));
847 }
848 
849 /*
850  * Allocate or reallocate mem_ranges array in incremental units
851  * of PAGE_SIZE.
852  */
853 static int fadump_alloc_mem_ranges(struct fadump_mrange_info *mrange_info)
854 {
855 	struct fadump_memory_range *new_array;
856 	u64 new_size;
857 
858 	new_size = mrange_info->mem_ranges_sz + PAGE_SIZE;
859 	pr_debug("Allocating %llu bytes of memory for %s memory ranges\n",
860 		 new_size, mrange_info->name);
861 
862 	new_array = krealloc(mrange_info->mem_ranges, new_size, GFP_KERNEL);
863 	if (new_array == NULL) {
864 		pr_err("Insufficient memory for setting up %s memory ranges\n",
865 		       mrange_info->name);
866 		fadump_free_mem_ranges(mrange_info);
867 		return -ENOMEM;
868 	}
869 
870 	mrange_info->mem_ranges = new_array;
871 	mrange_info->mem_ranges_sz = new_size;
872 	mrange_info->max_mem_ranges = (new_size /
873 				       sizeof(struct fadump_memory_range));
874 	return 0;
875 }
876 static inline int fadump_add_mem_range(struct fadump_mrange_info *mrange_info,
877 				       u64 base, u64 end)
878 {
879 	struct fadump_memory_range *mem_ranges = mrange_info->mem_ranges;
880 	bool is_adjacent = false;
881 	u64 start, size;
882 
883 	if (base == end)
884 		return 0;
885 
886 	/*
887 	 * Fold adjacent memory ranges to bring down the memory ranges/
888 	 * PT_LOAD segments count.
889 	 */
890 	if (mrange_info->mem_range_cnt) {
891 		start = mem_ranges[mrange_info->mem_range_cnt - 1].base;
892 		size  = mem_ranges[mrange_info->mem_range_cnt - 1].size;
893 
894 		/*
895 		 * Boot memory area needs separate PT_LOAD segment(s) as it
896 		 * is moved to a different location at the time of crash.
897 		 * So, fold only if the region is not boot memory area.
898 		 */
899 		if ((start + size) == base && start >= fw_dump.boot_mem_top)
900 			is_adjacent = true;
901 	}
902 	if (!is_adjacent) {
903 		/* resize the array on reaching the limit */
904 		if (mrange_info->mem_range_cnt == mrange_info->max_mem_ranges) {
905 			int ret;
906 
907 			if (mrange_info->is_static) {
908 				pr_err("Reached array size limit for %s memory ranges\n",
909 				       mrange_info->name);
910 				return -ENOSPC;
911 			}
912 
913 			ret = fadump_alloc_mem_ranges(mrange_info);
914 			if (ret)
915 				return ret;
916 
917 			/* Update to the new resized array */
918 			mem_ranges = mrange_info->mem_ranges;
919 		}
920 
921 		start = base;
922 		mem_ranges[mrange_info->mem_range_cnt].base = start;
923 		mrange_info->mem_range_cnt++;
924 	}
925 
926 	mem_ranges[mrange_info->mem_range_cnt - 1].size = (end - start);
927 	pr_debug("%s_memory_range[%d] [%#016llx-%#016llx], %#llx bytes\n",
928 		 mrange_info->name, (mrange_info->mem_range_cnt - 1),
929 		 start, end - 1, (end - start));
930 	return 0;
931 }
932 
933 static int fadump_exclude_reserved_area(u64 start, u64 end)
934 {
935 	u64 ra_start, ra_end;
936 	int ret = 0;
937 
938 	ra_start = fw_dump.reserve_dump_area_start;
939 	ra_end = ra_start + fw_dump.reserve_dump_area_size;
940 
941 	if ((ra_start < end) && (ra_end > start)) {
942 		if ((start < ra_start) && (end > ra_end)) {
943 			ret = fadump_add_mem_range(&crash_mrange_info,
