Loading...
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
38#include <asm/debugfs.h>
39#include <asm/page.h>
40#include <asm/prom.h>
41#include <asm/rtas.h>
42#include <asm/fadump.h>
43#include <asm/setup.h>
44
45static struct fw_dump fw_dump;
46static struct fadump_mem_struct fdm;
47static const struct fadump_mem_struct *fdm_active;
48
49static DEFINE_MUTEX(fadump_mutex);
50struct fad_crash_memory_ranges crash_memory_ranges[INIT_CRASHMEM_RANGES];
51int crash_mem_ranges;
52
53/* Scan the Firmware Assisted dump configuration details. */
54int __init early_init_dt_scan_fw_dump(unsigned long node,
55 const char *uname, int depth, void *data)
56{
57 const __be32 *sections;
58 int i, num_sections;
59 int size;
60 const __be32 *token;
61
62 if (depth != 1 || strcmp(uname, "rtas") != 0)
63 return 0;
64
65 /*
66 * Check if Firmware Assisted dump is supported. if yes, check
67 * if dump has been initiated on last reboot.
68 */
69 token = of_get_flat_dt_prop(node, "ibm,configure-kernel-dump", NULL);
70 if (!token)
71 return 1;
72
73 fw_dump.fadump_supported = 1;
74 fw_dump.ibm_configure_kernel_dump = be32_to_cpu(*token);
75
76 /*
77 * The 'ibm,kernel-dump' rtas node is present only if there is
78 * dump data waiting for us.
79 */
80 fdm_active = of_get_flat_dt_prop(node, "ibm,kernel-dump", NULL);
81 if (fdm_active)
82 fw_dump.dump_active = 1;
83
84 /* Get the sizes required to store dump data for the firmware provided
85 * dump sections.
86 * For each dump section type supported, a 32bit cell which defines
87 * the ID of a supported section followed by two 32 bit cells which
88 * gives teh size of the section in bytes.
89 */
90 sections = of_get_flat_dt_prop(node, "ibm,configure-kernel-dump-sizes",
91 &size);
92
93 if (!sections)
94 return 1;
95
96 num_sections = size / (3 * sizeof(u32));
97
98 for (i = 0; i < num_sections; i++, sections += 3) {
99 u32 type = (u32)of_read_number(sections, 1);
100
101 switch (type) {
102 case FADUMP_CPU_STATE_DATA:
103 fw_dump.cpu_state_data_size =
104 of_read_ulong(§ions[1], 2);
105 break;
106 case FADUMP_HPTE_REGION:
107 fw_dump.hpte_region_size =
108 of_read_ulong(§ions[1], 2);
109 break;
110 }
111 }
112
113 return 1;
114}
115
116/*
117 * If fadump is registered, check if the memory provided
118 * falls within boot memory area.
119 */
120int is_fadump_boot_memory_area(u64 addr, ulong size)
121{
122 if (!fw_dump.dump_registered)
123 return 0;
124
125 return (addr + size) > RMA_START && addr <= fw_dump.boot_memory_size;
126}
127
128int should_fadump_crash(void)
129{
130 if (!fw_dump.dump_registered || !fw_dump.fadumphdr_addr)
131 return 0;
132 return 1;
133}
134
135int is_fadump_active(void)
136{
137 return fw_dump.dump_active;
138}
139
140/*
141 * Returns 1, if there are no holes in boot memory area,
142 * 0 otherwise.
143 */
144static int is_boot_memory_area_contiguous(void)
145{
146 struct memblock_region *reg;
147 unsigned long tstart, tend;
148 unsigned long start_pfn = PHYS_PFN(RMA_START);
149 unsigned long end_pfn = PHYS_PFN(RMA_START + fw_dump.boot_memory_size);
150 unsigned int ret = 0;
151
152 for_each_memblock(memory, reg) {
153 tstart = max(start_pfn, memblock_region_memory_base_pfn(reg));
154 tend = min(end_pfn, memblock_region_memory_end_pfn(reg));
155 if (tstart < tend) {
156 /* Memory hole from start_pfn to tstart */
157 if (tstart > start_pfn)
158 break;
159
160 if (tend == end_pfn) {
161 ret = 1;
162 break;
163 }
164
165 start_pfn = tend + 1;
166 }
167 }
168
169 return ret;
170}
171
172/* Print firmware assisted dump configurations for debugging purpose. */
173static void fadump_show_config(void)
174{
175 pr_debug("Support for firmware-assisted dump (fadump): %s\n",
176 (fw_dump.fadump_supported ? "present" : "no support"));
177
178 if (!fw_dump.fadump_supported)
179 return;
180
181 pr_debug("Fadump enabled : %s\n",
182 (fw_dump.fadump_enabled ? "yes" : "no"));
183 pr_debug("Dump Active : %s\n",
184 (fw_dump.dump_active ? "yes" : "no"));
185 pr_debug("Dump section sizes:\n");
186 pr_debug(" CPU state data size: %lx\n", fw_dump.cpu_state_data_size);
187 pr_debug(" HPTE region size : %lx\n", fw_dump.hpte_region_size);
188 pr_debug("Boot memory size : %lx\n", fw_dump.boot_memory_size);
189}
190
191static unsigned long init_fadump_mem_struct(struct fadump_mem_struct *fdm,
192 unsigned long addr)
193{
194 if (!fdm)
195 return 0;
196
197 memset(fdm, 0, sizeof(struct fadump_mem_struct));
198 addr = addr & PAGE_MASK;
199
200 fdm->header.dump_format_version = cpu_to_be32(0x00000001);
201 fdm->header.dump_num_sections = cpu_to_be16(3);
202 fdm->header.dump_status_flag = 0;
203 fdm->header.offset_first_dump_section =
204 cpu_to_be32((u32)offsetof(struct fadump_mem_struct, cpu_state_data));
205
206 /*
207 * Fields for disk dump option.
208 * We are not using disk dump option, hence set these fields to 0.
209 */
210 fdm->header.dd_block_size = 0;
211 fdm->header.dd_block_offset = 0;
212 fdm->header.dd_num_blocks = 0;
213 fdm->header.dd_offset_disk_path = 0;
214
215 /* set 0 to disable an automatic dump-reboot. */
216 fdm->header.max_time_auto = 0;
217
218 /* Kernel dump sections */
219 /* cpu state data section. */
220 fdm->cpu_state_data.request_flag = cpu_to_be32(FADUMP_REQUEST_FLAG);
221 fdm->cpu_state_data.source_data_type = cpu_to_be16(FADUMP_CPU_STATE_DATA);
222 fdm->cpu_state_data.source_address = 0;
223 fdm->cpu_state_data.source_len = cpu_to_be64(fw_dump.cpu_state_data_size);
224 fdm->cpu_state_data.destination_address = cpu_to_be64(addr);
225 addr += fw_dump.cpu_state_data_size;
226
227 /* hpte region section */
228 fdm->hpte_region.request_flag = cpu_to_be32(FADUMP_REQUEST_FLAG);
229 fdm->hpte_region.source_data_type = cpu_to_be16(FADUMP_HPTE_REGION);
230 fdm->hpte_region.source_address = 0;
231 fdm->hpte_region.source_len = cpu_to_be64(fw_dump.hpte_region_size);
232 fdm->hpte_region.destination_address = cpu_to_be64(addr);
233 addr += fw_dump.hpte_region_size;
234
235 /* RMA region section */
236 fdm->rmr_region.request_flag = cpu_to_be32(FADUMP_REQUEST_FLAG);
237 fdm->rmr_region.source_data_type = cpu_to_be16(FADUMP_REAL_MODE_REGION);
238 fdm->rmr_region.source_address = cpu_to_be64(RMA_START);
239 fdm->rmr_region.source_len = cpu_to_be64(fw_dump.boot_memory_size);
240 fdm->rmr_region.destination_address = cpu_to_be64(addr);
241 addr += fw_dump.boot_memory_size;
242
243 return addr;
244}
245
246/**
247 * fadump_calculate_reserve_size(): reserve variable boot area 5% of System RAM
248 *
249 * Function to find the largest memory size we need to reserve during early
250 * boot process. This will be the size of the memory that is required for a
251 * kernel to boot successfully.
252 *
253 * This function has been taken from phyp-assisted dump feature implementation.
254 *
255 * returns larger of 256MB or 5% rounded down to multiples of 256MB.
256 *
257 * TODO: Come up with better approach to find out more accurate memory size
258 * that is required for a kernel to boot successfully.
259 *
260 */
261static inline unsigned long fadump_calculate_reserve_size(void)
262{
263 int ret;
264 unsigned long long base, size;
265
266 if (fw_dump.reserve_bootvar)
267 pr_warn("'fadump_reserve_mem=' parameter is deprecated in favor of 'crashkernel=' parameter.\n");
268
269 /*
270 * Check if the size is specified through crashkernel= cmdline
271 * option. If yes, then use that but ignore base as fadump reserves
272 * memory at a predefined offset.
273 */
274 ret = parse_crashkernel(boot_command_line, memblock_phys_mem_size(),
275 &size, &base);
276 if (ret == 0 && size > 0) {
277 unsigned long max_size;
278
279 if (fw_dump.reserve_bootvar)
280 pr_info("Using 'crashkernel=' parameter for memory reservation.\n");
281
282 fw_dump.reserve_bootvar = (unsigned long)size;
283
284 /*
285 * Adjust if the boot memory size specified is above
286 * the upper limit.
287 */
288 max_size = memblock_phys_mem_size() / MAX_BOOT_MEM_RATIO;
289 if (fw_dump.reserve_bootvar > max_size) {
290 fw_dump.reserve_bootvar = max_size;
291 pr_info("Adjusted boot memory size to %luMB\n",
292 (fw_dump.reserve_bootvar >> 20));
293 }
294
295 return fw_dump.reserve_bootvar;
296 } else if (fw_dump.reserve_bootvar) {
297 /*
298 * 'fadump_reserve_mem=' is being used to reserve memory
299 * for firmware-assisted dump.
300 */
301 return fw_dump.reserve_bootvar;
302 }
303
304 /* divide by 20 to get 5% of value */
305 size = memblock_phys_mem_size() / 20;
306
307 /* round it down in multiples of 256 */
308 size = size & ~0x0FFFFFFFUL;
309
310 /* Truncate to memory_limit. We don't want to over reserve the memory.*/
311 if (memory_limit && size > memory_limit)
312 size = memory_limit;
313
314 return (size > MIN_BOOT_MEM ? size : MIN_BOOT_MEM);
315}
316
317/*
318 * Calculate the total memory size required to be reserved for
319 * firmware-assisted dump registration.
320 */
321static unsigned long get_fadump_area_size(void)
322{
323 unsigned long size = 0;
324
325 size += fw_dump.cpu_state_data_size;
326 size += fw_dump.hpte_region_size;
327 size += fw_dump.boot_memory_size;
328 size += sizeof(struct fadump_crash_info_header);
329 size += sizeof(struct elfhdr); /* ELF core header.*/
330 size += sizeof(struct elf_phdr); /* place holder for cpu notes */
331 /* Program headers for crash memory regions. */
332 size += sizeof(struct elf_phdr) * (memblock_num_regions(memory) + 2);
333
334 size = PAGE_ALIGN(size);
335 return size;
336}
337
338int __init fadump_reserve_mem(void)
339{
340 unsigned long base, size, memory_boundary;
341
342 if (!fw_dump.fadump_enabled)
343 return 0;
344
345 if (!fw_dump.fadump_supported) {
346 printk(KERN_INFO "Firmware-assisted dump is not supported on"
347 " this hardware\n");
348 fw_dump.fadump_enabled = 0;
349 return 0;
350 }
351 /*
352 * Initialize boot memory size
353 * If dump is active then we have already calculated the size during
354 * first kernel.
355 */
356 if (fdm_active)
357 fw_dump.boot_memory_size = be64_to_cpu(fdm_active->rmr_region.source_len);
358 else
359 fw_dump.boot_memory_size = fadump_calculate_reserve_size();
360
361 /*
362 * Calculate the memory boundary.
