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>
26 #include <asm/processor.h>
27 #include <asm/hardirq.h>
29 #include <asm/hw_irq.h>
32 #include <linux/kdebug.h>
34 #include <asm/reboot.h>
35 #include <asm/virtext.h>
37 /* Alignment required for elf header segment */
38 #define ELF_CORE_HEADER_ALIGN 4096
40 /* This primarily represents number of split ranges due to exclusion */
41 #define CRASH_MAX_RANGES 16
43 struct crash_mem_range {
48 unsigned int nr_ranges;
49 struct crash_mem_range ranges[CRASH_MAX_RANGES];
52 /* Misc data about ram ranges needed to prepare elf headers */
53 struct crash_elf_data {
56 * Total number of ram ranges we have after various adjustments for
57 * GART, crash reserved region etc.
59 unsigned int max_nr_ranges;
60 unsigned long gart_start, gart_end;
62 /* Pointer to elf header */
64 /* Pointer to next phdr */
69 /* Used while preparing memory map entries for second kernel */
70 struct crash_memmap_data {
71 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_vm(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();
128 disable_local_APIC();
131 static void kdump_nmi_shootdown_cpus(void)
134 nmi_shootdown_cpus(kdump_nmi_callback);
136 disable_local_APIC();
140 static void kdump_nmi_shootdown_cpus(void)
142 /* There are no cpus to shootdown */
146 void native_machine_crash_shutdown(struct pt_regs *regs)
148 /* This function is only called after the system
149 * has panicked or is otherwise in a critical state.
150 * The minimum amount of code to allow a kexec'd kernel
151 * to run successfully needs to happen here.
153 * In practice this means shooting down the other cpus in
156 /* The kernel is broken so disable interrupts */
159 kdump_nmi_shootdown_cpus();
162 * VMCLEAR VMCSs loaded on this cpu if needed.
164 cpu_crash_vmclear_loaded_vmcss();
166 /* Booting kdump kernel with VMX or SVM enabled won't work,
167 * because (among other limitations) we can't disable paging
168 * with the virt flags.
170 cpu_emergency_vmxoff();
171 cpu_emergency_svm_disable();
173 #ifdef CONFIG_X86_IO_APIC
174 /* Prevent crash_kexec() from deadlocking on ioapic_lock. */
179 #ifdef CONFIG_HPET_TIMER
182 crash_save_cpu(regs, safe_smp_processor_id());
185 #ifdef CONFIG_KEXEC_FILE
186 static int get_nr_ram_ranges_callback(unsigned long start_pfn,
187 unsigned long nr_pfn, void *arg)
189 int *nr_ranges = arg;
195 static int get_gart_ranges_callback(u64 start, u64 end, void *arg)
197 struct crash_elf_data *ced = arg;
199 ced->gart_start = start;
202 /* Not expecting more than 1 gart aperture */
207 /* Gather all the required information to prepare elf headers for ram regions */
208 static void fill_up_crash_elf_data(struct crash_elf_data *ced,
209 struct kimage *image)
211 unsigned int nr_ranges = 0;
215 walk_system_ram_range(0, -1, &nr_ranges,
216 get_nr_ram_ranges_callback);
218 ced->max_nr_ranges = nr_ranges;
221 * We don't create ELF headers for GART aperture as an attempt
222 * to dump this memory in second kernel leads to hang/crash.
223 * If gart aperture is present, one needs to exclude that region
224 * and that could lead to need of extra phdr.
226 walk_iomem_res("GART", IORESOURCE_MEM, 0, -1,
227 ced, get_gart_ranges_callback);
230 * If we have gart region, excluding that could potentially split
231 * a memory range, resulting in extra header. Account for that.
