2 * S390 kdump implementation
4 * Copyright IBM Corp. 2011
5 * Author(s): Michael Holzheu <holzheu@linux.vnet.ibm.com>
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/asm-offsets.h>
17 #include <linux/memblock.h>
18 #include <asm/os_info.h>
23 #define PTR_ADD(x, y) (((char *) (x)) + ((unsigned long) (y)))
24 #define PTR_SUB(x, y) (((char *) (x)) - ((unsigned long) (y)))
25 #define PTR_DIFF(x, y) ((unsigned long)(((char *) (x)) - ((unsigned long) (y))))
27 static struct memblock_region oldmem_region;
29 static struct memblock_type oldmem_type = {
33 .regions = &oldmem_region,
37 struct list_head list;
49 __vector128 vxrs_high[16];
52 static LIST_HEAD(dump_save_areas);
55 * Allocate a save area
57 struct save_area * __init save_area_alloc(bool is_boot_cpu)
61 sa = (void *) memblock_alloc(sizeof(*sa), 8);
65 list_add(&sa->list, &dump_save_areas);
67 list_add_tail(&sa->list, &dump_save_areas);
72 * Return the address of the save area for the boot CPU
74 struct save_area * __init save_area_boot_cpu(void)
76 if (list_empty(&dump_save_areas))
78 return list_first_entry(&dump_save_areas, struct save_area, list);
82 * Copy CPU registers into the save area
84 void __init save_area_add_regs(struct save_area *sa, void *regs)
88 lc = (struct lowcore *)(regs - __LC_FPREGS_SAVE_AREA);
89 memcpy(&sa->psw, &lc->psw_save_area, sizeof(sa->psw));
90 memcpy(&sa->ctrs, &lc->cregs_save_area, sizeof(sa->ctrs));
91 memcpy(&sa->gprs, &lc->gpregs_save_area, sizeof(sa->gprs));
92 memcpy(&sa->acrs, &lc->access_regs_save_area, sizeof(sa->acrs));
93 memcpy(&sa->fprs, &lc->floating_pt_save_area, sizeof(sa->fprs));
94 memcpy(&sa->fpc, &lc->fpt_creg_save_area, sizeof(sa->fpc));
95 memcpy(&sa->prefix, &lc->prefixreg_save_area, sizeof(sa->prefix));
96 memcpy(&sa->todpreg, &lc->tod_progreg_save_area, sizeof(sa->todpreg));
97 memcpy(&sa->timer, &lc->cpu_timer_save_area, sizeof(sa->timer));
98 memcpy(&sa->todcmp, &lc->clock_comp_save_area, sizeof(sa->todcmp));
102 * Copy vector registers into the save area
104 void __init save_area_add_vxrs(struct save_area *sa, __vector128 *vxrs)
108 /* Copy lower halves of vector registers 0-15 */
109 for (i = 0; i < 16; i++)
110 memcpy(&sa->vxrs_low[i], &vxrs[i].u[2], 8);
111 /* Copy vector registers 16-31 */
112 memcpy(sa->vxrs_high, vxrs + 16, 16 * sizeof(__vector128));
116 * Return physical address for virtual address
118 static inline void *load_real_addr(void *addr)
120 unsigned long real_addr;
127 : "=a" (real_addr) : "a" (addr) : "cc");
128 return (void *)real_addr;
132 * Copy memory of the old, dumped system to a kernel space virtual address
134 int copy_oldmem_kernel(void *dst, void *src, size_t count)
136 unsigned long from, len;
142 if (!OLDMEM_BASE && from < sclp.hsa_size) {
143 /* Copy from zfcpdump HSA area */
144 len = min(count, sclp.hsa_size - from);
145 rc = memcpy_hsa_kernel(dst, from, len);
149 /* Check for swapped kdump oldmem areas */
150 if (OLDMEM_BASE && from - OLDMEM_BASE < OLDMEM_SIZE) {
152 len = min(count, OLDMEM_SIZE - from);
153 } else if (OLDMEM_BASE && from < OLDMEM_SIZE) {
154 len = min(count, OLDMEM_SIZE - from);
159 if (is_vmalloc_or_module_addr(dst)) {
160 ra = load_real_addr(dst);
161 len = min(PAGE_SIZE - offset_in_page(ra), len);
