|
| 1 | +/* |
| 2 | + * Firmware Assisted dump: A robust mechanism to get reliable kernel crash |
| 3 | + * dump with assistance from firmware. This approach does not use kexec, |
| 4 | + * instead firmware assists in booting the kdump kernel while preserving |
| 5 | + * memory contents. The most of the code implementation has been adapted |
| 6 | + * from phyp assisted dump implementation written by Linas Vepstas and |
| 7 | + * Manish Ahuja |
| 8 | + * |
| 9 | + * This program is free software; you can redistribute it and/or modify |
| 10 | + * it under the terms of the GNU General Public License as published by |
| 11 | + * the Free Software Foundation; either version 2 of the License, or |
| 12 | + * (at your option) any later version. |
| 13 | + * |
| 14 | + * This program is distributed in the hope that it will be useful, |
| 15 | + * but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 16 | + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
| 17 | + * GNU General Public License for more details. |
| 18 | + * |
| 19 | + * You should have received a copy of the GNU General Public License |
| 20 | + * along with this program; if not, write to the Free Software |
| 21 | + * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. |
| 22 | + * |
| 23 | + * Copyright 2011 IBM Corporation |
| 24 | + * Author: Mahesh Salgaonkar <[email protected]> |
| 25 | + */ |
| 26 | + |
| 27 | +#undef DEBUG |
| 28 | +#define pr_fmt(fmt) "fadump: " fmt |
| 29 | + |
| 30 | +#include <linux/string.h> |
| 31 | +#include <linux/memblock.h> |
| 32 | + |
| 33 | +#include <asm/page.h> |
| 34 | +#include <asm/prom.h> |
| 35 | +#include <asm/rtas.h> |
| 36 | +#include <asm/fadump.h> |
| 37 | + |
| 38 | +static struct fw_dump fw_dump; |
| 39 | + |
| 40 | +/* Scan the Firmware Assisted dump configuration details. */ |
| 41 | +int __init early_init_dt_scan_fw_dump(unsigned long node, |
| 42 | + const char *uname, int depth, void *data) |
| 43 | +{ |
| 44 | + __be32 *sections; |
| 45 | + int i, num_sections; |
| 46 | + unsigned long size; |
| 47 | + const int *token; |
| 48 | + |
| 49 | + if (depth != 1 || strcmp(uname, "rtas") != 0) |
| 50 | + return 0; |
| 51 | + |
| 52 | + /* |
| 53 | + * Check if Firmware Assisted dump is supported. if yes, check |
| 54 | + * if dump has been initiated on last reboot. |
| 55 | + */ |
| 56 | + token = of_get_flat_dt_prop(node, "ibm,configure-kernel-dump", NULL); |
| 57 | + if (!token) |
| 58 | + return 0; |
| 59 | + |
| 60 | + fw_dump.fadump_supported = 1; |
| 61 | + fw_dump.ibm_configure_kernel_dump = *token; |
| 62 | + |
| 63 | + /* |
| 64 | + * The 'ibm,kernel-dump' rtas node is present only if there is |
| 65 | + * dump data waiting for us. |
| 66 | + */ |
| 67 | + if (of_get_flat_dt_prop(node, "ibm,kernel-dump", NULL)) |
| 68 | + fw_dump.dump_active = 1; |
| 69 | + |
| 70 | + /* Get the sizes required to store dump data for the firmware provided |
| 71 | + * dump sections. |
| 72 | + * For each dump section type supported, a 32bit cell which defines |
| 73 | + * the ID of a supported section followed by two 32 bit cells which |
| 74 | + * gives teh size of the section in bytes. |
| 75 | + */ |
