11a59d1b8SThomas Gleixner // SPDX-License-Identifier: GPL-2.0-or-later
2eb39c880SMahesh Salgaonkar /*
3eb39c880SMahesh Salgaonkar * Firmware Assisted dump: A robust mechanism to get reliable kernel crash
4eb39c880SMahesh Salgaonkar * dump with assistance from firmware. This approach does not use kexec,
5eb39c880SMahesh Salgaonkar * instead firmware assists in booting the kdump kernel while preserving
6eb39c880SMahesh Salgaonkar * memory contents. The most of the code implementation has been adapted
7eb39c880SMahesh Salgaonkar * from phyp assisted dump implementation written by Linas Vepstas and
8eb39c880SMahesh Salgaonkar * Manish Ahuja
9eb39c880SMahesh Salgaonkar *
10eb39c880SMahesh Salgaonkar * Copyright 2011 IBM Corporation
11eb39c880SMahesh Salgaonkar * Author: Mahesh Salgaonkar <mahesh@linux.vnet.ibm.com>
12eb39c880SMahesh Salgaonkar */
13eb39c880SMahesh Salgaonkar
14eb39c880SMahesh Salgaonkar #undef DEBUG
15eb39c880SMahesh Salgaonkar #define pr_fmt(fmt) "fadump: " fmt
16eb39c880SMahesh Salgaonkar
17eb39c880SMahesh Salgaonkar #include <linux/string.h>
18eb39c880SMahesh Salgaonkar #include <linux/memblock.h>
193ccc00a7SMahesh Salgaonkar #include <linux/delay.h>
203ccc00a7SMahesh Salgaonkar #include <linux/seq_file.h>
212df173d9SMahesh Salgaonkar #include <linux/crash_dump.h>
22b500afffSMahesh Salgaonkar #include <linux/kobject.h>
23b500afffSMahesh Salgaonkar #include <linux/sysfs.h>
24a5818313SHari Bathini #include <linux/slab.h>
25a4e92ce8SMahesh Salgaonkar #include <linux/cma.h>
2645d0ba52SChristophe Leroy #include <linux/hugetlb.h>
27dbf77fedSAneesh Kumar K.V #include <linux/debugfs.h>
28e6f6390aSChristophe Leroy #include <linux/of.h>
29e6f6390aSChristophe Leroy #include <linux/of_fdt.h>
30eb39c880SMahesh Salgaonkar
31eb39c880SMahesh Salgaonkar #include <asm/page.h>
32eb39c880SMahesh Salgaonkar #include <asm/fadump.h>
33ca986d7fSHari Bathini #include <asm/fadump-internal.h>
34cad3c834SStephen Rothwell #include <asm/setup.h>
35cbd3d5baSMichael Ellerman #include <asm/interrupt.h>
36eb39c880SMahesh Salgaonkar
37ba608c4fSSourabh Jain /*
38ba608c4fSSourabh Jain * The CPU who acquired the lock to trigger the fadump crash should
39ba608c4fSSourabh Jain * wait for other CPUs to enter.
40ba608c4fSSourabh Jain *
41ba608c4fSSourabh Jain * The timeout is in milliseconds.
42ba608c4fSSourabh Jain */
43ba608c4fSSourabh Jain #define CRASH_TIMEOUT 500
44ba608c4fSSourabh Jain
45eb39c880SMahesh Salgaonkar static struct fw_dump fw_dump;
463ccc00a7SMahesh Salgaonkar
47b2a815a5SHari Bathini static void __init fadump_reserve_crash_area(u64 base);
48b2a815a5SHari Bathini
49bec53196SHari Bathini #ifndef CONFIG_PRESERVE_FA_DUMP
505f987caeSMichael Ellerman
512e341f56SMichael Ellerman static struct kobject *fadump_kobj;
522e341f56SMichael Ellerman
535f987caeSMichael Ellerman static atomic_t cpus_in_fadump;
543ccc00a7SMahesh Salgaonkar static DEFINE_MUTEX(fadump_mutex);
555f987caeSMichael Ellerman
562e341f56SMichael Ellerman static struct fadump_mrange_info crash_mrange_info = { "crash", NULL, 0, 0, 0, false };
5702c04e37SHari Bathini
5802c04e37SHari Bathini #define RESERVED_RNGS_SZ 16384 /* 16K - 128 entries */
5902c04e37SHari Bathini #define RESERVED_RNGS_CNT (RESERVED_RNGS_SZ / \
6002c04e37SHari Bathini sizeof(struct fadump_memory_range))
6102c04e37SHari Bathini static struct fadump_memory_range rngs[RESERVED_RNGS_CNT];
622e341f56SMichael Ellerman static struct fadump_mrange_info
632e341f56SMichael Ellerman reserved_mrange_info = { "reserved", rngs, RESERVED_RNGS_SZ, 0, RESERVED_RNGS_CNT, true };
6402c04e37SHari Bathini
6502c04e37SHari Bathini static void __init early_init_dt_scan_reserved_ranges(unsigned long node);
66eb39c880SMahesh Salgaonkar
67a4e92ce8SMahesh Salgaonkar #ifdef CONFIG_CMA
680226e552SHari Bathini static struct cma *fadump_cma;
690226e552SHari Bathini
70a4e92ce8SMahesh Salgaonkar /*
71a4e92ce8SMahesh Salgaonkar * fadump_cma_init() - Initialize CMA area from a fadump reserved memory
72a4e92ce8SMahesh Salgaonkar *
73a4e92ce8SMahesh Salgaonkar * This function initializes CMA area from fadump reserved memory.
74a4e92ce8SMahesh Salgaonkar * The total size of fadump reserved memory covers for boot memory size
75a4e92ce8SMahesh Salgaonkar * + cpu data size + hpte size and metadata.
76a4e92ce8SMahesh Salgaonkar * Initialize only the area equivalent to boot memory size for CMA use.
77887f56a0SRandy Dunlap * The remaining portion of fadump reserved memory will be not given
78887f56a0SRandy Dunlap * to CMA and pages for those will stay reserved. boot memory size is
79a4e92ce8SMahesh Salgaonkar * aligned per CMA requirement to satisy cma_init_reserved_mem() call.
80a4e92ce8SMahesh Salgaonkar * But for some reason even if it fails we still have the memory reservation
81a4e92ce8SMahesh Salgaonkar * with us and we can still continue doing fadump.
82a4e92ce8SMahesh Salgaonkar */
fadump_cma_init(void)832e341f56SMichael Ellerman static int __init fadump_cma_init(void)
84a4e92ce8SMahesh Salgaonkar {
85a4e92ce8SMahesh Salgaonkar unsigned long long base, size;
86a4e92ce8SMahesh Salgaonkar int rc;
87a4e92ce8SMahesh Salgaonkar
88a4e92ce8SMahesh Salgaonkar if (!fw_dump.fadump_enabled)
89a4e92ce8SMahesh Salgaonkar return 0;
90a4e92ce8SMahesh Salgaonkar
91a4e92ce8SMahesh Salgaonkar /*
92a4e92ce8SMahesh Salgaonkar * Do not use CMA if user has provided fadump=nocma kernel parameter.
93a4e92ce8SMahesh Salgaonkar * Return 1 to continue with fadump old behaviour.
94a4e92ce8SMahesh Salgaonkar */
95a4e92ce8SMahesh Salgaonkar if (fw_dump.nocma)
96a4e92ce8SMahesh Salgaonkar return 1;
97a4e92ce8SMahesh Salgaonkar
98a4e92ce8SMahesh Salgaonkar base = fw_dump.reserve_dump_area_start;
99a4e92ce8SMahesh Salgaonkar size = fw_dump.boot_memory_size;
100a4e92ce8SMahesh Salgaonkar
101a4e92ce8SMahesh Salgaonkar if (!size)
102a4e92ce8SMahesh Salgaonkar return 0;
103a4e92ce8SMahesh Salgaonkar
104a4e92ce8SMahesh Salgaonkar rc = cma_init_reserved_mem(base, size, 0, "fadump_cma", &fadump_cma);
105a4e92ce8SMahesh Salgaonkar if (rc) {
106a4e92ce8SMahesh Salgaonkar pr_err("Failed to init cma area for firmware-assisted dump,%d\n", rc);
107a4e92ce8SMahesh Salgaonkar /*
108a4e92ce8SMahesh Salgaonkar * Though the CMA init has failed we still have memory
109a4e92ce8SMahesh Salgaonkar * reservation with us. The reserved memory will be
110a4e92ce8SMahesh Salgaonkar * blocked from production system usage. Hence return 1,
111a4e92ce8SMahesh Salgaonkar * so that we can continue with fadump.
112a4e92ce8SMahesh Salgaonkar */
113a4e92ce8SMahesh Salgaonkar return 1;
114a4e92ce8SMahesh Salgaonkar }
115a4e92ce8SMahesh Salgaonkar
116a4e92ce8SMahesh Salgaonkar /*
117ee97347fSHari Bathini * If CMA activation fails, keep the pages reserved, instead of
118ee97347fSHari Bathini * exposing them to buddy allocator. Same as 'fadump=nocma' case.
119ee97347fSHari Bathini */
120ee97347fSHari Bathini cma_reserve_pages_on_error(fadump_cma);
121ee97347fSHari Bathini
122ee97347fSHari Bathini /*
123a4e92ce8SMahesh Salgaonkar * So we now have successfully initialized cma area for fadump.
124a4e92ce8SMahesh Salgaonkar */
125a4e92ce8SMahesh Salgaonkar pr_info("Initialized 0x%lx bytes cma area at %ldMB from 0x%lx "
126a4e92ce8SMahesh Salgaonkar "bytes of memory reserved for firmware-assisted dump\n",
127a4e92ce8SMahesh Salgaonkar cma_get_size(fadump_cma),
128a4e92ce8SMahesh Salgaonkar (unsigned long)cma_get_base(fadump_cma) >> 20,
129a4e92ce8SMahesh Salgaonkar fw_dump.reserve_dump_area_size);
130a4e92ce8SMahesh Salgaonkar return 1;
131a4e92ce8SMahesh Salgaonkar }
132a4e92ce8SMahesh Salgaonkar #else
fadump_cma_init(void)133a4e92ce8SMahesh Salgaonkar static int __init fadump_cma_init(void) { return 1; }
134a4e92ce8SMahesh Salgaonkar #endif /* CONFIG_CMA */
135a4e92ce8SMahesh Salgaonkar
136eb39c880SMahesh Salgaonkar /* Scan the Firmware Assisted dump configuration details. */
early_init_dt_scan_fw_dump(unsigned long node,const char * uname,int depth,void * data)137f3512011SHari Bathini int __init early_init_dt_scan_fw_dump(unsigned long node, const char *uname,
138f3512011SHari Bathini int depth, void *data)
139eb39c880SMahesh Salgaonkar {
14002c04e37SHari Bathini if (depth == 0) {
14102c04e37SHari Bathini early_init_dt_scan_reserved_ranges(node);
14202c04e37SHari Bathini return 0;
14302c04e37SHari Bathini }
14402c04e37SHari Bathini
14541df5928SHari Bathini if (depth != 1)
146eb39c880SMahesh Salgaonkar return 0;
147eb39c880SMahesh Salgaonkar
14841df5928SHari Bathini if (strcmp(uname, "rtas") == 0) {
149d3833a70SHari Bathini rtas_fadump_dt_scan(&fw_dump, node);
150eb39c880SMahesh Salgaonkar return 1;
151eb39c880SMahesh Salgaonkar }
152eb39c880SMahesh Salgaonkar
15341df5928SHari Bathini if (strcmp(uname, "ibm,opal") == 0) {
15441df5928SHari Bathini opal_fadump_dt_scan(&fw_dump, node);
15541df5928SHari Bathini return 1;
15641df5928SHari Bathini }
15741df5928SHari Bathini
15841df5928SHari Bathini return 0;
15941df5928SHari Bathini }
16041df5928SHari Bathini
161eae0dfccSHari Bathini /*
162eae0dfccSHari Bathini * If fadump is registered, check if the memory provided
1630db6896fSMahesh Salgaonkar * falls within boot memory area and reserved memory area.
164eae0dfccSHari Bathini */
is_fadump_memory_area(u64 addr,unsigned long size)165becd91d9SHari Bathini int is_fadump_memory_area(u64 addr, unsigned long size)
166eae0dfccSHari Bathini {
167becd91d9SHari Bathini u64 d_start, d_end;
1680db6896fSMahesh Salgaonkar
169eae0dfccSHari Bathini if (!fw_dump.dump_registered)
170eae0dfccSHari Bathini return 0;
171eae0dfccSHari Bathini
172becd91d9SHari Bathini if (!size)
173becd91d9SHari Bathini return 0;
174becd91d9SHari Bathini
175becd91d9SHari Bathini d_start = fw_dump.reserve_dump_area_start;
176becd91d9SHari Bathini d_end = d_start + fw_dump.reserve_dump_area_size;
1770db6896fSMahesh Salgaonkar if (((addr + size) > d_start) && (addr <= d_end))
1780db6896fSMahesh Salgaonkar return 1;
1790db6896fSMahesh Salgaonkar
1807dee93a9SHari Bathini return (addr <= fw_dump.boot_mem_top);
181eae0dfccSHari Bathini }
182eae0dfccSHari Bathini
should_fadump_crash(void)1836fcd6baaSNicholas Piggin int should_fadump_crash(void)
1846fcd6baaSNicholas Piggin {
1856fcd6baaSNicholas Piggin if (!fw_dump.dump_registered || !fw_dump.fadumphdr_addr)
1866fcd6baaSNicholas Piggin return 0;
1876fcd6baaSNicholas Piggin return 1;
1886fcd6baaSNicholas Piggin }
1896fcd6baaSNicholas Piggin
is_fadump_active(void)1903ccc00a7SMahesh Salgaonkar int is_fadump_active(void)
1913ccc00a7SMahesh Salgaonkar {
1923ccc00a7SMahesh Salgaonkar return fw_dump.dump_active;
1933ccc00a7SMahesh Salgaonkar }
1943ccc00a7SMahesh Salgaonkar
195a5a05b91SHari Bathini /*
196961cf26aSHari Bathini * Returns true, if there are no holes in memory area between d_start to d_end,
197961cf26aSHari Bathini * false otherwise.
198a5a05b91SHari Bathini */
is_fadump_mem_area_contiguous(u64 d_start,u64 d_end)199961cf26aSHari Bathini static bool is_fadump_mem_area_contiguous(u64 d_start, u64 d_end)
200a5a05b91SHari Bathini {
201b10d6bcaSMike Rapoport phys_addr_t reg_start, reg_end;
202961cf26aSHari Bathini bool ret = false;
203b10d6bcaSMike Rapoport u64 i, start, end;
204a5a05b91SHari Bathini
205b10d6bcaSMike Rapoport for_each_mem_range(i, ®_start, ®_end) {
206b10d6bcaSMike Rapoport start = max_t(u64, d_start, reg_start);
207b10d6bcaSMike Rapoport end = min_t(u64, d_end, reg_end);
208961cf26aSHari Bathini if (d_start < end) {
209961cf26aSHari Bathini /* Memory hole from d_start to start */
210961cf26aSHari Bathini if (start > d_start)
211a5a05b91SHari Bathini break;
212a5a05b91SHari Bathini
213961cf26aSHari Bathini if (end == d_end) {
214961cf26aSHari Bathini ret = true;
215a5a05b91SHari Bathini break;
216a5a05b91SHari Bathini }
217a5a05b91SHari Bathini
218961cf26aSHari Bathini d_start = end + 1;
219a5a05b91SHari Bathini }
220a5a05b91SHari Bathini }
221a5a05b91SHari Bathini
222a5a05b91SHari Bathini return ret;
223a5a05b91SHari Bathini }
224a5a05b91SHari Bathini
225f86593beSMahesh Salgaonkar /*
226961cf26aSHari Bathini * Returns true, if there are no holes in boot memory area,
227961cf26aSHari Bathini * false otherwise.
