xref: /linux/arch/s390/kernel/crash_dump.c (revision 908fc4c2)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * S390 kdump implementation
4  *
5  * Copyright IBM Corp. 2011
6  * Author(s): Michael Holzheu <holzheu@linux.vnet.ibm.com>
7  */
8 
9 #include <linux/crash_dump.h>
10 #include <asm/lowcore.h>
11 #include <linux/kernel.h>
12 #include <linux/init.h>
13 #include <linux/mm.h>
14 #include <linux/gfp.h>
15 #include <linux/slab.h>
16 #include <linux/memblock.h>
17 #include <linux/elf.h>
18 #include <linux/uio.h>
19 #include <asm/asm-offsets.h>
20 #include <asm/os_info.h>
21 #include <asm/elf.h>
22 #include <asm/ipl.h>
23 #include <asm/sclp.h>
24 
25 #define PTR_ADD(x, y) (((char *) (x)) + ((unsigned long) (y)))
26 #define PTR_SUB(x, y) (((char *) (x)) - ((unsigned long) (y)))
27 #define PTR_DIFF(x, y) ((unsigned long)(((char *) (x)) - ((unsigned long) (y))))
28 
29 static struct memblock_region oldmem_region;
30 
31 static struct memblock_type oldmem_type = {
32 	.cnt = 1,
33 	.max = 1,
34 	.total_size = 0,
35 	.regions = &oldmem_region,
36 	.name = "oldmem",
37 };
38 
39 struct save_area {
40 	struct list_head list;
41 	u64 psw[2];
42 	u64 ctrs[16];
43 	u64 gprs[16];
44 	u32 acrs[16];
45 	u64 fprs[16];
46 	u32 fpc;
47 	u32 prefix;
48 	u64 todpreg;
49 	u64 timer;
50 	u64 todcmp;
51 	u64 vxrs_low[16];
52 	__vector128 vxrs_high[16];
53 };
54 
55 static LIST_HEAD(dump_save_areas);
56 
57 /*
58  * Allocate a save area
59  */
60 struct save_area * __init save_area_alloc(bool is_boot_cpu)
61 {
62 	struct save_area *sa;
63 
64 	sa = memblock_alloc(sizeof(*sa), 8);
65 	if (!sa)
66 		panic("Failed to allocate save area\n");
67 
68 	if (is_boot_cpu)
69 		list_add(&sa->list, &dump_save_areas);
70 	else
71 		list_add_tail(&sa->list, &dump_save_areas);
72 	return sa;
73 }
74 
75 /*
76  * Return the address of the save area for the boot CPU
77  */
78 struct save_area * __init save_area_boot_cpu(void)
79 {
80 	return list_first_entry_or_null(&dump_save_areas, struct save_area, list);
81 }
82 
83 /*
84  * Copy CPU registers into the save area
85  */
86 void __init save_area_add_regs(struct save_area *sa, void *regs)
87 {
88 	struct lowcore *lc;
89 
90 	lc = (struct lowcore *)(regs - __LC_FPREGS_SAVE_AREA);
91 	memcpy(&sa->psw, &lc->psw_save_area, sizeof(sa->psw));
92 	memcpy(&sa->ctrs, &lc->cregs_save_area, sizeof(sa->ctrs));
93 	memcpy(&sa->gprs, &lc->gpregs_save_area, sizeof(sa->gprs));
94 	memcpy(&sa->acrs, &lc->access_regs_save_area, sizeof(sa->acrs));
95 	memcpy(&sa->fprs, &lc->floating_pt_save_area, sizeof(sa->fprs));
96 	memcpy(&sa->fpc, &lc->fpt_creg_save_area, sizeof(sa->fpc));
97 	memcpy(&sa->prefix, &lc->prefixreg_save_area, sizeof(sa->prefix));
98 	memcpy(&sa->todpreg, &lc->tod_progreg_save_area, sizeof(sa->todpreg));
99 	memcpy(&sa->timer, &lc->cpu_timer_save_area, sizeof(sa->timer));
100 	memcpy(&sa->todcmp, &lc->clock_comp_save_area, sizeof(sa->todcmp));
101 }
102 
103 /*
