Loading...
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#include <asm/maccess.h>
25#include <asm/fpu/api.h>
26
27#define PTR_ADD(x, y) (((char *) (x)) + ((unsigned long) (y)))
28#define PTR_SUB(x, y) (((char *) (x)) - ((unsigned long) (y)))
29#define PTR_DIFF(x, y) ((unsigned long)(((char *) (x)) - ((unsigned long) (y))))
30
31static struct memblock_region oldmem_region;
32
33static struct memblock_type oldmem_type = {
34 .cnt = 1,
35 .max = 1,
36 .total_size = 0,
37 .regions = &oldmem_region,
38 .name = "oldmem",
39};
40
41struct save_area {
42 struct list_head list;
43 u64 psw[2];
44 u64 ctrs[16];
45 u64 gprs[16];
46 u32 acrs[16];
47 u64 fprs[16];
48 u32 fpc;
49 u32 prefix;
50 u32 todpreg;
51 u64 timer;
52 u64 todcmp;
53 u64 vxrs_low[16];
54 __vector128 vxrs_high[16];
55};
56
57static LIST_HEAD(dump_save_areas);
58
59/*
60 * Allocate a save area
61 */
62struct save_area * __init save_area_alloc(bool is_boot_cpu)
63{
64 struct save_area *sa;
65
66 sa = memblock_alloc(sizeof(*sa), 8);
67 if (!sa)
68 return NULL;
69
70 if (is_boot_cpu)
71 list_add(&sa->list, &dump_save_areas);
72 else
73 list_add_tail(&sa->list, &dump_save_areas);
74 return sa;
75}
76
77/*
78 * Return the address of the save area for the boot CPU
79 */
80struct save_area * __init save_area_boot_cpu(void)
81{
82 return list_first_entry_or_null(&dump_save_areas, struct save_area, list);
83}
84
85/*
86 * Copy CPU registers into the save area
87 */
88void __init save_area_add_regs(struct save_area *sa, void *regs)
89{
90 struct lowcore *lc;
91
92 lc = (struct lowcore *)(regs - __LC_FPREGS_SAVE_AREA);
93 memcpy(&sa->psw, &lc->psw_save_area, sizeof(sa->psw));
94 memcpy(&sa->ctrs, &lc->cregs_save_area, sizeof(sa->ctrs));
95 memcpy(&sa->gprs, &lc->gpregs_save_area, sizeof(sa->gprs));
96 memcpy(&sa->acrs, &lc->access_regs_save_area, sizeof(sa->acrs));
97 memcpy(&sa->fprs, &lc->floating_pt_save_area, sizeof(sa->fprs));
98 memcpy(&sa->fpc, &lc->fpt_creg_save_area, sizeof(sa->fpc));
99 memcpy(&sa->prefix, &lc->prefixreg_save_area, sizeof(sa->prefix));
100 memcpy(&sa->todpreg, &lc->tod_progreg_save_area, sizeof(sa->todpreg));
101 memcpy(&sa->timer, &lc->cpu_timer_save_area, sizeof(sa->timer));
102 memcpy(&sa->todcmp, &lc->clock_comp_save_area, sizeof(sa->todcmp));
103}
104
105/*
106 * Copy vector registers into the save area
107 */
108void __init save_area_add_vxrs(struct save_area *sa, __vector128 *vxrs)
109{
110 int i;
111
112 /* Copy lower halves of vector registers 0-15 */
113 for (i = 0; i < 16; i++)
114 sa->vxrs_low[i] = vxrs[i].low;
115 /* Copy vector registers 16-31 */
116 memcpy(sa->vxrs_high, vxrs + 16, 16 * sizeof(__vector128));
117}
118
119static size_t copy_oldmem_iter(struct iov_iter *iter, unsigned long src, size_t count)
120{
121 size_t len, copied, res = 0;
122
123 while (count) {
124 if (!oldmem_data.start && src < sclp.hsa_size) {
125 /* Copy from zfcp/nvme dump HSA area */
126 len = min(count, sclp.hsa_size - src);
127 copied = memcpy_hsa_iter(iter, src, len);
128 } else {
129 /* Check for swapped kdump oldmem areas */
130 if (oldmem_data.start && src - oldmem_data.start < oldmem_data.size) {
131 src -= oldmem_data.start;
132 len = min(count, oldmem_data.size - src);
133 } else if (oldmem_data.start && src < oldmem_data.size) {
134 len = min(count, oldmem_data.size - src);
135 src += oldmem_data.start;
136 } else {
137 len = count;
138 }
139 copied = memcpy_real_iter(iter, src, len);
140 }
141 count -= copied;
142 src += copied;
143 res += copied;
144 if (copied < len)
145 break;
146 }
147 return res;
148}
149
150int copy_oldmem_kernel(void *dst, unsigned long src, size_t count)
151{
152 struct iov_iter iter;
153 struct kvec kvec;
154
155 kvec.iov_base = dst;
156 kvec.iov_len = count;
157 iov_iter_kvec(&iter, ITER_DEST, &kvec, 1, count);
158 if (copy_oldmem_iter(&iter, src, count) < count)
159 return -EFAULT;
160 return 0;
161}
162
163/*
164 * Copy one page from "oldmem"
165 */
166ssize_t copy_oldmem_page(struct iov_iter *iter, unsigned long pfn, size_t csize,
167 unsigned long offset)
168{
169 unsigned long src;
170
171 src = pfn_to_phys(pfn) + offset;
172 return copy_oldmem_iter(iter, src, csize);
173}
174
175/*
176 * Remap "oldmem" for kdump
177 *
178 * For the kdump reserved memory this functions performs a swap operation:
179 * [0 - OLDMEM_SIZE] is mapped to [OLDMEM_BASE - OLDMEM_BASE + OLDMEM_SIZE]
180 */
181static int remap_oldmem_pfn_range_kdump(struct vm_area_struct *vma,
182 unsigned long from, unsigned long pfn,
183 unsigned long size, pgprot_t prot)
184{
185 unsigned long size_old;
186 int rc;
187
188 if (pfn < oldmem_data.size >> PAGE_SHIFT) {
189 size_old = min(size, oldmem_data.size - (pfn << PAGE_SHIFT));
190 rc = remap_pfn_range(vma, from,
191 pfn + (oldmem_data.start >> PAGE_SHIFT),
192 size_old, prot);
193 if (rc || size == size_old)
194 return rc;
195 size -= size_old;
196 from += size_old;
197 pfn += size_old >> PAGE_SHIFT;
198 }
199 return remap_pfn_range(vma, from, pfn, size, prot);
200}
201
202/*
203 * Remap "oldmem" for zfcp/nvme dump
204 *
205 * We only map available memory above HSA size. Memory below HSA size
206 * is read on demand using the copy_oldmem_page() function.
