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/export.h>
24 #include <linux/slab.h>
25 #include <linux/vmalloc.h>
26 #include <linux/overflow.h>
28 #include <asm/processor.h>
29 #include <asm/hardirq.h>
31 #include <asm/hw_irq.h>
33 #include <asm/io_apic.h>
35 #include <linux/kdebug.h>
37 #include <asm/reboot.h>
38 #include <asm/virtext.h>
39 #include <asm/intel_pt.h>
41 /* Alignment required for elf header segment */
42 #define ELF_CORE_HEADER_ALIGN 4096
44 /* This primarily represents number of split ranges due to exclusion */
45 #define CRASH_MAX_RANGES 16
47 struct crash_mem_range {
52 unsigned int nr_ranges;
53 struct crash_mem_range ranges[CRASH_MAX_RANGES];
56 /* Misc data about ram ranges needed to prepare elf headers */
57 struct crash_elf_data {
60 * Total number of ram ranges we have after various adjustments for
61 * crash reserved region, etc.
63 unsigned int max_nr_ranges;
65 /* Pointer to elf header */
67 /* Pointer to next phdr */
72 /* Used while preparing memory map entries for second kernel */
73 struct crash_memmap_data {
74 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(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();
130 * Disable Intel PT to stop its logging
132 cpu_emergency_stop_pt();
134 disable_local_APIC();
137 void kdump_nmi_shootdown_cpus(void)
139 nmi_shootdown_cpus(kdump_nmi_callback);
141 disable_local_APIC();
144 /* Override the weak function in kernel/panic.c */
145 void crash_smp_send_stop(void)
147 static int cpus_stopped;
152 if (smp_ops.crash_stop_other_cpus)
153 smp_ops.crash_stop_other_cpus();
161 void crash_smp_send_stop(void)
163 /* There are no cpus to shootdown */
167 void native_machine_crash_shutdown(struct pt_regs *regs)
169 /* This function is only called after the system
170 * has panicked or is otherwise in a critical state.
171 * The minimum amount of code to allow a kexec'd kernel
172 * to run successfully needs to happen here.
174 * In practice this means shooting down the other cpus in
177 /* The kernel is broken so disable interrupts */
180 crash_smp_send_stop();
183 * VMCLEAR VMCSs loaded on this cpu if needed.
185 cpu_crash_vmclear_loaded_vmcss();
187 /* Booting kdump kernel with VMX or SVM enabled won't work,
188 * because (among other limitations) we can't disable paging
189 * with the virt flags.
191 cpu_emergency_vmxoff();
192 cpu_emergency_svm_disable();
195 * Disable Intel PT to stop its logging
197 cpu_emergency_stop_pt();
199 #ifdef CONFIG_X86_IO_APIC
200 /* Prevent crash_kexec() from deadlocking on ioapic_lock. */
205 #ifdef CONFIG_HPET_TIMER
208 crash_save_cpu(regs, safe_smp_processor_id());
211 #ifdef CONFIG_KEXEC_FILE
212 static int get_nr_ram_ranges_callback(u64 start, u64 end, void *arg)
214 unsigned int *nr_ranges = arg;
221 /* Gather all the required information to prepare elf headers for ram regions */
222 static void fill_up_crash_elf_data(struct crash_elf_data *ced,
223 struct kimage *image)
225 unsigned int nr_ranges = 0;
229 walk_system_ram_res(0, -1, &nr_ranges,
230 get_nr_ram_ranges_callback);
232 ced->max_nr_ranges = nr_ranges;
234 /* Exclusion of crash region could split memory ranges */
235 ced->max_nr_ranges++;
237 /* If crashk_low_res is not 0, another range split possible */
238 if (crashk_low_res.end)
239 ced->max_nr_ranges++;
242 static int exclude_mem_range(struct crash_mem *mem,
243 unsigned long long mstart, unsigned long long mend)
246 unsigned long long start, end;
247 struct crash_mem_range temp_range = {0, 0};
249 for (i = 0; i < mem->nr_ranges; i++) {
250 start = mem->ranges[i].start;
251 end = mem->ranges[i].end;