944 						   start, ra_start);
945 			if (ret)
946 				return ret;
947 
948 			ret = fadump_add_mem_range(&crash_mrange_info,
949 						   ra_end, end);
950 		} else if (start < ra_start) {
951 			ret = fadump_add_mem_range(&crash_mrange_info,
952 						   start, ra_start);
953 		} else if (ra_end < end) {
954 			ret = fadump_add_mem_range(&crash_mrange_info,
955 						   ra_end, end);
956 		}
957 	} else
958 		ret = fadump_add_mem_range(&crash_mrange_info, start, end);
959 
960 	return ret;
961 }
962 
963 static int fadump_init_elfcore_header(char *bufp)
964 {
965 	struct elfhdr *elf;
966 
967 	elf = (struct elfhdr *) bufp;
968 	bufp += sizeof(struct elfhdr);
969 	memcpy(elf->e_ident, ELFMAG, SELFMAG);
970 	elf->e_ident[EI_CLASS] = ELF_CLASS;
971 	elf->e_ident[EI_DATA] = ELF_DATA;
972 	elf->e_ident[EI_VERSION] = EV_CURRENT;
973 	elf->e_ident[EI_OSABI] = ELF_OSABI;
974 	memset(elf->e_ident+EI_PAD, 0, EI_NIDENT-EI_PAD);
975 	elf->e_type = ET_CORE;
976 	elf->e_machine = ELF_ARCH;
977 	elf->e_version = EV_CURRENT;
978 	elf->e_entry = 0;
979 	elf->e_phoff = sizeof(struct elfhdr);
980 	elf->e_shoff = 0;
981 
982 	if (IS_ENABLED(CONFIG_PPC64_ELF_ABI_V2))
983 		elf->e_flags = 2;
984 	else if (IS_ENABLED(CONFIG_PPC64_ELF_ABI_V1))
985 		elf->e_flags = 1;
986 	else
987 		elf->e_flags = 0;
988 
989 	elf->e_ehsize = sizeof(struct elfhdr);
990 	elf->e_phentsize = sizeof(struct elf_phdr);
991 	elf->e_phnum = 0;
992 	elf->e_shentsize = 0;
993 	elf->e_shnum = 0;
994 	elf->e_shstrndx = 0;
995 
996 	return 0;
997 }
998 
999 /*
1000  * Traverse through memblock structure and setup crash memory ranges. These
1001  * ranges will be used create PT_LOAD program headers in elfcore header.
1002  */
1003 static int fadump_setup_crash_memory_ranges(void)
1004 {
1005 	u64 i, start, end;
1006 	int ret;
1007 
1008 	pr_debug("Setup crash memory ranges.\n");
1009 	crash_mrange_info.mem_range_cnt = 0;
1010 
1011 	/*
1012 	 * Boot memory region(s) registered with firmware are moved to
1013 	 * different location at the time of crash. Create separate program
1014 	 * header(s) for this memory chunk(s) with the correct offset.
1015 	 */
1016 	for (i = 0; i < fw_dump.boot_mem_regs_cnt; i++) {
1017 		start = fw_dump.boot_mem_addr[i];
1018 		end = start + fw_dump.boot_mem_sz[i];
1019 		ret = fadump_add_mem_range(&crash_mrange_info, start, end);
1020 		if (ret)
1021 			return ret;
1022 	}
1023 
1024 	for_each_mem_range(i, &start, &end) {
1025 		/*
1026 		 * skip the memory chunk that is already added
1027 		 * (0 through boot_memory_top).
1028 		 */
1029 		if (start < fw_dump.boot_mem_top) {
1030 			if (end > fw_dump.boot_mem_top)
1031 				start = fw_dump.boot_mem_top;
1032 			else
1033 				continue;
1034 		}
1035 
1036 		/* add this range excluding the reserved dump area. */
1037 		ret = fadump_exclude_reserved_area(start, end);
1038 		if (ret)
1039 			return ret;
1040 	}
1041 
1042 	return 0;
1043 }
1044 
1045 /*
1046  * If the given physical address falls within the boot memory region then
1047  * return the relocated address that points to the dump region reserved
1048  * for saving initial boot memory contents.