363 * If memory_limit is less than actual memory boundary then reserve
364 * the memory for fadump beyond the memory_limit and adjust the
365 * memory_limit accordingly, so that the running kernel can run with
366 * specified memory_limit.
367 */
368 if (memory_limit && memory_limit < memblock_end_of_DRAM()) {
369 size = get_fadump_area_size();
370 if ((memory_limit + size) < memblock_end_of_DRAM())
371 memory_limit += size;
372 else
373 memory_limit = memblock_end_of_DRAM();
374 printk(KERN_INFO "Adjusted memory_limit for firmware-assisted"
375 " dump, now %#016llx\n", memory_limit);
376 }
377 if (memory_limit)
378 memory_boundary = memory_limit;
379 else
380 memory_boundary = memblock_end_of_DRAM();
381
382 if (fw_dump.dump_active) {
383 printk(KERN_INFO "Firmware-assisted dump is active.\n");
384 /*
385 * If last boot has crashed then reserve all the memory
386 * above boot_memory_size so that we don't touch it until
387 * dump is written to disk by userspace tool. This memory
388 * will be released for general use once the dump is saved.
389 */
390 base = fw_dump.boot_memory_size;
391 size = memory_boundary - base;
392 memblock_reserve(base, size);
393 printk(KERN_INFO "Reserved %ldMB of memory at %ldMB "
394 "for saving crash dump\n",
395 (unsigned long)(size >> 20),
396 (unsigned long)(base >> 20));
397
398 fw_dump.fadumphdr_addr =
399 be64_to_cpu(fdm_active->rmr_region.destination_address) +
400 be64_to_cpu(fdm_active->rmr_region.source_len);
401 pr_debug("fadumphdr_addr = %p\n",
402 (void *) fw_dump.fadumphdr_addr);
403 } else {
404 size = get_fadump_area_size();
405
406 /*
407 * Reserve memory at an offset closer to bottom of the RAM to
408 * minimize the impact of memory hot-remove operation. We can't
409 * use memblock_find_in_range() here since it doesn't allocate
410 * from bottom to top.
411 */
412 for (base = fw_dump.boot_memory_size;
413 base <= (memory_boundary - size);
414 base += size) {
415 if (memblock_is_region_memory(base, size) &&
416 !memblock_is_region_reserved(base, size))
417 break;
418 }
419 if ((base > (memory_boundary - size)) ||
420 memblock_reserve(base, size)) {
421 pr_err("Failed to reserve memory\n");
422 return 0;
423 }
424
425 pr_info("Reserved %ldMB of memory at %ldMB for firmware-"
426 "assisted dump (System RAM: %ldMB)\n",
427 (unsigned long)(size >> 20),
428 (unsigned long)(base >> 20),
429 (unsigned long)(memblock_phys_mem_size() >> 20));
430 }
431
432 fw_dump.reserve_dump_area_start = base;
433 fw_dump.reserve_dump_area_size = size;
434 return 1;
435}
436
437unsigned long __init arch_reserved_kernel_pages(void)
438{
439 return memblock_reserved_size() / PAGE_SIZE;
440}
441
442/* Look for fadump= cmdline option. */
443static int __init early_fadump_param(char *p)
444{
445 if (!p)
446 return 1;
447
448 if (strncmp(p, "on", 2) == 0)
449 fw_dump.fadump_enabled = 1;
450 else if (strncmp(p, "off", 3) == 0)
451 fw_dump.fadump_enabled = 0;
452
453 return 0;
454}
455early_param("fadump", early_fadump_param);
456
457/*
458 * Look for fadump_reserve_mem= cmdline option
459 * TODO: Remove references to 'fadump_reserve_mem=' parameter,
460 * the sooner 'crashkernel=' parameter is accustomed to.
461 */
462static int __init early_fadump_reserve_mem(char *p)
463{
464 if (p)
465 fw_dump.reserve_bootvar = memparse(p, &p);
466 return 0;
467}
468early_param("fadump_reserve_mem", early_fadump_reserve_mem);
469
470static int register_fw_dump(struct fadump_mem_struct *fdm)
471{
472 int rc, err;
473 unsigned int wait_time;
474
475 pr_debug("Registering for firmware-assisted kernel dump...\n");
476
477 /* TODO: Add upper time limit for the delay */
478 do {
479 rc = rtas_call(fw_dump.ibm_configure_kernel_dump, 3, 1, NULL,
480 FADUMP_REGISTER, fdm,
481 sizeof(struct fadump_mem_struct));
482
483 wait_time = rtas_busy_delay_time(rc);
484 if (wait_time)
485 mdelay(wait_time);
486
487 } while (wait_time);
488
489 err = -EIO;
490 switch (rc) {
491 default:
492 pr_err("Failed to register. Unknown Error(%d).\n", rc);
493 break;
494 case -1:
495 printk(KERN_ERR "Failed to register firmware-assisted kernel"
496 " dump. Hardware Error(%d).\n", rc);
497 break;
498 case -3:
499 if (!is_boot_memory_area_contiguous())
500 pr_err("Can't have holes in boot memory area while "
501 "registering fadump\n");
502
503 printk(KERN_ERR "Failed to register firmware-assisted kernel"
504 " dump. Parameter Error(%d).\n", rc);
505 err = -EINVAL;
506 break;
507 case -9:
508 printk(KERN_ERR "firmware-assisted kernel dump is already "
509 " registered.");
510 fw_dump.dump_registered = 1;
511 err = -EEXIST;
512 break;
513 case 0:
514 printk(KERN_INFO "firmware-assisted kernel dump registration"
515 " is successful\n");
516 fw_dump.dump_registered = 1;
517 err = 0;
518 break;
519 }
520 return err;
521}
522
523void crash_fadump(struct pt_regs *regs, const char *str)
524{
525 struct fadump_crash_info_header *fdh = NULL;
526 int old_cpu, this_cpu;
527
528 if (!should_fadump_crash())
529 return;
530
531 /*
532 * old_cpu == -1 means this is the first CPU which has come here,
533 * go ahead and trigger fadump.
534 *
535 * old_cpu != -1 means some other CPU has already on it's way
536 * to trigger fadump, just keep looping here.
537 */
538 this_cpu = smp_processor_id();
539 old_cpu = cmpxchg(&crashing_cpu, -1, this_cpu);
540
541 if (old_cpu != -1) {
542 /*
543 * We can't loop here indefinitely. Wait as long as fadump
544 * is in force. If we race with fadump un-registration this
545 * loop will break and then we go down to normal panic path
546 * and reboot. If fadump is in force the first crashing
547 * cpu will definitely trigger fadump.
548 */
549 while (fw_dump.dump_registered)
550 cpu_relax();
551 return;
552 }
553
554 fdh = __va(fw_dump.fadumphdr_addr);
555 fdh->crashing_cpu = crashing_cpu;
556 crash_save_vmcoreinfo();
557
558 if (regs)
559 fdh->regs = *regs;
560 else
561 ppc_save_regs(&fdh->regs);
562
563 fdh->online_mask = *cpu_online_mask;
564
565 /* Call ibm,os-term rtas call to trigger firmware assisted dump */
566 rtas_os_term((char *)str);
567}
568
569#define GPR_MASK 0xffffff0000000000
570static inline int fadump_gpr_index(u64 id)
571{
572 int i = -1;
573 char str[3];
574
575 if ((id & GPR_MASK) == REG_ID("GPR")) {
576 /* get the digits at the end */
577 id &= ~GPR_MASK;
578 id >>= 24;
579 str[2] = '\0';
580 str[1] = id & 0xff;
581 str[0] = (id >> 8) & 0xff;
582 sscanf(str, "%d", &i);
583 if (i > 31)
584 i = -1;
585 }
586 return i;
587}
588
589static inline void fadump_set_regval(struct pt_regs *regs, u64 reg_id,
590 u64 reg_val)
591{
592 int i;
593
594 i = fadump_gpr_index(reg_id);
595 if (i >= 0)
596 regs->gpr[i] = (unsigned long)reg_val;
597 else if (reg_id == REG_ID("NIA"))
598 regs->nip = (unsigned long)reg_val;
599 else if (reg_id == REG_ID("MSR"))
600 regs->msr = (unsigned long)reg_val;
601 else if (reg_id == REG_ID("CTR"))
602 regs->ctr = (unsigned long)reg_val;
603 else if (reg_id == REG_ID("LR"))
604 regs->link = (unsigned long)reg_val;
605 else if (reg_id == REG_ID("XER"))
606 regs->xer = (unsigned long)reg_val;
607 else if (reg_id == REG_ID("CR"))
608 regs->ccr = (unsigned long)reg_val;
609 else if (reg_id == REG_ID("DAR"))
610 regs->dar = (unsigned long)reg_val;
611 else if (reg_id == REG_ID("DSISR"))
612 regs->dsisr = (unsigned long)reg_val;
613}
614
615static struct fadump_reg_entry*
616fadump_read_registers(struct fadump_reg_entry *reg_entry, struct pt_regs *regs)
617{
618 memset(regs, 0, sizeof(struct pt_regs));
619
620 while (be64_to_cpu(reg_entry->reg_id) != REG_ID("CPUEND")) {
621 fadump_set_regval(regs, be64_to_cpu(reg_entry->reg_id),
622 be64_to_cpu(reg_entry->reg_value));
623 reg_entry++;
624 }
625 reg_entry++;
626 return reg_entry;
627}
628
629static u32 *fadump_regs_to_elf_notes(u32 *buf, struct pt_regs *regs)
630{
631 struct elf_prstatus prstatus;
632
633 memset(&prstatus, 0, sizeof(prstatus));
634 /*
635 * FIXME: How do i get PID? Do I really need it?
636 * prstatus.pr_pid = ????
637 */
638 elf_core_copy_kernel_regs(&prstatus.pr_reg, regs);
639 buf = append_elf_note(buf, CRASH_CORE_NOTE_NAME, NT_PRSTATUS,
640 &prstatus, sizeof(prstatus));
641 return buf;
642}
643
644static void fadump_update_elfcore_header(char *bufp)
645{
646 struct elfhdr *elf;
647 struct elf_phdr *phdr;
648
649 elf = (struct elfhdr *)bufp;
650 bufp += sizeof(struct elfhdr);
651
652 /* First note is a place holder for cpu notes info. */
653 phdr = (struct elf_phdr *)bufp;
654
655 if (phdr->p_type == PT_NOTE) {
656 phdr->p_paddr = fw_dump.cpu_notes_buf;
657 phdr->p_offset = phdr->p_paddr;
658 phdr->p_filesz = fw_dump.cpu_notes_buf_size;
659 phdr->p_memsz = fw_dump.cpu_notes_buf_size;
660 }
661 return;
662}
663
664static void *fadump_cpu_notes_buf_alloc(unsigned long size)
665{
666 void *vaddr;
667 struct page *page;
668 unsigned long order, count, i;
669
670 order = get_order(size);
671 vaddr = (void *)__get_free_pages(GFP_KERNEL|__GFP_ZERO, order);
672 if (!vaddr)
673 return NULL;
674
675 count = 1 << order;
676 page = virt_to_page(vaddr);
677 for (i = 0; i < count; i++)
678 SetPageReserved(page + i);
679 return vaddr;
680}
681
682static void fadump_cpu_notes_buf_free(unsigned long vaddr, unsigned long size)
683{
684 struct page *page;
685 unsigned long order, count, i;
686
687 order = get_order(size);
688 count = 1 << order;
689 page = virt_to_page(vaddr);
690 for (i = 0; i < count; i++)
691 ClearPageReserved(page + i);
692 __free_pages(page, order);
693}
694
695/*
696 * Read CPU state dump data and convert it into ELF notes.
