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
28#include <asm/debugfs.h>
29#include <asm/page.h>
30#include <asm/prom.h>
31#include <asm/fadump.h>
32#include <asm/fadump-internal.h>
33#include <asm/setup.h>
34
35static struct fw_dump fw_dump;
36
37static void __init fadump_reserve_crash_area(u64 base);
38
39#ifndef CONFIG_PRESERVE_FA_DUMP
40static DEFINE_MUTEX(fadump_mutex);
41struct fadump_mrange_info crash_mrange_info = { "crash", NULL, 0, 0, 0 };
42struct fadump_mrange_info reserved_mrange_info = { "reserved", NULL, 0, 0, 0 };
43
44#ifdef CONFIG_CMA
45static struct cma *fadump_cma;
46
47/*
48 * fadump_cma_init() - Initialize CMA area from a fadump reserved memory
49 *
50 * This function initializes CMA area from fadump reserved memory.
51 * The total size of fadump reserved memory covers for boot memory size
52 * + cpu data size + hpte size and metadata.
53 * Initialize only the area equivalent to boot memory size for CMA use.
54 * The reamining portion of fadump reserved memory will be not given
55 * to CMA and pages for thoes will stay reserved. boot memory size is
56 * aligned per CMA requirement to satisy cma_init_reserved_mem() call.
57 * But for some reason even if it fails we still have the memory reservation
58 * with us and we can still continue doing fadump.
59 */
60int __init fadump_cma_init(void)
61{
62 unsigned long long base, size;
63 int rc;
64
65 if (!fw_dump.fadump_enabled)
66 return 0;
67
68 /*
69 * Do not use CMA if user has provided fadump=nocma kernel parameter.
70 * Return 1 to continue with fadump old behaviour.
71 */
72 if (fw_dump.nocma)
73 return 1;
74
75 base = fw_dump.reserve_dump_area_start;
76 size = fw_dump.boot_memory_size;
77
78 if (!size)
79 return 0;
80
81 rc = cma_init_reserved_mem(base, size, 0, "fadump_cma", &fadump_cma);
82 if (rc) {
83 pr_err("Failed to init cma area for firmware-assisted dump,%d\n", rc);
84 /*
85 * Though the CMA init has failed we still have memory
86 * reservation with us. The reserved memory will be
87 * blocked from production system usage. Hence return 1,
88 * so that we can continue with fadump.
89 */
90 return 1;
91 }
92
93 /*
94 * So we now have successfully initialized cma area for fadump.
95 */
96 pr_info("Initialized 0x%lx bytes cma area at %ldMB from 0x%lx "
97 "bytes of memory reserved for firmware-assisted dump\n",
98 cma_get_size(fadump_cma),
99 (unsigned long)cma_get_base(fadump_cma) >> 20,
100 fw_dump.reserve_dump_area_size);
101 return 1;
102}
103#else
104static int __init fadump_cma_init(void) { return 1; }
105#endif /* CONFIG_CMA */
106
107/* Scan the Firmware Assisted dump configuration details. */
108int __init early_init_dt_scan_fw_dump(unsigned long node, const char *uname,
109 int depth, void *data)
110{
111 if (depth != 1)
112 return 0;
113
114 if (strcmp(uname, "rtas") == 0) {
115 rtas_fadump_dt_scan(&fw_dump, node);
116 return 1;
117 }
118
119 if (strcmp(uname, "ibm,opal") == 0) {
120 opal_fadump_dt_scan(&fw_dump, node);
121 return 1;
122 }
123
124 return 0;
125}
126
127/*
128 * If fadump is registered, check if the memory provided
129 * falls within boot memory area and reserved memory area.
130 */
131int is_fadump_memory_area(u64 addr, unsigned long size)
132{
133 u64 d_start, d_end;
134
135 if (!fw_dump.dump_registered)
136 return 0;
137
138 if (!size)
139 return 0;
140
141 d_start = fw_dump.reserve_dump_area_start;
142 d_end = d_start + fw_dump.reserve_dump_area_size;
143 if (((addr + size) > d_start) && (addr <= d_end))
144 return 1;
145
146 return (addr <= fw_dump.boot_mem_top);
147}
148
149int should_fadump_crash(void)
150{
151 if (!fw_dump.dump_registered || !fw_dump.fadumphdr_addr)
152 return 0;
153 return 1;
154}
155
156int is_fadump_active(void)
157{
158 return fw_dump.dump_active;
159}
160
161/*
162 * Returns true, if there are no holes in memory area between d_start to d_end,
163 * false otherwise.
164 */
165static bool is_fadump_mem_area_contiguous(u64 d_start, u64 d_end)
166{
167 struct memblock_region *reg;
168 bool ret = false;
169 u64 start, end;
170
171 for_each_memblock(memory, reg) {
172 start = max_t(u64, d_start, reg->base);
173 end = min_t(u64, d_end, (reg->base + reg->size));
174 if (d_start < end) {
175 /* Memory hole from d_start to start */
176 if (start > d_start)
177 break;
178
179 if (end == d_end) {
180 ret = true;
181 break;
182 }
183
184 d_start = end + 1;
185 }
186 }
187
188 return ret;
189}
190
191/*
192 * Returns true, if there are no holes in boot memory area,
193 * false otherwise.
194 */
195bool is_fadump_boot_mem_contiguous(void)
196{
197 unsigned long d_start, d_end;
198 bool ret = false;
199 int i;
200
201 for (i = 0; i < fw_dump.boot_mem_regs_cnt; i++) {
202 d_start = fw_dump.boot_mem_addr[i];
203 d_end = d_start + fw_dump.boot_mem_sz[i];
204
205 ret = is_fadump_mem_area_contiguous(d_start, d_end);
206 if (!ret)
207 break;
208 }
209
210 return ret;
211}
212
213/*
214 * Returns true, if there are no holes in reserved memory area,
215 * false otherwise.
