Loading...
1/*
2 * Architecture specific (i386/x86_64) functions for kexec based crash dumps.
3 *
4 * Created by: Hariprasad Nellitheertha (hari@in.ibm.com)
5 *
6 * Copyright (C) IBM Corporation, 2004. All rights reserved.
7 * Copyright (C) Red Hat Inc., 2014. All rights reserved.
8 * Authors:
9 * Vivek Goyal <vgoyal@redhat.com>
10 *
11 */
12
13#define pr_fmt(fmt) "kexec: " fmt
14
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
27#include <asm/processor.h>
28#include <asm/hardirq.h>
29#include <asm/nmi.h>
30#include <asm/hw_irq.h>
31#include <asm/apic.h>
32#include <asm/io_apic.h>
33#include <asm/hpet.h>
34#include <linux/kdebug.h>
35#include <asm/cpu.h>
36#include <asm/reboot.h>
37#include <asm/virtext.h>
38#include <asm/intel_pt.h>
39
40/* Alignment required for elf header segment */
41#define ELF_CORE_HEADER_ALIGN 4096
42
43/* This primarily represents number of split ranges due to exclusion */
44#define CRASH_MAX_RANGES 16
45
46struct crash_mem_range {
47 u64 start, end;
48};
49
50struct crash_mem {
51 unsigned int nr_ranges;
52 struct crash_mem_range ranges[CRASH_MAX_RANGES];
53};
54
55/* Misc data about ram ranges needed to prepare elf headers */
56struct crash_elf_data {
57 struct kimage *image;
58 /*
59 * Total number of ram ranges we have after various adjustments for
60 * crash reserved region, etc.
61 */
62 unsigned int max_nr_ranges;
63
64 /* Pointer to elf header */
65 void *ehdr;
66 /* Pointer to next phdr */
67 void *bufp;
68 struct crash_mem mem;
69};
70
71/* Used while preparing memory map entries for second kernel */
72struct crash_memmap_data {
73 struct boot_params *params;
74 /* Type of memory */
75 unsigned int type;
76};
77
78/*
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.
82 *
83 * protected by rcu.
84 */
85crash_vmclear_fn __rcu *crash_vmclear_loaded_vmcss = NULL;
86EXPORT_SYMBOL_GPL(crash_vmclear_loaded_vmcss);
87unsigned long crash_zero_bytes;
88
89static inline void cpu_crash_vmclear_loaded_vmcss(void)
90{
91 crash_vmclear_fn *do_vmclear_operation = NULL;
92
93 rcu_read_lock();
94 do_vmclear_operation = rcu_dereference(crash_vmclear_loaded_vmcss);
95 if (do_vmclear_operation)
96 do_vmclear_operation();
97 rcu_read_unlock();
98}
99
100#if defined(CONFIG_SMP) && defined(CONFIG_X86_LOCAL_APIC)
101
102static void kdump_nmi_callback(int cpu, struct pt_regs *regs)
103{
104#ifdef CONFIG_X86_32
105 struct pt_regs fixed_regs;
106
107 if (!user_mode(regs)) {
108 crash_fixup_ss_esp(&fixed_regs, regs);
109 regs = &fixed_regs;
110 }
111#endif
112 crash_save_cpu(regs, cpu);
113
114 /*
115 * VMCLEAR VMCSs loaded on all cpus if needed.
116 */
117 cpu_crash_vmclear_loaded_vmcss();
118
119 /* Disable VMX or SVM if needed.
120 *
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.
124 */
125 cpu_emergency_vmxoff();
126 cpu_emergency_svm_disable();
127
128 /*
129 * Disable Intel PT to stop its logging
130 */
131 cpu_emergency_stop_pt();
132
133 disable_local_APIC();
134}
135
136void kdump_nmi_shootdown_cpus(void)
137{
138 nmi_shootdown_cpus(kdump_nmi_callback);
139
140 disable_local_APIC();
141}
142
143/* Override the weak function in kernel/panic.c */
144void crash_smp_send_stop(void)
145{
146 static int cpus_stopped;
147
148 if (cpus_stopped)
149 return;
150
151 if (smp_ops.crash_stop_other_cpus)
152 smp_ops.crash_stop_other_cpus();
153 else
154 smp_send_stop();
155
156 cpus_stopped = 1;
157}
158
159#else
160void crash_smp_send_stop(void)
161{
162 /* There are no cpus to shootdown */
163}
164#endif
165
166void native_machine_crash_shutdown(struct pt_regs *regs)
167{
168 /* This function is only called after the system
169 * has panicked or is otherwise in a critical state.
170 * The minimum amount of code to allow a kexec'd kernel
171 * to run successfully needs to happen here.
172 *
173 * In practice this means shooting down the other cpus in
174 * an SMP system.
175 */
176 /* The kernel is broken so disable interrupts */
177 local_irq_disable();
178
179 crash_smp_send_stop();
180
181 /*
182 * VMCLEAR VMCSs loaded on this cpu if needed.
183 */
184 cpu_crash_vmclear_loaded_vmcss();
185
186 /* Booting kdump kernel with VMX or SVM enabled won't work,
187 * because (among other limitations) we can't disable paging
188 * with the virt flags.
