1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * Architecture specific (i386/x86_64) functions for kexec based crash dumps.
4 *
5 * Created by: Hariprasad Nellitheertha (hari@in.ibm.com)
6 *
7 * Copyright (C) IBM Corporation, 2004. All rights reserved.
8 * Copyright (C) Red Hat Inc., 2014. All rights reserved.
9 * Authors:
10 * Vivek Goyal <vgoyal@redhat.com>
11 *
12 */
13
14#define pr_fmt(fmt) "kexec: " fmt
15
16#include <linux/types.h>
17#include <linux/kernel.h>
18#include <linux/smp.h>
19#include <linux/reboot.h>
20#include <linux/kexec.h>
21#include <linux/delay.h>
22#include <linux/elf.h>
23#include <linux/elfcore.h>
24#include <linux/export.h>
25#include <linux/slab.h>
26#include <linux/vmalloc.h>
27#include <linux/memblock.h>
28
29#include <asm/processor.h>
30#include <asm/hardirq.h>
31#include <asm/nmi.h>
32#include <asm/hw_irq.h>
33#include <asm/apic.h>
34#include <asm/e820/types.h>
35#include <asm/io_apic.h>
36#include <asm/hpet.h>
37#include <linux/kdebug.h>
38#include <asm/cpu.h>
39#include <asm/reboot.h>
40#include <asm/intel_pt.h>
41#include <asm/crash.h>
42#include <asm/cmdline.h>
43
44/* Used while preparing memory map entries for second kernel */
45struct crash_memmap_data {
46 struct boot_params *params;
47 /* Type of memory */
48 unsigned int type;
49};
50
51#if defined(CONFIG_SMP) && defined(CONFIG_X86_LOCAL_APIC)
52
53static void kdump_nmi_callback(int cpu, struct pt_regs *regs)
54{
55 crash_save_cpu(regs, cpu);
56
57 /*
58 * Disable Intel PT to stop its logging
59 */
60 cpu_emergency_stop_pt();
61
62 disable_local_APIC();
63}
64
65void kdump_nmi_shootdown_cpus(void)
66{
67 nmi_shootdown_cpus(callback: kdump_nmi_callback);
68
69 disable_local_APIC();
70}
71
72/* Override the weak function in kernel/panic.c */
73void crash_smp_send_stop(void)
74{
75 static int cpus_stopped;
76
77 if (cpus_stopped)
78 return;
79
80 if (smp_ops.crash_stop_other_cpus)
81 smp_ops.crash_stop_other_cpus();
82 else
83 smp_send_stop();
84
85 cpus_stopped = 1;
86}
87
88#else
89void crash_smp_send_stop(void)
90{
91 /* There are no cpus to shootdown */
92}
93#endif
94
95void native_machine_crash_shutdown(struct pt_regs *regs)
96{
97 /* This function is only called after the system
98 * has panicked or is otherwise in a critical state.
99 * The minimum amount of code to allow a kexec'd kernel
100 * to run successfully needs to happen here.
101 *
102 * In practice this means shooting down the other cpus in
103 * an SMP system.
104 */
105 /* The kernel is broken so disable interrupts */
106 local_irq_disable();
107
108 crash_smp_send_stop();
109
110 cpu_emergency_disable_virtualization();
111
112 /*
113 * Disable Intel PT to stop its logging
114 */
115 cpu_emergency_stop_pt();
116
117#ifdef CONFIG_X86_IO_APIC
118 /* Prevent crash_kexec() from deadlocking on ioapic_lock. */
119 ioapic_zap_locks();
120 clear_IO_APIC();
121#endif
122 lapic_shutdown();
123 restore_boot_irq_mode();
124#ifdef CONFIG_HPET_TIMER
125 hpet_disable();
126#endif
127 crash_save_cpu(regs, safe_smp_processor_id());
128}
129
130#if defined(CONFIG_KEXEC_FILE) || defined(CONFIG_CRASH_HOTPLUG)
131static int get_nr_ram_ranges_callback(struct resource *res, void *arg)
132{
133 unsigned int *nr_ranges = arg;
134
135 (*nr_ranges)++;
136 return 0;
137}
138
139/* Gather all the required information to prepare elf headers for ram regions */
140static struct crash_mem *fill_up_crash_elf_data(void)
141{
142 unsigned int nr_ranges = 0;
143 struct crash_mem *cmem;
144
145 walk_system_ram_res(start: 0, end: -1, arg: &nr_ranges, func: get_nr_ram_ranges_callback);
146 if (!nr_ranges)
147 return NULL;
148
149 /*
150 * Exclusion of crash region and/or crashk_low_res may cause
151 * another range split. So add extra two slots here.