165 if (memcpy_real(ra, (void *) from, len))
176 * Copy memory of the old, dumped system to a user space virtual address
178 int copy_oldmem_user(void __user *dst, void *src, size_t count)
180 unsigned long from, len;
185 if (!OLDMEM_BASE && from < sclp.hsa_size) {
186 /* Copy from zfcpdump HSA area */
187 len = min(count, sclp.hsa_size - from);
188 rc = memcpy_hsa_user(dst, from, len);
192 /* Check for swapped kdump oldmem areas */
193 if (OLDMEM_BASE && from - OLDMEM_BASE < OLDMEM_SIZE) {
195 len = min(count, OLDMEM_SIZE - from);
196 } else if (OLDMEM_BASE && from < OLDMEM_SIZE) {
197 len = min(count, OLDMEM_SIZE - from);
202 rc = copy_to_user_real(dst, (void *) from, count);
214 * Copy one page from "oldmem"
216 ssize_t copy_oldmem_page(unsigned long pfn, char *buf, size_t csize,
217 unsigned long offset, int userbuf)
224 src = (void *) (pfn << PAGE_SHIFT) + offset;
226 rc = copy_oldmem_user((void __force __user *) buf, src, csize);
228 rc = copy_oldmem_kernel((void *) buf, src, csize);
233 * Remap "oldmem" for kdump
235 * For the kdump reserved memory this functions performs a swap operation:
236 * [0 - OLDMEM_SIZE] is mapped to [OLDMEM_BASE - OLDMEM_BASE + OLDMEM_SIZE]
238 static int remap_oldmem_pfn_range_kdump(struct vm_area_struct *vma,
239 unsigned long from, unsigned long pfn,
240 unsigned long size, pgprot_t prot)
242 unsigned long size_old;
245 if (pfn < OLDMEM_SIZE >> PAGE_SHIFT) {
246 size_old = min(size, OLDMEM_SIZE - (pfn << PAGE_SHIFT));
247 rc = remap_pfn_range(vma, from,
248 pfn + (OLDMEM_BASE >> PAGE_SHIFT),
250 if (rc || size == size_old)
254 pfn += size_old >> PAGE_SHIFT;
256 return remap_pfn_range(vma, from, pfn, size, prot);
260 * Remap "oldmem" for zfcpdump
262 * We only map available memory above HSA size. Memory below HSA size
263 * is read on demand using the copy_oldmem_page() function.
265 static int remap_oldmem_pfn_range_zfcpdump(struct vm_area_struct *vma,
268 unsigned long size, pgprot_t prot)
270 unsigned long hsa_end = sclp.hsa_size;
271 unsigned long size_hsa;
273 if (pfn < hsa_end >> PAGE_SHIFT) {
274 size_hsa = min(size, hsa_end - (pfn << PAGE_SHIFT));
275 if (size == size_hsa)
279 pfn += size_hsa >> PAGE_SHIFT;
281 return remap_pfn_range(vma, from, pfn, size, prot);
285 * Remap "oldmem" for kdump or zfcpdump
287 int remap_oldmem_pfn_range(struct vm_area_struct *vma, unsigned long from,
288 unsigned long pfn, unsigned long size, pgprot_t prot)
291 return remap_oldmem_pfn_range_kdump(vma, from, pfn, size, prot);
293 return remap_oldmem_pfn_range_zfcpdump(vma, from, pfn, size,
298 * Alloc memory and panic in case of ENOMEM
300 static void *kzalloc_panic(int len)
304 rc = kzalloc(len, GFP_KERNEL);
306 panic("s390 kdump kzalloc (%d) failed", len);
311 * Initialize ELF note
313 static void *nt_init_name(void *buf, Elf64_Word type, void *desc, int d_len,
319 note = (Elf64_Nhdr *)buf;
320 note->n_namesz = strlen(name) + 1;
321 note->n_descsz = d_len;
323 len = sizeof(Elf64_Nhdr);
325 memcpy(buf + len, name, note->n_namesz);
326 len = roundup(len + note->n_namesz, 4);
328 memcpy(buf + len, desc, note->n_descsz);
329 len = roundup(len + note->n_descsz, 4);
331 return PTR_ADD(buf, len);
334 static inline void *nt_init(void *buf, Elf64_Word type, void *desc, int d_len)
