io_uring: ensure finish_wait() is always called in __io_uring_task_cancel()
[linux/fpc-iii.git] / arch / s390 / kernel / crash_dump.c
blob205b2e2648aae3dd502ea37a0418040e0b0dc8b9
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * S390 kdump implementation
5 * Copyright IBM Corp. 2011
6 * Author(s): Michael Holzheu <holzheu@linux.vnet.ibm.com>
7 */
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 <asm/asm-offsets.h>
19 #include <asm/os_info.h>
20 #include <asm/elf.h>
21 #include <asm/ipl.h>
22 #include <asm/sclp.h>
24 #define PTR_ADD(x, y) (((char *) (x)) + ((unsigned long) (y)))
25 #define PTR_SUB(x, y) (((char *) (x)) - ((unsigned long) (y)))
26 #define PTR_DIFF(x, y) ((unsigned long)(((char *) (x)) - ((unsigned long) (y))))
28 static struct memblock_region oldmem_region;
30 static struct memblock_type oldmem_type = {
31 .cnt = 1,
32 .max = 1,
33 .total_size = 0,
34 .regions = &oldmem_region,
35 .name = "oldmem",
38 struct save_area {
39 struct list_head list;
40 u64 psw[2];
41 u64 ctrs[16];
42 u64 gprs[16];
43 u32 acrs[16];
44 u64 fprs[16];
45 u32 fpc;
46 u32 prefix;
47 u64 todpreg;
48 u64 timer;
49 u64 todcmp;
50 u64 vxrs_low[16];
51 __vector128 vxrs_high[16];
54 static LIST_HEAD(dump_save_areas);
57 * Allocate a save area
59 struct save_area * __init save_area_alloc(bool is_boot_cpu)
61 struct save_area *sa;
63 sa = (void *) memblock_phys_alloc(sizeof(*sa), 8);
64 if (!sa)
65 panic("Failed to allocate save area\n");
67 if (is_boot_cpu)
68 list_add(&sa->list, &dump_save_areas);
69 else
70 list_add_tail(&sa->list, &dump_save_areas);
71 return sa;
75 * Return the address of the save area for the boot CPU
77 struct save_area * __init save_area_boot_cpu(void)
79 return list_first_entry_or_null(&dump_save_areas, struct save_area, list);
83 * Copy CPU registers into the save area
85 void __init save_area_add_regs(struct save_area *sa, void *regs)
87 struct lowcore *lc;
89 lc = (struct lowcore *)(regs - __LC_FPREGS_SAVE_AREA);
90 memcpy(&sa->psw, &lc->psw_save_area, sizeof(sa->psw));
91 memcpy(&sa->ctrs, &lc->cregs_save_area, sizeof(sa->ctrs));
92 memcpy(&sa->gprs, &lc->gpregs_save_area, sizeof(sa->gprs));
93 memcpy(&sa->acrs, &lc->access_regs_save_area, sizeof(sa->acrs));
94 memcpy(&sa->fprs, &lc->floating_pt_save_area, sizeof(sa->fprs));
95 memcpy(&sa->fpc, &lc->fpt_creg_save_area, sizeof(sa->fpc));
96 memcpy(&sa->prefix, &lc->prefixreg_save_area, sizeof(sa->prefix));
97 memcpy(&sa->todpreg, &lc->tod_progreg_save_area, sizeof(sa->todpreg));
98 memcpy(&sa->timer, &lc->cpu_timer_save_area, sizeof(sa->timer));
99 memcpy(&sa->todcmp, &lc->clock_comp_save_area, sizeof(sa->todcmp));
103 * Copy vector registers into the save area
105 void __init save_area_add_vxrs(struct save_area *sa, __vector128 *vxrs)
107 int i;
109 /* Copy lower halves of vector registers 0-15 */
110 for (i = 0; i < 16; i++)
111 memcpy(&sa->vxrs_low[i], &vxrs[i].u[2], 8);
112 /* Copy vector registers 16-31 */
113 memcpy(sa->vxrs_high, vxrs + 16, 16 * sizeof(__vector128));
117 * Return physical address for virtual address
119 static inline void *load_real_addr(void *addr)
121 unsigned long real_addr;
123 asm volatile(
124 " lra %0,0(%1)\n"
125 " jz 0f\n"
126 " la %0,0\n"
127 "0:"
128 : "=a" (real_addr) : "a" (addr) : "cc");
