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;
51 #if defined(CONFIG_SMP) && defined(CONFIG_X86_LOCAL_APIC)
53 static void kdump_nmi_callback(int cpu, struct pt_regs *regs)
55 crash_save_cpu(regs, cpu);
58 * Disable Intel PT to stop its logging
60 cpu_emergency_stop_pt();
65 void kdump_nmi_shootdown_cpus(void)
67 nmi_shootdown_cpus(kdump_nmi_callback);
72 /* Override the weak function in kernel/panic.c */
73 void crash_smp_send_stop(void)
75 static int cpus_stopped;
80 if (smp_ops.crash_stop_other_cpus)
81 smp_ops.crash_stop_other_cpus();
89 void crash_smp_send_stop(void)
91 /* There are no cpus to shootdown */
95 void native_machine_crash_shutdown(struct pt_regs *regs)
97 /* This function is only called after the system
98 * has panicked or is otherwise in a critical state.
99 * The minimum amount of code to allow a kexec'd kernel
100 * to run successfully needs to happen here.
102 * In practice this means shooting down the other cpus in
105 /* The kernel is broken so disable interrupts */
108 crash_smp_send_stop();
110 cpu_emergency_disable_virtualization();
113 * Disable Intel PT to stop its logging
115 cpu_emergency_stop_pt();
117 #ifdef CONFIG_X86_IO_APIC
118 /* Prevent crash_kexec() from deadlocking on ioapic_lock. */
123 restore_boot_irq_mode();
124 #ifdef CONFIG_HPET_TIMER
127 crash_save_cpu(regs, safe_smp_processor_id());
130 #if defined(CONFIG_KEXEC_FILE) || defined(CONFIG_CRASH_HOTPLUG)
131 static int get_nr_ram_ranges_callback(struct resource *res, void *arg)
133 unsigned int *nr_ranges = arg;
139 /* Gather all the required information to prepare elf headers for ram regions */
140 static struct crash_mem *fill_up_crash_elf_data(void)
142 unsigned int nr_ranges = 0;
143 struct crash_mem *cmem;
145 walk_system_ram_res(0, -1, &nr_ranges, get_nr_ram_ranges_callback);
150 * Exclusion of crash region and/or crashk_low_res may cause
151 * another range split. So add extra two slots here.
154 cmem = vzalloc(struct_size(cmem, ranges, nr_ranges));
158 cmem->max_nr_ranges = nr_ranges;
165 * Look for any unwanted ranges between mstart, mend and remove them. This
166 * might lead to split and split ranges are put in cmem->ranges[] array
168 static int elf_header_exclude_ranges(struct crash_mem *cmem)
172 /* Exclude the low 1M because it is always reserved */
173 ret = crash_exclude_mem_range(cmem, 0, SZ_1M - 1);
177 /* Exclude crashkernel region */
178 ret = crash_exclude_mem_range(cmem, crashk_res.start, crashk_res.end);
182 if (crashk_low_res.end)
183 ret = crash_exclude_mem_range(cmem, crashk_low_res.start,
189 static int prepare_elf64_ram_headers_callback(struct resource *res, void *arg)
191 struct crash_mem *cmem = arg;
193 cmem->ranges[cmem->nr_ranges].start = res->start;
194 cmem->ranges[cmem->nr_ranges].end = res->end;
200 /* Prepare elf headers. Return addr and size */
201 static int prepare_elf_headers(void **addr, unsigned long *sz,
202 unsigned long *nr_mem_ranges)
204 struct crash_mem *cmem;
207 cmem = fill_up_crash_elf_data();
211 ret = walk_system_ram_res(0, -1, cmem, prepare_elf64_ram_headers_callback);
215 /* Exclude unwanted mem ranges */
216 ret = elf_header_exclude_ranges(cmem);
220 /* Return the computed number of memory ranges, for hotplug usage */
221 *nr_mem_ranges = cmem->nr_ranges;
223 /* By default prepare 64bit headers */