216 */
217bool is_fadump_reserved_mem_contiguous(void)
218{
219 u64 d_start, d_end;
220
221 d_start = fw_dump.reserve_dump_area_start;
222 d_end = d_start + fw_dump.reserve_dump_area_size;
223 return is_fadump_mem_area_contiguous(d_start, d_end);
224}
225
226/* Print firmware assisted dump configurations for debugging purpose. */
227static void fadump_show_config(void)
228{
229 int i;
230
231 pr_debug("Support for firmware-assisted dump (fadump): %s\n",
232 (fw_dump.fadump_supported ? "present" : "no support"));
233
234 if (!fw_dump.fadump_supported)
235 return;
236
237 pr_debug("Fadump enabled : %s\n",
238 (fw_dump.fadump_enabled ? "yes" : "no"));
239 pr_debug("Dump Active : %s\n",
240 (fw_dump.dump_active ? "yes" : "no"));
241 pr_debug("Dump section sizes:\n");
242 pr_debug(" CPU state data size: %lx\n", fw_dump.cpu_state_data_size);
243 pr_debug(" HPTE region size : %lx\n", fw_dump.hpte_region_size);
244 pr_debug(" Boot memory size : %lx\n", fw_dump.boot_memory_size);
245 pr_debug(" Boot memory top : %llx\n", fw_dump.boot_mem_top);
246 pr_debug("Boot memory regions cnt: %llx\n", fw_dump.boot_mem_regs_cnt);
247 for (i = 0; i < fw_dump.boot_mem_regs_cnt; i++) {
248 pr_debug("[%03d] base = %llx, size = %llx\n", i,
249 fw_dump.boot_mem_addr[i], fw_dump.boot_mem_sz[i]);
250 }
251}
252
253/**
254 * fadump_calculate_reserve_size(): reserve variable boot area 5% of System RAM
255 *
256 * Function to find the largest memory size we need to reserve during early
257 * boot process. This will be the size of the memory that is required for a
258 * kernel to boot successfully.
259 *
260 * This function has been taken from phyp-assisted dump feature implementation.
261 *
262 * returns larger of 256MB or 5% rounded down to multiples of 256MB.
263 *
264 * TODO: Come up with better approach to find out more accurate memory size
265 * that is required for a kernel to boot successfully.
266 *
267 */
268static inline u64 fadump_calculate_reserve_size(void)
269{
270 u64 base, size, bootmem_min;
271 int ret;
272
273 if (fw_dump.reserve_bootvar)
274 pr_warn("'fadump_reserve_mem=' parameter is deprecated in favor of 'crashkernel=' parameter.\n");
275
276 /*
277 * Check if the size is specified through crashkernel= cmdline
278 * option. If yes, then use that but ignore base as fadump reserves
279 * memory at a predefined offset.
280 */
281 ret = parse_crashkernel(boot_command_line, memblock_phys_mem_size(),
282 &size, &base);
283 if (ret == 0 && size > 0) {
284 unsigned long max_size;
285
286 if (fw_dump.reserve_bootvar)
287 pr_info("Using 'crashkernel=' parameter for memory reservation.\n");
288
289 fw_dump.reserve_bootvar = (unsigned long)size;
290
291 /*
292 * Adjust if the boot memory size specified is above
293 * the upper limit.
294 */
295 max_size = memblock_phys_mem_size() / MAX_BOOT_MEM_RATIO;
296 if (fw_dump.reserve_bootvar > max_size) {
297 fw_dump.reserve_bootvar = max_size;
298 pr_info("Adjusted boot memory size to %luMB\n",
299 (fw_dump.reserve_bootvar >> 20));
300 }
301
302 return fw_dump.reserve_bootvar;
303 } else if (fw_dump.reserve_bootvar) {
304 /*
305 * 'fadump_reserve_mem=' is being used to reserve memory
306 * for firmware-assisted dump.
307 */
308 return fw_dump.reserve_bootvar;
309 }
310
311 /* divide by 20 to get 5% of value */
312 size = memblock_phys_mem_size() / 20;
313
314 /* round it down in multiples of 256 */
315 size = size & ~0x0FFFFFFFUL;
316
317 /* Truncate to memory_limit. We don't want to over reserve the memory.*/
318 if (memory_limit && size > memory_limit)
319 size = memory_limit;
320
321 bootmem_min = fw_dump.ops->fadump_get_bootmem_min();
322 return (size > bootmem_min ? size : bootmem_min);
323}
324
325/*
326 * Calculate the total memory size required to be reserved for
327 * firmware-assisted dump registration.
328 */
329static unsigned long get_fadump_area_size(void)
330{
331 unsigned long size = 0;
332
333 size += fw_dump.cpu_state_data_size;
334 size += fw_dump.hpte_region_size;
335 size += fw_dump.boot_memory_size;
336 size += sizeof(struct fadump_crash_info_header);
337 size += sizeof(struct elfhdr); /* ELF core header.*/
338 size += sizeof(struct elf_phdr); /* place holder for cpu notes */
339 /* Program headers for crash memory regions. */
340 size += sizeof(struct elf_phdr) * (memblock_num_regions(memory) + 2);
341
342 size = PAGE_ALIGN(size);
343
344 /* This is to hold kernel metadata on platforms that support it */
345 size += (fw_dump.ops->fadump_get_metadata_size ?
346 fw_dump.ops->fadump_get_metadata_size() : 0);
347 return size;
348}
349
350static int __init add_boot_mem_region(unsigned long rstart,
351 unsigned long rsize)
352{
353 int i = fw_dump.boot_mem_regs_cnt++;
354
355 if (fw_dump.boot_mem_regs_cnt > FADUMP_MAX_MEM_REGS) {
356 fw_dump.boot_mem_regs_cnt = FADUMP_MAX_MEM_REGS;
357 return 0;
358 }
359
360 pr_debug("Added boot memory range[%d] [%#016lx-%#016lx)\n",
361 i, rstart, (rstart + rsize));
362 fw_dump.boot_mem_addr[i] = rstart;
363 fw_dump.boot_mem_sz[i] = rsize;
364 return 1;
365}
366
367/*
368 * Firmware usually has a hard limit on the data it can copy per region.
369 * Honour that by splitting a memory range into multiple regions.
370 */
371static int __init add_boot_mem_regions(unsigned long mstart,
372 unsigned long msize)
373{
374 unsigned long rstart, rsize, max_size;
375 int ret = 1;
376
377 rstart = mstart;
378 max_size = fw_dump.max_copy_size ? fw_dump.max_copy_size : msize;
379 while (msize) {
380 if (msize > max_size)
381 rsize = max_size;
382 else
383 rsize = msize;
384
385 ret = add_boot_mem_region(rstart, rsize);
386 if (!ret)
387 break;
388
389 msize -= rsize;
390 rstart += rsize;
391 }
392
393 return ret;
394}
395
396static int __init fadump_get_boot_mem_regions(void)
397{
398 unsigned long base, size, cur_size, hole_size, last_end;
399 unsigned long mem_size = fw_dump.boot_memory_size;
400 struct memblock_region *reg;
401 int ret = 1;
402
403 fw_dump.boot_mem_regs_cnt = 0;
404
405 last_end = 0;
406 hole_size = 0;
407 cur_size = 0;
408 for_each_memblock(memory, reg) {
409 base = reg->base;
410 size = reg->size;
411 hole_size += (base - last_end);
412
413 if ((cur_size + size) >= mem_size) {
414 size = (mem_size - cur_size);
415 ret = add_boot_mem_regions(base, size);
416 break;
417 }
418
419 mem_size -= size;
420 cur_size += size;
421 ret = add_boot_mem_regions(base, size);
422 if (!ret)
423 break;
424
425 last_end = base + size;
426 }
427 fw_dump.boot_mem_top = PAGE_ALIGN(fw_dump.boot_memory_size + hole_size);
428
429 return ret;
430}
431
432int __init fadump_reserve_mem(void)
433{
434 u64 base, size, mem_boundary, bootmem_min, align = PAGE_SIZE;
435 bool is_memblock_bottom_up = memblock_bottom_up();
436 int ret = 1;
437
438 if (!fw_dump.fadump_enabled)
439 return 0;
440
441 if (!fw_dump.fadump_supported) {
442 pr_info("Firmware-Assisted Dump is not supported on this hardware\n");
443 goto error_out;
444 }
445
446 /*
447 * Initialize boot memory size
448 * If dump is active then we have already calculated the size during
449 * first kernel.