228961cf26aSHari Bathini */
is_fadump_boot_mem_contiguous(void)2297f0ad11dSHari Bathini bool is_fadump_boot_mem_contiguous(void)
230961cf26aSHari Bathini {
2317dee93a9SHari Bathini unsigned long d_start, d_end;
2327dee93a9SHari Bathini bool ret = false;
2337dee93a9SHari Bathini int i;
2347dee93a9SHari Bathini
2357dee93a9SHari Bathini for (i = 0; i < fw_dump.boot_mem_regs_cnt; i++) {
2367dee93a9SHari Bathini d_start = fw_dump.boot_mem_addr[i];
2377dee93a9SHari Bathini d_end = d_start + fw_dump.boot_mem_sz[i];
2387dee93a9SHari Bathini
2397dee93a9SHari Bathini ret = is_fadump_mem_area_contiguous(d_start, d_end);
2407dee93a9SHari Bathini if (!ret)
2417dee93a9SHari Bathini break;
2427dee93a9SHari Bathini }
2437dee93a9SHari Bathini
2447dee93a9SHari Bathini return ret;
245961cf26aSHari Bathini }
246961cf26aSHari Bathini
247961cf26aSHari Bathini /*
248f86593beSMahesh Salgaonkar * Returns true, if there are no holes in reserved memory area,
249f86593beSMahesh Salgaonkar * false otherwise.
250f86593beSMahesh Salgaonkar */
is_fadump_reserved_mem_contiguous(void)2517f0ad11dSHari Bathini bool is_fadump_reserved_mem_contiguous(void)
252f86593beSMahesh Salgaonkar {
253961cf26aSHari Bathini u64 d_start, d_end;
254f86593beSMahesh Salgaonkar
255961cf26aSHari Bathini d_start = fw_dump.reserve_dump_area_start;
256961cf26aSHari Bathini d_end = d_start + fw_dump.reserve_dump_area_size;
257961cf26aSHari Bathini return is_fadump_mem_area_contiguous(d_start, d_end);
258f86593beSMahesh Salgaonkar }
259f86593beSMahesh Salgaonkar
2603ccc00a7SMahesh Salgaonkar /* Print firmware assisted dump configurations for debugging purpose. */
fadump_show_config(void)261d276960dSNick Child static void __init fadump_show_config(void)
2623ccc00a7SMahesh Salgaonkar {
2637dee93a9SHari Bathini int i;
2647dee93a9SHari Bathini
2653ccc00a7SMahesh Salgaonkar pr_debug("Support for firmware-assisted dump (fadump): %s\n",
2663ccc00a7SMahesh Salgaonkar (fw_dump.fadump_supported ? "present" : "no support"));
2673ccc00a7SMahesh Salgaonkar
2683ccc00a7SMahesh Salgaonkar if (!fw_dump.fadump_supported)
2693ccc00a7SMahesh Salgaonkar return;
2703ccc00a7SMahesh Salgaonkar
2713ccc00a7SMahesh Salgaonkar pr_debug("Fadump enabled : %s\n",
2723ccc00a7SMahesh Salgaonkar (fw_dump.fadump_enabled ? "yes" : "no"));
2733ccc00a7SMahesh Salgaonkar pr_debug("Dump Active : %s\n",
2743ccc00a7SMahesh Salgaonkar (fw_dump.dump_active ? "yes" : "no"));
2753ccc00a7SMahesh Salgaonkar pr_debug("Dump section sizes:\n");
2763ccc00a7SMahesh Salgaonkar pr_debug(" CPU state data size: %lx\n", fw_dump.cpu_state_data_size);
2773ccc00a7SMahesh Salgaonkar pr_debug(" HPTE region size : %lx\n", fw_dump.hpte_region_size);
2783ccc00a7SMahesh Salgaonkar pr_debug(" Boot memory size : %lx\n", fw_dump.boot_memory_size);
2797dee93a9SHari Bathini pr_debug(" Boot memory top : %llx\n", fw_dump.boot_mem_top);
2807dee93a9SHari Bathini pr_debug("Boot memory regions cnt: %llx\n", fw_dump.boot_mem_regs_cnt);
2817dee93a9SHari Bathini for (i = 0; i < fw_dump.boot_mem_regs_cnt; i++) {
2827dee93a9SHari Bathini pr_debug("[%03d] base = %llx, size = %llx\n", i,
2837dee93a9SHari Bathini fw_dump.boot_mem_addr[i], fw_dump.boot_mem_sz[i]);
2847dee93a9SHari Bathini }
2853ccc00a7SMahesh Salgaonkar }
2863ccc00a7SMahesh Salgaonkar
287eb39c880SMahesh Salgaonkar /**
288eb39c880SMahesh Salgaonkar * fadump_calculate_reserve_size(): reserve variable boot area 5% of System RAM
289eb39c880SMahesh Salgaonkar *
290eb39c880SMahesh Salgaonkar * Function to find the largest memory size we need to reserve during early
291eb39c880SMahesh Salgaonkar * boot process. This will be the size of the memory that is required for a
292eb39c880SMahesh Salgaonkar * kernel to boot successfully.
293eb39c880SMahesh Salgaonkar *
294eb39c880SMahesh Salgaonkar * This function has been taken from phyp-assisted dump feature implementation.
295eb39c880SMahesh Salgaonkar *
296eb39c880SMahesh Salgaonkar * returns larger of 256MB or 5% rounded down to multiples of 256MB.
297eb39c880SMahesh Salgaonkar *
298eb39c880SMahesh Salgaonkar * TODO: Come up with better approach to find out more accurate memory size
299eb39c880SMahesh Salgaonkar * that is required for a kernel to boot successfully.
300eb39c880SMahesh Salgaonkar *
301eb39c880SMahesh Salgaonkar */
fadump_calculate_reserve_size(void)302fbced154SNathan Chancellor static __init u64 fadump_calculate_reserve_size(void)
303eb39c880SMahesh Salgaonkar {
3047b1b3b48SHari Bathini u64 base, size, bootmem_min;
30511550dc0SHari Bathini int ret;
306eb39c880SMahesh Salgaonkar
30781d9eca5SHari Bathini if (fw_dump.reserve_bootvar)
30881d9eca5SHari Bathini pr_warn("'fadump_reserve_mem=' parameter is deprecated in favor of 'crashkernel=' parameter.\n");
30981d9eca5SHari Bathini
310eb39c880SMahesh Salgaonkar /*
31111550dc0SHari Bathini * Check if the size is specified through crashkernel= cmdline
312e7467dc6SHari Bathini * option. If yes, then use that but ignore base as fadump reserves
313e7467dc6SHari Bathini * memory at a predefined offset.
314eb39c880SMahesh Salgaonkar */
31511550dc0SHari Bathini ret = parse_crashkernel(boot_command_line, memblock_phys_mem_size(),
31611550dc0SHari Bathini &size, &base);
31711550dc0SHari Bathini if (ret == 0 && size > 0) {
31848a316e3SHari Bathini unsigned long max_size;
31948a316e3SHari Bathini
32081d9eca5SHari Bathini if (fw_dump.reserve_bootvar)
32181d9eca5SHari Bathini pr_info("Using 'crashkernel=' parameter for memory reservation.\n");
32281d9eca5SHari Bathini
32311550dc0SHari Bathini fw_dump.reserve_bootvar = (unsigned long)size;
32448a316e3SHari Bathini
32548a316e3SHari Bathini /*
32648a316e3SHari Bathini * Adjust if the boot memory size specified is above
32748a316e3SHari Bathini * the upper limit.
32848a316e3SHari Bathini */
32948a316e3SHari Bathini max_size = memblock_phys_mem_size() / MAX_BOOT_MEM_RATIO;
33048a316e3SHari Bathini if (fw_dump.reserve_bootvar > max_size) {
33148a316e3SHari Bathini fw_dump.reserve_bootvar = max_size;
33248a316e3SHari Bathini pr_info("Adjusted boot memory size to %luMB\n",
33348a316e3SHari Bathini (fw_dump.reserve_bootvar >> 20));
33448a316e3SHari Bathini }
33548a316e3SHari Bathini
336eb39c880SMahesh Salgaonkar return fw_dump.reserve_bootvar;
33781d9eca5SHari Bathini } else if (fw_dump.reserve_bootvar) {
33881d9eca5SHari Bathini /*
33981d9eca5SHari Bathini * 'fadump_reserve_mem=' is being used to reserve memory
34081d9eca5SHari Bathini * for firmware-assisted dump.
34181d9eca5SHari Bathini */
34281d9eca5SHari Bathini return fw_dump.reserve_bootvar;
34311550dc0SHari Bathini }
344eb39c880SMahesh Salgaonkar
345eb39c880SMahesh Salgaonkar /* divide by 20 to get 5% of value */
34648a316e3SHari Bathini size = memblock_phys_mem_size() / 20;
347eb39c880SMahesh Salgaonkar
348eb39c880SMahesh Salgaonkar /* round it down in multiples of 256 */
349eb39c880SMahesh Salgaonkar size = size & ~0x0FFFFFFFUL;
350eb39c880SMahesh Salgaonkar
351eb39c880SMahesh Salgaonkar /* Truncate to memory_limit. We don't want to over reserve the memory.*/
352eb39c880SMahesh Salgaonkar if (memory_limit && size > memory_limit)
353eb39c880SMahesh Salgaonkar size = memory_limit;
354eb39c880SMahesh Salgaonkar
3557b1b3b48SHari Bathini bootmem_min = fw_dump.ops->fadump_get_bootmem_min();
3567b1b3b48SHari Bathini return (size > bootmem_min ? size : bootmem_min);
357eb39c880SMahesh Salgaonkar }
358eb39c880SMahesh Salgaonkar
359eb39c880SMahesh Salgaonkar /*
360eb39c880SMahesh Salgaonkar * Calculate the total memory size required to be reserved for
361eb39c880SMahesh Salgaonkar * firmware-assisted dump registration.
362eb39c880SMahesh Salgaonkar */
get_fadump_area_size(void)363d276960dSNick Child static unsigned long __init get_fadump_area_size(void)
364eb39c880SMahesh Salgaonkar {
365eb39c880SMahesh Salgaonkar unsigned long size = 0;
366eb39c880SMahesh Salgaonkar
367eb39c880SMahesh Salgaonkar size += fw_dump.cpu_state_data_size;
368eb39c880SMahesh Salgaonkar size += fw_dump.hpte_region_size;
3699cf3b3a3SHari Bathini /*
3709cf3b3a3SHari Bathini * Account for pagesize alignment of boot memory area destination address.
3719cf3b3a3SHari Bathini * This faciliates in mmap reading of first kernel's memory.
3729cf3b3a3SHari Bathini */
3739cf3b3a3SHari Bathini size = PAGE_ALIGN(size);
374eb39c880SMahesh Salgaonkar size += fw_dump.boot_memory_size;
3752df173d9SMahesh Salgaonkar size += sizeof(struct fadump_crash_info_header);
3762df173d9SMahesh Salgaonkar size += sizeof(struct elfhdr); /* ELF core header.*/
377ebaeb5aeSMahesh Salgaonkar size += sizeof(struct elf_phdr); /* place holder for cpu notes */
3782df173d9SMahesh Salgaonkar /* Program headers for crash memory regions. */
3792df173d9SMahesh Salgaonkar size += sizeof(struct elf_phdr) * (memblock_num_regions(memory) + 2);
380eb39c880SMahesh Salgaonkar
381eb39c880SMahesh Salgaonkar size = PAGE_ALIGN(size);
382742a265aSHari Bathini
383742a265aSHari Bathini /* This is to hold kernel metadata on platforms that support it */
384742a265aSHari Bathini size += (fw_dump.ops->fadump_get_metadata_size ?
385742a265aSHari Bathini fw_dump.ops->fadump_get_metadata_size() : 0);
386eb39c880SMahesh Salgaonkar return size;
387eb39c880SMahesh Salgaonkar }
388eb39c880SMahesh Salgaonkar
add_boot_mem_region(unsigned long rstart,unsigned long rsize)3897dee93a9SHari Bathini static int __init add_boot_mem_region(unsigned long rstart,
3907dee93a9SHari Bathini unsigned long rsize)
3917dee93a9SHari Bathini {
3927dee93a9SHari Bathini int i = fw_dump.boot_mem_regs_cnt++;
3937dee93a9SHari Bathini
3947dee93a9SHari Bathini if (fw_dump.boot_mem_regs_cnt > FADUMP_MAX_MEM_REGS) {
3957dee93a9SHari Bathini fw_dump.boot_mem_regs_cnt = FADUMP_MAX_MEM_REGS;
3967dee93a9SHari Bathini return 0;
3977dee93a9SHari Bathini }
3987dee93a9SHari Bathini
3997dee93a9SHari Bathini pr_debug("Added boot memory range[%d] [%#016lx-%#016lx)\n",
4007dee93a9SHari Bathini i, rstart, (rstart + rsize));
4017dee93a9SHari Bathini fw_dump.boot_mem_addr[i] = rstart;
4027dee93a9SHari Bathini fw_dump.boot_mem_sz[i] = rsize;
4037dee93a9SHari Bathini return 1;
4047dee93a9SHari Bathini }
4057dee93a9SHari Bathini
4067dee93a9SHari Bathini /*
4077dee93a9SHari Bathini * Firmware usually has a hard limit on the data it can copy per region.
4087dee93a9SHari Bathini * Honour that by splitting a memory range into multiple regions.
