]>
Commit | Line | Data |
---|---|---|
5033cba0 | 1 | /* |
62a31a03 | 2 | * Architecture specific (i386/x86_64) functions for kexec based crash dumps. |
5033cba0 EB |
3 | * |
4 | * Created by: Hariprasad Nellitheertha (hari@in.ibm.com) | |
5 | * | |
6 | * Copyright (C) IBM Corporation, 2004. All rights reserved. | |
dd5f7260 VG |
7 | * Copyright (C) Red Hat Inc., 2014. All rights reserved. |
8 | * Authors: | |
9 | * Vivek Goyal <vgoyal@redhat.com> | |
5033cba0 EB |
10 | * |
11 | */ | |
12 | ||
dd5f7260 VG |
13 | #define pr_fmt(fmt) "kexec: " fmt |
14 | ||
5033cba0 EB |
15 | #include <linux/types.h> |
16 | #include <linux/kernel.h> | |
17 | #include <linux/smp.h> | |
5033cba0 EB |
18 | #include <linux/reboot.h> |
19 | #include <linux/kexec.h> | |
5033cba0 EB |
20 | #include <linux/delay.h> |
21 | #include <linux/elf.h> | |
22 | #include <linux/elfcore.h> | |
186f4360 | 23 | #include <linux/export.h> |
dd5f7260 | 24 | #include <linux/slab.h> |
d6472302 | 25 | #include <linux/vmalloc.h> |
5033cba0 EB |
26 | |
27 | #include <asm/processor.h> | |
28 | #include <asm/hardirq.h> | |
29 | #include <asm/nmi.h> | |
30 | #include <asm/hw_irq.h> | |
19842d67 | 31 | #include <asm/apic.h> |
5520b7e7 | 32 | #include <asm/e820/types.h> |
8643e28d | 33 | #include <asm/io_apic.h> |
0c1b2724 | 34 | #include <asm/hpet.h> |
1eeb66a1 | 35 | #include <linux/kdebug.h> |
96b89dc6 | 36 | #include <asm/cpu.h> |
ed23dc6f | 37 | #include <asm/reboot.h> |
2340b62f | 38 | #include <asm/virtext.h> |
da06a43d | 39 | #include <asm/intel_pt.h> |
8e294786 | 40 | |
dd5f7260 VG |
41 | /* Alignment required for elf header segment */ |
42 | #define ELF_CORE_HEADER_ALIGN 4096 | |
43 | ||
44 | /* This primarily represents number of split ranges due to exclusion */ | |
45 | #define CRASH_MAX_RANGES 16 | |
46 | ||
47 | struct crash_mem_range { | |
48 | u64 start, end; | |
49 | }; | |
50 | ||
51 | struct crash_mem { | |
52 | unsigned int nr_ranges; | |
53 | struct crash_mem_range ranges[CRASH_MAX_RANGES]; | |
54 | }; | |
55 | ||
56 | /* Misc data about ram ranges needed to prepare elf headers */ | |
57 | struct crash_elf_data { | |
58 | struct kimage *image; | |
59 | /* | |
60 | * Total number of ram ranges we have after various adjustments for | |
f296f263 | 61 | * crash reserved region, etc. |
dd5f7260 VG |
62 | */ |
63 | unsigned int max_nr_ranges; | |
dd5f7260 VG |
64 | |
65 | /* Pointer to elf header */ | |
66 | void *ehdr; | |
67 | /* Pointer to next phdr */ | |
68 | void *bufp; | |
69 | struct crash_mem mem; | |
70 | }; | |
71 | ||
72 | /* Used while preparing memory map entries for second kernel */ | |
73 | struct crash_memmap_data { | |
74 | struct boot_params *params; | |
75 | /* Type of memory */ | |
76 | unsigned int type; | |
77 | }; | |
78 | ||
f23d1f4a ZY |
79 | /* |
80 | * This is used to VMCLEAR all VMCSs loaded on the | |
81 | * processor. And when loading kvm_intel module, the | |
82 | * callback function pointer will be assigned. | |
83 | * | |
84 | * protected by rcu. | |
85 | */ | |
0ca0d818 | 86 | crash_vmclear_fn __rcu *crash_vmclear_loaded_vmcss = NULL; |