207 */
208static int remap_oldmem_pfn_range_zfcpdump(struct vm_area_struct *vma,
209 unsigned long from,
210 unsigned long pfn,
211 unsigned long size, pgprot_t prot)
212{
213 unsigned long hsa_end = sclp.hsa_size;
214 unsigned long size_hsa;
215
216 if (pfn < hsa_end >> PAGE_SHIFT) {
217 size_hsa = min(size, hsa_end - (pfn << PAGE_SHIFT));
218 if (size == size_hsa)
219 return 0;
220 size -= size_hsa;
221 from += size_hsa;
222 pfn += size_hsa >> PAGE_SHIFT;
223 }
224 return remap_pfn_range(vma, from, pfn, size, prot);
225}
226
227/*
228 * Remap "oldmem" for kdump or zfcp/nvme dump
229 */
230int remap_oldmem_pfn_range(struct vm_area_struct *vma, unsigned long from,
231 unsigned long pfn, unsigned long size, pgprot_t prot)
232{
233 if (oldmem_data.start)
234 return remap_oldmem_pfn_range_kdump(vma, from, pfn, size, prot);
235 else
236 return remap_oldmem_pfn_range_zfcpdump(vma, from, pfn, size,
237 prot);
238}
239
240static const char *nt_name(Elf64_Word type)
241{
242 const char *name = "LINUX";
243
244 if (type == NT_PRPSINFO || type == NT_PRSTATUS || type == NT_PRFPREG)
245 name = KEXEC_CORE_NOTE_NAME;
246 return name;
247}
248
249/*
250 * Initialize ELF note
251 */
252static void *nt_init_name(void *buf, Elf64_Word type, void *desc, int d_len,
253 const char *name)
254{
255 Elf64_Nhdr *note;
256 u64 len;
257
258 note = (Elf64_Nhdr *)buf;
259 note->n_namesz = strlen(name) + 1;
260 note->n_descsz = d_len;
261 note->n_type = type;
262 len = sizeof(Elf64_Nhdr);
263
264 memcpy(buf + len, name, note->n_namesz);
265 len = roundup(len + note->n_namesz, 4);
266
267 memcpy(buf + len, desc, note->n_descsz);
268 len = roundup(len + note->n_descsz, 4);
269
270 return PTR_ADD(buf, len);
271}
272
273static inline void *nt_init(void *buf, Elf64_Word type, void *desc, int d_len)
274{
275 return nt_init_name(buf, type, desc, d_len, nt_name(type));
276}
277
278/*
279 * Calculate the size of ELF note
280 */
281static size_t nt_size_name(int d_len, const char *name)
282{
283 size_t size;
284
285 size = sizeof(Elf64_Nhdr);
286 size += roundup(strlen(name) + 1, 4);
287 size += roundup(d_len, 4);
288
289 return size;
290}
291
292static inline size_t nt_size(Elf64_Word type, int d_len)
293{
294 return nt_size_name(d_len, nt_name(type));
295}
296
297/*
298 * Fill ELF notes for one CPU with save area registers
299 */
300static void *fill_cpu_elf_notes(void *ptr, int cpu, struct save_area *sa)
301{
302 struct elf_prstatus nt_prstatus;
303 elf_fpregset_t nt_fpregset;
304
305 /* Prepare prstatus note */
306 memset(&nt_prstatus, 0, sizeof(nt_prstatus));
307 memcpy(&nt_prstatus.pr_reg.gprs, sa->gprs, sizeof(sa->gprs));
308 memcpy(&nt_prstatus.pr_reg.psw, sa->psw, sizeof(sa->psw));
309 memcpy(&nt_prstatus.pr_reg.acrs, sa->acrs, sizeof(sa->acrs));
310 nt_prstatus.common.pr_pid = cpu;
311 /* Prepare fpregset (floating point) note */
312 memset(&nt_fpregset, 0, sizeof(nt_fpregset));
313 memcpy(&nt_fpregset.fpc, &sa->fpc, sizeof(sa->fpc));
314 memcpy(&nt_fpregset.fprs, &sa->fprs, sizeof(sa->fprs));
315 /* Create ELF notes for the CPU */
316 ptr = nt_init(ptr, NT_PRSTATUS, &nt_prstatus, sizeof(nt_prstatus));
317 ptr = nt_init(ptr, NT_PRFPREG, &nt_fpregset, sizeof(nt_fpregset));
318 ptr = nt_init(ptr, NT_S390_TIMER, &sa->timer, sizeof(sa->timer));
319 ptr = nt_init(ptr, NT_S390_TODCMP, &sa->todcmp, sizeof(sa->todcmp));
320 ptr = nt_init(ptr, NT_S390_TODPREG, &sa->todpreg, sizeof(sa->todpreg));
321 ptr = nt_init(ptr, NT_S390_CTRS, &sa->ctrs, sizeof(sa->ctrs));
322 ptr = nt_init(ptr, NT_S390_PREFIX, &sa->prefix, sizeof(sa->prefix));
323 if (cpu_has_vx()) {
324 ptr = nt_init(ptr, NT_S390_VXRS_HIGH,