152 */
153 nr_ranges += 2;
154 cmem = vzalloc(struct_size(cmem, ranges, nr_ranges));
155 if (!cmem)
156 return NULL;
157
158 cmem->max_nr_ranges = nr_ranges;
159 cmem->nr_ranges = 0;
160
161 return cmem;
162}
163
164/*
165 * Look for any unwanted ranges between mstart, mend and remove them. This
166 * might lead to split and split ranges are put in cmem->ranges[] array
167 */
168static int elf_header_exclude_ranges(struct crash_mem *cmem)
169{
170 int ret = 0;
171
172 /* Exclude the low 1M because it is always reserved */
173 ret = crash_exclude_mem_range(mem: cmem, mstart: 0, mend: (1<<20)-1);
174 if (ret)
175 return ret;
176
177 /* Exclude crashkernel region */
178 ret = crash_exclude_mem_range(mem: cmem, mstart: crashk_res.start, mend: crashk_res.end);
179 if (ret)
180 return ret;
181
182 if (crashk_low_res.end)
183 ret = crash_exclude_mem_range(mem: cmem, mstart: crashk_low_res.start,
184 mend: crashk_low_res.end);
185
186 return ret;
187}
188
189static int prepare_elf64_ram_headers_callback(struct resource *res, void *arg)
190{
191 struct crash_mem *cmem = arg;
192
193 cmem->ranges[cmem->nr_ranges].start = res->start;
194 cmem->ranges[cmem->nr_ranges].end = res->end;
195 cmem->nr_ranges++;
196
197 return 0;
198}
199
200/* Prepare elf headers. Return addr and size */
201static int prepare_elf_headers(struct kimage *image, void **addr,
202 unsigned long *sz, unsigned long *nr_mem_ranges)
203{
204 struct crash_mem *cmem;
205 int ret;
206
207 cmem = fill_up_crash_elf_data();
208 if (!cmem)
209 return -ENOMEM;
210
211 ret = walk_system_ram_res(start: 0, end: -1, arg: cmem, func: prepare_elf64_ram_headers_callback);
212 if (ret)
213 goto out;
214
215 /* Exclude unwanted mem ranges */
216 ret = elf_header_exclude_ranges(cmem);
217 if (ret)
218 goto out;
219
220 /* Return the computed number of memory ranges, for hotplug usage */
221 *nr_mem_ranges = cmem->nr_ranges;
222
223 /* By default prepare 64bit headers */
224 ret = crash_prepare_elf64_headers(mem: cmem, IS_ENABLED(CONFIG_X86_64), addr, sz);
225
226out:
227 vfree(addr: cmem);
228 return ret;
229}
230#endif
231
232#ifdef CONFIG_KEXEC_FILE
233static int add_e820_entry(struct boot_params *params, struct e820_entry *entry)
234{
235 unsigned int nr_e820_entries;
236
237 nr_e820_entries = params->e820_entries;
238 if (nr_e820_entries >= E820_MAX_ENTRIES_ZEROPAGE)
239 return 1;
240
241 memcpy(&params->e820_table[nr_e820_entries], entry, sizeof(struct e820_entry));
242 params->e820_entries++;
243 return 0;
244}
245
246static int memmap_entry_callback(struct resource *res, void *arg)
247{
248 struct crash_memmap_data *cmd = arg;
249 struct boot_params *params = cmd->params;
250 struct e820_entry ei;
251
252 ei.addr = res->start;
253 ei.size = resource_size(res);
254 ei.type = cmd->type;
255 add_e820_entry(params, entry: &ei);
256
257 return 0;
258}
259
260static int memmap_exclude_ranges(struct kimage *image, struct crash_mem *cmem,
261 unsigned long long mstart,
262 unsigned long long mend)
263{
264 unsigned long start, end;
265
266 cmem->ranges[0].start = mstart;