129 return (void *)real_addr;
133 * Copy memory of the old, dumped system to a kernel space virtual address
135 int copy_oldmem_kernel(void *dst, void *src, size_t count)
137 unsigned long from, len;
138 void *ra;
139 int rc;
141 while (count) {
142 from = __pa(src);
143 if (!OLDMEM_BASE && from < sclp.hsa_size) {
144 /* Copy from zfcp/nvme dump HSA area */
145 len = min(count, sclp.hsa_size - from);
146 rc = memcpy_hsa_kernel(dst, from, len);
147 if (rc)
148 return rc;
149 } else {
150 /* Check for swapped kdump oldmem areas */
151 if (OLDMEM_BASE && from - OLDMEM_BASE < OLDMEM_SIZE) {
152 from -= OLDMEM_BASE;
153 len = min(count, OLDMEM_SIZE - from);
154 } else if (OLDMEM_BASE && from < OLDMEM_SIZE) {
155 len = min(count, OLDMEM_SIZE - from);
156 from += OLDMEM_BASE;
157 } else {
158 len = count;
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;
166 if (memcpy_real(ra, (void *) from, len))
167 return -EFAULT;
169 dst += len;
170 src += len;
171 count -= len;
173 return 0;
177 * Copy memory of the old, dumped system to a user space virtual address
179 static int copy_oldmem_user(void __user *dst, void *src, size_t count)
181 unsigned long from, len;
182 int rc;
184 while (count) {
185 from = __pa(src);
186 if (!OLDMEM_BASE && from < sclp.hsa_size) {
187 /* Copy from zfcp/nvme dump HSA area */
188 len = min(count, sclp.hsa_size - from);
189 rc = memcpy_hsa_user(dst, from, len);
190 if (rc)
191 return rc;
192 } else {
193 /* Check for swapped kdump oldmem areas */
194 if (OLDMEM_BASE && from - OLDMEM_BASE < OLDMEM_SIZE) {
195 from -= OLDMEM_BASE;
196 len = min(count, OLDMEM_SIZE - from);
197 } else if (OLDMEM_BASE && from < OLDMEM_SIZE) {
198 len = min(count, OLDMEM_SIZE - from);
199 from += OLDMEM_BASE;
200 } else {
201 len = count;
203 rc = copy_to_user_real(dst, (void *) from, count);
204 if (rc)
205 return rc;
207 dst += len;
208 src += len;
209 count -= len;
211 return 0;
215 * Copy one page from "oldmem"
217 ssize_t copy_oldmem_page(unsigned long pfn, char *buf, size_t csize,
218 unsigned long offset, int userbuf)
220 void *src;
221 int rc;
223 if (!csize)
224 return 0;
225 src = (void *) (pfn << PAGE_SHIFT) + offset;
226 if (userbuf)
227 rc = copy_oldmem_user((void __force __user *) buf, src, csize);
228 else
229 rc = copy_oldmem_kernel((void *) buf, src, csize);
230 return rc;
234 * Remap "oldmem" for kdump
236 * For the kdump reserved memory this functions performs a swap operation:
237 * [0 - OLDMEM_SIZE] is mapped to [OLDMEM_BASE - OLDMEM_BASE + OLDMEM_SIZE]
239 static int remap_oldmem_pfn_range_kdump(struct vm_area_struct *vma,
240 unsigned long from, unsigned long pfn,
241 unsigned long size, pgprot_t prot)
243 unsigned long size_old;
244 int rc;
246 if (pfn < OLDMEM_SIZE >> PAGE_SHIFT) {
247 size_old = min(size, OLDMEM_SIZE - (pfn << PAGE_SHIFT));
248 rc = remap_pfn_range(vma, from,
249 pfn + (OLDMEM_BASE >> PAGE_SHIFT),
250 size_old, prot);
251 if (rc || size == size_old)
252 return rc;
253 size -= size_old;
254 from += size_old;
255 pfn += size_old >> PAGE_SHIFT;
257 return remap_pfn_range(vma, from, pfn, size, prot);
261 * Remap "oldmem" for zfcp/nvme dump
263 * We only map available memory above HSA size. Memory below HSA size
264 * is read on demand using the copy_oldmem_page() function.