104  * Copy vector registers into the save area
105  */
106 void __init save_area_add_vxrs(struct save_area *sa, __vector128 *vxrs)
107 {
108 	int i;
109 
110 	/* Copy lower halves of vector registers 0-15 */
111 	for (i = 0; i < 16; i++)
112 		memcpy(&sa->vxrs_low[i], &vxrs[i].u[2], 8);
113 	/* Copy vector registers 16-31 */
114 	memcpy(sa->vxrs_high, vxrs + 16, 16 * sizeof(__vector128));
115 }
116 
117 /*
118  * Return physical address for virtual address
119  */
120 static inline void *load_real_addr(void *addr)
121 {
122 	unsigned long real_addr;
123 
124 	asm volatile(
125 		   "	lra     %0,0(%1)\n"
126 		   "	jz	0f\n"
127 		   "	la	%0,0\n"
128 		   "0:"
129 		   : "=a" (real_addr) : "a" (addr) : "cc");
130 	return (void *)real_addr;
131 }
132 
133 /*
134  * Copy memory of the old, dumped system to a kernel space virtual address
135  */
136 int copy_oldmem_kernel(void *dst, unsigned long src, size_t count)
137 {
138 	unsigned long len;
139 	void *ra;
140 	int rc;
141 
142 	while (count) {
143 		if (!oldmem_data.start && src < sclp.hsa_size) {
144 			/* Copy from zfcp/nvme dump HSA area */
145 			len = min(count, sclp.hsa_size - src);
146 			rc = memcpy_hsa_kernel(dst, src, len);
147 			if (rc)
148 				return rc;
149 		} else {
150 			/* Check for swapped kdump oldmem areas */
151 			if (oldmem_data.start && src - oldmem_data.start < oldmem_data.size) {
152 				src -= oldmem_data.start;
153 				len = min(count, oldmem_data.size - src);
154 			} else if (oldmem_data.start && src < oldmem_data.size) {
155 				len = min(count, oldmem_data.size - src);
156 				src += oldmem_data.start;
157 			} else {
158 				len = count;
159 			}
160 			if (is_vmalloc_or_module_addr(dst)) {
161 				ra = load_real_addr(dst);
162 				len = min(PAGE_SIZE - offset_in_page(ra), len);
163 			} else {
164 				ra = dst;
165 			}
166 			if (memcpy_real(ra, src, len))
167 				return -EFAULT;
168 		}
169 		dst += len;
170 		src += len;
171 		count -= len;
172 	}
173 	return 0;
174 }
175 
176 /*
177  * Copy memory of the old, dumped system to a user space virtual address
178  */
179 static int copy_oldmem_user(void __user *dst, unsigned long src, size_t count)
180 {
181 	unsigned long len;
182 	int rc;
183 
184 	while (count) {
185 		if (!oldmem_data.start && src < sclp.hsa_size) {
186 			/* Copy from zfcp/nvme dump HSA area */
187 			len = min(count, sclp.hsa_size - src);
188 			rc = memcpy_hsa_user(dst, src, len);
189 			if (rc)
190 				return rc;
191 		} else {
192 			/* Check for swapped kdump oldmem areas */
193 			if (oldmem_data.start && src - oldmem_data.start < oldmem_data.size) {
194 				src -= oldmem_data.start;
195 				len = min(count, oldmem_data.size - src);
196 			} else if (oldmem_data.start && src < oldmem_data.size) {
197 				len = min(count, oldmem_data.size - src);
198 				src += oldmem_data.start;
199 			} else {
200 				len = count;
201 			}
202 			rc = copy_to_user_real(dst, src, count);
203 			if (rc)
204 				return rc;
