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 * GART, crash reserved region etc.
62 unsigned int max_nr_ranges;
63 unsigned long gart_start, gart_end;
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 static void kdump_nmi_shootdown_cpus(void)
139 nmi_shootdown_cpus(kdump_nmi_callback);
141 disable_local_APIC();
145 static void kdump_nmi_shootdown_cpus(void)
147 /* There are no cpus to shootdown */
151 void native_machine_crash_shutdown(struct pt_regs *regs)
153 /* This function is only called after the system
154 * has panicked or is otherwise in a critical state.
155 * The minimum amount of code to allow a kexec'd kernel
156 * to run successfully needs to happen here.
158 * In practice this means shooting down the other cpus in
161 /* The kernel is broken so disable interrupts */
164 kdump_nmi_shootdown_cpus();
167 * VMCLEAR VMCSs loaded on this cpu if needed.
169 cpu_crash_vmclear_loaded_vmcss();
171 /* Booting kdump kernel with VMX or SVM enabled won't work,
172 * because (among other limitations) we can't disable paging
173 * with the virt flags.
175 cpu_emergency_vmxoff();
176 cpu_emergency_svm_disable();
179 * Disable Intel PT to stop its logging
181 cpu_emergency_stop_pt();
183 #ifdef CONFIG_X86_IO_APIC
184 /* Prevent crash_kexec() from deadlocking on ioapic_lock. */
189 #ifdef CONFIG_HPET_TIMER
192 crash_save_cpu(regs, safe_smp_processor_id());
195 #ifdef CONFIG_KEXEC_FILE
196 static int get_nr_ram_ranges_callback(u64 start, u64 end, void *arg)
198 unsigned int *nr_ranges = arg;
204 static int get_gart_ranges_callback(u64 start, u64 end, void *arg)
206 struct crash_elf_data *ced = arg;
208 ced->gart_start = start;
211 /* Not expecting more than 1 gart aperture */
216 /* Gather all the required information to prepare elf headers for ram regions */
217 static void fill_up_crash_elf_data(struct crash_elf_data *ced,
218 struct kimage *image)
220 unsigned int nr_ranges = 0;
224 walk_system_ram_res(0, -1, &nr_ranges,
225 get_nr_ram_ranges_callback);
227 ced->max_nr_ranges = nr_ranges;
230 * We don't create ELF headers for GART aperture as an attempt
231 * to dump this memory in second kernel leads to hang/crash.
232 * If gart aperture is present, one needs to exclude that region
233 * and that could lead to need of extra phdr.
235 walk_iomem_res("GART", IORESOURCE_MEM, 0, -1,
236 ced, get_gart_ranges_callback);
239 * If we have gart region, excluding that could potentially split
240 * a memory range, resulting in extra header. Account for that.
243 ced->max_nr_ranges++;
245 /* Exclusion of crash region could split memory ranges */
246 ced->max_nr_ranges++;
248 /* If crashk_low_res is not 0, another range split possible */
249 if (crashk_low_res.end)
250 ced->max_nr_ranges++;
253 static int exclude_mem_range(struct crash_mem *mem,
254 unsigned long long mstart, unsigned long long mend)
257 unsigned long long start, end;
258 struct crash_mem_range temp_range = {0, 0};
260 for (i = 0; i < mem->nr_ranges; i++) {
261 start = mem->ranges[i].start;
262 end = mem->ranges[i].end;
264 if (mstart > end || mend < start)
267 /* Truncate any area outside of range */
273 /* Found completely overlapping range */
274 if (mstart == start && mend == end) {
275 mem->ranges[i].start = 0;
276 mem->ranges[i].end = 0;
277 if (i < mem->nr_ranges - 1) {
278 /* Shift rest of the ranges to left */
279 for (j = i; j < mem->nr_ranges - 1; j++) {