266 static int remap_oldmem_pfn_range_zfcpdump(struct vm_area_struct *vma,
267 unsigned long from,
268 unsigned long pfn,
269 unsigned long size, pgprot_t prot)
271 unsigned long hsa_end = sclp.hsa_size;
272 unsigned long size_hsa;
274 if (pfn < hsa_end >> PAGE_SHIFT) {
275 size_hsa = min(size, hsa_end - (pfn << PAGE_SHIFT));
276 if (size == size_hsa)
277 return 0;
278 size -= size_hsa;
279 from += size_hsa;
280 pfn += size_hsa >> PAGE_SHIFT;
282 return remap_pfn_range(vma, from, pfn, size, prot);
286 * Remap "oldmem" for kdump or zfcp/nvme dump
288 int remap_oldmem_pfn_range(struct vm_area_struct *vma, unsigned long from,
289 unsigned long pfn, unsigned long size, pgprot_t prot)
291 if (OLDMEM_BASE)
292 return remap_oldmem_pfn_range_kdump(vma, from, pfn, size, prot);
293 else
294 return remap_oldmem_pfn_range_zfcpdump(vma, from, pfn, size,
295 prot);
298 static const char *nt_name(Elf64_Word type)
300 const char *name = "LINUX";
302 if (type == NT_PRPSINFO || type == NT_PRSTATUS || type == NT_PRFPREG)
303 name = KEXEC_CORE_NOTE_NAME;
304 return name;
308 * Initialize ELF note
310 static void *nt_init_name(void *buf, Elf64_Word type, void *desc, int d_len,
311 const char *name)
313 Elf64_Nhdr *note;
314 u64 len;
316 note = (Elf64_Nhdr *)buf;
317 note->n_namesz = strlen(name) + 1;
318 note->n_descsz = d_len;
319 note->n_type = type;
320 len = sizeof(Elf64_Nhdr);
322 memcpy(buf + len, name, note->n_namesz);
323 len = roundup(len + note->n_namesz, 4);
325 memcpy(buf + len, desc, note->n_descsz);
326 len = roundup(len + note->n_descsz, 4);
328 return PTR_ADD(buf, len);
331 static inline void *nt_init(void *buf, Elf64_Word type, void *desc, int d_len)
333 return nt_init_name(buf, type, desc, d_len, nt_name(type));
337 * Calculate the size of ELF note
339 static size_t nt_size_name(int d_len, const char *name)
341 size_t size;
343 size = sizeof(Elf64_Nhdr);
344 size += roundup(strlen(name) + 1, 4);
345 size += roundup(d_len, 4);
347 return size;
350 static inline size_t nt_size(Elf64_Word type, int d_len)
352 return nt_size_name(d_len, nt_name(type));
356 * Fill ELF notes for one CPU with save area registers
358 static void *fill_cpu_elf_notes(void *ptr, int cpu, struct save_area *sa)
360 struct elf_prstatus nt_prstatus;
361 elf_fpregset_t nt_fpregset;
363 /* Prepare prstatus note */
364 memset(&nt_prstatus, 0, sizeof(nt_prstatus));
365 memcpy(&nt_prstatus.pr_reg.gprs, sa->gprs, sizeof(sa->gprs));
366 memcpy(&nt_prstatus.pr_reg.psw, sa->psw, sizeof(sa->psw));
367 memcpy(&nt_prstatus.pr_reg.acrs, sa->acrs, sizeof(sa->acrs));
368 nt_prstatus.pr_pid = cpu;
369 /* Prepare fpregset (floating point) note */
370 memset(&nt_fpregset, 0, sizeof(nt_fpregset));
371 memcpy(&nt_fpregset.fpc, &sa->fpc, sizeof(sa->fpc));
372 memcpy(&nt_fpregset.fprs, &sa->fprs, sizeof(sa->fprs));
373 /* Create ELF notes for the CPU */