325 &sa->vxrs_high, sizeof(sa->vxrs_high));
326 ptr = nt_init(ptr, NT_S390_VXRS_LOW,
327 &sa->vxrs_low, sizeof(sa->vxrs_low));
328 }
329 return ptr;
330}
331
332/*
333 * Calculate size of ELF notes per cpu
334 */
335static size_t get_cpu_elf_notes_size(void)
336{
337 struct save_area *sa = NULL;
338 size_t size;
339
340 size = nt_size(NT_PRSTATUS, sizeof(struct elf_prstatus));
341 size += nt_size(NT_PRFPREG, sizeof(elf_fpregset_t));
342 size += nt_size(NT_S390_TIMER, sizeof(sa->timer));
343 size += nt_size(NT_S390_TODCMP, sizeof(sa->todcmp));
344 size += nt_size(NT_S390_TODPREG, sizeof(sa->todpreg));
345 size += nt_size(NT_S390_CTRS, sizeof(sa->ctrs));
346 size += nt_size(NT_S390_PREFIX, sizeof(sa->prefix));
347 if (cpu_has_vx()) {
348 size += nt_size(NT_S390_VXRS_HIGH, sizeof(sa->vxrs_high));
349 size += nt_size(NT_S390_VXRS_LOW, sizeof(sa->vxrs_low));
350 }
351
352 return size;
353}
354
355/*
356 * Initialize prpsinfo note (new kernel)
357 */
358static void *nt_prpsinfo(void *ptr)
359{
360 struct elf_prpsinfo prpsinfo;
361
362 memset(&prpsinfo, 0, sizeof(prpsinfo));
363 prpsinfo.pr_sname = 'R';
364 strcpy(prpsinfo.pr_fname, "vmlinux");
365 return nt_init(ptr, NT_PRPSINFO, &prpsinfo, sizeof(prpsinfo));
366}
367
368/*
369 * Get vmcoreinfo using lowcore->vmcore_info (new kernel)
370 */
371static void *get_vmcoreinfo_old(unsigned long *size)
372{
373 char nt_name[11], *vmcoreinfo;
374 unsigned long addr;
375 Elf64_Nhdr note;
376
377 if (copy_oldmem_kernel(&addr, __LC_VMCORE_INFO, sizeof(addr)))
378 return NULL;
379 memset(nt_name, 0, sizeof(nt_name));
380 if (copy_oldmem_kernel(¬e, addr, sizeof(note)))
381 return NULL;
382 if (copy_oldmem_kernel(nt_name, addr + sizeof(note),
383 sizeof(nt_name) - 1))
384 return NULL;
385 if (strcmp(nt_name, VMCOREINFO_NOTE_NAME) != 0)
386 return NULL;
387 vmcoreinfo = kzalloc(note.n_descsz, GFP_KERNEL);
388 if (!vmcoreinfo)
389 return NULL;
390 if (copy_oldmem_kernel(vmcoreinfo, addr + 24, note.n_descsz)) {
391 kfree(vmcoreinfo);
392 return NULL;
393 }
394 *size = note.n_descsz;
395 return vmcoreinfo;
396}
397
398/*
399 * Initialize vmcoreinfo note (new kernel)
400 */
401static void *nt_vmcoreinfo(void *ptr)
402{
403 const char *name = VMCOREINFO_NOTE_NAME;
404 unsigned long size;
405 void *vmcoreinfo;
406
407 vmcoreinfo = os_info_old_entry(OS_INFO_VMCOREINFO, &size);
408 if (vmcoreinfo)
409 return nt_init_name(ptr, 0, vmcoreinfo, size, name);
410
411 vmcoreinfo = get_vmcoreinfo_old(&size);
412 if (!vmcoreinfo)
413 return ptr;
414 ptr = nt_init_name(ptr, 0, vmcoreinfo, size, name);
415 kfree(vmcoreinfo);
416 return ptr;
417}
418
419static size_t nt_vmcoreinfo_size(void)
420{
421 const char *name = VMCOREINFO_NOTE_NAME;
422 unsigned long size;
423 void *vmcoreinfo;
424
425 vmcoreinfo = os_info_old_entry(OS_INFO_VMCOREINFO, &size);
426 if (vmcoreinfo)
427 return nt_size_name(size, name);
428
429 vmcoreinfo = get_vmcoreinfo_old(&size);
430 if (!vmcoreinfo)
431 return 0;
432
433 kfree(vmcoreinfo);
434 return nt_size_name(size, name);
435}
436
437/*
438 * Initialize final note (needed for /proc/vmcore code)
439 */
440static void *nt_final(void *ptr)
441{
442 Elf64_Nhdr *note;
443
444 note = (Elf64_Nhdr *) ptr;
445 note->n_namesz = 0;
446 note->n_descsz = 0;
447 note->n_type = 0;
448 return PTR_ADD(ptr, sizeof(Elf64_Nhdr));
449}
450
451/*
452 * Initialize ELF header (new kernel)
453 */
454static void *ehdr_init(Elf64_Ehdr *ehdr, int mem_chunk_cnt)
455{
456 memset(ehdr, 0, sizeof(*ehdr));
457 memcpy(ehdr->e_ident, ELFMAG, SELFMAG);
458 ehdr->e_ident[EI_CLASS] = ELFCLASS64;
459 ehdr->e_ident[EI_DATA] = ELFDATA2MSB;
460 ehdr->e_ident[EI_VERSION] = EV_CURRENT;