697 * The CPU dump starts with magic number "REGSAVE". NumCpusOffset should be
698 * used to access the data to allow for additional fields to be added without
699 * affecting compatibility. Each list of registers for a CPU starts with
700 * "CPUSTRT" and ends with "CPUEND". Each register entry is of 16 bytes,
701 * 8 Byte ASCII identifier and 8 Byte register value. The register entry
702 * with identifier "CPUSTRT" and "CPUEND" contains 4 byte cpu id as part
703 * of register value. For more details refer to PAPR document.
704 *
705 * Only for the crashing cpu we ignore the CPU dump data and get exact
706 * state from fadump crash info structure populated by first kernel at the
707 * time of crash.
708 */
709static int __init fadump_build_cpu_notes(const struct fadump_mem_struct *fdm)
710{
711 struct fadump_reg_save_area_header *reg_header;
712 struct fadump_reg_entry *reg_entry;
713 struct fadump_crash_info_header *fdh = NULL;
714 void *vaddr;
715 unsigned long addr;
716 u32 num_cpus, *note_buf;
717 struct pt_regs regs;
718 int i, rc = 0, cpu = 0;
719
720 if (!fdm->cpu_state_data.bytes_dumped)
721 return -EINVAL;
722
723 addr = be64_to_cpu(fdm->cpu_state_data.destination_address);
724 vaddr = __va(addr);
725
726 reg_header = vaddr;
727 if (be64_to_cpu(reg_header->magic_number) != REGSAVE_AREA_MAGIC) {
728 printk(KERN_ERR "Unable to read register save area.\n");
729 return -ENOENT;
730 }
731 pr_debug("--------CPU State Data------------\n");
732 pr_debug("Magic Number: %llx\n", be64_to_cpu(reg_header->magic_number));
733 pr_debug("NumCpuOffset: %x\n", be32_to_cpu(reg_header->num_cpu_offset));
734
735 vaddr += be32_to_cpu(reg_header->num_cpu_offset);
736 num_cpus = be32_to_cpu(*((__be32 *)(vaddr)));
737 pr_debug("NumCpus : %u\n", num_cpus);
738 vaddr += sizeof(u32);
739 reg_entry = (struct fadump_reg_entry *)vaddr;
740
741 /* Allocate buffer to hold cpu crash notes. */
742 fw_dump.cpu_notes_buf_size = num_cpus * sizeof(note_buf_t);
743 fw_dump.cpu_notes_buf_size = PAGE_ALIGN(fw_dump.cpu_notes_buf_size);
744 note_buf = fadump_cpu_notes_buf_alloc(fw_dump.cpu_notes_buf_size);
745 if (!note_buf) {
746 printk(KERN_ERR "Failed to allocate 0x%lx bytes for "
747 "cpu notes buffer\n", fw_dump.cpu_notes_buf_size);
748 return -ENOMEM;
749 }
750 fw_dump.cpu_notes_buf = __pa(note_buf);
751
752 pr_debug("Allocated buffer for cpu notes of size %ld at %p\n",
753 (num_cpus * sizeof(note_buf_t)), note_buf);
754
755 if (fw_dump.fadumphdr_addr)
756 fdh = __va(fw_dump.fadumphdr_addr);
757
758 for (i = 0; i < num_cpus; i++) {
759 if (be64_to_cpu(reg_entry->reg_id) != REG_ID("CPUSTRT")) {
760 printk(KERN_ERR "Unable to read CPU state data\n");
761 rc = -ENOENT;
762 goto error_out;
763 }
764 /* Lower 4 bytes of reg_value contains logical cpu id */
765 cpu = be64_to_cpu(reg_entry->reg_value) & FADUMP_CPU_ID_MASK;
766 if (fdh && !cpumask_test_cpu(cpu, &fdh->online_mask)) {
767 SKIP_TO_NEXT_CPU(reg_entry);
768 continue;
769 }
770 pr_debug("Reading register data for cpu %d...\n", cpu);
771 if (fdh && fdh->crashing_cpu == cpu) {
772 regs = fdh->regs;
773 note_buf = fadump_regs_to_elf_notes(note_buf, ®s);
774 SKIP_TO_NEXT_CPU(reg_entry);
775 } else {
776 reg_entry++;
777 reg_entry = fadump_read_registers(reg_entry, ®s);
778 note_buf = fadump_regs_to_elf_notes(note_buf, ®s);
779 }
780 }
781 final_note(note_buf);
782
783 if (fdh) {
784 pr_debug("Updating elfcore header (%llx) with cpu notes\n",
785 fdh->elfcorehdr_addr);
786 fadump_update_elfcore_header((char *)__va(fdh->elfcorehdr_addr));
787 }
788 return 0;
789
790error_out:
791 fadump_cpu_notes_buf_free((unsigned long)__va(fw_dump.cpu_notes_buf),
792 fw_dump.cpu_notes_buf_size);
793 fw_dump.cpu_notes_buf = 0;
794 fw_dump.cpu_notes_buf_size = 0;
795 return rc;
796
797}
798
799/*
800 * Validate and process the dump data stored by firmware before exporting
801 * it through '/proc/vmcore'.
802 */
803static int __init process_fadump(const struct fadump_mem_struct *fdm_active)
804{
805 struct fadump_crash_info_header *fdh;
806 int rc = 0;
807
808 if (!fdm_active || !fw_dump.fadumphdr_addr)
809 return -EINVAL;
810
811 /* Check if the dump data is valid. */
812 if ((be16_to_cpu(fdm_active->header.dump_status_flag) == FADUMP_ERROR_FLAG) ||
813 (fdm_active->cpu_state_data.error_flags != 0) ||
814 (fdm_active->rmr_region.error_flags != 0)) {
815 printk(KERN_ERR "Dump taken by platform is not valid\n");
816 return -EINVAL;
817 }
818 if ((fdm_active->rmr_region.bytes_dumped !=
819 fdm_active->rmr_region.source_len) ||
820 !fdm_active->cpu_state_data.bytes_dumped) {
821 printk(KERN_ERR "Dump taken by platform is incomplete\n");
822 return -EINVAL;
823 }
824
825 /* Validate the fadump crash info header */
826 fdh = __va(fw_dump.fadumphdr_addr);
827 if (fdh->magic_number != FADUMP_CRASH_INFO_MAGIC) {
828 printk(KERN_ERR "Crash info header is not valid.\n");
829 return -EINVAL;
830 }
831
832 rc = fadump_build_cpu_notes(fdm_active);
833 if (rc)
834 return rc;
835
836 /*
837 * We are done validating dump info and elfcore header is now ready
838 * to be exported. set elfcorehdr_addr so that vmcore module will
839 * export the elfcore header through '/proc/vmcore'.
840 */
841 elfcorehdr_addr = fdh->elfcorehdr_addr;
842
843 return 0;
844}
845
846static inline void fadump_add_crash_memory(unsigned long long base,
847 unsigned long long end)
848{
849 if (base == end)
850 return;
851
852 pr_debug("crash_memory_range[%d] [%#016llx-%#016llx], %#llx bytes\n",
853 crash_mem_ranges, base, end - 1, (end - base));
854 crash_memory_ranges[crash_mem_ranges].base = base;
855 crash_memory_ranges[crash_mem_ranges].size = end - base;
856 crash_mem_ranges++;
857}
858
859static void fadump_exclude_reserved_area(unsigned long long start,
860 unsigned long long end)
861{
862 unsigned long long ra_start, ra_end;
863
864 ra_start = fw_dump.reserve_dump_area_start;
865 ra_end = ra_start + fw_dump.reserve_dump_area_size;
866
867 if ((ra_start < end) && (ra_end > start)) {
868 if ((start < ra_start) && (end > ra_end)) {
869 fadump_add_crash_memory(start, ra_start);
870 fadump_add_crash_memory(ra_end, end);
871 } else if (start < ra_start) {
872 fadump_add_crash_memory(start, ra_start);
873 } else if (ra_end < end) {
874 fadump_add_crash_memory(ra_end, end);
875 }
876 } else
877 fadump_add_crash_memory(start, end);
878}
879
880static int fadump_init_elfcore_header(char *bufp)
881{
882 struct elfhdr *elf;
883
884 elf = (struct elfhdr *) bufp;
885 bufp += sizeof(struct elfhdr);
886 memcpy(elf->e_ident, ELFMAG, SELFMAG);
887 elf->e_ident[EI_CLASS] = ELF_CLASS;
888 elf->e_ident[EI_DATA] = ELF_DATA;
889 elf->e_ident[EI_VERSION] = EV_CURRENT;
890 elf->e_ident[EI_OSABI] = ELF_OSABI;
891 memset(elf->e_ident+EI_PAD, 0, EI_NIDENT-EI_PAD);
892 elf->e_type = ET_CORE;
893 elf->e_machine = ELF_ARCH;
894 elf->e_version = EV_CURRENT;
895 elf->e_entry = 0;
896 elf->e_phoff = sizeof(struct elfhdr);
897 elf->e_shoff = 0;
898#if defined(_CALL_ELF)
899 elf->e_flags = _CALL_ELF;
900#else
901 elf->e_flags = 0;
902#endif
903 elf->e_ehsize = sizeof(struct elfhdr);
904 elf->e_phentsize = sizeof(struct elf_phdr);
905 elf->e_phnum = 0;
906 elf->e_shentsize = 0;
907 elf->e_shnum = 0;
908 elf->e_shstrndx = 0;
909
910 return 0;
911}
912
913/*
914 * Traverse through memblock structure and setup crash memory ranges. These
915 * ranges will be used create PT_LOAD program headers in elfcore header.
916 */
917static void fadump_setup_crash_memory_ranges(void)
918{
919 struct memblock_region *reg;
920 unsigned long long start, end;
921
922 pr_debug("Setup crash memory ranges.\n");
923 crash_mem_ranges = 0;
924 /*
925 * add the first memory chunk (RMA_START through boot_memory_size) as
926 * a separate memory chunk. The reason is, at the time crash firmware
927 * will move the content of this memory chunk to different location
928 * specified during fadump registration. We need to create a separate
929 * program header for this chunk with the correct offset.
930 */
931 fadump_add_crash_memory(RMA_START, fw_dump.boot_memory_size);
932
933 for_each_memblock(memory, reg) {
934 start = (unsigned long long)reg->base;
935 end = start + (unsigned long long)reg->size;
936
937 /*
938 * skip the first memory chunk that is already added (RMA_START
939 * through boot_memory_size). This logic needs a relook if and
940 * when RMA_START changes to a non-zero value.
941 */
942 BUILD_BUG_ON(RMA_START != 0);
943 if (start < fw_dump.boot_memory_size) {
944 if (end > fw_dump.boot_memory_size)
945 start = fw_dump.boot_memory_size;
946 else
947 continue;
948 }
949
950 /* add this range excluding the reserved dump area. */
951 fadump_exclude_reserved_area(start, end);
952 }
953}
954
955/*
956 * If the given physical address falls within the boot memory region then
957 * return the relocated address that points to the dump region reserved
958 * for saving initial boot memory contents.
959 */
960static inline unsigned long fadump_relocate(unsigned long paddr)
961{
962 if (paddr > RMA_START && paddr < fw_dump.boot_memory_size)
963 return be64_to_cpu(fdm.rmr_region.destination_address) + paddr;
964 else
965 return paddr;
966}
967
968static int fadump_create_elfcore_headers(char *bufp)
969{
970 struct elfhdr *elf;
971 struct elf_phdr *phdr;
972 int i;
973
974 fadump_init_elfcore_header(bufp);
975 elf = (struct elfhdr *)bufp;
976 bufp += sizeof(struct elfhdr);
977
978 /*
979 * setup ELF PT_NOTE, place holder for cpu notes info. The notes info
980 * will be populated during second kernel boot after crash. Hence
981 * this PT_NOTE will always be the first elf note.
982 *
983 * NOTE: Any new ELF note addition should be placed after this note.
