1 // SPDX-License-Identifier: GPL-2.0-only
3 * Architecture specific (i386/x86_64) functions for kexec based crash dumps.
5 * Created by: Hariprasad Nellitheertha (hari@in.ibm.com)
7 * Copyright (C) IBM Corporation, 2004. All rights reserved.
8 * Copyright (C) Red Hat Inc., 2014. All rights reserved.
10 * Vivek Goyal <vgoyal@redhat.com>
14 #define pr_fmt(fmt) "kexec: " fmt
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>
29 #include <asm/processor.h>
30 #include <asm/hardirq.h>
32 #include <asm/hw_irq.h>
34 #include <asm/e820/types.h>
35 #include <asm/io_apic.h>
37 #include <linux/kdebug.h>
39 #include <asm/reboot.h>
40 #include <asm/intel_pt.h>
41 #include <asm/crash.h>
42 #include <asm/cmdline.h>
44 /* Used while preparing memory map entries for second kernel */
45 struct crash_memmap_data {
46 struct boot_params *params;
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.
58 crash_vmclear_fn __rcu *crash_vmclear_loaded_vmcss = NULL;
59 EXPORT_SYMBOL_GPL(crash_vmclear_loaded_vmcss);
61 static inline void cpu_crash_vmclear_loaded_vmcss(void)
63 crash_vmclear_fn *do_vmclear_operation = NULL;
66 do_vmclear_operation = rcu_dereference(crash_vmclear_loaded_vmcss);
67 if (do_vmclear_operation)
68 do_vmclear_operation();
73 * When the crashkernel option is specified, only use the low
74 * 1M for the real mode trampoline.
76 void __init crash_reserve_low_1M(void)
78 if (cmdline_find_option(boot_command_line, "crashkernel", NULL, 0) < 0)
81 memblock_reserve(0, 1<<20);
82 pr_info("Reserving the low 1M of memory for crashkernel\n");
85 #if defined(CONFIG_SMP) && defined(CONFIG_X86_LOCAL_APIC)
87 static void kdump_nmi_callback(int cpu, struct pt_regs *regs)
89 crash_save_cpu(regs, cpu);
92 * VMCLEAR VMCSs loaded on all cpus if needed.
94 cpu_crash_vmclear_loaded_vmcss();
97 * Disable Intel PT to stop its logging
99 cpu_emergency_stop_pt();
101 disable_local_APIC();
104 void kdump_nmi_shootdown_cpus(void)
106 nmi_shootdown_cpus(kdump_nmi_callback);
108 disable_local_APIC();
111 /* Override the weak function in kernel/panic.c */
112 void crash_smp_send_stop(void)
114 static int cpus_stopped;
119 if (smp_ops.crash_stop_other_cpus)
120 smp_ops.crash_stop_other_cpus();
128 void crash_smp_send_stop(void)
130 /* There are no cpus to shootdown */
134 void native_machine_crash_shutdown(struct pt_regs *regs)
136 /* This function is only called after the system
137 * has panicked or is otherwise in a critical state.
138 * The minimum amount of code to allow a kexec'd kernel
139 * to run successfully needs to happen here.
141 * In practice this means shooting down the other cpus in
144 /* The kernel is broken so disable interrupts */
147 crash_smp_send_stop();
150 * VMCLEAR VMCSs loaded on this cpu if needed.
152 cpu_crash_vmclear_loaded_vmcss();
154 cpu_emergency_disable_virtualization();
157 * Disable Intel PT to stop its logging
159 cpu_emergency_stop_pt();
161 #ifdef CONFIG_X86_IO_APIC
162 /* Prevent crash_kexec() from deadlocking on ioapic_lock. */
167 restore_boot_irq_mode();
168 #ifdef CONFIG_HPET_TIMER
171 crash_save_cpu(regs, safe_smp_processor_id());
174 #ifdef CONFIG_KEXEC_FILE
176 static unsigned long crash_zero_bytes;
178 static int get_nr_ram_ranges_callback(struct resource *res, void *arg)
180 unsigned int *nr_ranges = arg;
186 /* Gather all the required information to prepare elf headers for ram regions */
187 static struct crash_mem *fill_up_crash_elf_data(void)
189 unsigned int nr_ranges = 0;
190 struct crash_mem *cmem;
192 walk_system_ram_res(0, -1, &nr_ranges,
193 get_nr_ram_ranges_callback);
198 * Exclusion of crash region and/or crashk_low_res may cause
199 * another range split. So add extra two slots here.