267 cmem->ranges[0].end = mend;
268 cmem->nr_ranges = 1;
269
270 /* Exclude elf header region */
271 start = image->elf_load_addr;
272 end = start + image->elf_headers_sz - 1;
273 return crash_exclude_mem_range(mem: cmem, mstart: start, mend: end);
274}
275
276/* Prepare memory map for crash dump kernel */
277int crash_setup_memmap_entries(struct kimage *image, struct boot_params *params)
278{
279 int i, ret = 0;
280 unsigned long flags;
281 struct e820_entry ei;
282 struct crash_memmap_data cmd;
283 struct crash_mem *cmem;
284
285 cmem = vzalloc(struct_size(cmem, ranges, 1));
286 if (!cmem)
287 return -ENOMEM;
288
289 memset(&cmd, 0, sizeof(struct crash_memmap_data));
290 cmd.params = params;
291
292 /* Add the low 1M */
293 cmd.type = E820_TYPE_RAM;
294 flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
295 walk_iomem_res_desc(desc: IORES_DESC_NONE, flags, start: 0, end: (1<<20)-1, arg: &cmd,
296 func: memmap_entry_callback);
297
298 /* Add ACPI tables */
299 cmd.type = E820_TYPE_ACPI;
300 flags = IORESOURCE_MEM | IORESOURCE_BUSY;
301 walk_iomem_res_desc(desc: IORES_DESC_ACPI_TABLES, flags, start: 0, end: -1, arg: &cmd,
302 func: memmap_entry_callback);
303
304 /* Add ACPI Non-volatile Storage */
305 cmd.type = E820_TYPE_NVS;
306 walk_iomem_res_desc(desc: IORES_DESC_ACPI_NV_STORAGE, flags, start: 0, end: -1, arg: &cmd,
307 func: memmap_entry_callback);
308
309 /* Add e820 reserved ranges */
310 cmd.type = E820_TYPE_RESERVED;
311 flags = IORESOURCE_MEM;
312 walk_iomem_res_desc(desc: IORES_DESC_RESERVED, flags, start: 0, end: -1, arg: &cmd,
313 func: memmap_entry_callback);
314
315 /* Add crashk_low_res region */
316 if (crashk_low_res.end) {
317 ei.addr = crashk_low_res.start;
318 ei.size = resource_size(res: &crashk_low_res);
319 ei.type = E820_TYPE_RAM;
320 add_e820_entry(params, entry: &ei);
321 }
322
323 /* Exclude some ranges from crashk_res and add rest to memmap */
324 ret = memmap_exclude_ranges(image, cmem, mstart: crashk_res.start, mend: crashk_res.end);
325 if (ret)
326 goto out;
327
328 for (i = 0; i < cmem->nr_ranges; i++) {
329 ei.size = cmem->ranges[i].end - cmem->ranges[i].start + 1;
330
331 /* If entry is less than a page, skip it */
332 if (ei.size < PAGE_SIZE)
333 continue;
334 ei.addr = cmem->ranges[i].start;
335 ei.type = E820_TYPE_RAM;
336 add_e820_entry(params, entry: &ei);
337 }
338
339out:
340 vfree(addr: cmem);
341 return ret;
342}
343
344int crash_load_segments(struct kimage *image)
345{
346 int ret;
347 unsigned long pnum = 0;
348 struct kexec_buf kbuf = { .image = image, .buf_min = 0,
349 .buf_max = ULONG_MAX, .top_down = false };
350
351 /* Prepare elf headers and add a segment */
352 ret = prepare_elf_headers(image, addr: &kbuf.buffer, sz: &kbuf.bufsz, nr_mem_ranges: &pnum);
353 if (ret)
354 return ret;
355
356 image->elf_headers = kbuf.buffer;
357 image->elf_headers_sz = kbuf.bufsz;
358 kbuf.memsz = kbuf.bufsz;
359
360#ifdef CONFIG_CRASH_HOTPLUG
361 /*
362 * The elfcorehdr segment size accounts for VMCOREINFO, kernel_map,
363 * maximum CPUs and maximum memory ranges.