1049  */
1050 static inline unsigned long fadump_relocate(unsigned long paddr)
1051 {
1052 	unsigned long raddr, rstart, rend, rlast, hole_size;
1053 	int i;
1054 
1055 	hole_size = 0;
1056 	rlast = 0;
1057 	raddr = paddr;
1058 	for (i = 0; i < fw_dump.boot_mem_regs_cnt; i++) {
1059 		rstart = fw_dump.boot_mem_addr[i];
1060 		rend = rstart + fw_dump.boot_mem_sz[i];
1061 		hole_size += (rstart - rlast);
1062 
1063 		if (paddr >= rstart && paddr < rend) {
1064 			raddr += fw_dump.boot_mem_dest_addr - hole_size;
1065 			break;
1066 		}
1067 
1068 		rlast = rend;
1069 	}
1070 
1071 	pr_debug("vmcoreinfo: paddr = 0x%lx, raddr = 0x%lx\n", paddr, raddr);
1072 	return raddr;
1073 }
1074 
1075 static int fadump_create_elfcore_headers(char *bufp)
1076 {
1077 	unsigned long long raddr, offset;
1078 	struct elf_phdr *phdr;
1079 	struct elfhdr *elf;
1080 	int i, j;
1081 
1082 	fadump_init_elfcore_header(bufp);
1083 	elf = (struct elfhdr *)bufp;
1084 	bufp += sizeof(struct elfhdr);
1085 
1086 	/*
1087 	 * setup ELF PT_NOTE, place holder for cpu notes info. The notes info
1088 	 * will be populated during second kernel boot after crash. Hence
1089 	 * this PT_NOTE will always be the first elf note.
1090 	 *
1091 	 * NOTE: Any new ELF note addition should be placed after this note.
1092 	 */
1093 	phdr = (struct elf_phdr *)bufp;
1094 	bufp += sizeof(struct elf_phdr);
1095 	phdr->p_type = PT_NOTE;
1096 	phdr->p_flags = 0;
1097 	phdr->p_vaddr = 0;
1098 	phdr->p_align = 0;
1099 
1100 	phdr->p_offset = 0;
1101 	phdr->p_paddr = 0;
1102 	phdr->p_filesz = 0;
1103 	phdr->p_memsz = 0;
1104 
1105 	(elf->e_phnum)++;
1106 
1107 	/* setup ELF PT_NOTE for vmcoreinfo */
1108 	phdr = (struct elf_phdr *)bufp;
1109 	bufp += sizeof(struct elf_phdr);
1110 	phdr->p_type	= PT_NOTE;
1111 	phdr->p_flags	= 0;
1112 	phdr->p_vaddr	= 0;
1113 	phdr->p_align	= 0;
1114 
1115 	phdr->p_paddr	= fadump_relocate(paddr_vmcoreinfo_note());
1116 	phdr->p_offset	= phdr->p_paddr;
1117 	phdr->p_memsz	= phdr->p_filesz = VMCOREINFO_NOTE_SIZE;
1118 
1119 	/* Increment number of program headers. */
1120 	(elf->e_phnum)++;
1121 
1122 	/* setup PT_LOAD sections. */
1123 	j = 0;
1124 	offset = 0;
1125 	raddr = fw_dump.boot_mem_addr[0];
1126 	for (i = 0; i < crash_mrange_info.mem_range_cnt; i++) {
1127 		u64 mbase, msize;
1128 
1129 		mbase = crash_mrange_info.mem_ranges[i].base;
1130 		msize = crash_mrange_info.mem_ranges[i].size;
1131 		if (!msize)
1132 			continue;
1133 
1134 		phdr = (struct elf_phdr *)bufp;
1135 		bufp += sizeof(struct elf_phdr);
1136 		phdr->p_type	= PT_LOAD;
1137 		phdr->p_flags	= PF_R|PF_W|PF_X;
1138 		phdr->p_offset	= mbase;
1139 
1140 		if (mbase == raddr) {
1141 			/*
1142 			 * The entire real memory region will be moved by
1143 			 * firmware to the specified destination_address.
1144 			 * Hence set the correct offset.
1145 			 */
1146 			phdr->p_offset = fw_dump.boot_mem_dest_addr + offset;
1147 			if (j < (fw_dump.boot_mem_regs_cnt - 1)) {
1148 				offset += fw_dump.boot_mem_sz[j];
1149 				raddr = fw_dump.boot_mem_addr[++j];
1150 			}
1151 		}
1152 
1153 		phdr->p_paddr = mbase;
1154 		phdr->p_vaddr = (unsigned long)__va(mbase);
1155 		phdr->p_filesz = msize;
1156 		phdr->p_memsz = msize;
1157 		phdr->p_align = 0;
1158 
1159 		/* Increment number of program headers. */
1160 		(elf->e_phnum)++;
1161 	}
1162 	return 0;
1163 }
1164 
1165 static unsigned long init_fadump_header(unsigned long addr)
1166 {
1167 	struct fadump_crash_info_header *fdh;
1168 
1169 	if (!addr)
1170 		return 0;
1171 
1172 	fdh = __va(addr);
1173 	addr += sizeof(struct fadump_crash_info_header);
1174 
1175 	memset(fdh, 0, sizeof(struct fadump_crash_info_header));
1176 	fdh->magic_number = FADUMP_CRASH_INFO_MAGIC;
1177 	fdh->elfcorehdr_addr = addr;
1178 	/* We will set the crashing cpu id in crash_fadump() during crash. */
1179 	fdh->crashing_cpu = FADUMP_CPU_UNKNOWN;
1180 	/*
1181 	 * When LPAR is terminated by PYHP, ensure all possible CPUs'
1182 	 * register data is processed while exporting the vmcore.