374 ptr = nt_init(ptr, NT_PRSTATUS, &nt_prstatus, sizeof(nt_prstatus));
375 ptr = nt_init(ptr, NT_PRFPREG, &nt_fpregset, sizeof(nt_fpregset));
376 ptr = nt_init(ptr, NT_S390_TIMER, &sa->timer, sizeof(sa->timer));
377 ptr = nt_init(ptr, NT_S390_TODCMP, &sa->todcmp, sizeof(sa->todcmp));
378 ptr = nt_init(ptr, NT_S390_TODPREG, &sa->todpreg, sizeof(sa->todpreg));
379 ptr = nt_init(ptr, NT_S390_CTRS, &sa->ctrs, sizeof(sa->ctrs));
380 ptr = nt_init(ptr, NT_S390_PREFIX, &sa->prefix, sizeof(sa->prefix));
381 if (MACHINE_HAS_VX) {
382 ptr = nt_init(ptr, NT_S390_VXRS_HIGH,
383 &sa->vxrs_high, sizeof(sa->vxrs_high));
384 ptr = nt_init(ptr, NT_S390_VXRS_LOW,
385 &sa->vxrs_low, sizeof(sa->vxrs_low));
387 return ptr;
391 * Calculate size of ELF notes per cpu
393 static size_t get_cpu_elf_notes_size(void)
395 struct save_area *sa = NULL;
396 size_t size;
398 size = nt_size(NT_PRSTATUS, sizeof(struct elf_prstatus));
399 size += nt_size(NT_PRFPREG, sizeof(elf_fpregset_t));
400 size += nt_size(NT_S390_TIMER, sizeof(sa->timer));
401 size += nt_size(NT_S390_TODCMP, sizeof(sa->todcmp));
402 size += nt_size(NT_S390_TODPREG, sizeof(sa->todpreg));
403 size += nt_size(NT_S390_CTRS, sizeof(sa->ctrs));
404 size += nt_size(NT_S390_PREFIX, sizeof(sa->prefix));
405 if (MACHINE_HAS_VX) {
406 size += nt_size(NT_S390_VXRS_HIGH, sizeof(sa->vxrs_high));
407 size += nt_size(NT_S390_VXRS_LOW, sizeof(sa->vxrs_low));
410 return size;
414 * Initialize prpsinfo note (new kernel)
416 static void *nt_prpsinfo(void *ptr)
418 struct elf_prpsinfo prpsinfo;
420 memset(&prpsinfo, 0, sizeof(prpsinfo));
421 prpsinfo.pr_sname = 'R';
422 strcpy(prpsinfo.pr_fname, "vmlinux");
423 return nt_init(ptr, NT_PRPSINFO, &prpsinfo, sizeof(prpsinfo));
427 * Get vmcoreinfo using lowcore->vmcore_info (new kernel)
429 static void *get_vmcoreinfo_old(unsigned long *size)
431 char nt_name[11], *vmcoreinfo;
432 Elf64_Nhdr note;
433 void *addr;
435 if (copy_oldmem_kernel(&addr, &S390_lowcore.vmcore_info, sizeof(addr)))
436 return NULL;
437 memset(nt_name, 0, sizeof(nt_name));
438 if (copy_oldmem_kernel(&note, addr, sizeof(note)))
439 return NULL;
440 if (copy_oldmem_kernel(nt_name, addr + sizeof(note),
441 sizeof(nt_name) - 1))
442 return NULL;
443 if (strcmp(nt_name, VMCOREINFO_NOTE_NAME) != 0)
444 return NULL;
445 vmcoreinfo = kzalloc(note.n_descsz, GFP_KERNEL);
446 if (!vmcoreinfo)
447 return NULL;
448 if (copy_oldmem_kernel(vmcoreinfo, addr + 24, note.n_descsz)) {
449 kfree(vmcoreinfo);
450 return NULL;
452 *size = note.n_descsz;
453 return vmcoreinfo;
457 * Initialize vmcoreinfo note (new kernel)
459 static void *nt_vmcoreinfo(void *ptr)
461 const char *name = VMCOREINFO_NOTE_NAME;
462 unsigned long size;
463 void *vmcoreinfo;
465 vmcoreinfo = os_info_old_entry(OS_INFO_VMCOREINFO, &size);
466 if (vmcoreinfo)
467 return nt_init_name(ptr, 0, vmcoreinfo, size, name);