f23d1f4a | 87 | EXPORT_SYMBOL_GPL(crash_vmclear_loaded_vmcss); |
dd5f7260 | 88 | unsigned long crash_zero_bytes; |
f23d1f4a ZY |
89 | |
90 | static inline void cpu_crash_vmclear_loaded_vmcss(void) | |
91 | { | |
0ca0d818 | 92 | crash_vmclear_fn *do_vmclear_operation = NULL; |
f23d1f4a ZY |
93 | |
94 | rcu_read_lock(); | |
95 | do_vmclear_operation = rcu_dereference(crash_vmclear_loaded_vmcss); | |
96 | if (do_vmclear_operation) | |
97 | do_vmclear_operation(); | |
98 | rcu_read_unlock(); | |
99 | } | |
100 | ||
b2bbe71b EH |
101 | #if defined(CONFIG_SMP) && defined(CONFIG_X86_LOCAL_APIC) |
102 | ||
9c48f1c6 | 103 | static void kdump_nmi_callback(int cpu, struct pt_regs *regs) |
c4ac4263 | 104 | { |
1fb473d8 | 105 | #ifdef CONFIG_X86_32 |
4d55476c | 106 | struct pt_regs fixed_regs; |
a7d41820 | 107 | |
f39b6f0e | 108 | if (!user_mode(regs)) { |
a7d41820 EH |
109 | crash_fixup_ss_esp(&fixed_regs, regs); |
110 | regs = &fixed_regs; | |
111 | } | |
112 | #endif | |
113 | crash_save_cpu(regs, cpu); | |
114 | ||
f23d1f4a ZY |
115 | /* |
116 | * VMCLEAR VMCSs loaded on all cpus if needed. | |
117 | */ | |
118 | cpu_crash_vmclear_loaded_vmcss(); | |
119 | ||
2340b62f EH |
120 | /* Disable VMX or SVM if needed. |
121 | * | |
122 | * We need to disable virtualization on all CPUs. | |
123 | * Having VMX or SVM enabled on any CPU may break rebooting | |
124 | * after the kdump kernel has finished its task. | |
125 | */ | |
126 | cpu_emergency_vmxoff(); | |
127 | cpu_emergency_svm_disable(); | |
128 | ||
da06a43d TI |
129 | /* |
130 | * Disable Intel PT to stop its logging | |
131 | */ | |
132 | cpu_emergency_stop_pt(); | |
133 | ||
a7d41820 EH |
134 | disable_local_APIC(); |
135 | } | |
136 | ||
0ee59413 | 137 | void kdump_nmi_shootdown_cpus(void) |
d1e7b91c | 138 | { |
8e294786 | 139 | nmi_shootdown_cpus(kdump_nmi_callback); |
d1e7b91c | 140 | |
19842d67 | 141 | disable_local_APIC(); |
c4ac4263 | 142 | } |
d1e7b91c | 143 | |
0ee59413 HK |
144 | /* Override the weak function in kernel/panic.c */ |
145 | void crash_smp_send_stop(void) | |
146 | { | |
147 | static int cpus_stopped; | |
148 | ||
149 | if (cpus_stopped) | |
150 | return; | |
151 | ||
152 | if (smp_ops.crash_stop_other_cpus) | |
153 | smp_ops.crash_stop_other_cpus(); | |
154 | else | |
155 | smp_send_stop(); | |
156 | ||
157 | cpus_stopped = 1; | |
158 | } | |
159 | ||
c4ac4263 | 160 | #else |
0ee59413 | 161 | void crash_smp_send_stop(void) |
c4ac4263 EB |
162 | { |
163 | /* There are no cpus to shootdown */ | |
164 | } | |
165 | #endif | |
166 | ||
ed23dc6f | 167 | void native_machine_crash_shutdown(struct pt_regs *regs) |
5033cba0 EB |
168 | { |
169 | /* This function is only called after the system | |
f18190bd | 170 | * has panicked or is otherwise in a critical state. |
5033cba0 EB |
171 | * The minimum amount of code to allow a kexec'd kernel |
172 | * to run successfully needs to happen here. | |
173 | * | |
174 | * In practice this means shooting down the other cpus in | |
175 | * an SMP system. | |
