Merge branch 'akpm' (second patchbomb from Andrew Morton)
[deliverable/linux.git] / arch / x86 / kernel / crash.c
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
26 #include <asm/processor.h>
27 #include <asm/hardirq.h>
28 #include <asm/nmi.h>
29 #include <asm/hw_irq.h>
30 #include <asm/apic.h>
31 #include <asm/hpet.h>
32 #include <linux/kdebug.h>
33 #include <asm/cpu.h>
34 #include <asm/reboot.h>
35 #include <asm/virtext.h>
36
37 /* Alignment required for elf header segment */
38 #define ELF_CORE_HEADER_ALIGN 4096
39
40 /* This primarily represents number of split ranges due to exclusion */
41 #define CRASH_MAX_RANGES 16
42
43 struct crash_mem_range {
44 u64 start, end;
45 };
46
47 struct crash_mem {
48 unsigned int nr_ranges;
49 struct crash_mem_range ranges[CRASH_MAX_RANGES];
50 };
51
52 /* Misc data about ram ranges needed to prepare elf headers */
53 struct crash_elf_data {
54 struct kimage *image;
55 /*
56 * Total number of ram ranges we have after various adjustments for
57 * GART, crash reserved region etc.
58 */
59 unsigned int max_nr_ranges;
60 unsigned long gart_start, gart_end;
61
62 /* Pointer to elf header */
63 void *ehdr;
64 /* Pointer to next phdr */
65 void *bufp;
66 struct crash_mem mem;
67 };
68
69 /* Used while preparing memory map entries for second kernel */
70 struct crash_memmap_data {
71 struct boot_params *params;
72 /* Type of memory */
73 unsigned int type;
74 };
75
76 int in_crash_kexec;
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 */
85 crash_vmclear_fn __rcu *crash_vmclear_loaded_vmcss = NULL;
86 EXPORT_SYMBOL_GPL(crash_vmclear_loaded_vmcss);
87 unsigned long crash_zero_bytes;
88
89 static 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
102 static 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_vm(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 disable_local_APIC();
129 }
130
131 static void kdump_nmi_shootdown_cpus(void)
132 {
133 in_crash_kexec = 1;
134 nmi_shootdown_cpus(kdump_nmi_callback);
135
136 disable_local_APIC();
137 }
138
139 #else
140 static void kdump_nmi_shootdown_cpus(void)
141 {
142 /* There are no cpus to shootdown */
143 }
144 #endif
145
146 void native_machine_crash_shutdown(struct pt_regs *regs)
147 {
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.
152 *
153 * In practice this means shooting down the other cpus in
154 * an SMP system.
155 */
156 /* The kernel is broken so disable interrupts */
157 local_irq_disable();
158
159 kdump_nmi_shootdown_cpus();
160
161 /*
162 * VMCLEAR VMCSs loaded on this cpu if needed.
163 */
164 cpu_crash_vmclear_loaded_vmcss();
165
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.
169 */
170 cpu_emergency_vmxoff();
171 cpu_emergency_svm_disable();
172
173 #ifdef CONFIG_X86_IO_APIC
174 /* Prevent crash_kexec() from deadlocking on ioapic_lock. */
175 ioapic_zap_locks();
176 disable_IO_APIC();
177 #endif
178 lapic_shutdown();
179 #ifdef CONFIG_HPET_TIMER
180 hpet_disable();
181 #endif
182 crash_save_cpu(regs, safe_smp_processor_id());
183 }
184
185 #ifdef CONFIG_X86_64
186
187 static int get_nr_ram_ranges_callback(unsigned long start_pfn,
188 unsigned long nr_pfn, void *arg)
189 {
190 int *nr_ranges = arg;
191
192 (*nr_ranges)++;
193 return 0;
194 }
195
196 static int get_gart_ranges_callback(u64 start, u64 end, void *arg)
197 {
198 struct crash_elf_data *ced = arg;
199
200 ced->gart_start = start;
201 ced->gart_end = end;
202
203 /* Not expecting more than 1 gart aperture */
204 return 1;
205 }
206
207
208 /* Gather all the required information to prepare elf headers for ram regions */
209 static void fill_up_crash_elf_data(struct crash_elf_data *ced,
210 struct kimage *image)
211 {
212 unsigned int nr_ranges = 0;
213
214 ced->image = image;
215
216 walk_system_ram_range(0, -1, &nr_ranges,
217 get_nr_ram_ranges_callback);
218
219 ced->max_nr_ranges = nr_ranges;
220
221 /*
222 * We don't create ELF headers for GART aperture as an attempt
223 * to dump this memory in second kernel leads to hang/crash.