4097dee93a9SHari Bathini */
add_boot_mem_regions(unsigned long mstart,unsigned long msize)4107dee93a9SHari Bathini static int __init add_boot_mem_regions(unsigned long mstart,
4117dee93a9SHari Bathini unsigned long msize)
4127dee93a9SHari Bathini {
4137dee93a9SHari Bathini unsigned long rstart, rsize, max_size;
4147dee93a9SHari Bathini int ret = 1;
4157dee93a9SHari Bathini
4167dee93a9SHari Bathini rstart = mstart;
4177dee93a9SHari Bathini max_size = fw_dump.max_copy_size ? fw_dump.max_copy_size : msize;
4187dee93a9SHari Bathini while (msize) {
4197dee93a9SHari Bathini if (msize > max_size)
4207dee93a9SHari Bathini rsize = max_size;
4217dee93a9SHari Bathini else
4227dee93a9SHari Bathini rsize = msize;
4237dee93a9SHari Bathini
4247dee93a9SHari Bathini ret = add_boot_mem_region(rstart, rsize);
4257dee93a9SHari Bathini if (!ret)
4267dee93a9SHari Bathini break;
4277dee93a9SHari Bathini
4287dee93a9SHari Bathini msize -= rsize;
4297dee93a9SHari Bathini rstart += rsize;
4307dee93a9SHari Bathini }
4317dee93a9SHari Bathini
4327dee93a9SHari Bathini return ret;
4337dee93a9SHari Bathini }
4347dee93a9SHari Bathini
fadump_get_boot_mem_regions(void)4357dee93a9SHari Bathini static int __init fadump_get_boot_mem_regions(void)
4367dee93a9SHari Bathini {
437b10d6bcaSMike Rapoport unsigned long size, cur_size, hole_size, last_end;
4387dee93a9SHari Bathini unsigned long mem_size = fw_dump.boot_memory_size;
439b10d6bcaSMike Rapoport phys_addr_t reg_start, reg_end;
4407dee93a9SHari Bathini int ret = 1;
441b10d6bcaSMike Rapoport u64 i;
4427dee93a9SHari Bathini
4437dee93a9SHari Bathini fw_dump.boot_mem_regs_cnt = 0;
4447dee93a9SHari Bathini
4457dee93a9SHari Bathini last_end = 0;
4467dee93a9SHari Bathini hole_size = 0;
4477dee93a9SHari Bathini cur_size = 0;
448b10d6bcaSMike Rapoport for_each_mem_range(i, ®_start, ®_end) {
449b10d6bcaSMike Rapoport size = reg_end - reg_start;
450b10d6bcaSMike Rapoport hole_size += (reg_start - last_end);
4517dee93a9SHari Bathini
4527dee93a9SHari Bathini if ((cur_size + size) >= mem_size) {
4537dee93a9SHari Bathini size = (mem_size - cur_size);
454b10d6bcaSMike Rapoport ret = add_boot_mem_regions(reg_start, size);
4557dee93a9SHari Bathini break;
4567dee93a9SHari Bathini }
4577dee93a9SHari Bathini
4587dee93a9SHari Bathini mem_size -= size;
4597dee93a9SHari Bathini cur_size += size;
460b10d6bcaSMike Rapoport ret = add_boot_mem_regions(reg_start, size);
4617dee93a9SHari Bathini if (!ret)
4627dee93a9SHari Bathini break;
4637dee93a9SHari Bathini
464b10d6bcaSMike Rapoport last_end = reg_end;
4657dee93a9SHari Bathini }
4667dee93a9SHari Bathini fw_dump.boot_mem_top = PAGE_ALIGN(fw_dump.boot_memory_size + hole_size);
4677dee93a9SHari Bathini
4687dee93a9SHari Bathini return ret;
4697dee93a9SHari Bathini }
4707dee93a9SHari Bathini
471140777a3SHari Bathini /*
472140777a3SHari Bathini * Returns true, if the given range overlaps with reserved memory ranges
473140777a3SHari Bathini * starting at idx. Also, updates idx to index of overlapping memory range
474140777a3SHari Bathini * with the given memory range.
475140777a3SHari Bathini * False, otherwise.
476140777a3SHari Bathini */
overlaps_reserved_ranges(u64 base,u64 end,int * idx)477d276960dSNick Child static bool __init overlaps_reserved_ranges(u64 base, u64 end, int *idx)
478140777a3SHari Bathini {
479140777a3SHari Bathini bool ret = false;
480140777a3SHari Bathini int i;
481140777a3SHari Bathini
482140777a3SHari Bathini for (i = *idx; i < reserved_mrange_info.mem_range_cnt; i++) {
483140777a3SHari Bathini u64 rbase = reserved_mrange_info.mem_ranges[i].base;
484140777a3SHari Bathini u64 rend = rbase + reserved_mrange_info.mem_ranges[i].size;
485140777a3SHari Bathini
486140777a3SHari Bathini if (end <= rbase)
487140777a3SHari Bathini break;
488140777a3SHari Bathini
489140777a3SHari Bathini if ((end > rbase) && (base < rend)) {
490140777a3SHari Bathini *idx = i;
491140777a3SHari Bathini ret = true;
492140777a3SHari Bathini break;
493140777a3SHari Bathini }
494140777a3SHari Bathini }
495140777a3SHari Bathini
496140777a3SHari Bathini return ret;
497140777a3SHari Bathini }
498140777a3SHari Bathini
499140777a3SHari Bathini /*
500140777a3SHari Bathini * Locate a suitable memory area to reserve memory for FADump. While at it,
501140777a3SHari Bathini * lookup reserved-ranges & avoid overlap with them, as they are used by F/W.
502140777a3SHari Bathini */
fadump_locate_reserve_mem(u64 base,u64 size)503140777a3SHari Bathini static u64 __init fadump_locate_reserve_mem(u64 base, u64 size)
504140777a3SHari Bathini {
505140777a3SHari Bathini struct fadump_memory_range *mrngs;
506140777a3SHari Bathini phys_addr_t mstart, mend;
507140777a3SHari Bathini int idx = 0;
508140777a3SHari Bathini u64 i, ret = 0;
509140777a3SHari Bathini
510140777a3SHari Bathini mrngs = reserved_mrange_info.mem_ranges;
511140777a3SHari Bathini for_each_free_mem_range(i, NUMA_NO_NODE, MEMBLOCK_NONE,
512140777a3SHari Bathini &mstart, &mend, NULL) {
513140777a3SHari Bathini pr_debug("%llu) mstart: %llx, mend: %llx, base: %llx\n",
514140777a3SHari Bathini i, mstart, mend, base);
515140777a3SHari Bathini
516140777a3SHari Bathini if (mstart > base)
517140777a3SHari Bathini base = PAGE_ALIGN(mstart);
518140777a3SHari Bathini
519140777a3SHari Bathini while ((mend > base) && ((mend - base) >= size)) {
520140777a3SHari Bathini if (!overlaps_reserved_ranges(base, base+size, &idx)) {
521140777a3SHari Bathini ret = base;
522140777a3SHari Bathini goto out;
523140777a3SHari Bathini }
524140777a3SHari Bathini
525140777a3SHari Bathini base = mrngs[idx].base + mrngs[idx].size;
526140777a3SHari Bathini base = PAGE_ALIGN(base);
527140777a3SHari Bathini }
528140777a3SHari Bathini }
529140777a3SHari Bathini
530140777a3SHari Bathini out:
531140777a3SHari Bathini return ret;
532140777a3SHari Bathini }
533140777a3SHari Bathini
fadump_reserve_mem(void)534eb39c880SMahesh Salgaonkar int __init fadump_reserve_mem(void)
535eb39c880SMahesh Salgaonkar {
536140777a3SHari Bathini u64 base, size, mem_boundary, bootmem_min;
5376abec12cSHari Bathini int ret = 1;
538eb39c880SMahesh Salgaonkar
539eb39c880SMahesh Salgaonkar if (!fw_dump.fadump_enabled)
540eb39c880SMahesh Salgaonkar return 0;
541eb39c880SMahesh Salgaonkar
542eb39c880SMahesh Salgaonkar if (!fw_dump.fadump_supported) {
5436abec12cSHari Bathini pr_info("Firmware-Assisted Dump is not supported on this hardware\n");
5446abec12cSHari Bathini goto error_out;
545eb39c880SMahesh Salgaonkar }
546742a265aSHari Bathini
5473ccc00a7SMahesh Salgaonkar /*
5483ccc00a7SMahesh Salgaonkar * Initialize boot memory size
5493ccc00a7SMahesh Salgaonkar * If dump is active then we have already calculated the size during
5503ccc00a7SMahesh Salgaonkar * first kernel.
5513ccc00a7SMahesh Salgaonkar */
552f3512011SHari Bathini if (!fw_dump.dump_active) {
5536abec12cSHari Bathini fw_dump.boot_memory_size =
5546abec12cSHari Bathini PAGE_ALIGN(fadump_calculate_reserve_size());
555a4e92ce8SMahesh Salgaonkar #ifdef CONFIG_CMA
556579ca1a2SHari Bathini if (!fw_dump.nocma) {
557a4e92ce8SMahesh Salgaonkar fw_dump.boot_memory_size =
558140777a3SHari Bathini ALIGN(fw_dump.boot_memory_size,
559e16faf26SDavid Hildenbrand CMA_MIN_ALIGNMENT_BYTES);
560579ca1a2SHari Bathini }
561a4e92ce8SMahesh Salgaonkar #endif
5627b1b3b48SHari Bathini
5637b1b3b48SHari Bathini bootmem_min = fw_dump.ops->fadump_get_bootmem_min();
5647b1b3b48SHari Bathini if (fw_dump.boot_memory_size < bootmem_min) {
5657b1b3b48SHari Bathini pr_err("Can't enable fadump with boot memory size (0x%lx) less than 0x%llx\n",
5667b1b3b48SHari Bathini fw_dump.boot_memory_size, bootmem_min);
5677b1b3b48SHari Bathini goto error_out;
5687b1b3b48SHari Bathini }
5697dee93a9SHari Bathini
5707dee93a9SHari Bathini if (!fadump_get_boot_mem_regions()) {
5717dee93a9SHari Bathini pr_err("Too many holes in boot memory area to enable fadump\n");
5727dee93a9SHari Bathini goto error_out;
5737dee93a9SHari Bathini }
574a4e92ce8SMahesh Salgaonkar }
575eb39c880SMahesh Salgaonkar
576eb39c880SMahesh Salgaonkar /*
577eb39c880SMahesh Salgaonkar * Calculate the memory boundary.
578eb39c880SMahesh Salgaonkar * If memory_limit is less than actual memory boundary then reserve
579eb39c880SMahesh Salgaonkar * the memory for fadump beyond the memory_limit and adjust the
580eb39c880SMahesh Salgaonkar * memory_limit accordingly, so that the running kernel can run with
581eb39c880SMahesh Salgaonkar * specified memory_limit.
582eb39c880SMahesh Salgaonkar */
583eb39c880SMahesh Salgaonkar if (memory_limit && memory_limit < memblock_end_of_DRAM()) {
584eb39c880SMahesh Salgaonkar size = get_fadump_area_size();
585eb39c880SMahesh Salgaonkar if ((memory_limit + size) < memblock_end_of_DRAM())
586eb39c880SMahesh Salgaonkar memory_limit += size;
587eb39c880SMahesh Salgaonkar else
588eb39c880SMahesh Salgaonkar memory_limit = memblock_end_of_DRAM();
589eb39c880SMahesh Salgaonkar printk(KERN_INFO "Adjusted memory_limit for firmware-assisted"
590a84fcd46SSuzuki Poulose " dump, now %#016llx\n", memory_limit);
591eb39c880SMahesh Salgaonkar }
592eb39c880SMahesh Salgaonkar if (memory_limit)
5936abec12cSHari Bathini mem_boundary = memory_limit;
594eb39c880SMahesh Salgaonkar else
5956abec12cSHari Bathini mem_boundary = memblock_end_of_DRAM();
596eb39c880SMahesh Salgaonkar
5977dee93a9SHari Bathini base = fw_dump.boot_mem_top;
5988255da95SHari Bathini size = get_fadump_area_size();
5998255da95SHari Bathini fw_dump.reserve_dump_area_size = size;
600eb39c880SMahesh Salgaonkar if (fw_dump.dump_active) {
601b71a693dSMahesh Salgaonkar pr_info("Firmware-assisted dump is active.\n");
602b71a693dSMahesh Salgaonkar
60385975387SHari Bathini #ifdef CONFIG_HUGETLB_PAGE
60485975387SHari Bathini /*
60585975387SHari Bathini * FADump capture kernel doesn't care much about hugepages.
60685975387SHari Bathini * In fact, handling hugepages in capture kernel is asking for
60785975387SHari Bathini * trouble. So, disable HugeTLB support when fadump is active.
60885975387SHari Bathini */
60985975387SHari Bathini hugetlb_disabled = true;
61085975387SHari Bathini #endif
611eb39c880SMahesh Salgaonkar /*
612eb39c880SMahesh Salgaonkar * If last boot has crashed then reserve all the memory
613b2a815a5SHari Bathini * above boot memory size so that we don't touch it until
614eb39c880SMahesh Salgaonkar * dump is written to disk by userspace tool. This memory
615b2a815a5SHari Bathini * can be released for general use by invalidating fadump.
616eb39c880SMahesh Salgaonkar */
617b2a815a5SHari Bathini fadump_reserve_crash_area(base);
6182df173d9SMahesh Salgaonkar
619f3512011SHari Bathini pr_debug("fadumphdr_addr = %#016lx\n", fw_dump.fadumphdr_addr);
620f3512011SHari Bathini pr_debug("Reserve dump area start address: 0x%lx\n",
621f3512011SHari Bathini fw_dump.reserve_dump_area_start);
6228255da95SHari Bathini } else {
6238255da95SHari Bathini /*
624f6e6bedbSHari Bathini * Reserve memory at an offset closer to bottom of the RAM to
625579ca1a2SHari Bathini * minimize the impact of memory hot-remove operation.
626f6e6bedbSHari Bathini */
627140777a3SHari Bathini base = fadump_locate_reserve_mem(base, size);
628f6e6bedbSHari Bathini
6299a2921e5SHari Bathini if (!base || (base + size > mem_boundary)) {
630742a265aSHari Bathini pr_err("Failed to find memory chunk for reservation!\n");
631742a265aSHari Bathini goto error_out;
632742a265aSHari Bathini }
633742a265aSHari Bathini fw_dump.reserve_dump_area_start = base;
634742a265aSHari Bathini
635742a265aSHari Bathini /*
636742a265aSHari Bathini * Calculate the kernel metadata address and register it with
637742a265aSHari Bathini * f/w if the platform supports.
638742a265aSHari Bathini */
639742a265aSHari Bathini if (fw_dump.ops->fadump_setup_metadata &&
640742a265aSHari Bathini (fw_dump.ops->fadump_setup_metadata(&fw_dump) < 0))
641742a265aSHari Bathini goto error_out;
642742a265aSHari Bathini
643742a265aSHari Bathini if (memblock_reserve(base, size)) {
6446abec12cSHari Bathini pr_err("Failed to reserve memory!\n");
6456abec12cSHari Bathini goto error_out;
6466abec12cSHari Bathini }
6476abec12cSHari Bathini
6486abec12cSHari Bathini pr_info("Reserved %lldMB of memory at %#016llx (System RAM: %lldMB)\n",
6496abec12cSHari Bathini (size >> 20), base, (memblock_phys_mem_size() >> 20));
650f6e6bedbSHari Bathini
6516abec12cSHari Bathini ret = fadump_cma_init();
652a4e92ce8SMahesh Salgaonkar }
6536abec12cSHari Bathini
6546abec12cSHari Bathini return ret;
6556abec12cSHari Bathini error_out:
6566abec12cSHari Bathini fw_dump.fadump_enabled = 0;
657*d1eb75e0SSourabh Jain fw_dump.reserve_dump_area_size = 0;
6586abec12cSHari Bathini return 0;
659eb39c880SMahesh Salgaonkar }
660eb39c880SMahesh Salgaonkar
661eb39c880SMahesh Salgaonkar /* Look for fadump= cmdline option. */
early_fadump_param(char * p)662eb39c880SMahesh Salgaonkar static int __init early_fadump_param(char *p)
663eb39c880SMahesh Salgaonkar {
664eb39c880SMahesh Salgaonkar if (!p)
665eb39c880SMahesh Salgaonkar return 1;
666eb39c880SMahesh Salgaonkar
667eb39c880SMahesh Salgaonkar if (strncmp(p, "on", 2) == 0)
668eb39c880SMahesh Salgaonkar fw_dump.fadump_enabled = 1;
669eb39c880SMahesh Salgaonkar else if (strncmp(p, "off", 3) == 0)
670eb39c880SMahesh Salgaonkar fw_dump.fadump_enabled = 0;
671a4e92ce8SMahesh Salgaonkar else if (strncmp(p, "nocma", 5) == 0) {
672a4e92ce8SMahesh Salgaonkar fw_dump.fadump_enabled = 1;
673a4e92ce8SMahesh Salgaonkar fw_dump.nocma = 1;
674a4e92ce8SMahesh Salgaonkar }
675eb39c880SMahesh Salgaonkar
676eb39c880SMahesh Salgaonkar return 0;
677eb39c880SMahesh Salgaonkar }
678eb39c880SMahesh Salgaonkar early_param("fadump", early_fadump_param);
679eb39c880SMahesh Salgaonkar
68081d9eca5SHari Bathini /*
68181d9eca5SHari Bathini * Look for fadump_reserve_mem= cmdline option
68281d9eca5SHari Bathini * TODO: Remove references to 'fadump_reserve_mem=' parameter,
68381d9eca5SHari Bathini * the sooner 'crashkernel=' parameter is accustomed to.