202 cmem = vzalloc(struct_size(cmem, ranges, nr_ranges));
206 cmem->max_nr_ranges = nr_ranges;
213 * Look for any unwanted ranges between mstart, mend and remove them. This
214 * might lead to split and split ranges are put in cmem->ranges[] array
216 static int elf_header_exclude_ranges(struct crash_mem *cmem)
220 /* Exclude crashkernel region */
221 ret = crash_exclude_mem_range(cmem, crashk_res.start, crashk_res.end);
225 if (crashk_low_res.end) {
226 ret = crash_exclude_mem_range(cmem, crashk_low_res.start,
233 static int prepare_elf64_ram_headers_callback(struct resource *res, void *arg)
235 struct crash_mem *cmem = arg;
237 cmem->ranges[cmem->nr_ranges].start = res->start;
238 cmem->ranges[cmem->nr_ranges].end = res->end;
244 /* Prepare elf headers. Return addr and size */
245 static int prepare_elf_headers(struct kimage *image, void **addr,
248 struct crash_mem *cmem;
253 cmem = fill_up_crash_elf_data();
257 ret = walk_system_ram_res(0, -1, cmem,
258 prepare_elf64_ram_headers_callback);
262 /* Exclude unwanted mem ranges */
263 ret = elf_header_exclude_ranges(cmem);
267 /* By default prepare 64bit headers */
268 ret = crash_prepare_elf64_headers(cmem,
269 IS_ENABLED(CONFIG_X86_64), addr, sz);
274 * If a range matches backup region, adjust offset to backup
277 ehdr = (Elf64_Ehdr *)*addr;
278 phdr = (Elf64_Phdr *)(ehdr + 1);
279 for (i = 0; i < ehdr->e_phnum; phdr++, i++)
280 if (phdr->p_type == PT_LOAD &&
281 phdr->p_paddr == image->arch.backup_src_start &&
282 phdr->p_memsz == image->arch.backup_src_sz) {
283 phdr->p_offset = image->arch.backup_load_addr;
291 static int add_e820_entry(struct boot_params *params, struct e820_entry *entry)
293 unsigned int nr_e820_entries;
295 nr_e820_entries = params->e820_entries;
296 if (nr_e820_entries >= E820_MAX_ENTRIES_ZEROPAGE)
299 memcpy(¶ms->e820_table[nr_e820_entries], entry,
300 sizeof(struct e820_entry));
301 params->e820_entries++;
305 static int memmap_entry_callback(struct resource *res, void *arg)
307 struct crash_memmap_data *cmd = arg;
308 struct boot_params *params = cmd->params;
309 struct e820_entry ei;
311 ei.addr = res->start;
312 ei.size = resource_size(res);
314 add_e820_entry(params, &ei);
319 static int memmap_exclude_ranges(struct kimage *image, struct crash_mem *cmem,
320 unsigned long long mstart,
321 unsigned long long mend)
323 unsigned long start, end;
326 cmem->ranges[0].start = mstart;
327 cmem->ranges[0].end = mend;
330 /* Exclude Backup region */
331 start = image->arch.backup_load_addr;
332 end = start + image->arch.backup_src_sz - 1;
333 ret = crash_exclude_mem_range(cmem, start, end);
337 /* Exclude elf header region */
338 start = image->arch.elf_load_addr;
339 end = start + image->arch.elf_headers_sz - 1;
340 return crash_exclude_mem_range(cmem, start, end);
343 /* Prepare memory map for crash dump kernel */
344 int crash_setup_memmap_entries(struct kimage *image, struct boot_params *params)
348 struct e820_entry ei;
349 struct crash_memmap_data cmd;
350 struct crash_mem *cmem;
352 cmem = vzalloc(struct_size(cmem, ranges, 1));
356 memset(&cmd, 0, sizeof(struct crash_memmap_data));
359 /* Add first 640K segment */
360 ei.addr = image->arch.backup_src_start;
