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>
39 /* Alignment required for elf header segment */
40 #define ELF_CORE_HEADER_ALIGN 4096
42 /* This primarily represents number of split ranges due to exclusion */
43 #define CRASH_MAX_RANGES 16
45 struct crash_mem_range
{
50 unsigned int nr_ranges
;
51 struct crash_mem_range ranges
[CRASH_MAX_RANGES
];
54 /* Misc data about ram ranges needed to prepare elf headers */
55 struct crash_elf_data
{
58 * Total number of ram ranges we have after various adjustments for
59 * GART, crash reserved region etc.
61 unsigned int max_nr_ranges
;
62 unsigned long gart_start
, gart_end
;
64 /* Pointer to elf header */
66 /* Pointer to next phdr */
71 /* Used while preparing memory map entries for second kernel */
72 struct crash_memmap_data
{
73 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();
130 disable_local_APIC();
133 static void kdump_nmi_shootdown_cpus(void)
136 nmi_shootdown_cpus(kdump_nmi_callback
);
138 disable_local_APIC();
142 static void kdump_nmi_shootdown_cpus(void)
144 /* There are no cpus to shootdown */
148 void native_machine_crash_shutdown(struct pt_regs
*regs
)
150 /* This function is only called after the system
151 * has panicked or is otherwise in a critical state.
152 * The minimum amount of code to allow a kexec'd kernel
153 * to run successfully needs to happen here.
155 * In practice this means shooting down the other cpus in
158 /* The kernel is broken so disable interrupts */
161 kdump_nmi_shootdown_cpus();
164 * VMCLEAR VMCSs loaded on this cpu if needed.
166 cpu_crash_vmclear_loaded_vmcss();
168 /* Booting kdump kernel with VMX or SVM enabled won't work,
169 * because (among other limitations) we can't disable paging
170 * with the virt flags.
172 cpu_emergency_vmxoff();
173 cpu_emergency_svm_disable();
175 #ifdef CONFIG_X86_IO_APIC
176 /* Prevent crash_kexec() from deadlocking on ioapic_lock. */
181 #ifdef CONFIG_HPET_TIMER
184 crash_save_cpu(regs
, safe_smp_processor_id());
187 #ifdef CONFIG_KEXEC_FILE
188 static int get_nr_ram_ranges_callback(unsigned long start_pfn
,
189 unsigned long nr_pfn
, void *arg
)
191 int *nr_ranges
= arg
;
197 static int get_gart_ranges_callback(u64 start
, u64 end
, void *arg
)
199 struct crash_elf_data
*ced
= arg
;
201 ced
->gart_start
= start
;
204 /* Not expecting more than 1 gart aperture */
209 /* Gather all the required information to prepare elf headers for ram regions */
210 static void fill_up_crash_elf_data(struct crash_elf_data
*ced
,
211 struct kimage
*image
)
213 unsigned int nr_ranges
= 0;
217 walk_system_ram_range(0, -1, &nr_ranges
,
218 get_nr_ram_ranges_callback
);
220 ced
->max_nr_ranges
= nr_ranges
;
223 * We don't create ELF headers for GART aperture as an attempt
224 * to dump this memory in second kernel leads to hang/crash.
225 * If gart aperture is present, one needs to exclude that region
226 * and that could lead to need of extra phdr.
228 walk_iomem_res("GART", IORESOURCE_MEM
, 0, -1,
229 ced
, get_gart_ranges_callback
);
232 * If we have gart region, excluding that could potentially split
233 * a memory range, resulting in extra header. Account for that.