450 */
451 if (!fw_dump.dump_active) {
452 fw_dump.boot_memory_size =
453 PAGE_ALIGN(fadump_calculate_reserve_size());
454#ifdef CONFIG_CMA
455 if (!fw_dump.nocma) {
456 align = FADUMP_CMA_ALIGNMENT;
457 fw_dump.boot_memory_size =
458 ALIGN(fw_dump.boot_memory_size, align);
459 }
460#endif
461
462 bootmem_min = fw_dump.ops->fadump_get_bootmem_min();
463 if (fw_dump.boot_memory_size < bootmem_min) {
464 pr_err("Can't enable fadump with boot memory size (0x%lx) less than 0x%llx\n",
465 fw_dump.boot_memory_size, bootmem_min);
466 goto error_out;
467 }
468
469 if (!fadump_get_boot_mem_regions()) {
470 pr_err("Too many holes in boot memory area to enable fadump\n");
471 goto error_out;
472 }
473 }
474
475 /*
476 * Calculate the memory boundary.
477 * If memory_limit is less than actual memory boundary then reserve
478 * the memory for fadump beyond the memory_limit and adjust the
479 * memory_limit accordingly, so that the running kernel can run with
480 * specified memory_limit.
481 */
482 if (memory_limit && memory_limit < memblock_end_of_DRAM()) {
483 size = get_fadump_area_size();
484 if ((memory_limit + size) < memblock_end_of_DRAM())
485 memory_limit += size;
486 else
487 memory_limit = memblock_end_of_DRAM();
488 printk(KERN_INFO "Adjusted memory_limit for firmware-assisted"
489 " dump, now %#016llx\n", memory_limit);
490 }
491 if (memory_limit)
492 mem_boundary = memory_limit;
493 else
494 mem_boundary = memblock_end_of_DRAM();
495
496 base = fw_dump.boot_mem_top;
497 size = get_fadump_area_size();
498 fw_dump.reserve_dump_area_size = size;
499 if (fw_dump.dump_active) {
500 pr_info("Firmware-assisted dump is active.\n");
501
502#ifdef CONFIG_HUGETLB_PAGE
503 /*
504 * FADump capture kernel doesn't care much about hugepages.
505 * In fact, handling hugepages in capture kernel is asking for
506 * trouble. So, disable HugeTLB support when fadump is active.
507 */
508 hugetlb_disabled = true;
509#endif
510 /*
511 * If last boot has crashed then reserve all the memory
512 * above boot memory size so that we don't touch it until
513 * dump is written to disk by userspace tool. This memory
514 * can be released for general use by invalidating fadump.
515 */
516 fadump_reserve_crash_area(base);
517
518 pr_debug("fadumphdr_addr = %#016lx\n", fw_dump.fadumphdr_addr);
519 pr_debug("Reserve dump area start address: 0x%lx\n",
520 fw_dump.reserve_dump_area_start);
521 } else {
522 /*
523 * Reserve memory at an offset closer to bottom of the RAM to
524 * minimize the impact of memory hot-remove operation.
525 */
526 memblock_set_bottom_up(true);
527 base = memblock_find_in_range(base, mem_boundary, size, align);
528
529 /* Restore the previous allocation mode */
530 memblock_set_bottom_up(is_memblock_bottom_up);
531
532 if (!base) {
533 pr_err("Failed to find memory chunk for reservation!\n");
534 goto error_out;
535 }
536 fw_dump.reserve_dump_area_start = base;
537
538 /*
539 * Calculate the kernel metadata address and register it with
540 * f/w if the platform supports.
541 */
542 if (fw_dump.ops->fadump_setup_metadata &&
543 (fw_dump.ops->fadump_setup_metadata(&fw_dump) < 0))
544 goto error_out;
545
546 if (memblock_reserve(base, size)) {
547 pr_err("Failed to reserve memory!\n");
548 goto error_out;
549 }
550
551 pr_info("Reserved %lldMB of memory at %#016llx (System RAM: %lldMB)\n",
552 (size >> 20), base, (memblock_phys_mem_size() >> 20));
553
554 ret = fadump_cma_init();
555 }
556
557 return ret;
558error_out:
559 fw_dump.fadump_enabled = 0;
560 return 0;
561}
562
563/* Look for fadump= cmdline option. */
564static int __init early_fadump_param(char *p)
565{
566 if (!p)
567 return 1;
568
569 if (strncmp(p, "on", 2) == 0)
570 fw_dump.fadump_enabled = 1;
571 else if (strncmp(p, "off", 3) == 0)
572 fw_dump.fadump_enabled = 0;
573 else if (strncmp(p, "nocma", 5) == 0) {
574 fw_dump.fadump_enabled = 1;
575 fw_dump.nocma = 1;
576 }
577
578 return 0;
579}
580early_param("fadump", early_fadump_param);
581
582/*
583 * Look for fadump_reserve_mem= cmdline option
584 * TODO: Remove references to 'fadump_reserve_mem=' parameter,
585 * the sooner 'crashkernel=' parameter is accustomed to.
586 */
587static int __init early_fadump_reserve_mem(char *p)
588{
589 if (p)
590 fw_dump.reserve_bootvar = memparse(p, &p);
591 return 0;
592}
593early_param("fadump_reserve_mem", early_fadump_reserve_mem);
594
595void crash_fadump(struct pt_regs *regs, const char *str)
596{
597 struct fadump_crash_info_header *fdh = NULL;
598 int old_cpu, this_cpu;
599
600 if (!should_fadump_crash())
601 return;
602
603 /*
604 * old_cpu == -1 means this is the first CPU which has come here,
605 * go ahead and trigger fadump.
606 *
607 * old_cpu != -1 means some other CPU has already on it's way
608 * to trigger fadump, just keep looping here.