205 		}
206 		dst += len;
207 		src += len;
208 		count -= len;
209 	}
210 	return 0;
211 }
212 
213 /*
214  * Copy one page from "oldmem"
215  */
216 ssize_t copy_oldmem_page(struct iov_iter *iter, unsigned long pfn, size_t csize,
217 			 unsigned long offset)
218 {
219 	unsigned long src;
220 	int rc;
221 
222 	if (!(iter_is_iovec(iter) || iov_iter_is_kvec(iter)))
223 		return -EINVAL;
224 	/* Multi-segment iterators are not supported */
225 	if (iter->nr_segs > 1)
226 		return -EINVAL;
227 	if (!csize)
228 		return 0;
229 	src = pfn_to_phys(pfn) + offset;
230 
231 	/* XXX: pass the iov_iter down to a common function */
232 	if (iter_is_iovec(iter))
233 		rc = copy_oldmem_user(iter->iov->iov_base, src, csize);
234 	else
235 		rc = copy_oldmem_kernel(iter->kvec->iov_base, src, csize);
236 	if (rc < 0)
237 		return rc;
238 	iov_iter_advance(iter, csize);
239 	return csize;
240 }
241 
242 /*
243  * Remap "oldmem" for kdump
244  *
245  * For the kdump reserved memory this functions performs a swap operation:
246  * [0 - OLDMEM_SIZE] is mapped to [OLDMEM_BASE - OLDMEM_BASE + OLDMEM_SIZE]
247  */
248 static int remap_oldmem_pfn_range_kdump(struct vm_area_struct *vma,
249 					unsigned long from, unsigned long pfn,
250 					unsigned long size, pgprot_t prot)
251 {
252 	unsigned long size_old;
253 	int rc;
254 
255 	if (pfn < oldmem_data.size >> PAGE_SHIFT) {
256 		size_old = min(size, oldmem_data.size - (pfn << PAGE_SHIFT));
257 		rc = remap_pfn_range(vma, from,
258 				     pfn + (oldmem_data.start >> PAGE_SHIFT),
259 				     size_old, prot);
260 		if (rc || size == size_old)
261 			return rc;
262 		size -= size_old;
263 		from += size_old;
264 		pfn += size_old >> PAGE_SHIFT;
265 	}
266 	return remap_pfn_range(vma, from, pfn, size, prot);
267 }
268 
269 /*
270  * Remap "oldmem" for zfcp/nvme dump
271  *
272  * We only map available memory above HSA size. Memory below HSA size
273  * is read on demand using the copy_oldmem_page() function.
274  */
275 static int remap_oldmem_pfn_range_zfcpdump(struct vm_area_struct *vma,
276 					   unsigned long from,
277 					   unsigned long pfn,
278 					   unsigned long size, pgprot_t prot)
279 {
280 	unsigned long hsa_end = sclp.hsa_size;
281 	unsigned long size_hsa;
282 
283 	if (pfn < hsa_end >> PAGE_SHIFT) {
284 		size_hsa = min(size, hsa_end - (pfn << PAGE_SHIFT));
285 		if (size == size_hsa)
286 			return 0;
287 		size -= size_hsa;
288 		from += size_hsa;
289 		pfn += size_hsa >> PAGE_SHIFT;
290 	}
291 	return remap_pfn_range(vma, from, pfn, size, prot);
292 }
293 
294 /*
295  * Remap "oldmem" for kdump or zfcp/nvme dump
296  */
297 int remap_oldmem_pfn_range(struct vm_area_struct *vma, unsigned long from,
298 			   unsigned long pfn, unsigned long size, pgprot_t prot)
299 {
300 	if (oldmem_data.start)
301 		return remap_oldmem_pfn_range_kdump(vma, from, pfn, size, prot);
302 	else
303 		return remap_oldmem_pfn_range_zfcpdump(vma, from, pfn, size,
