2 * Architecture specific (i386/x86_64) functions for kexec based crash dumps.
4 * Created by: Hariprasad Nellitheertha (hari@in.ibm.com)
6 * Copyright (C) IBM Corporation, 2004. All rights reserved.
7 * Copyright (C) Red Hat Inc., 2014. All rights reserved.
9 * Vivek Goyal <vgoyal@redhat.com>
13 #define pr_fmt(fmt) "kexec: " fmt
15 #include <linux/types.h>
16 #include <linux/kernel.h>
17 #include <linux/smp.h>
18 #include <linux/reboot.h>
19 #include <linux/kexec.h>
20 #include <linux/delay.h>
21 #include <linux/elf.h>
22 #include <linux/elfcore.h>
23 #include <linux/module.h>
24 #include <linux/slab.h>
25 #include <linux/vmalloc.h>
27 #include <asm/processor.h>
28 #include <asm/hardirq.h>
30 #include <asm/hw_irq.h>
32 #include <asm/io_apic.h>
34 #include <linux/kdebug.h>
36 #include <asm/reboot.h>
37 #include <asm/virtext.h>
38 #include <asm/intel_pt.h>
40 /* Alignment required for elf header segment */
41 #define ELF_CORE_HEADER_ALIGN 4096
43 /* This primarily represents number of split ranges due to exclusion */
44 #define CRASH_MAX_RANGES 16
46 struct crash_mem_range {
51 unsigned int nr_ranges;
52 struct crash_mem_range ranges[CRASH_MAX_RANGES];
55 /* Misc data about ram ranges needed to prepare elf headers */
56 struct crash_elf_data {
59 * Total number of ram ranges we have after various adjustments for
60 * crash reserved region, etc.
62 unsigned int max_nr_ranges;
64 /* Pointer to elf header */
66 /* Pointer to next phdr */
71 /* Used while preparing memory map entries for second kernel */
72 struct crash_memmap_data {
73 struct boot_params *params;
79 * This is used to VMCLEAR all VMCSs loaded on the
80 * processor. And when loading kvm_intel module, the
81 * callback function pointer will be assigned.
85 crash_vmclear_fn __rcu *crash_vmclear_loaded_vmcss = NULL;
86 EXPORT_SYMBOL_GPL(crash_vmclear_loaded_vmcss);
87 unsigned long crash_zero_bytes;
89 static inline void cpu_crash_vmclear_loaded_vmcss(void)
91 crash_vmclear_fn *do_vmclear_operation = NULL;
94 do_vmclear_operation = rcu_dereference(crash_vmclear_loaded_vmcss);
95 if (do_vmclear_operation)
96 do_vmclear_operation();
100 #if defined(CONFIG_SMP) && defined(CONFIG_X86_LOCAL_APIC)
102 static void kdump_nmi_callback(int cpu, struct pt_regs *regs)
105 struct pt_regs fixed_regs;
107 if (!user_mode(regs)) {
108 crash_fixup_ss_esp(&fixed_regs, regs);
112 crash_save_cpu(regs, cpu);
115 * VMCLEAR VMCSs loaded on all cpus if needed.
117 cpu_crash_vmclear_loaded_vmcss();
119 /* Disable VMX or SVM if needed.
121 * We need to disable virtualization on all CPUs.
122 * Having VMX or SVM enabled on any CPU may break rebooting
123 * after the kdump kernel has finished its task.
125 cpu_emergency_vmxoff();
126 cpu_emergency_svm_disable();
129 * Disable Intel PT to stop its logging
131 cpu_emergency_stop_pt();
133 disable_local_APIC();
136 static void kdump_nmi_shootdown_cpus(void)
138 nmi_shootdown_cpus(kdump_nmi_callback);
140 disable_local_APIC();
144 static void kdump_nmi_shootdown_cpus(void)
146 /* There are no cpus to shootdown */
150 void native_machine_crash_shutdown(struct pt_regs *regs)
152 /* This function is only called after the system
153 * has panicked or is otherwise in a critical state.
154 * The minimum amount of code to allow a kexec'd kernel
155 * to run successfully needs to happen here.
157 * In practice this means shooting down the other cpus in
160 /* The kernel is broken so disable interrupts */
163 kdump_nmi_shootdown_cpus();
166 * VMCLEAR VMCSs loaded on this cpu if needed.
168 cpu_crash_vmclear_loaded_vmcss();
170 /* Booting kdump kernel with VMX or SVM enabled won't work,
171 * because (among other limitations) we can't disable paging
172 * with the virt flags.