461 memset(ehdr->e_ident + EI_PAD, 0, EI_NIDENT - EI_PAD);
462 ehdr->e_type = ET_CORE;
463 ehdr->e_machine = EM_S390;
464 ehdr->e_version = EV_CURRENT;
465 ehdr->e_phoff = sizeof(Elf64_Ehdr);
466 ehdr->e_ehsize = sizeof(Elf64_Ehdr);
467 ehdr->e_phentsize = sizeof(Elf64_Phdr);
468 ehdr->e_phnum = mem_chunk_cnt + 1;
469 return ehdr + 1;
470}
471
472/*
473 * Return CPU count for ELF header (new kernel)
474 */
475static int get_cpu_cnt(void)
476{
477 struct save_area *sa;
478 int cpus = 0;
479
480 list_for_each_entry(sa, &dump_save_areas, list)
481 if (sa->prefix != 0)
482 cpus++;
483 return cpus;
484}
485
486/*
487 * Return memory chunk count for ELF header (new kernel)
488 */
489static int get_mem_chunk_cnt(void)
490{
491 int cnt = 0;
492 u64 idx;
493
494 for_each_physmem_range(idx, &oldmem_type, NULL, NULL)
495 cnt++;
496 return cnt;
497}
498
499/*
500 * Initialize ELF loads (new kernel)
501 */
502static void loads_init(Elf64_Phdr *phdr)
503{
504 phys_addr_t start, end;
505 u64 idx;
506
507 for_each_physmem_range(idx, &oldmem_type, &start, &end) {
508 phdr->p_filesz = end - start;
509 phdr->p_type = PT_LOAD;
510 phdr->p_offset = start;
511 phdr->p_vaddr = (unsigned long)__va(start);
512 phdr->p_paddr = start;
513 phdr->p_memsz = end - start;
514 phdr->p_flags = PF_R | PF_W | PF_X;
515 phdr->p_align = PAGE_SIZE;
516 phdr++;
517 }
518}
519
520/*
521 * Initialize notes (new kernel)
522 */
523static void *notes_init(Elf64_Phdr *phdr, void *ptr, u64 notes_offset)
524{
525 struct save_area *sa;
526 void *ptr_start = ptr;
527 int cpu;
528
529 ptr = nt_prpsinfo(ptr);
530
531 cpu = 1;
532 list_for_each_entry(sa, &dump_save_areas, list)
533 if (sa->prefix != 0)
534 ptr = fill_cpu_elf_notes(ptr, cpu++, sa);
535 ptr = nt_vmcoreinfo(ptr);
536 ptr = nt_final(ptr);
537 memset(phdr, 0, sizeof(*phdr));
538 phdr->p_type = PT_NOTE;
539 phdr->p_offset = notes_offset;
540 phdr->p_filesz = (unsigned long) PTR_SUB(ptr, ptr_start);
541 phdr->p_memsz = phdr->p_filesz;
542 return ptr;
543}
544
545static size_t get_elfcorehdr_size(int mem_chunk_cnt)
546{
547 size_t size;
548
549 size = sizeof(Elf64_Ehdr);
550 /* PT_NOTES */
551 size += sizeof(Elf64_Phdr);
552 /* nt_prpsinfo */
553 size += nt_size(NT_PRPSINFO, sizeof(struct elf_prpsinfo));
554 /* regsets */
555 size += get_cpu_cnt() * get_cpu_elf_notes_size();
556 /* nt_vmcoreinfo */
557 size += nt_vmcoreinfo_size();
558 /* nt_final */
559 size += sizeof(Elf64_Nhdr);
560 /* PT_LOADS */
561 size += mem_chunk_cnt * sizeof(Elf64_Phdr);
562
563 return size;
564}
565
566/*
567 * Create ELF core header (new kernel)
568 */
569int elfcorehdr_alloc(unsigned long long *addr, unsigned long long *size)
570{
571 Elf64_Phdr *phdr_notes, *phdr_loads;
572 size_t alloc_size;
573 int mem_chunk_cnt;
574 void *ptr, *hdr;
575 u64 hdr_off;
576
577 /* If we are not in kdump or zfcp/nvme dump mode return */
578 if (!oldmem_data.start && !is_ipl_type_dump())
579 return 0;
580 /* If we cannot get HSA size for zfcp/nvme dump return error */
581 if (is_ipl_type_dump() && !sclp.hsa_size)
582 return -ENODEV;
583
584 /* For kdump, exclude previous crashkernel memory */
585 if (oldmem_data.start) {
586 oldmem_region.base = oldmem_data.start;
587 oldmem_region.size = oldmem_data.size;
588 oldmem_type.total_size = oldmem_data.size;
589 }
590
591 mem_chunk_cnt = get_mem_chunk_cnt();
592
593 alloc_size = get_elfcorehdr_size(mem_chunk_cnt);
594
595 hdr = kzalloc(alloc_size, GFP_KERNEL);
596
597 /* Without elfcorehdr /proc/vmcore cannot be created. Thus creating
598 * a dump with this crash kernel will fail. Panic now to allow other
599 * dump mechanisms to take over.