176 | */ | |
c4ac4263 EB |
177 | /* The kernel is broken so disable interrupts */ |
178 | local_irq_disable(); | |
a3ea8ac8 | 179 | |
0ee59413 | 180 | crash_smp_send_stop(); |
2340b62f | 181 | |
f23d1f4a ZY |
182 | /* |
183 | * VMCLEAR VMCSs loaded on this cpu if needed. | |
184 | */ | |
185 | cpu_crash_vmclear_loaded_vmcss(); | |
186 | ||
2340b62f EH |
187 | /* Booting kdump kernel with VMX or SVM enabled won't work, |
188 | * because (among other limitations) we can't disable paging | |
189 | * with the virt flags. | |
190 | */ | |
191 | cpu_emergency_vmxoff(); | |
192 | cpu_emergency_svm_disable(); | |
193 | ||
da06a43d TI |
194 | /* |
195 | * Disable Intel PT to stop its logging | |
196 | */ | |
197 | cpu_emergency_stop_pt(); | |
198 | ||
17405453 YY |
199 | #ifdef CONFIG_X86_IO_APIC |
200 | /* Prevent crash_kexec() from deadlocking on ioapic_lock. */ | |
201 | ioapic_zap_locks(); | |
19842d67 | 202 | disable_IO_APIC(); |
0c1b2724 | 203 | #endif |
522e6646 | 204 | lapic_shutdown(); |
0c1b2724 OH |
205 | #ifdef CONFIG_HPET_TIMER |
206 | hpet_disable(); | |
19842d67 | 207 | #endif |
85916f81 | 208 | crash_save_cpu(regs, safe_smp_processor_id()); |
5033cba0 | 209 | } |
dd5f7260 | 210 | |
74ca317c | 211 | #ifdef CONFIG_KEXEC_FILE |
e3c41e37 | 212 | static int get_nr_ram_ranges_callback(u64 start, u64 end, void *arg) |
dd5f7260 | 213 | { |
e3c41e37 | 214 | unsigned int *nr_ranges = arg; |
dd5f7260 VG |
215 | |
216 | (*nr_ranges)++; | |
217 | return 0; | |
218 | } | |
219 | ||
dd5f7260 VG |
220 | |
221 | /* Gather all the required information to prepare elf headers for ram regions */ | |
222 | static void fill_up_crash_elf_data(struct crash_elf_data *ced, | |
223 | struct kimage *image) | |
224 | { | |
225 | unsigned int nr_ranges = 0; | |
226 | ||
227 | ced->image = image; | |
228 | ||
e3c41e37 | 229 | walk_system_ram_res(0, -1, &nr_ranges, |
dd5f7260 VG |
230 | get_nr_ram_ranges_callback); |
231 | ||
232 | ced->max_nr_ranges = nr_ranges; | |
233 | ||
dd5f7260 VG |
234 | /* Exclusion of crash region could split memory ranges */ |
235 | ced->max_nr_ranges++; | |
236 | ||
237 | /* If crashk_low_res is not 0, another range split possible */ | |
a2d6aa8f | 238 | if (crashk_low_res.end) |
dd5f7260 VG |
239 | ced->max_nr_ranges++; |
240 | } | |
241 | ||
242 | static int exclude_mem_range(struct crash_mem *mem, | |
243 | unsigned long long mstart, unsigned long long mend) | |
244 | { | |
245 | int i, j; | |
246 | unsigned long long start, end; | |
247 | struct crash_mem_range temp_range = {0, 0}; | |
248 | ||
249 | for (i = 0; i < mem->nr_ranges; i++) { | |
250 | start = mem->ranges[i].start; | |
251 | end = mem->ranges[i].end; | |
252 | ||
253 | if (mstart > end || mend < start) | |
254 | continue; | |
255 | ||
256 | /* Truncate any area outside of range */ | |
257 | if (mstart < start) | |
258 | mstart = start; | |
259 | if (mend > end) | |
260 | mend = end; | |
261 | ||
262 | /* Found completely overlapping range */ | |
263 | if (mstart == start && mend == end) { | |
264 | mem->ranges[i].start = 0; | |