280 mem->ranges[j].start =
281 mem->ranges[j+1].start;
283 mem->ranges[j+1].end;
290 if (mstart > start && mend < end) {
291 /* Split original range */
292 mem->ranges[i].end = mstart - 1;
293 temp_range.start = mend + 1;
294 temp_range.end = end;
295 } else if (mstart != start)
296 mem->ranges[i].end = mstart - 1;
298 mem->ranges[i].start = mend + 1;
302 /* If a split happend, add the split to array */
307 if (i == CRASH_MAX_RANGES - 1) {
308 pr_err("Too many crash ranges after split\n");
312 /* Location where new range should go */
314 if (j < mem->nr_ranges) {
315 /* Move over all ranges one slot towards the end */
316 for (i = mem->nr_ranges - 1; i >= j; i--)
317 mem->ranges[i + 1] = mem->ranges[i];
320 mem->ranges[j].start = temp_range.start;
321 mem->ranges[j].end = temp_range.end;
327 * Look for any unwanted ranges between mstart, mend and remove them. This
328 * might lead to split and split ranges are put in ced->mem.ranges[] array
330 static int elf_header_exclude_ranges(struct crash_elf_data *ced,
331 unsigned long long mstart, unsigned long long mend)
333 struct crash_mem *cmem = &ced->mem;
336 memset(cmem->ranges, 0, sizeof(cmem->ranges));
338 cmem->ranges[0].start = mstart;
339 cmem->ranges[0].end = mend;
342 /* Exclude crashkernel region */
343 ret = exclude_mem_range(cmem, crashk_res.start, crashk_res.end);
347 if (crashk_low_res.end) {
348 ret = exclude_mem_range(cmem, crashk_low_res.start, crashk_low_res.end);
353 /* Exclude GART region */
355 ret = exclude_mem_range(cmem, ced->gart_start, ced->gart_end);
363 static int prepare_elf64_ram_headers_callback(u64 start, u64 end, void *arg)
365 struct crash_elf_data *ced = arg;
368 unsigned long mstart, mend;
369 struct kimage *image = ced->image;
370 struct crash_mem *cmem;
375 /* Exclude unwanted mem ranges */
376 ret = elf_header_exclude_ranges(ced, start, end);
380 /* Go through all the ranges in ced->mem.ranges[] and prepare phdr */
383 for (i = 0; i < cmem->nr_ranges; i++) {
384 mstart = cmem->ranges[i].start;
385 mend = cmem->ranges[i].end;
388 ced->bufp += sizeof(Elf64_Phdr);
390 phdr->p_type = PT_LOAD;
391 phdr->p_flags = PF_R|PF_W|PF_X;
392 phdr->p_offset = mstart;
395 * If a range matches backup region, adjust offset to backup
398 if (mstart == image->arch.backup_src_start &&
399 (mend - mstart + 1) == image->arch.backup_src_sz)
400 phdr->p_offset = image->arch.backup_load_addr;
402 phdr->p_paddr = mstart;
403 phdr->p_vaddr = (unsigned long long) __va(mstart);
404 phdr->p_filesz = phdr->p_memsz = mend - mstart + 1;
407 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",
408 phdr, phdr->p_vaddr, phdr->p_paddr, phdr->p_filesz,
409 ehdr->e_phnum, phdr->p_offset);
415 static int prepare_elf64_headers(struct crash_elf_data *ced,
416 void **addr, unsigned long *sz)
420 unsigned long nr_cpus = num_possible_cpus(), nr_phdr, elf_sz;
421 unsigned char *buf, *bufp;
423 unsigned long long notes_addr;
426 /* extra phdr for vmcoreinfo elf note */
427 nr_phdr = nr_cpus + 1;
428 nr_phdr += ced->max_nr_ranges;
431 * kexec-tools creates an extra PT_LOAD phdr for kernel text mapping
432 * area on x86_64 (ffffffff80000000 - ffffffffa0000000).
433 * I think this is required by tools like gdb. So same physical
434 * memory will be mapped in two elf headers. One will contain kernel
435 * text virtual addresses and other will have __va(physical) addresses.