189 */
190 cpu_emergency_vmxoff();
191 cpu_emergency_svm_disable();
192
193 /*
194 * Disable Intel PT to stop its logging
195 */
196 cpu_emergency_stop_pt();
197
198#ifdef CONFIG_X86_IO_APIC
199 /* Prevent crash_kexec() from deadlocking on ioapic_lock. */
200 ioapic_zap_locks();
201 disable_IO_APIC();
202#endif
203 lapic_shutdown();
204#ifdef CONFIG_HPET_TIMER
205 hpet_disable();
206#endif
207 crash_save_cpu(regs, safe_smp_processor_id());
208}
209
210#ifdef CONFIG_KEXEC_FILE
211static int get_nr_ram_ranges_callback(u64 start, u64 end, void *arg)
212{
213 unsigned int *nr_ranges = arg;
214
215 (*nr_ranges)++;
216 return 0;
217}
218
219
220/* Gather all the required information to prepare elf headers for ram regions */
221static void fill_up_crash_elf_data(struct crash_elf_data *ced,
222 struct kimage *image)
223{
224 unsigned int nr_ranges = 0;
225
226 ced->image = image;
227
228 walk_system_ram_res(0, -1, &nr_ranges,
229 get_nr_ram_ranges_callback);
230
231 ced->max_nr_ranges = nr_ranges;
232
233 /* Exclusion of crash region could split memory ranges */
234 ced->max_nr_ranges++;
235
236 /* If crashk_low_res is not 0, another range split possible */
237 if (crashk_low_res.end)
238 ced->max_nr_ranges++;
239}
240
241static int exclude_mem_range(struct crash_mem *mem,
242 unsigned long long mstart, unsigned long long mend)
243{
244 int i, j;
245 unsigned long long start, end;
246 struct crash_mem_range temp_range = {0, 0};
247
248 for (i = 0; i < mem->nr_ranges; i++) {
249 start = mem->ranges[i].start;
250 end = mem->ranges[i].end;
251
252 if (mstart > end || mend < start)
253 continue;
254
255 /* Truncate any area outside of range */
256 if (mstart < start)
257 mstart = start;
258 if (mend > end)
259 mend = end;
260
261 /* Found completely overlapping range */
262 if (mstart == start && mend == end) {
263 mem->ranges[i].start = 0;
264 mem->ranges[i].end = 0;
265 if (i < mem->nr_ranges - 1) {
266 /* Shift rest of the ranges to left */
267 for (j = i; j < mem->nr_ranges - 1; j++) {
268 mem->ranges[j].start =
269 mem->ranges[j+1].start;
270 mem->ranges[j].end =
271 mem->ranges[j+1].end;
272 }
273 }
274 mem->nr_ranges--;
275 return 0;
276 }
277
278 if (mstart > start && mend < end) {
279 /* Split original range */
280 mem->ranges[i].end = mstart - 1;
281 temp_range.start = mend + 1;
282 temp_range.end = end;
283 } else if (mstart != start)
284 mem->ranges[i].end = mstart - 1;
285 else
286 mem->ranges[i].start = mend + 1;
287 break;
288 }
289
290 /* If a split happend, add the split to array */
291 if (!temp_range.end)
292 return 0;
293
294 /* Split happened */
295 if (i == CRASH_MAX_RANGES - 1) {
296 pr_err("Too many crash ranges after split\n");
297 return -ENOMEM;
298 }
299
300 /* Location where new range should go */
301 j = i + 1;
302 if (j < mem->nr_ranges) {
303 /* Move over all ranges one slot towards the end */
304 for (i = mem->nr_ranges - 1; i >= j; i--)
305 mem->ranges[i + 1] = mem->ranges[i];
306 }
307
308 mem->ranges[j].start = temp_range.start;
309 mem->ranges[j].end = temp_range.end;
310 mem->nr_ranges++;
311 return 0;
312}
313
314/*
315 * Look for any unwanted ranges between mstart, mend and remove them. This
316 * might lead to split and split ranges are put in ced->mem.ranges[] array
317 */
318static int elf_header_exclude_ranges(struct crash_elf_data *ced,
319 unsigned long long mstart, unsigned long long mend)
320{
321 struct crash_mem *cmem = &ced->mem;
322 int ret = 0;
323
324 memset(cmem->ranges, 0, sizeof(cmem->ranges));
325
326 cmem->ranges[0].start = mstart;
327 cmem->ranges[0].end = mend;
328 cmem->nr_ranges = 1;
329
330 /* Exclude crashkernel region */
331 ret = exclude_mem_range(cmem, crashk_res.start, crashk_res.end);
332 if (ret)
333 return ret;
334
335 if (crashk_low_res.end) {
336 ret = exclude_mem_range(cmem, crashk_low_res.start, crashk_low_res.end);
337 if (ret)
338 return ret;
339 }
340
341 return ret;
342}
343
344static int prepare_elf64_ram_headers_callback(u64 start, u64 end, void *arg)
345{
346 struct crash_elf_data *ced = arg;
347 Elf64_Ehdr *ehdr;
348 Elf64_Phdr *phdr;
349 unsigned long mstart, mend;
350 struct kimage *image = ced->image;
351 struct crash_mem *cmem;
352 int ret, i;
353
354 ehdr = ced->ehdr;
355
356 /* Exclude unwanted mem ranges */
357 ret = elf_header_exclude_ranges(ced, start, end);
358 if (ret)
359 return ret;
360
361 /* Go through all the ranges in ced->mem.ranges[] and prepare phdr */
362 cmem = &ced->mem;
363
364 for (i = 0; i < cmem->nr_ranges; i++) {
365 mstart = cmem->ranges[i].start;
366 mend = cmem->ranges[i].end;
367
368 phdr = ced->bufp;
369 ced->bufp += sizeof(Elf64_Phdr);
370
371 phdr->p_type = PT_LOAD;
372 phdr->p_flags = PF_R|PF_W|PF_X;
373 phdr->p_offset = mstart;
374
375 /*
376 * If a range matches backup region, adjust offset to backup
377 * segment.