265 | mem->ranges[i].end = 0; | |
266 | if (i < mem->nr_ranges - 1) { | |
267 | /* Shift rest of the ranges to left */ | |
268 | for (j = i; j < mem->nr_ranges - 1; j++) { | |
269 | mem->ranges[j].start = | |
270 | mem->ranges[j+1].start; | |
271 | mem->ranges[j].end = | |
272 | mem->ranges[j+1].end; | |
273 | } | |
274 | } | |
275 | mem->nr_ranges--; | |
276 | return 0; | |
277 | } | |
278 | ||
279 | if (mstart > start && mend < end) { | |
280 | /* Split original range */ | |
281 | mem->ranges[i].end = mstart - 1; | |
282 | temp_range.start = mend + 1; | |
283 | temp_range.end = end; | |
284 | } else if (mstart != start) | |
285 | mem->ranges[i].end = mstart - 1; | |
286 | else | |
287 | mem->ranges[i].start = mend + 1; | |
288 | break; | |
289 | } | |
290 | ||
291 | /* If a split happend, add the split to array */ | |
292 | if (!temp_range.end) | |
293 | return 0; | |
294 | ||
295 | /* Split happened */ | |
296 | if (i == CRASH_MAX_RANGES - 1) { | |
297 | pr_err("Too many crash ranges after split\n"); | |
298 | return -ENOMEM; | |
299 | } | |
300 | ||
301 | /* Location where new range should go */ | |
302 | j = i + 1; | |
303 | if (j < mem->nr_ranges) { | |
304 | /* Move over all ranges one slot towards the end */ | |
305 | for (i = mem->nr_ranges - 1; i >= j; i--) | |
306 | mem->ranges[i + 1] = mem->ranges[i]; | |
307 | } | |
308 | ||
309 | mem->ranges[j].start = temp_range.start; | |
310 | mem->ranges[j].end = temp_range.end; | |
311 | mem->nr_ranges++; | |
312 | return 0; | |
313 | } | |
314 | ||
315 | /* | |
316 | * Look for any unwanted ranges between mstart, mend and remove them. This | |
317 | * might lead to split and split ranges are put in ced->mem.ranges[] array | |
318 | */ | |
319 | static int elf_header_exclude_ranges(struct crash_elf_data *ced, | |
320 | unsigned long long mstart, unsigned long long mend) | |
321 | { | |
322 | struct crash_mem *cmem = &ced->mem; | |
323 | int ret = 0; | |
324 | ||
325 | memset(cmem->ranges, 0, sizeof(cmem->ranges)); | |
326 | ||
327 | cmem->ranges[0].start = mstart; | |
328 | cmem->ranges[0].end = mend; | |
329 | cmem->nr_ranges = 1; | |
330 | ||
331 | /* Exclude crashkernel region */ | |
332 | ret = exclude_mem_range(cmem, crashk_res.start, crashk_res.end); | |
333 | if (ret) | |
334 | return ret; | |
335 | ||
a2d6aa8f BH |
336 | if (crashk_low_res.end) { |
337 | ret = exclude_mem_range(cmem, crashk_low_res.start, crashk_low_res.end); | |
338 | if (ret) | |
339 | return ret; | |
340 | } | |
dd5f7260 | 341 | |
dd5f7260 VG |
342 | return ret; |
343 | } | |
344 | ||
345 | static int prepare_elf64_ram_headers_callback(u64 start, u64 end, void *arg) | |
346 | { | |
347 | struct crash_elf_data *ced = arg; | |
348 | Elf64_Ehdr *ehdr; | |
349 | Elf64_Phdr *phdr; | |
350 | unsigned long mstart, mend; | |
351 | struct kimage *image = ced->image; | |
352 | struct crash_mem *cmem; | |
353 | int ret, i; | |
354 | ||
355 | ehdr = ced->ehdr; | |
356 | ||
357 | /* Exclude unwanted mem ranges */ | |
358 | ret = elf_header_exclude_ranges(ced, start, end); | |