609 */
610 this_cpu = smp_processor_id();
611 old_cpu = cmpxchg(&crashing_cpu, -1, this_cpu);
612
613 if (old_cpu != -1) {
614 /*
615 * We can't loop here indefinitely. Wait as long as fadump
616 * is in force. If we race with fadump un-registration this
617 * loop will break and then we go down to normal panic path
618 * and reboot. If fadump is in force the first crashing
619 * cpu will definitely trigger fadump.
620 */
621 while (fw_dump.dump_registered)
622 cpu_relax();
623 return;
624 }
625
626 fdh = __va(fw_dump.fadumphdr_addr);
627 fdh->crashing_cpu = crashing_cpu;
628 crash_save_vmcoreinfo();
629
630 if (regs)
631 fdh->regs = *regs;
632 else
633 ppc_save_regs(&fdh->regs);
634
635 fdh->online_mask = *cpu_online_mask;
636
637 fw_dump.ops->fadump_trigger(fdh, str);
638}
639
640u32 *fadump_regs_to_elf_notes(u32 *buf, struct pt_regs *regs)
641{
642 struct elf_prstatus prstatus;
643
644 memset(&prstatus, 0, sizeof(prstatus));
645 /*
646 * FIXME: How do i get PID? Do I really need it?
647 * prstatus.pr_pid = ????
648 */
649 elf_core_copy_kernel_regs(&prstatus.pr_reg, regs);
650 buf = append_elf_note(buf, CRASH_CORE_NOTE_NAME, NT_PRSTATUS,
651 &prstatus, sizeof(prstatus));
652 return buf;
653}
654
655void fadump_update_elfcore_header(char *bufp)
656{
657 struct elfhdr *elf;
658 struct elf_phdr *phdr;
659
660 elf = (struct elfhdr *)bufp;
661 bufp += sizeof(struct elfhdr);
662
663 /* First note is a place holder for cpu notes info. */
664 phdr = (struct elf_phdr *)bufp;
665
666 if (phdr->p_type == PT_NOTE) {
667 phdr->p_paddr = __pa(fw_dump.cpu_notes_buf_vaddr);
668 phdr->p_offset = phdr->p_paddr;
669 phdr->p_filesz = fw_dump.cpu_notes_buf_size;
670 phdr->p_memsz = fw_dump.cpu_notes_buf_size;
671 }
672 return;
673}
674
675static void *fadump_alloc_buffer(unsigned long size)
676{
677 unsigned long count, i;
678 struct page *page;
679 void *vaddr;
680
681 vaddr = alloc_pages_exact(size, GFP_KERNEL | __GFP_ZERO);
682 if (!vaddr)
683 return NULL;
684
685 count = PAGE_ALIGN(size) / PAGE_SIZE;
686 page = virt_to_page(vaddr);
687 for (i = 0; i < count; i++)
688 mark_page_reserved(page + i);
689 return vaddr;
690}
691
692static void fadump_free_buffer(unsigned long vaddr, unsigned long size)
693{
694 free_reserved_area((void *)vaddr, (void *)(vaddr + size), -1, NULL);
695}
696
697s32 fadump_setup_cpu_notes_buf(u32 num_cpus)
698{
699 /* Allocate buffer to hold cpu crash notes. */
700 fw_dump.cpu_notes_buf_size = num_cpus * sizeof(note_buf_t);
701 fw_dump.cpu_notes_buf_size = PAGE_ALIGN(fw_dump.cpu_notes_buf_size);
702 fw_dump.cpu_notes_buf_vaddr =
703 (unsigned long)fadump_alloc_buffer(fw_dump.cpu_notes_buf_size);
704 if (!fw_dump.cpu_notes_buf_vaddr) {
705 pr_err("Failed to allocate %ld bytes for CPU notes buffer\n",
706 fw_dump.cpu_notes_buf_size);
707 return -ENOMEM;
708 }
709
710 pr_debug("Allocated buffer for cpu notes of size %ld at 0x%lx\n",
711 fw_dump.cpu_notes_buf_size,
712 fw_dump.cpu_notes_buf_vaddr);
713 return 0;
714}
715
716void fadump_free_cpu_notes_buf(void)
717{
718 if (!fw_dump.cpu_notes_buf_vaddr)
719 return;
720
721 fadump_free_buffer(fw_dump.cpu_notes_buf_vaddr,
722 fw_dump.cpu_notes_buf_size);
723 fw_dump.cpu_notes_buf_vaddr = 0;
724 fw_dump.cpu_notes_buf_size = 0;
725}
726
727static void fadump_free_mem_ranges(struct fadump_mrange_info *mrange_info)
728{
729 kfree(mrange_info->mem_ranges);
730 mrange_info->mem_ranges = NULL;
731 mrange_info->mem_ranges_sz = 0;
732 mrange_info->max_mem_ranges = 0;
733}
734
735/*
736 * Allocate or reallocate mem_ranges array in incremental units
737 * of PAGE_SIZE.
738 */
739static int fadump_alloc_mem_ranges(struct fadump_mrange_info *mrange_info)
740{
741 struct fadump_memory_range *new_array;
742 u64 new_size;
743
744 new_size = mrange_info->mem_ranges_sz + PAGE_SIZE;
745 pr_debug("Allocating %llu bytes of memory for %s memory ranges\n",
746 new_size, mrange_info->name);
747
748 new_array = krealloc(mrange_info->mem_ranges, new_size, GFP_KERNEL);
749 if (new_array == NULL) {
750 pr_err("Insufficient memory for setting up %s memory ranges\n",
751 mrange_info->name);
752 fadump_free_mem_ranges(mrange_info);
753 return -ENOMEM;
754 }
755
756 mrange_info->mem_ranges = new_array;
757 mrange_info->mem_ranges_sz = new_size;
758 mrange_info->max_mem_ranges = (new_size /
759 sizeof(struct fadump_memory_range));
760 return 0;
761}
762
763static inline int fadump_add_mem_range(struct fadump_mrange_info *mrange_info,
764 u64 base, u64 end)
765{
766 struct fadump_memory_range *mem_ranges = mrange_info->mem_ranges;
767 bool is_adjacent = false;
768 u64 start, size;
769
770 if (base == end)
771 return 0;
772
773 /*
774 * Fold adjacent memory ranges to bring down the memory ranges/
775 * PT_LOAD segments count.