224 * If gart aperture is present, one needs to exclude that region
225 * and that could lead to need of extra phdr.
226 */
227 walk_iomem_res("GART", IORESOURCE_MEM, 0, -1,
228 ced, get_gart_ranges_callback);
229
230 /*
231 * If we have gart region, excluding that could potentially split
232 * a memory range, resulting in extra header. Account for that.
233 */
234 if (ced->gart_end)
235 ced->max_nr_ranges++;
236
237 /* Exclusion of crash region could split memory ranges */
238 ced->max_nr_ranges++;
239
240 /* If crashk_low_res is not 0, another range split possible */
241 if (crashk_low_res.end != 0)
242 ced->max_nr_ranges++;
243 }
244
245 static int exclude_mem_range(struct crash_mem *mem,
246 unsigned long long mstart, unsigned long long mend)
247 {
248 int i, j;
249 unsigned long long start, end;
250 struct crash_mem_range temp_range = {0, 0};
251
252 for (i = 0; i < mem->nr_ranges; i++) {
253 start = mem->ranges[i].start;
254 end = mem->ranges[i].end;
255
256 if (mstart > end || mend < start)
257 continue;
258
259 /* Truncate any area outside of range */
260 if (mstart < start)
261 mstart = start;
262 if (mend > end)
263 mend = end;
264
265 /* Found completely overlapping range */
266 if (mstart == start && mend == end) {
267 mem->ranges[i].start = 0;
268 mem->ranges[i].end = 0;
269 if (i < mem->nr_ranges - 1) {
270 /* Shift rest of the ranges to left */
271 for (j = i; j < mem->nr_ranges - 1; j++) {
272 mem->ranges[j].start =
273 mem->ranges[j+1].start;
274 mem->ranges[j].end =
275 mem->ranges[j+1].end;
276 }
277 }
278 mem->nr_ranges--;
279 return 0;
280 }
281
282 if (mstart > start && mend < end) {
283 /* Split original range */
284 mem->ranges[i].end = mstart - 1;
285 temp_range.start = mend + 1;
286 temp_range.end = end;
287 } else if (mstart != start)
288 mem->ranges[i].end = mstart - 1;
289 else
290 mem->ranges[i].start = mend + 1;
291 break;
292 }
293
294 /* If a split happend, add the split to array */
295 if (!temp_range.end)
296 return 0;
297
298 /* Split happened */
299 if (i == CRASH_MAX_RANGES - 1) {
300 pr_err("Too many crash ranges after split\n");
301 return -ENOMEM;
302 }
303
304 /* Location where new range should go */
305 j = i + 1;
306 if (j < mem->nr_ranges) {
307 /* Move over all ranges one slot towards the end */
308 for (i = mem->nr_ranges - 1; i >= j; i--)
309 mem->ranges[i + 1] = mem->ranges[i];
310 }
311
312 mem->ranges[j].start = temp_range.start;
313 mem->ranges[j].end = temp_range.end;
314 mem->nr_ranges++;
315 return 0;
316 }
317
318 /*
319 * Look for any unwanted ranges between mstart, mend and remove them. This
320 * might lead to split and split ranges are put in ced->mem.ranges[] array
321 */
322 static int elf_header_exclude_ranges(struct crash_elf_data *ced,
323 unsigned long long mstart, unsigned long long mend)
324 {
325 struct crash_mem *cmem = &ced->mem;
326 int ret = 0;
327
328 memset(cmem->ranges, 0, sizeof(cmem->ranges));
329
330 cmem->ranges[0].start = mstart;
331 cmem->ranges[0].end = mend;
332 cmem->nr_ranges = 1;
333
334 /* Exclude crashkernel region */
335 ret = exclude_mem_range(cmem, crashk_res.start, crashk_res.end);
336 if (ret)
337 return ret;
338
339 ret = exclude_mem_range(cmem, crashk_low_res.start, crashk_low_res.end);
340 if (ret)
341 return ret;
342
343 /* Exclude GART region */
344 if (ced->gart_end) {
345 ret = exclude_mem_range(cmem, ced->gart_start, ced->gart_end);
346 if (ret)
347 return ret;
348 }
349
350 return ret;
351 }
352
353 static int prepare_elf64_ram_headers_callback(u64 start, u64 end, void *arg)
354 {
355 struct crash_elf_data *ced = arg;
356 Elf64_Ehdr *ehdr;
357 Elf64_Phdr *phdr;
358 unsigned long mstart, mend;
359 struct kimage *image = ced->image;
360 struct crash_mem *cmem;
361 int ret, i;
362
363 ehdr = ced->ehdr;
364
365 /* Exclude unwanted mem ranges */
366 ret = elf_header_exclude_ranges(ced, start, end);
367 if (ret)
368 return ret;
369
370 /* Go through all the ranges in ced->mem.ranges[] and prepare phdr */
371 cmem = &ced->mem;
372
373 for (i = 0; i < cmem->nr_ranges; i++) {
374 mstart = cmem->ranges[i].start;
375 mend = cmem->ranges[i].end;
376
377 phdr = ced->bufp;
378 ced->bufp += sizeof(Elf64_Phdr);
379
380 phdr->p_type = PT_LOAD;
381 phdr->p_flags = PF_R|PF_W|PF_X;
382 phdr->p_offset = mstart;
383
384 /*
385 * If a range matches backup region, adjust offset to backup
386 * segment.