304 						       prot);
305 }
306 
307 static const char *nt_name(Elf64_Word type)
308 {
309 	const char *name = "LINUX";
310 
311 	if (type == NT_PRPSINFO || type == NT_PRSTATUS || type == NT_PRFPREG)
312 		name = KEXEC_CORE_NOTE_NAME;
313 	return name;
314 }
315 
316 /*
317  * Initialize ELF note
318  */
319 static void *nt_init_name(void *buf, Elf64_Word type, void *desc, int d_len,
320 			  const char *name)
321 {
322 	Elf64_Nhdr *note;
323 	u64 len;
324 
325 	note = (Elf64_Nhdr *)buf;
326 	note->n_namesz = strlen(name) + 1;
327 	note->n_descsz = d_len;
328 	note->n_type = type;
329 	len = sizeof(Elf64_Nhdr);
330 
331 	memcpy(buf + len, name, note->n_namesz);
332 	len = roundup(len + note->n_namesz, 4);
333 
334 	memcpy(buf + len, desc, note->n_descsz);
335 	len = roundup(len + note->n_descsz, 4);
336 
337 	return PTR_ADD(buf, len);
338 }
339 
340 static inline void *nt_init(void *buf, Elf64_Word type, void *desc, int d_len)
341 {
342 	return nt_init_name(buf, type, desc, d_len, nt_name(type));
343 }
344 
345 /*
346  * Calculate the size of ELF note
347  */
348 static size_t nt_size_name(int d_len, const char *name)
349 {
350 	size_t size;
351 
352 	size = sizeof(Elf64_Nhdr);
353 	size += roundup(strlen(name) + 1, 4);
354 	size += roundup(d_len, 4);
355 
356 	return size;
357 }
358 
359 static inline size_t nt_size(Elf64_Word type, int d_len)
360 {
361 	return nt_size_name(d_len, nt_name(type));
362 }
363 
364 /*
365  * Fill ELF notes for one CPU with save area registers
366  */
367 static void *fill_cpu_elf_notes(void *ptr, int cpu, struct save_area *sa)
368 {
369 	struct elf_prstatus nt_prstatus;
370 	elf_fpregset_t nt_fpregset;
371 
372 	/* Prepare prstatus note */
373 	memset(&nt_prstatus, 0, sizeof(nt_prstatus));
374 	memcpy(&nt_prstatus.pr_reg.gprs, sa->gprs, sizeof(sa->gprs));
375 	memcpy(&nt_prstatus.pr_reg.psw, sa->psw, sizeof(sa->psw));
376 	memcpy(&nt_prstatus.pr_reg.acrs, sa->acrs, sizeof(sa->acrs));
377 	nt_prstatus.common.pr_pid = cpu;
378 	/* Prepare fpregset (floating point) note */
379 	memset(&nt_fpregset, 0, sizeof(nt_fpregset));
380 	memcpy(&nt_fpregset.fpc, &sa->fpc, sizeof(sa->fpc));
381 	memcpy(&nt_fpregset.fprs, &sa->fprs, sizeof(sa->fprs));
382 	/* Create ELF notes for the CPU */
383 	ptr = nt_init(ptr, NT_PRSTATUS, &nt_prstatus, sizeof(nt_prstatus));
384 	ptr = nt_init(ptr, NT_PRFPREG, &nt_fpregset, sizeof(nt_fpregset));
385 	ptr = nt_init(ptr, NT_S390_TIMER, &sa->timer, sizeof(sa->timer));
386 	ptr = nt_init(ptr, NT_S390_TODCMP, &sa->todcmp, sizeof(sa->todcmp));
387 	ptr = nt_init(ptr, NT_S390_TODPREG, &sa->todpreg, sizeof(sa->todpreg));
388 	ptr = nt_init(ptr, NT_S390_CTRS, &sa->ctrs, sizeof(sa->ctrs));
389 	ptr = nt_init(ptr, NT_S390_PREFIX, &sa->prefix, sizeof(sa->prefix));
390 	if (MACHINE_HAS_VX) {
391 		ptr = nt_init(ptr, NT_S390_VXRS_HIGH,
392 			      &sa->vxrs_high, sizeof(sa->vxrs_high));
393 		ptr = nt_init(ptr, NT_S390_VXRS_LOW,
394 			      &sa->vxrs_low, sizeof(sa->vxrs_low));