174 cpu_emergency_vmxoff();
175 cpu_emergency_svm_disable();
178 * Disable Intel PT to stop its logging
180 cpu_emergency_stop_pt();
182 #ifdef CONFIG_X86_IO_APIC
183 /* Prevent crash_kexec() from deadlocking on ioapic_lock. */
188 #ifdef CONFIG_HPET_TIMER
191 crash_save_cpu(regs, safe_smp_processor_id());
194 #ifdef CONFIG_KEXEC_FILE
195 static int get_nr_ram_ranges_callback(u64 start, u64 end, void *arg)
197 unsigned int *nr_ranges = arg;
204 /* Gather all the required information to prepare elf headers for ram regions */
205 static void fill_up_crash_elf_data(struct crash_elf_data *ced,
206 struct kimage *image)
208 unsigned int nr_ranges = 0;
212 walk_system_ram_res(0, -1, &nr_ranges,
213 get_nr_ram_ranges_callback);
215 ced->max_nr_ranges = nr_ranges;
217 /* Exclusion of crash region could split memory ranges */
218 ced->max_nr_ranges++;
220 /* If crashk_low_res is not 0, another range split possible */
221 if (crashk_low_res.end)
222 ced->max_nr_ranges++;
225 static int exclude_mem_range(struct crash_mem *mem,
226 unsigned long long mstart, unsigned long long mend)
229 unsigned long long start, end;
230 struct crash_mem_range temp_range = {0, 0};
232 for (i = 0; i < mem->nr_ranges; i++) {
233 start = mem->ranges[i].start;
234 end = mem->ranges[i].end;
236 if (mstart > end || mend < start)
239 /* Truncate any area outside of range */
245 /* Found completely overlapping range */
246 if (mstart == start && mend == end) {
247 mem->ranges[i].start = 0;
248 mem->ranges[i].end = 0;
249 if (i < mem->nr_ranges - 1) {
250 /* Shift rest of the ranges to left */
251 for (j = i; j < mem->nr_ranges - 1; j++) {
252 mem->ranges[j].start =
253 mem->ranges[j+1].start;
255 mem->ranges[j+1].end;
262 if (mstart > start && mend < end) {
263 /* Split original range */
264 mem->ranges[i].end = mstart - 1;
265 temp_range.start = mend + 1;
266 temp_range.end = end;
267 } else if (mstart != start)
268 mem->ranges[i].end = mstart - 1;
270 mem->ranges[i].start = mend + 1;
274 /* If a split happend, add the split to array */
279 if (i == CRASH_MAX_RANGES - 1) {
280 pr_err("Too many crash ranges after split\n");
284 /* Location where new range should go */
286 if (j < mem->nr_ranges) {
287 /* Move over all ranges one slot towards the end */
288 for (i = mem->nr_ranges - 1; i >= j; i--)
289 mem->ranges[i + 1] = mem->ranges[i];
292 mem->ranges[j].start = temp_range.start;
293 mem->ranges[j].end = temp_range.end;
299 * Look for any unwanted ranges between mstart, mend and remove them. This
300 * might lead to split and split ranges are put in ced->mem.ranges[] array
302 static int elf_header_exclude_ranges(struct crash_elf_data *ced,
303 unsigned long long mstart, unsigned long long mend)
305 struct crash_mem *cmem = &ced->mem;
308 memset(cmem->ranges, 0, sizeof(cmem->ranges));
310 cmem->ranges[0].start = mstart;
311 cmem->ranges[0].end = mend;
314 /* Exclude crashkernel region */
315 ret = exclude_mem_range(cmem, crashk_res.start, crashk_res.end);
319 if (crashk_low_res.end) {
320 ret = exclude_mem_range(cmem, crashk_low_res.start, crashk_low_res.end);
328 static int prepare_elf64_ram_headers_callback(u64 start, u64 end, void *arg)
330 struct crash_elf_data *ced = arg;
333 unsigned long mstart, mend;
334 struct kimage *image = ced->image;
335 struct crash_mem *cmem;
340 /* Exclude unwanted mem ranges */
341 ret = elf_header_exclude_ranges(ced, start, end);
345 /* Go through all the ranges in ced->mem.ranges[] and prepare phdr */
348 for (i = 0; i < cmem->nr_ranges; i++) {
349 mstart = cmem->ranges[i].start;
350 mend = cmem->ranges[i].end;
353 ced->bufp += sizeof(Elf64_Phdr);
355 phdr->p_type = PT_LOAD;
356 phdr->p_flags = PF_R|PF_W|PF_X;
357 phdr->p_offset = mstart;
360 * If a range matches backup region, adjust offset to backup
363 if (mstart == image->arch.backup_src_start &&
364 (mend - mstart + 1) == image->arch.backup_src_sz)
365 phdr->p_offset = image->arch.backup_load_addr;
367 phdr->p_paddr = mstart;
368 phdr->p_vaddr = (unsigned long long) __va(mstart);
369 phdr->p_filesz = phdr->p_memsz = mend - mstart + 1;
372 pr_debug("Crash PT_LOAD elf header. phdr=%p vaddr=0x%llx, paddr=0x%llx, sz=0x%llx e_phnum=%d p_offset=0x%llx\n",
373 phdr, phdr->p_vaddr, phdr->p_paddr, phdr->p_filesz,
374 ehdr->e_phnum, phdr->p_offset);
380 static int prepare_elf64_headers(struct crash_elf_data *ced,
381 void **addr, unsigned long *sz)
385 unsigned long nr_cpus = num_possible_cpus(), nr_phdr, elf_sz;
386 unsigned char *buf, *bufp;
388 unsigned long long notes_addr;
391 /* extra phdr for vmcoreinfo elf note */
392 nr_phdr = nr_cpus + 1;
393 nr_phdr += ced->max_nr_ranges;
396 * kexec-tools creates an extra PT_LOAD phdr for kernel text mapping
397 * area on x86_64 (ffffffff80000000 - ffffffffa0000000).