224 ret = crash_prepare_elf64_headers(cmem, IS_ENABLED(CONFIG_X86_64), addr, sz);
232 #ifdef CONFIG_KEXEC_FILE
233 static int add_e820_entry(struct boot_params *params, struct e820_entry *entry)
235 unsigned int nr_e820_entries;
237 nr_e820_entries = params->e820_entries;
238 if (nr_e820_entries >= E820_MAX_ENTRIES_ZEROPAGE)
241 memcpy(¶ms->e820_table[nr_e820_entries], entry, sizeof(struct e820_entry));
242 params->e820_entries++;
246 static int memmap_entry_callback(struct resource *res, void *arg)
248 struct crash_memmap_data *cmd = arg;
249 struct boot_params *params = cmd->params;
250 struct e820_entry ei;
252 ei.addr = res->start;
253 ei.size = resource_size(res);
255 add_e820_entry(params, &ei);
260 static int memmap_exclude_ranges(struct kimage *image, struct crash_mem *cmem,
261 unsigned long long mstart,
262 unsigned long long mend)
264 unsigned long start, end;
266 cmem->ranges[0].start = mstart;
267 cmem->ranges[0].end = mend;
270 /* Exclude elf header region */
271 start = image->elf_load_addr;
272 end = start + image->elf_headers_sz - 1;
273 return crash_exclude_mem_range(cmem, start, end);
276 /* Prepare memory map for crash dump kernel */
277 int crash_setup_memmap_entries(struct kimage *image, struct boot_params *params)
281 struct e820_entry ei;
282 struct crash_memmap_data cmd;
283 struct crash_mem *cmem;
285 cmem = vzalloc(struct_size(cmem, ranges, 1));
289 memset(&cmd, 0, sizeof(struct crash_memmap_data));
293 cmd.type = E820_TYPE_RAM;
294 flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
295 walk_iomem_res_desc(IORES_DESC_NONE, flags, 0, (1<<20)-1, &cmd,
296 memmap_entry_callback);
298 /* Add ACPI tables */
299 cmd.type = E820_TYPE_ACPI;
300 flags = IORESOURCE_MEM | IORESOURCE_BUSY;
301 walk_iomem_res_desc(IORES_DESC_ACPI_TABLES, flags, 0, -1, &cmd,
302 memmap_entry_callback);
304 /* Add ACPI Non-volatile Storage */
305 cmd.type = E820_TYPE_NVS;
306 walk_iomem_res_desc(IORES_DESC_ACPI_NV_STORAGE, flags, 0, -1, &cmd,
307 memmap_entry_callback);
309 /* Add e820 reserved ranges */
310 cmd.type = E820_TYPE_RESERVED;
311 flags = IORESOURCE_MEM;
312 walk_iomem_res_desc(IORES_DESC_RESERVED, flags, 0, -1, &cmd,
313 memmap_entry_callback);
315 /* Add crashk_low_res region */
316 if (crashk_low_res.end) {
317 ei.addr = crashk_low_res.start;
318 ei.size = resource_size(&crashk_low_res);
319 ei.type = E820_TYPE_RAM;
320 add_e820_entry(params, &ei);
323 /* Exclude some ranges from crashk_res and add rest to memmap */
324 ret = memmap_exclude_ranges(image, cmem, crashk_res.start, crashk_res.end);
328 for (i = 0; i < cmem->nr_ranges; i++) {
329 ei.size = cmem->ranges[i].end - cmem->ranges[i].start + 1;
331 /* If entry is less than a page, skip it */
332 if (ei.size < PAGE_SIZE)
334 ei.addr = cmem->ranges[i].start;
335 ei.type = E820_TYPE_RAM;
336 add_e820_entry(params, &ei);
344 int crash_load_segments(struct kimage *image)
347 unsigned long pnum = 0;
348 struct kexec_buf kbuf = { .image = image, .buf_min = 0,
349 .buf_max = ULONG_MAX, .top_down = false };
351 /* Prepare elf headers and add a segment */
352 ret = prepare_elf_headers(&kbuf.buffer, &kbuf.bufsz, &pnum);
356 image->elf_headers = kbuf.buffer;
357 image->elf_headers_sz = kbuf.bufsz;
358 kbuf.memsz = kbuf.bufsz;
360 #ifdef CONFIG_CRASH_HOTPLUG
362 * The elfcorehdr segment size accounts for VMCOREINFO, kernel_map,
363 * maximum CPUs and maximum memory ranges.