395 	}
396 	return ptr;
397 }
398 
399 /*
400  * Calculate size of ELF notes per cpu
401  */
402 static size_t get_cpu_elf_notes_size(void)
403 {
404 	struct save_area *sa = NULL;
405 	size_t size;
406 
407 	size =	nt_size(NT_PRSTATUS, sizeof(struct elf_prstatus));
408 	size +=  nt_size(NT_PRFPREG, sizeof(elf_fpregset_t));
409 	size +=  nt_size(NT_S390_TIMER, sizeof(sa->timer));
410 	size +=  nt_size(NT_S390_TODCMP, sizeof(sa->todcmp));
411 	size +=  nt_size(NT_S390_TODPREG, sizeof(sa->todpreg));
412 	size +=  nt_size(NT_S390_CTRS, sizeof(sa->ctrs));
413 	size +=  nt_size(NT_S390_PREFIX, sizeof(sa->prefix));
414 	if (MACHINE_HAS_VX) {
415 		size += nt_size(NT_S390_VXRS_HIGH, sizeof(sa->vxrs_high));
416 		size += nt_size(NT_S390_VXRS_LOW, sizeof(sa->vxrs_low));
417 	}
418 
419 	return size;
420 }
421 
422 /*
423  * Initialize prpsinfo note (new kernel)
424  */
425 static void *nt_prpsinfo(void *ptr)
426 {
427 	struct elf_prpsinfo prpsinfo;
428 
429 	memset(&prpsinfo, 0, sizeof(prpsinfo));
430 	prpsinfo.pr_sname = 'R';
431 	strcpy(prpsinfo.pr_fname, "vmlinux");
432 	return nt_init(ptr, NT_PRPSINFO, &prpsinfo, sizeof(prpsinfo));
433 }
434 
435 /*
436  * Get vmcoreinfo using lowcore->vmcore_info (new kernel)
437  */
438 static void *get_vmcoreinfo_old(unsigned long *size)
439 {
440 	char nt_name[11], *vmcoreinfo;
441 	unsigned long addr;
442 	Elf64_Nhdr note;
443 
444 	if (copy_oldmem_kernel(&addr, __LC_VMCORE_INFO, sizeof(addr)))
445 		return NULL;
446 	memset(nt_name, 0, sizeof(nt_name));
447 	if (copy_oldmem_kernel(&note, addr, sizeof(note)))
448 		return NULL;
449 	if (copy_oldmem_kernel(nt_name, addr + sizeof(note),
450 			       sizeof(nt_name) - 1))
451 		return NULL;
452 	if (strcmp(nt_name, VMCOREINFO_NOTE_NAME) != 0)
453 		return NULL;
454 	vmcoreinfo = kzalloc(note.n_descsz, GFP_KERNEL);
455 	if (!vmcoreinfo)
456 		return NULL;
457 	if (copy_oldmem_kernel(vmcoreinfo, addr + 24, note.n_descsz)) {
458 		kfree(vmcoreinfo);
459 		return NULL;
460 	}
461 	*size = note.n_descsz;
462 	return vmcoreinfo;
463 }
464 
465 /*
466  * Initialize vmcoreinfo note (new kernel)
467  */
468 static void *nt_vmcoreinfo(void *ptr)
469 {
470 	const char *name = VMCOREINFO_NOTE_NAME;
471 	unsigned long size;
472 	void *vmcoreinfo;
473 
474 	vmcoreinfo = os_info_old_entry(OS_INFO_VMCOREINFO, &size);
475 	if (vmcoreinfo)
476 		return nt_init_name(ptr, 0, vmcoreinfo, size, name);
477 
478 	vmcoreinfo = get_vmcoreinfo_old(&size);
479 	if (!vmcoreinfo)
480 		return ptr;
481 	ptr = nt_init_name(ptr, 0, vmcoreinfo, size, name);
482 	kfree(vmcoreinfo);
483 	return ptr;
484 }
485 
486 static size_t nt_vmcoreinfo_size(void)
487 {
488 	const char *name = VMCOREINFO_NOTE_NAME;
489 	unsigned long size;
490 	void *vmcoreinfo;
491 
492 	vmcoreinfo = os_info_old_entry(OS_INFO_VMCOREINFO, &size);
493 	if (vmcoreinfo)
494 		return nt_size_name(size, name);
495 
496 	vmcoreinfo = get_vmcoreinfo_old(&size);
497 	if (!vmcoreinfo)
