1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * Architecture specific (i386/x86_64) functions for kexec based crash dumps. 4 * 5 * Created by: Hariprasad Nellitheertha (hari@in.ibm.com) 6 * 7 * Copyright (C) IBM Corporation, 2004. All rights reserved. 8 * Copyright (C) Red Hat Inc., 2014. All rights reserved. 9 * Authors: 10 * Vivek Goyal <vgoyal@redhat.com> 11 * 12 */ 13 14 #define pr_fmt(fmt) "kexec: " fmt 15 16 #include <linux/types.h> 17 #include <linux/kernel.h> 18 #include <linux/smp.h> 19 #include <linux/reboot.h> 20 #include <linux/kexec.h> 21 #include <linux/delay.h> 22 #include <linux/elf.h> 23 #include <linux/elfcore.h> 24 #include <linux/export.h> 25 #include <linux/slab.h> 26 #include <linux/vmalloc.h> 27 #include <linux/memblock.h> 28 29 #include <asm/processor.h> 30 #include <asm/hardirq.h> 31 #include <asm/nmi.h> 32 #include <asm/hw_irq.h> 33 #include <asm/apic.h> 34 #include <asm/e820/types.h> 35 #include <asm/io_apic.h> 36 #include <asm/hpet.h> 37 #include <linux/kdebug.h> 38 #include <asm/cpu.h> 39 #include <asm/reboot.h> 40 #include <asm/intel_pt.h> 41 #include <asm/crash.h> 42 #include <asm/cmdline.h> 43 44 /* Used while preparing memory map entries for second kernel */ 45 struct crash_memmap_data { 46 struct boot_params *params; 47 /* Type of memory */ 48 unsigned int type; 49 }; 50 51 /* 52 * This is used to VMCLEAR all VMCSs loaded on the 53 * processor. And when loading kvm_intel module, the 54 * callback function pointer will be assigned. 55 * 56 * protected by rcu. 57 */ 58 crash_vmclear_fn __rcu *crash_vmclear_loaded_vmcss = NULL; 59 EXPORT_SYMBOL_GPL(crash_vmclear_loaded_vmcss); 60 61 static inline void cpu_crash_vmclear_loaded_vmcss(void) 62 { 63 crash_vmclear_fn *do_vmclear_operation = NULL; 64 65 rcu_read_lock(); 66 do_vmclear_operation = rcu_dereference(crash_vmclear_loaded_vmcss); 67 if (do_vmclear_operation) 68 do_vmclear_operation(); 69 rcu_read_unlock(); 70 } 71 72 #if defined(CONFIG_SMP) && defined(CONFIG_X86_LOCAL_APIC) 73 74 static void kdump_nmi_callback(int cpu, struct pt_regs *regs) 75 { 76 crash_save_cpu(regs, cpu); 77 78 /* 79 * VMCLEAR VMCSs loaded on all cpus if needed. 80 */ 81 cpu_crash_vmclear_loaded_vmcss(); 82 83 /* 84 * Disable Intel PT to stop its logging 85 */ 86 cpu_emergency_stop_pt(); 87 88 disable_local_APIC(); 89 } 90 91 void kdump_nmi_shootdown_cpus(void) 92 { 93 nmi_shootdown_cpus(kdump_nmi_callback); 94 95 disable_local_APIC(); 96 } 97 98 /* Override the weak function in kernel/panic.c */ 99 void crash_smp_send_stop(void) 100 { 101 static int cpus_stopped; 102 103 if (cpus_stopped) 104 return; 105 106 if (smp_ops.crash_stop_other_cpus) 107 smp_ops.crash_stop_other_cpus(); 108 else 109 smp_send_stop(); 110 111 cpus_stopped = 1; 112 } 113 114 #else 115 void crash_smp_send_stop(void) 116 { 117 /* There are no cpus to shootdown */ 118 } 119 #endif 120 121 void native_machine_crash_shutdown(struct pt_regs *regs) 122 { 123 /* This function is only called after the system 124 * has panicked or is otherwise in a critical state. 125 * The minimum amount of code to allow a kexec'd kernel 126 * to run successfully needs to happen here. 127 * 128 * In practice this means shooting down the other cpus in 129 * an SMP system. 130 */ 131 /* The kernel is broken so disable interrupts */ 132 local_irq_disable(); 133 134 crash_smp_send_stop(); 135 136 /* 137 * VMCLEAR VMCSs loaded on this cpu if needed. 