1183 	 */
1184 	fdh->cpu_mask = *cpu_possible_mask;
1185 
1186 	return addr;
1187 }
1188 
1189 static int register_fadump(void)
1190 {
1191 	unsigned long addr;
1192 	void *vaddr;
1193 	int ret;
1194 
1195 	/*
1196 	 * If no memory is reserved then we can not register for firmware-
1197 	 * assisted dump.
1198 	 */
1199 	if (!fw_dump.reserve_dump_area_size)
1200 		return -ENODEV;
1201 
1202 	ret = fadump_setup_crash_memory_ranges();
1203 	if (ret)
1204 		return ret;
1205 
1206 	addr = fw_dump.fadumphdr_addr;
1207 
1208 	/* Initialize fadump crash info header. */
1209 	addr = init_fadump_header(addr);
1210 	vaddr = __va(addr);
1211 
1212 	pr_debug("Creating ELF core headers at %#016lx\n", addr);
1213 	fadump_create_elfcore_headers(vaddr);
1214 
1215 	/* register the future kernel dump with firmware. */
1216 	pr_debug("Registering for firmware-assisted kernel dump...\n");
1217 	return fw_dump.ops->fadump_register(&fw_dump);
1218 }
1219 
1220 void fadump_cleanup(void)
1221 {
1222 	if (!fw_dump.fadump_supported)
1223 		return;
1224 
1225 	/* Invalidate the registration only if dump is active. */
1226 	if (fw_dump.dump_active) {
1227 		pr_debug("Invalidating firmware-assisted dump registration\n");
1228 		fw_dump.ops->fadump_invalidate(&fw_dump);
1229 	} else if (fw_dump.dump_registered) {
1230 		/* Un-register Firmware-assisted dump if it was registered. */
1231 		fw_dump.ops->fadump_unregister(&fw_dump);
1232 		fadump_free_mem_ranges(&crash_mrange_info);
1233 	}
1234 
1235 	if (fw_dump.ops->fadump_cleanup)
1236 		fw_dump.ops->fadump_cleanup(&fw_dump);
1237 }
1238 
1239 static void fadump_free_reserved_memory(unsigned long start_pfn,
1240 					unsigned long end_pfn)
1241 {
1242 	unsigned long pfn;
1243 	unsigned long time_limit = jiffies + HZ;
1244 
1245 	pr_info("freeing reserved memory (0x%llx - 0x%llx)\n",
1246 		PFN_PHYS(start_pfn), PFN_PHYS(end_pfn));
1247 
1248 	for (pfn = start_pfn; pfn < end_pfn; pfn++) {
1249 		free_reserved_page(pfn_to_page(pfn));
1250 
1251 		if (time_after(jiffies, time_limit)) {
1252 			cond_resched();
1253 			time_limit = jiffies + HZ;
1254 		}
1255 	}
1256 }
1257 
1258 /*
1259  * Skip memory holes and free memory that was actually reserved.
1260  */
1261 static void fadump_release_reserved_area(u64 start, u64 end)
1262 {
1263 	unsigned long reg_spfn, reg_epfn;
1264 	u64 tstart, tend, spfn, epfn;
1265 	int i;
1266 
1267 	spfn = PHYS_PFN(start);
1268 	epfn = PHYS_PFN(end);
1269 
1270 	for_each_mem_pfn_range(i, MAX_NUMNODES, &reg_spfn, &reg_epfn, NULL) {
1271 		tstart = max_t(u64, spfn, reg_spfn);
1272 		tend   = min_t(u64, epfn, reg_epfn);
1273 
1274 		if (tstart < tend) {
1275 			fadump_free_reserved_memory(tstart, tend);
1276 
1277 			if (tend == epfn)
1278 				break;
1279 
1280 			spfn = tend;
1281 		}
1282 	}
1283 }
1284 
1285 /*
1286  * Sort the mem ranges in-place and merge adjacent ranges
1287  * to minimize the memory ranges count.