253 if (mstart > end || mend < start)
256 /* Truncate any area outside of range */
262 /* Found completely overlapping range */
263 if (mstart == start && mend == end) {
264 mem->ranges[i].start = 0;
265 mem->ranges[i].end = 0;
266 if (i < mem->nr_ranges - 1) {
267 /* Shift rest of the ranges to left */
268 for (j = i; j < mem->nr_ranges - 1; j++) {
269 mem->ranges[j].start =
270 mem->ranges[j+1].start;
272 mem->ranges[j+1].end;
279 if (mstart > start && mend < end) {
280 /* Split original range */
281 mem->ranges[i].end = mstart - 1;
282 temp_range.start = mend + 1;
283 temp_range.end = end;
284 } else if (mstart != start)
285 mem->ranges[i].end = mstart - 1;
287 mem->ranges[i].start = mend + 1;
291 /* If a split happend, add the split to array */
296 if (i == CRASH_MAX_RANGES - 1) {
297 pr_err("Too many crash ranges after split\n");
301 /* Location where new range should go */
303 if (j < mem->nr_ranges) {
304 /* Move over all ranges one slot towards the end */
305 for (i = mem->nr_ranges - 1; i >= j; i--)
306 mem->ranges[i + 1] = mem->ranges[i];
309 mem->ranges[j].start = temp_range.start;
310 mem->ranges[j].end = temp_range.end;
316 * Look for any unwanted ranges between mstart, mend and remove them. This
317 * might lead to split and split ranges are put in ced->mem.ranges[] array
319 static int elf_header_exclude_ranges(struct crash_elf_data *ced,
320 unsigned long long mstart, unsigned long long mend)
322 struct crash_mem *cmem = &ced->mem;
325 memset(cmem->ranges, 0, sizeof(cmem->ranges));
327 cmem->ranges[0].start = mstart;
328 cmem->ranges[0].end = mend;
331 /* Exclude crashkernel region */
332 ret = exclude_mem_range(cmem, crashk_res.start, crashk_res.end);
336 if (crashk_low_res.end) {
337 ret = exclude_mem_range(cmem, crashk_low_res.start, crashk_low_res.end);
345 static int prepare_elf64_ram_headers_callback(u64 start, u64 end, void *arg)
347 struct crash_elf_data *ced = arg;
350 unsigned long mstart, mend;
351 struct kimage *image = ced->image;
352 struct crash_mem *cmem;
357 /* Exclude unwanted mem ranges */
358 ret = elf_header_exclude_ranges(ced, start, end);
362 /* Go through all the ranges in ced->mem.ranges[] and prepare phdr */
365 for (i = 0; i < cmem->nr_ranges; i++) {
366 mstart = cmem->ranges[i].start;
367 mend = cmem->ranges[i].end;
370 ced->bufp += sizeof(Elf64_Phdr);
372 phdr->p_type = PT_LOAD;
373 phdr->p_flags = PF_R|PF_W|PF_X;
374 phdr->p_offset = mstart;
377 * If a range matches backup region, adjust offset to backup
380 if (mstart == image->arch.backup_src_start &&
381 (mend - mstart + 1) == image->arch.backup_src_sz)
382 phdr->p_offset = image->arch.backup_load_addr;
384 phdr->p_paddr = mstart;
385 phdr->p_vaddr = (unsigned long long) __va(mstart);
386 phdr->p_filesz = phdr->p_memsz = mend - mstart + 1;
389 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",
390 phdr, phdr->p_vaddr, phdr->p_paddr, phdr->p_filesz,
391 ehdr->e_phnum, phdr->p_offset);
397 static int prepare_elf64_headers(struct crash_elf_data *ced,
398 void **addr, unsigned long *sz)
402 unsigned long nr_cpus = num_possible_cpus(), nr_phdr, elf_sz;
403 unsigned char *buf, *bufp;
405 unsigned long long notes_addr;
408 /* extra phdr for vmcoreinfo elf note */
409 nr_phdr = nr_cpus + 1;
410 nr_phdr += ced->max_nr_ranges;
413 * kexec-tools creates an extra PT_LOAD phdr for kernel text mapping
414 * area on x86_64 (ffffffff80000000 - ffffffffa0000000).
415 * I think this is required by tools like gdb. So same physical
416 * memory will be mapped in two elf headers. One will contain kernel
417 * text virtual addresses and other will have __va(physical) addresses.