776 */
777 if (mrange_info->mem_range_cnt) {
778 start = mem_ranges[mrange_info->mem_range_cnt - 1].base;
779 size = mem_ranges[mrange_info->mem_range_cnt - 1].size;
780
781 if ((start + size) == base)
782 is_adjacent = true;
783 }
784 if (!is_adjacent) {
785 /* resize the array on reaching the limit */
786 if (mrange_info->mem_range_cnt == mrange_info->max_mem_ranges) {
787 int ret;
788
789 ret = fadump_alloc_mem_ranges(mrange_info);
790 if (ret)
791 return ret;
792
793 /* Update to the new resized array */
794 mem_ranges = mrange_info->mem_ranges;
795 }
796
797 start = base;
798 mem_ranges[mrange_info->mem_range_cnt].base = start;
799 mrange_info->mem_range_cnt++;
800 }
801
802 mem_ranges[mrange_info->mem_range_cnt - 1].size = (end - start);
803 pr_debug("%s_memory_range[%d] [%#016llx-%#016llx], %#llx bytes\n",
804 mrange_info->name, (mrange_info->mem_range_cnt - 1),
805 start, end - 1, (end - start));
806 return 0;
807}
808
809static int fadump_exclude_reserved_area(u64 start, u64 end)
810{
811 u64 ra_start, ra_end;
812 int ret = 0;
813
814 ra_start = fw_dump.reserve_dump_area_start;
815 ra_end = ra_start + fw_dump.reserve_dump_area_size;
816
817 if ((ra_start < end) && (ra_end > start)) {
818 if ((start < ra_start) && (end > ra_end)) {
819 ret = fadump_add_mem_range(&crash_mrange_info,
820 start, ra_start);
821 if (ret)
822 return ret;
823
824 ret = fadump_add_mem_range(&crash_mrange_info,
825 ra_end, end);
826 } else if (start < ra_start) {
827 ret = fadump_add_mem_range(&crash_mrange_info,
828 start, ra_start);
829 } else if (ra_end < end) {
830 ret = fadump_add_mem_range(&crash_mrange_info,
831 ra_end, end);
832 }
833 } else
834 ret = fadump_add_mem_range(&crash_mrange_info, start, end);
835
836 return ret;
837}
838
839static int fadump_init_elfcore_header(char *bufp)
840{
841 struct elfhdr *elf;
842
843 elf = (struct elfhdr *) bufp;
844 bufp += sizeof(struct elfhdr);
845 memcpy(elf->e_ident, ELFMAG, SELFMAG);
846 elf->e_ident[EI_CLASS] = ELF_CLASS;
847 elf->e_ident[EI_DATA] = ELF_DATA;
848 elf->e_ident[EI_VERSION] = EV_CURRENT;
849 elf->e_ident[EI_OSABI] = ELF_OSABI;
850 memset(elf->e_ident+EI_PAD, 0, EI_NIDENT-EI_PAD);
851 elf->e_type = ET_CORE;
852 elf->e_machine = ELF_ARCH;
853 elf->e_version = EV_CURRENT;
854 elf->e_entry = 0;
855 elf->e_phoff = sizeof(struct elfhdr);
856 elf->e_shoff = 0;
857#if defined(_CALL_ELF)
858 elf->e_flags = _CALL_ELF;
859#else
860 elf->e_flags = 0;
861#endif
862 elf->e_ehsize = sizeof(struct elfhdr);
863 elf->e_phentsize = sizeof(struct elf_phdr);
864 elf->e_phnum = 0;
865 elf->e_shentsize = 0;
866 elf->e_shnum = 0;
867 elf->e_shstrndx = 0;
868
869 return 0;
870}
871
872/*
873 * Traverse through memblock structure and setup crash memory ranges. These
874 * ranges will be used create PT_LOAD program headers in elfcore header.
875 */
876static int fadump_setup_crash_memory_ranges(void)
877{
878 struct memblock_region *reg;
879 u64 start, end;
880 int i, ret;
881
882 pr_debug("Setup crash memory ranges.\n");
883 crash_mrange_info.mem_range_cnt = 0;
884
885 /*
886 * Boot memory region(s) registered with firmware are moved to
887 * different location at the time of crash. Create separate program
888 * header(s) for this memory chunk(s) with the correct offset.
889 */
890 for (i = 0; i < fw_dump.boot_mem_regs_cnt; i++) {
891 start = fw_dump.boot_mem_addr[i];
892 end = start + fw_dump.boot_mem_sz[i];
893 ret = fadump_add_mem_range(&crash_mrange_info, start, end);
894 if (ret)
895 return ret;
896 }
897
898 for_each_memblock(memory, reg) {
899 start = (u64)reg->base;
900 end = start + (u64)reg->size;
901
902 /*
903 * skip the memory chunk that is already added
904 * (0 through boot_memory_top).
905 */
906 if (start < fw_dump.boot_mem_top) {
907 if (end > fw_dump.boot_mem_top)
908 start = fw_dump.boot_mem_top;
909 else
910 continue;
911 }
912
913 /* add this range excluding the reserved dump area. */
914 ret = fadump_exclude_reserved_area(start, end);
915 if (ret)
916 return ret;
917 }
918
919 return 0;
920}
921
922/*
923 * If the given physical address falls within the boot memory region then
924 * return the relocated address that points to the dump region reserved
925 * for saving initial boot memory contents.
926 */
927static inline unsigned long fadump_relocate(unsigned long paddr)
928{
929 unsigned long raddr, rstart, rend, rlast, hole_size;
930 int i;
931
932 hole_size = 0;
933 rlast = 0;
934 raddr = paddr;
935 for (i = 0; i < fw_dump.boot_mem_regs_cnt; i++) {
936 rstart = fw_dump.boot_mem_addr[i];
937 rend = rstart + fw_dump.boot_mem_sz[i];
938 hole_size += (rstart - rlast);
939
940 if (paddr >= rstart && paddr < rend) {
941 raddr += fw_dump.boot_mem_dest_addr - hole_size;
942 break;
943 }
944
945 rlast = rend;
946 }
947
948 pr_debug("vmcoreinfo: paddr = 0x%lx, raddr = 0x%lx\n", paddr, raddr);
949 return raddr;
950}
951
952static int fadump_create_elfcore_headers(char *bufp)
953{
954 unsigned long long raddr, offset;
955 struct elf_phdr *phdr;
956 struct elfhdr *elf;
957 int i, j;
958
959 fadump_init_elfcore_header(bufp);
960 elf = (struct elfhdr *)bufp;
961 bufp += sizeof(struct elfhdr);
962
963 /*
964 * setup ELF PT_NOTE, place holder for cpu notes info. The notes info
965 * will be populated during second kernel boot after crash. Hence
966 * this PT_NOTE will always be the first elf note.
967 *
968 * NOTE: Any new ELF note addition should be placed after this note.