68481d9eca5SHari Bathini */
early_fadump_reserve_mem(char * p)68581d9eca5SHari Bathini static int __init early_fadump_reserve_mem(char *p)
68681d9eca5SHari Bathini {
68781d9eca5SHari Bathini if (p)
68881d9eca5SHari Bathini fw_dump.reserve_bootvar = memparse(p, &p);
68981d9eca5SHari Bathini return 0;
69081d9eca5SHari Bathini }
69181d9eca5SHari Bathini early_param("fadump_reserve_mem", early_fadump_reserve_mem);
69281d9eca5SHari Bathini
crash_fadump(struct pt_regs * regs,const char * str)693ebaeb5aeSMahesh Salgaonkar void crash_fadump(struct pt_regs *regs, const char *str)
694ebaeb5aeSMahesh Salgaonkar {
695ba608c4fSSourabh Jain unsigned int msecs;
696ebaeb5aeSMahesh Salgaonkar struct fadump_crash_info_header *fdh = NULL;
697f2a5e8f0SMahesh Salgaonkar int old_cpu, this_cpu;
698ba608c4fSSourabh Jain /* Do not include first CPU */
699ba608c4fSSourabh Jain unsigned int ncpus = num_online_cpus() - 1;
700ebaeb5aeSMahesh Salgaonkar
7016fcd6baaSNicholas Piggin if (!should_fadump_crash())
702ebaeb5aeSMahesh Salgaonkar return;
703ebaeb5aeSMahesh Salgaonkar
704f2a5e8f0SMahesh Salgaonkar /*
705f2a5e8f0SMahesh Salgaonkar * old_cpu == -1 means this is the first CPU which has come here,
706f2a5e8f0SMahesh Salgaonkar * go ahead and trigger fadump.
707f2a5e8f0SMahesh Salgaonkar *
708f2a5e8f0SMahesh Salgaonkar * old_cpu != -1 means some other CPU has already on it's way
709f2a5e8f0SMahesh Salgaonkar * to trigger fadump, just keep looping here.
710f2a5e8f0SMahesh Salgaonkar */
711f2a5e8f0SMahesh Salgaonkar this_cpu = smp_processor_id();
712f2a5e8f0SMahesh Salgaonkar old_cpu = cmpxchg(&crashing_cpu, -1, this_cpu);
713f2a5e8f0SMahesh Salgaonkar
714f2a5e8f0SMahesh Salgaonkar if (old_cpu != -1) {
715ba608c4fSSourabh Jain atomic_inc(&cpus_in_fadump);
716ba608c4fSSourabh Jain
717f2a5e8f0SMahesh Salgaonkar /*
718f2a5e8f0SMahesh Salgaonkar * We can't loop here indefinitely. Wait as long as fadump
719f2a5e8f0SMahesh Salgaonkar * is in force. If we race with fadump un-registration this
720f2a5e8f0SMahesh Salgaonkar * loop will break and then we go down to normal panic path
721f2a5e8f0SMahesh Salgaonkar * and reboot. If fadump is in force the first crashing
722f2a5e8f0SMahesh Salgaonkar * cpu will definitely trigger fadump.
723f2a5e8f0SMahesh Salgaonkar */
724f2a5e8f0SMahesh Salgaonkar while (fw_dump.dump_registered)
725f2a5e8f0SMahesh Salgaonkar cpu_relax();
726f2a5e8f0SMahesh Salgaonkar return;
727f2a5e8f0SMahesh Salgaonkar }
728f2a5e8f0SMahesh Salgaonkar
729ebaeb5aeSMahesh Salgaonkar fdh = __va(fw_dump.fadumphdr_addr);
730ebaeb5aeSMahesh Salgaonkar fdh->crashing_cpu = crashing_cpu;
731ebaeb5aeSMahesh Salgaonkar crash_save_vmcoreinfo();
732ebaeb5aeSMahesh Salgaonkar
733ebaeb5aeSMahesh Salgaonkar if (regs)
734ebaeb5aeSMahesh Salgaonkar fdh->regs = *regs;
735ebaeb5aeSMahesh Salgaonkar else
736ebaeb5aeSMahesh Salgaonkar ppc_save_regs(&fdh->regs);
737ebaeb5aeSMahesh Salgaonkar
7386584cec0SHari Bathini fdh->cpu_mask = *cpu_online_mask;
739ebaeb5aeSMahesh Salgaonkar
740ba608c4fSSourabh Jain /*
741ba608c4fSSourabh Jain * If we came in via system reset, wait a while for the secondary
742ba608c4fSSourabh Jain * CPUs to enter.
743ba608c4fSSourabh Jain */
7447153d4bfSXiongwei Song if (TRAP(&(fdh->regs)) == INTERRUPT_SYSTEM_RESET) {
745ba608c4fSSourabh Jain msecs = CRASH_TIMEOUT;
746ba608c4fSSourabh Jain while ((atomic_read(&cpus_in_fadump) < ncpus) && (--msecs > 0))
747ba608c4fSSourabh Jain mdelay(1);
748ba608c4fSSourabh Jain }
749ba608c4fSSourabh Jain
75041a65d16SHari Bathini fw_dump.ops->fadump_trigger(fdh, str);
751ebaeb5aeSMahesh Salgaonkar }
752ebaeb5aeSMahesh Salgaonkar
fadump_regs_to_elf_notes(u32 * buf,struct pt_regs * regs)753d276960dSNick Child u32 *__init fadump_regs_to_elf_notes(u32 *buf, struct pt_regs *regs)
754ebaeb5aeSMahesh Salgaonkar {
755ebaeb5aeSMahesh Salgaonkar struct elf_prstatus prstatus;
756ebaeb5aeSMahesh Salgaonkar
757ebaeb5aeSMahesh Salgaonkar memset(&prstatus, 0, sizeof(prstatus));
758ebaeb5aeSMahesh Salgaonkar /*
759ebaeb5aeSMahesh Salgaonkar * FIXME: How do i get PID? Do I really need it?
760ebaeb5aeSMahesh Salgaonkar * prstatus.pr_pid = ????
761ebaeb5aeSMahesh Salgaonkar */
7629554e908SBrian Gerst elf_core_copy_regs(&prstatus.pr_reg, regs);
76322bd0177SHari Bathini buf = append_elf_note(buf, CRASH_CORE_NOTE_NAME, NT_PRSTATUS,
764ebaeb5aeSMahesh Salgaonkar &prstatus, sizeof(prstatus));
765ebaeb5aeSMahesh Salgaonkar return buf;
766ebaeb5aeSMahesh Salgaonkar }
767ebaeb5aeSMahesh Salgaonkar
fadump_update_elfcore_header(char * bufp)768d276960dSNick Child void __init fadump_update_elfcore_header(char *bufp)
769ebaeb5aeSMahesh Salgaonkar {
770ebaeb5aeSMahesh Salgaonkar struct elf_phdr *phdr;
771ebaeb5aeSMahesh Salgaonkar
772ebaeb5aeSMahesh Salgaonkar bufp += sizeof(struct elfhdr);
773ebaeb5aeSMahesh Salgaonkar
774ebaeb5aeSMahesh Salgaonkar /* First note is a place holder for cpu notes info. */
775ebaeb5aeSMahesh Salgaonkar phdr = (struct elf_phdr *)bufp;
776ebaeb5aeSMahesh Salgaonkar
777ebaeb5aeSMahesh Salgaonkar if (phdr->p_type == PT_NOTE) {
778961cf26aSHari Bathini phdr->p_paddr = __pa(fw_dump.cpu_notes_buf_vaddr);
779ebaeb5aeSMahesh Salgaonkar phdr->p_offset = phdr->p_paddr;
780ebaeb5aeSMahesh Salgaonkar phdr->p_filesz = fw_dump.cpu_notes_buf_size;
781ebaeb5aeSMahesh Salgaonkar phdr->p_memsz = fw_dump.cpu_notes_buf_size;
782ebaeb5aeSMahesh Salgaonkar }
783ebaeb5aeSMahesh Salgaonkar return;
784ebaeb5aeSMahesh Salgaonkar }
785ebaeb5aeSMahesh Salgaonkar
fadump_alloc_buffer(unsigned long size)786d276960dSNick Child static void *__init fadump_alloc_buffer(unsigned long size)
787ebaeb5aeSMahesh Salgaonkar {
78872aa6517SHari Bathini unsigned long count, i;
789ebaeb5aeSMahesh Salgaonkar struct page *page;
79072aa6517SHari Bathini void *vaddr;
791ebaeb5aeSMahesh Salgaonkar
79272aa6517SHari Bathini vaddr = alloc_pages_exact(size, GFP_KERNEL | __GFP_ZERO);
793ebaeb5aeSMahesh Salgaonkar if (!vaddr)
794ebaeb5aeSMahesh Salgaonkar return NULL;
795ebaeb5aeSMahesh Salgaonkar
79672aa6517SHari Bathini count = PAGE_ALIGN(size) / PAGE_SIZE;
797ebaeb5aeSMahesh Salgaonkar page = virt_to_page(vaddr);
798ebaeb5aeSMahesh Salgaonkar for (i = 0; i < count; i++)
79972aa6517SHari Bathini mark_page_reserved(page + i);
800ebaeb5aeSMahesh Salgaonkar return vaddr;
801ebaeb5aeSMahesh Salgaonkar }
802ebaeb5aeSMahesh Salgaonkar
fadump_free_buffer(unsigned long vaddr,unsigned long size)803961cf26aSHari Bathini static void fadump_free_buffer(unsigned long vaddr, unsigned long size)
804ebaeb5aeSMahesh Salgaonkar {
80572aa6517SHari Bathini free_reserved_area((void *)vaddr, (void *)(vaddr + size), -1, NULL);
806ebaeb5aeSMahesh Salgaonkar }
807ebaeb5aeSMahesh Salgaonkar
fadump_setup_cpu_notes_buf(u32 num_cpus)808d276960dSNick Child s32 __init fadump_setup_cpu_notes_buf(u32 num_cpus)
809961cf26aSHari Bathini {
810961cf26aSHari Bathini /* Allocate buffer to hold cpu crash notes. */
811961cf26aSHari Bathini fw_dump.cpu_notes_buf_size = num_cpus * sizeof(note_buf_t);
812961cf26aSHari Bathini fw_dump.cpu_notes_buf_size = PAGE_ALIGN(fw_dump.cpu_notes_buf_size);
813961cf26aSHari Bathini fw_dump.cpu_notes_buf_vaddr =
814961cf26aSHari Bathini (unsigned long)fadump_alloc_buffer(fw_dump.cpu_notes_buf_size);
815961cf26aSHari Bathini if (!fw_dump.cpu_notes_buf_vaddr) {
816961cf26aSHari Bathini pr_err("Failed to allocate %ld bytes for CPU notes buffer\n",
817961cf26aSHari Bathini fw_dump.cpu_notes_buf_size);
818961cf26aSHari Bathini return -ENOMEM;
819961cf26aSHari Bathini }
820961cf26aSHari Bathini
821961cf26aSHari Bathini pr_debug("Allocated buffer for cpu notes of size %ld at 0x%lx\n",
822961cf26aSHari Bathini fw_dump.cpu_notes_buf_size,
823961cf26aSHari Bathini fw_dump.cpu_notes_buf_vaddr);
824961cf26aSHari Bathini return 0;
825961cf26aSHari Bathini }
826961cf26aSHari Bathini
fadump_free_cpu_notes_buf(void)8277f0ad11dSHari Bathini void fadump_free_cpu_notes_buf(void)
828961cf26aSHari Bathini {
829961cf26aSHari Bathini if (!fw_dump.cpu_notes_buf_vaddr)
830961cf26aSHari Bathini return;
831961cf26aSHari Bathini
832961cf26aSHari Bathini fadump_free_buffer(fw_dump.cpu_notes_buf_vaddr,
833961cf26aSHari Bathini fw_dump.cpu_notes_buf_size);
834961cf26aSHari Bathini fw_dump.cpu_notes_buf_vaddr = 0;
835961cf26aSHari Bathini fw_dump.cpu_notes_buf_size = 0;
836961cf26aSHari Bathini }
837961cf26aSHari Bathini
fadump_free_mem_ranges(struct fadump_mrange_info * mrange_info)838e4fc48fbSHari Bathini static void fadump_free_mem_ranges(struct fadump_mrange_info *mrange_info)
8391bd6a1c4SHari Bathini {
84002c04e37SHari Bathini if (mrange_info->is_static) {
84102c04e37SHari Bathini mrange_info->mem_range_cnt = 0;
84202c04e37SHari Bathini return;
84302c04e37SHari Bathini }
84402c04e37SHari Bathini
845e4fc48fbSHari Bathini kfree(mrange_info->mem_ranges);
84602c04e37SHari Bathini memset((void *)((u64)mrange_info + RNG_NAME_SZ), 0,
84702c04e37SHari Bathini (sizeof(struct fadump_mrange_info) - RNG_NAME_SZ));
8481bd6a1c4SHari Bathini }
8491bd6a1c4SHari Bathini
8501bd6a1c4SHari Bathini /*
851e4fc48fbSHari Bathini * Allocate or reallocate mem_ranges array in incremental units
8521bd6a1c4SHari Bathini * of PAGE_SIZE.
8531bd6a1c4SHari Bathini */
fadump_alloc_mem_ranges(struct fadump_mrange_info * mrange_info)854e4fc48fbSHari Bathini static int fadump_alloc_mem_ranges(struct fadump_mrange_info *mrange_info)
8551bd6a1c4SHari Bathini {
856e4fc48fbSHari Bathini struct fadump_memory_range *new_array;
8571bd6a1c4SHari Bathini u64 new_size;
8581bd6a1c4SHari Bathini
859e4fc48fbSHari Bathini new_size = mrange_info->mem_ranges_sz + PAGE_SIZE;
860e4fc48fbSHari Bathini pr_debug("Allocating %llu bytes of memory for %s memory ranges\n",
861e4fc48fbSHari Bathini new_size, mrange_info->name);
8621bd6a1c4SHari Bathini
863e4fc48fbSHari Bathini new_array = krealloc(mrange_info->mem_ranges, new_size, GFP_KERNEL);
8641bd6a1c4SHari Bathini if (new_array == NULL) {
865e4fc48fbSHari Bathini pr_err("Insufficient memory for setting up %s memory ranges\n",
866e4fc48fbSHari Bathini mrange_info->name);
867e4fc48fbSHari Bathini fadump_free_mem_ranges(mrange_info);
8681bd6a1c4SHari Bathini return -ENOMEM;
8691bd6a1c4SHari Bathini }
8701bd6a1c4SHari Bathini
871e4fc48fbSHari Bathini mrange_info->mem_ranges = new_array;
872e4fc48fbSHari Bathini mrange_info->mem_ranges_sz = new_size;
873e4fc48fbSHari Bathini mrange_info->max_mem_ranges = (new_size /
874e4fc48fbSHari Bathini sizeof(struct fadump_memory_range));
8751bd6a1c4SHari Bathini return 0;
8761bd6a1c4SHari Bathini }
fadump_add_mem_range(struct fadump_mrange_info * mrange_info,u64 base,u64 end)877e4fc48fbSHari Bathini static inline int fadump_add_mem_range(struct fadump_mrange_info *mrange_info,
878e4fc48fbSHari Bathini u64 base, u64 end)
8792df173d9SMahesh Salgaonkar {
880e4fc48fbSHari Bathini struct fadump_memory_range *mem_ranges = mrange_info->mem_ranges;
881ced1bf52SHari Bathini bool is_adjacent = false;
882e4fc48fbSHari Bathini u64 start, size;
883ced1bf52SHari Bathini
8842df173d9SMahesh Salgaonkar if (base == end)
8851bd6a1c4SHari Bathini return 0;
8861bd6a1c4SHari Bathini
887ced1bf52SHari Bathini /*
888ced1bf52SHari Bathini * Fold adjacent memory ranges to bring down the memory ranges/
889ced1bf52SHari Bathini * PT_LOAD segments count.