439 elf_sz = sizeof(Elf64_Ehdr) + nr_phdr * sizeof(Elf64_Phdr);
440 elf_sz = ALIGN(elf_sz, ELF_CORE_HEADER_ALIGN);
442 buf = vzalloc(elf_sz);
447 ehdr = (Elf64_Ehdr *)bufp;
448 bufp += sizeof(Elf64_Ehdr);
449 memcpy(ehdr->e_ident, ELFMAG, SELFMAG);
450 ehdr->e_ident[EI_CLASS] = ELFCLASS64;
451 ehdr->e_ident[EI_DATA] = ELFDATA2LSB;
452 ehdr->e_ident[EI_VERSION] = EV_CURRENT;
453 ehdr->e_ident[EI_OSABI] = ELF_OSABI;
454 memset(ehdr->e_ident + EI_PAD, 0, EI_NIDENT - EI_PAD);
455 ehdr->e_type = ET_CORE;
456 ehdr->e_machine = ELF_ARCH;
457 ehdr->e_version = EV_CURRENT;
458 ehdr->e_phoff = sizeof(Elf64_Ehdr);
459 ehdr->e_ehsize = sizeof(Elf64_Ehdr);
460 ehdr->e_phentsize = sizeof(Elf64_Phdr);
462 /* Prepare one phdr of type PT_NOTE for each present cpu */
463 for_each_present_cpu(cpu) {
464 phdr = (Elf64_Phdr *)bufp;
465 bufp += sizeof(Elf64_Phdr);
466 phdr->p_type = PT_NOTE;
467 notes_addr = per_cpu_ptr_to_phys(per_cpu_ptr(crash_notes, cpu));
468 phdr->p_offset = phdr->p_paddr = notes_addr;
469 phdr->p_filesz = phdr->p_memsz = sizeof(note_buf_t);
473 /* Prepare one PT_NOTE header for vmcoreinfo */
474 phdr = (Elf64_Phdr *)bufp;
475 bufp += sizeof(Elf64_Phdr);
476 phdr->p_type = PT_NOTE;
477 phdr->p_offset = phdr->p_paddr = paddr_vmcoreinfo_note();
478 phdr->p_filesz = phdr->p_memsz = sizeof(vmcoreinfo_note);
482 /* Prepare PT_LOAD type program header for kernel text region */
483 phdr = (Elf64_Phdr *)bufp;
484 bufp += sizeof(Elf64_Phdr);
485 phdr->p_type = PT_LOAD;
486 phdr->p_flags = PF_R|PF_W|PF_X;
487 phdr->p_vaddr = (Elf64_Addr)_text;
488 phdr->p_filesz = phdr->p_memsz = _end - _text;
489 phdr->p_offset = phdr->p_paddr = __pa_symbol(_text);
493 /* Prepare PT_LOAD headers for system ram chunks. */
496 ret = walk_system_ram_res(0, -1, ced,
497 prepare_elf64_ram_headers_callback);
506 /* Prepare elf headers. Return addr and size */
507 static int prepare_elf_headers(struct kimage *image, void **addr,
510 struct crash_elf_data *ced;
513 ced = kzalloc(sizeof(*ced), GFP_KERNEL);
517 fill_up_crash_elf_data(ced, image);
519 /* By default prepare 64bit headers */
520 ret = prepare_elf64_headers(ced, addr, sz);
525 static int add_e820_entry(struct boot_params *params, struct e820entry *entry)
527 unsigned int nr_e820_entries;
529 nr_e820_entries = params->e820_entries;
530 if (nr_e820_entries >= E820MAX)
533 memcpy(¶ms->e820_map[nr_e820_entries], entry,
534 sizeof(struct e820entry));
535 params->e820_entries++;
539 static int memmap_entry_callback(u64 start, u64 end, void *arg)
541 struct crash_memmap_data *cmd = arg;
542 struct boot_params *params = cmd->params;
546 ei.size = end - start + 1;
548 add_e820_entry(params, &ei);
553 static int memmap_exclude_ranges(struct kimage *image, struct crash_mem *cmem,
554 unsigned long long mstart,
555 unsigned long long mend)
557 unsigned long start, end;
560 cmem->ranges[0].start = mstart;
561 cmem->ranges[0].end = mend;
564 /* Exclude Backup region */
565 start = image->arch.backup_load_addr;
566 end = start + image->arch.backup_src_sz - 1;
567 ret = exclude_mem_range(cmem, start, end);
571 /* Exclude elf header region */
572 start = image->arch.elf_load_addr;