359 | if (ret) | |
360 | return ret; | |
361 | ||
362 | /* Go through all the ranges in ced->mem.ranges[] and prepare phdr */ | |
363 | cmem = &ced->mem; | |
364 | ||
365 | for (i = 0; i < cmem->nr_ranges; i++) { | |
366 | mstart = cmem->ranges[i].start; | |
367 | mend = cmem->ranges[i].end; | |
368 | ||
369 | phdr = ced->bufp; | |
370 | ced->bufp += sizeof(Elf64_Phdr); | |
371 | ||
372 | phdr->p_type = PT_LOAD; | |
373 | phdr->p_flags = PF_R|PF_W|PF_X; | |
374 | phdr->p_offset = mstart; | |
375 | ||
376 | /* | |
377 | * If a range matches backup region, adjust offset to backup | |
378 | * segment. | |
379 | */ | |
380 | if (mstart == image->arch.backup_src_start && | |
381 | (mend - mstart + 1) == image->arch.backup_src_sz) | |
382 | phdr->p_offset = image->arch.backup_load_addr; | |
383 | ||
384 | phdr->p_paddr = mstart; | |
385 | phdr->p_vaddr = (unsigned long long) __va(mstart); | |
386 | phdr->p_filesz = phdr->p_memsz = mend - mstart + 1; | |
387 | phdr->p_align = 0; | |
388 | ehdr->e_phnum++; | |
389 | 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", | |
390 | phdr, phdr->p_vaddr, phdr->p_paddr, phdr->p_filesz, | |
391 | ehdr->e_phnum, phdr->p_offset); | |
392 | } | |
393 | ||
394 | return ret; | |
395 | } | |
396 | ||
397 | static int prepare_elf64_headers(struct crash_elf_data *ced, | |
398 | void **addr, unsigned long *sz) | |
399 | { | |
400 | Elf64_Ehdr *ehdr; | |
401 | Elf64_Phdr *phdr; | |
402 | unsigned long nr_cpus = num_possible_cpus(), nr_phdr, elf_sz; | |
403 | unsigned char *buf, *bufp; | |
404 | unsigned int cpu; | |
405 | unsigned long long notes_addr; | |
406 | int ret; | |
407 | ||
408 | /* extra phdr for vmcoreinfo elf note */ | |
409 | nr_phdr = nr_cpus + 1; | |
410 | nr_phdr += ced->max_nr_ranges; | |
411 | ||
412 | /* | |
413 | * kexec-tools creates an extra PT_LOAD phdr for kernel text mapping | |
414 | * area on x86_64 (ffffffff80000000 - ffffffffa0000000). | |
415 | * I think this is required by tools like gdb. So same physical | |
416 | * memory will be mapped in two elf headers. One will contain kernel | |
417 | * text virtual addresses and other will have __va(physical) addresses. | |
418 | */ | |
419 | ||
420 | nr_phdr++; | |
421 | elf_sz = sizeof(Elf64_Ehdr) + nr_phdr * sizeof(Elf64_Phdr); | |
422 | elf_sz = ALIGN(elf_sz, ELF_CORE_HEADER_ALIGN); | |
423 | ||
424 | buf = vzalloc(elf_sz); | |
425 | if (!buf) | |
426 | return -ENOMEM; | |
427 | ||
428 | bufp = buf; | |
429 | ehdr = (Elf64_Ehdr *)bufp; | |
430 | bufp += sizeof(Elf64_Ehdr); | |
431 | memcpy(ehdr->e_ident, ELFMAG, SELFMAG); | |
432 | ehdr->e_ident[EI_CLASS] = ELFCLASS64; | |
433 | ehdr->e_ident[EI_DATA] = ELFDATA2LSB; | |
434 | ehdr->e_ident[EI_VERSION] = EV_CURRENT; | |
435 | ehdr->e_ident[EI_OSABI] = ELF_OSABI; | |
436 | memset(ehdr->e_ident + EI_PAD, 0, EI_NIDENT - EI_PAD); | |
437 | ehdr->e_type = ET_CORE; | |
438 | ehdr->e_machine = ELF_ARCH; | |
439 | ehdr->e_version = EV_CURRENT; | |
440 | ehdr->e_phoff = sizeof(Elf64_Ehdr); | |
441 | ehdr->e_ehsize = sizeof(Elf64_Ehdr); | |