336 return nt_init_name(buf, type, desc, d_len, KEXEC_CORE_NOTE_NAME);
340 * Fill ELF notes for one CPU with save area registers
342 static void *fill_cpu_elf_notes(void *ptr, int cpu, struct save_area *sa)
344 struct elf_prstatus nt_prstatus;
345 elf_fpregset_t nt_fpregset;
347 /* Prepare prstatus note */
348 memset(&nt_prstatus, 0, sizeof(nt_prstatus));
349 memcpy(&nt_prstatus.pr_reg.gprs, sa->gprs, sizeof(sa->gprs));
350 memcpy(&nt_prstatus.pr_reg.psw, sa->psw, sizeof(sa->psw));
351 memcpy(&nt_prstatus.pr_reg.acrs, sa->acrs, sizeof(sa->acrs));
352 nt_prstatus.pr_pid = cpu;
353 /* Prepare fpregset (floating point) note */
354 memset(&nt_fpregset, 0, sizeof(nt_fpregset));
355 memcpy(&nt_fpregset.fpc, &sa->fpc, sizeof(sa->fpc));
356 memcpy(&nt_fpregset.fprs, &sa->fprs, sizeof(sa->fprs));
357 /* Create ELF notes for the CPU */
358 ptr = nt_init(ptr, NT_PRSTATUS, &nt_prstatus, sizeof(nt_prstatus));
359 ptr = nt_init(ptr, NT_PRFPREG, &nt_fpregset, sizeof(nt_fpregset));
360 ptr = nt_init(ptr, NT_S390_TIMER, &sa->timer, sizeof(sa->timer));
361 ptr = nt_init(ptr, NT_S390_TODCMP, &sa->todcmp, sizeof(sa->todcmp));
362 ptr = nt_init(ptr, NT_S390_TODPREG, &sa->todpreg, sizeof(sa->todpreg));
363 ptr = nt_init(ptr, NT_S390_CTRS, &sa->ctrs, sizeof(sa->ctrs));
364 ptr = nt_init(ptr, NT_S390_PREFIX, &sa->prefix, sizeof(sa->prefix));
365 if (MACHINE_HAS_VX) {
366 ptr = nt_init(ptr, NT_S390_VXRS_HIGH,
367 &sa->vxrs_high, sizeof(sa->vxrs_high));
368 ptr = nt_init(ptr, NT_S390_VXRS_LOW,
369 &sa->vxrs_low, sizeof(sa->vxrs_low));
375 * Initialize prpsinfo note (new kernel)
377 static void *nt_prpsinfo(void *ptr)
379 struct elf_prpsinfo prpsinfo;
381 memset(&prpsinfo, 0, sizeof(prpsinfo));
382 prpsinfo.pr_sname = 'R';
383 strcpy(prpsinfo.pr_fname, "vmlinux");
384 return nt_init(ptr, NT_PRPSINFO, &prpsinfo, sizeof(prpsinfo));
388 * Get vmcoreinfo using lowcore->vmcore_info (new kernel)
390 static void *get_vmcoreinfo_old(unsigned long *size)
392 char nt_name[11], *vmcoreinfo;
396 if (copy_oldmem_kernel(&addr, &S390_lowcore.vmcore_info, sizeof(addr)))
398 memset(nt_name, 0, sizeof(nt_name));
399 if (copy_oldmem_kernel(¬e, addr, sizeof(note)))
401 if (copy_oldmem_kernel(nt_name, addr + sizeof(note),
402 sizeof(nt_name) - 1))
404 if (strcmp(nt_name, "VMCOREINFO") != 0)
406 vmcoreinfo = kzalloc_panic(note.n_descsz);
407 if (copy_oldmem_kernel(vmcoreinfo, addr + 24, note.n_descsz))
409 *size = note.n_descsz;
414 * Initialize vmcoreinfo note (new kernel)
416 static void *nt_vmcoreinfo(void *ptr)
421 vmcoreinfo = os_info_old_entry(OS_INFO_VMCOREINFO, &size);
423 vmcoreinfo = get_vmcoreinfo_old(&size);
426 return nt_init_name(ptr, 0, vmcoreinfo, size, "VMCOREINFO");
430 * Initialize ELF header (new kernel)
432 static void *ehdr_init(Elf64_Ehdr *ehdr, int mem_chunk_cnt)
434 memset(ehdr, 0, sizeof(*ehdr));
435 memcpy(ehdr->e_ident, ELFMAG, SELFMAG);
436 ehdr->e_ident[EI_CLASS] = ELFCLASS64;
437 ehdr->e_ident[EI_DATA] = ELFDATA2MSB;
438 ehdr->e_ident[EI_VERSION] = EV_CURRENT;
439 memset(ehdr->e_ident + EI_PAD, 0, EI_NIDENT - EI_PAD);
440 ehdr->e_type = ET_CORE;
441 ehdr->e_machine = EM_S390;
442 ehdr->e_version = EV_CURRENT;
443 ehdr->e_phoff = sizeof(Elf64_Ehdr);