378 */
379 if (mstart == image->arch.backup_src_start &&
380 (mend - mstart + 1) == image->arch.backup_src_sz)
381 phdr->p_offset = image->arch.backup_load_addr;
382
383 phdr->p_paddr = mstart;
384 phdr->p_vaddr = (unsigned long long) __va(mstart);
385 phdr->p_filesz = phdr->p_memsz = mend - mstart + 1;
386 phdr->p_align = 0;
387 ehdr->e_phnum++;
388 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",
389 phdr, phdr->p_vaddr, phdr->p_paddr, phdr->p_filesz,
390 ehdr->e_phnum, phdr->p_offset);
391 }
392
393 return ret;
394}
395
396static int prepare_elf64_headers(struct crash_elf_data *ced,
397 void **addr, unsigned long *sz)
398{
399 Elf64_Ehdr *ehdr;
400 Elf64_Phdr *phdr;
401 unsigned long nr_cpus = num_possible_cpus(), nr_phdr, elf_sz;
402 unsigned char *buf, *bufp;
403 unsigned int cpu;
404 unsigned long long notes_addr;
405 int ret;
406
407 /* extra phdr for vmcoreinfo elf note */
408 nr_phdr = nr_cpus + 1;
409 nr_phdr += ced->max_nr_ranges;
410
411 /*
412 * kexec-tools creates an extra PT_LOAD phdr for kernel text mapping
413 * area on x86_64 (ffffffff80000000 - ffffffffa0000000).
414 * I think this is required by tools like gdb. So same physical
415 * memory will be mapped in two elf headers. One will contain kernel
416 * text virtual addresses and other will have __va(physical) addresses.
417 */
418
419 nr_phdr++;
420 elf_sz = sizeof(Elf64_Ehdr) + nr_phdr * sizeof(Elf64_Phdr);
421 elf_sz = ALIGN(elf_sz, ELF_CORE_HEADER_ALIGN);
422
423 buf = vzalloc(elf_sz);
424 if (!buf)
425 return -ENOMEM;
426
427 bufp = buf;
428 ehdr = (Elf64_Ehdr *)bufp;
429 bufp += sizeof(Elf64_Ehdr);
430 memcpy(ehdr->e_ident, ELFMAG, SELFMAG);
431 ehdr->e_ident[EI_CLASS] = ELFCLASS64;
432 ehdr->e_ident[EI_DATA] = ELFDATA2LSB;
433 ehdr->e_ident[EI_VERSION] = EV_CURRENT;
434 ehdr->e_ident[EI_OSABI] = ELF_OSABI;
435 memset(ehdr->e_ident + EI_PAD, 0, EI_NIDENT - EI_PAD);
436 ehdr->e_type = ET_CORE;
437 ehdr->e_machine = ELF_ARCH;
438 ehdr->e_version = EV_CURRENT;
439 ehdr->e_phoff = sizeof(Elf64_Ehdr);
440 ehdr->e_ehsize = sizeof(Elf64_Ehdr);
441 ehdr->e_phentsize = sizeof(Elf64_Phdr);
442
443 /* Prepare one phdr of type PT_NOTE for each present cpu */
444 for_each_present_cpu(cpu) {
445 phdr = (Elf64_Phdr *)bufp;
446 bufp += sizeof(Elf64_Phdr);
447 phdr->p_type = PT_NOTE;
448 notes_addr = per_cpu_ptr_to_phys(per_cpu_ptr(crash_notes, cpu));
449 phdr->p_offset = phdr->p_paddr = notes_addr;
450 phdr->p_filesz = phdr->p_memsz = sizeof(note_buf_t);
451 (ehdr->e_phnum)++;
452 }
453
454 /* Prepare one PT_NOTE header for vmcoreinfo */
455 phdr = (Elf64_Phdr *)bufp;
456 bufp += sizeof(Elf64_Phdr);
457 phdr->p_type = PT_NOTE;
458 phdr->p_offset = phdr->p_paddr = paddr_vmcoreinfo_note();
459 phdr->p_filesz = phdr->p_memsz = sizeof(vmcoreinfo_note);
460 (ehdr->e_phnum)++;
461
462#ifdef CONFIG_X86_64
463 /* Prepare PT_LOAD type program header for kernel text region */
464 phdr = (Elf64_Phdr *)bufp;
465 bufp += sizeof(Elf64_Phdr);
466 phdr->p_type = PT_LOAD;
467 phdr->p_flags = PF_R|PF_W|PF_X;
468 phdr->p_vaddr = (Elf64_Addr)_text;
469 phdr->p_filesz = phdr->p_memsz = _end - _text;
470 phdr->p_offset = phdr->p_paddr = __pa_symbol(_text);
471 (ehdr->e_phnum)++;
472#endif
473
474 /* Prepare PT_LOAD headers for system ram chunks. */
475 ced->ehdr = ehdr;
476 ced->bufp = bufp;
477 ret = walk_system_ram_res(0, -1, ced,
478 prepare_elf64_ram_headers_callback);
479 if (ret < 0)
480 return ret;
481
482 *addr = buf;
483 *sz = elf_sz;
484 return 0;
485}
486
487/* Prepare elf headers. Return addr and size */
488static int prepare_elf_headers(struct kimage *image, void **addr,
489 unsigned long *sz)
490{
491 struct crash_elf_data *ced;
492 int ret;
493
494 ced = kzalloc(sizeof(*ced), GFP_KERNEL);
495 if (!ced)
496 return -ENOMEM;
497
498 fill_up_crash_elf_data(ced, image);
499
500 /* By default prepare 64bit headers */
501 ret = prepare_elf64_headers(ced, addr, sz);
502 kfree(ced);
503 return ret;
504}
505
506static int add_e820_entry(struct boot_params *params, struct e820entry *entry)
507{
508 unsigned int nr_e820_entries;
509
510 nr_e820_entries = params->e820_entries;
511 if (nr_e820_entries >= E820MAX)
512 return 1;
513
514 memcpy(¶ms->e820_map[nr_e820_entries], entry,
515 sizeof(struct e820entry));
516 params->e820_entries++;
517 return 0;
518}
519
520static int memmap_entry_callback(u64 start, u64 end, void *arg)
521{
522 struct crash_memmap_data *cmd = arg;
523 struct boot_params *params = cmd->params;
524 struct e820entry ei;
525
526 ei.addr = start;
527 ei.size = end - start + 1;
528 ei.type = cmd->type;