984 */
985 phdr = (struct elf_phdr *)bufp;
986 bufp += sizeof(struct elf_phdr);
987 phdr->p_type = PT_NOTE;
988 phdr->p_flags = 0;
989 phdr->p_vaddr = 0;
990 phdr->p_align = 0;
991
992 phdr->p_offset = 0;
993 phdr->p_paddr = 0;
994 phdr->p_filesz = 0;
995 phdr->p_memsz = 0;
996
997 (elf->e_phnum)++;
998
999 /* setup ELF PT_NOTE for vmcoreinfo */
1000 phdr = (struct elf_phdr *)bufp;
1001 bufp += sizeof(struct elf_phdr);
1002 phdr->p_type = PT_NOTE;
1003 phdr->p_flags = 0;
1004 phdr->p_vaddr = 0;
1005 phdr->p_align = 0;
1006
1007 phdr->p_paddr = fadump_relocate(paddr_vmcoreinfo_note());
1008 phdr->p_offset = phdr->p_paddr;
1009 phdr->p_memsz = phdr->p_filesz = VMCOREINFO_NOTE_SIZE;
1010
1011 /* Increment number of program headers. */
1012 (elf->e_phnum)++;
1013
1014 /* setup PT_LOAD sections. */
1015
1016 for (i = 0; i < crash_mem_ranges; i++) {
1017 unsigned long long mbase, msize;
1018 mbase = crash_memory_ranges[i].base;
1019 msize = crash_memory_ranges[i].size;
1020
1021 if (!msize)
1022 continue;
1023
1024 phdr = (struct elf_phdr *)bufp;
1025 bufp += sizeof(struct elf_phdr);
1026 phdr->p_type = PT_LOAD;
1027 phdr->p_flags = PF_R|PF_W|PF_X;
1028 phdr->p_offset = mbase;
1029
1030 if (mbase == RMA_START) {
1031 /*
1032 * The entire RMA region will be moved by firmware
1033 * to the specified destination_address. Hence set
1034 * the correct offset.
1035 */
1036 phdr->p_offset = be64_to_cpu(fdm.rmr_region.destination_address);
1037 }
1038
1039 phdr->p_paddr = mbase;
1040 phdr->p_vaddr = (unsigned long)__va(mbase);
1041 phdr->p_filesz = msize;
1042 phdr->p_memsz = msize;
1043 phdr->p_align = 0;
1044
1045 /* Increment number of program headers. */
1046 (elf->e_phnum)++;
1047 }
1048 return 0;
1049}
1050
1051static unsigned long init_fadump_header(unsigned long addr)
1052{
1053 struct fadump_crash_info_header *fdh;
1054
1055 if (!addr)
1056 return 0;
1057
1058 fw_dump.fadumphdr_addr = addr;
1059 fdh = __va(addr);
1060 addr += sizeof(struct fadump_crash_info_header);
1061
1062 memset(fdh, 0, sizeof(struct fadump_crash_info_header));
1063 fdh->magic_number = FADUMP_CRASH_INFO_MAGIC;
1064 fdh->elfcorehdr_addr = addr;
1065 /* We will set the crashing cpu id in crash_fadump() during crash. */
1066 fdh->crashing_cpu = CPU_UNKNOWN;
1067
1068 return addr;
1069}
1070
1071static int register_fadump(void)
1072{
1073 unsigned long addr;
1074 void *vaddr;
1075
1076 /*
1077 * If no memory is reserved then we can not register for firmware-
1078 * assisted dump.
1079 */
1080 if (!fw_dump.reserve_dump_area_size)
1081 return -ENODEV;
1082
1083 fadump_setup_crash_memory_ranges();
1084
1085 addr = be64_to_cpu(fdm.rmr_region.destination_address) + be64_to_cpu(fdm.rmr_region.source_len);
1086 /* Initialize fadump crash info header. */
1087 addr = init_fadump_header(addr);
1088 vaddr = __va(addr);
1089
1090 pr_debug("Creating ELF core headers at %#016lx\n", addr);
1091 fadump_create_elfcore_headers(vaddr);
1092
1093 /* register the future kernel dump with firmware. */
1094 return register_fw_dump(&fdm);
1095}
1096
1097static int fadump_unregister_dump(struct fadump_mem_struct *fdm)
1098{
1099 int rc = 0;
1100 unsigned int wait_time;
1101
1102 pr_debug("Un-register firmware-assisted dump\n");
1103
1104 /* TODO: Add upper time limit for the delay */
1105 do {
1106 rc = rtas_call(fw_dump.ibm_configure_kernel_dump, 3, 1, NULL,
1107 FADUMP_UNREGISTER, fdm,
1108 sizeof(struct fadump_mem_struct));
1109
1110 wait_time = rtas_busy_delay_time(rc);
1111 if (wait_time)
1112 mdelay(wait_time);
1113 } while (wait_time);
1114
1115 if (rc) {
1116 printk(KERN_ERR "Failed to un-register firmware-assisted dump."
1117 " unexpected error(%d).\n", rc);
1118 return rc;
1119 }
1120 fw_dump.dump_registered = 0;
1121 return 0;
1122}
1123
1124static int fadump_invalidate_dump(struct fadump_mem_struct *fdm)
1125{
1126 int rc = 0;
1127 unsigned int wait_time;
1128
1129 pr_debug("Invalidating firmware-assisted dump registration\n");
1130
1131 /* TODO: Add upper time limit for the delay */
1132 do {
1133 rc = rtas_call(fw_dump.ibm_configure_kernel_dump, 3, 1, NULL,
1134 FADUMP_INVALIDATE, fdm,
1135 sizeof(struct fadump_mem_struct));
1136
1137 wait_time = rtas_busy_delay_time(rc);
1138 if (wait_time)
1139 mdelay(wait_time);
1140 } while (wait_time);
1141
1142 if (rc) {
1143 pr_err("Failed to invalidate firmware-assisted dump registration. Unexpected error (%d).\n", rc);
1144 return rc;
1145 }
1146 fw_dump.dump_active = 0;
1147 fdm_active = NULL;
1148 return 0;
1149}
1150
1151void fadump_cleanup(void)
1152{
1153 /* Invalidate the registration only if dump is active. */
1154 if (fw_dump.dump_active) {
1155 init_fadump_mem_struct(&fdm,
1156 be64_to_cpu(fdm_active->cpu_state_data.destination_address));
1157 fadump_invalidate_dump(&fdm);
1158 }
1159}
1160
1161static void fadump_free_reserved_memory(unsigned long start_pfn,
1162 unsigned long end_pfn)
1163{
1164 unsigned long pfn;
1165 unsigned long time_limit = jiffies + HZ;
1166
1167 pr_info("freeing reserved memory (0x%llx - 0x%llx)\n",
1168 PFN_PHYS(start_pfn), PFN_PHYS(end_pfn));
1169
1170 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
1171 free_reserved_page(pfn_to_page(pfn));
1172
1173 if (time_after(jiffies, time_limit)) {
1174 cond_resched();
1175 time_limit = jiffies + HZ;
1176 }
1177 }
1178}
1179
1180/*
1181 * Skip memory holes and free memory that was actually reserved.
1182 */
1183static void fadump_release_reserved_area(unsigned long start, unsigned long end)
1184{
1185 struct memblock_region *reg;
1186 unsigned long tstart, tend;
1187 unsigned long start_pfn = PHYS_PFN(start);
1188 unsigned long end_pfn = PHYS_PFN(end);
1189
1190 for_each_memblock(memory, reg) {
1191 tstart = max(start_pfn, memblock_region_memory_base_pfn(reg));
1192 tend = min(end_pfn, memblock_region_memory_end_pfn(reg));
1193 if (tstart < tend) {
1194 fadump_free_reserved_memory(tstart, tend);
1195
1196 if (tend == end_pfn)
1197 break;
1198
1199 start_pfn = tend + 1;
1200 }
1201 }
1202}
1203
1204/*
1205 * Release the memory that was reserved in early boot to preserve the memory
1206 * contents. The released memory will be available for general use.
1207 */
1208static void fadump_release_memory(unsigned long begin, unsigned long end)
1209{
1210 unsigned long ra_start, ra_end;
1211
1212 ra_start = fw_dump.reserve_dump_area_start;
1213 ra_end = ra_start + fw_dump.reserve_dump_area_size;
1214
1215 /*
1216 * exclude the dump reserve area. Will reuse it for next
1217 * fadump registration.
1218 */
1219 if (begin < ra_end && end > ra_start) {
1220 if (begin < ra_start)
1221 fadump_release_reserved_area(begin, ra_start);
1222 if (end > ra_end)
1223 fadump_release_reserved_area(ra_end, end);
1224 } else
1225 fadump_release_reserved_area(begin, end);
1226}
1227
1228static void fadump_invalidate_release_mem(void)
1229{
1230 unsigned long reserved_area_start, reserved_area_end;
1231 unsigned long destination_address;
1232
1233 mutex_lock(&fadump_mutex);
1234 if (!fw_dump.dump_active) {
1235 mutex_unlock(&fadump_mutex);
1236 return;
1237 }
1238
1239 destination_address = be64_to_cpu(fdm_active->cpu_state_data.destination_address);
1240 fadump_cleanup();
1241 mutex_unlock(&fadump_mutex);
1242
1243 /*
1244 * Save the current reserved memory bounds we will require them
1245 * later for releasing the memory for general use.
1246 */
1247 reserved_area_start = fw_dump.reserve_dump_area_start;
1248 reserved_area_end = reserved_area_start +
1249 fw_dump.reserve_dump_area_size;
1250 /*
1251 * Setup reserve_dump_area_start and its size so that we can
1252 * reuse this reserved memory for Re-registration.
1253 */
1254 fw_dump.reserve_dump_area_start = destination_address;
1255 fw_dump.reserve_dump_area_size = get_fadump_area_size();
1256
1257 fadump_release_memory(reserved_area_start, reserved_area_end);
1258 if (fw_dump.cpu_notes_buf) {
1259 fadump_cpu_notes_buf_free(
1260 (unsigned long)__va(fw_dump.cpu_notes_buf),
1261 fw_dump.cpu_notes_buf_size);
1262 fw_dump.cpu_notes_buf = 0;
1263 fw_dump.cpu_notes_buf_size = 0;
1264 }
1265 /* Initialize the kernel dump memory structure for FAD registration. */
1266 init_fadump_mem_struct(&fdm, fw_dump.reserve_dump_area_start);
1267}
1268
1269static ssize_t fadump_release_memory_store(struct kobject *kobj,
1270 struct kobj_attribute *attr,
1271 const char *buf, size_t count)
1272{
1273 int input = -1;
1274
1275 if (!fw_dump.dump_active)
1276 return -EPERM;
1277
1278 if (kstrtoint(buf, 0, &input))
1279 return -EINVAL;
1280
1281 if (input == 1) {
1282 /*
1283 * Take away the '/proc/vmcore'. We are releasing the dump
1284 * memory, hence it will not be valid anymore.