398 * I think this is required by tools like gdb. So same physical
399 * memory will be mapped in two elf headers. One will contain kernel
400 * text virtual addresses and other will have __va(physical) addresses.
404 elf_sz = sizeof(Elf64_Ehdr) + nr_phdr * sizeof(Elf64_Phdr);
405 elf_sz = ALIGN(elf_sz, ELF_CORE_HEADER_ALIGN);
407 buf = vzalloc(elf_sz);
412 ehdr = (Elf64_Ehdr *)bufp;
413 bufp += sizeof(Elf64_Ehdr);
414 memcpy(ehdr->e_ident, ELFMAG, SELFMAG);
415 ehdr->e_ident[EI_CLASS] = ELFCLASS64;
416 ehdr->e_ident[EI_DATA] = ELFDATA2LSB;
417 ehdr->e_ident[EI_VERSION] = EV_CURRENT;
418 ehdr->e_ident[EI_OSABI] = ELF_OSABI;
419 memset(ehdr->e_ident + EI_PAD, 0, EI_NIDENT - EI_PAD);
420 ehdr->e_type = ET_CORE;
421 ehdr->e_machine = ELF_ARCH;
422 ehdr->e_version = EV_CURRENT;
423 ehdr->e_phoff = sizeof(Elf64_Ehdr);
424 ehdr->e_ehsize = sizeof(Elf64_Ehdr);
425 ehdr->e_phentsize = sizeof(Elf64_Phdr);
427 /* Prepare one phdr of type PT_NOTE for each present cpu */
428 for_each_present_cpu(cpu) {
429 phdr = (Elf64_Phdr *)bufp;
430 bufp += sizeof(Elf64_Phdr);
431 phdr->p_type = PT_NOTE;
432 notes_addr = per_cpu_ptr_to_phys(per_cpu_ptr(crash_notes, cpu));
433 phdr->p_offset = phdr->p_paddr = notes_addr;
434 phdr->p_filesz = phdr->p_memsz = sizeof(note_buf_t);
438 /* Prepare one PT_NOTE header for vmcoreinfo */
439 phdr = (Elf64_Phdr *)bufp;
440 bufp += sizeof(Elf64_Phdr);
441 phdr->p_type = PT_NOTE;
442 phdr->p_offset = phdr->p_paddr = paddr_vmcoreinfo_note();
443 phdr->p_filesz = phdr->p_memsz = sizeof(vmcoreinfo_note);
447 /* Prepare PT_LOAD type program header for kernel text region */
448 phdr = (Elf64_Phdr *)bufp;
449 bufp += sizeof(Elf64_Phdr);
450 phdr->p_type = PT_LOAD;
451 phdr->p_flags = PF_R|PF_W|PF_X;
452 phdr->p_vaddr = (Elf64_Addr)_text;
453 phdr->p_filesz = phdr->p_memsz = _end - _text;
454 phdr->p_offset = phdr->p_paddr = __pa_symbol(_text);
458 /* Prepare PT_LOAD headers for system ram chunks. */
461 ret = walk_system_ram_res(0, -1, ced,
462 prepare_elf64_ram_headers_callback);
471 /* Prepare elf headers. Return addr and size */
472 static int prepare_elf_headers(struct kimage *image, void **addr,
475 struct crash_elf_data *ced;
478 ced = kzalloc(sizeof(*ced), GFP_KERNEL);
482 fill_up_crash_elf_data(ced, image);
484 /* By default prepare 64bit headers */
485 ret = prepare_elf64_headers(ced, addr, sz);
490 static int add_e820_entry(struct boot_params *params, struct e820entry *entry)
492 unsigned int nr_e820_entries;
494 nr_e820_entries = params->e820_entries;
495 if (nr_e820_entries >= E820MAX)
498 memcpy(¶ms->e820_map[nr_e820_entries], entry,
499 sizeof(struct e820entry));
500 params->e820_entries++;
504 static int memmap_entry_callback(u64 start, u64 end, void *arg)
506 struct crash_memmap_data *cmd = arg;
507 struct boot_params *params = cmd->params;
511 ei.size = end - start + 1;
513 add_e820_entry(params, &ei);
518 static int memmap_exclude_ranges(struct kimage *image, struct crash_mem *cmem,
519 unsigned long long mstart,
520 unsigned long long mend)
522 unsigned long start, end;
525 cmem->ranges[0].start = mstart;
526 cmem->ranges[0].end = mend;
529 /* Exclude Backup region */
530 start = image->arch.backup_load_addr;
531 end = start + image->arch.backup_src_sz - 1;