573 end = start + image->arch.elf_headers_sz - 1;
574 return exclude_mem_range(cmem, start, end);
577 /* Prepare memory map for crash dump kernel */
578 int crash_setup_memmap_entries(struct kimage *image, struct boot_params *params)
583 struct crash_memmap_data cmd;
584 struct crash_mem *cmem;
586 cmem = vzalloc(sizeof(struct crash_mem));
590 memset(&cmd, 0, sizeof(struct crash_memmap_data));
593 /* Add first 640K segment */
594 ei.addr = image->arch.backup_src_start;
595 ei.size = image->arch.backup_src_sz;
597 add_e820_entry(params, &ei);
599 /* Add ACPI tables */
600 cmd.type = E820_ACPI;
601 flags = IORESOURCE_MEM | IORESOURCE_BUSY;
602 walk_iomem_res("ACPI Tables", flags, 0, -1, &cmd,
603 memmap_entry_callback);
605 /* Add ACPI Non-volatile Storage */
607 walk_iomem_res("ACPI Non-volatile Storage", flags, 0, -1, &cmd,
608 memmap_entry_callback);
610 /* Add crashk_low_res region */
611 if (crashk_low_res.end) {
612 ei.addr = crashk_low_res.start;
613 ei.size = crashk_low_res.end - crashk_low_res.start + 1;
615 add_e820_entry(params, &ei);
618 /* Exclude some ranges from crashk_res and add rest to memmap */
619 ret = memmap_exclude_ranges(image, cmem, crashk_res.start,
624 for (i = 0; i < cmem->nr_ranges; i++) {
625 ei.size = cmem->ranges[i].end - cmem->ranges[i].start + 1;
627 /* If entry is less than a page, skip it */
628 if (ei.size < PAGE_SIZE)
630 ei.addr = cmem->ranges[i].start;
632 add_e820_entry(params, &ei);
640 static int determine_backup_region(u64 start, u64 end, void *arg)
642 struct kimage *image = arg;
644 image->arch.backup_src_start = start;
645 image->arch.backup_src_sz = end - start + 1;
647 /* Expecting only one range for backup region */
651 int crash_load_segments(struct kimage *image)
653 unsigned long src_start, src_sz, elf_sz;
658 * Determine and load a segment for backup area. First 640K RAM
659 * region is backup source
662 ret = walk_system_ram_res(KEXEC_BACKUP_SRC_START, KEXEC_BACKUP_SRC_END,
663 image, determine_backup_region);
665 /* Zero or postive return values are ok */
669 src_start = image->arch.backup_src_start;
670 src_sz = image->arch.backup_src_sz;
672 /* Add backup segment. */
675 * Ideally there is no source for backup segment. This is
676 * copied in purgatory after crash. Just add a zero filled
677 * segment for now to make sure checksum logic works fine.
679 ret = kexec_add_buffer(image, (char *)&crash_zero_bytes,
680 sizeof(crash_zero_bytes), src_sz,
682 &image->arch.backup_load_addr);
685 pr_debug("Loaded backup region at 0x%lx backup_start=0x%lx memsz=0x%lx\n",
686 image->arch.backup_load_addr, src_start, src_sz);
689 /* Prepare elf headers and add a segment */
690 ret = prepare_elf_headers(image, &elf_addr, &elf_sz);
694 image->arch.elf_headers = elf_addr;
695 image->arch.elf_headers_sz = elf_sz;
697 ret = kexec_add_buffer(image, (char *)elf_addr, elf_sz, elf_sz,
698 ELF_CORE_HEADER_ALIGN, 0, -1, 0,
699 &image->arch.elf_load_addr);
701 vfree((void *)image->arch.elf_headers);
704 pr_debug("Loaded ELF headers at 0x%lx bufsz=0x%lx memsz=0x%lx\n",
705 image->arch.elf_load_addr, elf_sz, elf_sz);
709 #endif /* CONFIG_KEXEC_FILE */