1285 */
1286#ifdef CONFIG_PROC_VMCORE
1287 vmcore_cleanup();
1288#endif
1289 fadump_invalidate_release_mem();
1290
1291 } else
1292 return -EINVAL;
1293 return count;
1294}
1295
1296static ssize_t fadump_enabled_show(struct kobject *kobj,
1297 struct kobj_attribute *attr,
1298 char *buf)
1299{
1300 return sprintf(buf, "%d\n", fw_dump.fadump_enabled);
1301}
1302
1303static ssize_t fadump_register_show(struct kobject *kobj,
1304 struct kobj_attribute *attr,
1305 char *buf)
1306{
1307 return sprintf(buf, "%d\n", fw_dump.dump_registered);
1308}
1309
1310static ssize_t fadump_register_store(struct kobject *kobj,
1311 struct kobj_attribute *attr,
1312 const char *buf, size_t count)
1313{
1314 int ret = 0;
1315 int input = -1;
1316
1317 if (!fw_dump.fadump_enabled || fdm_active)
1318 return -EPERM;
1319
1320 if (kstrtoint(buf, 0, &input))
1321 return -EINVAL;
1322
1323 mutex_lock(&fadump_mutex);
1324
1325 switch (input) {
1326 case 0:
1327 if (fw_dump.dump_registered == 0) {
1328 goto unlock_out;
1329 }
1330 /* Un-register Firmware-assisted dump */
1331 fadump_unregister_dump(&fdm);
1332 break;
1333 case 1:
1334 if (fw_dump.dump_registered == 1) {
1335 ret = -EEXIST;
1336 goto unlock_out;
1337 }
1338 /* Register Firmware-assisted dump */
1339 ret = register_fadump();
1340 break;
1341 default:
1342 ret = -EINVAL;
1343 break;
1344 }
1345
1346unlock_out:
1347 mutex_unlock(&fadump_mutex);
1348 return ret < 0 ? ret : count;
1349}
1350
1351static int fadump_region_show(struct seq_file *m, void *private)
1352{
1353 const struct fadump_mem_struct *fdm_ptr;
1354
1355 if (!fw_dump.fadump_enabled)
1356 return 0;
1357
1358 mutex_lock(&fadump_mutex);
1359 if (fdm_active)
1360 fdm_ptr = fdm_active;
1361 else {
1362 mutex_unlock(&fadump_mutex);
1363 fdm_ptr = &fdm;
1364 }
1365
1366 seq_printf(m,
1367 "CPU : [%#016llx-%#016llx] %#llx bytes, "
1368 "Dumped: %#llx\n",
1369 be64_to_cpu(fdm_ptr->cpu_state_data.destination_address),
1370 be64_to_cpu(fdm_ptr->cpu_state_data.destination_address) +
1371 be64_to_cpu(fdm_ptr->cpu_state_data.source_len) - 1,
1372 be64_to_cpu(fdm_ptr->cpu_state_data.source_len),
1373 be64_to_cpu(fdm_ptr->cpu_state_data.bytes_dumped));
1374 seq_printf(m,
1375 "HPTE: [%#016llx-%#016llx] %#llx bytes, "
1376 "Dumped: %#llx\n",
1377 be64_to_cpu(fdm_ptr->hpte_region.destination_address),
1378 be64_to_cpu(fdm_ptr->hpte_region.destination_address) +
1379 be64_to_cpu(fdm_ptr->hpte_region.source_len) - 1,
1380 be64_to_cpu(fdm_ptr->hpte_region.source_len),
1381 be64_to_cpu(fdm_ptr->hpte_region.bytes_dumped));
1382 seq_printf(m,
1383 "DUMP: [%#016llx-%#016llx] %#llx bytes, "
1384 "Dumped: %#llx\n",
1385 be64_to_cpu(fdm_ptr->rmr_region.destination_address),
1386 be64_to_cpu(fdm_ptr->rmr_region.destination_address) +
1387 be64_to_cpu(fdm_ptr->rmr_region.source_len) - 1,
1388 be64_to_cpu(fdm_ptr->rmr_region.source_len),
1389 be64_to_cpu(fdm_ptr->rmr_region.bytes_dumped));
1390
1391 if (!fdm_active ||
1392 (fw_dump.reserve_dump_area_start ==
1393 be64_to_cpu(fdm_ptr->cpu_state_data.destination_address)))
1394 goto out;
1395
1396 /* Dump is active. Show reserved memory region. */
1397 seq_printf(m,
1398 " : [%#016llx-%#016llx] %#llx bytes, "
1399 "Dumped: %#llx\n",
1400 (unsigned long long)fw_dump.reserve_dump_area_start,
1401 be64_to_cpu(fdm_ptr->cpu_state_data.destination_address) - 1,
1402 be64_to_cpu(fdm_ptr->cpu_state_data.destination_address) -
1403 fw_dump.reserve_dump_area_start,
1404 be64_to_cpu(fdm_ptr->cpu_state_data.destination_address) -
1405 fw_dump.reserve_dump_area_start);
1406out:
1407 if (fdm_active)
1408 mutex_unlock(&fadump_mutex);
1409 return 0;
1410}
1411
1412static struct kobj_attribute fadump_release_attr = __ATTR(fadump_release_mem,
1413 0200, NULL,
1414 fadump_release_memory_store);
1415static struct kobj_attribute fadump_attr = __ATTR(fadump_enabled,
1416 0444, fadump_enabled_show,
1417 NULL);
1418static struct kobj_attribute fadump_register_attr = __ATTR(fadump_registered,
1419 0644, fadump_register_show,
1420 fadump_register_store);
1421
1422static int fadump_region_open(struct inode *inode, struct file *file)
1423{
1424 return single_open(file, fadump_region_show, inode->i_private);
1425}
1426
1427static const struct file_operations fadump_region_fops = {
1428 .open = fadump_region_open,
1429 .read = seq_read,
1430 .llseek = seq_lseek,
1431 .release = single_release,
1432};
1433
1434static void fadump_init_files(void)
1435{
1436 struct dentry *debugfs_file;
1437 int rc = 0;
1438
1439 rc = sysfs_create_file(kernel_kobj, &fadump_attr.attr);
1440 if (rc)
1441 printk(KERN_ERR "fadump: unable to create sysfs file"
1442 " fadump_enabled (%d)\n", rc);
1443
1444 rc = sysfs_create_file(kernel_kobj, &fadump_register_attr.attr);
1445 if (rc)
1446 printk(KERN_ERR "fadump: unable to create sysfs file"
1447 " fadump_registered (%d)\n", rc);
1448
1449 debugfs_file = debugfs_create_file("fadump_region", 0444,
1450 powerpc_debugfs_root, NULL,
1451 &fadump_region_fops);
1452 if (!debugfs_file)
1453 printk(KERN_ERR "fadump: unable to create debugfs file"
1454 " fadump_region\n");
1455
1456 if (fw_dump.dump_active) {
1457 rc = sysfs_create_file(kernel_kobj, &fadump_release_attr.attr);
1458 if (rc)
1459 printk(KERN_ERR "fadump: unable to create sysfs file"
1460 " fadump_release_mem (%d)\n", rc);
1461 }
1462 return;
1463}
1464
1465/*
1466 * Prepare for firmware-assisted dump.
1467 */
1468int __init setup_fadump(void)
1469{
1470 if (!fw_dump.fadump_enabled)
1471 return 0;
1472
1473 if (!fw_dump.fadump_supported) {
1474 printk(KERN_ERR "Firmware-assisted dump is not supported on"
1475 " this hardware\n");
1476 return 0;
1477 }
1478
1479 fadump_show_config();
1480 /*
1481 * If dump data is available then see if it is valid and prepare for
1482 * saving it to the disk.
1483 */
1484 if (fw_dump.dump_active) {
1485 /*
1486 * if dump process fails then invalidate the registration
1487 * and release memory before proceeding for re-registration.
1488 */
1489 if (process_fadump(fdm_active) < 0)
1490 fadump_invalidate_release_mem();
1491 }
1492 /* Initialize the kernel dump memory structure for FAD registration. */
1493 else if (fw_dump.reserve_dump_area_size)
1494 init_fadump_mem_struct(&fdm, fw_dump.reserve_dump_area_start);
1495 fadump_init_files();
1496
1497 return 1;
1498}
1499subsys_initcall(setup_fadump);
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
45static struct fw_dump fw_dump;
46
47static void __init fadump_reserve_crash_area(u64 base);
48
49#ifndef CONFIG_PRESERVE_FA_DUMP
50
51static struct kobject *fadump_kobj;
52
53static atomic_t cpus_in_fadump;
54static DEFINE_MUTEX(fadump_mutex);
55
56static 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))
61static struct fadump_memory_range rngs[RESERVED_RNGS_CNT];
62static struct fadump_mrange_info
63reserved_mrange_info = { "reserved", rngs, RESERVED_RNGS_SZ, 0, RESERVED_RNGS_CNT, true };
64
65static void __init early_init_dt_scan_reserved_ranges(unsigned long node);
66
67#ifdef CONFIG_CMA
68static 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 */
83static 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
133static int __init fadump_cma_init(void) { return 1; }
134#endif /* CONFIG_CMA */
135
136/* Scan the Firmware Assisted dump configuration details. */
137int __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 */
165int 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
183int should_fadump_crash(void)
184{
185 if (!fw_dump.dump_registered || !fw_dump.fadumphdr_addr)
186 return 0;
187 return 1;
188}
189
190int 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 */
199static 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, ®_start, ®_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 */
229bool 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 */
251bool 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. */
261static 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 */
302static __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, NULL, NULL);
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 */
363static 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
389static 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 */
410static 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
435static 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, ®_start, ®_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 */
477static 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 */
503static 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
530out:
531 return ret;
532}
533
534int __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;
655error_out:
656 fw_dump.fadump_enabled = 0;
657 fw_dump.reserve_dump_area_size = 0;
658 return 0;
659}
660
661/* Look for fadump= cmdline option. */
662static int __init early_fadump_param(char *p)
663{
664 if (!p)
665 return 1;
666
667 if (strncmp(p, "on", 2) == 0)
668 fw_dump.fadump_enabled = 1;
669 else if (strncmp(p, "off", 3) == 0)
670 fw_dump.fadump_enabled = 0;
671 else if (strncmp(p, "nocma", 5) == 0) {
672 fw_dump.fadump_enabled = 1;
673 fw_dump.nocma = 1;
674 }
675
676 return 0;
677}
678early_param("fadump", early_fadump_param);
679
680/*
681 * Look for fadump_reserve_mem= cmdline option
682 * TODO: Remove references to 'fadump_reserve_mem=' parameter,
683 * the sooner 'crashkernel=' parameter is accustomed to.
684 */
685static int __init early_fadump_reserve_mem(char *p)
686{
687 if (p)
688 fw_dump.reserve_bootvar = memparse(p, &p);
689 return 0;
690}
691early_param("fadump_reserve_mem", early_fadump_reserve_mem);
692
693void crash_fadump(struct pt_regs *regs, const char *str)
694{
695 unsigned int msecs;
696 struct fadump_crash_info_header *fdh = NULL;
697 int old_cpu, this_cpu;
698 /* Do not include first CPU */
699 unsigned int ncpus = num_online_cpus() - 1;
700
701 if (!should_fadump_crash())
702 return;
703
704 /*
705 * old_cpu == -1 means this is the first CPU which has come here,
706 * go ahead and trigger fadump.
707 *
708 * old_cpu != -1 means some other CPU has already on it's way
709 * to trigger fadump, just keep looping here.
710 */
711 this_cpu = smp_processor_id();
712 old_cpu = cmpxchg(&crashing_cpu, -1, this_cpu);
713
714 if (old_cpu != -1) {
715 atomic_inc(&cpus_in_fadump);
716
717 /*
718 * We can't loop here indefinitely. Wait as long as fadump
719 * is in force. If we race with fadump un-registration this
720 * loop will break and then we go down to normal panic path
721 * and reboot. If fadump is in force the first crashing
722 * cpu will definitely trigger fadump.
723 */
724 while (fw_dump.dump_registered)
725 cpu_relax();
726 return;
727 }
728
729 fdh = __va(fw_dump.fadumphdr_addr);
730 fdh->crashing_cpu = crashing_cpu;
731 crash_save_vmcoreinfo();
732
733 if (regs)
734 fdh->regs = *regs;
735 else
736 ppc_save_regs(&fdh->regs);
737
738 fdh->cpu_mask = *cpu_online_mask;
739
740 /*
741 * If we came in via system reset, wait a while for the secondary
742 * CPUs to enter.
743 */
744 if (TRAP(&(fdh->regs)) == INTERRUPT_SYSTEM_RESET) {
745 msecs = CRASH_TIMEOUT;
746 while ((atomic_read(&cpus_in_fadump) < ncpus) && (--msecs > 0))
747 mdelay(1);
748 }
749
750 fw_dump.ops->fadump_trigger(fdh, str);
751}
752
753u32 *__init fadump_regs_to_elf_notes(u32 *buf, struct pt_regs *regs)
754{
755 struct elf_prstatus prstatus;
756
757 memset(&prstatus, 0, sizeof(prstatus));
758 /*
759 * FIXME: How do i get PID? Do I really need it?
