2 * Architecture specific (i386/x86_64) functions for kexec based crash dumps.
6 * Copyright (C) IBM Corporation, 2004. All rights reserved.
7 * Copyright (C) Red Hat Inc., 2014. All rights reserved.
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>
26 #include <asm/processor.h>
27 #include <asm/hardirq.h>
29 #include <asm/hw_irq.h>
31 #include <asm/io_apic.h>
33 #include <linux/kdebug.h>
35 #include <asm/reboot.h>
36 #include <asm/virtext.h>
38 /* Alignment required for elf header segment */
39 #define ELF_CORE_HEADER_ALIGN 4096
41 /* This primarily represents number of split ranges due to exclusion */
42 #define CRASH_MAX_RANGES 16
44 struct crash_mem_range {
49 unsigned int nr_ranges;
50 struct crash_mem_range ranges[CRASH_MAX_RANGES];
53 /* Misc data about ram ranges needed to prepare elf headers */
54 struct crash_elf_data {
57 * Total number of ram ranges we have after various adjustments for
58 * GART, crash reserved region etc.
60 unsigned int max_nr_ranges;
61 unsigned long gart_start, gart_end;
63 /* Pointer to elf header */
65 /* Pointer to next phdr */
70 /* Used while preparing memory map entries for second kernel */
71 struct crash_memmap_data {
72 struct boot_params *params;
80 * This is used to VMCLEAR all VMCSs loaded on the
81 * processor. And when loading kvm_intel module, the
82 * callback function pointer will be assigned.
86 crash_vmclear_fn __rcu *crash_vmclear_loaded_vmcss = NULL;
87 EXPORT_SYMBOL_GPL(crash_vmclear_loaded_vmcss);
88 unsigned long crash_zero_bytes;
90 static inline void cpu_crash_vmclear_loaded_vmcss(void)
92 crash_vmclear_fn *do_vmclear_operation = NULL;
95 do_vmclear_operation = rcu_dereference(crash_vmclear_loaded_vmcss);
96 if (do_vmclear_operation)
97 do_vmclear_operation();
101 #if defined(CONFIG_SMP) && defined(CONFIG_X86_LOCAL_APIC)
103 static void kdump_nmi_callback(int cpu, struct pt_regs *regs)
106 struct pt_regs fixed_regs;
108 if (!user_mode_vm(regs)) {
109 crash_fixup_ss_esp(&fixed_regs, regs);
113 crash_save_cpu(regs, cpu);
116 * VMCLEAR VMCSs loaded on all cpus if needed.
118 cpu_crash_vmclear_loaded_vmcss();
120 /* Disable VMX or SVM if needed.
122 * We need to disable virtualization on all CPUs.
123 * Having VMX or SVM enabled on any CPU may break rebooting
124 * after the kdump kernel has finished its task.
126 cpu_emergency_vmxoff();
127 cpu_emergency_svm_disable();
129 disable_local_APIC();
132 static void kdump_nmi_shootdown_cpus(void)
135 nmi_shootdown_cpus(kdump_nmi_callback);
137 disable_local_APIC();
141 static void kdump_nmi_shootdown_cpus(void)
143 /* There are no cpus to shootdown */
147 void native_machine_crash_shutdown(struct pt_regs *regs)
149 /* This function is only called after the system
150 * has panicked or is otherwise in a critical state.
151 * The minimum amount of code to allow a kexec'd kernel
152 * to run successfully needs to happen here.
154 * In practice this means shooting down the other cpus in
157 /* The kernel is broken so disable interrupts */
160 kdump_nmi_shootdown_cpus();
163 * VMCLEAR VMCSs loaded on this cpu if needed.
165 cpu_crash_vmclear_loaded_vmcss();
167 /* Booting kdump kernel with VMX or SVM enabled won't work,
168 * because (among other limitations) we can't disable paging
169 * with the virt flags.
171 cpu_emergency_vmxoff();
172 cpu_emergency_svm_disable();
174 #ifdef CONFIG_X86_IO_APIC
175 /* Prevent crash_kexec() from deadlocking on ioapic_lock. */
180 #ifdef CONFIG_HPET_TIMER
183 crash_save_cpu(regs, safe_smp_processor_id());
186 #ifdef CONFIG_KEXEC_FILE
187 static int get_nr_ram_ranges_callback(unsigned long start_pfn,
188 unsigned long nr_pfn, void *arg)
190 int *nr_ranges = arg;
196 static int get_gart_ranges_callback(u64 start, u64 end, void *arg)
198 struct crash_elf_data *ced = arg;
200 ced->gart_start = start;
203 /* Not expecting more than 1 gart aperture */
208 /* Gather all the required information to prepare elf headers for ram regions */
209 static void fill_up_crash_elf_data(struct crash_elf_data *ced,
210 struct kimage *image)
212 unsigned int nr_ranges = 0;
216 walk_system_ram_range(0, -1, &nr_ranges,
217 get_nr_ram_ranges_callback);
219 ced->max_nr_ranges = nr_ranges;
222 * We don't create ELF headers for GART aperture as an attempt
223 * to dump this memory in second kernel leads to hang/crash.