469 vmcoreinfo = get_vmcoreinfo_old(&size);
470 if (!vmcoreinfo)
471 return ptr;
472 ptr = nt_init_name(ptr, 0, vmcoreinfo, size, name);
473 kfree(vmcoreinfo);
474 return ptr;
477 static size_t nt_vmcoreinfo_size(void)
479 const char *name = VMCOREINFO_NOTE_NAME;
480 unsigned long size;
481 void *vmcoreinfo;
483 vmcoreinfo = os_info_old_entry(OS_INFO_VMCOREINFO, &size);
484 if (vmcoreinfo)
485 return nt_size_name(size, name);
487 vmcoreinfo = get_vmcoreinfo_old(&size);
488 if (!vmcoreinfo)
489 return 0;
491 kfree(vmcoreinfo);
492 return nt_size_name(size, name);
496 * Initialize final note (needed for /proc/vmcore code)
498 static void *nt_final(void *ptr)
500 Elf64_Nhdr *note;
502 note = (Elf64_Nhdr *) ptr;
503 note->n_namesz = 0;
504 note->n_descsz = 0;
505 note->n_type = 0;
506 return PTR_ADD(ptr, sizeof(Elf64_Nhdr));
510 * Initialize ELF header (new kernel)
512 static void *ehdr_init(Elf64_Ehdr *ehdr, int mem_chunk_cnt)
514 memset(ehdr, 0, sizeof(*ehdr));
515 memcpy(ehdr->e_ident, ELFMAG, SELFMAG);
516 ehdr->e_ident[EI_CLASS] = ELFCLASS64;
517 ehdr->e_ident[EI_DATA] = ELFDATA2MSB;
518 ehdr->e_ident[EI_VERSION] = EV_CURRENT;
519 memset(ehdr->e_ident + EI_PAD, 0, EI_NIDENT - EI_PAD);
520 ehdr->e_type = ET_CORE;
521 ehdr->e_machine = EM_S390;
522 ehdr->e_version = EV_CURRENT;
523 ehdr->e_phoff = sizeof(Elf64_Ehdr);
524 ehdr->e_ehsize = sizeof(Elf64_Ehdr);
525 ehdr->e_phentsize = sizeof(Elf64_Phdr);
526 ehdr->e_phnum = mem_chunk_cnt + 1;
527 return ehdr + 1;
531 * Return CPU count for ELF header (new kernel)
533 static int get_cpu_cnt(void)
535 struct save_area *sa;
536 int cpus = 0;
538 list_for_each_entry(sa, &dump_save_areas, list)
539 if (sa->prefix != 0)
540 cpus++;
541 return cpus;
545 * Return memory chunk count for ELF header (new kernel)
547 static int get_mem_chunk_cnt(void)
549 int cnt = 0;
550 u64 idx;
552 for_each_physmem_range(idx, &oldmem_type, NULL, NULL)
553 cnt++;
554 return cnt;
558 * Initialize ELF loads (new kernel)
560 static void loads_init(Elf64_Phdr *phdr, u64 loads_offset)
562 phys_addr_t start, end;
563 u64 idx;
565 for_each_physmem_range(idx, &oldmem_type, &start, &end) {
566 phdr->p_filesz = end - start;
567 phdr->p_type = PT_LOAD;
568 phdr->p_offset = start;
569 phdr->p_vaddr = start;
570 phdr->p_paddr = start;
571 phdr->p_memsz = end - start;
572 phdr->p_flags = PF_R | PF_W | PF_X;
573 phdr->p_align = PAGE_SIZE;
574 phdr++;
579 * Initialize notes (new kernel)
581 static void *notes_init(Elf64_Phdr *phdr, void *ptr, u64 notes_offset)
583 struct save_area *sa;
584 void *ptr_start = ptr;
585 int cpu;
587 ptr = nt_prpsinfo(ptr);
589 cpu = 1;
590 list_for_each_entry(sa, &dump_save_areas, list)
591 if (sa->prefix != 0)
592 ptr = fill_cpu_elf_notes(ptr, cpu++, sa);
593 ptr = nt_vmcoreinfo(ptr);
594 ptr = nt_final(ptr);
595 memset(phdr, 0, sizeof(*phdr));
596 phdr->p_type = PT_NOTE;
597 phdr->p_offset = notes_offset;