234 ced->max_nr_ranges++;
236 /* Exclusion of crash region could split memory ranges */
237 ced->max_nr_ranges++;
239 /* If crashk_low_res is not 0, another range split possible */
240 if (crashk_low_res.end)
241 ced->max_nr_ranges++;
244 static int exclude_mem_range(struct crash_mem *mem,
245 unsigned long long mstart, unsigned long long mend)
248 unsigned long long start, end;
249 struct crash_mem_range temp_range = {0, 0};
251 for (i = 0; i < mem->nr_ranges; i++) {
252 start = mem->ranges[i].start;
253 end = mem->ranges[i].end;
255 if (mstart > end || mend < start)
258 /* Truncate any area outside of range */
264 /* Found completely overlapping range */
265 if (mstart == start && mend == end) {
266 mem->ranges[i].start = 0;
267 mem->ranges[i].end = 0;
268 if (i < mem->nr_ranges - 1) {
269 /* Shift rest of the ranges to left */
270 for (j = i; j < mem->nr_ranges - 1; j++) {
271 mem->ranges[j].start =
272 mem->ranges[j+1].start;
274 mem->ranges[j+1].end;
281 if (mstart > start && mend < end) {
282 /* Split original range */
283 mem->ranges[i].end = mstart - 1;
284 temp_range.start = mend + 1;
285 temp_range.end = end;
286 } else if (mstart != start)
287 mem->ranges[i].end = mstart - 1;
289 mem->ranges[i].start = mend + 1;
293 /* If a split happend, add the split to array */
298 if (i == CRASH_MAX_RANGES - 1) {
299 pr_err("Too many crash ranges after split\n");
303 /* Location where new range should go */
305 if (j < mem->nr_ranges) {
306 /* Move over all ranges one slot towards the end */
307 for (i = mem->nr_ranges - 1; i >= j; i--)
308 mem->ranges[i + 1] = mem->ranges[i];
311 mem->ranges[j].start = temp_range.start;
312 mem->ranges[j].end = temp_range.end;
318 * Look for any unwanted ranges between mstart, mend and remove them. This
319 * might lead to split and split ranges are put in ced->mem.ranges[] array
321 static int elf_header_exclude_ranges(struct crash_elf_data *ced,
322 unsigned long long mstart, unsigned long long mend)
324 struct crash_mem *cmem = &ced->mem;
327 memset(cmem->ranges, 0, sizeof(cmem->ranges));
329 cmem->ranges[0].start = mstart;
330 cmem->ranges[0].end = mend;
333 /* Exclude crashkernel region */
334 ret = exclude_mem_range(cmem, crashk_res.start, crashk_res.end);
338 if (crashk_low_res.end) {
339 ret = exclude_mem_range(cmem, crashk_low_res.start, crashk_low_res.end);
344 /* Exclude GART region */
346 ret = exclude_mem_range(cmem, ced->gart_start, ced->gart_end);
354 static int prepare_elf64_ram_headers_callback(u64 start, u64 end, void *arg)
356 struct crash_elf_data *ced = arg;
359 unsigned long mstart, mend;
360 struct kimage *image = ced->image;
361 struct crash_mem *cmem;
366 /* Exclude unwanted mem ranges */
367 ret = elf_header_exclude_ranges(ced, start, end);
371 /* Go through all the ranges in ced->mem.ranges[] and prepare phdr */
374 for (i = 0; i < cmem->nr_ranges; i++) {
375 mstart = cmem->ranges[i].start;
376 mend = cmem->ranges[i].end;
379 ced->bufp += sizeof(Elf64_Phdr);
381 phdr->p_type = PT_LOAD;
382 phdr->p_flags = PF_R|PF_W|PF_X;
383 phdr->p_offset = mstart;
386 * If a range matches backup region, adjust offset to backup
389 if (mstart == image->arch.backup_src_start &&
390 (mend - mstart + 1) == image->arch.backup_src_sz)
391 phdr->p_offset = image->arch.backup_load_addr;
393 phdr->p_paddr = mstart;
394 phdr->p_vaddr = (unsigned long long) __va(mstart);
395 phdr->p_filesz = phdr->p_memsz = mend - mstart + 1;
398 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",
399 phdr, phdr->p_vaddr, phdr->p_paddr, phdr->p_filesz,
400 ehdr->e_phnum, phdr->p_offset);
406 static int prepare_elf64_headers(struct crash_elf_data *ced,
407 void **addr, unsigned long *sz)
411 unsigned long nr_cpus = num_possible_cpus(), nr_phdr, elf_sz;
412 unsigned char *buf, *bufp;
414 unsigned long long notes_addr;
417 /* extra phdr for vmcoreinfo elf note */
418 nr_phdr = nr_cpus + 1;
419 nr_phdr += ced->max_nr_ranges;
422 * kexec-tools creates an extra PT_LOAD phdr for kernel text mapping
423 * area on x86_64 (ffffffff80000000 - ffffffffa0000000).
