Loading...
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/bootparam.h>
30#include <asm/processor.h>
31#include <asm/hardirq.h>
32#include <asm/nmi.h>
33#include <asm/hw_irq.h>
34#include <asm/apic.h>
35#include <asm/e820/types.h>
36#include <asm/io_apic.h>
37#include <asm/hpet.h>
38#include <linux/kdebug.h>
39#include <asm/cpu.h>
40#include <asm/reboot.h>
41#include <asm/intel_pt.h>
42#include <asm/crash.h>
43#include <asm/cmdline.h>
44#include <asm/sev.h>
45
46/* Used while preparing memory map entries for second kernel */
47struct crash_memmap_data {
48 struct boot_params *params;
49 /* Type of memory */
50 unsigned int type;
51};
52
53#if defined(CONFIG_SMP) && defined(CONFIG_X86_LOCAL_APIC)
54
55static void kdump_nmi_callback(int cpu, struct pt_regs *regs)
56{
57 crash_save_cpu(regs, cpu);
58
59 /*
60 * Disable Intel PT to stop its logging
61 */
62 cpu_emergency_stop_pt();
63
64 kdump_sev_callback();
65
66 disable_local_APIC();
67}
68
69void kdump_nmi_shootdown_cpus(void)
70{
71 nmi_shootdown_cpus(kdump_nmi_callback);
72
73 disable_local_APIC();
74}
75
76/* Override the weak function in kernel/panic.c */
77void crash_smp_send_stop(void)
78{
79 static int cpus_stopped;
80
81 if (cpus_stopped)
82 return;
83
84 if (smp_ops.crash_stop_other_cpus)
85 smp_ops.crash_stop_other_cpus();
86 else
87 smp_send_stop();
88
89 cpus_stopped = 1;
90}
91
92#else
93void crash_smp_send_stop(void)
94{
95 /* There are no cpus to shootdown */
96}
97#endif
98
99void native_machine_crash_shutdown(struct pt_regs *regs)
100{
101 /* This function is only called after the system
102 * has panicked or is otherwise in a critical state.
103 * The minimum amount of code to allow a kexec'd kernel
104 * to run successfully needs to happen here.
105 *
106 * In practice this means shooting down the other cpus in
107 * an SMP system.
108 */
109 /* The kernel is broken so disable interrupts */
110 local_irq_disable();
111
112 crash_smp_send_stop();
113
114 cpu_emergency_disable_virtualization();
115
116 /*
117 * Disable Intel PT to stop its logging
118 */
119 cpu_emergency_stop_pt();
120
121#ifdef CONFIG_X86_IO_APIC
122 /* Prevent crash_kexec() from deadlocking on ioapic_lock. */
123 ioapic_zap_locks();
124 clear_IO_APIC();
125#endif
126 lapic_shutdown();
127 restore_boot_irq_mode();
128#ifdef CONFIG_HPET_TIMER
129 hpet_disable();
130#endif
131
132 /*
133 * Non-crash kexec calls enc_kexec_begin() while scheduling is still
134 * active. This allows the callback to wait until all in-flight
135 * shared<->private conversions are complete. In a crash scenario,
136 * enc_kexec_begin() gets called after all but one CPU have been shut
137 * down and interrupts have been disabled. This allows the callback to
138 * detect a race with the conversion and report it.
139 */
140 x86_platform.guest.enc_kexec_begin();
141 x86_platform.guest.enc_kexec_finish();
142
143 crash_save_cpu(regs, safe_smp_processor_id());
144}
145
146#if defined(CONFIG_KEXEC_FILE) || defined(CONFIG_CRASH_HOTPLUG)
147static int get_nr_ram_ranges_callback(struct resource *res, void *arg)
148{
149 unsigned int *nr_ranges = arg;
150
151 (*nr_ranges)++;
152 return 0;
153}
154
155/* Gather all the required information to prepare elf headers for ram regions */
156static struct crash_mem *fill_up_crash_elf_data(void)
157{
158 unsigned int nr_ranges = 0;
159 struct crash_mem *cmem;
160
161 walk_system_ram_res(0, -1, &nr_ranges, get_nr_ram_ranges_callback);
162 if (!nr_ranges)
163 return NULL;
164
165 /*
166 * Exclusion of crash region and/or crashk_low_res may cause
167 * another range split. So add extra two slots here.