890ced1bf52SHari Bathini */
891e4fc48fbSHari Bathini if (mrange_info->mem_range_cnt) {
892e4fc48fbSHari Bathini start = mem_ranges[mrange_info->mem_range_cnt - 1].base;
893e4fc48fbSHari Bathini size = mem_ranges[mrange_info->mem_range_cnt - 1].size;
894ced1bf52SHari Bathini
89515eb77f8SHari Bathini /*
89615eb77f8SHari Bathini * Boot memory area needs separate PT_LOAD segment(s) as it
89715eb77f8SHari Bathini * is moved to a different location at the time of crash.
89815eb77f8SHari Bathini * So, fold only if the region is not boot memory area.
89915eb77f8SHari Bathini */
90015eb77f8SHari Bathini if ((start + size) == base && start >= fw_dump.boot_mem_top)
901ced1bf52SHari Bathini is_adjacent = true;
902ced1bf52SHari Bathini }
903ced1bf52SHari Bathini if (!is_adjacent) {
904ced1bf52SHari Bathini /* resize the array on reaching the limit */
905e4fc48fbSHari Bathini if (mrange_info->mem_range_cnt == mrange_info->max_mem_ranges) {
9061bd6a1c4SHari Bathini int ret;
9071bd6a1c4SHari Bathini
90802c04e37SHari Bathini if (mrange_info->is_static) {
90902c04e37SHari Bathini pr_err("Reached array size limit for %s memory ranges\n",
91002c04e37SHari Bathini mrange_info->name);
91102c04e37SHari Bathini return -ENOSPC;
91202c04e37SHari Bathini }
91302c04e37SHari Bathini
914e4fc48fbSHari Bathini ret = fadump_alloc_mem_ranges(mrange_info);
9151bd6a1c4SHari Bathini if (ret)
9161bd6a1c4SHari Bathini return ret;
917e4fc48fbSHari Bathini
918e4fc48fbSHari Bathini /* Update to the new resized array */
919e4fc48fbSHari Bathini mem_ranges = mrange_info->mem_ranges;
9201bd6a1c4SHari Bathini }
9212df173d9SMahesh Salgaonkar
922ced1bf52SHari Bathini start = base;
923e4fc48fbSHari Bathini mem_ranges[mrange_info->mem_range_cnt].base = start;
924e4fc48fbSHari Bathini mrange_info->mem_range_cnt++;
925ced1bf52SHari Bathini }
926ced1bf52SHari Bathini
927e4fc48fbSHari Bathini mem_ranges[mrange_info->mem_range_cnt - 1].size = (end - start);
928e4fc48fbSHari Bathini pr_debug("%s_memory_range[%d] [%#016llx-%#016llx], %#llx bytes\n",
929e4fc48fbSHari Bathini mrange_info->name, (mrange_info->mem_range_cnt - 1),
930e4fc48fbSHari Bathini start, end - 1, (end - start));
9311bd6a1c4SHari Bathini return 0;
9322df173d9SMahesh Salgaonkar }
9332df173d9SMahesh Salgaonkar
fadump_exclude_reserved_area(u64 start,u64 end)934e4fc48fbSHari Bathini static int fadump_exclude_reserved_area(u64 start, u64 end)
9352df173d9SMahesh Salgaonkar {
936e4fc48fbSHari Bathini u64 ra_start, ra_end;
9371bd6a1c4SHari Bathini int ret = 0;
9382df173d9SMahesh Salgaonkar
9392df173d9SMahesh Salgaonkar ra_start = fw_dump.reserve_dump_area_start;
9402df173d9SMahesh Salgaonkar ra_end = ra_start + fw_dump.reserve_dump_area_size;
9412df173d9SMahesh Salgaonkar
9422df173d9SMahesh Salgaonkar if ((ra_start < end) && (ra_end > start)) {
9432df173d9SMahesh Salgaonkar if ((start < ra_start) && (end > ra_end)) {
944e4fc48fbSHari Bathini ret = fadump_add_mem_range(&crash_mrange_info,
945e4fc48fbSHari Bathini start, ra_start);
9461bd6a1c4SHari Bathini if (ret)
9471bd6a1c4SHari Bathini return ret;
9481bd6a1c4SHari Bathini
949e4fc48fbSHari Bathini ret = fadump_add_mem_range(&crash_mrange_info,
950e4fc48fbSHari Bathini ra_end, end);
9512df173d9SMahesh Salgaonkar } else if (start < ra_start) {
952e4fc48fbSHari Bathini ret = fadump_add_mem_range(&crash_mrange_info,
953e4fc48fbSHari Bathini start, ra_start);
9542df173d9SMahesh Salgaonkar } else if (ra_end < end) {
955e4fc48fbSHari Bathini ret = fadump_add_mem_range(&crash_mrange_info,
956e4fc48fbSHari Bathini ra_end, end);
9572df173d9SMahesh Salgaonkar }
9582df173d9SMahesh Salgaonkar } else
959e4fc48fbSHari Bathini ret = fadump_add_mem_range(&crash_mrange_info, start, end);
9601bd6a1c4SHari Bathini
9611bd6a1c4SHari Bathini return ret;
9622df173d9SMahesh Salgaonkar }
9632df173d9SMahesh Salgaonkar
fadump_init_elfcore_header(char * bufp)9642df173d9SMahesh Salgaonkar static int fadump_init_elfcore_header(char *bufp)
9652df173d9SMahesh Salgaonkar {
9662df173d9SMahesh Salgaonkar struct elfhdr *elf;
9672df173d9SMahesh Salgaonkar
9682df173d9SMahesh Salgaonkar elf = (struct elfhdr *) bufp;
9692df173d9SMahesh Salgaonkar bufp += sizeof(struct elfhdr);
9702df173d9SMahesh Salgaonkar memcpy(elf->e_ident, ELFMAG, SELFMAG);
9712df173d9SMahesh Salgaonkar elf->e_ident[EI_CLASS] = ELF_CLASS;
9722df173d9SMahesh Salgaonkar elf->e_ident[EI_DATA] = ELF_DATA;
9732df173d9SMahesh Salgaonkar elf->e_ident[EI_VERSION] = EV_CURRENT;
9742df173d9SMahesh Salgaonkar elf->e_ident[EI_OSABI] = ELF_OSABI;
9752df173d9SMahesh Salgaonkar memset(elf->e_ident+EI_PAD, 0, EI_NIDENT-EI_PAD);
9762df173d9SMahesh Salgaonkar elf->e_type = ET_CORE;
9772df173d9SMahesh Salgaonkar elf->e_machine = ELF_ARCH;
9782df173d9SMahesh Salgaonkar elf->e_version = EV_CURRENT;
9792df173d9SMahesh Salgaonkar elf->e_entry = 0;
9802df173d9SMahesh Salgaonkar elf->e_phoff = sizeof(struct elfhdr);
9812df173d9SMahesh Salgaonkar elf->e_shoff = 0;
9825b89492cSChristophe Leroy
9835b89492cSChristophe Leroy if (IS_ENABLED(CONFIG_PPC64_ELF_ABI_V2))
9845b89492cSChristophe Leroy elf->e_flags = 2;
9855b89492cSChristophe Leroy else if (IS_ENABLED(CONFIG_PPC64_ELF_ABI_V1))
9865b89492cSChristophe Leroy elf->e_flags = 1;
9875b89492cSChristophe Leroy else
988d8bced27SDaniel Axtens elf->e_flags = 0;
9895b89492cSChristophe Leroy
9902df173d9SMahesh Salgaonkar elf->e_ehsize = sizeof(struct elfhdr);
9912df173d9SMahesh Salgaonkar elf->e_phentsize = sizeof(struct elf_phdr);
9922df173d9SMahesh Salgaonkar elf->e_phnum = 0;
9932df173d9SMahesh Salgaonkar elf->e_shentsize = 0;
9942df173d9SMahesh Salgaonkar elf->e_shnum = 0;
9952df173d9SMahesh Salgaonkar elf->e_shstrndx = 0;
9962df173d9SMahesh Salgaonkar
9972df173d9SMahesh Salgaonkar return 0;
9982df173d9SMahesh Salgaonkar }
9992df173d9SMahesh Salgaonkar
10002df173d9SMahesh Salgaonkar /*
10012df173d9SMahesh Salgaonkar * Traverse through memblock structure and setup crash memory ranges. These
10022df173d9SMahesh Salgaonkar * ranges will be used create PT_LOAD program headers in elfcore header.
10032df173d9SMahesh Salgaonkar */
fadump_setup_crash_memory_ranges(void)10041bd6a1c4SHari Bathini static int fadump_setup_crash_memory_ranges(void)
10052df173d9SMahesh Salgaonkar {
1006b10d6bcaSMike Rapoport u64 i, start, end;
1007b10d6bcaSMike Rapoport int ret;
10082df173d9SMahesh Salgaonkar
10092df173d9SMahesh Salgaonkar pr_debug("Setup crash memory ranges.\n");
1010e4fc48fbSHari Bathini crash_mrange_info.mem_range_cnt = 0;
1011ced1bf52SHari Bathini
10122df173d9SMahesh Salgaonkar /*
10137dee93a9SHari Bathini * Boot memory region(s) registered with firmware are moved to
10147dee93a9SHari Bathini * different location at the time of crash. Create separate program
10157dee93a9SHari Bathini * header(s) for this memory chunk(s) with the correct offset.
10162df173d9SMahesh Salgaonkar */
10177dee93a9SHari Bathini for (i = 0; i < fw_dump.boot_mem_regs_cnt; i++) {
10187dee93a9SHari Bathini start = fw_dump.boot_mem_addr[i];
10197dee93a9SHari Bathini end = start + fw_dump.boot_mem_sz[i];
10207dee93a9SHari Bathini ret = fadump_add_mem_range(&crash_mrange_info, start, end);
10211bd6a1c4SHari Bathini if (ret)
10221bd6a1c4SHari Bathini return ret;
10237dee93a9SHari Bathini }
10242df173d9SMahesh Salgaonkar
1025b10d6bcaSMike Rapoport for_each_mem_range(i, &start, &end) {
1026a77af552SHari Bathini /*
10277dee93a9SHari Bathini * skip the memory chunk that is already added
10287dee93a9SHari Bathini * (0 through boot_memory_top).
1029a77af552SHari Bathini */
10307dee93a9SHari Bathini if (start < fw_dump.boot_mem_top) {
10317dee93a9SHari Bathini if (end > fw_dump.boot_mem_top)
10327dee93a9SHari Bathini start = fw_dump.boot_mem_top;
1033a77af552SHari Bathini else
1034a77af552SHari Bathini continue;
1035a77af552SHari Bathini }
10362df173d9SMahesh Salgaonkar
10372df173d9SMahesh Salgaonkar /* add this range excluding the reserved dump area. */
10381bd6a1c4SHari Bathini ret = fadump_exclude_reserved_area(start, end);
10391bd6a1c4SHari Bathini if (ret)
10401bd6a1c4SHari Bathini return ret;
10412df173d9SMahesh Salgaonkar }
10421bd6a1c4SHari Bathini
10431bd6a1c4SHari Bathini return 0;
10442df173d9SMahesh Salgaonkar }
10452df173d9SMahesh Salgaonkar
1046d34c5f26SMahesh Salgaonkar /*
1047d34c5f26SMahesh Salgaonkar * If the given physical address falls within the boot memory region then
1048d34c5f26SMahesh Salgaonkar * return the relocated address that points to the dump region reserved
1049d34c5f26SMahesh Salgaonkar * for saving initial boot memory contents.
1050d34c5f26SMahesh Salgaonkar */
fadump_relocate(unsigned long paddr)1051d34c5f26SMahesh Salgaonkar static inline unsigned long fadump_relocate(unsigned long paddr)
1052d34c5f26SMahesh Salgaonkar {
10537dee93a9SHari Bathini unsigned long raddr, rstart, rend, rlast, hole_size;
10547dee93a9SHari Bathini int i;
10557dee93a9SHari Bathini
10567dee93a9SHari Bathini hole_size = 0;
10577dee93a9SHari Bathini rlast = 0;
10587dee93a9SHari Bathini raddr = paddr;
10597dee93a9SHari Bathini for (i = 0; i < fw_dump.boot_mem_regs_cnt; i++) {
10607dee93a9SHari Bathini rstart = fw_dump.boot_mem_addr[i];
10617dee93a9SHari Bathini rend = rstart + fw_dump.boot_mem_sz[i];
10627dee93a9SHari Bathini hole_size += (rstart - rlast);
10637dee93a9SHari Bathini
10647dee93a9SHari Bathini if (paddr >= rstart && paddr < rend) {
10657dee93a9SHari Bathini raddr += fw_dump.boot_mem_dest_addr - hole_size;
10667dee93a9SHari Bathini break;
10677dee93a9SHari Bathini }
10687dee93a9SHari Bathini
10697dee93a9SHari Bathini rlast = rend;
10707dee93a9SHari Bathini }
10717dee93a9SHari Bathini
10727dee93a9SHari Bathini pr_debug("vmcoreinfo: paddr = 0x%lx, raddr = 0x%lx\n", paddr, raddr);
10737dee93a9SHari Bathini return raddr;
1074d34c5f26SMahesh Salgaonkar }
1075d34c5f26SMahesh Salgaonkar
fadump_create_elfcore_headers(char * bufp)10762df173d9SMahesh Salgaonkar static int fadump_create_elfcore_headers(char *bufp)
10772df173d9SMahesh Salgaonkar {
10787dee93a9SHari Bathini unsigned long long raddr, offset;
10792df173d9SMahesh Salgaonkar struct elf_phdr *phdr;
10807dee93a9SHari Bathini struct elfhdr *elf;
10817dee93a9SHari Bathini int i, j;
10822df173d9SMahesh Salgaonkar
10832df173d9SMahesh Salgaonkar fadump_init_elfcore_header(bufp);
10842df173d9SMahesh Salgaonkar elf = (struct elfhdr *)bufp;
10852df173d9SMahesh Salgaonkar bufp += sizeof(struct elfhdr);
10862df173d9SMahesh Salgaonkar
1087ebaeb5aeSMahesh Salgaonkar /*
1088ebaeb5aeSMahesh Salgaonkar * setup ELF PT_NOTE, place holder for cpu notes info. The notes info
1089ebaeb5aeSMahesh Salgaonkar * will be populated during second kernel boot after crash. Hence
1090ebaeb5aeSMahesh Salgaonkar * this PT_NOTE will always be the first elf note.