444 ehdr->e_ehsize = sizeof(Elf64_Ehdr);
445 ehdr->e_phentsize = sizeof(Elf64_Phdr);
446 ehdr->e_phnum = mem_chunk_cnt + 1;
451 * Return CPU count for ELF header (new kernel)
453 static int get_cpu_cnt(void)
455 struct save_area *sa;
458 list_for_each_entry(sa, &dump_save_areas, list)
465 * Return memory chunk count for ELF header (new kernel)
467 static int get_mem_chunk_cnt(void)
472 for_each_mem_range(idx, &memblock.physmem, &oldmem_type, NUMA_NO_NODE,
473 MEMBLOCK_NONE, NULL, NULL, NULL)
479 * Initialize ELF loads (new kernel)
481 static void loads_init(Elf64_Phdr *phdr, u64 loads_offset)
483 phys_addr_t start, end;
486 for_each_mem_range(idx, &memblock.physmem, &oldmem_type, NUMA_NO_NODE,
487 MEMBLOCK_NONE, &start, &end, NULL) {
488 phdr->p_filesz = end - start;
489 phdr->p_type = PT_LOAD;
490 phdr->p_offset = start;
491 phdr->p_vaddr = start;
492 phdr->p_paddr = start;
493 phdr->p_memsz = end - start;
494 phdr->p_flags = PF_R | PF_W | PF_X;
495 phdr->p_align = PAGE_SIZE;
501 * Initialize notes (new kernel)
503 static void *notes_init(Elf64_Phdr *phdr, void *ptr, u64 notes_offset)
505 struct save_area *sa;
506 void *ptr_start = ptr;
509 ptr = nt_prpsinfo(ptr);
512 list_for_each_entry(sa, &dump_save_areas, list)
514 ptr = fill_cpu_elf_notes(ptr, cpu++, sa);
515 ptr = nt_vmcoreinfo(ptr);
516 memset(phdr, 0, sizeof(*phdr));
517 phdr->p_type = PT_NOTE;
518 phdr->p_offset = notes_offset;
519 phdr->p_filesz = (unsigned long) PTR_SUB(ptr, ptr_start);
520 phdr->p_memsz = phdr->p_filesz;
525 * Create ELF core header (new kernel)
527 int elfcorehdr_alloc(unsigned long long *addr, unsigned long long *size)
529 Elf64_Phdr *phdr_notes, *phdr_loads;
535 /* If we are not in kdump or zfcpdump mode return */
536 if (!OLDMEM_BASE && ipl_info.type != IPL_TYPE_FCP_DUMP)
538 /* If we cannot get HSA size for zfcpdump return error */
539 if (ipl_info.type == IPL_TYPE_FCP_DUMP && !sclp.hsa_size)
542 /* For kdump, exclude previous crashkernel memory */
544 oldmem_region.base = OLDMEM_BASE;
545 oldmem_region.size = OLDMEM_SIZE;
546 oldmem_type.total_size = OLDMEM_SIZE;
549 mem_chunk_cnt = get_mem_chunk_cnt();
551 alloc_size = 0x1000 + get_cpu_cnt() * 0x4a0 +
552 mem_chunk_cnt * sizeof(Elf64_Phdr);
553 hdr = kzalloc_panic(alloc_size);
554 /* Init elf header */
555 ptr = ehdr_init(hdr, mem_chunk_cnt);
556 /* Init program headers */
558 ptr = PTR_ADD(ptr, sizeof(Elf64_Phdr));
560 ptr = PTR_ADD(ptr, sizeof(Elf64_Phdr) * mem_chunk_cnt);
562 hdr_off = PTR_DIFF(ptr, hdr);
563 ptr = notes_init(phdr_notes, ptr, ((unsigned long) hdr) + hdr_off);
565 hdr_off = PTR_DIFF(ptr, hdr);
566 loads_init(phdr_loads, hdr_off);
567 *addr = (unsigned long long) hdr;
568 *size = (unsigned long long) hdr_off;
569 BUG_ON(elfcorehdr_size > alloc_size);
574 * Free ELF core header (new kernel)
576 void elfcorehdr_free(unsigned long long addr)
578 kfree((void *)(unsigned long)addr);
582 * Read from ELF header
584 ssize_t elfcorehdr_read(char *buf, size_t count, u64 *ppos)
586 void *src = (void *)(unsigned long)*ppos;
588 memcpy(buf, src, count);
594 * Read from ELF notes data
596 ssize_t elfcorehdr_read_notes(char *buf, size_t count, u64 *ppos)
598 void *src = (void *)(unsigned long)*ppos;
600 memcpy(buf, src, count);