600 */
601 if (!hdr)
602 panic("s390 kdump allocating elfcorehdr failed");
603
604 /* Init elf header */
605 ptr = ehdr_init(hdr, mem_chunk_cnt);
606 /* Init program headers */
607 phdr_notes = ptr;
608 ptr = PTR_ADD(ptr, sizeof(Elf64_Phdr));
609 phdr_loads = ptr;
610 ptr = PTR_ADD(ptr, sizeof(Elf64_Phdr) * mem_chunk_cnt);
611 /* Init notes */
612 hdr_off = PTR_DIFF(ptr, hdr);
613 ptr = notes_init(phdr_notes, ptr, ((unsigned long) hdr) + hdr_off);
614 /* Init loads */
615 hdr_off = PTR_DIFF(ptr, hdr);
616 loads_init(phdr_loads);
617 *addr = (unsigned long long) hdr;
618 *size = (unsigned long long) hdr_off;
619 BUG_ON(elfcorehdr_size > alloc_size);
620 return 0;
621}
622
623/*
624 * Free ELF core header (new kernel)
625 */
626void elfcorehdr_free(unsigned long long addr)
627{
628 kfree((void *)(unsigned long)addr);
629}
630
631/*
632 * Read from ELF header
633 */
634ssize_t elfcorehdr_read(char *buf, size_t count, u64 *ppos)
635{
636 void *src = (void *)(unsigned long)*ppos;
637
638 memcpy(buf, src, count);
639 *ppos += count;
640 return count;
641}
642
643/*
644 * Read from ELF notes data
645 */
646ssize_t elfcorehdr_read_notes(char *buf, size_t count, u64 *ppos)
647{
648 void *src = (void *)(unsigned long)*ppos;
649
650 memcpy(buf, src, count);
651 *ppos += count;
652 return count;
653}
1/*
2 * S390 kdump implementation
3 *
4 * Copyright IBM Corp. 2011
5 * Author(s): Michael Holzheu <holzheu@linux.vnet.ibm.com>
6 */
7
8#include <linux/crash_dump.h>
9#include <asm/lowcore.h>
10#include <linux/kernel.h>
11#include <linux/module.h>
12#include <linux/gfp.h>
13#include <linux/slab.h>
14#include <linux/bootmem.h>
15#include <linux/elf.h>
16#include <asm/os_info.h>
17#include <asm/elf.h>
18#include <asm/ipl.h>
19#include <asm/sclp.h>
20
21#define PTR_ADD(x, y) (((char *) (x)) + ((unsigned long) (y)))
22#define PTR_SUB(x, y) (((char *) (x)) - ((unsigned long) (y)))
23#define PTR_DIFF(x, y) ((unsigned long)(((char *) (x)) - ((unsigned long) (y))))
24
25struct dump_save_areas dump_save_areas;
26
27/*
28 * Allocate and add a save area for a CPU
29 */
30struct save_area *dump_save_area_create(int cpu)
31{
32 struct save_area **save_areas, *save_area;
33
34 save_area = kmalloc(sizeof(*save_area), GFP_KERNEL);
35 if (!save_area)
36 return NULL;
37 if (cpu + 1 > dump_save_areas.count) {
38 dump_save_areas.count = cpu + 1;
39 save_areas = krealloc(dump_save_areas.areas,
40 dump_save_areas.count * sizeof(void *),
41 GFP_KERNEL | __GFP_ZERO);
42 if (!save_areas) {
43 kfree(save_area);
44 return NULL;
45 }
46 dump_save_areas.areas = save_areas;
47 }
48 dump_save_areas.areas[cpu] = save_area;
49 return save_area;
50}
51
52/*
53 * Return physical address for virtual address
54 */
55static inline void *load_real_addr(void *addr)
56{
57 unsigned long real_addr;
58
59 asm volatile(
60 " lra %0,0(%1)\n"
61 " jz 0f\n"
62 " la %0,0\n"
63 "0:"
64 : "=a" (real_addr) : "a" (addr) : "cc");
65 return (void *)real_addr;
66}
67
68/*
69 * Copy real to virtual or real memory
70 */
71static int copy_from_realmem(void *dest, void *src, size_t count)
72{
73 unsigned long size;
74
75 if (!count)
76 return 0;
77 if (!is_vmalloc_or_module_addr(dest))
78 return memcpy_real(dest, src, count);
79 do {
80 size = min(count, PAGE_SIZE - (__pa(dest) & ~PAGE_MASK));
81 if (memcpy_real(load_real_addr(dest), src, size))
82 return -EFAULT;
83 count -= size;
84 dest += size;
85 src += size;
86 } while (count);
87 return 0;
88}
89
90/*
91 * Pointer to ELF header in new kernel
92 */
93static void *elfcorehdr_newmem;
94
95/*
96 * Copy one page from zfcpdump "oldmem"
97 *
98 * For pages below HSA size memory from the HSA is copied. Otherwise
99 * real memory copy is used.
100 */
101static ssize_t copy_oldmem_page_zfcpdump(char *buf, size_t csize,
102 unsigned long src, int userbuf)
103{
104 int rc;
105
106 if (src < sclp_get_hsa_size()) {
107 rc = memcpy_hsa(buf, src, csize, userbuf);
108 } else {
109 if (userbuf)
110 rc = copy_to_user_real((void __force __user *) buf,
111 (void *) src, csize);
112 else
113 rc = memcpy_real(buf, (void *) src, csize);
114 }
115 return rc ? rc : csize;
116}
117
118/*
119 * Copy one page from kdump "oldmem"
120 *
121 * For the kdump reserved memory this functions performs a swap operation:
122 * - [OLDMEM_BASE - OLDMEM_BASE + OLDMEM_SIZE] is mapped to [0 - OLDMEM_SIZE].
123 * - [0 - OLDMEM_SIZE] is mapped to [OLDMEM_BASE - OLDMEM_BASE + OLDMEM_SIZE]
124 */
125static ssize_t copy_oldmem_page_kdump(char *buf, size_t csize,
126 unsigned long src, int userbuf)
127
128{
129 int rc;
130
131 if (src < OLDMEM_SIZE)
132 src += OLDMEM_BASE;
133 else if (src > OLDMEM_BASE &&
134 src < OLDMEM_BASE + OLDMEM_SIZE)
135 src -= OLDMEM_BASE;
136 if (userbuf)
137 rc = copy_to_user_real((void __force __user *) buf,
138 (void *) src, csize);
139 else
140 rc = copy_from_realmem(buf, (void *) src, csize);
141 return (rc == 0) ? rc : csize;
142}
143
144/*
145 * Copy one page from "oldmem"
146 */
147ssize_t copy_oldmem_page(unsigned long pfn, char *buf, size_t csize,
148 unsigned long offset, int userbuf)
149{
150 unsigned long src;
151
152 if (!csize)
153 return 0;
154 src = (pfn << PAGE_SHIFT) + offset;
155 if (OLDMEM_BASE)
156 return copy_oldmem_page_kdump(buf, csize, src, userbuf);
157 else
158 return copy_oldmem_page_zfcpdump(buf, csize, src, userbuf);
159}
160
161/*
162 * Remap "oldmem" for kdump
163 *
164 * For the kdump reserved memory this functions performs a swap operation:
165 * [0 - OLDMEM_SIZE] is mapped to [OLDMEM_BASE - OLDMEM_BASE + OLDMEM_SIZE]
166 */
167static int remap_oldmem_pfn_range_kdump(struct vm_area_struct *vma,
168 unsigned long from, unsigned long pfn,
169 unsigned long size, pgprot_t prot)
170{
171 unsigned long size_old;
172 int rc;
173
174 if (pfn < OLDMEM_SIZE >> PAGE_SHIFT) {
175 size_old = min(size, OLDMEM_SIZE - (pfn << PAGE_SHIFT));
176 rc = remap_pfn_range(vma, from,
177 pfn + (OLDMEM_BASE >> PAGE_SHIFT),
178 size_old, prot);
179 if (rc || size == size_old)
180 return rc;
181 size -= size_old;
182 from += size_old;
183 pfn += size_old >> PAGE_SHIFT;
184 }
185 return remap_pfn_range(vma, from, pfn, size, prot);
186}
187
188/*
189 * Remap "oldmem" for zfcpdump
190 *
191 * We only map available memory above HSA size. Memory below HSA size
192 * is read on demand using the copy_oldmem_page() function.