1091ebaeb5aeSMahesh Salgaonkar *
1092ebaeb5aeSMahesh Salgaonkar * NOTE: Any new ELF note addition should be placed after this note.
1093ebaeb5aeSMahesh Salgaonkar */
1094ebaeb5aeSMahesh Salgaonkar phdr = (struct elf_phdr *)bufp;
1095ebaeb5aeSMahesh Salgaonkar bufp += sizeof(struct elf_phdr);
1096ebaeb5aeSMahesh Salgaonkar phdr->p_type = PT_NOTE;
1097ebaeb5aeSMahesh Salgaonkar phdr->p_flags = 0;
1098ebaeb5aeSMahesh Salgaonkar phdr->p_vaddr = 0;
1099ebaeb5aeSMahesh Salgaonkar phdr->p_align = 0;
1100ebaeb5aeSMahesh Salgaonkar
1101ebaeb5aeSMahesh Salgaonkar phdr->p_offset = 0;
1102ebaeb5aeSMahesh Salgaonkar phdr->p_paddr = 0;
1103ebaeb5aeSMahesh Salgaonkar phdr->p_filesz = 0;
1104ebaeb5aeSMahesh Salgaonkar phdr->p_memsz = 0;
1105ebaeb5aeSMahesh Salgaonkar
1106ebaeb5aeSMahesh Salgaonkar (elf->e_phnum)++;
1107ebaeb5aeSMahesh Salgaonkar
1108d34c5f26SMahesh Salgaonkar /* setup ELF PT_NOTE for vmcoreinfo */
1109d34c5f26SMahesh Salgaonkar phdr = (struct elf_phdr *)bufp;
1110d34c5f26SMahesh Salgaonkar bufp += sizeof(struct elf_phdr);
1111d34c5f26SMahesh Salgaonkar phdr->p_type = PT_NOTE;
1112d34c5f26SMahesh Salgaonkar phdr->p_flags = 0;
1113d34c5f26SMahesh Salgaonkar phdr->p_vaddr = 0;
1114d34c5f26SMahesh Salgaonkar phdr->p_align = 0;
1115d34c5f26SMahesh Salgaonkar
1116d34c5f26SMahesh Salgaonkar phdr->p_paddr = fadump_relocate(paddr_vmcoreinfo_note());
1117d34c5f26SMahesh Salgaonkar phdr->p_offset = phdr->p_paddr;
11185203f499SXunlei Pang phdr->p_memsz = phdr->p_filesz = VMCOREINFO_NOTE_SIZE;
1119d34c5f26SMahesh Salgaonkar
1120d34c5f26SMahesh Salgaonkar /* Increment number of program headers. */
1121d34c5f26SMahesh Salgaonkar (elf->e_phnum)++;
1122d34c5f26SMahesh Salgaonkar
11232df173d9SMahesh Salgaonkar /* setup PT_LOAD sections. */
11247dee93a9SHari Bathini j = 0;
11257dee93a9SHari Bathini offset = 0;
11267dee93a9SHari Bathini raddr = fw_dump.boot_mem_addr[0];
1127e4fc48fbSHari Bathini for (i = 0; i < crash_mrange_info.mem_range_cnt; i++) {
1128e4fc48fbSHari Bathini u64 mbase, msize;
11292df173d9SMahesh Salgaonkar
1130e4fc48fbSHari Bathini mbase = crash_mrange_info.mem_ranges[i].base;
1131e4fc48fbSHari Bathini msize = crash_mrange_info.mem_ranges[i].size;
11322df173d9SMahesh Salgaonkar if (!msize)
11332df173d9SMahesh Salgaonkar continue;
11342df173d9SMahesh Salgaonkar
11352df173d9SMahesh Salgaonkar phdr = (struct elf_phdr *)bufp;
11362df173d9SMahesh Salgaonkar bufp += sizeof(struct elf_phdr);
11372df173d9SMahesh Salgaonkar phdr->p_type = PT_LOAD;
11382df173d9SMahesh Salgaonkar phdr->p_flags = PF_R|PF_W|PF_X;
11392df173d9SMahesh Salgaonkar phdr->p_offset = mbase;
11402df173d9SMahesh Salgaonkar
11417dee93a9SHari Bathini if (mbase == raddr) {
11422df173d9SMahesh Salgaonkar /*
1143becd91d9SHari Bathini * The entire real memory region will be moved by
1144becd91d9SHari Bathini * firmware to the specified destination_address.
1145becd91d9SHari Bathini * Hence set the correct offset.
11462df173d9SMahesh Salgaonkar */
11477dee93a9SHari Bathini phdr->p_offset = fw_dump.boot_mem_dest_addr + offset;
11487dee93a9SHari Bathini if (j < (fw_dump.boot_mem_regs_cnt - 1)) {
11497dee93a9SHari Bathini offset += fw_dump.boot_mem_sz[j];
11507dee93a9SHari Bathini raddr = fw_dump.boot_mem_addr[++j];
11517dee93a9SHari Bathini }
11522df173d9SMahesh Salgaonkar }
11532df173d9SMahesh Salgaonkar
11542df173d9SMahesh Salgaonkar phdr->p_paddr = mbase;
11552df173d9SMahesh Salgaonkar phdr->p_vaddr = (unsigned long)__va(mbase);
11562df173d9SMahesh Salgaonkar phdr->p_filesz = msize;
11572df173d9SMahesh Salgaonkar phdr->p_memsz = msize;
11582df173d9SMahesh Salgaonkar phdr->p_align = 0;
11592df173d9SMahesh Salgaonkar
11602df173d9SMahesh Salgaonkar /* Increment number of program headers. */
11612df173d9SMahesh Salgaonkar (elf->e_phnum)++;
11622df173d9SMahesh Salgaonkar }
11632df173d9SMahesh Salgaonkar return 0;
11642df173d9SMahesh Salgaonkar }
11652df173d9SMahesh Salgaonkar
init_fadump_header(unsigned long addr)11662df173d9SMahesh Salgaonkar static unsigned long init_fadump_header(unsigned long addr)
11672df173d9SMahesh Salgaonkar {
11682df173d9SMahesh Salgaonkar struct fadump_crash_info_header *fdh;
11692df173d9SMahesh Salgaonkar
11702df173d9SMahesh Salgaonkar if (!addr)
11712df173d9SMahesh Salgaonkar return 0;
11722df173d9SMahesh Salgaonkar
11732df173d9SMahesh Salgaonkar fdh = __va(addr);
11742df173d9SMahesh Salgaonkar addr += sizeof(struct fadump_crash_info_header);
11752df173d9SMahesh Salgaonkar
11762df173d9SMahesh Salgaonkar memset(fdh, 0, sizeof(struct fadump_crash_info_header));
11772df173d9SMahesh Salgaonkar fdh->magic_number = FADUMP_CRASH_INFO_MAGIC;
11782df173d9SMahesh Salgaonkar fdh->elfcorehdr_addr = addr;
1179ebaeb5aeSMahesh Salgaonkar /* We will set the crashing cpu id in crash_fadump() during crash. */
11800226e552SHari Bathini fdh->crashing_cpu = FADUMP_CPU_UNKNOWN;
11816584cec0SHari Bathini /*
11826584cec0SHari Bathini * When LPAR is terminated by PYHP, ensure all possible CPUs'
11836584cec0SHari Bathini * register data is processed while exporting the vmcore.
11846584cec0SHari Bathini */
11856584cec0SHari Bathini fdh->cpu_mask = *cpu_possible_mask;
11862df173d9SMahesh Salgaonkar
11872df173d9SMahesh Salgaonkar return addr;
11882df173d9SMahesh Salgaonkar }
11892df173d9SMahesh Salgaonkar
register_fadump(void)119098b8cd7fSMichal Suchanek static int register_fadump(void)
11913ccc00a7SMahesh Salgaonkar {
11922df173d9SMahesh Salgaonkar unsigned long addr;
11932df173d9SMahesh Salgaonkar void *vaddr;
11941bd6a1c4SHari Bathini int ret;
11952df173d9SMahesh Salgaonkar
11963ccc00a7SMahesh Salgaonkar /*
11973ccc00a7SMahesh Salgaonkar * If no memory is reserved then we can not register for firmware-
11983ccc00a7SMahesh Salgaonkar * assisted dump.
11993ccc00a7SMahesh Salgaonkar */
12003ccc00a7SMahesh Salgaonkar if (!fw_dump.reserve_dump_area_size)
120198b8cd7fSMichal Suchanek return -ENODEV;
12023ccc00a7SMahesh Salgaonkar
12031bd6a1c4SHari Bathini ret = fadump_setup_crash_memory_ranges();
12041bd6a1c4SHari Bathini if (ret)
12051bd6a1c4SHari Bathini return ret;
12062df173d9SMahesh Salgaonkar
120741a65d16SHari Bathini addr = fw_dump.fadumphdr_addr;
120841a65d16SHari Bathini
12092df173d9SMahesh Salgaonkar /* Initialize fadump crash info header. */
12102df173d9SMahesh Salgaonkar addr = init_fadump_header(addr);
12112df173d9SMahesh Salgaonkar vaddr = __va(addr);
12122df173d9SMahesh Salgaonkar
12132df173d9SMahesh Salgaonkar pr_debug("Creating ELF core headers at %#016lx\n", addr);
12142df173d9SMahesh Salgaonkar fadump_create_elfcore_headers(vaddr);
12152df173d9SMahesh Salgaonkar
12163ccc00a7SMahesh Salgaonkar /* register the future kernel dump with firmware. */
121741a65d16SHari Bathini pr_debug("Registering for firmware-assisted kernel dump...\n");
121841a65d16SHari Bathini return fw_dump.ops->fadump_register(&fw_dump);
12193ccc00a7SMahesh Salgaonkar }
12203ccc00a7SMahesh Salgaonkar
fadump_cleanup(void)1221b500afffSMahesh Salgaonkar void fadump_cleanup(void)
1222b500afffSMahesh Salgaonkar {
12232790d01dSHari Bathini if (!fw_dump.fadump_supported)
12242790d01dSHari Bathini return;
12252790d01dSHari Bathini
1226b500afffSMahesh Salgaonkar /* Invalidate the registration only if dump is active. */
1227b500afffSMahesh Salgaonkar if (fw_dump.dump_active) {
1228f3512011SHari Bathini pr_debug("Invalidating firmware-assisted dump registration\n");
1229f3512011SHari Bathini fw_dump.ops->fadump_invalidate(&fw_dump);
1230722cde76SMahesh Salgaonkar } else if (fw_dump.dump_registered) {
1231722cde76SMahesh Salgaonkar /* Un-register Firmware-assisted dump if it was registered. */
123241a65d16SHari Bathini fw_dump.ops->fadump_unregister(&fw_dump);
1233e4fc48fbSHari Bathini fadump_free_mem_ranges(&crash_mrange_info);
1234b500afffSMahesh Salgaonkar }
12352790d01dSHari Bathini
12362790d01dSHari Bathini if (fw_dump.ops->fadump_cleanup)
12372790d01dSHari Bathini fw_dump.ops->fadump_cleanup(&fw_dump);
1238b500afffSMahesh Salgaonkar }
1239b500afffSMahesh Salgaonkar
fadump_free_reserved_memory(unsigned long start_pfn,unsigned long end_pfn)124068fa6478SHari Bathini static void fadump_free_reserved_memory(unsigned long start_pfn,
124168fa6478SHari Bathini unsigned long end_pfn)
124268fa6478SHari Bathini {
124368fa6478SHari Bathini unsigned long pfn;
124468fa6478SHari Bathini unsigned long time_limit = jiffies + HZ;
124568fa6478SHari Bathini
124668fa6478SHari Bathini pr_info("freeing reserved memory (0x%llx - 0x%llx)\n",
124768fa6478SHari Bathini PFN_PHYS(start_pfn), PFN_PHYS(end_pfn));
124868fa6478SHari Bathini
124968fa6478SHari Bathini for (pfn = start_pfn; pfn < end_pfn; pfn++) {
125068fa6478SHari Bathini free_reserved_page(pfn_to_page(pfn));
125168fa6478SHari Bathini
125268fa6478SHari Bathini if (time_after(jiffies, time_limit)) {
125368fa6478SHari Bathini cond_resched();
125468fa6478SHari Bathini time_limit = jiffies + HZ;
125568fa6478SHari Bathini }
125668fa6478SHari Bathini }
125768fa6478SHari Bathini }
125868fa6478SHari Bathini
125968fa6478SHari Bathini /*
126068fa6478SHari Bathini * Skip memory holes and free memory that was actually reserved.
126168fa6478SHari Bathini */
fadump_release_reserved_area(u64 start,u64 end)1262dda9dbfeSHari Bathini static void fadump_release_reserved_area(u64 start, u64 end)
126368fa6478SHari Bathini {
1264b10d6bcaSMike Rapoport unsigned long reg_spfn, reg_epfn;
1265b10d6bcaSMike Rapoport u64 tstart, tend, spfn, epfn;
1266b10d6bcaSMike Rapoport int i;
126768fa6478SHari Bathini
1268dda9dbfeSHari Bathini spfn = PHYS_PFN(start);
1269dda9dbfeSHari Bathini epfn = PHYS_PFN(end);
1270c9118e6cSMike Rapoport
1271c9118e6cSMike Rapoport for_each_mem_pfn_range(i, MAX_NUMNODES, ®_spfn, ®_epfn, NULL) {
1272c9118e6cSMike Rapoport tstart = max_t(u64, spfn, reg_spfn);
1273c9118e6cSMike Rapoport tend = min_t(u64, epfn, reg_epfn);
1274c9118e6cSMike Rapoport
127568fa6478SHari Bathini if (tstart < tend) {
127668fa6478SHari Bathini fadump_free_reserved_memory(tstart, tend);
127768fa6478SHari Bathini
1278dda9dbfeSHari Bathini if (tend == epfn)
127968fa6478SHari Bathini break;
128068fa6478SHari Bathini
1281dda9dbfeSHari Bathini spfn = tend;
128268fa6478SHari Bathini }
128368fa6478SHari Bathini }
128468fa6478SHari Bathini }
128568fa6478SHari Bathini
1286b500afffSMahesh Salgaonkar /*
1287dda9dbfeSHari Bathini * Sort the mem ranges in-place and merge adjacent ranges
1288dda9dbfeSHari Bathini * to minimize the memory ranges count.