1288  */
1289 static void sort_and_merge_mem_ranges(struct fadump_mrange_info *mrange_info)
1290 {
1291 	struct fadump_memory_range *mem_ranges;
1292 	u64 base, size;
1293 	int i, j, idx;
1294 
1295 	if (!reserved_mrange_info.mem_range_cnt)
1296 		return;
1297 
1298 	/* Sort the memory ranges */
1299 	mem_ranges = mrange_info->mem_ranges;
1300 	for (i = 0; i < mrange_info->mem_range_cnt; i++) {
1301 		idx = i;
1302 		for (j = (i + 1); j < mrange_info->mem_range_cnt; j++) {
1303 			if (mem_ranges[idx].base > mem_ranges[j].base)
1304 				idx = j;
1305 		}
1306 		if (idx != i)
1307 			swap(mem_ranges[idx], mem_ranges[i]);
1308 	}
1309 
1310 	/* Merge adjacent reserved ranges */
1311 	idx = 0;
1312 	for (i = 1; i < mrange_info->mem_range_cnt; i++) {
1313 		base = mem_ranges[i-1].base;
1314 		size = mem_ranges[i-1].size;
1315 		if (mem_ranges[i].base == (base + size))
1316 			mem_ranges[idx].size += mem_ranges[i].size;
1317 		else {
1318 			idx++;
1319 			if (i == idx)
1320 				continue;
1321 
1322 			mem_ranges[idx] = mem_ranges[i];
1323 		}
1324 	}
1325 	mrange_info->mem_range_cnt = idx + 1;
1326 }
1327 
1328 /*
1329  * Scan reserved-ranges to consider them while reserving/releasing
1330  * memory for FADump.
1331  */
1332 static void __init early_init_dt_scan_reserved_ranges(unsigned long node)
1333 {
1334 	const __be32 *prop;
1335 	int len, ret = -1;
1336 	unsigned long i;
1337 
1338 	/* reserved-ranges already scanned */
1339 	if (reserved_mrange_info.mem_range_cnt != 0)
1340 		return;
1341 
1342 	prop = of_get_flat_dt_prop(node, "reserved-ranges", &len);
1343 	if (!prop)
1344 		return;
1345 
1346 	/*
1347 	 * Each reserved range is an (address,size) pair, 2 cells each,
1348 	 * totalling 4 cells per range.
1349 	 */
1350 	for (i = 0; i < len / (sizeof(*prop) * 4); i++) {
1351 		u64 base, size;
1352 
1353 		base = of_read_number(prop + (i * 4) + 0, 2);
1354 		size = of_read_number(prop + (i * 4) + 2, 2);
1355 
1356 		if (size) {
1357 			ret = fadump_add_mem_range(&reserved_mrange_info,
1358 						   base, base + size);
1359 			if (ret < 0) {
1360 				pr_warn("some reserved ranges are ignored!\n");
1361 				break;
1362 			}
1363 		}
1364 	}
1365 
1366 	/* Compact reserved ranges */
1367 	sort_and_merge_mem_ranges(&reserved_mrange_info);
1368 }
1369 
1370 /*
1371  * Release the memory that was reserved during early boot to preserve the
1372  * crash'ed kernel's memory contents except reserved dump area (permanent
1373  * reservation) and reserved ranges used by F/W. The released memory will
1374  * be available for general use.
1375  */
1376 static void fadump_release_memory(u64 begin, u64 end)
1377 {
1378 	u64 ra_start, ra_end, tstart;
1379 	int i, ret;
1380 
1381 	ra_start = fw_dump.reserve_dump_area_start;
1382 	ra_end = ra_start + fw_dump.reserve_dump_area_size;
1383 
1384 	/*
1385 	 * If reserved ranges array limit is hit, overwrite the last reserved
1386 	 * memory range with reserved dump area to ensure it is excluded from
1387 	 * the memory being released (reused for next FADump registration).