442 | ehdr->e_phentsize = sizeof(Elf64_Phdr); | |
443 | ||
444 | /* Prepare one phdr of type PT_NOTE for each present cpu */ | |
445 | for_each_present_cpu(cpu) { | |
446 | phdr = (Elf64_Phdr *)bufp; | |
447 | bufp += sizeof(Elf64_Phdr); | |
448 | phdr->p_type = PT_NOTE; | |
449 | notes_addr = per_cpu_ptr_to_phys(per_cpu_ptr(crash_notes, cpu)); | |
450 | phdr->p_offset = phdr->p_paddr = notes_addr; | |
451 | phdr->p_filesz = phdr->p_memsz = sizeof(note_buf_t); | |
452 | (ehdr->e_phnum)++; | |
453 | } | |
454 | ||
455 | /* Prepare one PT_NOTE header for vmcoreinfo */ | |
456 | phdr = (Elf64_Phdr *)bufp; | |
457 | bufp += sizeof(Elf64_Phdr); | |
458 | phdr->p_type = PT_NOTE; | |
459 | phdr->p_offset = phdr->p_paddr = paddr_vmcoreinfo_note(); | |
203e9e41 | 460 | phdr->p_filesz = phdr->p_memsz = VMCOREINFO_NOTE_SIZE; |
dd5f7260 VG |
461 | (ehdr->e_phnum)++; |
462 | ||
463 | #ifdef CONFIG_X86_64 | |
464 | /* Prepare PT_LOAD type program header for kernel text region */ | |
465 | phdr = (Elf64_Phdr *)bufp; | |
466 | bufp += sizeof(Elf64_Phdr); | |
467 | phdr->p_type = PT_LOAD; | |
468 | phdr->p_flags = PF_R|PF_W|PF_X; | |
469 | phdr->p_vaddr = (Elf64_Addr)_text; | |
470 | phdr->p_filesz = phdr->p_memsz = _end - _text; | |
471 | phdr->p_offset = phdr->p_paddr = __pa_symbol(_text); | |
472 | (ehdr->e_phnum)++; | |
473 | #endif | |
474 | ||
475 | /* Prepare PT_LOAD headers for system ram chunks. */ | |
476 | ced->ehdr = ehdr; | |
477 | ced->bufp = bufp; | |
478 | ret = walk_system_ram_res(0, -1, ced, | |
479 | prepare_elf64_ram_headers_callback); | |
480 | if (ret < 0) | |
481 | return ret; | |
482 | ||
483 | *addr = buf; | |
484 | *sz = elf_sz; | |
485 | return 0; | |
486 | } | |
487 | ||
488 | /* Prepare elf headers. Return addr and size */ | |
489 | static int prepare_elf_headers(struct kimage *image, void **addr, | |
490 | unsigned long *sz) | |
491 | { | |
492 | struct crash_elf_data *ced; | |
493 | int ret; | |
494 | ||
495 | ced = kzalloc(sizeof(*ced), GFP_KERNEL); | |
496 | if (!ced) | |
497 | return -ENOMEM; | |
498 | ||
499 | fill_up_crash_elf_data(ced, image); | |
500 | ||
501 | /* By default prepare 64bit headers */ | |
502 | ret = prepare_elf64_headers(ced, addr, sz); | |
503 | kfree(ced); | |
504 | return ret; | |
505 | } | |
506 | ||
8ec67d97 | 507 | static int add_e820_entry(struct boot_params *params, struct e820_entry *entry) |
dd5f7260 VG |
508 | { |
509 | unsigned int nr_e820_entries; | |
510 | ||
511 | nr_e820_entries = params->e820_entries; | |
08b46d5d | 512 | if (nr_e820_entries >= E820_MAX_ENTRIES_ZEROPAGE) |
dd5f7260 VG |
513 | return 1; |
514 | ||
61a50101 | 515 | memcpy(¶ms->e820_table[nr_e820_entries], entry, |
8ec67d97 | 516 | sizeof(struct e820_entry)); |
dd5f7260 VG |
517 | params->e820_entries++; |
518 | return 0; | |
519 | } | |
520 | ||
521 | static int memmap_entry_callback(u64 start, u64 end, void *arg) | |
522 | { | |
523 | struct crash_memmap_data *cmd = arg; | |
524 | struct boot_params *params = cmd->params; | |
8ec67d97 | 525 | struct e820_entry ei; |
dd5f7260 VG |
526 | |
527 | ei.addr = start; | |