224 * If gart aperture is present, one needs to exclude that region
225 * and that could lead to need of extra phdr.
227 walk_iomem_res("GART", IORESOURCE_MEM, 0, -1,
228 ced, get_gart_ranges_callback);
231 * If we have gart region, excluding that could potentially split
232 * a memory range, resulting in extra header. Account for that.
235 ced->max_nr_ranges++;
237 /* Exclusion of crash region could split memory ranges */
238 ced->max_nr_ranges++;
240 /* If crashk_low_res is not 0, another range split possible */
241 if (crashk_low_res.end)
242 ced->max_nr_ranges++;
245 static int exclude_mem_range(struct crash_mem *mem,
246 unsigned long long mstart, unsigned long long mend)
249 unsigned long long start, end;
250 struct crash_mem_range temp_range = {0, 0};
252 for (i = 0; i < mem->nr_ranges; i++) {
253 start = mem->ranges[i].start;
254 end = mem->ranges[i].end;
256 if (mstart > end || mend < start)
259 /* Truncate any area outside of range */
265 /* Found completely overlapping range */
266 if (mstart == start && mend == end) {
267 mem->ranges[i].start = 0;
268 mem->ranges[i].end = 0;
269 if (i < mem->nr_ranges - 1) {
270 /* Shift rest of the ranges to left */
271 for (j = i; j < mem->nr_ranges - 1; j++) {
272 mem->ranges[j].start =
273 mem->ranges[j+1].start;
275 mem->ranges[j+1].end;
282 if (mstart > start && mend < end) {
283 /* Split original range */
284 mem->ranges[i].end = mstart - 1;
285 temp_range.start = mend + 1;
286 temp_range.end = end;
287 } else if (mstart != start)
288 mem->ranges[i].end = mstart - 1;
290 mem->ranges[i].start = mend + 1;
294 /* If a split happend, add the split to array */
299 if (i == CRASH_MAX_RANGES - 1) {
300 pr_err("Too many crash ranges after split\n");
304 /* Location where new range should go */
306 if (j < mem->nr_ranges) {
307 /* Move over all ranges one slot towards the end */
308 for (i = mem->nr_ranges - 1; i >= j; i--)
309 mem->ranges[i + 1] = mem->ranges[i];
312 mem->ranges[j].start = temp_range.start;
313 mem->ranges[j].end = temp_range.end;
319 * Look for any unwanted ranges between mstart, mend and remove them. This
320 * might lead to split and split ranges are put in ced->mem.ranges[] array
322 static int elf_header_exclude_ranges(struct crash_elf_data *ced,
323 unsigned long long mstart, unsigned long long mend)
325 struct crash_mem *cmem = &ced->mem;
328 memset(cmem->ranges, 0, sizeof(cmem->ranges));
330 cmem->ranges[0].start = mstart;
331 cmem->ranges[0].end = mend;
334 /* Exclude crashkernel region */
335 ret = exclude_mem_range(cmem, crashk_res.start, crashk_res.end);
339 if (crashk_low_res.end) {
340 ret = exclude_mem_range(cmem, crashk_low_res.start, crashk_low_res.end);
345 /* Exclude GART region */
347 ret = exclude_mem_range(cmem, ced->gart_start, ced->gart_end);
355 static int prepare_elf64_ram_headers_callback(u64 start, u64 end, void *arg)
357 struct crash_elf_data *ced = arg;
360 unsigned long mstart, mend;
361 struct kimage *image = ced->image;
362 struct crash_mem *cmem;
367 /* Exclude unwanted mem ranges */
368 ret = elf_header_exclude_ranges(ced, start, end);
372 /* Go through all the ranges in ced->mem.ranges[] and prepare phdr */
375 for (i = 0; i < cmem->nr_ranges; i++) {
376 mstart = cmem->ranges[i].start;
377 mend = cmem->ranges[i].end;
380 ced->bufp += sizeof(Elf64_Phdr);
382 phdr->p_type = PT_LOAD;
383 phdr->p_flags = PF_R|PF_W|PF_X;
384 phdr->p_offset = mstart;
387 * If a range matches backup region, adjust offset to backup
390 if (mstart == image->arch.backup_src_start &&
391 (mend - mstart + 1) == image->arch.backup_src_sz)
392 phdr->p_offset = image->arch.backup_load_addr;
394 phdr->p_paddr = mstart;
395 phdr->p_vaddr = (unsigned long long) __va(mstart);
396 phdr->p_filesz = phdr->p_memsz = mend - mstart + 1;
399 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",
400 phdr, phdr->p_vaddr, phdr->p_paddr, phdr->p_filesz,
401 ehdr->e_phnum, phdr->p_offset);
407 static int prepare_elf64_headers(struct crash_elf_data *ced,
408 void **addr, unsigned long *sz)
412 unsigned long nr_cpus = num_possible_cpus(), nr_phdr, elf_sz;
413 unsigned char *buf, *bufp;
415 unsigned long long notes_addr;
418 /* extra phdr for vmcoreinfo elf note */
419 nr_phdr = nr_cpus + 1;
420 nr_phdr += ced->max_nr_ranges;
423 * kexec-tools creates an extra PT_LOAD phdr for kernel text mapping
424 * area on x86_64 (ffffffff80000000 - ffffffffa0000000).