364 */
365 if (IS_ENABLED(CONFIG_MEMORY_HOTPLUG))
366 pnum = 2 + CONFIG_NR_CPUS_DEFAULT + CONFIG_CRASH_MAX_MEMORY_RANGES;
367 else
368 pnum += 2 + CONFIG_NR_CPUS_DEFAULT;
369
370 if (pnum < (unsigned long)PN_XNUM) {
371 kbuf.memsz = pnum * sizeof(Elf64_Phdr);
372 kbuf.memsz += sizeof(Elf64_Ehdr);
373
374 image->elfcorehdr_index = image->nr_segments;
375
376 /* Mark as usable to crash kernel, else crash kernel fails on boot */
377 image->elf_headers_sz = kbuf.memsz;
378 } else {
379 pr_err("number of Phdrs %lu exceeds max\n", pnum);
380 }
381#endif
382
383 kbuf.buf_align = ELF_CORE_HEADER_ALIGN;
384 kbuf.mem = KEXEC_BUF_MEM_UNKNOWN;
385 ret = kexec_add_buffer(kbuf: &kbuf);
386 if (ret)
387 return ret;
388 image->elf_load_addr = kbuf.mem;
389 pr_debug("Loaded ELF headers at 0x%lx bufsz=0x%lx memsz=0x%lx\n",
390 image->elf_load_addr, kbuf.bufsz, kbuf.memsz);
391
392 return ret;
393}
394#endif /* CONFIG_KEXEC_FILE */
395
396#ifdef CONFIG_CRASH_HOTPLUG
397
398#undef pr_fmt
399#define pr_fmt(fmt) "crash hp: " fmt
400
401/* These functions provide the value for the sysfs crash_hotplug nodes */
402#ifdef CONFIG_HOTPLUG_CPU
403int arch_crash_hotplug_cpu_support(void)
404{
405 return crash_check_update_elfcorehdr();
406}
407#endif
408
409#ifdef CONFIG_MEMORY_HOTPLUG
410int arch_crash_hotplug_memory_support(void)
411{
412 return crash_check_update_elfcorehdr();
413}
414#endif
415
416unsigned int arch_crash_get_elfcorehdr_size(void)
417{
418 unsigned int sz;
419
420 /* kernel_map, VMCOREINFO and maximum CPUs */
421 sz = 2 + CONFIG_NR_CPUS_DEFAULT;
422 if (IS_ENABLED(CONFIG_MEMORY_HOTPLUG))
423 sz += CONFIG_CRASH_MAX_MEMORY_RANGES;
424 sz *= sizeof(Elf64_Phdr);
425 return sz;
426}
427
428/**
429 * arch_crash_handle_hotplug_event() - Handle hotplug elfcorehdr changes
430 * @image: a pointer to kexec_crash_image
431 *
432 * Prepare the new elfcorehdr and replace the existing elfcorehdr.
433 */
434void arch_crash_handle_hotplug_event(struct kimage *image)
435{
436 void *elfbuf = NULL, *old_elfcorehdr;
437 unsigned long nr_mem_ranges;
438 unsigned long mem, memsz;
439 unsigned long elfsz = 0;
440
441 /*
442 * As crash_prepare_elf64_headers() has already described all
443 * possible CPUs, there is no need to update the elfcorehdr
444 * for additional CPU changes.
445 */
446 if ((image->file_mode || image->elfcorehdr_updated) &&
447 ((image->hp_action == KEXEC_CRASH_HP_ADD_CPU) ||
448 (image->hp_action == KEXEC_CRASH_HP_REMOVE_CPU)))
449 return;
450
451 /*
452 * Create the new elfcorehdr reflecting the changes to CPU and/or
453 * memory resources.
454 */
455 if (prepare_elf_headers(image, addr: &elfbuf, sz: &elfsz, nr_mem_ranges: &nr_mem_ranges)) {
456 pr_err("unable to create new elfcorehdr");
457 goto out;
458 }
459
460 /*
461 * Obtain address and size of the elfcorehdr segment, and
462 * check it against the new elfcorehdr buffer.
463 */
464 mem = image->segment[image->elfcorehdr_index].mem;
465 memsz = image->segment[image->elfcorehdr_index].memsz;
466 if (elfsz > memsz) {
467 pr_err("update elfcorehdr elfsz %lu > memsz %lu",
468 elfsz, memsz);
469 goto out;
470 }
471
472 /*
473 * Copy new elfcorehdr over the old elfcorehdr at destination.
474 */
475 old_elfcorehdr = kmap_local_page(pfn_to_page(mem >> PAGE_SHIFT));
476 if (!old_elfcorehdr) {
477 pr_err("mapping elfcorehdr segment failed\n");
478 goto out;
479 }
480
481 /*
482 * Temporarily invalidate the crash image while the
483 * elfcorehdr is updated.
484 */
485 xchg(&kexec_crash_image, NULL);
486 memcpy_flushcache(dst: old_elfcorehdr, src: elfbuf, cnt: elfsz);
487 xchg(&kexec_crash_image, image);
488 kunmap_local(old_elfcorehdr);
489 pr_debug("updated elfcorehdr\n");
490
491out:
492 vfree(addr: elfbuf);
493}
494#endif
495

source code of linux/arch/x86/kernel/crash.c