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/e820/types.h>
34 #include <asm/io_apic.h>
36 #include <linux/kdebug.h>
38 #include <asm/reboot.h>
39 #include <asm/virtext.h>
40 #include <asm/intel_pt.h>
42 /* Alignment required for elf header segment */
43 #define ELF_CORE_HEADER_ALIGN 4096
45 /* This primarily represents number of split ranges due to exclusion */
46 #define CRASH_MAX_RANGES 16
48 struct crash_mem_range {
53 unsigned int nr_ranges;
54 struct crash_mem_range ranges[CRASH_MAX_RANGES];
57 /* Misc data about ram ranges needed to prepare elf headers */
58 struct crash_elf_data {
61 * Total number of ram ranges we have after various adjustments for
62 * crash reserved region, etc.
64 unsigned int max_nr_ranges;
66 /* Pointer to elf header */
68 /* Pointer to next phdr */
73 /* Used while preparing memory map entries for second kernel */
74 struct crash_memmap_data {
75 struct boot_params *params;
81 * This is used to VMCLEAR all VMCSs loaded on the
82 * processor. And when loading kvm_intel module, the
83 * callback function pointer will be assigned.
87 crash_vmclear_fn __rcu *crash_vmclear_loaded_vmcss = NULL;
88 EXPORT_SYMBOL_GPL(crash_vmclear_loaded_vmcss);
89 unsigned long crash_zero_bytes;
91 static inline void cpu_crash_vmclear_loaded_vmcss(void)
93 crash_vmclear_fn *do_vmclear_operation = NULL;
96 do_vmclear_operation = rcu_dereference(crash_vmclear_loaded_vmcss);
97 if (do_vmclear_operation)
98 do_vmclear_operation();
102 #if defined(CONFIG_SMP) && defined(CONFIG_X86_LOCAL_APIC)
104 static void kdump_nmi_callback(int cpu, struct pt_regs *regs)
107 struct pt_regs fixed_regs;
109 if (!user_mode(regs)) {
110 crash_fixup_ss_esp(&fixed_regs, regs);
114 crash_save_cpu(regs, cpu);
117 * VMCLEAR VMCSs loaded on all cpus if needed.
119 cpu_crash_vmclear_loaded_vmcss();
121 /* Disable VMX or SVM if needed.
123 * We need to disable virtualization on all CPUs.
124 * Having VMX or SVM enabled on any CPU may break rebooting
125 * after the kdump kernel has finished its task.
127 cpu_emergency_vmxoff();
128 cpu_emergency_svm_disable();
131 * Disable Intel PT to stop its logging
133 cpu_emergency_stop_pt();
135 disable_local_APIC();
138 void kdump_nmi_shootdown_cpus(void)
140 nmi_shootdown_cpus(kdump_nmi_callback);
142 disable_local_APIC();
145 /* Override the weak function in kernel/panic.c */
146 void crash_smp_send_stop(void)
148 static int cpus_stopped;
153 if (smp_ops.crash_stop_other_cpus)
154 smp_ops.crash_stop_other_cpus();
162 void crash_smp_send_stop(void)
164 /* There are no cpus to shootdown */
168 void native_machine_crash_shutdown(struct pt_regs *regs)
170 /* This function is only called after the system
171 * has panicked or is otherwise in a critical state.
172 * The minimum amount of code to allow a kexec'd kernel
173 * to run successfully needs to happen here.
175 * In practice this means shooting down the other cpus in
178 /* The kernel is broken so disable interrupts */
181 crash_smp_send_stop();
184 * VMCLEAR VMCSs loaded on this cpu if needed.
186 cpu_crash_vmclear_loaded_vmcss();
188 /* Booting kdump kernel with VMX or SVM enabled won't work,
189 * because (among other limitations) we can't disable paging
190 * with the virt flags.