969 */
970 phdr = (struct elf_phdr *)bufp;
971 bufp += sizeof(struct elf_phdr);
972 phdr->p_type = PT_NOTE;
973 phdr->p_flags = 0;
974 phdr->p_vaddr = 0;
975 phdr->p_align = 0;
976
977 phdr->p_offset = 0;
978 phdr->p_paddr = 0;
979 phdr->p_filesz = 0;
980 phdr->p_memsz = 0;
981
982 (elf->e_phnum)++;
983
984 /* setup ELF PT_NOTE for vmcoreinfo */
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_paddr = fadump_relocate(paddr_vmcoreinfo_note());
993 phdr->p_offset = phdr->p_paddr;
994 phdr->p_memsz = phdr->p_filesz = VMCOREINFO_NOTE_SIZE;
995
996 /* Increment number of program headers. */
997 (elf->e_phnum)++;
998
999 /* setup PT_LOAD sections. */
1000 j = 0;
1001 offset = 0;
1002 raddr = fw_dump.boot_mem_addr[0];
1003 for (i = 0; i < crash_mrange_info.mem_range_cnt; i++) {
1004 u64 mbase, msize;
1005
1006 mbase = crash_mrange_info.mem_ranges[i].base;
1007 msize = crash_mrange_info.mem_ranges[i].size;
1008 if (!msize)
1009 continue;
1010
1011 phdr = (struct elf_phdr *)bufp;
1012 bufp += sizeof(struct elf_phdr);
1013 phdr->p_type = PT_LOAD;
1014 phdr->p_flags = PF_R|PF_W|PF_X;
1015 phdr->p_offset = mbase;
1016
1017 if (mbase == raddr) {
1018 /*
1019 * The entire real memory region will be moved by
1020 * firmware to the specified destination_address.
1021 * Hence set the correct offset.
1022 */
1023 phdr->p_offset = fw_dump.boot_mem_dest_addr + offset;
1024 if (j < (fw_dump.boot_mem_regs_cnt - 1)) {
1025 offset += fw_dump.boot_mem_sz[j];
1026 raddr = fw_dump.boot_mem_addr[++j];
1027 }
1028 }
1029
1030 phdr->p_paddr = mbase;
1031 phdr->p_vaddr = (unsigned long)__va(mbase);
1032 phdr->p_filesz = msize;
1033 phdr->p_memsz = msize;
1034 phdr->p_align = 0;
1035
1036 /* Increment number of program headers. */
1037 (elf->e_phnum)++;
1038 }
1039 return 0;
1040}
1041
1042static unsigned long init_fadump_header(unsigned long addr)
1043{
1044 struct fadump_crash_info_header *fdh;
1045
1046 if (!addr)
1047 return 0;
1048
1049 fdh = __va(addr);
1050 addr += sizeof(struct fadump_crash_info_header);
1051
1052 memset(fdh, 0, sizeof(struct fadump_crash_info_header));
1053 fdh->magic_number = FADUMP_CRASH_INFO_MAGIC;
1054 fdh->elfcorehdr_addr = addr;
1055 /* We will set the crashing cpu id in crash_fadump() during crash. */
1056 fdh->crashing_cpu = FADUMP_CPU_UNKNOWN;
1057
1058 return addr;
1059}
1060
1061static int register_fadump(void)
1062{
1063 unsigned long addr;
1064 void *vaddr;
1065 int ret;
1066
1067 /*
1068 * If no memory is reserved then we can not register for firmware-
1069 * assisted dump.
1070 */
1071 if (!fw_dump.reserve_dump_area_size)
1072 return -ENODEV;
1073
1074 ret = fadump_setup_crash_memory_ranges();
1075 if (ret)
1076 return ret;
1077
1078 addr = fw_dump.fadumphdr_addr;
1079
1080 /* Initialize fadump crash info header. */
1081 addr = init_fadump_header(addr);
1082 vaddr = __va(addr);
1083
1084 pr_debug("Creating ELF core headers at %#016lx\n", addr);
1085 fadump_create_elfcore_headers(vaddr);
1086
1087 /* register the future kernel dump with firmware. */
1088 pr_debug("Registering for firmware-assisted kernel dump...\n");
1089 return fw_dump.ops->fadump_register(&fw_dump);
1090}
1091
1092void fadump_cleanup(void)
1093{
1094 if (!fw_dump.fadump_supported)
1095 return;
1096
1097 /* Invalidate the registration only if dump is active. */
1098 if (fw_dump.dump_active) {
1099 pr_debug("Invalidating firmware-assisted dump registration\n");
1100 fw_dump.ops->fadump_invalidate(&fw_dump);
1101 } else if (fw_dump.dump_registered) {
1102 /* Un-register Firmware-assisted dump if it was registered. */
1103 fw_dump.ops->fadump_unregister(&fw_dump);
1104 fadump_free_mem_ranges(&crash_mrange_info);
1105 }
1106
1107 if (fw_dump.ops->fadump_cleanup)
1108 fw_dump.ops->fadump_cleanup(&fw_dump);
1109}
1110
1111static void fadump_free_reserved_memory(unsigned long start_pfn,
1112 unsigned long end_pfn)
1113{
1114 unsigned long pfn;
1115 unsigned long time_limit = jiffies + HZ;
1116
1117 pr_info("freeing reserved memory (0x%llx - 0x%llx)\n",
1118 PFN_PHYS(start_pfn), PFN_PHYS(end_pfn));
1119
1120 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
1121 free_reserved_page(pfn_to_page(pfn));
1122
1123 if (time_after(jiffies, time_limit)) {
1124 cond_resched();
1125 time_limit = jiffies + HZ;
1126 }
1127 }
1128}
1129
1130/*
1131 * Skip memory holes and free memory that was actually reserved.
1132 */
1133static void fadump_release_reserved_area(u64 start, u64 end)
1134{
1135 u64 tstart, tend, spfn, epfn;
1136 struct memblock_region *reg;
1137
1138 spfn = PHYS_PFN(start);
1139 epfn = PHYS_PFN(end);
1140 for_each_memblock(memory, reg) {
1141 tstart = max_t(u64, spfn, memblock_region_memory_base_pfn(reg));
1142 tend = min_t(u64, epfn, memblock_region_memory_end_pfn(reg));
1143 if (tstart < tend) {
1144 fadump_free_reserved_memory(tstart, tend);
1145
1146 if (tend == epfn)
1147 break;
1148
1149 spfn = tend;
1150 }
1151 }
1152}
1153
1154/*
1155 * Sort the mem ranges in-place and merge adjacent ranges
1156 * to minimize the memory ranges count.