529 add_e820_entry(params, &ei);
530
531 return 0;
532}
533
534static int memmap_exclude_ranges(struct kimage *image, struct crash_mem *cmem,
535 unsigned long long mstart,
536 unsigned long long mend)
537{
538 unsigned long start, end;
539 int ret = 0;
540
541 cmem->ranges[0].start = mstart;
542 cmem->ranges[0].end = mend;
543 cmem->nr_ranges = 1;
544
545 /* Exclude Backup region */
546 start = image->arch.backup_load_addr;
547 end = start + image->arch.backup_src_sz - 1;
548 ret = exclude_mem_range(cmem, start, end);
549 if (ret)
550 return ret;
551
552 /* Exclude elf header region */
553 start = image->arch.elf_load_addr;
554 end = start + image->arch.elf_headers_sz - 1;
555 return exclude_mem_range(cmem, start, end);
556}
557
558/* Prepare memory map for crash dump kernel */
559int crash_setup_memmap_entries(struct kimage *image, struct boot_params *params)
560{
561 int i, ret = 0;
562 unsigned long flags;
563 struct e820entry ei;
564 struct crash_memmap_data cmd;
565 struct crash_mem *cmem;
566
567 cmem = vzalloc(sizeof(struct crash_mem));
568 if (!cmem)
569 return -ENOMEM;
570
571 memset(&cmd, 0, sizeof(struct crash_memmap_data));
572 cmd.params = params;
573
574 /* Add first 640K segment */
575 ei.addr = image->arch.backup_src_start;
576 ei.size = image->arch.backup_src_sz;
577 ei.type = E820_RAM;
578 add_e820_entry(params, &ei);
579
580 /* Add ACPI tables */
581 cmd.type = E820_ACPI;
582 flags = IORESOURCE_MEM | IORESOURCE_BUSY;
583 walk_iomem_res_desc(IORES_DESC_ACPI_TABLES, flags, 0, -1, &cmd,
584 memmap_entry_callback);
585
586 /* Add ACPI Non-volatile Storage */
587 cmd.type = E820_NVS;
588 walk_iomem_res_desc(IORES_DESC_ACPI_NV_STORAGE, flags, 0, -1, &cmd,
589 memmap_entry_callback);
590
591 /* Add crashk_low_res region */
592 if (crashk_low_res.end) {
593 ei.addr = crashk_low_res.start;
594 ei.size = crashk_low_res.end - crashk_low_res.start + 1;
595 ei.type = E820_RAM;
596 add_e820_entry(params, &ei);
597 }
598
599 /* Exclude some ranges from crashk_res and add rest to memmap */
600 ret = memmap_exclude_ranges(image, cmem, crashk_res.start,
601 crashk_res.end);
602 if (ret)
603 goto out;
604
605 for (i = 0; i < cmem->nr_ranges; i++) {
606 ei.size = cmem->ranges[i].end - cmem->ranges[i].start + 1;
607
608 /* If entry is less than a page, skip it */
609 if (ei.size < PAGE_SIZE)
610 continue;
611 ei.addr = cmem->ranges[i].start;
612 ei.type = E820_RAM;
613 add_e820_entry(params, &ei);
614 }
615
616out:
617 vfree(cmem);
618 return ret;
619}
620
621static int determine_backup_region(u64 start, u64 end, void *arg)
622{
623 struct kimage *image = arg;
624
625 image->arch.backup_src_start = start;
626 image->arch.backup_src_sz = end - start + 1;
627
628 /* Expecting only one range for backup region */
629 return 1;
630}
631
632int crash_load_segments(struct kimage *image)
633{
634 int ret;
635 struct kexec_buf kbuf = { .image = image, .buf_min = 0,
636 .buf_max = ULONG_MAX, .top_down = false };
637
638 /*
639 * Determine and load a segment for backup area. First 640K RAM
640 * region is backup source
641 */
642
643 ret = walk_system_ram_res(KEXEC_BACKUP_SRC_START, KEXEC_BACKUP_SRC_END,
644 image, determine_backup_region);
645
646 /* Zero or postive return values are ok */
647 if (ret < 0)
648 return ret;
649
650 /* Add backup segment. */
651 if (image->arch.backup_src_sz) {
652 kbuf.buffer = &crash_zero_bytes;
653 kbuf.bufsz = sizeof(crash_zero_bytes);
654 kbuf.memsz = image->arch.backup_src_sz;
655 kbuf.buf_align = PAGE_SIZE;
656 /*
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.
660 */
661 ret = kexec_add_buffer(&kbuf);
662 if (ret)
663 return ret;
664 image->arch.backup_load_addr = kbuf.mem;
665 pr_debug("Loaded backup region at 0x%lx backup_start=0x%lx memsz=0x%lx\n",
666 image->arch.backup_load_addr,
667 image->arch.backup_src_start, kbuf.memsz);
668 }
669
670 /* Prepare elf headers and add a segment */
671 ret = prepare_elf_headers(image, &kbuf.buffer, &kbuf.bufsz);
672 if (ret)
673 return ret;
674
675 image->arch.elf_headers = kbuf.buffer;
676 image->arch.elf_headers_sz = kbuf.bufsz;
677
678 kbuf.memsz = kbuf.bufsz;
679 kbuf.buf_align = ELF_CORE_HEADER_ALIGN;
680 ret = kexec_add_buffer(&kbuf);
681 if (ret) {
682 vfree((void *)image->arch.elf_headers);
683 return ret;
684 }
685 image->arch.elf_load_addr = kbuf.mem;
686 pr_debug("Loaded ELF headers at 0x%lx bufsz=0x%lx memsz=0x%lx\n",
687 image->arch.elf_load_addr, kbuf.bufsz, kbuf.bufsz);
688
689 return ret;
690}
691#endif /* CONFIG_KEXEC_FILE */
1/*
2 * Architecture specific (i386/x86_64) functions for kexec based crash dumps.