192 cpu_emergency_vmxoff();
193 cpu_emergency_svm_disable();
196 * Disable Intel PT to stop its logging
198 cpu_emergency_stop_pt();
200 #ifdef CONFIG_X86_IO_APIC
201 /* Prevent crash_kexec() from deadlocking on ioapic_lock. */
206 #ifdef CONFIG_HPET_TIMER
209 crash_save_cpu(regs, safe_smp_processor_id());
212 #ifdef CONFIG_KEXEC_FILE
213 static int get_nr_ram_ranges_callback(u64 start, u64 end, void *arg)
215 unsigned int *nr_ranges = arg;
222 /* Gather all the required information to prepare elf headers for ram regions */
223 static void fill_up_crash_elf_data(struct crash_elf_data *ced,
224 struct kimage *image)
226 unsigned int nr_ranges = 0;
230 walk_system_ram_res(0, -1, &nr_ranges,
231 get_nr_ram_ranges_callback);
233 ced->max_nr_ranges = nr_ranges;
235 /* Exclusion of crash region could split memory ranges */
236 ced->max_nr_ranges++;
238 /* If crashk_low_res is not 0, another range split possible */
239 if (crashk_low_res.end)
240 ced->max_nr_ranges++;
243 static int exclude_mem_range(struct crash_mem *mem,
244 unsigned long long mstart, unsigned long long mend)
247 unsigned long long start, end;
248 struct crash_mem_range temp_range = {0, 0};
250 for (i = 0; i < mem->nr_ranges; i++) {
251 start = mem->ranges[i].start;
252 end = mem->ranges[i].end;
254 if (mstart > end || mend < start)
257 /* Truncate any area outside of range */
263 /* Found completely overlapping range */
264 if (mstart == start && mend == end) {
265 mem->ranges[i].start = 0;
266 mem->ranges[i].end = 0;
267 if (i < mem->nr_ranges - 1) {
268 /* Shift rest of the ranges to left */
269 for (j = i; j < mem->nr_ranges - 1; j++) {
270 mem->ranges[j].start =
271 mem->ranges[j+1].start;
273 mem->ranges[j+1].end;
280 if (mstart > start && mend < end) {
281 /* Split original range */
282 mem->ranges[i].end = mstart - 1;
283 temp_range.start = mend + 1;
284 temp_range.end = end;
285 } else if (mstart != start)
286 mem->ranges[i].end = mstart - 1;
288 mem->ranges[i].start = mend + 1;
292 /* If a split happend, add the split to array */
297 if (i == CRASH_MAX_RANGES - 1) {
298 pr_err("Too many crash ranges after split\n");
302 /* Location where new range should go */
304 if (j < mem->nr_ranges) {
305 /* Move over all ranges one slot towards the end */
306 for (i = mem->nr_ranges - 1; i >= j; i--)
307 mem->ranges[i + 1] = mem->ranges[i];
310 mem->ranges[j].start = temp_range.start;
311 mem->ranges[j].end = temp_range.end;
317 * Look for any unwanted ranges between mstart, mend and remove them. This
318 * might lead to split and split ranges are put in ced->mem.ranges[] array
320 static int elf_header_exclude_ranges(struct crash_elf_data *ced,
321 unsigned long long mstart, unsigned long long mend)
323 struct crash_mem *cmem = &ced->mem;
326 memset(cmem->ranges, 0, sizeof(cmem->ranges));
328 cmem->ranges[0].start = mstart;
329 cmem->ranges[0].end = mend;
332 /* Exclude crashkernel region */
333 ret = exclude_mem_range(cmem, crashk_res.start, crashk_res.end);
337 if (crashk_low_res.end) {
338 ret = exclude_mem_range(cmem, crashk_low_res.start, crashk_low_res.end);
346 static int prepare_elf64_ram_headers_callback(u64 start, u64 end, void *arg)
348 struct crash_elf_data *ced = arg;
351 unsigned long mstart, mend;
352 struct kimage *image = ced->image;
353 struct crash_mem *cmem;
358 /* Exclude unwanted mem ranges */
359 ret = elf_header_exclude_ranges(ced, start, end);
363 /* Go through all the ranges in ced->mem.ranges[] and prepare phdr */
366 for (i = 0; i < cmem->nr_ranges; i++) {
367 mstart = cmem->ranges[i].start;
368 mend = cmem->ranges[i].end;
371 ced->bufp += sizeof(Elf64_Phdr);
373 phdr->p_type = PT_LOAD;
374 phdr->p_flags = PF_R|PF_W|PF_X;
375 phdr->p_offset = mstart;
378 * If a range matches backup region, adjust offset to backup
381 if (mstart == image->arch.backup_src_start &&
382 (mend - mstart + 1) == image->arch.backup_src_sz)
383 phdr->p_offset = image->arch.backup_load_addr;
385 phdr->p_paddr = mstart;
386 phdr->p_vaddr = (unsigned long long) __va(mstart);
387 phdr->p_filesz = phdr->p_memsz = mend - mstart + 1;
390 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",
391 phdr, phdr->p_vaddr, phdr->p_paddr, phdr->p_filesz,
392 ehdr->e_phnum, phdr->p_offset);
398 static int prepare_elf64_headers(struct crash_elf_data *ced,
399 void **addr, unsigned long *sz)
403 unsigned long nr_cpus = num_possible_cpus(), nr_phdr, elf_sz;
404 unsigned char *buf, *bufp;
406 unsigned long long notes_addr;
409 /* extra phdr for vmcoreinfo elf note */
410 nr_phdr = nr_cpus + 1;
411 nr_phdr += ced->max_nr_ranges;
414 * kexec-tools creates an extra PT_LOAD phdr for kernel text mapping
415 * area on x86_64 (ffffffff80000000 - ffffffffa0000000).