| 76 | + sections = of_get_flat_dt_prop(node, "ibm,configure-kernel-dump-sizes", |
| 77 | + &size); |
| 78 | + |
| 79 | + if (!sections) |
| 80 | + return 0; |
| 81 | + |
| 82 | + num_sections = size / (3 * sizeof(u32)); |
| 83 | + |
| 84 | + for (i = 0; i < num_sections; i++, sections += 3) { |
| 85 | + u32 type = (u32)of_read_number(sections, 1); |
| 86 | + |
| 87 | + switch (type) { |
| 88 | + case FADUMP_CPU_STATE_DATA: |
| 89 | + fw_dump.cpu_state_data_size = |
| 90 | + of_read_ulong(§ions[1], 2); |
| 91 | + break; |
| 92 | + case FADUMP_HPTE_REGION: |
| 93 | + fw_dump.hpte_region_size = |
| 94 | + of_read_ulong(§ions[1], 2); |
| 95 | + break; |
| 96 | + } |
| 97 | + } |
| 98 | + return 1; |
| 99 | +} |
| 100 | + |
| 101 | +/** |
| 102 | + * fadump_calculate_reserve_size(): reserve variable boot area 5% of System RAM |
| 103 | + * |
| 104 | + * Function to find the largest memory size we need to reserve during early |
| 105 | + * boot process. This will be the size of the memory that is required for a |
| 106 | + * kernel to boot successfully. |
| 107 | + * |
| 108 | + * This function has been taken from phyp-assisted dump feature implementation. |
| 109 | + * |
| 110 | + * returns larger of 256MB or 5% rounded down to multiples of 256MB. |
| 111 | + * |
| 112 | + * TODO: Come up with better approach to find out more accurate memory size |
| 113 | + * that is required for a kernel to boot successfully. |
| 114 | + * |
| 115 | + */ |
| 116 | +static inline unsigned long fadump_calculate_reserve_size(void) |
| 117 | +{ |
| 118 | + unsigned long size; |
| 119 | + |
| 120 | + /* |
| 121 | + * Check if the size is specified through fadump_reserve_mem= cmdline |
| 122 | + * option. If yes, then use that. |
| 123 | + */ |
| 124 | + if (fw_dump.reserve_bootvar) |
| 125 | + return fw_dump.reserve_bootvar; |
| 126 | + |
| 127 | + /* divide by 20 to get 5% of value */ |
| 128 | + size = memblock_end_of_DRAM() / 20; |
| 129 | + |
| 130 | + /* round it down in multiples of 256 */ |
| 131 | + size = size & ~0x0FFFFFFFUL; |
| 132 | + |
| 133 | + /* Truncate to memory_limit. We don't want to over reserve the memory.*/ |
| 134 | + if (memory_limit && size > memory_limit) |
| 135 | + size = memory_limit; |
| 136 | + |
| 137 | + return (size > MIN_BOOT_MEM ? size : MIN_BOOT_MEM); |
| 138 | +} |
| 139 | + |
| 140 | +/* |
| 141 | + * Calculate the total memory size required to be reserved for |
| 142 | + * firmware-assisted dump registration. |
| 143 | + */ |
| 144 | +static unsigned long get_fadump_area_size(void) |
| 145 | +{ |
| 146 | + unsigned long size = 0; |
| 147 | + |
| 148 | + size += fw_dump.cpu_state_data_size; |
| 149 | + size += fw_dump.hpte_region_size; |
| 150 | + size += fw_dump.boot_memory_size; |
| 151 | + |
| 152 | + size = PAGE_ALIGN(size); |
| 153 | + return size; |
| 154 | +} |
| 155 | + |
| 156 | +int __init fadump_reserve_mem(void) |
| 157 | +{ |
| 158 | + unsigned long base, size, memory_boundary; |
| 159 | + |
| 160 | + if (!fw_dump.fadump_enabled) |
| 161 | + return 0; |
| 162 | + |
| 163 | + if (!fw_dump.fadump_supported) { |
| 164 | + printk(KERN_INFO "Firmware-assisted dump is not supported on" |
| 165 | + " this hardware\n"); |
| 166 | + fw_dump.fadump_enabled = 0; |