3 *
4 * Created by: Hariprasad Nellitheertha (hari@in.ibm.com)
5 *
6 * Copyright (C) IBM Corporation, 2004. All rights reserved.
7 * Copyright (C) Red Hat Inc., 2014. All rights reserved.
8 * Authors:
9 * Vivek Goyal <vgoyal@redhat.com>
10 *
11 */
12
13#define pr_fmt(fmt) "kexec: " fmt
14
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>
26
27#include <asm/processor.h>
28#include <asm/hardirq.h>
29#include <asm/nmi.h>
30#include <asm/hw_irq.h>
31#include <asm/apic.h>
32#include <asm/io_apic.h>
33#include <asm/hpet.h>
34#include <linux/kdebug.h>
35#include <asm/cpu.h>
36#include <asm/reboot.h>
37#include <asm/virtext.h>
38#include <asm/intel_pt.h>
39
40/* Alignment required for elf header segment */
41#define ELF_CORE_HEADER_ALIGN 4096
42
43/* This primarily represents number of split ranges due to exclusion */
44#define CRASH_MAX_RANGES 16
45
46struct crash_mem_range {
47 u64 start, end;
48};
49
50struct crash_mem {
51 unsigned int nr_ranges;
52 struct crash_mem_range ranges[CRASH_MAX_RANGES];
53};
54
55/* Misc data about ram ranges needed to prepare elf headers */
56struct crash_elf_data {
57 struct kimage *image;
58 /*
59 * Total number of ram ranges we have after various adjustments for
60 * crash reserved region, etc.
61 */
62 unsigned int max_nr_ranges;
63
64 /* Pointer to elf header */
65 void *ehdr;
66 /* Pointer to next phdr */
67 void *bufp;
68 struct crash_mem mem;
69};
70
71/* Used while preparing memory map entries for second kernel */
72struct crash_memmap_data {
73 struct boot_params *params;
74 /* Type of memory */
75 unsigned int type;
76};
77
78/*
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.
82 *
83 * protected by rcu.
84 */
85crash_vmclear_fn __rcu *crash_vmclear_loaded_vmcss = NULL;
86EXPORT_SYMBOL_GPL(crash_vmclear_loaded_vmcss);
87unsigned long crash_zero_bytes;
88
89static inline void cpu_crash_vmclear_loaded_vmcss(void)
90{
91 crash_vmclear_fn *do_vmclear_operation = NULL;
92
93 rcu_read_lock();
94 do_vmclear_operation = rcu_dereference(crash_vmclear_loaded_vmcss);
95 if (do_vmclear_operation)
96 do_vmclear_operation();
97 rcu_read_unlock();
98}
99
100#if defined(CONFIG_SMP) && defined(CONFIG_X86_LOCAL_APIC)
101
102static void kdump_nmi_callback(int cpu, struct pt_regs *regs)
103{
104#ifdef CONFIG_X86_32
105 struct pt_regs fixed_regs;
106
107 if (!user_mode(regs)) {
108 crash_fixup_ss_esp(&fixed_regs, regs);
109 regs = &fixed_regs;
110 }
111#endif
112 crash_save_cpu(regs, cpu);
113
114 /*
115 * VMCLEAR VMCSs loaded on all cpus if needed.
116 */
117 cpu_crash_vmclear_loaded_vmcss();
118
119 /* Disable VMX or SVM if needed.
120 *
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.
124 */
125 cpu_emergency_vmxoff();
126 cpu_emergency_svm_disable();
127
128 /*
129 * Disable Intel PT to stop its logging
130 */
131 cpu_emergency_stop_pt();
132
133 disable_local_APIC();
134}
135
136static void kdump_nmi_shootdown_cpus(void)
137{
138 nmi_shootdown_cpus(kdump_nmi_callback);
139
140 disable_local_APIC();
141}
142
143#else
144static void kdump_nmi_shootdown_cpus(void)
145{
146 /* There are no cpus to shootdown */
147}
148#endif
149
150void native_machine_crash_shutdown(struct pt_regs *regs)
151{
152 /* This function is only called after the system
153 * has panicked or is otherwise in a critical state.
154 * The minimum amount of code to allow a kexec'd kernel
155 * to run successfully needs to happen here.
156 *
157 * In practice this means shooting down the other cpus in
158 * an SMP system.
159 */
160 /* The kernel is broken so disable interrupts */
161 local_irq_disable();
162
163 kdump_nmi_shootdown_cpus();
164
165 /*
166 * VMCLEAR VMCSs loaded on this cpu if needed.
167 */
168 cpu_crash_vmclear_loaded_vmcss();
169
170 /* Booting kdump kernel with VMX or SVM enabled won't work,
171 * because (among other limitations) we can't disable paging
172 * with the virt flags.