416 * I think this is required by tools like gdb. So same physical
417 * memory will be mapped in two elf headers. One will contain kernel
418 * text virtual addresses and other will have __va(physical) addresses.
422 elf_sz = sizeof(Elf64_Ehdr) + nr_phdr * sizeof(Elf64_Phdr);
423 elf_sz = ALIGN(elf_sz, ELF_CORE_HEADER_ALIGN);
425 buf = vzalloc(elf_sz);
430 ehdr = (Elf64_Ehdr *)bufp;
431 bufp += sizeof(Elf64_Ehdr);
432 memcpy(ehdr->e_ident, ELFMAG, SELFMAG);
433 ehdr->e_ident[EI_CLASS] = ELFCLASS64;
434 ehdr->e_ident[EI_DATA] = ELFDATA2LSB;
435 ehdr->e_ident[EI_VERSION] = EV_CURRENT;
436 ehdr->e_ident[EI_OSABI] = ELF_OSABI;
437 memset(ehdr->e_ident + EI_PAD, 0, EI_NIDENT - EI_PAD);
438 ehdr->e_type = ET_CORE;
439 ehdr->e_machine = ELF_ARCH;
440 ehdr->e_version = EV_CURRENT;
441 ehdr->e_phoff = sizeof(Elf64_Ehdr);
442 ehdr->e_ehsize = sizeof(Elf64_Ehdr);
443 ehdr->e_phentsize = sizeof(Elf64_Phdr);
445 /* Prepare one phdr of type PT_NOTE for each present cpu */
446 for_each_present_cpu(cpu) {
447 phdr = (Elf64_Phdr *)bufp;
448 bufp += sizeof(Elf64_Phdr);
449 phdr->p_type = PT_NOTE;
450 notes_addr = per_cpu_ptr_to_phys(per_cpu_ptr(crash_notes, cpu));
451 phdr->p_offset = phdr->p_paddr = notes_addr;
452 phdr->p_filesz = phdr->p_memsz = sizeof(note_buf_t);
456 /* Prepare one PT_NOTE header for vmcoreinfo */
457 phdr = (Elf64_Phdr *)bufp;
458 bufp += sizeof(Elf64_Phdr);
459 phdr->p_type = PT_NOTE;
460 phdr->p_offset = phdr->p_paddr = paddr_vmcoreinfo_note();
461 phdr->p_filesz = phdr->p_memsz = VMCOREINFO_NOTE_SIZE;
465 /* Prepare PT_LOAD type program header for kernel text region */
466 phdr = (Elf64_Phdr *)bufp;
467 bufp += sizeof(Elf64_Phdr);
468 phdr->p_type = PT_LOAD;
469 phdr->p_flags = PF_R|PF_W|PF_X;
470 phdr->p_vaddr = (Elf64_Addr)_text;
471 phdr->p_filesz = phdr->p_memsz = _end - _text;
472 phdr->p_offset = phdr->p_paddr = __pa_symbol(_text);
476 /* Prepare PT_LOAD headers for system ram chunks. */
479 ret = walk_system_ram_res(0, -1, ced,
480 prepare_elf64_ram_headers_callback);
489 /* Prepare elf headers. Return addr and size */
490 static int prepare_elf_headers(struct kimage *image, void **addr,
493 struct crash_elf_data *ced;
496 ced = kzalloc(sizeof(*ced), GFP_KERNEL);
500 fill_up_crash_elf_data(ced, image);
502 /* By default prepare 64bit headers */
503 ret = prepare_elf64_headers(ced, addr, sz);
508 static int add_e820_entry(struct boot_params *params, struct e820_entry *entry)
510 unsigned int nr_e820_entries;
512 nr_e820_entries = params->e820_entries;
513 if (nr_e820_entries >= E820_MAX_ENTRIES_ZEROPAGE)
516 memcpy(¶ms->e820_table[nr_e820_entries], entry,
517 sizeof(struct e820_entry));
518 params->e820_entries++;
522 static int memmap_entry_callback(u64 start, u64 end, void *arg)
524 struct crash_memmap_data *cmd = arg;
525 struct boot_params *params = cmd->params;
526 struct e820_entry ei;
529 ei.size = end - start + 1;
531 add_e820_entry(params, &ei);
536 static int memmap_exclude_ranges(struct kimage *image, struct crash_mem *cmem,
537 unsigned long long mstart,
538 unsigned long long mend)
540 unsigned long start, end;
543 cmem->ranges[0].start = mstart;
544 cmem->ranges[0].end = mend;
547 /* Exclude Backup region */
548 start = image->arch.backup_load_addr;
549 end = start + image->arch.backup_src_sz - 1;
550 ret = exclude_mem_range(cmem, start, end);
554 /* Exclude elf header region */
555 start = image->arch.elf_load_addr;
556 end = start + image->arch.elf_headers_sz - 1;