498 		return 0;
499 
500 	kfree(vmcoreinfo);
501 	return nt_size_name(size, name);
502 }
503 
504 /*
505  * Initialize final note (needed for /proc/vmcore code)
506  */
507 static void *nt_final(void *ptr)
508 {
509 	Elf64_Nhdr *note;
510 
511 	note = (Elf64_Nhdr *) ptr;
512 	note->n_namesz = 0;
513 	note->n_descsz = 0;
514 	note->n_type = 0;
515 	return PTR_ADD(ptr, sizeof(Elf64_Nhdr));
516 }
517 
518 /*
519  * Initialize ELF header (new kernel)
520  */
521 static void *ehdr_init(Elf64_Ehdr *ehdr, int mem_chunk_cnt)
522 {
523 	memset(ehdr, 0, sizeof(*ehdr));
524 	memcpy(ehdr->e_ident, ELFMAG, SELFMAG);
525 	ehdr->e_ident[EI_CLASS] = ELFCLASS64;
526 	ehdr->e_ident[EI_DATA] = ELFDATA2MSB;
527 	ehdr->e_ident[EI_VERSION] = EV_CURRENT;
528 	memset(ehdr->e_ident + EI_PAD, 0, EI_NIDENT - EI_PAD);
529 	ehdr->e_type = ET_CORE;
530 	ehdr->e_machine = EM_S390;
531 	ehdr->e_version = EV_CURRENT;
532 	ehdr->e_phoff = sizeof(Elf64_Ehdr);
533 	ehdr->e_ehsize = sizeof(Elf64_Ehdr);
534 	ehdr->e_phentsize = sizeof(Elf64_Phdr);
535 	ehdr->e_phnum = mem_chunk_cnt + 1;
536 	return ehdr + 1;
537 }
538 
539 /*
540  * Return CPU count for ELF header (new kernel)
541  */
542 static int get_cpu_cnt(void)
543 {
544 	struct save_area *sa;
545 	int cpus = 0;
546 
547 	list_for_each_entry(sa, &dump_save_areas, list)
548 		if (sa->prefix != 0)
549 			cpus++;
550 	return cpus;
551 }
552 
553 /*
554  * Return memory chunk count for ELF header (new kernel)
555  */
556 static int get_mem_chunk_cnt(void)
557 {
558 	int cnt = 0;
559 	u64 idx;
560 
561 	for_each_physmem_range(idx, &oldmem_type, NULL, NULL)
562 		cnt++;
563 	return cnt;
564 }
565 
566 /*
567  * Initialize ELF loads (new kernel)
568  */
569 static void loads_init(Elf64_Phdr *phdr, u64 loads_offset)
570 {
571 	phys_addr_t start, end;
572 	u64 idx;
573 
574 	for_each_physmem_range(idx, &oldmem_type, &start, &end) {
575 		phdr->p_filesz = end - start;
576 		phdr->p_type = PT_LOAD;
577 		phdr->p_offset = start;
578 		phdr->p_vaddr = start;
579 		phdr->p_paddr = start;
580 		phdr->p_memsz = end - start;
581 		phdr->p_flags = PF_R | PF_W | PF_X;
582 		phdr->p_align = PAGE_SIZE;
583 		phdr++;
584 	}
585 }
586 
587 /*
588  * Initialize notes (new kernel)
589  */
590 static void *notes_init(Elf64_Phdr *phdr, void *ptr, u64 notes_offset)
591 {
592 	struct save_area *sa;
593 	void *ptr_start = ptr;
594 	int cpu;
595 
596 	ptr = nt_prpsinfo(ptr);
597 
598 	cpu = 1;
599 	list_for_each_entry(sa, &dump_save_areas, list)
600 		if (sa->prefix != 0)
601 			ptr = fill_cpu_elf_notes(ptr, cpu++, sa);
602 	ptr = nt_vmcoreinfo(ptr);
603 	ptr = nt_final(ptr);
604 	memset(phdr, 0, sizeof(*phdr));
605 	phdr->p_type = PT_NOTE;
606 	phdr->p_offset = notes_offset;
607 	phdr->p_filesz = (unsigned long) PTR_SUB(ptr, ptr_start);
608 	phdr->p_memsz = phdr->p_filesz;
609 	return ptr;
610 }
611 
612 static size_t get_elfcorehdr_size(int mem_chunk_cnt)
613 {
614 	size_t size;
615 
616 	size = sizeof(Elf64_Ehdr);