425 * I think this is required by tools like gdb. So same physical
426 * memory will be mapped in two elf headers. One will contain kernel
427 * text virtual addresses and other will have __va(physical) addresses.
431 elf_sz = sizeof(Elf64_Ehdr) + nr_phdr * sizeof(Elf64_Phdr);
432 elf_sz = ALIGN(elf_sz, ELF_CORE_HEADER_ALIGN);
434 buf = vzalloc(elf_sz);
439 ehdr = (Elf64_Ehdr *)bufp;
440 bufp += sizeof(Elf64_Ehdr);
441 memcpy(ehdr->e_ident, ELFMAG, SELFMAG);
442 ehdr->e_ident[EI_CLASS] = ELFCLASS64;
443 ehdr->e_ident[EI_DATA] = ELFDATA2LSB;
444 ehdr->e_ident[EI_VERSION] = EV_CURRENT;
445 ehdr->e_ident[EI_OSABI] = ELF_OSABI;
446 memset(ehdr->e_ident + EI_PAD, 0, EI_NIDENT - EI_PAD);
447 ehdr->e_type = ET_CORE;
448 ehdr->e_machine = ELF_ARCH;
449 ehdr->e_version = EV_CURRENT;
450 ehdr->e_phoff = sizeof(Elf64_Ehdr);
451 ehdr->e_ehsize = sizeof(Elf64_Ehdr);
452 ehdr->e_phentsize = sizeof(Elf64_Phdr);
454 /* Prepare one phdr of type PT_NOTE for each present cpu */
455 for_each_present_cpu(cpu) {
456 phdr = (Elf64_Phdr *)bufp;
457 bufp += sizeof(Elf64_Phdr);
458 phdr->p_type = PT_NOTE;
459 notes_addr = per_cpu_ptr_to_phys(per_cpu_ptr(crash_notes, cpu));
460 phdr->p_offset = phdr->p_paddr = notes_addr;
461 phdr->p_filesz = phdr->p_memsz = sizeof(note_buf_t);
465 /* Prepare one PT_NOTE header for vmcoreinfo */
466 phdr = (Elf64_Phdr *)bufp;
467 bufp += sizeof(Elf64_Phdr);
468 phdr->p_type = PT_NOTE;
469 phdr->p_offset = phdr->p_paddr = paddr_vmcoreinfo_note();
470 phdr->p_filesz = phdr->p_memsz = sizeof(vmcoreinfo_note);
474 /* Prepare PT_LOAD type program header for kernel text region */
475 phdr = (Elf64_Phdr *)bufp;
476 bufp += sizeof(Elf64_Phdr);
477 phdr->p_type = PT_LOAD;
478 phdr->p_flags = PF_R|PF_W|PF_X;
479 phdr->p_vaddr = (Elf64_Addr)_text;
480 phdr->p_filesz = phdr->p_memsz = _end - _text;
481 phdr->p_offset = phdr->p_paddr = __pa_symbol(_text);
485 /* Prepare PT_LOAD headers for system ram chunks. */
488 ret = walk_system_ram_res(0, -1, ced,
489 prepare_elf64_ram_headers_callback);
498 /* Prepare elf headers. Return addr and size */
499 static int prepare_elf_headers(struct kimage *image, void **addr,
502 struct crash_elf_data *ced;
505 ced = kzalloc(sizeof(*ced), GFP_KERNEL);
509 fill_up_crash_elf_data(ced, image);
511 /* By default prepare 64bit headers */
512 ret = prepare_elf64_headers(ced, addr, sz);
517 static int add_e820_entry(struct boot_params *params, struct e820entry *entry)
519 unsigned int nr_e820_entries;
521 nr_e820_entries = params->e820_entries;
522 if (nr_e820_entries >= E820MAX)
525 memcpy(¶ms->e820_map[nr_e820_entries], entry,
526 sizeof(struct e820entry));
527 params->e820_entries++;
531 static int memmap_entry_callback(u64 start, u64 end, void *arg)
533 struct crash_memmap_data *cmd = arg;
534 struct boot_params *params = cmd->params;
538 ei.size = end - start + 1;
540 add_e820_entry(params, &ei);
545 static int memmap_exclude_ranges(struct kimage *image, struct crash_mem *cmem,
546 unsigned long long mstart,
547 unsigned long long mend)
549 unsigned long start, end;
552 cmem->ranges[0].start = mstart;
553 cmem->ranges[0].end = mend;
556 /* Exclude Backup region */
557 start = image->arch.backup_load_addr;
558 end = start + image->arch.backup_src_sz - 1;