557 return exclude_mem_range(cmem, start, end);
560 /* Prepare memory map for crash dump kernel */
561 int crash_setup_memmap_entries(struct kimage *image, struct boot_params *params)
565 struct e820_entry ei;
566 struct crash_memmap_data cmd;
567 struct crash_mem *cmem;
569 cmem = vzalloc(struct_size(cmem, ranges, 1));
573 memset(&cmd, 0, sizeof(struct crash_memmap_data));
576 /* Add first 640K segment */
577 ei.addr = image->arch.backup_src_start;
578 ei.size = image->arch.backup_src_sz;
579 ei.type = E820_TYPE_RAM;
580 add_e820_entry(params, &ei);
582 /* Add ACPI tables */
583 cmd.type = E820_TYPE_ACPI;
584 flags = IORESOURCE_MEM | IORESOURCE_BUSY;
585 walk_iomem_res_desc(IORES_DESC_ACPI_TABLES, flags, 0, -1, &cmd,
586 memmap_entry_callback);
588 /* Add ACPI Non-volatile Storage */
589 cmd.type = E820_TYPE_NVS;
590 walk_iomem_res_desc(IORES_DESC_ACPI_NV_STORAGE, flags, 0, -1, &cmd,
591 memmap_entry_callback);
593 /* Add crashk_low_res region */
594 if (crashk_low_res.end) {
595 ei.addr = crashk_low_res.start;
596 ei.size = crashk_low_res.end - crashk_low_res.start + 1;
597 ei.type = E820_TYPE_RAM;
598 add_e820_entry(params, &ei);
601 /* Exclude some ranges from crashk_res and add rest to memmap */
602 ret = memmap_exclude_ranges(image, cmem, crashk_res.start,
607 for (i = 0; i < cmem->nr_ranges; i++) {
608 ei.size = cmem->ranges[i].end - cmem->ranges[i].start + 1;
610 /* If entry is less than a page, skip it */
611 if (ei.size < PAGE_SIZE)
613 ei.addr = cmem->ranges[i].start;
614 ei.type = E820_TYPE_RAM;
615 add_e820_entry(params, &ei);
623 static int determine_backup_region(u64 start, u64 end, void *arg)
625 struct kimage *image = arg;
627 image->arch.backup_src_start = start;
628 image->arch.backup_src_sz = end - start + 1;
630 /* Expecting only one range for backup region */
634 int crash_load_segments(struct kimage *image)
637 struct kexec_buf kbuf = { .image = image, .buf_min = 0,
638 .buf_max = ULONG_MAX, .top_down = false };
641 * Determine and load a segment for backup area. First 640K RAM
642 * region is backup source
645 ret = walk_system_ram_res(KEXEC_BACKUP_SRC_START, KEXEC_BACKUP_SRC_END,
646 image, determine_backup_region);
648 /* Zero or postive return values are ok */
652 /* Add backup segment. */
653 if (image->arch.backup_src_sz) {
654 kbuf.buffer = &crash_zero_bytes;
655 kbuf.bufsz = sizeof(crash_zero_bytes);
656 kbuf.memsz = image->arch.backup_src_sz;
657 kbuf.buf_align = PAGE_SIZE;
659 * Ideally there is no source for backup segment. This is
660 * copied in purgatory after crash. Just add a zero filled
661 * segment for now to make sure checksum logic works fine.
663 ret = kexec_add_buffer(&kbuf);
666 image->arch.backup_load_addr = kbuf.mem;
667 pr_debug("Loaded backup region at 0x%lx backup_start=0x%lx memsz=0x%lx\n",
668 image->arch.backup_load_addr,
669 image->arch.backup_src_start, kbuf.memsz);
672 /* Prepare elf headers and add a segment */
673 ret = prepare_elf_headers(image, &kbuf.buffer, &kbuf.bufsz);
677 image->arch.elf_headers = kbuf.buffer;
678 image->arch.elf_headers_sz = kbuf.bufsz;
680 kbuf.memsz = kbuf.bufsz;
681 kbuf.buf_align = ELF_CORE_HEADER_ALIGN;
682 ret = kexec_add_buffer(&kbuf);
684 vfree((void *)image->arch.elf_headers);
687 image->arch.elf_load_addr = kbuf.mem;
688 pr_debug("Loaded ELF headers at 0x%lx bufsz=0x%lx memsz=0x%lx\n",
689 image->arch.elf_load_addr, kbuf.bufsz, kbuf.bufsz);
693 #endif /* CONFIG_KEXEC_FILE */