1388 	 */
1389 	if (reserved_mrange_info.mem_range_cnt ==
1390 	    reserved_mrange_info.max_mem_ranges)
1391 		reserved_mrange_info.mem_range_cnt--;
1392 
1393 	ret = fadump_add_mem_range(&reserved_mrange_info, ra_start, ra_end);
1394 	if (ret != 0)
1395 		return;
1396 
1397 	/* Get the reserved ranges list in order first. */
1398 	sort_and_merge_mem_ranges(&reserved_mrange_info);
1399 
1400 	/* Exclude reserved ranges and release remaining memory */
1401 	tstart = begin;
1402 	for (i = 0; i < reserved_mrange_info.mem_range_cnt; i++) {
1403 		ra_start = reserved_mrange_info.mem_ranges[i].base;
1404 		ra_end = ra_start + reserved_mrange_info.mem_ranges[i].size;
1405 
1406 		if (tstart >= ra_end)
1407 			continue;
1408 
1409 		if (tstart < ra_start)
1410 			fadump_release_reserved_area(tstart, ra_start);
1411 		tstart = ra_end;
1412 	}
1413 
1414 	if (tstart < end)
1415 		fadump_release_reserved_area(tstart, end);
1416 }
1417 
1418 static void fadump_invalidate_release_mem(void)
1419 {
1420 	mutex_lock(&fadump_mutex);
1421 	if (!fw_dump.dump_active) {
1422 		mutex_unlock(&fadump_mutex);
1423 		return;
1424 	}
1425 
1426 	fadump_cleanup();
1427 	mutex_unlock(&fadump_mutex);
1428 
1429 	fadump_release_memory(fw_dump.boot_mem_top, memblock_end_of_DRAM());
1430 	fadump_free_cpu_notes_buf();
1431 
1432 	/*
1433 	 * Setup kernel metadata and initialize the kernel dump
1434 	 * memory structure for FADump re-registration.
1435 	 */
1436 	if (fw_dump.ops->fadump_setup_metadata &&
1437 	    (fw_dump.ops->fadump_setup_metadata(&fw_dump) < 0))
1438 		pr_warn("Failed to setup kernel metadata!\n");
1439 	fw_dump.ops->fadump_init_mem_struct(&fw_dump);
1440 }
1441 
1442 static ssize_t release_mem_store(struct kobject *kobj,
1443 				 struct kobj_attribute *attr,
1444 				 const char *buf, size_t count)
1445 {
1446 	int input = -1;
1447 
1448 	if (!fw_dump.dump_active)
1449 		return -EPERM;
1450 
1451 	if (kstrtoint(buf, 0, &input))
1452 		return -EINVAL;
1453 
1454 	if (input == 1) {
1455 		/*
1456 		 * Take away the '/proc/vmcore'. We are releasing the dump
1457 		 * memory, hence it will not be valid anymore.
1458 		 */
1459 #ifdef CONFIG_PROC_VMCORE
1460 		vmcore_cleanup();
1461 #endif
1462 		fadump_invalidate_release_mem();
1463 
1464 	} else
1465 		return -EINVAL;
1466 	return count;
1467 }
1468 
1469 /* Release the reserved memory and disable the FADump */
1470 static void __init unregister_fadump(void)
1471 {
1472 	fadump_cleanup();
1473 	fadump_release_memory(fw_dump.reserve_dump_area_start,
1474 			      fw_dump.reserve_dump_area_size);
1475 	fw_dump.fadump_enabled = 0;
1476 	kobject_put(fadump_kobj);
1477 }
1478 
1479 static ssize_t enabled_show(struct kobject *kobj,
1480 			    struct kobj_attribute *attr,
1481 			    char *buf)
1482 {
1483 	return sprintf(buf, "%d\n", fw_dump.fadump_enabled);
1484 }
1485 
1486 static ssize_t mem_reserved_show(struct kobject *kobj,
1487 				 struct kobj_attribute *attr,
1488 				 char *buf)
1489 {
1490 	return sprintf(buf, "%ld\n", fw_dump.reserve_dump_area_size);
1491 }
1492 
1493 static ssize_t registered_show(struct kobject *kobj,
1494 			       struct kobj_attribute *attr,
1495 			       char *buf)
1496 {
1497 	return sprintf(buf, "%d\n", fw_dump.dump_registered);
1498 }
1499 
1500 static ssize_t registered_store(struct kobject *kobj,
1501 				struct kobj_attribute *attr,
1502 				const char *buf, size_t count)
1503 {
1504 	int ret = 0;
1505 	int input = -1;
1506 
1507 	if (!fw_dump.fadump_enabled || fw_dump.dump_active)
1508 		return -EPERM;
1509 
1510 	if (kstrtoint(buf, 0, &input))
1511 		return -EINVAL;
1512 