760 * prstatus.pr_pid = ????
761 */
762 elf_core_copy_regs(&prstatus.pr_reg, regs);
763 buf = append_elf_note(buf, CRASH_CORE_NOTE_NAME, NT_PRSTATUS,
764 &prstatus, sizeof(prstatus));
765 return buf;
766}
767
768void __init fadump_update_elfcore_header(char *bufp)
769{
770 struct elf_phdr *phdr;
771
772 bufp += sizeof(struct elfhdr);
773
774 /* First note is a place holder for cpu notes info. */
775 phdr = (struct elf_phdr *)bufp;
776
777 if (phdr->p_type == PT_NOTE) {
778 phdr->p_paddr = __pa(fw_dump.cpu_notes_buf_vaddr);
779 phdr->p_offset = phdr->p_paddr;
780 phdr->p_filesz = fw_dump.cpu_notes_buf_size;
781 phdr->p_memsz = fw_dump.cpu_notes_buf_size;
782 }
783 return;
784}
785
786static void *__init fadump_alloc_buffer(unsigned long size)
787{
788 unsigned long count, i;
789 struct page *page;
790 void *vaddr;
791
792 vaddr = alloc_pages_exact(size, GFP_KERNEL | __GFP_ZERO);
793 if (!vaddr)
794 return NULL;
795
796 count = PAGE_ALIGN(size) / PAGE_SIZE;
797 page = virt_to_page(vaddr);
798 for (i = 0; i < count; i++)
799 mark_page_reserved(page + i);
800 return vaddr;
801}
802
803static void fadump_free_buffer(unsigned long vaddr, unsigned long size)
804{
805 free_reserved_area((void *)vaddr, (void *)(vaddr + size), -1, NULL);
806}
807
808s32 __init fadump_setup_cpu_notes_buf(u32 num_cpus)
809{
810 /* Allocate buffer to hold cpu crash notes. */
811 fw_dump.cpu_notes_buf_size = num_cpus * sizeof(note_buf_t);
812 fw_dump.cpu_notes_buf_size = PAGE_ALIGN(fw_dump.cpu_notes_buf_size);
813 fw_dump.cpu_notes_buf_vaddr =
814 (unsigned long)fadump_alloc_buffer(fw_dump.cpu_notes_buf_size);
815 if (!fw_dump.cpu_notes_buf_vaddr) {
816 pr_err("Failed to allocate %ld bytes for CPU notes buffer\n",
817 fw_dump.cpu_notes_buf_size);
818 return -ENOMEM;
819 }
820
821 pr_debug("Allocated buffer for cpu notes of size %ld at 0x%lx\n",
822 fw_dump.cpu_notes_buf_size,
823 fw_dump.cpu_notes_buf_vaddr);
824 return 0;
825}
826
827void fadump_free_cpu_notes_buf(void)
828{
829 if (!fw_dump.cpu_notes_buf_vaddr)
830 return;
831
832 fadump_free_buffer(fw_dump.cpu_notes_buf_vaddr,
833 fw_dump.cpu_notes_buf_size);
834 fw_dump.cpu_notes_buf_vaddr = 0;
835 fw_dump.cpu_notes_buf_size = 0;
836}
837
838static void fadump_free_mem_ranges(struct fadump_mrange_info *mrange_info)
839{
840 if (mrange_info->is_static) {
841 mrange_info->mem_range_cnt = 0;
842 return;
843 }
844
845 kfree(mrange_info->mem_ranges);
846 memset((void *)((u64)mrange_info + RNG_NAME_SZ), 0,
847 (sizeof(struct fadump_mrange_info) - RNG_NAME_SZ));
848}
849
850/*
851 * Allocate or reallocate mem_ranges array in incremental units
852 * of PAGE_SIZE.
853 */
854static int fadump_alloc_mem_ranges(struct fadump_mrange_info *mrange_info)
855{
856 struct fadump_memory_range *new_array;
857 u64 new_size;
858
859 new_size = mrange_info->mem_ranges_sz + PAGE_SIZE;
860 pr_debug("Allocating %llu bytes of memory for %s memory ranges\n",
861 new_size, mrange_info->name);
862
863 new_array = krealloc(mrange_info->mem_ranges, new_size, GFP_KERNEL);
864 if (new_array == NULL) {
865 pr_err("Insufficient memory for setting up %s memory ranges\n",
866 mrange_info->name);
867 fadump_free_mem_ranges(mrange_info);
868 return -ENOMEM;
869 }
870
871 mrange_info->mem_ranges = new_array;
872 mrange_info->mem_ranges_sz = new_size;
873 mrange_info->max_mem_ranges = (new_size /
874 sizeof(struct fadump_memory_range));
875 return 0;
876}
877static inline int fadump_add_mem_range(struct fadump_mrange_info *mrange_info,
878 u64 base, u64 end)
879{
880 struct fadump_memory_range *mem_ranges = mrange_info->mem_ranges;
881 bool is_adjacent = false;
882 u64 start, size;
883
884 if (base == end)
885 return 0;
886
887 /*
888 * Fold adjacent memory ranges to bring down the memory ranges/
889 * PT_LOAD segments count.
890 */
891 if (mrange_info->mem_range_cnt) {
892 start = mem_ranges[mrange_info->mem_range_cnt - 1].base;
893 size = mem_ranges[mrange_info->mem_range_cnt - 1].size;
894
895 /*
896 * Boot memory area needs separate PT_LOAD segment(s) as it
897 * is moved to a different location at the time of crash.
898 * So, fold only if the region is not boot memory area.
899 */
900 if ((start + size) == base && start >= fw_dump.boot_mem_top)
901 is_adjacent = true;
902 }
903 if (!is_adjacent) {
904 /* resize the array on reaching the limit */
905 if (mrange_info->mem_range_cnt == mrange_info->max_mem_ranges) {
906 int ret;
907
908 if (mrange_info->is_static) {
909 pr_err("Reached array size limit for %s memory ranges\n",
910 mrange_info->name);
911 return -ENOSPC;
912 }
913
914 ret = fadump_alloc_mem_ranges(mrange_info);
915 if (ret)
916 return ret;
917
918 /* Update to the new resized array */
919 mem_ranges = mrange_info->mem_ranges;
920 }
921
922 start = base;
923 mem_ranges[mrange_info->mem_range_cnt].base = start;
924 mrange_info->mem_range_cnt++;
925 }
926
927 mem_ranges[mrange_info->mem_range_cnt - 1].size = (end - start);
928 pr_debug("%s_memory_range[%d] [%#016llx-%#016llx], %#llx bytes\n",
929 mrange_info->name, (mrange_info->mem_range_cnt - 1),
930 start, end - 1, (end - start));
931 return 0;
932}
933
934static int fadump_exclude_reserved_area(u64 start, u64 end)
935{
936 u64 ra_start, ra_end;
937 int ret = 0;
938
939 ra_start = fw_dump.reserve_dump_area_start;
940 ra_end = ra_start + fw_dump.reserve_dump_area_size;
941
942 if ((ra_start < end) && (ra_end > start)) {
943 if ((start < ra_start) && (end > ra_end)) {
944 ret = fadump_add_mem_range(&crash_mrange_info,
945 start, ra_start);
946 if (ret)
947 return ret;
948
949 ret = fadump_add_mem_range(&crash_mrange_info,
950 ra_end, end);
951 } else if (start < ra_start) {
952 ret = fadump_add_mem_range(&crash_mrange_info,
953 start, ra_start);
954 } else if (ra_end < end) {
955 ret = fadump_add_mem_range(&crash_mrange_info,
956 ra_end, end);
957 }
958 } else
959 ret = fadump_add_mem_range(&crash_mrange_info, start, end);
960
961 return ret;
962}
963
964static int fadump_init_elfcore_header(char *bufp)
965{
966 struct elfhdr *elf;
967
968 elf = (struct elfhdr *) bufp;
969 bufp += sizeof(struct elfhdr);
970 memcpy(elf->e_ident, ELFMAG, SELFMAG);
971 elf->e_ident[EI_CLASS] = ELF_CLASS;
972 elf->e_ident[EI_DATA] = ELF_DATA;
973 elf->e_ident[EI_VERSION] = EV_CURRENT;
974 elf->e_ident[EI_OSABI] = ELF_OSABI;
975 memset(elf->e_ident+EI_PAD, 0, EI_NIDENT-EI_PAD);
976 elf->e_type = ET_CORE;
977 elf->e_machine = ELF_ARCH;
978 elf->e_version = EV_CURRENT;
979 elf->e_entry = 0;
980 elf->e_phoff = sizeof(struct elfhdr);
981 elf->e_shoff = 0;
982
983 if (IS_ENABLED(CONFIG_PPC64_ELF_ABI_V2))
984 elf->e_flags = 2;
985 else if (IS_ENABLED(CONFIG_PPC64_ELF_ABI_V1))
986 elf->e_flags = 1;
987 else
988 elf->e_flags = 0;
989
990 elf->e_ehsize = sizeof(struct elfhdr);
991 elf->e_phentsize = sizeof(struct elf_phdr);
992 elf->e_phnum = 0;
993 elf->e_shentsize = 0;
994 elf->e_shnum = 0;
995 elf->e_shstrndx = 0;
996
997 return 0;
998}
999
1000/*
1001 * Traverse through memblock structure and setup crash memory ranges. These
1002 * ranges will be used create PT_LOAD program headers in elfcore header.
1003 */
1004static int fadump_setup_crash_memory_ranges(void)
1005{
1006 u64 i, start, end;
1007 int ret;
1008
1009 pr_debug("Setup crash memory ranges.\n");
1010 crash_mrange_info.mem_range_cnt = 0;
1011
1012 /*
1013 * Boot memory region(s) registered with firmware are moved to
1014 * different location at the time of crash. Create separate program
1015 * header(s) for this memory chunk(s) with the correct offset.
1016 */
1017 for (i = 0; i < fw_dump.boot_mem_regs_cnt; i++) {
1018 start = fw_dump.boot_mem_addr[i];
1019 end = start + fw_dump.boot_mem_sz[i];
1020 ret = fadump_add_mem_range(&crash_mrange_info, start, end);
1021 if (ret)
1022 return ret;
1023 }
1024
1025 for_each_mem_range(i, &start, &end) {
1026 /*
1027 * skip the memory chunk that is already added
1028 * (0 through boot_memory_top).
1029 */
1030 if (start < fw_dump.boot_mem_top) {
1031 if (end > fw_dump.boot_mem_top)
1032 start = fw_dump.boot_mem_top;
1033 else
1034 continue;
1035 }
1036
1037 /* add this range excluding the reserved dump area. */
1038 ret = fadump_exclude_reserved_area(start, end);
1039 if (ret)
1040 return ret;
1041 }
1042
1043 return 0;
1044}
1045
1046/*
1047 * If the given physical address falls within the boot memory region then
1048 * return the relocated address that points to the dump region reserved
1049 * for saving initial boot memory contents.
1050 */
1051static inline unsigned long fadump_relocate(unsigned long paddr)
1052{
1053 unsigned long raddr, rstart, rend, rlast, hole_size;
1054 int i;
1055
1056 hole_size = 0;
1057 rlast = 0;
1058 raddr = paddr;
1059 for (i = 0; i < fw_dump.boot_mem_regs_cnt; i++) {
1060 rstart = fw_dump.boot_mem_addr[i];
1061 rend = rstart + fw_dump.boot_mem_sz[i];
1062 hole_size += (rstart - rlast);
1063
1064 if (paddr >= rstart && paddr < rend) {
1065 raddr += fw_dump.boot_mem_dest_addr - hole_size;
1066 break;
1067 }
1068
1069 rlast = rend;
1070 }
1071
1072 pr_debug("vmcoreinfo: paddr = 0x%lx, raddr = 0x%lx\n", paddr, raddr);
1073 return raddr;
1074}
1075
1076static int fadump_create_elfcore_headers(char *bufp)
1077{
1078 unsigned long long raddr, offset;
1079 struct elf_phdr *phdr;
1080 struct elfhdr *elf;
1081 int i, j;
1082
1083 fadump_init_elfcore_header(bufp);
1084 elf = (struct elfhdr *)bufp;
1085 bufp += sizeof(struct elfhdr);
1086
1087 /*
1088 * setup ELF PT_NOTE, place holder for cpu notes info. The notes info
1089 * will be populated during second kernel boot after crash. Hence
1090 * this PT_NOTE will always be the first elf note.