387 */
388 if (mstart == image->arch.backup_src_start &&
389 (mend - mstart + 1) == image->arch.backup_src_sz)
390 phdr->p_offset = image->arch.backup_load_addr;
391
392 phdr->p_paddr = mstart;
393 phdr->p_vaddr = (unsigned long long) __va(mstart);
394 phdr->p_filesz = phdr->p_memsz = mend - mstart + 1;
395 phdr->p_align = 0;
396 ehdr->e_phnum++;
397 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",
398 phdr, phdr->p_vaddr, phdr->p_paddr, phdr->p_filesz,
399 ehdr->e_phnum, phdr->p_offset);
400 }
401
402 return ret;
403 }
404
405 static int prepare_elf64_headers(struct crash_elf_data *ced,
406 void **addr, unsigned long *sz)
407 {
408 Elf64_Ehdr *ehdr;
409 Elf64_Phdr *phdr;
410 unsigned long nr_cpus = num_possible_cpus(), nr_phdr, elf_sz;
411 unsigned char *buf, *bufp;
412 unsigned int cpu;
413 unsigned long long notes_addr;
414 int ret;
415
416 /* extra phdr for vmcoreinfo elf note */
417 nr_phdr = nr_cpus + 1;
418 nr_phdr += ced->max_nr_ranges;
419
420 /*
421 * kexec-tools creates an extra PT_LOAD phdr for kernel text mapping
422 * area on x86_64 (ffffffff80000000 - ffffffffa0000000).
423 * I think this is required by tools like gdb. So same physical
424 * memory will be mapped in two elf headers. One will contain kernel
425 * text virtual addresses and other will have __va(physical) addresses.
426 */
427
428 nr_phdr++;
429 elf_sz = sizeof(Elf64_Ehdr) + nr_phdr * sizeof(Elf64_Phdr);
430 elf_sz = ALIGN(elf_sz, ELF_CORE_HEADER_ALIGN);
431
432 buf = vzalloc(elf_sz);
433 if (!buf)
434 return -ENOMEM;
435
436 bufp = buf;
437 ehdr = (Elf64_Ehdr *)bufp;
438 bufp += sizeof(Elf64_Ehdr);
439 memcpy(ehdr->e_ident, ELFMAG, SELFMAG);
440 ehdr->e_ident[EI_CLASS] = ELFCLASS64;
441 ehdr->e_ident[EI_DATA] = ELFDATA2LSB;
442 ehdr->e_ident[EI_VERSION] = EV_CURRENT;
443 ehdr->e_ident[EI_OSABI] = ELF_OSABI;
444 memset(ehdr->e_ident + EI_PAD, 0, EI_NIDENT - EI_PAD);
445 ehdr->e_type = ET_CORE;
446 ehdr->e_machine = ELF_ARCH;
447 ehdr->e_version = EV_CURRENT;
448 ehdr->e_phoff = sizeof(Elf64_Ehdr);
449 ehdr->e_ehsize = sizeof(Elf64_Ehdr);
450 ehdr->e_phentsize = sizeof(Elf64_Phdr);
451
452 /* Prepare one phdr of type PT_NOTE for each present cpu */
453 for_each_present_cpu(cpu) {
454 phdr = (Elf64_Phdr *)bufp;
455 bufp += sizeof(Elf64_Phdr);
456 phdr->p_type = PT_NOTE;
457 notes_addr = per_cpu_ptr_to_phys(per_cpu_ptr(crash_notes, cpu));
458 phdr->p_offset = phdr->p_paddr = notes_addr;
459 phdr->p_filesz = phdr->p_memsz = sizeof(note_buf_t);
460 (ehdr->e_phnum)++;
461 }
462
463 /* Prepare one PT_NOTE header for vmcoreinfo */
464 phdr = (Elf64_Phdr *)bufp;
465 bufp += sizeof(Elf64_Phdr);
466 phdr->p_type = PT_NOTE;
467 phdr->p_offset = phdr->p_paddr = paddr_vmcoreinfo_note();
468 phdr->p_filesz = phdr->p_memsz = sizeof(vmcoreinfo_note);
469 (ehdr->e_phnum)++;
470
471 #ifdef CONFIG_X86_64
472 /* Prepare PT_LOAD type program header for kernel text region */
473 phdr = (Elf64_Phdr *)bufp;
474 bufp += sizeof(Elf64_Phdr);
475 phdr->p_type = PT_LOAD;
476 phdr->p_flags = PF_R|PF_W|PF_X;
477 phdr->p_vaddr = (Elf64_Addr)_text;