| 167 | + return 0; |
| 168 | + } |
| 169 | + /* Initialize boot memory size */ |
| 170 | + fw_dump.boot_memory_size = fadump_calculate_reserve_size(); |
| 171 | + |
| 172 | + /* |
| 173 | + * Calculate the memory boundary. |
| 174 | + * If memory_limit is less than actual memory boundary then reserve |
| 175 | + * the memory for fadump beyond the memory_limit and adjust the |
| 176 | + * memory_limit accordingly, so that the running kernel can run with |
| 177 | + * specified memory_limit. |
| 178 | + */ |
| 179 | + if (memory_limit && memory_limit < memblock_end_of_DRAM()) { |
| 180 | + size = get_fadump_area_size(); |
| 181 | + if ((memory_limit + size) < memblock_end_of_DRAM()) |
| 182 | + memory_limit += size; |
| 183 | + else |
| 184 | + memory_limit = memblock_end_of_DRAM(); |
| 185 | + printk(KERN_INFO "Adjusted memory_limit for firmware-assisted" |
| 186 | + " dump, now %#016llx\n", |
| 187 | + (unsigned long long)memory_limit); |
| 188 | + } |
| 189 | + if (memory_limit) |
| 190 | + memory_boundary = memory_limit; |
| 191 | + else |
| 192 | + memory_boundary = memblock_end_of_DRAM(); |
| 193 | + |
| 194 | + if (fw_dump.dump_active) { |
| 195 | + printk(KERN_INFO "Firmware-assisted dump is active.\n"); |
| 196 | + /* |
| 197 | + * If last boot has crashed then reserve all the memory |
| 198 | + * above boot_memory_size so that we don't touch it until |
| 199 | + * dump is written to disk by userspace tool. This memory |
| 200 | + * will be released for general use once the dump is saved. |
| 201 | + */ |
| 202 | + base = fw_dump.boot_memory_size; |
| 203 | + size = memory_boundary - base; |
| 204 | + memblock_reserve(base, size); |
| 205 | + printk(KERN_INFO "Reserved %ldMB of memory at %ldMB " |
| 206 | + "for saving crash dump\n", |
| 207 | + (unsigned long)(size >> 20), |
| 208 | + (unsigned long)(base >> 20)); |
| 209 | + } else { |
| 210 | + /* Reserve the memory at the top of memory. */ |
| 211 | + size = get_fadump_area_size(); |
| 212 | + base = memory_boundary - size; |
| 213 | + memblock_reserve(base, size); |
| 214 | + printk(KERN_INFO "Reserved %ldMB of memory at %ldMB " |
| 215 | + "for firmware-assisted dump\n", |
| 216 | + (unsigned long)(size >> 20), |
| 217 | + (unsigned long)(base >> 20)); |
| 218 | + } |
| 219 | + fw_dump.reserve_dump_area_start = base; |
| 220 | + fw_dump.reserve_dump_area_size = size; |
| 221 | + return 1; |
| 222 | +} |
| 223 | + |
| 224 | +/* Look for fadump= cmdline option. */ |
| 225 | +static int __init early_fadump_param(char *p) |
| 226 | +{ |
| 227 | + if (!p) |
| 228 | + return 1; |
| 229 | + |
| 230 | + if (strncmp(p, "on", 2) == 0) |
| 231 | + fw_dump.fadump_enabled = 1; |
| 232 | + else if (strncmp(p, "off", 3) == 0) |
| 233 | + fw_dump.fadump_enabled = 0; |
| 234 | + |
| 235 | + return 0; |
| 236 | +} |
| 237 | +early_param("fadump", early_fadump_param); |
| 238 | + |
| 239 | +/* Look for fadump_reserve_mem= cmdline option */ |
| 240 | +static int __init early_fadump_reserve_mem(char *p) |
| 241 | +{ |
| 242 | + if (p) |
| 243 | + fw_dump.reserve_bootvar = memparse(p, &p); |
| 244 | + return 0; |
| 245 | +} |
| 246 | +early_param("fadump_reserve_mem", early_fadump_reserve_mem); |
0 commit comments