1513 	mutex_lock(&fadump_mutex);
1514 
1515 	switch (input) {
1516 	case 0:
1517 		if (fw_dump.dump_registered == 0) {
1518 			goto unlock_out;
1519 		}
1520 
1521 		/* Un-register Firmware-assisted dump */
1522 		pr_debug("Un-register firmware-assisted dump\n");
1523 		fw_dump.ops->fadump_unregister(&fw_dump);
1524 		break;
1525 	case 1:
1526 		if (fw_dump.dump_registered == 1) {
1527 			/* Un-register Firmware-assisted dump */
1528 			fw_dump.ops->fadump_unregister(&fw_dump);
1529 		}
1530 		/* Register Firmware-assisted dump */
1531 		ret = register_fadump();
1532 		break;
1533 	default:
1534 		ret = -EINVAL;
1535 		break;
1536 	}
1537 
1538 unlock_out:
1539 	mutex_unlock(&fadump_mutex);
1540 	return ret < 0 ? ret : count;
1541 }
1542 
1543 static int fadump_region_show(struct seq_file *m, void *private)
1544 {
1545 	if (!fw_dump.fadump_enabled)
1546 		return 0;
1547 
1548 	mutex_lock(&fadump_mutex);
1549 	fw_dump.ops->fadump_region_show(&fw_dump, m);
1550 	mutex_unlock(&fadump_mutex);
1551 	return 0;
1552 }
1553 
1554 static struct kobj_attribute release_attr = __ATTR_WO(release_mem);
1555 static struct kobj_attribute enable_attr = __ATTR_RO(enabled);
1556 static struct kobj_attribute register_attr = __ATTR_RW(registered);
1557 static struct kobj_attribute mem_reserved_attr = __ATTR_RO(mem_reserved);
1558 
1559 static struct attribute *fadump_attrs[] = {
1560 	&enable_attr.attr,
1561 	&register_attr.attr,
1562 	&mem_reserved_attr.attr,
1563 	NULL,
1564 };
1565 
1566 ATTRIBUTE_GROUPS(fadump);
1567 
1568 DEFINE_SHOW_ATTRIBUTE(fadump_region);
1569 
1570 static void __init fadump_init_files(void)
1571 {
1572 	int rc = 0;
1573 
1574 	fadump_kobj = kobject_create_and_add("fadump", kernel_kobj);
1575 	if (!fadump_kobj) {
1576 		pr_err("failed to create fadump kobject\n");
1577 		return;
1578 	}
1579 
1580 	debugfs_create_file("fadump_region", 0444, arch_debugfs_dir, NULL,
1581 			    &fadump_region_fops);
1582 
1583 	if (fw_dump.dump_active) {
1584 		rc = sysfs_create_file(fadump_kobj, &release_attr.attr);
1585 		if (rc)
1586 			pr_err("unable to create release_mem sysfs file (%d)\n",
1587 			       rc);
1588 	}
1589 
1590 	rc = sysfs_create_groups(fadump_kobj, fadump_groups);
1591 	if (rc) {
1592 		pr_err("sysfs group creation failed (%d), unregistering FADump",
1593 		       rc);
1594 		unregister_fadump();
1595 		return;
1596 	}
1597 
1598 	/*
1599 	 * The FADump sysfs are moved from kernel_kobj to fadump_kobj need to
1600 	 * create symlink at old location to maintain backward compatibility.
1601 	 *
1602 	 *      - fadump_enabled -> fadump/enabled
1603 	 *      - fadump_registered -> fadump/registered
1604 	 *      - fadump_release_mem -> fadump/release_mem
1605 	 */
1606 	rc = compat_only_sysfs_link_entry_to_kobj(kernel_kobj, fadump_kobj,
1607 						  "enabled", "fadump_enabled");
1608 	if (rc) {
1609 		pr_err("unable to create fadump_enabled symlink (%d)", rc);
1610 		return;
1611 	}
1612 
1613 	rc = compat_only_sysfs_link_entry_to_kobj(kernel_kobj, fadump_kobj,
1614 						  "registered",
1615 						  "fadump_registered");
1616 	if (rc) {
1617 		pr_err("unable to create fadump_registered symlink (%d)", rc);
1618 		sysfs_remove_link(kernel_kobj, "fadump_enabled");
1619 		return;
1620 	}
1621 
1622 	if (fw_dump.dump_active) {
1623 		rc = compat_only_sysfs_link_entry_to_kobj(kernel_kobj,
1624 							  fadump_kobj,
1625 							  "release_mem",
1626 							  "fadump_release_mem");
1627 		if (rc)
1628 			pr_err("unable to create fadump_release_mem symlink (%d)",
1629 			       rc);
1630 	}
1631 	return;
1632 }
1633 
1634 /*
1635  * Prepare for firmware-assisted dump.