168 */
169 nr_ranges += 2;
170 cmem = vzalloc(struct_size(cmem, ranges, nr_ranges));
171 if (!cmem)
172 return NULL;
173
174 cmem->max_nr_ranges = nr_ranges;
175 cmem->nr_ranges = 0;
176
177 return cmem;
178}
179
180/*
181 * Look for any unwanted ranges between mstart, mend and remove them. This
182 * might lead to split and split ranges are put in cmem->ranges[] array
183 */
184static int elf_header_exclude_ranges(struct crash_mem *cmem)
185{
186 int ret = 0;
187
188 /* Exclude the low 1M because it is always reserved */
189 ret = crash_exclude_mem_range(cmem, 0, SZ_1M - 1);
190 if (ret)
191 return ret;
192
193 /* Exclude crashkernel region */
194 ret = crash_exclude_mem_range(cmem, crashk_res.start, crashk_res.end);
195 if (ret)
196 return ret;
197
198 if (crashk_low_res.end)
199 ret = crash_exclude_mem_range(cmem, crashk_low_res.start,
200 crashk_low_res.end);
201
202 return ret;
203}
204
205static int prepare_elf64_ram_headers_callback(struct resource *res, void *arg)
206{
207 struct crash_mem *cmem = arg;
208
209 cmem->ranges[cmem->nr_ranges].start = res->start;
210 cmem->ranges[cmem->nr_ranges].end = res->end;
211 cmem->nr_ranges++;
212
213 return 0;
214}
215
216/* Prepare elf headers. Return addr and size */
217static int prepare_elf_headers(void **addr, unsigned long *sz,
218 unsigned long *nr_mem_ranges)
219{
220 struct crash_mem *cmem;
221 int ret;
222
223 cmem = fill_up_crash_elf_data();
224 if (!cmem)
225 return -ENOMEM;
226
227 ret = walk_system_ram_res(0, -1, cmem, prepare_elf64_ram_headers_callback);
228 if (ret)
229 goto out;
230
231 /* Exclude unwanted mem ranges */
232 ret = elf_header_exclude_ranges(cmem);
233 if (ret)
234 goto out;
235
236 /* Return the computed number of memory ranges, for hotplug usage */
237 *nr_mem_ranges = cmem->nr_ranges;
238
239 /* By default prepare 64bit headers */
240 ret = crash_prepare_elf64_headers(cmem, IS_ENABLED(CONFIG_X86_64), addr, sz);
241
242out:
243 vfree(cmem);
244 return ret;
245}
246#endif
247
248#ifdef CONFIG_KEXEC_FILE
249static int add_e820_entry(struct boot_params *params, struct e820_entry *entry)
250{
251 unsigned int nr_e820_entries;
252
253 nr_e820_entries = params->e820_entries;
254 if (nr_e820_entries >= E820_MAX_ENTRIES_ZEROPAGE)
255 return 1;
256
257 memcpy(¶ms->e820_table[nr_e820_entries], entry, sizeof(struct e820_entry));
258 params->e820_entries++;
259 return 0;
260}
261
262static int memmap_entry_callback(struct resource *res, void *arg)
263{
264 struct crash_memmap_data *cmd = arg;
265 struct boot_params *params = cmd->params;
266 struct e820_entry ei;
267
268 ei.addr = res->start;
269 ei.size = resource_size(res);
270 ei.type = cmd->type;
271 add_e820_entry(params, &ei);
272
273 return 0;
274}
275
276static int memmap_exclude_ranges(struct kimage *image, struct crash_mem *cmem,
277 unsigned long long mstart,
278 unsigned long long mend)
279{
280 unsigned long start, end;
281
282 cmem->ranges[0].start = mstart;
283 cmem->ranges[0].end = mend;
284 cmem->nr_ranges = 1;
285
286 /* Exclude elf header region */
287 start = image->elf_load_addr;
288 end = start + image->elf_headers_sz - 1;
289 return crash_exclude_mem_range(cmem, start, end);
290}
291
292/* Prepare memory map for crash dump kernel */
293int crash_setup_memmap_entries(struct kimage *image, struct boot_params *params)
294{
295 int i, ret = 0;
296 unsigned long flags;
297 struct e820_entry ei;
298 struct crash_memmap_data cmd;
299 struct crash_mem *cmem;
300
301 cmem = vzalloc(struct_size(cmem, ranges, 1));
302 if (!cmem)
303 return -ENOMEM;
304
305 memset(&cmd, 0, sizeof(struct crash_memmap_data));
306 cmd.params = params;
307
308 /* Add the low 1M */
309 cmd.type = E820_TYPE_RAM;
310 flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