1289b500afffSMahesh Salgaonkar */
sort_and_merge_mem_ranges(struct fadump_mrange_info * mrange_info)1290dda9dbfeSHari Bathini static void sort_and_merge_mem_ranges(struct fadump_mrange_info *mrange_info)
1291b500afffSMahesh Salgaonkar {
1292dda9dbfeSHari Bathini struct fadump_memory_range *mem_ranges;
1293dda9dbfeSHari Bathini u64 base, size;
1294dda9dbfeSHari Bathini int i, j, idx;
1295dda9dbfeSHari Bathini
1296dda9dbfeSHari Bathini if (!reserved_mrange_info.mem_range_cnt)
1297dda9dbfeSHari Bathini return;
1298dda9dbfeSHari Bathini
1299dda9dbfeSHari Bathini /* Sort the memory ranges */
1300dda9dbfeSHari Bathini mem_ranges = mrange_info->mem_ranges;
1301dda9dbfeSHari Bathini for (i = 0; i < mrange_info->mem_range_cnt; i++) {
1302dda9dbfeSHari Bathini idx = i;
1303dda9dbfeSHari Bathini for (j = (i + 1); j < mrange_info->mem_range_cnt; j++) {
1304dda9dbfeSHari Bathini if (mem_ranges[idx].base > mem_ranges[j].base)
1305dda9dbfeSHari Bathini idx = j;
1306dda9dbfeSHari Bathini }
130720776319SJiapeng Chong if (idx != i)
130820776319SJiapeng Chong swap(mem_ranges[idx], mem_ranges[i]);
1309dda9dbfeSHari Bathini }
1310dda9dbfeSHari Bathini
1311dda9dbfeSHari Bathini /* Merge adjacent reserved ranges */
1312dda9dbfeSHari Bathini idx = 0;
1313dda9dbfeSHari Bathini for (i = 1; i < mrange_info->mem_range_cnt; i++) {
1314dda9dbfeSHari Bathini base = mem_ranges[i-1].base;
1315dda9dbfeSHari Bathini size = mem_ranges[i-1].size;
1316dda9dbfeSHari Bathini if (mem_ranges[i].base == (base + size))
1317dda9dbfeSHari Bathini mem_ranges[idx].size += mem_ranges[i].size;
1318dda9dbfeSHari Bathini else {
1319dda9dbfeSHari Bathini idx++;
1320dda9dbfeSHari Bathini if (i == idx)
1321dda9dbfeSHari Bathini continue;
1322dda9dbfeSHari Bathini
1323dda9dbfeSHari Bathini mem_ranges[idx] = mem_ranges[i];
1324dda9dbfeSHari Bathini }
1325dda9dbfeSHari Bathini }
1326dda9dbfeSHari Bathini mrange_info->mem_range_cnt = idx + 1;
1327dda9dbfeSHari Bathini }
1328dda9dbfeSHari Bathini
1329dda9dbfeSHari Bathini /*
1330dda9dbfeSHari Bathini * Scan reserved-ranges to consider them while reserving/releasing
1331dda9dbfeSHari Bathini * memory for FADump.
1332dda9dbfeSHari Bathini */
early_init_dt_scan_reserved_ranges(unsigned long node)133302c04e37SHari Bathini static void __init early_init_dt_scan_reserved_ranges(unsigned long node)
1334dda9dbfeSHari Bathini {
1335dda9dbfeSHari Bathini const __be32 *prop;
1336dda9dbfeSHari Bathini int len, ret = -1;
1337dda9dbfeSHari Bathini unsigned long i;
1338dda9dbfeSHari Bathini
133902c04e37SHari Bathini /* reserved-ranges already scanned */
134002c04e37SHari Bathini if (reserved_mrange_info.mem_range_cnt != 0)
134102c04e37SHari Bathini return;
1342dda9dbfeSHari Bathini
134302c04e37SHari Bathini prop = of_get_flat_dt_prop(node, "reserved-ranges", &len);
1344dda9dbfeSHari Bathini if (!prop)
134502c04e37SHari Bathini return;
1346dda9dbfeSHari Bathini
1347dda9dbfeSHari Bathini /*
1348dda9dbfeSHari Bathini * Each reserved range is an (address,size) pair, 2 cells each,
1349dda9dbfeSHari Bathini * totalling 4 cells per range.
1350dda9dbfeSHari Bathini */
1351dda9dbfeSHari Bathini for (i = 0; i < len / (sizeof(*prop) * 4); i++) {
1352dda9dbfeSHari Bathini u64 base, size;
1353dda9dbfeSHari Bathini
1354dda9dbfeSHari Bathini base = of_read_number(prop + (i * 4) + 0, 2);
1355dda9dbfeSHari Bathini size = of_read_number(prop + (i * 4) + 2, 2);
1356dda9dbfeSHari Bathini
1357dda9dbfeSHari Bathini if (size) {
1358dda9dbfeSHari Bathini ret = fadump_add_mem_range(&reserved_mrange_info,
1359dda9dbfeSHari Bathini base, base + size);
1360dda9dbfeSHari Bathini if (ret < 0) {
1361dda9dbfeSHari Bathini pr_warn("some reserved ranges are ignored!\n");
1362dda9dbfeSHari Bathini break;
1363dda9dbfeSHari Bathini }
1364dda9dbfeSHari Bathini }
1365dda9dbfeSHari Bathini }
1366dda9dbfeSHari Bathini
136702c04e37SHari Bathini /* Compact reserved ranges */
136802c04e37SHari Bathini sort_and_merge_mem_ranges(&reserved_mrange_info);
1369dda9dbfeSHari Bathini }
1370dda9dbfeSHari Bathini
1371dda9dbfeSHari Bathini /*
1372dda9dbfeSHari Bathini * Release the memory that was reserved during early boot to preserve the
1373dda9dbfeSHari Bathini * crash'ed kernel's memory contents except reserved dump area (permanent
1374dda9dbfeSHari Bathini * reservation) and reserved ranges used by F/W. The released memory will
1375dda9dbfeSHari Bathini * be available for general use.
1376dda9dbfeSHari Bathini */
fadump_release_memory(u64 begin,u64 end)1377dda9dbfeSHari Bathini static void fadump_release_memory(u64 begin, u64 end)
1378dda9dbfeSHari Bathini {
1379dda9dbfeSHari Bathini u64 ra_start, ra_end, tstart;
1380dda9dbfeSHari Bathini int i, ret;
1381dda9dbfeSHari Bathini
1382b500afffSMahesh Salgaonkar ra_start = fw_dump.reserve_dump_area_start;
1383b500afffSMahesh Salgaonkar ra_end = ra_start + fw_dump.reserve_dump_area_size;
1384b500afffSMahesh Salgaonkar
1385b500afffSMahesh Salgaonkar /*
138602c04e37SHari Bathini * If reserved ranges array limit is hit, overwrite the last reserved
138702c04e37SHari Bathini * memory range with reserved dump area to ensure it is excluded from
138802c04e37SHari Bathini * the memory being released (reused for next FADump registration).
1389dda9dbfeSHari Bathini */
139002c04e37SHari Bathini if (reserved_mrange_info.mem_range_cnt ==
139102c04e37SHari Bathini reserved_mrange_info.max_mem_ranges)
139202c04e37SHari Bathini reserved_mrange_info.mem_range_cnt--;
1393dda9dbfeSHari Bathini
139402c04e37SHari Bathini ret = fadump_add_mem_range(&reserved_mrange_info, ra_start, ra_end);
139502c04e37SHari Bathini if (ret != 0)
1396dda9dbfeSHari Bathini return;
1397dda9dbfeSHari Bathini
1398dda9dbfeSHari Bathini /* Get the reserved ranges list in order first. */
1399dda9dbfeSHari Bathini sort_and_merge_mem_ranges(&reserved_mrange_info);
1400dda9dbfeSHari Bathini
1401dda9dbfeSHari Bathini /* Exclude reserved ranges and release remaining memory */
1402dda9dbfeSHari Bathini tstart = begin;
1403dda9dbfeSHari Bathini for (i = 0; i < reserved_mrange_info.mem_range_cnt; i++) {
1404dda9dbfeSHari Bathini ra_start = reserved_mrange_info.mem_ranges[i].base;
1405dda9dbfeSHari Bathini ra_end = ra_start + reserved_mrange_info.mem_ranges[i].size;
1406dda9dbfeSHari Bathini
1407dda9dbfeSHari Bathini if (tstart >= ra_end)
1408dda9dbfeSHari Bathini continue;
1409dda9dbfeSHari Bathini
1410dda9dbfeSHari Bathini if (tstart < ra_start)
1411dda9dbfeSHari Bathini fadump_release_reserved_area(tstart, ra_start);
1412dda9dbfeSHari Bathini tstart = ra_end;
1413dda9dbfeSHari Bathini }
1414dda9dbfeSHari Bathini
1415dda9dbfeSHari Bathini if (tstart < end)
1416dda9dbfeSHari Bathini fadump_release_reserved_area(tstart, end);
1417b500afffSMahesh Salgaonkar }
1418b500afffSMahesh Salgaonkar
fadump_invalidate_release_mem(void)1419b500afffSMahesh Salgaonkar static void fadump_invalidate_release_mem(void)
1420b500afffSMahesh Salgaonkar {
1421b500afffSMahesh Salgaonkar mutex_lock(&fadump_mutex);
1422b500afffSMahesh Salgaonkar if (!fw_dump.dump_active) {
1423b500afffSMahesh Salgaonkar mutex_unlock(&fadump_mutex);
1424b500afffSMahesh Salgaonkar return;
1425b500afffSMahesh Salgaonkar }
1426b500afffSMahesh Salgaonkar
1427b500afffSMahesh Salgaonkar fadump_cleanup();
1428b500afffSMahesh Salgaonkar mutex_unlock(&fadump_mutex);
1429b500afffSMahesh Salgaonkar
14307dee93a9SHari Bathini fadump_release_memory(fw_dump.boot_mem_top, memblock_end_of_DRAM());
1431961cf26aSHari Bathini fadump_free_cpu_notes_buf();
1432961cf26aSHari Bathini
1433a4e2e2caSHari Bathini /*
1434a4e2e2caSHari Bathini * Setup kernel metadata and initialize the kernel dump
1435a4e2e2caSHari Bathini * memory structure for FADump re-registration.
1436a4e2e2caSHari Bathini */
1437a4e2e2caSHari Bathini if (fw_dump.ops->fadump_setup_metadata &&
1438a4e2e2caSHari Bathini (fw_dump.ops->fadump_setup_metadata(&fw_dump) < 0))
1439a4e2e2caSHari Bathini pr_warn("Failed to setup kernel metadata!\n");
144041a65d16SHari Bathini fw_dump.ops->fadump_init_mem_struct(&fw_dump);
1441b500afffSMahesh Salgaonkar }
1442b500afffSMahesh Salgaonkar
release_mem_store(struct kobject * kobj,struct kobj_attribute * attr,const char * buf,size_t count)1443d418b19fSSourabh Jain static ssize_t release_mem_store(struct kobject *kobj,
1444b500afffSMahesh Salgaonkar struct kobj_attribute *attr,
1445b500afffSMahesh Salgaonkar const char *buf, size_t count)
1446b500afffSMahesh Salgaonkar {
1447dcdc4679SMichal Suchanek int input = -1;
1448dcdc4679SMichal Suchanek
1449b500afffSMahesh Salgaonkar if (!fw_dump.dump_active)
1450b500afffSMahesh Salgaonkar return -EPERM;
1451b500afffSMahesh Salgaonkar
1452dcdc4679SMichal Suchanek if (kstrtoint(buf, 0, &input))
1453dcdc4679SMichal Suchanek return -EINVAL;
1454dcdc4679SMichal Suchanek
1455dcdc4679SMichal Suchanek if (input == 1) {
1456b500afffSMahesh Salgaonkar /*
1457b500afffSMahesh Salgaonkar * Take away the '/proc/vmcore'. We are releasing the dump
1458b500afffSMahesh Salgaonkar * memory, hence it will not be valid anymore.