1157 */
1158static void sort_and_merge_mem_ranges(struct fadump_mrange_info *mrange_info)
1159{
1160 struct fadump_memory_range *mem_ranges;
1161 struct fadump_memory_range tmp_range;
1162 u64 base, size;
1163 int i, j, idx;
1164
1165 if (!reserved_mrange_info.mem_range_cnt)
1166 return;
1167
1168 /* Sort the memory ranges */
1169 mem_ranges = mrange_info->mem_ranges;
1170 for (i = 0; i < mrange_info->mem_range_cnt; i++) {
1171 idx = i;
1172 for (j = (i + 1); j < mrange_info->mem_range_cnt; j++) {
1173 if (mem_ranges[idx].base > mem_ranges[j].base)
1174 idx = j;
1175 }
1176 if (idx != i) {
1177 tmp_range = mem_ranges[idx];
1178 mem_ranges[idx] = mem_ranges[i];
1179 mem_ranges[i] = tmp_range;
1180 }
1181 }
1182
1183 /* Merge adjacent reserved ranges */
1184 idx = 0;
1185 for (i = 1; i < mrange_info->mem_range_cnt; i++) {
1186 base = mem_ranges[i-1].base;
1187 size = mem_ranges[i-1].size;
1188 if (mem_ranges[i].base == (base + size))
1189 mem_ranges[idx].size += mem_ranges[i].size;
1190 else {
1191 idx++;
1192 if (i == idx)
1193 continue;
1194
1195 mem_ranges[idx] = mem_ranges[i];
1196 }
1197 }
1198 mrange_info->mem_range_cnt = idx + 1;
1199}
1200
1201/*
1202 * Scan reserved-ranges to consider them while reserving/releasing
1203 * memory for FADump.
1204 */
1205static inline int fadump_scan_reserved_mem_ranges(void)
1206{
1207 struct device_node *root;
1208 const __be32 *prop;
1209 int len, ret = -1;
1210 unsigned long i;
1211
1212 root = of_find_node_by_path("/");
1213 if (!root)
1214 return ret;
1215
1216 prop = of_get_property(root, "reserved-ranges", &len);
1217 if (!prop)
1218 return ret;
1219
1220 /*
1221 * Each reserved range is an (address,size) pair, 2 cells each,
1222 * totalling 4 cells per range.
1223 */
1224 for (i = 0; i < len / (sizeof(*prop) * 4); i++) {
1225 u64 base, size;
1226
1227 base = of_read_number(prop + (i * 4) + 0, 2);
1228 size = of_read_number(prop + (i * 4) + 2, 2);
1229
1230 if (size) {
1231 ret = fadump_add_mem_range(&reserved_mrange_info,
1232 base, base + size);
1233 if (ret < 0) {
1234 pr_warn("some reserved ranges are ignored!\n");
1235 break;
1236 }
1237 }
1238 }
1239
1240 return ret;
1241}
1242
1243/*
1244 * Release the memory that was reserved during early boot to preserve the
1245 * crash'ed kernel's memory contents except reserved dump area (permanent
1246 * reservation) and reserved ranges used by F/W. The released memory will
1247 * be available for general use.
1248 */
1249static void fadump_release_memory(u64 begin, u64 end)
1250{
1251 u64 ra_start, ra_end, tstart;
1252 int i, ret;
1253
1254 fadump_scan_reserved_mem_ranges();
1255
1256 ra_start = fw_dump.reserve_dump_area_start;
1257 ra_end = ra_start + fw_dump.reserve_dump_area_size;
1258
1259 /*
1260 * Add reserved dump area to reserved ranges list
1261 * and exclude all these ranges while releasing memory.
1262 */
1263 ret = fadump_add_mem_range(&reserved_mrange_info, ra_start, ra_end);
1264 if (ret != 0) {
1265 /*
1266 * Not enough memory to setup reserved ranges but the system is
1267 * running shortage of memory. So, release all the memory except
1268 * Reserved dump area (reused for next fadump registration).
1269 */
1270 if (begin < ra_end && end > ra_start) {
1271 if (begin < ra_start)
1272 fadump_release_reserved_area(begin, ra_start);
1273 if (end > ra_end)
1274 fadump_release_reserved_area(ra_end, end);
1275 } else
1276 fadump_release_reserved_area(begin, end);
1277
1278 return;
1279 }
1280
1281 /* Get the reserved ranges list in order first. */
1282 sort_and_merge_mem_ranges(&reserved_mrange_info);
1283
1284 /* Exclude reserved ranges and release remaining memory */
1285 tstart = begin;
1286 for (i = 0; i < reserved_mrange_info.mem_range_cnt; i++) {
1287 ra_start = reserved_mrange_info.mem_ranges[i].base;
1288 ra_end = ra_start + reserved_mrange_info.mem_ranges[i].size;
1289
1290 if (tstart >= ra_end)
1291 continue;
1292
1293 if (tstart < ra_start)
1294 fadump_release_reserved_area(tstart, ra_start);
1295 tstart = ra_end;
1296 }
1297
1298 if (tstart < end)
1299 fadump_release_reserved_area(tstart, end);
1300}
1301
1302static void fadump_invalidate_release_mem(void)
1303{
1304 mutex_lock(&fadump_mutex);
1305 if (!fw_dump.dump_active) {
1306 mutex_unlock(&fadump_mutex);
1307 return;
1308 }
1309
1310 fadump_cleanup();
1311 mutex_unlock(&fadump_mutex);
1312
1313 fadump_release_memory(fw_dump.boot_mem_top, memblock_end_of_DRAM());
1314 fadump_free_cpu_notes_buf();
1315
1316 /*
1317 * Setup kernel metadata and initialize the kernel dump
1318 * memory structure for FADump re-registration.
1319 */
1320 if (fw_dump.ops->fadump_setup_metadata &&
1321 (fw_dump.ops->fadump_setup_metadata(&fw_dump) < 0))
1322 pr_warn("Failed to setup kernel metadata!\n");
1323 fw_dump.ops->fadump_init_mem_struct(&fw_dump);
1324}
1325
1326static ssize_t fadump_release_memory_store(struct kobject *kobj,
1327 struct kobj_attribute *attr,
1328 const char *buf, size_t count)
1329{
1330 int input = -1;
1331
1332 if (!fw_dump.dump_active)
1333 return -EPERM;
1334
1335 if (kstrtoint(buf, 0, &input))
1336 return -EINVAL;
1337
1338 if (input == 1) {
1339 /*
1340 * Take away the '/proc/vmcore'. We are releasing the dump
1341 * memory, hence it will not be valid anymore.