478 phdr->p_filesz = phdr->p_memsz = _end - _text;
479 phdr->p_offset = phdr->p_paddr = __pa_symbol(_text);
480 (ehdr->e_phnum)++;
481 #endif
482
483 /* Prepare PT_LOAD headers for system ram chunks. */
484 ced->ehdr = ehdr;
485 ced->bufp = bufp;
486 ret = walk_system_ram_res(0, -1, ced,
487 prepare_elf64_ram_headers_callback);
488 if (ret < 0)
489 return ret;
490
491 *addr = buf;
492 *sz = elf_sz;
493 return 0;
494 }
495
496 /* Prepare elf headers. Return addr and size */
497 static int prepare_elf_headers(struct kimage *image, void **addr,
498 unsigned long *sz)
499 {
500 struct crash_elf_data *ced;
501 int ret;
502
503 ced = kzalloc(sizeof(*ced), GFP_KERNEL);
504 if (!ced)
505 return -ENOMEM;
506
507 fill_up_crash_elf_data(ced, image);
508
509 /* By default prepare 64bit headers */
510 ret = prepare_elf64_headers(ced, addr, sz);
511 kfree(ced);
512 return ret;
513 }
514
515 static int add_e820_entry(struct boot_params *params, struct e820entry *entry)
516 {
517 unsigned int nr_e820_entries;
518
519 nr_e820_entries = params->e820_entries;
520 if (nr_e820_entries >= E820MAX)
521 return 1;
522
523 memcpy(&params->e820_map[nr_e820_entries], entry,
524 sizeof(struct e820entry));
525 params->e820_entries++;
526 return 0;
527 }
528
529 static int memmap_entry_callback(u64 start, u64 end, void *arg)
530 {
531 struct crash_memmap_data *cmd = arg;
532 struct boot_params *params = cmd->params;
533 struct e820entry ei;
534
535 ei.addr = start;
536 ei.size = end - start + 1;
537 ei.type = cmd->type;
538 add_e820_entry(params, &ei);
539
540 return 0;
541 }
542
543 static int memmap_exclude_ranges(struct kimage *image, struct crash_mem *cmem,
544 unsigned long long mstart,
545 unsigned long long mend)
546 {
547 unsigned long start, end;
548 int ret = 0;
549
550 cmem->ranges[0].start = mstart;
551 cmem->ranges[0].end = mend;
552 cmem->nr_ranges = 1;
553
554 /* Exclude Backup region */
555 start = image->arch.backup_load_addr;
556 end = start + image->arch.backup_src_sz - 1;
557 ret = exclude_mem_range(cmem, start, end);
558 if (ret)
559 return ret;
560
561 /* Exclude elf header region */
562 start = image->arch.elf_load_addr;
563 end = start + image->arch.elf_headers_sz - 1;
564 return exclude_mem_range(cmem, start, end);
565 }
566
567 /* Prepare memory map for crash dump kernel */
568 int crash_setup_memmap_entries(struct kimage *image, struct boot_params *params)
569 {
570 int i, ret = 0;
571 unsigned long flags;
572 struct e820entry ei;
573 struct crash_memmap_data cmd;
574 struct crash_mem *cmem;
575
576 cmem = vzalloc(sizeof(struct crash_mem));
577 if (!cmem)
578 return -ENOMEM;
579
580 memset(&cmd, 0, sizeof(struct crash_memmap_data));
581 cmd.params = params;
582
583 /* Add first 640K segment */
584 ei.addr = image->arch.backup_src_start;
585 ei.size = image->arch.backup_src_sz;
586 ei.type = E820_RAM;
587 add_e820_entry(params, &ei);
588
589 /* Add ACPI tables */
590 cmd.type = E820_ACPI;
591 flags = IORESOURCE_MEM | IORESOURCE_BUSY;
592 walk_iomem_res("ACPI Tables", flags, 0, -1, &cmd,