1459b500afffSMahesh Salgaonkar */
14602685f826SMichael Ellerman #ifdef CONFIG_PROC_VMCORE
1461b500afffSMahesh Salgaonkar vmcore_cleanup();
14622685f826SMichael Ellerman #endif
1463b500afffSMahesh Salgaonkar fadump_invalidate_release_mem();
1464b500afffSMahesh Salgaonkar
1465b500afffSMahesh Salgaonkar } else
1466b500afffSMahesh Salgaonkar return -EINVAL;
1467b500afffSMahesh Salgaonkar return count;
1468b500afffSMahesh Salgaonkar }
1469b500afffSMahesh Salgaonkar
1470d418b19fSSourabh Jain /* Release the reserved memory and disable the FADump */
unregister_fadump(void)1471d276960dSNick Child static void __init unregister_fadump(void)
1472d418b19fSSourabh Jain {
1473d418b19fSSourabh Jain fadump_cleanup();
1474d418b19fSSourabh Jain fadump_release_memory(fw_dump.reserve_dump_area_start,
1475d418b19fSSourabh Jain fw_dump.reserve_dump_area_size);
1476d418b19fSSourabh Jain fw_dump.fadump_enabled = 0;
1477d418b19fSSourabh Jain kobject_put(fadump_kobj);
1478d418b19fSSourabh Jain }
1479d418b19fSSourabh Jain
enabled_show(struct kobject * kobj,struct kobj_attribute * attr,char * buf)1480d418b19fSSourabh Jain static ssize_t enabled_show(struct kobject *kobj,
14813ccc00a7SMahesh Salgaonkar struct kobj_attribute *attr,
14823ccc00a7SMahesh Salgaonkar char *buf)
14833ccc00a7SMahesh Salgaonkar {
14843ccc00a7SMahesh Salgaonkar return sprintf(buf, "%d\n", fw_dump.fadump_enabled);
14853ccc00a7SMahesh Salgaonkar }
14863ccc00a7SMahesh Salgaonkar
mem_reserved_show(struct kobject * kobj,struct kobj_attribute * attr,char * buf)1487d8e73458SSourabh Jain static ssize_t mem_reserved_show(struct kobject *kobj,
1488d8e73458SSourabh Jain struct kobj_attribute *attr,
1489d8e73458SSourabh Jain char *buf)
1490d8e73458SSourabh Jain {
1491d8e73458SSourabh Jain return sprintf(buf, "%ld\n", fw_dump.reserve_dump_area_size);
1492d8e73458SSourabh Jain }
1493d8e73458SSourabh Jain
registered_show(struct kobject * kobj,struct kobj_attribute * attr,char * buf)1494d418b19fSSourabh Jain static ssize_t registered_show(struct kobject *kobj,
14953ccc00a7SMahesh Salgaonkar struct kobj_attribute *attr,
14963ccc00a7SMahesh Salgaonkar char *buf)
14973ccc00a7SMahesh Salgaonkar {
14983ccc00a7SMahesh Salgaonkar return sprintf(buf, "%d\n", fw_dump.dump_registered);
14993ccc00a7SMahesh Salgaonkar }
15003ccc00a7SMahesh Salgaonkar
registered_store(struct kobject * kobj,struct kobj_attribute * attr,const char * buf,size_t count)1501d418b19fSSourabh Jain static ssize_t registered_store(struct kobject *kobj,
15023ccc00a7SMahesh Salgaonkar struct kobj_attribute *attr,
15033ccc00a7SMahesh Salgaonkar const char *buf, size_t count)
15043ccc00a7SMahesh Salgaonkar {
15053ccc00a7SMahesh Salgaonkar int ret = 0;
1506dcdc4679SMichal Suchanek int input = -1;
15073ccc00a7SMahesh Salgaonkar
1508f3512011SHari Bathini if (!fw_dump.fadump_enabled || fw_dump.dump_active)
15093ccc00a7SMahesh Salgaonkar return -EPERM;
15103ccc00a7SMahesh Salgaonkar
1511dcdc4679SMichal Suchanek if (kstrtoint(buf, 0, &input))
1512dcdc4679SMichal Suchanek return -EINVAL;
1513dcdc4679SMichal Suchanek
15143ccc00a7SMahesh Salgaonkar mutex_lock(&fadump_mutex);
15153ccc00a7SMahesh Salgaonkar
1516dcdc4679SMichal Suchanek switch (input) {
1517dcdc4679SMichal Suchanek case 0:
15183ccc00a7SMahesh Salgaonkar if (fw_dump.dump_registered == 0) {
15193ccc00a7SMahesh Salgaonkar goto unlock_out;
15203ccc00a7SMahesh Salgaonkar }
1521f3512011SHari Bathini
15223ccc00a7SMahesh Salgaonkar /* Un-register Firmware-assisted dump */
152341a65d16SHari Bathini pr_debug("Un-register firmware-assisted dump\n");
152441a65d16SHari Bathini fw_dump.ops->fadump_unregister(&fw_dump);
15253ccc00a7SMahesh Salgaonkar break;
1526dcdc4679SMichal Suchanek case 1:
15273ccc00a7SMahesh Salgaonkar if (fw_dump.dump_registered == 1) {
15280823c68bSHari Bathini /* Un-register Firmware-assisted dump */
152941a65d16SHari Bathini fw_dump.ops->fadump_unregister(&fw_dump);
15303ccc00a7SMahesh Salgaonkar }
15313ccc00a7SMahesh Salgaonkar /* Register Firmware-assisted dump */
153298b8cd7fSMichal Suchanek ret = register_fadump();
15333ccc00a7SMahesh Salgaonkar break;
15343ccc00a7SMahesh Salgaonkar default:
15353ccc00a7SMahesh Salgaonkar ret = -EINVAL;
15363ccc00a7SMahesh Salgaonkar break;
15373ccc00a7SMahesh Salgaonkar }
15383ccc00a7SMahesh Salgaonkar
15393ccc00a7SMahesh Salgaonkar unlock_out:
15403ccc00a7SMahesh Salgaonkar mutex_unlock(&fadump_mutex);
15413ccc00a7SMahesh Salgaonkar return ret < 0 ? ret : count;
15423ccc00a7SMahesh Salgaonkar }
15433ccc00a7SMahesh Salgaonkar
fadump_region_show(struct seq_file * m,void * private)15443ccc00a7SMahesh Salgaonkar static int fadump_region_show(struct seq_file *m, void *private)
15453ccc00a7SMahesh Salgaonkar {
15463ccc00a7SMahesh Salgaonkar if (!fw_dump.fadump_enabled)
15473ccc00a7SMahesh Salgaonkar return 0;
15483ccc00a7SMahesh Salgaonkar
1549b500afffSMahesh Salgaonkar mutex_lock(&fadump_mutex);
155041a65d16SHari Bathini fw_dump.ops->fadump_region_show(&fw_dump, m);
1551b500afffSMahesh Salgaonkar mutex_unlock(&fadump_mutex);
15523ccc00a7SMahesh Salgaonkar return 0;
15533ccc00a7SMahesh Salgaonkar }
15543ccc00a7SMahesh Salgaonkar
1555d418b19fSSourabh Jain static struct kobj_attribute release_attr = __ATTR_WO(release_mem);
1556d418b19fSSourabh Jain static struct kobj_attribute enable_attr = __ATTR_RO(enabled);
1557d418b19fSSourabh Jain static struct kobj_attribute register_attr = __ATTR_RW(registered);
1558d8e73458SSourabh Jain static struct kobj_attribute mem_reserved_attr = __ATTR_RO(mem_reserved);
1559d418b19fSSourabh Jain
1560d418b19fSSourabh Jain static struct attribute *fadump_attrs[] = {
1561d418b19fSSourabh Jain &enable_attr.attr,
1562d418b19fSSourabh Jain ®ister_attr.attr,
1563d8e73458SSourabh Jain &mem_reserved_attr.attr,
1564d418b19fSSourabh Jain NULL,
1565d418b19fSSourabh Jain };
1566d418b19fSSourabh Jain
1567d418b19fSSourabh Jain ATTRIBUTE_GROUPS(fadump);
15683ccc00a7SMahesh Salgaonkar
1569f6cee260SYangtao Li DEFINE_SHOW_ATTRIBUTE(fadump_region);
15703ccc00a7SMahesh Salgaonkar
fadump_init_files(void)1571d276960dSNick Child static void __init fadump_init_files(void)
15723ccc00a7SMahesh Salgaonkar {
15733ccc00a7SMahesh Salgaonkar int rc = 0;
15743ccc00a7SMahesh Salgaonkar
1575d418b19fSSourabh Jain fadump_kobj = kobject_create_and_add("fadump", kernel_kobj);
1576d418b19fSSourabh Jain if (!fadump_kobj) {
1577d418b19fSSourabh Jain pr_err("failed to create fadump kobject\n");
1578d418b19fSSourabh Jain return;
1579d418b19fSSourabh Jain }
1580860286cfSGreg Kroah-Hartman
1581dbf77fedSAneesh Kumar K.V debugfs_create_file("fadump_region", 0444, arch_debugfs_dir, NULL,
15823ccc00a7SMahesh Salgaonkar &fadump_region_fops);
1583b500afffSMahesh Salgaonkar
1584b500afffSMahesh Salgaonkar if (fw_dump.dump_active) {
1585d418b19fSSourabh Jain rc = sysfs_create_file(fadump_kobj, &release_attr.attr);
1586b500afffSMahesh Salgaonkar if (rc)
1587d418b19fSSourabh Jain pr_err("unable to create release_mem sysfs file (%d)\n",
1588d418b19fSSourabh Jain rc);
1589d418b19fSSourabh Jain }
1590d418b19fSSourabh Jain
1591d418b19fSSourabh Jain rc = sysfs_create_groups(fadump_kobj, fadump_groups);
1592d418b19fSSourabh Jain if (rc) {
1593d418b19fSSourabh Jain pr_err("sysfs group creation failed (%d), unregistering FADump",
1594d418b19fSSourabh Jain rc);
1595d418b19fSSourabh Jain unregister_fadump();
1596d418b19fSSourabh Jain return;
1597d418b19fSSourabh Jain }
1598d418b19fSSourabh Jain
1599d418b19fSSourabh Jain /*
1600d418b19fSSourabh Jain * The FADump sysfs are moved from kernel_kobj to fadump_kobj need to
1601d418b19fSSourabh Jain * create symlink at old location to maintain backward compatibility.
1602d418b19fSSourabh Jain *
1603d418b19fSSourabh Jain * - fadump_enabled -> fadump/enabled
1604d418b19fSSourabh Jain * - fadump_registered -> fadump/registered
1605d418b19fSSourabh Jain * - fadump_release_mem -> fadump/release_mem
1606d418b19fSSourabh Jain */
1607d418b19fSSourabh Jain rc = compat_only_sysfs_link_entry_to_kobj(kernel_kobj, fadump_kobj,
1608d418b19fSSourabh Jain "enabled", "fadump_enabled");
1609d418b19fSSourabh Jain if (rc) {
1610d418b19fSSourabh Jain pr_err("unable to create fadump_enabled symlink (%d)", rc);
1611d418b19fSSourabh Jain return;
1612d418b19fSSourabh Jain }
1613d418b19fSSourabh Jain
1614d418b19fSSourabh Jain rc = compat_only_sysfs_link_entry_to_kobj(kernel_kobj, fadump_kobj,
1615d418b19fSSourabh Jain "registered",
1616d418b19fSSourabh Jain "fadump_registered");
1617d418b19fSSourabh Jain if (rc) {
1618d418b19fSSourabh Jain pr_err("unable to create fadump_registered symlink (%d)", rc);
1619d418b19fSSourabh Jain sysfs_remove_link(kernel_kobj, "fadump_enabled");
1620d418b19fSSourabh Jain return;
1621d418b19fSSourabh Jain }
1622d418b19fSSourabh Jain
1623d418b19fSSourabh Jain if (fw_dump.dump_active) {
1624d418b19fSSourabh Jain rc = compat_only_sysfs_link_entry_to_kobj(kernel_kobj,
1625d418b19fSSourabh Jain fadump_kobj,
1626d418b19fSSourabh Jain "release_mem",
1627d418b19fSSourabh Jain "fadump_release_mem");
1628d418b19fSSourabh Jain if (rc)
1629d418b19fSSourabh Jain pr_err("unable to create fadump_release_mem symlink (%d)",
1630d418b19fSSourabh Jain rc);
1631b500afffSMahesh Salgaonkar }
16323ccc00a7SMahesh Salgaonkar return;
16333ccc00a7SMahesh Salgaonkar }
16343ccc00a7SMahesh Salgaonkar
16353ccc00a7SMahesh Salgaonkar /*
16363ccc00a7SMahesh Salgaonkar * Prepare for firmware-assisted dump.
16373ccc00a7SMahesh Salgaonkar */
setup_fadump(void)16383ccc00a7SMahesh Salgaonkar int __init setup_fadump(void)
16393ccc00a7SMahesh Salgaonkar {
1640565f9bc0SMichal Suchanek if (!fw_dump.fadump_supported)
16413ccc00a7SMahesh Salgaonkar return 0;
16423ccc00a7SMahesh Salgaonkar
1643565f9bc0SMichal Suchanek fadump_init_files();
16443ccc00a7SMahesh Salgaonkar fadump_show_config();
1645565f9bc0SMichal Suchanek
1646565f9bc0SMichal Suchanek if (!fw_dump.fadump_enabled)
1647565f9bc0SMichal Suchanek return 1;
1648565f9bc0SMichal Suchanek
16492df173d9SMahesh Salgaonkar /*
16502df173d9SMahesh Salgaonkar * If dump data is available then see if it is valid and prepare for
16512df173d9SMahesh Salgaonkar * saving it to the disk.
16522df173d9SMahesh Salgaonkar */
1653b500afffSMahesh Salgaonkar if (fw_dump.dump_active) {
1654b500afffSMahesh Salgaonkar /*
1655b500afffSMahesh Salgaonkar * if dump process fails then invalidate the registration
1656b500afffSMahesh Salgaonkar * and release memory before proceeding for re-registration.
1657b500afffSMahesh Salgaonkar */
1658f3512011SHari Bathini if (fw_dump.ops->fadump_process(&fw_dump) < 0)
1659b500afffSMahesh Salgaonkar fadump_invalidate_release_mem();
1660b500afffSMahesh Salgaonkar }
1661607451ceSHari Bathini /* Initialize the kernel dump memory structure and register with f/w */
1662607451ceSHari Bathini else if (fw_dump.reserve_dump_area_size) {
166341a65d16SHari Bathini fw_dump.ops->fadump_init_mem_struct(&fw_dump);
1664607451ceSHari Bathini register_fadump();
1665607451ceSHari Bathini }
1666f3512011SHari Bathini
166706e629c2SHari Bathini /*
166806e629c2SHari Bathini * In case of panic, fadump is triggered via ppc_panic_event()
166906e629c2SHari Bathini * panic notifier. Setting crash_kexec_post_notifiers to 'true'
167006e629c2SHari Bathini * lets panic() function take crash friendly path before panic
167106e629c2SHari Bathini * notifiers are invoked.
167206e629c2SHari Bathini */
167306e629c2SHari Bathini crash_kexec_post_notifiers = true;
167406e629c2SHari Bathini
16753ccc00a7SMahesh Salgaonkar return 1;
16763ccc00a7SMahesh Salgaonkar }
1677607451ceSHari Bathini /*
1678607451ceSHari Bathini * Use subsys_initcall_sync() here because there is dependency with
16791fd02f66SJulia Lawall * crash_save_vmcoreinfo_init(), which must run first to ensure vmcoreinfo initialization
16801fd02f66SJulia Lawall * is done before registering with f/w.
1681607451ceSHari Bathini */
1682607451ceSHari Bathini subsys_initcall_sync(setup_fadump);
1683bec53196SHari Bathini #else /* !CONFIG_PRESERVE_FA_DUMP */
1684bec53196SHari Bathini
1685bec53196SHari Bathini /* Scan the Firmware Assisted dump configuration details. */
early_init_dt_scan_fw_dump(unsigned long node,const char * uname,int depth,void * data)1686bec53196SHari Bathini int __init early_init_dt_scan_fw_dump(unsigned long node, const char *uname,
1687bec53196SHari Bathini int depth, void *data)
1688bec53196SHari Bathini {
1689bec53196SHari Bathini if ((depth != 1) || (strcmp(uname, "ibm,opal") != 0))
1690bec53196SHari Bathini return 0;
1691bec53196SHari Bathini
1692bec53196SHari Bathini opal_fadump_dt_scan(&fw_dump, node);
1693bec53196SHari Bathini return 1;
1694bec53196SHari Bathini }
1695bec53196SHari Bathini
1696bec53196SHari Bathini /*
1697bec53196SHari Bathini * When dump is active but PRESERVE_FA_DUMP is enabled on the kernel,
1698bec53196SHari Bathini * preserve crash data. The subsequent memory preserving kernel boot
1699bec53196SHari Bathini * is likely to process this crash data.
1700bec53196SHari Bathini */
fadump_reserve_mem(void)1701bec53196SHari Bathini int __init fadump_reserve_mem(void)
1702bec53196SHari Bathini {
1703bec53196SHari Bathini if (fw_dump.dump_active) {
1704bec53196SHari Bathini /*
1705bec53196SHari Bathini * If last boot has crashed then reserve all the memory
1706bec53196SHari Bathini * above boot memory to preserve crash data.
1707bec53196SHari Bathini */
1708bec53196SHari Bathini pr_info("Preserving crash data for processing in next boot.\n");
1709bec53196SHari Bathini fadump_reserve_crash_area(fw_dump.boot_mem_top);
1710bec53196SHari Bathini } else
1711bec53196SHari Bathini pr_debug("FADump-aware kernel..\n");
1712bec53196SHari Bathini
1713bec53196SHari Bathini return 1;
1714bec53196SHari Bathini }
1715bec53196SHari Bathini #endif /* CONFIG_PRESERVE_FA_DUMP */
1716b2a815a5SHari Bathini
1717b2a815a5SHari Bathini /* Preserve everything above the base address */
fadump_reserve_crash_area(u64 base)1718b2a815a5SHari Bathini static void __init fadump_reserve_crash_area(u64 base)
1719b2a815a5SHari Bathini {
1720b10d6bcaSMike Rapoport u64 i, mstart, mend, msize;
1721b2a815a5SHari Bathini
1722b10d6bcaSMike Rapoport for_each_mem_range(i, &mstart, &mend) {
1723b10d6bcaSMike Rapoport msize = mend - mstart;
1724b2a815a5SHari Bathini
1725b2a815a5SHari Bathini if ((mstart + msize) < base)
1726b2a815a5SHari Bathini continue;
1727b2a815a5SHari Bathini
1728b2a815a5SHari Bathini if (mstart < base) {
1729b2a815a5SHari Bathini msize -= (base - mstart);
1730b2a815a5SHari Bathini mstart = base;
1731b2a815a5SHari Bathini }
1732b2a815a5SHari Bathini
1733b2a815a5SHari Bathini pr_info("Reserving %lluMB of memory at %#016llx for preserving crash data",
1734b2a815a5SHari Bathini (msize >> 20), mstart);
1735b2a815a5SHari Bathini memblock_reserve(mstart, msize);
1736b2a815a5SHari Bathini }
1737b2a815a5SHari Bathini }
1738bec53196SHari Bathini
arch_reserved_kernel_pages(void)1739bec53196SHari Bathini unsigned long __init arch_reserved_kernel_pages(void)
1740bec53196SHari Bathini {
1741bec53196SHari Bathini return memblock_reserved_size() / PAGE_SIZE;
1742bec53196SHari Bathini }
1743