311 walk_iomem_res_desc(IORES_DESC_NONE, flags, 0, (1<<20)-1, &cmd,
312 memmap_entry_callback);
313
314 /* Add ACPI tables */
315 cmd.type = E820_TYPE_ACPI;
316 flags = IORESOURCE_MEM | IORESOURCE_BUSY;
317 walk_iomem_res_desc(IORES_DESC_ACPI_TABLES, flags, 0, -1, &cmd,
318 memmap_entry_callback);
319
320 /* Add ACPI Non-volatile Storage */
321 cmd.type = E820_TYPE_NVS;
322 walk_iomem_res_desc(IORES_DESC_ACPI_NV_STORAGE, flags, 0, -1, &cmd,
323 memmap_entry_callback);
324
325 /* Add e820 reserved ranges */
326 cmd.type = E820_TYPE_RESERVED;
327 flags = IORESOURCE_MEM;
328 walk_iomem_res_desc(IORES_DESC_RESERVED, flags, 0, -1, &cmd,
329 memmap_entry_callback);
330
331 /* Add crashk_low_res region */
332 if (crashk_low_res.end) {
333 ei.addr = crashk_low_res.start;
334 ei.size = resource_size(&crashk_low_res);
335 ei.type = E820_TYPE_RAM;
336 add_e820_entry(params, &ei);
337 }
338
339 /* Exclude some ranges from crashk_res and add rest to memmap */
340 ret = memmap_exclude_ranges(image, cmem, crashk_res.start, crashk_res.end);
341 if (ret)
342 goto out;
343
344 for (i = 0; i < cmem->nr_ranges; i++) {
345 ei.size = cmem->ranges[i].end - cmem->ranges[i].start + 1;
346
347 /* If entry is less than a page, skip it */
348 if (ei.size < PAGE_SIZE)
349 continue;
350 ei.addr = cmem->ranges[i].start;
351 ei.type = E820_TYPE_RAM;
352 add_e820_entry(params, &ei);
353 }
354
355out:
356 vfree(cmem);
357 return ret;
358}
359
360int crash_load_segments(struct kimage *image)
361{
362 int ret;
363 unsigned long pnum = 0;
364 struct kexec_buf kbuf = { .image = image, .buf_min = 0,
365 .buf_max = ULONG_MAX, .top_down = false };
366
367 /* Prepare elf headers and add a segment */
368 ret = prepare_elf_headers(&kbuf.buffer, &kbuf.bufsz, &pnum);
369 if (ret)
370 return ret;
371
372 image->elf_headers = kbuf.buffer;
373 image->elf_headers_sz = kbuf.bufsz;
374 kbuf.memsz = kbuf.bufsz;
375
376#ifdef CONFIG_CRASH_HOTPLUG
377 /*
378 * The elfcorehdr segment size accounts for VMCOREINFO, kernel_map,
379 * maximum CPUs and maximum memory ranges.
380 */
381 if (IS_ENABLED(CONFIG_MEMORY_HOTPLUG))
382 pnum = 2 + CONFIG_NR_CPUS_DEFAULT + CONFIG_CRASH_MAX_MEMORY_RANGES;
383 else
384 pnum += 2 + CONFIG_NR_CPUS_DEFAULT;
385
386 if (pnum < (unsigned long)PN_XNUM) {
387 kbuf.memsz = pnum * sizeof(Elf64_Phdr);
388 kbuf.memsz += sizeof(Elf64_Ehdr);
389
390 image->elfcorehdr_index = image->nr_segments;
391
392 /* Mark as usable to crash kernel, else crash kernel fails on boot */
393 image->elf_headers_sz = kbuf.memsz;
394 } else {
395 pr_err("number of Phdrs %lu exceeds max\n", pnum);
396 }
397#endif
398
399 kbuf.buf_align = ELF_CORE_HEADER_ALIGN;
400 kbuf.mem = KEXEC_BUF_MEM_UNKNOWN;
401 ret = kexec_add_buffer(&kbuf);
402 if (ret)
403 return ret;
404 image->elf_load_addr = kbuf.mem;
405 kexec_dprintk("Loaded ELF headers at 0x%lx bufsz=0x%lx memsz=0x%lx\n",
406 image->elf_load_addr, kbuf.bufsz, kbuf.memsz);
407
408 return ret;
409}
410#endif /* CONFIG_KEXEC_FILE */
411
412#ifdef CONFIG_CRASH_HOTPLUG
413
414#undef pr_fmt
415#define pr_fmt(fmt) "crash hp: " fmt
416
417int arch_crash_hotplug_support(struct kimage *image, unsigned long kexec_flags)
418{
419
420#ifdef CONFIG_KEXEC_FILE
421 if (image->file_mode)
422 return 1;
423#endif
424 /*
425 * Initially, crash hotplug support for kexec_load was added
426 * with the KEXEC_UPDATE_ELFCOREHDR flag. Later, this
427 * functionality was expanded to accommodate multiple kexec
428 * segment updates, leading to the introduction of the
429 * KEXEC_CRASH_HOTPLUG_SUPPORT kexec flag bit. Consequently,
430 * when the kexec tool sends either of these flags, it indicates
431 * that the required kexec segment (elfcorehdr) is excluded from
432 * the SHA calculation.