173 */
174 cpu_emergency_vmxoff();
175 cpu_emergency_svm_disable();
176
177 /*
178 * Disable Intel PT to stop its logging
179 */
180 cpu_emergency_stop_pt();
181
182#ifdef CONFIG_X86_IO_APIC
183 /* Prevent crash_kexec() from deadlocking on ioapic_lock. */
184 ioapic_zap_locks();
185 disable_IO_APIC();
186#endif
187 lapic_shutdown();
188#ifdef CONFIG_HPET_TIMER
189 hpet_disable();
190#endif
191 crash_save_cpu(regs, safe_smp_processor_id());
192}
193
194#ifdef CONFIG_KEXEC_FILE
195static int get_nr_ram_ranges_callback(u64 start, u64 end, void *arg)
196{
197 unsigned int *nr_ranges = arg;
198
199 (*nr_ranges)++;
200 return 0;
201}
202
203
204/* Gather all the required information to prepare elf headers for ram regions */
205static void fill_up_crash_elf_data(struct crash_elf_data *ced,
206 struct kimage *image)
207{
208 unsigned int nr_ranges = 0;
209
210 ced->image = image;
211
212 walk_system_ram_res(0, -1, &nr_ranges,
213 get_nr_ram_ranges_callback);
214
215 ced->max_nr_ranges = nr_ranges;
216
217 /* Exclusion of crash region could split memory ranges */
218 ced->max_nr_ranges++;
219
220 /* If crashk_low_res is not 0, another range split possible */
221 if (crashk_low_res.end)
222 ced->max_nr_ranges++;
223}
224
225static int exclude_mem_range(struct crash_mem *mem,
226 unsigned long long mstart, unsigned long long mend)
227{
228 int i, j;
229 unsigned long long start, end;
230 struct crash_mem_range temp_range = {0, 0};
231
232 for (i = 0; i < mem->nr_ranges; i++) {
233 start = mem->ranges[i].start;
234 end = mem->ranges[i].end;
235
236 if (mstart > end || mend < start)
237 continue;
238
239 /* Truncate any area outside of range */
240 if (mstart < start)
241 mstart = start;
242 if (mend > end)
243 mend = end;
244
245 /* Found completely overlapping range */
246 if (mstart == start && mend == end) {
247 mem->ranges[i].start = 0;
248 mem->ranges[i].end = 0;
249 if (i < mem->nr_ranges - 1) {
250 /* Shift rest of the ranges to left */
251 for (j = i; j < mem->nr_ranges - 1; j++) {
252 mem->ranges[j].start =
253 mem->ranges[j+1].start;
254 mem->ranges[j].end =
255 mem->ranges[j+1].end;
256 }
257 }
258 mem->nr_ranges--;
259 return 0;
260 }
261
262 if (mstart > start && mend < end) {
263 /* Split original range */
264 mem->ranges[i].end = mstart - 1;
265 temp_range.start = mend + 1;
266 temp_range.end = end;
267 } else if (mstart != start)
268 mem->ranges[i].end = mstart - 1;
269 else
270 mem->ranges[i].start = mend + 1;
271 break;
272 }
273
274 /* If a split happend, add the split to array */
275 if (!temp_range.end)
276 return 0;
277
278 /* Split happened */
279 if (i == CRASH_MAX_RANGES - 1) {
280 pr_err("Too many crash ranges after split\n");
281 return -ENOMEM;
282 }
283
284 /* Location where new range should go */
285 j = i + 1;
286 if (j < mem->nr_ranges) {
287 /* Move over all ranges one slot towards the end */
288 for (i = mem->nr_ranges - 1; i >= j; i--)
289 mem->ranges[i + 1] = mem->ranges[i];
290 }
291
292 mem->ranges[j].start = temp_range.start;
293 mem->ranges[j].end = temp_range.end;
294 mem->nr_ranges++;
295 return 0;
296}
297
298/*
299 * Look for any unwanted ranges between mstart, mend and remove them. This
300 * might lead to split and split ranges are put in ced->mem.ranges[] array
301 */
302static int elf_header_exclude_ranges(struct crash_elf_data *ced,
303 unsigned long long mstart, unsigned long long mend)
304{
305 struct crash_mem *cmem = &ced->mem;
306 int ret = 0;
307
308 memset(cmem->ranges, 0, sizeof(cmem->ranges));
309
310 cmem->ranges[0].start = mstart;
311 cmem->ranges[0].end = mend;
312 cmem->nr_ranges = 1;
313
314 /* Exclude crashkernel region */
315 ret = exclude_mem_range(cmem, crashk_res.start, crashk_res.end);
316 if (ret)
317 return ret;
318
319 if (crashk_low_res.end) {
320 ret = exclude_mem_range(cmem, crashk_low_res.start, crashk_low_res.end);
321 if (ret)
322 return ret;
323 }
324
325 return ret;
326}
327
328static int prepare_elf64_ram_headers_callback(u64 start, u64 end, void *arg)
329{
330 struct crash_elf_data *ced = arg;
331 Elf64_Ehdr *ehdr;
332 Elf64_Phdr *phdr;
333 unsigned long mstart, mend;
334 struct kimage *image = ced->image;
335 struct crash_mem *cmem;
336 int ret, i;
337
338 ehdr = ced->ehdr;
339
340 /* Exclude unwanted mem ranges */
341 ret = elf_header_exclude_ranges(ced, start, end);
342 if (ret)
343 return ret;
344
345 /* Go through all the ranges in ced->mem.ranges[] and prepare phdr */
346 cmem = &ced->mem;
347
348 for (i = 0; i < cmem->nr_ranges; i++) {
349 mstart = cmem->ranges[i].start;
350 mend = cmem->ranges[i].end;
351
352 phdr = ced->bufp;
353 ced->bufp += sizeof(Elf64_Phdr);
354
355 phdr->p_type = PT_LOAD;
356 phdr->p_flags = PF_R|PF_W|PF_X;
357 phdr->p_offset = mstart;
358
359 /*
360 * If a range matches backup region, adjust offset to backup
361 * segment.