617 	/* PT_NOTES */
618 	size += sizeof(Elf64_Phdr);
619 	/* nt_prpsinfo */
620 	size += nt_size(NT_PRPSINFO, sizeof(struct elf_prpsinfo));
621 	/* regsets */
622 	size += get_cpu_cnt() * get_cpu_elf_notes_size();
623 	/* nt_vmcoreinfo */
624 	size += nt_vmcoreinfo_size();
625 	/* nt_final */
626 	size += sizeof(Elf64_Nhdr);
627 	/* PT_LOADS */
628 	size += mem_chunk_cnt * sizeof(Elf64_Phdr);
629 
630 	return size;
631 }
632 
633 /*
634  * Create ELF core header (new kernel)
635  */
636 int elfcorehdr_alloc(unsigned long long *addr, unsigned long long *size)
637 {
638 	Elf64_Phdr *phdr_notes, *phdr_loads;
639 	int mem_chunk_cnt;
640 	void *ptr, *hdr;
641 	u32 alloc_size;
642 	u64 hdr_off;
643 
644 	/* If we are not in kdump or zfcp/nvme dump mode return */
645 	if (!oldmem_data.start && !is_ipl_type_dump())
646 		return 0;
647 	/* If we cannot get HSA size for zfcp/nvme dump return error */
648 	if (is_ipl_type_dump() && !sclp.hsa_size)
649 		return -ENODEV;
650 
651 	/* For kdump, exclude previous crashkernel memory */
652 	if (oldmem_data.start) {
653 		oldmem_region.base = oldmem_data.start;
654 		oldmem_region.size = oldmem_data.size;
655 		oldmem_type.total_size = oldmem_data.size;
656 	}
657 
658 	mem_chunk_cnt = get_mem_chunk_cnt();
659 
660 	alloc_size = get_elfcorehdr_size(mem_chunk_cnt);
661 
662 	hdr = kzalloc(alloc_size, GFP_KERNEL);
663 
664 	/* Without elfcorehdr /proc/vmcore cannot be created. Thus creating
665 	 * a dump with this crash kernel will fail. Panic now to allow other
666 	 * dump mechanisms to take over.
667 	 */
668 	if (!hdr)
669 		panic("s390 kdump allocating elfcorehdr failed");
670 
671 	/* Init elf header */
672 	ptr = ehdr_init(hdr, mem_chunk_cnt);
673 	/* Init program headers */
674 	phdr_notes = ptr;
675 	ptr = PTR_ADD(ptr, sizeof(Elf64_Phdr));
676 	phdr_loads = ptr;
677 	ptr = PTR_ADD(ptr, sizeof(Elf64_Phdr) * mem_chunk_cnt);
678 	/* Init notes */
679 	hdr_off = PTR_DIFF(ptr, hdr);
680 	ptr = notes_init(phdr_notes, ptr, ((unsigned long) hdr) + hdr_off);
681 	/* Init loads */
682 	hdr_off = PTR_DIFF(ptr, hdr);
683 	loads_init(phdr_loads, hdr_off);
684 	*addr = (unsigned long long) hdr;
685 	*size = (unsigned long long) hdr_off;
686 	BUG_ON(elfcorehdr_size > alloc_size);
687 	return 0;
688 }
689 
690 /*
691  * Free ELF core header (new kernel)
692  */
693 void elfcorehdr_free(unsigned long long addr)
694 {
695 	kfree((void *)(unsigned long)addr);
696 }
697 
698 /*
699  * Read from ELF header
700  */
701 ssize_t elfcorehdr_read(char *buf, size_t count, u64 *ppos)
702 {
703 	void *src = (void *)(unsigned long)*ppos;
704 
705 	memcpy(buf, src, count);
706 	*ppos += count;
707 	return count;
708 }
709 
710 /*
711  * Read from ELF notes data
712  */
713 ssize_t elfcorehdr_read_notes(char *buf, size_t count, u64 *ppos)
714 {
715 	void *src = (void *)(unsigned long)*ppos;
716 
717 	memcpy(buf, src, count);
718 	*ppos += count;
719 	return count;
720 }
721