236 ced
->max_nr_ranges
++;
238 /* Exclusion of crash region could split memory ranges */
239 ced
->max_nr_ranges
++;
241 /* If crashk_low_res is not 0, another range split possible */
242 if (crashk_low_res
.end
)
243 ced
->max_nr_ranges
++;
246 static int exclude_mem_range(struct crash_mem
*mem
,
247 unsigned long long mstart
, unsigned long long mend
)
250 unsigned long long start
, end
;
251 struct crash_mem_range temp_range
= {0, 0};
253 for (i
= 0; i
< mem
->nr_ranges
; i
++) {
254 start
= mem
->ranges
[i
].start
;
255 end
= mem
->ranges
[i
].end
;
257 if (mstart
> end
|| mend
< start
)
260 /* Truncate any area outside of range */
266 /* Found completely overlapping range */
267 if (mstart
== start
&& mend
== end
) {
268 mem
->ranges
[i
].start
= 0;
269 mem
->ranges
[i
].end
= 0;
270 if (i
< mem
->nr_ranges
- 1) {
271 /* Shift rest of the ranges to left */
272 for (j
= i
; j
< mem
->nr_ranges
- 1; j
++) {
273 mem
->ranges
[j
].start
=
274 mem
->ranges
[j
+1].start
;
276 mem
->ranges
[j
+1].end
;
283 if (mstart
> start
&& mend
< end
) {
284 /* Split original range */
285 mem
->ranges
[i
].end
= mstart
- 1;
286 temp_range
.start
= mend
+ 1;
287 temp_range
.end
= end
;
288 } else if (mstart
!= start
)
289 mem
->ranges
[i
].end
= mstart
- 1;
291 mem
->ranges
[i
].start
= mend
+ 1;
295 /* If a split happend, add the split to array */
300 if (i
== CRASH_MAX_RANGES
- 1) {
301 pr_err("Too many crash ranges after split\n");
305 /* Location where new range should go */
307 if (j
< mem
->nr_ranges
) {
308 /* Move over all ranges one slot towards the end */
309 for (i
= mem
->nr_ranges
- 1; i
>= j
; i
--)
310 mem
->ranges
[i
+ 1] = mem
->ranges
[i
];
313 mem
->ranges
[j
].start
= temp_range
.start
;
314 mem
->ranges
[j
].end
= temp_range
.end
;
320 * Look for any unwanted ranges between mstart, mend and remove them. This
321 * might lead to split and split ranges are put in ced->mem.ranges[] array
323 static int elf_header_exclude_ranges(struct crash_elf_data
*ced
,
324 unsigned long long mstart
, unsigned long long mend
)
326 struct crash_mem
*cmem
= &ced
->mem
;
329 memset(cmem
->ranges
, 0, sizeof(cmem
->ranges
));
331 cmem
->ranges
[0].start
= mstart
;
332 cmem
->ranges
[0].end
= mend
;
335 /* Exclude crashkernel region */
336 ret
= exclude_mem_range(cmem
, crashk_res
.start
, crashk_res
.end
);
340 if (crashk_low_res
.end
) {
341 ret
= exclude_mem_range(cmem
, crashk_low_res
.start
, crashk_low_res
.end
);
346 /* Exclude GART region */
348 ret
= exclude_mem_range(cmem
, ced
->gart_start
, ced
->gart_end
);
356 static int prepare_elf64_ram_headers_callback(u64 start
, u64 end
, void *arg
)
358 struct crash_elf_data
*ced
= arg
;
361 unsigned long mstart
, mend
;
362 struct kimage
*image
= ced
->image
;
363 struct crash_mem
*cmem
;
368 /* Exclude unwanted mem ranges */
369 ret
= elf_header_exclude_ranges(ced
, start
, end
);
373 /* Go through all the ranges in ced->mem.ranges[] and prepare phdr */
376 for (i
= 0; i
< cmem
->nr_ranges
; i
++) {
377 mstart
= cmem
->ranges
[i
].start
;
378 mend
= cmem
->ranges
[i
].end
;
381 ced
->bufp
+= sizeof(Elf64_Phdr
);
383 phdr
->p_type
= PT_LOAD
;
384 phdr
->p_flags
= PF_R
|PF_W
|PF_X
;
385 phdr
->p_offset
= mstart
;
388 * If a range matches backup region, adjust offset to backup
391 if (mstart
== image
->arch
.backup_src_start
&&
392 (mend
- mstart
+ 1) == image
->arch
.backup_src_sz
)
393 phdr
->p_offset
= image
->arch
.backup_load_addr
;
395 phdr
->p_paddr
= mstart
;
396 phdr
->p_vaddr
= (unsigned long long) __va(mstart
);
397 phdr
->p_filesz
= phdr
->p_memsz
= mend
- mstart
+ 1;
400 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",
401 phdr
, phdr
->p_vaddr
, phdr
->p_paddr
, phdr
->p_filesz
,
402 ehdr
->e_phnum
, phdr
->p_offset
);
408 static int prepare_elf64_headers(struct crash_elf_data
*ced
,
409 void **addr
, unsigned long *sz
)
413 unsigned long nr_cpus
= num_possible_cpus(), nr_phdr
, elf_sz
;
414 unsigned char *buf
, *bufp
;
416 unsigned long long notes_addr
;
419 /* extra phdr for vmcoreinfo elf note */
420 nr_phdr
= nr_cpus
+ 1;
421 nr_phdr
+= ced
->max_nr_ranges
;
424 * kexec-tools creates an extra PT_LOAD phdr for kernel text mapping
425 * area on x86_64 (ffffffff80000000 - ffffffffa0000000).
426 * I think this is required by tools like gdb. So same physical
427 * memory will be mapped in two elf headers. One will contain kernel
428 * text virtual addresses and other will have __va(physical) addresses.