362 */
363 if (mstart == image->arch.backup_src_start &&
364 (mend - mstart + 1) == image->arch.backup_src_sz)
365 phdr->p_offset = image->arch.backup_load_addr;
366
367 phdr->p_paddr = mstart;
368 phdr->p_vaddr = (unsigned long long) __va(mstart);
369 phdr->p_filesz = phdr->p_memsz = mend - mstart + 1;
370 phdr->p_align = 0;
371 ehdr->e_phnum++;
372 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",
373 phdr, phdr->p_vaddr, phdr->p_paddr, phdr->p_filesz,
374 ehdr->e_phnum, phdr->p_offset);
375 }
376
377 return ret;
378}
379
380static int prepare_elf64_headers(struct crash_elf_data *ced,
381 void **addr, unsigned long *sz)
382{
383 Elf64_Ehdr *ehdr;
384 Elf64_Phdr *phdr;
385 unsigned long nr_cpus = num_possible_cpus(), nr_phdr, elf_sz;
386 unsigned char *buf, *bufp;
387 unsigned int cpu;
388 unsigned long long notes_addr;
389 int ret;
390
391 /* extra phdr for vmcoreinfo elf note */
392 nr_phdr = nr_cpus + 1;
393 nr_phdr += ced->max_nr_ranges;
394
395 /*
396 * kexec-tools creates an extra PT_LOAD phdr for kernel text mapping
397 * area on x86_64 (ffffffff80000000 - ffffffffa0000000).
398 * I think this is required by tools like gdb. So same physical
399 * memory will be mapped in two elf headers. One will contain kernel
400 * text virtual addresses and other will have __va(physical) addresses.
401 */
402
403 nr_phdr++;
404 elf_sz = sizeof(Elf64_Ehdr) + nr_phdr * sizeof(Elf64_Phdr);
405 elf_sz = ALIGN(elf_sz, ELF_CORE_HEADER_ALIGN);
406
407 buf = vzalloc(elf_sz);
408 if (!buf)
409 return -ENOMEM;
410
411 bufp = buf;
412 ehdr = (Elf64_Ehdr *)bufp;
413 bufp += sizeof(Elf64_Ehdr);
414 memcpy(ehdr->e_ident, ELFMAG, SELFMAG);
415 ehdr->e_ident[EI_CLASS] = ELFCLASS64;
416 ehdr->e_ident[EI_DATA] = ELFDATA2LSB;
417 ehdr->e_ident[EI_VERSION] = EV_CURRENT;
418 ehdr->e_ident[EI_OSABI] = ELF_OSABI;
419 memset(ehdr->e_ident + EI_PAD, 0, EI_NIDENT - EI_PAD);
420 ehdr->e_type = ET_CORE;
421 ehdr->e_machine = ELF_ARCH;
422 ehdr->e_version = EV_CURRENT;
423 ehdr->e_phoff = sizeof(Elf64_Ehdr);
424 ehdr->e_ehsize = sizeof(Elf64_Ehdr);
425 ehdr->e_phentsize = sizeof(Elf64_Phdr);
426
427 /* Prepare one phdr of type PT_NOTE for each present cpu */
428 for_each_present_cpu(cpu) {
429 phdr = (Elf64_Phdr *)bufp;
430 bufp += sizeof(Elf64_Phdr);
431 phdr->p_type = PT_NOTE;
432 notes_addr = per_cpu_ptr_to_phys(per_cpu_ptr(crash_notes, cpu));
433 phdr->p_offset = phdr->p_paddr = notes_addr;
434 phdr->p_filesz = phdr->p_memsz = sizeof(note_buf_t);
435 (ehdr->e_phnum)++;
436 }
437
438 /* Prepare one PT_NOTE header for vmcoreinfo */
439 phdr = (Elf64_Phdr *)bufp;
440 bufp += sizeof(Elf64_Phdr);
441 phdr->p_type = PT_NOTE;
442 phdr->p_offset = phdr->p_paddr = paddr_vmcoreinfo_note();
443 phdr->p_filesz = phdr->p_memsz = sizeof(vmcoreinfo_note);
444 (ehdr->e_phnum)++;
445
446#ifdef CONFIG_X86_64
447 /* Prepare PT_LOAD type program header for kernel text region */
448 phdr = (Elf64_Phdr *)bufp;
449 bufp += sizeof(Elf64_Phdr);
450 phdr->p_type = PT_LOAD;
451 phdr->p_flags = PF_R|PF_W|PF_X;
452 phdr->p_vaddr = (Elf64_Addr)_text;
453 phdr->p_filesz = phdr->p_memsz = _end - _text;
454 phdr->p_offset = phdr->p_paddr = __pa_symbol(_text);
455 (ehdr->e_phnum)++;
456#endif
457
458 /* Prepare PT_LOAD headers for system ram chunks. */
459 ced->ehdr = ehdr;
460 ced->bufp = bufp;
461 ret = walk_system_ram_res(0, -1, ced,
462 prepare_elf64_ram_headers_callback);
463 if (ret < 0)
464 return ret;
465
466 *addr = buf;
467 *sz = elf_sz;
468 return 0;
469}
470
471/* Prepare elf headers. Return addr and size */
472static int prepare_elf_headers(struct kimage *image, void **addr,
473 unsigned long *sz)
474{
475 struct crash_elf_data *ced;
476 int ret;
477
478 ced = kzalloc(sizeof(*ced), GFP_KERNEL);
479 if (!ced)
480 return -ENOMEM;
481
482 fill_up_crash_elf_data(ced, image);
483
484 /* By default prepare 64bit headers */
485 ret = prepare_elf64_headers(ced, addr, sz);
486 kfree(ced);
487 return ret;
488}
489
490static int add_e820_entry(struct boot_params *params, struct e820entry *entry)
491{
492 unsigned int nr_e820_entries;
493
494 nr_e820_entries = params->e820_entries;
495 if (nr_e820_entries >= E820MAX)
496 return 1;
497
498 memcpy(¶ms->e820_map[nr_e820_entries], entry,
499 sizeof(struct e820entry));
500 params->e820_entries++;
501 return 0;
502}
503
504static int memmap_entry_callback(u64 start, u64 end, void *arg)
505{
506 struct crash_memmap_data *cmd = arg;
507 struct boot_params *params = cmd->params;
508 struct e820entry ei;
509
510 ei.addr = start;
511 ei.size = end - start + 1;