1342 */
1343#ifdef CONFIG_PROC_VMCORE
1344 vmcore_cleanup();
1345#endif
1346 fadump_invalidate_release_mem();
1347
1348 } else
1349 return -EINVAL;
1350 return count;
1351}
1352
1353static ssize_t fadump_enabled_show(struct kobject *kobj,
1354 struct kobj_attribute *attr,
1355 char *buf)
1356{
1357 return sprintf(buf, "%d\n", fw_dump.fadump_enabled);
1358}
1359
1360static ssize_t fadump_register_show(struct kobject *kobj,
1361 struct kobj_attribute *attr,
1362 char *buf)
1363{
1364 return sprintf(buf, "%d\n", fw_dump.dump_registered);
1365}
1366
1367static ssize_t fadump_register_store(struct kobject *kobj,
1368 struct kobj_attribute *attr,
1369 const char *buf, size_t count)
1370{
1371 int ret = 0;
1372 int input = -1;
1373
1374 if (!fw_dump.fadump_enabled || fw_dump.dump_active)
1375 return -EPERM;
1376
1377 if (kstrtoint(buf, 0, &input))
1378 return -EINVAL;
1379
1380 mutex_lock(&fadump_mutex);
1381
1382 switch (input) {
1383 case 0:
1384 if (fw_dump.dump_registered == 0) {
1385 goto unlock_out;
1386 }
1387
1388 /* Un-register Firmware-assisted dump */
1389 pr_debug("Un-register firmware-assisted dump\n");
1390 fw_dump.ops->fadump_unregister(&fw_dump);
1391 break;
1392 case 1:
1393 if (fw_dump.dump_registered == 1) {
1394 /* Un-register Firmware-assisted dump */
1395 fw_dump.ops->fadump_unregister(&fw_dump);
1396 }
1397 /* Register Firmware-assisted dump */
1398 ret = register_fadump();
1399 break;
1400 default:
1401 ret = -EINVAL;
1402 break;
1403 }
1404
1405unlock_out:
1406 mutex_unlock(&fadump_mutex);
1407 return ret < 0 ? ret : count;
1408}
1409
1410static int fadump_region_show(struct seq_file *m, void *private)
1411{
1412 if (!fw_dump.fadump_enabled)
1413 return 0;
1414
1415 mutex_lock(&fadump_mutex);
1416 fw_dump.ops->fadump_region_show(&fw_dump, m);
1417 mutex_unlock(&fadump_mutex);
1418 return 0;
1419}
1420
1421static struct kobj_attribute fadump_release_attr = __ATTR(fadump_release_mem,
1422 0200, NULL,
1423 fadump_release_memory_store);
1424static struct kobj_attribute fadump_attr = __ATTR(fadump_enabled,
1425 0444, fadump_enabled_show,
1426 NULL);
1427static struct kobj_attribute fadump_register_attr = __ATTR(fadump_registered,
1428 0644, fadump_register_show,
1429 fadump_register_store);
1430
1431DEFINE_SHOW_ATTRIBUTE(fadump_region);
1432
1433static void fadump_init_files(void)
1434{
1435 struct dentry *debugfs_file;
1436 int rc = 0;
1437
1438 rc = sysfs_create_file(kernel_kobj, &fadump_attr.attr);
1439 if (rc)
1440 printk(KERN_ERR "fadump: unable to create sysfs file"
1441 " fadump_enabled (%d)\n", rc);
1442
1443 rc = sysfs_create_file(kernel_kobj, &fadump_register_attr.attr);
1444 if (rc)
1445 printk(KERN_ERR "fadump: unable to create sysfs file"
1446 " fadump_registered (%d)\n", rc);
1447
1448 debugfs_file = debugfs_create_file("fadump_region", 0444,
1449 powerpc_debugfs_root, NULL,
1450 &fadump_region_fops);
1451 if (!debugfs_file)
1452 printk(KERN_ERR "fadump: unable to create debugfs file"
1453 " fadump_region\n");
1454
1455 if (fw_dump.dump_active) {
1456 rc = sysfs_create_file(kernel_kobj, &fadump_release_attr.attr);
1457 if (rc)
1458 printk(KERN_ERR "fadump: unable to create sysfs file"
1459 " fadump_release_mem (%d)\n", rc);
1460 }
1461 return;
1462}
1463
1464/*
1465 * Prepare for firmware-assisted dump.
1466 */
1467int __init setup_fadump(void)
1468{
1469 if (!fw_dump.fadump_enabled)
1470 return 0;
1471
1472 if (!fw_dump.fadump_supported) {
1473 printk(KERN_ERR "Firmware-assisted dump is not supported on"
1474 " this hardware\n");
1475 return 0;
1476 }
1477
1478 fadump_show_config();
1479 /*
1480 * If dump data is available then see if it is valid and prepare for
1481 * saving it to the disk.
1482 */
1483 if (fw_dump.dump_active) {
1484 /*
1485 * if dump process fails then invalidate the registration
1486 * and release memory before proceeding for re-registration.
1487 */
1488 if (fw_dump.ops->fadump_process(&fw_dump) < 0)
1489 fadump_invalidate_release_mem();
1490 }
1491 /* Initialize the kernel dump memory structure for FAD registration. */
1492 else if (fw_dump.reserve_dump_area_size)
1493 fw_dump.ops->fadump_init_mem_struct(&fw_dump);
1494
1495 fadump_init_files();
1496
1497 return 1;
1498}
1499subsys_initcall(setup_fadump);
1500#else /* !CONFIG_PRESERVE_FA_DUMP */
1501
1502/* Scan the Firmware Assisted dump configuration details. */
1503int __init early_init_dt_scan_fw_dump(unsigned long node, const char *uname,
1504 int depth, void *data)
1505{
1506 if ((depth != 1) || (strcmp(uname, "ibm,opal") != 0))
1507 return 0;
1508
1509 opal_fadump_dt_scan(&fw_dump, node);
1510 return 1;
1511}
1512
1513/*
1514 * When dump is active but PRESERVE_FA_DUMP is enabled on the kernel,
1515 * preserve crash data. The subsequent memory preserving kernel boot
1516 * is likely to process this crash data.
1517 */
1518int __init fadump_reserve_mem(void)
1519{
1520 if (fw_dump.dump_active) {
1521 /*
1522 * If last boot has crashed then reserve all the memory
1523 * above boot memory to preserve crash data.
1524 */
1525 pr_info("Preserving crash data for processing in next boot.\n");
1526 fadump_reserve_crash_area(fw_dump.boot_mem_top);
1527 } else
1528 pr_debug("FADump-aware kernel..\n");
1529
1530 return 1;
1531}
1532#endif /* CONFIG_PRESERVE_FA_DUMP */
1533
1534/* Preserve everything above the base address */
1535static void __init fadump_reserve_crash_area(u64 base)
1536{
1537 struct memblock_region *reg;
1538 u64 mstart, msize;
1539
1540 for_each_memblock(memory, reg) {
1541 mstart = reg->base;
1542 msize = reg->size;
1543
1544 if ((mstart + msize) < base)
1545 continue;
1546
1547 if (mstart < base) {
1548 msize -= (base - mstart);
1549 mstart = base;
1550 }
1551
1552 pr_info("Reserving %lluMB of memory at %#016llx for preserving crash data",
1553 (msize >> 20), mstart);
1554 memblock_reserve(mstart, msize);
1555 }
1556}
1557
1558unsigned long __init arch_reserved_kernel_pages(void)
1559{
1560 return memblock_reserved_size() / PAGE_SIZE;
1561}