1091 *
1092 * NOTE: Any new ELF note addition should be placed after this note.
1093 */
1094 phdr = (struct elf_phdr *)bufp;
1095 bufp += sizeof(struct elf_phdr);
1096 phdr->p_type = PT_NOTE;
1097 phdr->p_flags = 0;
1098 phdr->p_vaddr = 0;
1099 phdr->p_align = 0;
1100
1101 phdr->p_offset = 0;
1102 phdr->p_paddr = 0;
1103 phdr->p_filesz = 0;
1104 phdr->p_memsz = 0;
1105
1106 (elf->e_phnum)++;
1107
1108 /* setup ELF PT_NOTE for vmcoreinfo */
1109 phdr = (struct elf_phdr *)bufp;
1110 bufp += sizeof(struct elf_phdr);
1111 phdr->p_type = PT_NOTE;
1112 phdr->p_flags = 0;
1113 phdr->p_vaddr = 0;
1114 phdr->p_align = 0;
1115
1116 phdr->p_paddr = fadump_relocate(paddr_vmcoreinfo_note());
1117 phdr->p_offset = phdr->p_paddr;
1118 phdr->p_memsz = phdr->p_filesz = VMCOREINFO_NOTE_SIZE;
1119
1120 /* Increment number of program headers. */
1121 (elf->e_phnum)++;
1122
1123 /* setup PT_LOAD sections. */
1124 j = 0;
1125 offset = 0;
1126 raddr = fw_dump.boot_mem_addr[0];
1127 for (i = 0; i < crash_mrange_info.mem_range_cnt; i++) {
1128 u64 mbase, msize;
1129
1130 mbase = crash_mrange_info.mem_ranges[i].base;
1131 msize = crash_mrange_info.mem_ranges[i].size;
1132 if (!msize)
1133 continue;
1134
1135 phdr = (struct elf_phdr *)bufp;
1136 bufp += sizeof(struct elf_phdr);
1137 phdr->p_type = PT_LOAD;
1138 phdr->p_flags = PF_R|PF_W|PF_X;
1139 phdr->p_offset = mbase;
1140
1141 if (mbase == raddr) {
1142 /*
1143 * The entire real memory region will be moved by
1144 * firmware to the specified destination_address.
1145 * Hence set the correct offset.
1146 */
1147 phdr->p_offset = fw_dump.boot_mem_dest_addr + offset;
1148 if (j < (fw_dump.boot_mem_regs_cnt - 1)) {
1149 offset += fw_dump.boot_mem_sz[j];
1150 raddr = fw_dump.boot_mem_addr[++j];
1151 }
1152 }
1153
1154 phdr->p_paddr = mbase;
1155 phdr->p_vaddr = (unsigned long)__va(mbase);
1156 phdr->p_filesz = msize;
1157 phdr->p_memsz = msize;
1158 phdr->p_align = 0;
1159
1160 /* Increment number of program headers. */
1161 (elf->e_phnum)++;
1162 }
1163 return 0;
1164}
1165
1166static unsigned long init_fadump_header(unsigned long addr)
1167{
1168 struct fadump_crash_info_header *fdh;
1169
1170 if (!addr)
1171 return 0;
1172
1173 fdh = __va(addr);
1174 addr += sizeof(struct fadump_crash_info_header);
1175
1176 memset(fdh, 0, sizeof(struct fadump_crash_info_header));
1177 fdh->magic_number = FADUMP_CRASH_INFO_MAGIC;
1178 fdh->elfcorehdr_addr = addr;
1179 /* We will set the crashing cpu id in crash_fadump() during crash. */
1180 fdh->crashing_cpu = FADUMP_CPU_UNKNOWN;
1181 /*
1182 * When LPAR is terminated by PYHP, ensure all possible CPUs'
1183 * register data is processed while exporting the vmcore.
1184 */
1185 fdh->cpu_mask = *cpu_possible_mask;
1186
1187 return addr;
1188}
1189
1190static int register_fadump(void)
1191{
1192 unsigned long addr;
1193 void *vaddr;
1194 int ret;
1195
1196 /*
1197 * If no memory is reserved then we can not register for firmware-
1198 * assisted dump.
1199 */
1200 if (!fw_dump.reserve_dump_area_size)
1201 return -ENODEV;
1202
1203 ret = fadump_setup_crash_memory_ranges();
1204 if (ret)
1205 return ret;
1206
1207 addr = fw_dump.fadumphdr_addr;
1208
1209 /* Initialize fadump crash info header. */
1210 addr = init_fadump_header(addr);
1211 vaddr = __va(addr);
1212
1213 pr_debug("Creating ELF core headers at %#016lx\n", addr);
1214 fadump_create_elfcore_headers(vaddr);
1215
1216 /* register the future kernel dump with firmware. */
1217 pr_debug("Registering for firmware-assisted kernel dump...\n");
1218 return fw_dump.ops->fadump_register(&fw_dump);
1219}
1220
1221void fadump_cleanup(void)
1222{
1223 if (!fw_dump.fadump_supported)
1224 return;
1225
1226 /* Invalidate the registration only if dump is active. */
1227 if (fw_dump.dump_active) {
1228 pr_debug("Invalidating firmware-assisted dump registration\n");
1229 fw_dump.ops->fadump_invalidate(&fw_dump);
1230 } else if (fw_dump.dump_registered) {
1231 /* Un-register Firmware-assisted dump if it was registered. */
1232 fw_dump.ops->fadump_unregister(&fw_dump);
1233 fadump_free_mem_ranges(&crash_mrange_info);
1234 }
1235
1236 if (fw_dump.ops->fadump_cleanup)
1237 fw_dump.ops->fadump_cleanup(&fw_dump);
1238}
1239
1240static void fadump_free_reserved_memory(unsigned long start_pfn,
1241 unsigned long end_pfn)
1242{
1243 unsigned long pfn;
1244 unsigned long time_limit = jiffies + HZ;
1245
1246 pr_info("freeing reserved memory (0x%llx - 0x%llx)\n",
1247 PFN_PHYS(start_pfn), PFN_PHYS(end_pfn));
1248
1249 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
1250 free_reserved_page(pfn_to_page(pfn));
1251
1252 if (time_after(jiffies, time_limit)) {
1253 cond_resched();
1254 time_limit = jiffies + HZ;
1255 }
1256 }
1257}
1258
1259/*
1260 * Skip memory holes and free memory that was actually reserved.
1261 */
1262static void fadump_release_reserved_area(u64 start, u64 end)
1263{
1264 unsigned long reg_spfn, reg_epfn;
1265 u64 tstart, tend, spfn, epfn;
1266 int i;
1267
1268 spfn = PHYS_PFN(start);
1269 epfn = PHYS_PFN(end);
1270
1271 for_each_mem_pfn_range(i, MAX_NUMNODES, ®_spfn, ®_epfn, NULL) {
1272 tstart = max_t(u64, spfn, reg_spfn);
1273 tend = min_t(u64, epfn, reg_epfn);
1274
1275 if (tstart < tend) {
1276 fadump_free_reserved_memory(tstart, tend);
1277
1278 if (tend == epfn)
1279 break;
1280
1281 spfn = tend;
1282 }
1283 }
1284}
1285
1286/*
1287 * Sort the mem ranges in-place and merge adjacent ranges
1288 * to minimize the memory ranges count.
1289 */
1290static void sort_and_merge_mem_ranges(struct fadump_mrange_info *mrange_info)
1291{
1292 struct fadump_memory_range *mem_ranges;
1293 u64 base, size;
1294 int i, j, idx;
1295
1296 if (!reserved_mrange_info.mem_range_cnt)
1297 return;
1298
1299 /* Sort the memory ranges */
1300 mem_ranges = mrange_info->mem_ranges;
1301 for (i = 0; i < mrange_info->mem_range_cnt; i++) {
1302 idx = i;
1303 for (j = (i + 1); j < mrange_info->mem_range_cnt; j++) {
1304 if (mem_ranges[idx].base > mem_ranges[j].base)
1305 idx = j;
1306 }
1307 if (idx != i)
1308 swap(mem_ranges[idx], mem_ranges[i]);
1309 }
1310
1311 /* Merge adjacent reserved ranges */
1312 idx = 0;
1313 for (i = 1; i < mrange_info->mem_range_cnt; i++) {
1314 base = mem_ranges[i-1].base;
1315 size = mem_ranges[i-1].size;
1316 if (mem_ranges[i].base == (base + size))
1317 mem_ranges[idx].size += mem_ranges[i].size;
1318 else {
1319 idx++;
1320 if (i == idx)
1321 continue;
1322
1323 mem_ranges[idx] = mem_ranges[i];
1324 }
1325 }
1326 mrange_info->mem_range_cnt = idx + 1;
1327}
1328
1329/*
1330 * Scan reserved-ranges to consider them while reserving/releasing
1331 * memory for FADump.
1332 */
1333static void __init early_init_dt_scan_reserved_ranges(unsigned long node)
1334{
1335 const __be32 *prop;
1336 int len, ret = -1;
1337 unsigned long i;
1338
1339 /* reserved-ranges already scanned */
1340 if (reserved_mrange_info.mem_range_cnt != 0)
1341 return;
1342
1343 prop = of_get_flat_dt_prop(node, "reserved-ranges", &len);
1344 if (!prop)
1345 return;
1346
1347 /*
1348 * Each reserved range is an (address,size) pair, 2 cells each,
1349 * totalling 4 cells per range.
1350 */
1351 for (i = 0; i < len / (sizeof(*prop) * 4); i++) {
1352 u64 base, size;
1353
1354 base = of_read_number(prop + (i * 4) + 0, 2);
1355 size = of_read_number(prop + (i * 4) + 2, 2);
1356
1357 if (size) {
1358 ret = fadump_add_mem_range(&reserved_mrange_info,
1359 base, base + size);
1360 if (ret < 0) {
1361 pr_warn("some reserved ranges are ignored!\n");
1362 break;
1363 }
1364 }
1365 }
1366
1367 /* Compact reserved ranges */
1368 sort_and_merge_mem_ranges(&reserved_mrange_info);
1369}
1370
1371/*
1372 * Release the memory that was reserved during early boot to preserve the
1373 * crash'ed kernel's memory contents except reserved dump area (permanent
1374 * reservation) and reserved ranges used by F/W. The released memory will
1375 * be available for general use.
1376 */
1377static void fadump_release_memory(u64 begin, u64 end)
1378{
1379 u64 ra_start, ra_end, tstart;
1380 int i, ret;
1381
1382 ra_start = fw_dump.reserve_dump_area_start;
1383 ra_end = ra_start + fw_dump.reserve_dump_area_size;
1384
1385 /*
1386 * If reserved ranges array limit is hit, overwrite the last reserved
1387 * memory range with reserved dump area to ensure it is excluded from
1388 * the memory being released (reused for next FADump registration).