593 memmap_entry_callback);
594
595 /* Add ACPI Non-volatile Storage */
596 cmd.type = E820_NVS;
597 walk_iomem_res("ACPI Non-volatile Storage", flags, 0, -1, &cmd,
598 memmap_entry_callback);
599
600 /* Add crashk_low_res region */
601 if (crashk_low_res.end) {
602 ei.addr = crashk_low_res.start;
603 ei.size = crashk_low_res.end - crashk_low_res.start + 1;
604 ei.type = E820_RAM;
605 add_e820_entry(params, &ei);
606 }
607
608 /* Exclude some ranges from crashk_res and add rest to memmap */
609 ret = memmap_exclude_ranges(image, cmem, crashk_res.start,
610 crashk_res.end);
611 if (ret)
612 goto out;
613
614 for (i = 0; i < cmem->nr_ranges; i++) {
615 ei.size = cmem->ranges[i].end - cmem->ranges[i].start + 1;
616
617 /* If entry is less than a page, skip it */
618 if (ei.size < PAGE_SIZE)
619 continue;
620 ei.addr = cmem->ranges[i].start;
621 ei.type = E820_RAM;
622 add_e820_entry(params, &ei);
623 }
624
625 out:
626 vfree(cmem);
627 return ret;
628 }
629
630 static int determine_backup_region(u64 start, u64 end, void *arg)
631 {
632 struct kimage *image = arg;
633
634 image->arch.backup_src_start = start;
635 image->arch.backup_src_sz = end - start + 1;
636
637 /* Expecting only one range for backup region */
638 return 1;
639 }
640
641 int crash_load_segments(struct kimage *image)
642 {
643 unsigned long src_start, src_sz, elf_sz;
644 void *elf_addr;
645 int ret;
646
647 /*
648 * Determine and load a segment for backup area. First 640K RAM
649 * region is backup source
650 */
651
652 ret = walk_system_ram_res(KEXEC_BACKUP_SRC_START, KEXEC_BACKUP_SRC_END,
653 image, determine_backup_region);
654
655 /* Zero or postive return values are ok */
656 if (ret < 0)
657 return ret;
658
659 src_start = image->arch.backup_src_start;
660 src_sz = image->arch.backup_src_sz;
661
662 /* Add backup segment. */
663 if (src_sz) {
664 /*
665 * Ideally there is no source for backup segment. This is
666 * copied in purgatory after crash. Just add a zero filled
667 * segment for now to make sure checksum logic works fine.
668 */
669 ret = kexec_add_buffer(image, (char *)&crash_zero_bytes,
670 sizeof(crash_zero_bytes), src_sz,
671 PAGE_SIZE, 0, -1, 0,
672 &image->arch.backup_load_addr);
673 if (ret)
674 return ret;
675 pr_debug("Loaded backup region at 0x%lx backup_start=0x%lx memsz=0x%lx\n",
676 image->arch.backup_load_addr, src_start, src_sz);
677 }
678
679 /* Prepare elf headers and add a segment */
680 ret = prepare_elf_headers(image, &elf_addr, &elf_sz);
681 if (ret)
682 return ret;
683
684 image->arch.elf_headers = elf_addr;
685 image->arch.elf_headers_sz = elf_sz;
686
687 ret = kexec_add_buffer(image, (char *)elf_addr, elf_sz, elf_sz,
688 ELF_CORE_HEADER_ALIGN, 0, -1, 0,
689 &image->arch.elf_load_addr);
690 if (ret) {
691 vfree((void *)image->arch.elf_headers);
692 return ret;
693 }
694 pr_debug("Loaded ELF headers at 0x%lx bufsz=0x%lx memsz=0x%lx\n",
695 image->arch.elf_load_addr, elf_sz, elf_sz);
696
697 return ret;
698 }
699
700 #endif /* CONFIG_X86_64 */
This page took 0.046383 seconds and 5 git commands to generate.