424 * I think this is required by tools like gdb. So same physical
425 * memory will be mapped in two elf headers. One will contain kernel
426 * text virtual addresses and other will have __va(physical) addresses.
430 elf_sz = sizeof(Elf64_Ehdr) + nr_phdr * sizeof(Elf64_Phdr);
431 elf_sz = ALIGN(elf_sz, ELF_CORE_HEADER_ALIGN);
433 buf = vzalloc(elf_sz);
438 ehdr = (Elf64_Ehdr *)bufp;
439 bufp += sizeof(Elf64_Ehdr);
440 memcpy(ehdr->e_ident, ELFMAG, SELFMAG);
441 ehdr->e_ident[EI_CLASS] = ELFCLASS64;
442 ehdr->e_ident[EI_DATA] = ELFDATA2LSB;
443 ehdr->e_ident[EI_VERSION] = EV_CURRENT;
444 ehdr->e_ident[EI_OSABI] = ELF_OSABI;
445 memset(ehdr->e_ident + EI_PAD, 0, EI_NIDENT - EI_PAD);
446 ehdr->e_type = ET_CORE;
447 ehdr->e_machine = ELF_ARCH;
448 ehdr->e_version = EV_CURRENT;
449 ehdr->e_phoff = sizeof(Elf64_Ehdr);
450 ehdr->e_ehsize = sizeof(Elf64_Ehdr);
451 ehdr->e_phentsize = sizeof(Elf64_Phdr);
453 /* Prepare one phdr of type PT_NOTE for each present cpu */
454 for_each_present_cpu(cpu) {
455 phdr = (Elf64_Phdr *)bufp;
456 bufp += sizeof(Elf64_Phdr);
457 phdr->p_type = PT_NOTE;
458 notes_addr = per_cpu_ptr_to_phys(per_cpu_ptr(crash_notes, cpu));
459 phdr->p_offset = phdr->p_paddr = notes_addr;
460 phdr->p_filesz = phdr->p_memsz = sizeof(note_buf_t);
464 /* Prepare one PT_NOTE header for vmcoreinfo */
465 phdr = (Elf64_Phdr *)bufp;
466 bufp += sizeof(Elf64_Phdr);
467 phdr->p_type = PT_NOTE;
468 phdr->p_offset = phdr->p_paddr = paddr_vmcoreinfo_note();
469 phdr->p_filesz = phdr->p_memsz = sizeof(vmcoreinfo_note);
473 /* Prepare PT_LOAD type program header for kernel text region */
474 phdr = (Elf64_Phdr *)bufp;
475 bufp += sizeof(Elf64_Phdr);
476 phdr->p_type = PT_LOAD;
477 phdr->p_flags = PF_R|PF_W|PF_X;
478 phdr->p_vaddr = (Elf64_Addr)_text;
479 phdr->p_filesz = phdr->p_memsz = _end - _text;
480 phdr->p_offset = phdr->p_paddr = __pa_symbol(_text);
484 /* Prepare PT_LOAD headers for system ram chunks. */
487 ret = walk_system_ram_res(0, -1, ced,
488 prepare_elf64_ram_headers_callback);
497 /* Prepare elf headers. Return addr and size */
498 static int prepare_elf_headers(struct kimage *image, void **addr,
501 struct crash_elf_data *ced;
504 ced = kzalloc(sizeof(*ced), GFP_KERNEL);
508 fill_up_crash_elf_data(ced, image);
510 /* By default prepare 64bit headers */
511 ret = prepare_elf64_headers(ced, addr, sz);
516 static int add_e820_entry(struct boot_params *params, struct e820entry *entry)
518 unsigned int nr_e820_entries;
520 nr_e820_entries = params->e820_entries;
521 if (nr_e820_entries >= E820MAX)
524 memcpy(¶ms->e820_map[nr_e820_entries], entry,
525 sizeof(struct e820entry));
526 params->e820_entries++;
530 static int memmap_entry_callback(u64 start, u64 end, void *arg)
532 struct crash_memmap_data *cmd = arg;
533 struct boot_params *params = cmd->params;
537 ei.size = end - start + 1;
539 add_e820_entry(params, &ei);
544 static int memmap_exclude_ranges(struct kimage *image, struct crash_mem *cmem,
545 unsigned long long mstart,
546 unsigned long long mend)
548 unsigned long start, end;
551 cmem->ranges[0].start = mstart;
552 cmem->ranges[0].end = mend;
555 /* Exclude Backup region */
556 start = image->arch.backup_load_addr;
557 end = start + image->arch.backup_src_sz - 1;