421 elf_sz = sizeof(Elf64_Ehdr) + nr_phdr * sizeof(Elf64_Phdr);
422 elf_sz = ALIGN(elf_sz, ELF_CORE_HEADER_ALIGN);
424 buf = vzalloc(elf_sz);
429 ehdr = (Elf64_Ehdr *)bufp;
430 bufp += sizeof(Elf64_Ehdr);
431 memcpy(ehdr->e_ident, ELFMAG, SELFMAG);
432 ehdr->e_ident[EI_CLASS] = ELFCLASS64;
433 ehdr->e_ident[EI_DATA] = ELFDATA2LSB;
434 ehdr->e_ident[EI_VERSION] = EV_CURRENT;
435 ehdr->e_ident[EI_OSABI] = ELF_OSABI;
436 memset(ehdr->e_ident + EI_PAD, 0, EI_NIDENT - EI_PAD);
437 ehdr->e_type = ET_CORE;
438 ehdr->e_machine = ELF_ARCH;
439 ehdr->e_version = EV_CURRENT;
440 ehdr->e_phoff = sizeof(Elf64_Ehdr);
441 ehdr->e_ehsize = sizeof(Elf64_Ehdr);
442 ehdr->e_phentsize = sizeof(Elf64_Phdr);
444 /* Prepare one phdr of type PT_NOTE for each present cpu */
445 for_each_present_cpu(cpu) {
446 phdr = (Elf64_Phdr *)bufp;
447 bufp += sizeof(Elf64_Phdr);
448 phdr->p_type = PT_NOTE;
449 notes_addr = per_cpu_ptr_to_phys(per_cpu_ptr(crash_notes, cpu));
450 phdr->p_offset = phdr->p_paddr = notes_addr;
451 phdr->p_filesz = phdr->p_memsz = sizeof(note_buf_t);
455 /* Prepare one PT_NOTE header for vmcoreinfo */
456 phdr = (Elf64_Phdr *)bufp;
457 bufp += sizeof(Elf64_Phdr);
458 phdr->p_type = PT_NOTE;
459 phdr->p_offset = phdr->p_paddr = paddr_vmcoreinfo_note();
460 phdr->p_filesz = phdr->p_memsz = sizeof(vmcoreinfo_note);
464 /* Prepare PT_LOAD type program header for kernel text region */
465 phdr = (Elf64_Phdr *)bufp;
466 bufp += sizeof(Elf64_Phdr);
467 phdr->p_type = PT_LOAD;
468 phdr->p_flags = PF_R|PF_W|PF_X;
469 phdr->p_vaddr = (Elf64_Addr)_text;
470 phdr->p_filesz = phdr->p_memsz = _end - _text;
471 phdr->p_offset = phdr->p_paddr = __pa_symbol(_text);
475 /* Prepare PT_LOAD headers for system ram chunks. */
478 ret = walk_system_ram_res(0, -1, ced,
479 prepare_elf64_ram_headers_callback);
488 /* Prepare elf headers. Return addr and size */
489 static int prepare_elf_headers(struct kimage *image, void **addr,
492 struct crash_elf_data *ced;
495 ced = kzalloc(sizeof(*ced), GFP_KERNEL);
499 fill_up_crash_elf_data(ced, image);
501 /* By default prepare 64bit headers */
502 ret = prepare_elf64_headers(ced, addr, sz);
507 static int add_e820_entry(struct boot_params *params, struct e820entry *entry)
509 unsigned int nr_e820_entries;
511 nr_e820_entries = params->e820_entries;
512 if (nr_e820_entries >= E820MAX)
515 memcpy(¶ms->e820_map[nr_e820_entries], entry,
516 sizeof(struct e820entry));
517 params->e820_entries++;
521 static int memmap_entry_callback(u64 start, u64 end, void *arg)
523 struct crash_memmap_data *cmd = arg;
524 struct boot_params *params = cmd->params;
528 ei.size = end - start + 1;
530 add_e820_entry(params, &ei);
535 static int memmap_exclude_ranges(struct kimage *image, struct crash_mem *cmem,
536 unsigned long long mstart,
537 unsigned long long mend)
539 unsigned long start, end;
542 cmem->ranges[0].start = mstart;
543 cmem->ranges[0].end = mend;
546 /* Exclude Backup region */
547 start = image->arch.backup_load_addr;
548 end = start + image->arch.backup_src_sz - 1;
549 ret = exclude_mem_range(cmem, start, end);
553 /* Exclude elf header region */
554 start = image->arch.elf_load_addr;