512 ei.type = cmd->type;
513 add_e820_entry(params, &ei);
514
515 return 0;
516}
517
518static int memmap_exclude_ranges(struct kimage *image, struct crash_mem *cmem,
519 unsigned long long mstart,
520 unsigned long long mend)
521{
522 unsigned long start, end;
523 int ret = 0;
524
525 cmem->ranges[0].start = mstart;
526 cmem->ranges[0].end = mend;
527 cmem->nr_ranges = 1;
528
529 /* Exclude Backup region */
530 start = image->arch.backup_load_addr;
531 end = start + image->arch.backup_src_sz - 1;
532 ret = exclude_mem_range(cmem, start, end);
533 if (ret)
534 return ret;
535
536 /* Exclude elf header region */
537 start = image->arch.elf_load_addr;
538 end = start + image->arch.elf_headers_sz - 1;
539 return exclude_mem_range(cmem, start, end);
540}
541
542/* Prepare memory map for crash dump kernel */
543int crash_setup_memmap_entries(struct kimage *image, struct boot_params *params)
544{
545 int i, ret = 0;
546 unsigned long flags;
547 struct e820entry ei;
548 struct crash_memmap_data cmd;
549 struct crash_mem *cmem;
550
551 cmem = vzalloc(sizeof(struct crash_mem));
552 if (!cmem)
553 return -ENOMEM;
554
555 memset(&cmd, 0, sizeof(struct crash_memmap_data));
556 cmd.params = params;
557
558 /* Add first 640K segment */
559 ei.addr = image->arch.backup_src_start;
560 ei.size = image->arch.backup_src_sz;
561 ei.type = E820_RAM;
562 add_e820_entry(params, &ei);
563
564 /* Add ACPI tables */
565 cmd.type = E820_ACPI;
566 flags = IORESOURCE_MEM | IORESOURCE_BUSY;
567 walk_iomem_res_desc(IORES_DESC_ACPI_TABLES, flags, 0, -1, &cmd,
568 memmap_entry_callback);
569
570 /* Add ACPI Non-volatile Storage */
571 cmd.type = E820_NVS;
572 walk_iomem_res_desc(IORES_DESC_ACPI_NV_STORAGE, flags, 0, -1, &cmd,
573 memmap_entry_callback);
574
575 /* Add crashk_low_res region */
576 if (crashk_low_res.end) {
577 ei.addr = crashk_low_res.start;
578 ei.size = crashk_low_res.end - crashk_low_res.start + 1;
579 ei.type = E820_RAM;
580 add_e820_entry(params, &ei);
581 }
582
583 /* Exclude some ranges from crashk_res and add rest to memmap */
584 ret = memmap_exclude_ranges(image, cmem, crashk_res.start,
585 crashk_res.end);
586 if (ret)
587 goto out;
588
589 for (i = 0; i < cmem->nr_ranges; i++) {
590 ei.size = cmem->ranges[i].end - cmem->ranges[i].start + 1;
591
592 /* If entry is less than a page, skip it */
593 if (ei.size < PAGE_SIZE)
594 continue;
595 ei.addr = cmem->ranges[i].start;
596 ei.type = E820_RAM;
597 add_e820_entry(params, &ei);
598 }
599
600out:
601 vfree(cmem);
602 return ret;
603}
604
605static int determine_backup_region(u64 start, u64 end, void *arg)
606{
607 struct kimage *image = arg;
608
609 image->arch.backup_src_start = start;
610 image->arch.backup_src_sz = end - start + 1;
611
612 /* Expecting only one range for backup region */
613 return 1;
614}
615
616int crash_load_segments(struct kimage *image)
617{
618 unsigned long src_start, src_sz, elf_sz;
619 void *elf_addr;
620 int ret;
621
622 /*
623 * Determine and load a segment for backup area. First 640K RAM
624 * region is backup source
625 */
626
627 ret = walk_system_ram_res(KEXEC_BACKUP_SRC_START, KEXEC_BACKUP_SRC_END,
628 image, determine_backup_region);
629
630 /* Zero or postive return values are ok */
631 if (ret < 0)
632 return ret;
633
634 src_start = image->arch.backup_src_start;
635 src_sz = image->arch.backup_src_sz;
636
637 /* Add backup segment. */
638 if (src_sz) {
639 /*
640 * Ideally there is no source for backup segment. This is
641 * copied in purgatory after crash. Just add a zero filled
642 * segment for now to make sure checksum logic works fine.
643 */
644 ret = kexec_add_buffer(image, (char *)&crash_zero_bytes,
645 sizeof(crash_zero_bytes), src_sz,
646 PAGE_SIZE, 0, -1, 0,
647 &image->arch.backup_load_addr);
648 if (ret)
649 return ret;
650 pr_debug("Loaded backup region at 0x%lx backup_start=0x%lx memsz=0x%lx\n",
651 image->arch.backup_load_addr, src_start, src_sz);
652 }
653
654 /* Prepare elf headers and add a segment */
655 ret = prepare_elf_headers(image, &elf_addr, &elf_sz);
656 if (ret)
657 return ret;
658
659 image->arch.elf_headers = elf_addr;
660 image->arch.elf_headers_sz = elf_sz;
661
662 ret = kexec_add_buffer(image, (char *)elf_addr, elf_sz, elf_sz,
663 ELF_CORE_HEADER_ALIGN, 0, -1, 0,
664 &image->arch.elf_load_addr);
665 if (ret) {
666 vfree((void *)image->arch.elf_headers);
667 return ret;
668 }
669 pr_debug("Loaded ELF headers at 0x%lx bufsz=0x%lx memsz=0x%lx\n",
670 image->arch.elf_load_addr, elf_sz, elf_sz);
671
672 return ret;
673}
674#endif /* CONFIG_KEXEC_FILE */