558 ret = exclude_mem_range(cmem, start, end);
562 /* Exclude elf header region */
563 start = image->arch.elf_load_addr;
564 end = start + image->arch.elf_headers_sz - 1;
565 return exclude_mem_range(cmem, start, end);
568 /* Prepare memory map for crash dump kernel */
569 int crash_setup_memmap_entries(struct kimage *image, struct boot_params *params)
574 struct crash_memmap_data cmd;
575 struct crash_mem *cmem;
577 cmem = vzalloc(sizeof(struct crash_mem));
581 memset(&cmd, 0, sizeof(struct crash_memmap_data));
584 /* Add first 640K segment */
585 ei.addr = image->arch.backup_src_start;
586 ei.size = image->arch.backup_src_sz;
588 add_e820_entry(params, &ei);
590 /* Add ACPI tables */
591 cmd.type = E820_ACPI;
592 flags = IORESOURCE_MEM | IORESOURCE_BUSY;
593 walk_iomem_res("ACPI Tables", flags, 0, -1, &cmd,
594 memmap_entry_callback);
596 /* Add ACPI Non-volatile Storage */
598 walk_iomem_res("ACPI Non-volatile Storage", flags, 0, -1, &cmd,
599 memmap_entry_callback);
601 /* Add crashk_low_res region */
602 if (crashk_low_res.end) {
603 ei.addr = crashk_low_res.start;
604 ei.size = crashk_low_res.end - crashk_low_res.start + 1;
606 add_e820_entry(params, &ei);
609 /* Exclude some ranges from crashk_res and add rest to memmap */
610 ret = memmap_exclude_ranges(image, cmem, crashk_res.start,
615 for (i = 0; i < cmem->nr_ranges; i++) {
616 ei.size = cmem->ranges[i].end - cmem->ranges[i].start + 1;
618 /* If entry is less than a page, skip it */
619 if (ei.size < PAGE_SIZE)
621 ei.addr = cmem->ranges[i].start;
623 add_e820_entry(params, &ei);
631 static int determine_backup_region(u64 start, u64 end, void *arg)
633 struct kimage *image = arg;
635 image->arch.backup_src_start = start;
636 image->arch.backup_src_sz = end - start + 1;
638 /* Expecting only one range for backup region */
642 int crash_load_segments(struct kimage *image)
644 unsigned long src_start, src_sz, elf_sz;
649 * Determine and load a segment for backup area. First 640K RAM
650 * region is backup source
653 ret = walk_system_ram_res(KEXEC_BACKUP_SRC_START, KEXEC_BACKUP_SRC_END,
654 image, determine_backup_region);
656 /* Zero or postive return values are ok */
660 src_start = image->arch.backup_src_start;
661 src_sz = image->arch.backup_src_sz;
663 /* Add backup segment. */
666 * Ideally there is no source for backup segment. This is
667 * copied in purgatory after crash. Just add a zero filled
668 * segment for now to make sure checksum logic works fine.
670 ret = kexec_add_buffer(image, (char *)&crash_zero_bytes,
671 sizeof(crash_zero_bytes), src_sz,
673 &image->arch.backup_load_addr);
676 pr_debug("Loaded backup region at 0x%lx backup_start=0x%lx memsz=0x%lx\n",
677 image->arch.backup_load_addr, src_start, src_sz);
680 /* Prepare elf headers and add a segment */
681 ret = prepare_elf_headers(image, &elf_addr, &elf_sz);
685 image->arch.elf_headers = elf_addr;
686 image->arch.elf_headers_sz = elf_sz;
688 ret = kexec_add_buffer(image, (char *)elf_addr, elf_sz, elf_sz,
689 ELF_CORE_HEADER_ALIGN, 0, -1, 0,
690 &image->arch.elf_load_addr);
692 vfree((void *)image->arch.elf_headers);
695 pr_debug("Loaded ELF headers at 0x%lx bufsz=0x%lx memsz=0x%lx\n",
696 image->arch.elf_load_addr, elf_sz, elf_sz);
700 #endif /* CONFIG_KEXEC_FILE */