532 ret = exclude_mem_range(cmem, start, end);
536 /* Exclude elf header region */
537 start = image->arch.elf_load_addr;
538 end = start + image->arch.elf_headers_sz - 1;
539 return exclude_mem_range(cmem, start, end);
542 /* Prepare memory map for crash dump kernel */
543 int crash_setup_memmap_entries(struct kimage *image, struct boot_params *params)
548 struct crash_memmap_data cmd;
549 struct crash_mem *cmem;
551 cmem = vzalloc(sizeof(struct crash_mem));
555 memset(&cmd, 0, sizeof(struct crash_memmap_data));
558 /* Add first 640K segment */
559 ei.addr = image->arch.backup_src_start;
560 ei.size = image->arch.backup_src_sz;
562 add_e820_entry(params, &ei);
564 /* Add ACPI tables */
565 cmd.type = E820_ACPI;
566 flags = IORESOURCE_MEM | IORESOURCE_BUSY;
567 walk_iomem_res_desc(IORES_DESC_ACPI_TABLES, flags, 0, -1, &cmd,
568 memmap_entry_callback);
570 /* Add ACPI Non-volatile Storage */
572 walk_iomem_res_desc(IORES_DESC_ACPI_NV_STORAGE, flags, 0, -1, &cmd,
573 memmap_entry_callback);
575 /* Add crashk_low_res region */
576 if (crashk_low_res.end) {
577 ei.addr = crashk_low_res.start;
578 ei.size = crashk_low_res.end - crashk_low_res.start + 1;
580 add_e820_entry(params, &ei);
583 /* Exclude some ranges from crashk_res and add rest to memmap */
584 ret = memmap_exclude_ranges(image, cmem, crashk_res.start,
589 for (i = 0; i < cmem->nr_ranges; i++) {
590 ei.size = cmem->ranges[i].end - cmem->ranges[i].start + 1;
592 /* If entry is less than a page, skip it */
593 if (ei.size < PAGE_SIZE)
595 ei.addr = cmem->ranges[i].start;
597 add_e820_entry(params, &ei);
605 static int determine_backup_region(u64 start, u64 end, void *arg)
607 struct kimage *image = arg;
609 image->arch.backup_src_start = start;
610 image->arch.backup_src_sz = end - start + 1;
612 /* Expecting only one range for backup region */
616 int crash_load_segments(struct kimage *image)
618 unsigned long src_start, src_sz, elf_sz;
623 * Determine and load a segment for backup area. First 640K RAM
624 * region is backup source
627 ret = walk_system_ram_res(KEXEC_BACKUP_SRC_START, KEXEC_BACKUP_SRC_END,
628 image, determine_backup_region);
630 /* Zero or postive return values are ok */
634 src_start = image->arch.backup_src_start;
635 src_sz = image->arch.backup_src_sz;
637 /* Add backup segment. */
640 * Ideally there is no source for backup segment. This is
641 * copied in purgatory after crash. Just add a zero filled
642 * segment for now to make sure checksum logic works fine.
644 ret = kexec_add_buffer(image, (char *)&crash_zero_bytes,
645 sizeof(crash_zero_bytes), src_sz,
647 &image->arch.backup_load_addr);
650 pr_debug("Loaded backup region at 0x%lx backup_start=0x%lx memsz=0x%lx\n",
651 image->arch.backup_load_addr, src_start, src_sz);
654 /* Prepare elf headers and add a segment */
655 ret = prepare_elf_headers(image, &elf_addr, &elf_sz);
659 image->arch.elf_headers = elf_addr;
660 image->arch.elf_headers_sz = elf_sz;
662 ret = kexec_add_buffer(image, (char *)elf_addr, elf_sz, elf_sz,
663 ELF_CORE_HEADER_ALIGN, 0, -1, 0,
664 &image->arch.elf_load_addr);
666 vfree((void *)image->arch.elf_headers);
669 pr_debug("Loaded ELF headers at 0x%lx bufsz=0x%lx memsz=0x%lx\n",
670 image->arch.elf_load_addr, elf_sz, elf_sz);
674 #endif /* CONFIG_KEXEC_FILE */