1389 */
1390 if (reserved_mrange_info.mem_range_cnt ==
1391 reserved_mrange_info.max_mem_ranges)
1392 reserved_mrange_info.mem_range_cnt--;
1393
1394 ret = fadump_add_mem_range(&reserved_mrange_info, ra_start, ra_end);
1395 if (ret != 0)
1396 return;
1397
1398 /* Get the reserved ranges list in order first. */
1399 sort_and_merge_mem_ranges(&reserved_mrange_info);
1400
1401 /* Exclude reserved ranges and release remaining memory */
1402 tstart = begin;
1403 for (i = 0; i < reserved_mrange_info.mem_range_cnt; i++) {
1404 ra_start = reserved_mrange_info.mem_ranges[i].base;
1405 ra_end = ra_start + reserved_mrange_info.mem_ranges[i].size;
1406
1407 if (tstart >= ra_end)
1408 continue;
1409
1410 if (tstart < ra_start)
1411 fadump_release_reserved_area(tstart, ra_start);
1412 tstart = ra_end;
1413 }
1414
1415 if (tstart < end)
1416 fadump_release_reserved_area(tstart, end);
1417}
1418
1419static void fadump_invalidate_release_mem(void)
1420{
1421 mutex_lock(&fadump_mutex);
1422 if (!fw_dump.dump_active) {
1423 mutex_unlock(&fadump_mutex);
1424 return;
1425 }
1426
1427 fadump_cleanup();
1428 mutex_unlock(&fadump_mutex);
1429
1430 fadump_release_memory(fw_dump.boot_mem_top, memblock_end_of_DRAM());
1431 fadump_free_cpu_notes_buf();
1432
1433 /*
1434 * Setup kernel metadata and initialize the kernel dump
1435 * memory structure for FADump re-registration.
1436 */
1437 if (fw_dump.ops->fadump_setup_metadata &&
1438 (fw_dump.ops->fadump_setup_metadata(&fw_dump) < 0))
1439 pr_warn("Failed to setup kernel metadata!\n");
1440 fw_dump.ops->fadump_init_mem_struct(&fw_dump);
1441}
1442
1443static ssize_t release_mem_store(struct kobject *kobj,
1444 struct kobj_attribute *attr,
1445 const char *buf, size_t count)
1446{
1447 int input = -1;
1448
1449 if (!fw_dump.dump_active)
1450 return -EPERM;
1451
1452 if (kstrtoint(buf, 0, &input))
1453 return -EINVAL;
1454
1455 if (input == 1) {
1456 /*
1457 * Take away the '/proc/vmcore'. We are releasing the dump
1458 * memory, hence it will not be valid anymore.
1459 */
1460#ifdef CONFIG_PROC_VMCORE
1461 vmcore_cleanup();
1462#endif
1463 fadump_invalidate_release_mem();
1464
1465 } else
1466 return -EINVAL;
1467 return count;
1468}
1469
1470/* Release the reserved memory and disable the FADump */
1471static void __init unregister_fadump(void)
1472{
1473 fadump_cleanup();
1474 fadump_release_memory(fw_dump.reserve_dump_area_start,
1475 fw_dump.reserve_dump_area_size);
1476 fw_dump.fadump_enabled = 0;
1477 kobject_put(fadump_kobj);
1478}
1479
1480static ssize_t enabled_show(struct kobject *kobj,
1481 struct kobj_attribute *attr,
1482 char *buf)
1483{
1484 return sprintf(buf, "%d\n", fw_dump.fadump_enabled);
1485}
1486
1487static ssize_t mem_reserved_show(struct kobject *kobj,
1488 struct kobj_attribute *attr,
1489 char *buf)
1490{
1491 return sprintf(buf, "%ld\n", fw_dump.reserve_dump_area_size);
1492}
1493
1494static ssize_t registered_show(struct kobject *kobj,
1495 struct kobj_attribute *attr,
1496 char *buf)
1497{
1498 return sprintf(buf, "%d\n", fw_dump.dump_registered);
1499}
1500
1501static ssize_t registered_store(struct kobject *kobj,
1502 struct kobj_attribute *attr,
1503 const char *buf, size_t count)
1504{
1505 int ret = 0;
1506 int input = -1;
1507
1508 if (!fw_dump.fadump_enabled || fw_dump.dump_active)
1509 return -EPERM;
1510
1511 if (kstrtoint(buf, 0, &input))
1512 return -EINVAL;
1513
1514 mutex_lock(&fadump_mutex);
1515
1516 switch (input) {
1517 case 0:
1518 if (fw_dump.dump_registered == 0) {
1519 goto unlock_out;
1520 }
1521
1522 /* Un-register Firmware-assisted dump */
1523 pr_debug("Un-register firmware-assisted dump\n");
1524 fw_dump.ops->fadump_unregister(&fw_dump);
1525 break;
1526 case 1:
1527 if (fw_dump.dump_registered == 1) {
1528 /* Un-register Firmware-assisted dump */
1529 fw_dump.ops->fadump_unregister(&fw_dump);
1530 }
1531 /* Register Firmware-assisted dump */
1532 ret = register_fadump();
1533 break;
1534 default:
1535 ret = -EINVAL;
1536 break;
1537 }
1538
1539unlock_out:
1540 mutex_unlock(&fadump_mutex);
1541 return ret < 0 ? ret : count;
1542}
1543
1544static int fadump_region_show(struct seq_file *m, void *private)
1545{
1546 if (!fw_dump.fadump_enabled)
1547 return 0;
1548
1549 mutex_lock(&fadump_mutex);
1550 fw_dump.ops->fadump_region_show(&fw_dump, m);
1551 mutex_unlock(&fadump_mutex);
1552 return 0;
1553}
1554
1555static struct kobj_attribute release_attr = __ATTR_WO(release_mem);
1556static struct kobj_attribute enable_attr = __ATTR_RO(enabled);
1557static struct kobj_attribute register_attr = __ATTR_RW(registered);
1558static struct kobj_attribute mem_reserved_attr = __ATTR_RO(mem_reserved);
1559
1560static struct attribute *fadump_attrs[] = {
1561 &enable_attr.attr,
1562 ®ister_attr.attr,
1563 &mem_reserved_attr.attr,
1564 NULL,
1565};
1566
1567ATTRIBUTE_GROUPS(fadump);
1568
1569DEFINE_SHOW_ATTRIBUTE(fadump_region);
1570
1571static void __init fadump_init_files(void)
1572{
1573 int rc = 0;
1574
1575 fadump_kobj = kobject_create_and_add("fadump", kernel_kobj);
1576 if (!fadump_kobj) {
1577 pr_err("failed to create fadump kobject\n");
1578 return;
1579 }
1580
1581 debugfs_create_file("fadump_region", 0444, arch_debugfs_dir, NULL,
1582 &fadump_region_fops);
1583
1584 if (fw_dump.dump_active) {
1585 rc = sysfs_create_file(fadump_kobj, &release_attr.attr);
1586 if (rc)
1587 pr_err("unable to create release_mem sysfs file (%d)\n",
1588 rc);
1589 }
1590
1591 rc = sysfs_create_groups(fadump_kobj, fadump_groups);
1592 if (rc) {
1593 pr_err("sysfs group creation failed (%d), unregistering FADump",
1594 rc);
1595 unregister_fadump();
1596 return;
1597 }
1598
1599 /*
1600 * The FADump sysfs are moved from kernel_kobj to fadump_kobj need to
1601 * create symlink at old location to maintain backward compatibility.
1602 *
1603 * - fadump_enabled -> fadump/enabled
1604 * - fadump_registered -> fadump/registered
1605 * - fadump_release_mem -> fadump/release_mem
1606 */
1607 rc = compat_only_sysfs_link_entry_to_kobj(kernel_kobj, fadump_kobj,
1608 "enabled", "fadump_enabled");
1609 if (rc) {
1610 pr_err("unable to create fadump_enabled symlink (%d)", rc);
1611 return;
1612 }
1613
1614 rc = compat_only_sysfs_link_entry_to_kobj(kernel_kobj, fadump_kobj,
1615 "registered",
1616 "fadump_registered");
1617 if (rc) {
1618 pr_err("unable to create fadump_registered symlink (%d)", rc);
1619 sysfs_remove_link(kernel_kobj, "fadump_enabled");
1620 return;
1621 }
1622
1623 if (fw_dump.dump_active) {
1624 rc = compat_only_sysfs_link_entry_to_kobj(kernel_kobj,
1625 fadump_kobj,
1626 "release_mem",
1627 "fadump_release_mem");
1628 if (rc)
1629 pr_err("unable to create fadump_release_mem symlink (%d)",
1630 rc);
1631 }
1632 return;
1633}
1634
1635/*
1636 * Prepare for firmware-assisted dump.
1637 */
1638int __init setup_fadump(void)
1639{
1640 if (!fw_dump.fadump_supported)
1641 return 0;
1642
1643 fadump_init_files();
1644 fadump_show_config();
1645
1646 if (!fw_dump.fadump_enabled)
1647 return 1;
1648
1649 /*
1650 * If dump data is available then see if it is valid and prepare for
1651 * saving it to the disk.
1652 */
1653 if (fw_dump.dump_active) {
1654 /*
1655 * if dump process fails then invalidate the registration
1656 * and release memory before proceeding for re-registration.
1657 */
1658 if (fw_dump.ops->fadump_process(&fw_dump) < 0)
1659 fadump_invalidate_release_mem();
1660 }
1661 /* Initialize the kernel dump memory structure and register with f/w */
1662 else if (fw_dump.reserve_dump_area_size) {
1663 fw_dump.ops->fadump_init_mem_struct(&fw_dump);
1664 register_fadump();
1665 }
1666
1667 /*
1668 * In case of panic, fadump is triggered via ppc_panic_event()
1669 * panic notifier. Setting crash_kexec_post_notifiers to 'true'
1670 * lets panic() function take crash friendly path before panic
1671 * notifiers are invoked.
1672 */
1673 crash_kexec_post_notifiers = true;
1674
1675 return 1;
1676}
1677/*
1678 * Use subsys_initcall_sync() here because there is dependency with
1679 * crash_save_vmcoreinfo_init(), which must run first to ensure vmcoreinfo initialization
1680 * is done before registering with f/w.
1681 */
1682subsys_initcall_sync(setup_fadump);
1683#else /* !CONFIG_PRESERVE_FA_DUMP */
1684
1685/* Scan the Firmware Assisted dump configuration details. */
1686int __init early_init_dt_scan_fw_dump(unsigned long node, const char *uname,
1687 int depth, void *data)
1688{
1689 if ((depth != 1) || (strcmp(uname, "ibm,opal") != 0))
1690 return 0;
1691
1692 opal_fadump_dt_scan(&fw_dump, node);
1693 return 1;
1694}
1695
1696/*
1697 * When dump is active but PRESERVE_FA_DUMP is enabled on the kernel,
1698 * preserve crash data. The subsequent memory preserving kernel boot
1699 * is likely to process this crash data.
1700 */
1701int __init fadump_reserve_mem(void)
1702{
1703 if (fw_dump.dump_active) {
1704 /*
1705 * If last boot has crashed then reserve all the memory
1706 * above boot memory to preserve crash data.
1707 */
1708 pr_info("Preserving crash data for processing in next boot.\n");
1709 fadump_reserve_crash_area(fw_dump.boot_mem_top);
1710 } else
1711 pr_debug("FADump-aware kernel..\n");
1712
1713 return 1;
1714}
1715#endif /* CONFIG_PRESERVE_FA_DUMP */
1716
1717/* Preserve everything above the base address */
1718static void __init fadump_reserve_crash_area(u64 base)
1719{
1720 u64 i, mstart, mend, msize;
1721
1722 for_each_mem_range(i, &mstart, &mend) {
1723 msize = mend - mstart;
1724
1725 if ((mstart + msize) < base)
1726 continue;
1727
1728 if (mstart < base) {
1729 msize -= (base - mstart);
1730 mstart = base;
1731 }
1732
1733 pr_info("Reserving %lluMB of memory at %#016llx for preserving crash data",
1734 (msize >> 20), mstart);
1735 memblock_reserve(mstart, msize);
1736 }
1737}
1738
1739unsigned long __init arch_reserved_kernel_pages(void)
1740{
1741 return memblock_reserved_size() / PAGE_SIZE;
1742}