193 */
194static int remap_oldmem_pfn_range_zfcpdump(struct vm_area_struct *vma,
195 unsigned long from,
196 unsigned long pfn,
197 unsigned long size, pgprot_t prot)
198{
199 unsigned long hsa_end = sclp_get_hsa_size();
200 unsigned long size_hsa;
201
202 if (pfn < hsa_end >> PAGE_SHIFT) {
203 size_hsa = min(size, hsa_end - (pfn << PAGE_SHIFT));
204 if (size == size_hsa)
205 return 0;
206 size -= size_hsa;
207 from += size_hsa;
208 pfn += size_hsa >> PAGE_SHIFT;
209 }
210 return remap_pfn_range(vma, from, pfn, size, prot);
211}
212
213/*
214 * Remap "oldmem" for kdump or zfcpdump
215 */
216int remap_oldmem_pfn_range(struct vm_area_struct *vma, unsigned long from,
217 unsigned long pfn, unsigned long size, pgprot_t prot)
218{
219 if (OLDMEM_BASE)
220 return remap_oldmem_pfn_range_kdump(vma, from, pfn, size, prot);
221 else
222 return remap_oldmem_pfn_range_zfcpdump(vma, from, pfn, size,
223 prot);
224}
225
226/*
227 * Copy memory from old kernel
228 */
229int copy_from_oldmem(void *dest, void *src, size_t count)
230{
231 unsigned long copied = 0;
232 int rc;
233
234 if (OLDMEM_BASE) {
235 if ((unsigned long) src < OLDMEM_SIZE) {
236 copied = min(count, OLDMEM_SIZE - (unsigned long) src);
237 rc = copy_from_realmem(dest, src + OLDMEM_BASE, copied);
238 if (rc)
239 return rc;
240 }
241 } else {
242 unsigned long hsa_end = sclp_get_hsa_size();
243 if ((unsigned long) src < hsa_end) {
244 copied = min(count, hsa_end - (unsigned long) src);
245 rc = memcpy_hsa(dest, (unsigned long) src, copied, 0);
246 if (rc)
247 return rc;
248 }
249 }
250 return copy_from_realmem(dest + copied, src + copied, count - copied);
251}
252
253/*
254 * Alloc memory and panic in case of ENOMEM
255 */
256static void *kzalloc_panic(int len)
257{
258 void *rc;
259
260 rc = kzalloc(len, GFP_KERNEL);
261 if (!rc)
262 panic("s390 kdump kzalloc (%d) failed", len);
263 return rc;
264}
265
266/*
267 * Get memory layout and create hole for oldmem
268 */
269static struct mem_chunk *get_memory_layout(void)
270{
271 struct mem_chunk *chunk_array;
272
273 chunk_array = kzalloc_panic(MEMORY_CHUNKS * sizeof(struct mem_chunk));
274 detect_memory_layout(chunk_array, 0);
275 create_mem_hole(chunk_array, OLDMEM_BASE, OLDMEM_SIZE);
276 return chunk_array;
277}
278
279/*
280 * Initialize ELF note
281 */
282static void *nt_init(void *buf, Elf64_Word type, void *desc, int d_len,
283 const char *name)
284{
285 Elf64_Nhdr *note;
286 u64 len;
287
288 note = (Elf64_Nhdr *)buf;
289 note->n_namesz = strlen(name) + 1;
290 note->n_descsz = d_len;
291 note->n_type = type;
292 len = sizeof(Elf64_Nhdr);
293
294 memcpy(buf + len, name, note->n_namesz);
295 len = roundup(len + note->n_namesz, 4);
296
297 memcpy(buf + len, desc, note->n_descsz);
298 len = roundup(len + note->n_descsz, 4);
299
300 return PTR_ADD(buf, len);
301}
302
303/*
304 * Initialize prstatus note
305 */
306static void *nt_prstatus(void *ptr, struct save_area *sa)
307{
308 struct elf_prstatus nt_prstatus;
309 static int cpu_nr = 1;
310
311 memset(&nt_prstatus, 0, sizeof(nt_prstatus));
312 memcpy(&nt_prstatus.pr_reg.gprs, sa->gp_regs, sizeof(sa->gp_regs));
313 memcpy(&nt_prstatus.pr_reg.psw, sa->psw, sizeof(sa->psw));
314 memcpy(&nt_prstatus.pr_reg.acrs, sa->acc_regs, sizeof(sa->acc_regs));
315 nt_prstatus.pr_pid = cpu_nr;
316 cpu_nr++;
317
318 return nt_init(ptr, NT_PRSTATUS, &nt_prstatus, sizeof(nt_prstatus),
319 "CORE");
320}
321
322/*
323 * Initialize fpregset (floating point) note
324 */
325static void *nt_fpregset(void *ptr, struct save_area *sa)
326{
327 elf_fpregset_t nt_fpregset;
328
329 memset(&nt_fpregset, 0, sizeof(nt_fpregset));
330 memcpy(&nt_fpregset.fpc, &sa->fp_ctrl_reg, sizeof(sa->fp_ctrl_reg));
331 memcpy(&nt_fpregset.fprs, &sa->fp_regs, sizeof(sa->fp_regs));
332
333 return nt_init(ptr, NT_PRFPREG, &nt_fpregset, sizeof(nt_fpregset),
334 "CORE");
335}
336
337/*
338 * Initialize timer note
339 */
340static void *nt_s390_timer(void *ptr, struct save_area *sa)
341{
342 return nt_init(ptr, NT_S390_TIMER, &sa->timer, sizeof(sa->timer),
343 KEXEC_CORE_NOTE_NAME);
344}