528 | ei.size = end - start + 1; | |
529 | ei.type = cmd->type; | |
530 | add_e820_entry(params, &ei); | |
531 | ||
532 | return 0; | |
533 | } | |
534 | ||
535 | static int memmap_exclude_ranges(struct kimage *image, struct crash_mem *cmem, | |
536 | unsigned long long mstart, | |
537 | unsigned long long mend) | |
538 | { | |
539 | unsigned long start, end; | |
540 | int ret = 0; | |
541 | ||
542 | cmem->ranges[0].start = mstart; | |
543 | cmem->ranges[0].end = mend; | |
544 | cmem->nr_ranges = 1; | |
545 | ||
546 | /* Exclude Backup region */ | |
547 | start = image->arch.backup_load_addr; | |
548 | end = start + image->arch.backup_src_sz - 1; | |
549 | ret = exclude_mem_range(cmem, start, end); | |
550 | if (ret) | |
551 | return ret; | |
552 | ||
553 | /* Exclude elf header region */ | |
554 | start = image->arch.elf_load_addr; | |
555 | end = start + image->arch.elf_headers_sz - 1; | |
556 | return exclude_mem_range(cmem, start, end); | |
557 | } | |
558 | ||
559 | /* Prepare memory map for crash dump kernel */ | |
560 | int crash_setup_memmap_entries(struct kimage *image, struct boot_params *params) | |
561 | { | |
562 | int i, ret = 0; | |
563 | unsigned long flags; | |
8ec67d97 | 564 | struct e820_entry ei; |
dd5f7260 VG |
565 | struct crash_memmap_data cmd; |
566 | struct crash_mem *cmem; | |
567 | ||
568 | cmem = vzalloc(sizeof(struct crash_mem)); | |
569 | if (!cmem) | |
570 | return -ENOMEM; | |
571 | ||
572 | memset(&cmd, 0, sizeof(struct crash_memmap_data)); | |
573 | cmd.params = params; | |
574 | ||
575 | /* Add first 640K segment */ | |
576 | ei.addr = image->arch.backup_src_start; | |
577 | ei.size = image->arch.backup_src_sz; | |
09821ff1 | 578 | ei.type = E820_TYPE_RAM; |
dd5f7260 VG |
579 | add_e820_entry(params, &ei); |
580 | ||
581 | /* Add ACPI tables */ | |
09821ff1 | 582 | cmd.type = E820_TYPE_ACPI; |
dd5f7260 | 583 | flags = IORESOURCE_MEM | IORESOURCE_BUSY; |
f0f4711a | 584 | walk_iomem_res_desc(IORES_DESC_ACPI_TABLES, flags, 0, -1, &cmd, |
dd5f7260 VG |
585 | memmap_entry_callback); |
586 | ||
587 | /* Add ACPI Non-volatile Storage */ | |
09821ff1 | 588 | cmd.type = E820_TYPE_NVS; |
f0f4711a | 589 | walk_iomem_res_desc(IORES_DESC_ACPI_NV_STORAGE, flags, 0, -1, &cmd, |
dd5f7260 VG |
590 | memmap_entry_callback); |
591 | ||
592 | /* Add crashk_low_res region */ | |
593 | if (crashk_low_res.end) { | |
594 | ei.addr = crashk_low_res.start; | |
595 | ei.size = crashk_low_res.end - crashk_low_res.start + 1; | |
09821ff1 | 596 | ei.type = E820_TYPE_RAM; |
dd5f7260 VG |
597 | add_e820_entry(params, &ei); |
598 | } | |
599 | ||
600 | /* Exclude some ranges from crashk_res and add rest to memmap */ | |
601 | ret = memmap_exclude_ranges(image, cmem, crashk_res.start, | |
602 | crashk_res.end); | |
603 | if (ret) | |
604 | goto out; | |
605 | ||
606 | for (i = 0; i < cmem->nr_ranges; i++) { | |
607 | ei.size = cmem->ranges[i].end - cmem->ranges[i].start + 1; | |
608 | ||
609 | /* If entry is less than a page, skip it */ | |
610 | if (ei.size < PAGE_SIZE) | |