559 ret = exclude_mem_range(cmem, start, end);
563 /* Exclude elf header region */
564 start = image->arch.elf_load_addr;
565 end = start + image->arch.elf_headers_sz - 1;
566 return exclude_mem_range(cmem, start, end);
569 /* Prepare memory map for crash dump kernel */
570 int crash_setup_memmap_entries(struct kimage *image, struct boot_params *params)
575 struct crash_memmap_data cmd;
576 struct crash_mem *cmem;
578 cmem = vzalloc(sizeof(struct crash_mem));
582 memset(&cmd, 0, sizeof(struct crash_memmap_data));
585 /* Add first 640K segment */
586 ei.addr = image->arch.backup_src_start;
587 ei.size = image->arch.backup_src_sz;
589 add_e820_entry(params, &ei);
591 /* Add ACPI tables */
592 cmd.type = E820_ACPI;
593 flags = IORESOURCE_MEM | IORESOURCE_BUSY;
594 walk_iomem_res("ACPI Tables", flags, 0, -1, &cmd,
595 memmap_entry_callback);
597 /* Add ACPI Non-volatile Storage */
599 walk_iomem_res("ACPI Non-volatile Storage", flags, 0, -1, &cmd,
600 memmap_entry_callback);
602 /* Add crashk_low_res region */
603 if (crashk_low_res.end) {
604 ei.addr = crashk_low_res.start;
605 ei.size = crashk_low_res.end - crashk_low_res.start + 1;
607 add_e820_entry(params, &ei);
610 /* Exclude some ranges from crashk_res and add rest to memmap */
611 ret = memmap_exclude_ranges(image, cmem, crashk_res.start,
616 for (i = 0; i < cmem->nr_ranges; i++) {
617 ei.size = cmem->ranges[i].end - cmem->ranges[i].start + 1;
619 /* If entry is less than a page, skip it */
620 if (ei.size < PAGE_SIZE)
622 ei.addr = cmem->ranges[i].start;
624 add_e820_entry(params, &ei);
632 static int determine_backup_region(u64 start, u64 end, void *arg)
634 struct kimage *image = arg;
636 image->arch.backup_src_start = start;
637 image->arch.backup_src_sz = end - start + 1;
639 /* Expecting only one range for backup region */
643 int crash_load_segments(struct kimage *image)
645 unsigned long src_start, src_sz, elf_sz;
650 * Determine and load a segment for backup area. First 640K RAM
651 * region is backup source
654 ret = walk_system_ram_res(KEXEC_BACKUP_SRC_START, KEXEC_BACKUP_SRC_END,
655 image, determine_backup_region);
657 /* Zero or postive return values are ok */
661 src_start = image->arch.backup_src_start;
662 src_sz = image->arch.backup_src_sz;
664 /* Add backup segment. */
667 * Ideally there is no source for backup segment. This is
668 * copied in purgatory after crash. Just add a zero filled
669 * segment for now to make sure checksum logic works fine.
671 ret = kexec_add_buffer(image, (char *)&crash_zero_bytes,
672 sizeof(crash_zero_bytes), src_sz,
674 &image->arch.backup_load_addr);
677 pr_debug("Loaded backup region at 0x%lx backup_start=0x%lx memsz=0x%lx\n",
678 image->arch.backup_load_addr, src_start, src_sz);
681 /* Prepare elf headers and add a segment */
682 ret = prepare_elf_headers(image, &elf_addr, &elf_sz);
686 image->arch.elf_headers = elf_addr;
687 image->arch.elf_headers_sz = elf_sz;
689 ret = kexec_add_buffer(image, (char *)elf_addr, elf_sz, elf_sz,
690 ELF_CORE_HEADER_ALIGN, 0, -1, 0,
691 &image->arch.elf_load_addr);
693 vfree((void *)image->arch.elf_headers);
696 pr_debug("Loaded ELF headers at 0x%lx bufsz=0x%lx memsz=0x%lx\n",
697 image->arch.elf_load_addr, elf_sz, elf_sz);
701 #endif /* CONFIG_KEXEC_FILE */