432 elf_sz
= sizeof(Elf64_Ehdr
) + nr_phdr
* sizeof(Elf64_Phdr
);
433 elf_sz
= ALIGN(elf_sz
, ELF_CORE_HEADER_ALIGN
);
435 buf
= vzalloc(elf_sz
);
440 ehdr
= (Elf64_Ehdr
*)bufp
;
441 bufp
+= sizeof(Elf64_Ehdr
);
442 memcpy(ehdr
->e_ident
, ELFMAG
, SELFMAG
);
443 ehdr
->e_ident
[EI_CLASS
] = ELFCLASS64
;
444 ehdr
->e_ident
[EI_DATA
] = ELFDATA2LSB
;
445 ehdr
->e_ident
[EI_VERSION
] = EV_CURRENT
;
446 ehdr
->e_ident
[EI_OSABI
] = ELF_OSABI
;
447 memset(ehdr
->e_ident
+ EI_PAD
, 0, EI_NIDENT
- EI_PAD
);
448 ehdr
->e_type
= ET_CORE
;
449 ehdr
->e_machine
= ELF_ARCH
;
450 ehdr
->e_version
= EV_CURRENT
;
451 ehdr
->e_phoff
= sizeof(Elf64_Ehdr
);
452 ehdr
->e_ehsize
= sizeof(Elf64_Ehdr
);
453 ehdr
->e_phentsize
= sizeof(Elf64_Phdr
);
455 /* Prepare one phdr of type PT_NOTE for each present cpu */
456 for_each_present_cpu(cpu
) {
457 phdr
= (Elf64_Phdr
*)bufp
;
458 bufp
+= sizeof(Elf64_Phdr
);
459 phdr
->p_type
= PT_NOTE
;
460 notes_addr
= per_cpu_ptr_to_phys(per_cpu_ptr(crash_notes
, cpu
));
461 phdr
->p_offset
= phdr
->p_paddr
= notes_addr
;
462 phdr
->p_filesz
= phdr
->p_memsz
= sizeof(note_buf_t
);
466 /* Prepare one PT_NOTE header for vmcoreinfo */
467 phdr
= (Elf64_Phdr
*)bufp
;
468 bufp
+= sizeof(Elf64_Phdr
);
469 phdr
->p_type
= PT_NOTE
;
470 phdr
->p_offset
= phdr
->p_paddr
= paddr_vmcoreinfo_note();
471 phdr
->p_filesz
= phdr
->p_memsz
= sizeof(vmcoreinfo_note
);
475 /* Prepare PT_LOAD type program header for kernel text region */
476 phdr
= (Elf64_Phdr
*)bufp
;
477 bufp
+= sizeof(Elf64_Phdr
);
478 phdr
->p_type
= PT_LOAD
;
479 phdr
->p_flags
= PF_R
|PF_W
|PF_X
;
480 phdr
->p_vaddr
= (Elf64_Addr
)_text
;
481 phdr
->p_filesz
= phdr
->p_memsz
= _end
- _text
;
482 phdr
->p_offset
= phdr
->p_paddr
= __pa_symbol(_text
);
486 /* Prepare PT_LOAD headers for system ram chunks. */
489 ret
= walk_system_ram_res(0, -1, ced
,
490 prepare_elf64_ram_headers_callback
);
499 /* Prepare elf headers. Return addr and size */
500 static int prepare_elf_headers(struct kimage
*image
, void **addr
,
503 struct crash_elf_data
*ced
;
506 ced
= kzalloc(sizeof(*ced
), GFP_KERNEL
);
510 fill_up_crash_elf_data(ced
, image
);
512 /* By default prepare 64bit headers */
513 ret
= prepare_elf64_headers(ced
, addr
, sz
);
518 static int add_e820_entry(struct boot_params
*params
, struct e820entry
*entry
)
520 unsigned int nr_e820_entries
;
522 nr_e820_entries
= params
->e820_entries
;
523 if (nr_e820_entries
>= E820MAX
)
526 memcpy(¶ms
->e820_map
[nr_e820_entries
], entry
,
527 sizeof(struct e820entry
));
528 params
->e820_entries
++;
532 static int memmap_entry_callback(u64 start
, u64 end
, void *arg
)
534 struct crash_memmap_data
*cmd
= arg
;
535 struct boot_params
*params
= cmd
->params
;
539 ei
.size
= end
- start
+ 1;
541 add_e820_entry(params
, &ei
);
546 static int memmap_exclude_ranges(struct kimage
*image
, struct crash_mem
*cmem
,
547 unsigned long long mstart
,
548 unsigned long long mend
)
550 unsigned long start
, end
;
553 cmem
->ranges
[0].start
= mstart
;
554 cmem
->ranges
[0].end
= mend
;
557 /* Exclude Backup region */
558 start
= image
->arch
.backup_load_addr
;
559 end
= start
+ image
->arch
.backup_src_sz
- 1;
560 ret
= exclude_mem_range(cmem
, start
, end
);
564 /* Exclude elf header region */
565 start
= image
->arch