345
346/*
347 * Initialize TOD clock comparator note
348 */
349static void *nt_s390_tod_cmp(void *ptr, struct save_area *sa)
350{
351 return nt_init(ptr, NT_S390_TODCMP, &sa->clk_cmp,
352 sizeof(sa->clk_cmp), KEXEC_CORE_NOTE_NAME);
353}
354
355/*
356 * Initialize TOD programmable register note
357 */
358static void *nt_s390_tod_preg(void *ptr, struct save_area *sa)
359{
360 return nt_init(ptr, NT_S390_TODPREG, &sa->tod_reg,
361 sizeof(sa->tod_reg), KEXEC_CORE_NOTE_NAME);
362}
363
364/*
365 * Initialize control register note
366 */
367static void *nt_s390_ctrs(void *ptr, struct save_area *sa)
368{
369 return nt_init(ptr, NT_S390_CTRS, &sa->ctrl_regs,
370 sizeof(sa->ctrl_regs), KEXEC_CORE_NOTE_NAME);
371}
372
373/*
374 * Initialize prefix register note
375 */
376static void *nt_s390_prefix(void *ptr, struct save_area *sa)
377{
378 return nt_init(ptr, NT_S390_PREFIX, &sa->pref_reg,
379 sizeof(sa->pref_reg), KEXEC_CORE_NOTE_NAME);
380}
381
382/*
383 * Fill ELF notes for one CPU with save area registers
384 */
385void *fill_cpu_elf_notes(void *ptr, struct save_area *sa)
386{
387 ptr = nt_prstatus(ptr, sa);
388 ptr = nt_fpregset(ptr, sa);
389 ptr = nt_s390_timer(ptr, sa);
390 ptr = nt_s390_tod_cmp(ptr, sa);
391 ptr = nt_s390_tod_preg(ptr, sa);
392 ptr = nt_s390_ctrs(ptr, sa);
393 ptr = nt_s390_prefix(ptr, sa);
394 return ptr;
395}
396
397/*
398 * Initialize prpsinfo note (new kernel)
399 */
400static void *nt_prpsinfo(void *ptr)
401{
402 struct elf_prpsinfo prpsinfo;
403
404 memset(&prpsinfo, 0, sizeof(prpsinfo));
405 prpsinfo.pr_sname = 'R';
406 strcpy(prpsinfo.pr_fname, "vmlinux");
407 return nt_init(ptr, NT_PRPSINFO, &prpsinfo, sizeof(prpsinfo),
408 KEXEC_CORE_NOTE_NAME);
409}
410
411/*
412 * Get vmcoreinfo using lowcore->vmcore_info (new kernel)
413 */
414static void *get_vmcoreinfo_old(unsigned long *size)
415{
416 char nt_name[11], *vmcoreinfo;
417 Elf64_Nhdr note;
418 void *addr;
419
420 if (copy_from_oldmem(&addr, &S390_lowcore.vmcore_info, sizeof(addr)))
421 return NULL;
422 memset(nt_name, 0, sizeof(nt_name));
423 if (copy_from_oldmem(¬e, addr, sizeof(note)))
424 return NULL;
425 if (copy_from_oldmem(nt_name, addr + sizeof(note), sizeof(nt_name) - 1))
426 return NULL;
427 if (strcmp(nt_name, "VMCOREINFO") != 0)
428 return NULL;
429 vmcoreinfo = kzalloc_panic(note.n_descsz);
430 if (copy_from_oldmem(vmcoreinfo, addr + 24, note.n_descsz))
431 return NULL;
432 *size = note.n_descsz;
433 return vmcoreinfo;
434}
435
436/*
437 * Initialize vmcoreinfo note (new kernel)
438 */
439static void *nt_vmcoreinfo(void *ptr)
440{
441 unsigned long size;
442 void *vmcoreinfo;
443
444 vmcoreinfo = os_info_old_entry(OS_INFO_VMCOREINFO, &size);
445 if (!vmcoreinfo)
446 vmcoreinfo = get_vmcoreinfo_old(&size);
447 if (!vmcoreinfo)
448 return ptr;
449 return nt_init(ptr, 0, vmcoreinfo, size, "VMCOREINFO");
450}
451
452/*
453 * Initialize ELF header (new kernel)
454 */
455static void *ehdr_init(Elf64_Ehdr *ehdr, int mem_chunk_cnt)
456{
457 memset(ehdr, 0, sizeof(*ehdr));
458 memcpy(ehdr->e_ident, ELFMAG, SELFMAG);
459 ehdr->e_ident[EI_CLASS] = ELFCLASS64;
460 ehdr->e_ident[EI_DATA] = ELFDATA2MSB;
461 ehdr->e_ident[EI_VERSION] = EV_CURRENT;
462 memset(ehdr->e_ident + EI_PAD, 0, EI_NIDENT - EI_PAD);
463 ehdr->e_type = ET_CORE;
464 ehdr->e_machine = EM_S390;
465 ehdr->e_version = EV_CURRENT;
466 ehdr->e_phoff = sizeof(Elf64_Ehdr);
467 ehdr->e_ehsize = sizeof(Elf64_Ehdr);
468 ehdr->e_phentsize = sizeof(Elf64_Phdr);
469 ehdr->e_phnum = mem_chunk_cnt + 1;
470 return ehdr + 1;
471}
472
473/*
474 * Return CPU count for ELF header (new kernel)
475 */
476static int get_cpu_cnt(void)
477{
478 int i, cpus = 0;
479
480 for (i = 0; i < dump_save_areas.count; i++) {
481 if (dump_save_areas.areas[i]->pref_reg == 0)
482 continue;
483 cpus++;
484 }
485 return cpus;
486}
487
488/*
489 * Return memory chunk count for ELF header (new kernel)
490 */
491static int get_mem_chunk_cnt(void)
492{
493 struct mem_chunk *chunk_array, *mem_chunk;
494 int i, cnt = 0;
495
496 chunk_array = get_memory_layout();
497 for (i = 0; i < MEMORY_CHUNKS; i++) {