365 if (IS_ENABLED(CONFIG_MEMORY_HOTPLUG))
366 pnum = 2 + CONFIG_NR_CPUS_DEFAULT + CONFIG_CRASH_MAX_MEMORY_RANGES;
368 pnum += 2 + CONFIG_NR_CPUS_DEFAULT;
370 if (pnum < (unsigned long)PN_XNUM) {
371 kbuf.memsz = pnum * sizeof(Elf64_Phdr);
372 kbuf.memsz += sizeof(Elf64_Ehdr);
374 image->elfcorehdr_index = image->nr_segments;
376 /* Mark as usable to crash kernel, else crash kernel fails on boot */
377 image->elf_headers_sz = kbuf.memsz;
379 pr_err("number of Phdrs %lu exceeds max\n", pnum);
383 kbuf.buf_align = ELF_CORE_HEADER_ALIGN;
384 kbuf.mem = KEXEC_BUF_MEM_UNKNOWN;
385 ret = kexec_add_buffer(&kbuf);
388 image->elf_load_addr = kbuf.mem;
389 kexec_dprintk("Loaded ELF headers at 0x%lx bufsz=0x%lx memsz=0x%lx\n",
390 image->elf_load_addr, kbuf.bufsz, kbuf.memsz);
394 #endif /* CONFIG_KEXEC_FILE */
396 #ifdef CONFIG_CRASH_HOTPLUG
399 #define pr_fmt(fmt) "crash hp: " fmt
401 /* These functions provide the value for the sysfs crash_hotplug nodes */
402 #ifdef CONFIG_HOTPLUG_CPU
403 int arch_crash_hotplug_cpu_support(void)
405 return crash_check_update_elfcorehdr();
409 #ifdef CONFIG_MEMORY_HOTPLUG
410 int arch_crash_hotplug_memory_support(void)
412 return crash_check_update_elfcorehdr();
416 unsigned int arch_crash_get_elfcorehdr_size(void)
420 /* kernel_map, VMCOREINFO and maximum CPUs */
421 sz = 2 + CONFIG_NR_CPUS_DEFAULT;
422 if (IS_ENABLED(CONFIG_MEMORY_HOTPLUG))
423 sz += CONFIG_CRASH_MAX_MEMORY_RANGES;
424 sz *= sizeof(Elf64_Phdr);
429 * arch_crash_handle_hotplug_event() - Handle hotplug elfcorehdr changes
430 * @image: a pointer to kexec_crash_image
432 * Prepare the new elfcorehdr and replace the existing elfcorehdr.
434 void arch_crash_handle_hotplug_event(struct kimage *image)
436 void *elfbuf = NULL, *old_elfcorehdr;
437 unsigned long nr_mem_ranges;
438 unsigned long mem, memsz;
439 unsigned long elfsz = 0;
442 * As crash_prepare_elf64_headers() has already described all
443 * possible CPUs, there is no need to update the elfcorehdr
444 * for additional CPU changes.
446 if ((image->file_mode || image->elfcorehdr_updated) &&
447 ((image->hp_action == KEXEC_CRASH_HP_ADD_CPU) ||
448 (image->hp_action == KEXEC_CRASH_HP_REMOVE_CPU)))
452 * Create the new elfcorehdr reflecting the changes to CPU and/or
455 if (prepare_elf_headers(&elfbuf, &elfsz, &nr_mem_ranges)) {
456 pr_err("unable to create new elfcorehdr");
461 * Obtain address and size of the elfcorehdr segment, and
462 * check it against the new elfcorehdr buffer.
464 mem = image->segment[image->elfcorehdr_index].mem;
465 memsz = image->segment[image->elfcorehdr_index].memsz;
467 pr_err("update elfcorehdr elfsz %lu > memsz %lu",
473 * Copy new elfcorehdr over the old elfcorehdr at destination.
475 old_elfcorehdr = kmap_local_page(pfn_to_page(mem >> PAGE_SHIFT));
476 if (!old_elfcorehdr) {
477 pr_err("mapping elfcorehdr segment failed\n");
482 * Temporarily invalidate the crash image while the
483 * elfcorehdr is updated.
485 xchg(&kexec_crash_image, NULL);
486 memcpy_flushcache(old_elfcorehdr, elfbuf, elfsz);
487 xchg(&kexec_crash_image, image);
488 kunmap_local(old_elfcorehdr);
489 pr_debug("updated elfcorehdr\n");