.elf_load_addr
;
566 end
= start
+ image
->arch
.elf_headers_sz
- 1;
567 return exclude_mem_range(cmem
, start
, end
);
570 /* Prepare memory map for crash dump kernel */
571 int crash_setup_memmap_entries(struct kimage
*image
, struct boot_params
*params
)
576 struct crash_memmap_data cmd
;
577 struct crash_mem
*cmem
;
579 cmem
= vzalloc(sizeof(struct crash_mem
));
583 memset(&cmd
, 0, sizeof(struct crash_memmap_data
));
586 /* Add first 640K segment */
587 ei
.addr
= image
->arch
.backup_src_start
;
588 ei
.size
= image
->arch
.backup_src_sz
;
590 add_e820_entry(params
, &ei
);
592 /* Add ACPI tables */
593 cmd
.type
= E820_ACPI
;
594 flags
= IORESOURCE_MEM
| IORESOURCE_BUSY
;
595 walk_iomem_res("ACPI Tables", flags
, 0, -1, &cmd
,
596 memmap_entry_callback
);
598 /* Add ACPI Non-volatile Storage */
600 walk_iomem_res("ACPI Non-volatile Storage", flags
, 0, -1, &cmd
,
601 memmap_entry_callback
);
603 /* Add crashk_low_res region */
604 if (crashk_low_res
.end
) {
605 ei
.addr
= crashk_low_res
.start
;
606 ei
.size
= crashk_low_res
.end
- crashk_low_res
.start
+ 1;
608 add_e820_entry(params
, &ei
);
611 /* Exclude some ranges from crashk_res and add rest to memmap */
612 ret
= memmap_exclude_ranges(image
, cmem
, crashk_res
.start
,
617 for (i
= 0; i
< cmem
->nr_ranges
; i
++) {
618 ei
.size
= cmem
->ranges
[i
].end
- cmem
->ranges
[i
].start
+ 1;
620 /* If entry is less than a page, skip it */
621 if (ei
.size
< PAGE_SIZE
)
623 ei
.addr
= cmem
->ranges
[i
].start
;
625 add_e820_entry(params
, &ei
);
633 static int determine_backup_region(u64 start
, u64 end
, void *arg
)
635 struct kimage
*image
= arg
;
637 image
->arch
.backup_src_start
= start
;
638 image
->arch
.backup_src_sz
= end
- start
+ 1;
640 /* Expecting only one range for backup region */
644 int crash_load_segments(struct kimage
*image
)
646 unsigned long src_start
, src_sz
, elf_sz
;
651 * Determine and load a segment for backup area. First 640K RAM
652 * region is backup source
655 ret
= walk_system_ram_res(KEXEC_BACKUP_SRC_START
, KEXEC_BACKUP_SRC_END
,
656 image
, determine_backup_region
);
658 /* Zero or postive return values are ok */
662 src_start
= image
->arch
.backup_src_start
;
663 src_sz
= image
->arch
.backup_src_sz
;
665 /* Add backup segment. */
668 * Ideally there is no source for backup segment. This is
669 * copied in purgatory after crash. Just add a zero filled
670 * segment for now to make sure checksum logic works fine.
672 ret
= kexec_add_buffer(image
, (char *)&crash_zero_bytes
,
673 sizeof(crash_zero_bytes
), src_sz
,
675 &image
->arch
.backup_load_addr
);
678 pr_debug("Loaded backup region at 0x%lx backup_start=0x%lx memsz=0x%lx\n",
679 image
->arch
.backup_load_addr
, src_start
, src_sz
);
682 /* Prepare elf headers and add a segment */
683 ret
= prepare_elf_headers(image
, &elf_addr
, &elf_sz
);
687 image
->arch
.elf_headers
= elf_addr
;
688 image
->arch
.elf_headers_sz
= elf_sz
;
690 ret
= kexec_add_buffer(image
, (char *)elf_addr
, elf_sz
, elf_sz
,
691 ELF_CORE_HEADER_ALIGN
, 0, -1, 0,
692 &image
->arch
.elf_load_addr
);
694 vfree((void *)image
->arch
.elf_headers
);
697 pr_debug("Loaded ELF headers at 0x%lx bufsz=0x%lx memsz=0x%lx\n",
698 image
->arch
.elf_load_addr
, elf_sz
, elf_sz
);
702 #endif /* CONFIG_KEXEC_FILE */