498 mem_chunk = &chunk_array[i];
499 if (chunk_array[i].type != CHUNK_READ_WRITE &&
500 chunk_array[i].type != CHUNK_READ_ONLY)
501 continue;
502 if (mem_chunk->size == 0)
503 continue;
504 cnt++;
505 }
506 kfree(chunk_array);
507 return cnt;
508}
509
510/*
511 * Initialize ELF loads (new kernel)
512 */
513static int loads_init(Elf64_Phdr *phdr, u64 loads_offset)
514{
515 struct mem_chunk *chunk_array, *mem_chunk;
516 int i;
517
518 chunk_array = get_memory_layout();
519 for (i = 0; i < MEMORY_CHUNKS; i++) {
520 mem_chunk = &chunk_array[i];
521 if (mem_chunk->size == 0)
522 continue;
523 if (chunk_array[i].type != CHUNK_READ_WRITE &&
524 chunk_array[i].type != CHUNK_READ_ONLY)
525 continue;
526 else
527 phdr->p_filesz = mem_chunk->size;
528 phdr->p_type = PT_LOAD;
529 phdr->p_offset = mem_chunk->addr;
530 phdr->p_vaddr = mem_chunk->addr;
531 phdr->p_paddr = mem_chunk->addr;
532 phdr->p_memsz = mem_chunk->size;
533 phdr->p_flags = PF_R | PF_W | PF_X;
534 phdr->p_align = PAGE_SIZE;
535 phdr++;
536 }
537 kfree(chunk_array);
538 return i;
539}
540
541/*
542 * Initialize notes (new kernel)
543 */
544static void *notes_init(Elf64_Phdr *phdr, void *ptr, u64 notes_offset)
545{
546 struct save_area *sa;
547 void *ptr_start = ptr;
548 int i;
549
550 ptr = nt_prpsinfo(ptr);
551
552 for (i = 0; i < dump_save_areas.count; i++) {
553 sa = dump_save_areas.areas[i];
554 if (sa->pref_reg == 0)
555 continue;
556 ptr = fill_cpu_elf_notes(ptr, sa);
557 }
558 ptr = nt_vmcoreinfo(ptr);
559 memset(phdr, 0, sizeof(*phdr));
560 phdr->p_type = PT_NOTE;
561 phdr->p_offset = notes_offset;
562 phdr->p_filesz = (unsigned long) PTR_SUB(ptr, ptr_start);
563 phdr->p_memsz = phdr->p_filesz;
564 return ptr;
565}
566
567/*
568 * Create ELF core header (new kernel)
569 */
570int elfcorehdr_alloc(unsigned long long *addr, unsigned long long *size)
571{
572 Elf64_Phdr *phdr_notes, *phdr_loads;
573 int mem_chunk_cnt;
574 void *ptr, *hdr;
575 u32 alloc_size;
576 u64 hdr_off;
577
578 /* If we are not in kdump or zfcpdump mode return */
579 if (!OLDMEM_BASE && ipl_info.type != IPL_TYPE_FCP_DUMP)
580 return 0;
581 /* If elfcorehdr= has been passed via cmdline, we use that one */
582 if (elfcorehdr_addr != ELFCORE_ADDR_MAX)
583 return 0;
584 /* If we cannot get HSA size for zfcpdump return error */
585 if (ipl_info.type == IPL_TYPE_FCP_DUMP && !sclp_get_hsa_size())
586 return -ENODEV;
587 mem_chunk_cnt = get_mem_chunk_cnt();
588
589 alloc_size = 0x1000 + get_cpu_cnt() * 0x300 +
590 mem_chunk_cnt * sizeof(Elf64_Phdr);
591 hdr = kzalloc_panic(alloc_size);
592 /* Init elf header */
593 ptr = ehdr_init(hdr, mem_chunk_cnt);
594 /* Init program headers */
595 phdr_notes = ptr;
596 ptr = PTR_ADD(ptr, sizeof(Elf64_Phdr));
597 phdr_loads = ptr;
598 ptr = PTR_ADD(ptr, sizeof(Elf64_Phdr) * mem_chunk_cnt);
599 /* Init notes */
600 hdr_off = PTR_DIFF(ptr, hdr);
601 ptr = notes_init(phdr_notes, ptr, ((unsigned long) hdr) + hdr_off);
602 /* Init loads */
603 hdr_off = PTR_DIFF(ptr, hdr);
604 loads_init(phdr_loads, hdr_off);
605 *addr = (unsigned long long) hdr;
606 elfcorehdr_newmem = hdr;
607 *size = (unsigned long long) hdr_off;
608 BUG_ON(elfcorehdr_size > alloc_size);
609 return 0;
610}
611
612/*
613 * Free ELF core header (new kernel)
614 */
615void elfcorehdr_free(unsigned long long addr)
616{
617 if (!elfcorehdr_newmem)
618 return;
619 kfree((void *)(unsigned long)addr);
620}
621
622/*
623 * Read from ELF header
624 */
625ssize_t elfcorehdr_read(char *buf, size_t count, u64 *ppos)
626{
627 void *src = (void *)(unsigned long)*ppos;
628
629 src = elfcorehdr_newmem ? src : src - OLDMEM_BASE;
630 memcpy(buf, src, count);
631 *ppos += count;
632 return count;
633}
634
635/*
636 * Read from ELF notes data
637 */
638ssize_t elfcorehdr_read_notes(char *buf, size_t count, u64 *ppos)
639{
640 void *src = (void *)(unsigned long)*ppos;
641 int rc;
642
643 if (elfcorehdr_newmem) {
644 memcpy(buf, src, count);
645 } else {
646 rc = copy_from_oldmem(buf, src, count);
647 if (rc)
648 return rc;
649 }
650 *ppos += count;
651 return count;
652}