555 end = start + image->arch.elf_headers_sz - 1;
556 return exclude_mem_range(cmem, start, end);
559 /* Prepare memory map for crash dump kernel */
560 int crash_setup_memmap_entries(struct kimage *image, struct boot_params *params)
565 struct crash_memmap_data cmd;
566 struct crash_mem *cmem;
568 cmem = vzalloc(struct_size(cmem, ranges, 1));
572 memset(&cmd, 0, sizeof(struct crash_memmap_data));
575 /* Add first 640K segment */
576 ei.addr = image->arch.backup_src_start;
577 ei.size = image->arch.backup_src_sz;
579 add_e820_entry(params, &ei);
581 /* Add ACPI tables */
582 cmd.type = E820_ACPI;
583 flags = IORESOURCE_MEM | IORESOURCE_BUSY;
584 walk_iomem_res_desc(IORES_DESC_ACPI_TABLES, flags, 0, -1, &cmd,
585 memmap_entry_callback);
587 /* Add ACPI Non-volatile Storage */
589 walk_iomem_res_desc(IORES_DESC_ACPI_NV_STORAGE, flags, 0, -1, &cmd,
590 memmap_entry_callback);
592 /* Add crashk_low_res region */
593 if (crashk_low_res.end) {
594 ei.addr = crashk_low_res.start;
595 ei.size = crashk_low_res.end - crashk_low_res.start + 1;
597 add_e820_entry(params, &ei);
600 /* Exclude some ranges from crashk_res and add rest to memmap */
601 ret = memmap_exclude_ranges(image, cmem, crashk_res.start,
606 for (i = 0; i < cmem->nr_ranges; i++) {
607 ei.size = cmem->ranges[i].end - cmem->ranges[i].start + 1;
609 /* If entry is less than a page, skip it */
610 if (ei.size < PAGE_SIZE)
612 ei.addr = cmem->ranges[i].start;
614 add_e820_entry(params, &ei);
622 static int determine_backup_region(u64 start, u64 end, void *arg)
624 struct kimage *image = arg;
626 image->arch.backup_src_start = start;
627 image->arch.backup_src_sz = end - start + 1;
629 /* Expecting only one range for backup region */
633 int crash_load_segments(struct kimage *image)
635 unsigned long src_start, src_sz, elf_sz;
640 * Determine and load a segment for backup area. First 640K RAM
641 * region is backup source
644 ret = walk_system_ram_res(KEXEC_BACKUP_SRC_START, KEXEC_BACKUP_SRC_END,
645 image, determine_backup_region);
647 /* Zero or postive return values are ok */
651 src_start = image->arch.backup_src_start;
652 src_sz = image->arch.backup_src_sz;
654 /* Add backup segment. */
657 * Ideally there is no source for backup segment. This is
658 * copied in purgatory after crash. Just add a zero filled
659 * segment for now to make sure checksum logic works fine.
661 ret = kexec_add_buffer(image, (char *)&crash_zero_bytes,
662 sizeof(crash_zero_bytes), src_sz,
664 &image->arch.backup_load_addr);
667 pr_debug("Loaded backup region at 0x%lx backup_start=0x%lx memsz=0x%lx\n",
668 image->arch.backup_load_addr, src_start, src_sz);
671 /* Prepare elf headers and add a segment */
672 ret = prepare_elf_headers(image, &elf_addr, &elf_sz);
676 image->arch.elf_headers = elf_addr;
677 image->arch.elf_headers_sz = elf_sz;
679 ret = kexec_add_buffer(image, (char *)elf_addr, elf_sz, elf_sz,
680 ELF_CORE_HEADER_ALIGN, 0, -1, 0,
681 &image->arch.elf_load_addr);
683 vfree((void *)image->arch.elf_headers);
686 pr_debug("Loaded ELF headers at 0x%lx bufsz=0x%lx memsz=0x%lx\n",
687 image->arch.elf_load_addr, elf_sz, elf_sz);
691 #endif /* CONFIG_KEXEC_FILE */