598 phdr->p_filesz = (unsigned long) PTR_SUB(ptr, ptr_start);
599 phdr->p_memsz = phdr->p_filesz;
600 return ptr;
603 static size_t get_elfcorehdr_size(int mem_chunk_cnt)
605 size_t size;
607 size = sizeof(Elf64_Ehdr);
608 /* PT_NOTES */
609 size += sizeof(Elf64_Phdr);
610 /* nt_prpsinfo */
611 size += nt_size(NT_PRPSINFO, sizeof(struct elf_prpsinfo));
612 /* regsets */
613 size += get_cpu_cnt() * get_cpu_elf_notes_size();
614 /* nt_vmcoreinfo */
615 size += nt_vmcoreinfo_size();
616 /* nt_final */
617 size += sizeof(Elf64_Nhdr);
618 /* PT_LOADS */
619 size += mem_chunk_cnt * sizeof(Elf64_Phdr);
621 return size;
625 * Create ELF core header (new kernel)
627 int elfcorehdr_alloc(unsigned long long *addr, unsigned long long *size)
629 Elf64_Phdr *phdr_notes, *phdr_loads;
630 int mem_chunk_cnt;
631 void *ptr, *hdr;
632 u32 alloc_size;
633 u64 hdr_off;
635 /* If we are not in kdump or zfcp/nvme dump mode return */
636 if (!OLDMEM_BASE && !is_ipl_type_dump())
637 return 0;
638 /* If we cannot get HSA size for zfcp/nvme dump return error */
639 if (is_ipl_type_dump() && !sclp.hsa_size)
640 return -ENODEV;
642 /* For kdump, exclude previous crashkernel memory */
643 if (OLDMEM_BASE) {
644 oldmem_region.base = OLDMEM_BASE;
645 oldmem_region.size = OLDMEM_SIZE;
646 oldmem_type.total_size = OLDMEM_SIZE;
649 mem_chunk_cnt = get_mem_chunk_cnt();
651 alloc_size = get_elfcorehdr_size(mem_chunk_cnt);
653 hdr = kzalloc(alloc_size, GFP_KERNEL);
655 /* Without elfcorehdr /proc/vmcore cannot be created. Thus creating
656 * a dump with this crash kernel will fail. Panic now to allow other
657 * dump mechanisms to take over.
659 if (!hdr)
660 panic("s390 kdump allocating elfcorehdr failed");
662 /* Init elf header */
663 ptr = ehdr_init(hdr, mem_chunk_cnt);
664 /* Init program headers */
665 phdr_notes = ptr;
666 ptr = PTR_ADD(ptr, sizeof(Elf64_Phdr));
667 phdr_loads = ptr;
668 ptr = PTR_ADD(ptr, sizeof(Elf64_Phdr) * mem_chunk_cnt);
669 /* Init notes */
670 hdr_off = PTR_DIFF(ptr, hdr);
671 ptr = notes_init(phdr_notes, ptr, ((unsigned long) hdr) + hdr_off);
672 /* Init loads */
673 hdr_off = PTR_DIFF(ptr, hdr);
674 loads_init(phdr_loads, hdr_off);
675 *addr = (unsigned long long) hdr;
676 *size = (unsigned long long) hdr_off;
677 BUG_ON(elfcorehdr_size > alloc_size);
678 return 0;
682 * Free ELF core header (new kernel)
684 void elfcorehdr_free(unsigned long long addr)
686 kfree((void *)(unsigned long)addr);
690 * Read from ELF header
692 ssize_t elfcorehdr_read(char *buf, size_t count, u64 *ppos)
694 void *src = (void *)(unsigned long)*ppos;
696 memcpy(buf, src, count);
697 *ppos += count;
698 return count;
702 * Read from ELF notes data
704 ssize_t elfcorehdr_read_notes(char *buf, size_t count, u64 *ppos)
706 void *src = (void *)(unsigned long)*ppos;
708 memcpy(buf, src, count);
709 *ppos += count;
710 return count;