138 */ 139 cpu_crash_vmclear_loaded_vmcss(); 140 141 cpu_emergency_disable_virtualization(); 142 143 /* 144 * Disable Intel PT to stop its logging 145 */ 146 cpu_emergency_stop_pt(); 147 148 #ifdef CONFIG_X86_IO_APIC 149 /* Prevent crash_kexec() from deadlocking on ioapic_lock. */ 150 ioapic_zap_locks(); 151 clear_IO_APIC(); 152 #endif 153 lapic_shutdown(); 154 restore_boot_irq_mode(); 155 #ifdef CONFIG_HPET_TIMER 156 hpet_disable(); 157 #endif 158 crash_save_cpu(regs, safe_smp_processor_id()); 159 } 160 161 #ifdef CONFIG_KEXEC_FILE 162 163 static int get_nr_ram_ranges_callback(struct resource *res, void *arg) 164 { 165 unsigned int *nr_ranges = arg; 166 167 (*nr_ranges)++; 168 return 0; 169 } 170 171 /* Gather all the required information to prepare elf headers for ram regions */ 172 static struct crash_mem *fill_up_crash_elf_data(void) 173 { 174 unsigned int nr_ranges = 0; 175 struct crash_mem *cmem; 176 177 walk_system_ram_res(0, -1, &nr_ranges, get_nr_ram_ranges_callback); 178 if (!nr_ranges) 179 return NULL; 180 181 /* 182 * Exclusion of crash region and/or crashk_low_res may cause 183 * another range split. So add extra two slots here. 184 */ 185 nr_ranges += 2; 186 cmem = vzalloc(struct_size(cmem, ranges, nr_ranges)); 187 if (!cmem) 188 return NULL; 189 190 cmem->max_nr_ranges = nr_ranges; 191 cmem->nr_ranges = 0; 192 193 return cmem; 194 } 195 196 /* 197 * Look for any unwanted ranges between mstart, mend and remove them. This 198 * might lead to split and split ranges are put in cmem->ranges[] array 199 */ 200 static int elf_header_exclude_ranges(struct crash_mem *cmem) 201 { 202 int ret = 0; 203 204 /* Exclude the low 1M because it is always reserved */ 205 ret = crash_exclude_mem_range(cmem, 0, (1<<20)-1); 206 if (ret) 207 return ret; 208 209 /* Exclude crashkernel region */ 210 ret = crash_exclude_mem_range(cmem, crashk_res.start, crashk_res.end); 211 if (ret) 212 return ret; 213 214 if (crashk_low_res.end) 215 ret = crash_exclude_mem_range(cmem, crashk_low_res.start, 216 crashk_low_res.end); 217 218 return ret; 219 } 220 221 static int prepare_elf64_ram_headers_callback(struct resource *res, void *arg) 222 { 223 struct crash_mem *cmem = arg; 224 225 cmem->ranges[cmem->nr_ranges].start = res->start; 226 cmem->ranges[cmem->nr_ranges].end = res->end; 227 cmem->nr_ranges++; 228 229 return 0; 230 } 231 232 /* Prepare elf headers. Return addr and size */ 233 static int prepare_elf_headers(struct kimage *image, void **addr, 234 unsigned long *sz) 235 { 236 struct crash_mem *cmem; 237 int ret; 238 239 cmem = fill_up_crash_elf_data(); 240 if (!cmem) 241 return -ENOMEM; 242 243 ret = walk_system_ram_res(0, -1, cmem, prepare_elf64_ram_headers_callback); 244 if (ret) 245 goto out; 246 247 /* Exclude unwanted mem ranges */ 248 ret = elf_header_exclude_ranges(cmem); 249 if (ret) 250 goto out; 251 252 /* By default prepare 64bit headers */ 253 ret = crash_prepare_elf64_headers(cmem, IS_ENABLED(CONFIG_X86_64), addr, sz); 254 255 out: 256 vfree(cmem); 257 return ret; 258 } 259 260 static int add_e820_entry(struct boot_params *params, struct e820_entry *entry) 261 { 262 unsigned int nr_e820_entries; 263 264 nr_e820_entries = params->e820_entries; 265 if (nr_e820_entries >= E820_MAX_ENTRIES_ZEROPAGE) 266 return 1; 267 268 memcpy(¶ms->e820_table[nr_e820_entries], entry, sizeof(struct e820_entry)); 269 params->e820_entries++; 270 return 0; 271 } 272 273 static int memmap_entry_callback(struct resource *res, void *arg) 274 { 275 struct crash_memmap_data *cmd = arg; 276 struct