611 | continue; | |
612 | ei.addr = cmem->ranges[i].start; | |
09821ff1 | 613 | ei.type = E820_TYPE_RAM; |
dd5f7260 VG |
614 | add_e820_entry(params, &ei); |
615 | } | |
616 | ||
617 | out: | |
618 | vfree(cmem); | |
619 | return ret; | |
620 | } | |
621 | ||
622 | static int determine_backup_region(u64 start, u64 end, void *arg) | |
623 | { | |
624 | struct kimage *image = arg; | |
625 | ||
626 | image->arch.backup_src_start = start; | |
627 | image->arch.backup_src_sz = end - start + 1; | |
628 | ||
629 | /* Expecting only one range for backup region */ | |
630 | return 1; | |
631 | } | |
632 | ||
633 | int crash_load_segments(struct kimage *image) | |
634 | { | |
dd5f7260 | 635 | int ret; |
ec2b9bfa TJB |
636 | struct kexec_buf kbuf = { .image = image, .buf_min = 0, |
637 | .buf_max = ULONG_MAX, .top_down = false }; | |
dd5f7260 VG |
638 | |
639 | /* | |
640 | * Determine and load a segment for backup area. First 640K RAM | |
641 | * region is backup source | |
642 | */ | |
643 | ||
644 | ret = walk_system_ram_res(KEXEC_BACKUP_SRC_START, KEXEC_BACKUP_SRC_END, | |
645 | image, determine_backup_region); | |
646 | ||
647 | /* Zero or postive return values are ok */ | |
648 | if (ret < 0) | |
649 | return ret; | |
650 | ||
dd5f7260 | 651 | /* Add backup segment. */ |
ec2b9bfa TJB |
652 | if (image->arch.backup_src_sz) { |
653 | kbuf.buffer = &crash_zero_bytes; | |
654 | kbuf.bufsz = sizeof(crash_zero_bytes); | |
655 | kbuf.memsz = image->arch.backup_src_sz; | |
656 | kbuf.buf_align = PAGE_SIZE; | |
dd5f7260 VG |
657 | /* |
658 | * Ideally there is no source for backup segment. This is | |
659 | * copied in purgatory after crash. Just add a zero filled | |
660 | * segment for now to make sure checksum logic works fine. | |
661 | */ | |
ec2b9bfa | 662 | ret = kexec_add_buffer(&kbuf); |
dd5f7260 VG |
663 | if (ret) |
664 | return ret; | |
ec2b9bfa | 665 | image->arch.backup_load_addr = kbuf.mem; |
dd5f7260 | 666 | pr_debug("Loaded backup region at 0x%lx backup_start=0x%lx memsz=0x%lx\n", |
ec2b9bfa TJB |
667 | image->arch.backup_load_addr, |
668 | image->arch.backup_src_start, kbuf.memsz); | |
dd5f7260 VG |
669 | } |
670 | ||
671 | /* Prepare elf headers and add a segment */ | |
ec2b9bfa | 672 | ret = prepare_elf_headers(image, &kbuf.buffer, &kbuf.bufsz); |
dd5f7260 VG |
673 | if (ret) |
674 | return ret; | |
675 | ||
ec2b9bfa TJB |
676 | image->arch.elf_headers = kbuf.buffer; |
677 | image->arch.elf_headers_sz = kbuf.bufsz; | |
dd5f7260 | 678 | |
ec2b9bfa TJB |
679 | kbuf.memsz = kbuf.bufsz; |
680 | kbuf.buf_align = ELF_CORE_HEADER_ALIGN; | |
681 | ret = kexec_add_buffer(&kbuf); | |
dd5f7260 VG |
682 | if (ret) { |
683 | vfree((void *)image->arch.elf_headers); | |
684 | return ret; | |
685 | } | |
ec2b9bfa | 686 | image->arch.elf_load_addr = kbuf.mem; |
dd5f7260 | 687 | pr_debug("Loaded ELF headers at 0x%lx bufsz=0x%lx memsz=0x%lx\n", |
ec2b9bfa | 688 | image->arch.elf_load_addr, kbuf.bufsz, kbuf.bufsz); |
dd5f7260 VG |
689 | |
690 | return ret; | |
691 | } | |
74ca317c | 692 | #endif /* CONFIG_KEXEC_FILE */ |