361 ei.size = image->arch.backup_src_sz;
362 ei.type = E820_TYPE_RAM;
363 add_e820_entry(params, &ei);
365 /* Add ACPI tables */
366 cmd.type = E820_TYPE_ACPI;
367 flags = IORESOURCE_MEM | IORESOURCE_BUSY;
368 walk_iomem_res_desc(IORES_DESC_ACPI_TABLES, flags, 0, -1, &cmd,
369 memmap_entry_callback);
371 /* Add ACPI Non-volatile Storage */
372 cmd.type = E820_TYPE_NVS;
373 walk_iomem_res_desc(IORES_DESC_ACPI_NV_STORAGE, flags, 0, -1, &cmd,
374 memmap_entry_callback);
376 /* Add e820 reserved ranges */
377 cmd.type = E820_TYPE_RESERVED;
378 flags = IORESOURCE_MEM;
379 walk_iomem_res_desc(IORES_DESC_RESERVED, flags, 0, -1, &cmd,
380 memmap_entry_callback);
382 /* Add crashk_low_res region */
383 if (crashk_low_res.end) {
384 ei.addr = crashk_low_res.start;
385 ei.size = crashk_low_res.end - crashk_low_res.start + 1;
386 ei.type = E820_TYPE_RAM;
387 add_e820_entry(params, &ei);
390 /* Exclude some ranges from crashk_res and add rest to memmap */
391 ret = memmap_exclude_ranges(image, cmem, crashk_res.start,
396 for (i = 0; i < cmem->nr_ranges; i++) {
397 ei.size = cmem->ranges[i].end - cmem->ranges[i].start + 1;
399 /* If entry is less than a page, skip it */
400 if (ei.size < PAGE_SIZE)
402 ei.addr = cmem->ranges[i].start;
403 ei.type = E820_TYPE_RAM;
404 add_e820_entry(params, &ei);
412 static int determine_backup_region(struct resource *res, void *arg)
414 struct kimage *image = arg;
416 image->arch.backup_src_start = res->start;
417 image->arch.backup_src_sz = resource_size(res);
419 /* Expecting only one range for backup region */
423 int crash_load_segments(struct kimage *image)
426 struct kexec_buf kbuf = { .image = image, .buf_min = 0,
427 .buf_max = ULONG_MAX, .top_down = false };
430 * Determine and load a segment for backup area. First 640K RAM
431 * region is backup source
434 ret = walk_system_ram_res(KEXEC_BACKUP_SRC_START, KEXEC_BACKUP_SRC_END,
435 image, determine_backup_region);
437 /* Zero or postive return values are ok */
441 /* Add backup segment. */
442 if (image->arch.backup_src_sz) {
443 kbuf.buffer = &crash_zero_bytes;
444 kbuf.bufsz = sizeof(crash_zero_bytes);
445 kbuf.memsz = image->arch.backup_src_sz;
446 kbuf.buf_align = PAGE_SIZE;
448 * Ideally there is no source for backup segment. This is
449 * copied in purgatory after crash. Just add a zero filled
450 * segment for now to make sure checksum logic works fine.
452 ret = kexec_add_buffer(&kbuf);
455 image->arch.backup_load_addr = kbuf.mem;
456 pr_debug("Loaded backup region at 0x%lx backup_start=0x%lx memsz=0x%lx\n",
457 image->arch.backup_load_addr,
458 image->arch.backup_src_start, kbuf.memsz);
461 /* Prepare elf headers and add a segment */
462 ret = prepare_elf_headers(image, &kbuf.buffer, &kbuf.bufsz);
466 image->arch.elf_headers = kbuf.buffer;
467 image->arch.elf_headers_sz = kbuf.bufsz;
469 kbuf.memsz = kbuf.bufsz;
470 kbuf.buf_align = ELF_CORE_HEADER_ALIGN;
471 kbuf.mem = KEXEC_BUF_MEM_UNKNOWN;
472 ret = kexec_add_buffer(&kbuf);
474 vfree((void *)image->arch.elf_headers);
477 image->arch.elf_load_addr = kbuf.mem;
478 pr_debug("Loaded ELF headers at 0x%lx bufsz=0x%lx memsz=0x%lx\n",
479 image->arch.elf_load_addr, kbuf.bufsz, kbuf.bufsz);
483 #endif /* CONFIG_KEXEC_FILE */