boot_params *params = cmd->params; 277 struct e820_entry ei; 278 279 ei.addr = res->start; 280 ei.size = resource_size(res); 281 ei.type = cmd->type; 282 add_e820_entry(params, &ei); 283 284 return 0; 285 } 286 287 static int memmap_exclude_ranges(struct kimage *image, struct crash_mem *cmem, 288 unsigned long long mstart, 289 unsigned long long mend) 290 { 291 unsigned long start, end; 292 293 cmem->ranges[0].start = mstart; 294 cmem->ranges[0].end = mend; 295 cmem->nr_ranges = 1; 296 297 /* Exclude elf header region */ 298 start = image->elf_load_addr; 299 end = start + image->elf_headers_sz - 1; 300 return crash_exclude_mem_range(cmem, start, end); 301 } 302 303 /* Prepare memory map for crash dump kernel */ 304 int crash_setup_memmap_entries(struct kimage *image, struct boot_params *params) 305 { 306 int i, ret = 0; 307 unsigned long flags; 308 struct e820_entry ei; 309 struct crash_memmap_data cmd; 310 struct crash_mem *cmem; 311 312 cmem = vzalloc(struct_size(cmem, ranges, 1)); 313 if (!cmem) 314 return -ENOMEM; 315 316 memset(&cmd, 0, sizeof(struct crash_memmap_data)); 317 cmd.params = params; 318 319 /* Add the low 1M */ 320 cmd.type = E820_TYPE_RAM; 321 flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY; 322 walk_iomem_res_desc(IORES_DESC_NONE, flags, 0, (1<<20)-1, &cmd, 323 memmap_entry_callback); 324 325 /* Add ACPI tables */ 326 cmd.type = E820_TYPE_ACPI; 327 flags = IORESOURCE_MEM | IORESOURCE_BUSY; 328 walk_iomem_res_desc(IORES_DESC_ACPI_TABLES, flags, 0, -1, &cmd, 329 memmap_entry_callback); 330 331 /* Add ACPI Non-volatile Storage */ 332 cmd.type = E820_TYPE_NVS; 333 walk_iomem_res_desc(IORES_DESC_ACPI_NV_STORAGE, flags, 0, -1, &cmd, 334 memmap_entry_callback); 335 336 /* Add e820 reserved ranges */ 337 cmd.type = E820_TYPE_RESERVED; 338 flags = IORESOURCE_MEM; 339 walk_iomem_res_desc(IORES_DESC_RESERVED, flags, 0, -1, &cmd, 340 memmap_entry_callback); 341 342 /* Add crashk_low_res region */ 343 if (crashk_low_res.end) { 344 ei.addr = crashk_low_res.start; 345 ei.size = resource_size(&crashk_low_res); 346 ei.type = E820_TYPE_RAM; 347 add_e820_entry(params, &ei); 348 } 349 350 /* Exclude some ranges from crashk_res and add rest to memmap */ 351 ret = memmap_exclude_ranges(image, cmem, crashk_res.start, crashk_res.end); 352 if (ret) 353 goto out; 354 355 for (i = 0; i < cmem->nr_ranges; i++) { 356 ei.size = cmem->ranges[i].end - cmem->ranges[i].start + 1; 357 358 /* If entry is less than a page, skip it */ 359 if (ei.size < PAGE_SIZE) 360 continue; 361 ei.addr = cmem->ranges[i].start; 362 ei.type = E820_TYPE_RAM; 363 add_e820_entry(params, &ei); 364 } 365 366 out: 367 vfree(cmem); 368 return ret; 369 } 370 371 int crash_load_segments(struct kimage *image) 372 { 373 int ret; 374 struct kexec_buf kbuf = { .image = image, .buf_min = 0, 375 .buf_max = ULONG_MAX, .top_down = false }; 376 377 /* Prepare elf headers and add a segment */ 378 ret = prepare_elf_headers(image, &kbuf.buffer, &kbuf.bufsz); 379 if (ret) 380 return ret; 381 382 image->elf_headers = kbuf.buffer; 383 image->elf_headers_sz = kbuf.bufsz; 384 385 kbuf.memsz = kbuf.bufsz; 386 kbuf.buf_align = ELF_CORE_HEADER_ALIGN; 387 kbuf.mem = KEXEC_BUF_MEM_UNKNOWN; 388 ret = kexec_add_buffer(&kbuf); 389 if (ret) 390 return ret; 391 image->elf_load_addr = kbuf.mem; 392 pr_debug("Loaded ELF headers at 0x%lx bufsz=0x%lx memsz=0x%lx\n", 393 image->elf_load_addr, kbuf.bufsz, kbuf.memsz); 394 395 return ret; 396 } 397 #endif /* CONFIG_KEXEC_FILE */ 398