1636  */
1637 int __init setup_fadump(void)
1638 {
1639 	if (!fw_dump.fadump_supported)
1640 		return 0;
1641 
1642 	fadump_init_files();
1643 	fadump_show_config();
1644 
1645 	if (!fw_dump.fadump_enabled)
1646 		return 1;
1647 
1648 	/*
1649 	 * If dump data is available then see if it is valid and prepare for
1650 	 * saving it to the disk.
1651 	 */
1652 	if (fw_dump.dump_active) {
1653 		/*
1654 		 * if dump process fails then invalidate the registration
1655 		 * and release memory before proceeding for re-registration.
1656 		 */
1657 		if (fw_dump.ops->fadump_process(&fw_dump) < 0)
1658 			fadump_invalidate_release_mem();
1659 	}
1660 	/* Initialize the kernel dump memory structure and register with f/w */
1661 	else if (fw_dump.reserve_dump_area_size) {
1662 		fw_dump.ops->fadump_init_mem_struct(&fw_dump);
1663 		register_fadump();
1664 	}
1665 
1666 	/*
1667 	 * In case of panic, fadump is triggered via ppc_panic_event()
1668 	 * panic notifier. Setting crash_kexec_post_notifiers to 'true'
1669 	 * lets panic() function take crash friendly path before panic
1670 	 * notifiers are invoked.
1671 	 */
1672 	crash_kexec_post_notifiers = true;
1673 
1674 	return 1;
1675 }
1676 /*
1677  * Use subsys_initcall_sync() here because there is dependency with
1678  * crash_save_vmcoreinfo_init(), which must run first to ensure vmcoreinfo initialization
1679  * is done before registering with f/w.
1680  */
1681 subsys_initcall_sync(setup_fadump);
1682 #else /* !CONFIG_PRESERVE_FA_DUMP */
1683 
1684 /* Scan the Firmware Assisted dump configuration details. */
1685 int __init early_init_dt_scan_fw_dump(unsigned long node, const char *uname,
1686 				      int depth, void *data)
1687 {
1688 	if ((depth != 1) || (strcmp(uname, "ibm,opal") != 0))
1689 		return 0;
1690 
1691 	opal_fadump_dt_scan(&fw_dump, node);
1692 	return 1;
1693 }
1694 
1695 /*
1696  * When dump is active but PRESERVE_FA_DUMP is enabled on the kernel,
1697  * preserve crash data. The subsequent memory preserving kernel boot
1698  * is likely to process this crash data.
1699  */
1700 int __init fadump_reserve_mem(void)
1701 {
1702 	if (fw_dump.dump_active) {
1703 		/*
1704 		 * If last boot has crashed then reserve all the memory
1705 		 * above boot memory to preserve crash data.
1706 		 */
1707 		pr_info("Preserving crash data for processing in next boot.\n");
1708 		fadump_reserve_crash_area(fw_dump.boot_mem_top);
1709 	} else
1710 		pr_debug("FADump-aware kernel..\n");
1711 
1712 	return 1;
1713 }
1714 #endif /* CONFIG_PRESERVE_FA_DUMP */
1715 
1716 /* Preserve everything above the base address */
1717 static void __init fadump_reserve_crash_area(u64 base)
1718 {
1719 	u64 i, mstart, mend, msize;
1720 
1721 	for_each_mem_range(i, &mstart, &mend) {
1722 		msize  = mend - mstart;
1723 
1724 		if ((mstart + msize) < base)
1725 			continue;
1726 
1727 		if (mstart < base) {
1728 			msize -= (base - mstart);
1729 			mstart = base;
1730 		}
1731 
1732 		pr_info("Reserving %lluMB of memory at %#016llx for preserving crash data",
1733 			(msize >> 20), mstart);
1734 		memblock_reserve(mstart, msize);
1735 	}
1736 }
1737 
1738 unsigned long __init arch_reserved_kernel_pages(void)
1739 {
1740 	return memblock_reserved_size() / PAGE_SIZE;
1741 }
1742