433 */
434 return (kexec_flags & KEXEC_UPDATE_ELFCOREHDR ||
435 kexec_flags & KEXEC_CRASH_HOTPLUG_SUPPORT);
436}
437
438unsigned int arch_crash_get_elfcorehdr_size(void)
439{
440 unsigned int sz;
441
442 /* kernel_map, VMCOREINFO and maximum CPUs */
443 sz = 2 + CONFIG_NR_CPUS_DEFAULT;
444 if (IS_ENABLED(CONFIG_MEMORY_HOTPLUG))
445 sz += CONFIG_CRASH_MAX_MEMORY_RANGES;
446 sz *= sizeof(Elf64_Phdr);
447 return sz;
448}
449
450/**
451 * arch_crash_handle_hotplug_event() - Handle hotplug elfcorehdr changes
452 * @image: a pointer to kexec_crash_image
453 * @arg: struct memory_notify handler for memory hotplug case and
454 * NULL for CPU hotplug case.
455 *
456 * Prepare the new elfcorehdr and replace the existing elfcorehdr.
457 */
458void arch_crash_handle_hotplug_event(struct kimage *image, void *arg)
459{
460 void *elfbuf = NULL, *old_elfcorehdr;
461 unsigned long nr_mem_ranges;
462 unsigned long mem, memsz;
463 unsigned long elfsz = 0;
464
465 /*
466 * As crash_prepare_elf64_headers() has already described all
467 * possible CPUs, there is no need to update the elfcorehdr
468 * for additional CPU changes.
469 */
470 if ((image->file_mode || image->elfcorehdr_updated) &&
471 ((image->hp_action == KEXEC_CRASH_HP_ADD_CPU) ||
472 (image->hp_action == KEXEC_CRASH_HP_REMOVE_CPU)))
473 return;
474
475 /*
476 * Create the new elfcorehdr reflecting the changes to CPU and/or
477 * memory resources.
478 */
479 if (prepare_elf_headers(&elfbuf, &elfsz, &nr_mem_ranges)) {
480 pr_err("unable to create new elfcorehdr");
481 goto out;
482 }
483
484 /*
485 * Obtain address and size of the elfcorehdr segment, and
486 * check it against the new elfcorehdr buffer.
487 */
488 mem = image->segment[image->elfcorehdr_index].mem;
489 memsz = image->segment[image->elfcorehdr_index].memsz;
490 if (elfsz > memsz) {
491 pr_err("update elfcorehdr elfsz %lu > memsz %lu",
492 elfsz, memsz);
493 goto out;
494 }
495
496 /*
497 * Copy new elfcorehdr over the old elfcorehdr at destination.
498 */
499 old_elfcorehdr = kmap_local_page(pfn_to_page(mem >> PAGE_SHIFT));
500 if (!old_elfcorehdr) {
501 pr_err("mapping elfcorehdr segment failed\n");
502 goto out;
503 }
504
505 /*
506 * Temporarily invalidate the crash image while the
507 * elfcorehdr is updated.
508 */
509 xchg(&kexec_crash_image, NULL);
510 memcpy_flushcache(old_elfcorehdr, elfbuf, elfsz);
511 xchg(&kexec_crash_image, image);
512 kunmap_local(old_elfcorehdr);
513 pr_debug("updated elfcorehdr\n");
514
515out:
516 vfree(elfbuf);
517}
518#endif
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 */