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);
63 list_add(&sa->list, &dump_save_areas);
65 list_add_tail(&sa->list, &dump_save_areas);
70 * Return the address of the save area for the boot CPU
72 struct save_area * __init save_area_boot_cpu(void)
74 return list_first_entry_or_null(&dump_save_areas, struct save_area, list);
78 * Copy CPU registers into the save area
80 void __init save_area_add_regs(struct save_area *sa, void *regs)
84 lc = (struct lowcore *)(regs - __LC_FPREGS_SAVE_AREA);
85 memcpy(&sa->psw, &lc->psw_save_area, sizeof(sa->psw));
86 memcpy(&sa->ctrs, &lc->cregs_save_area, sizeof(sa->ctrs));
87 memcpy(&sa->gprs, &lc->gpregs_save_area, sizeof(sa->gprs));
88 memcpy(&sa->acrs, &lc->access_regs_save_area, sizeof(sa->acrs));
89 memcpy(&sa->fprs, &lc->floating_pt_save_area, sizeof(sa->fprs));
90 memcpy(&sa->fpc, &lc->fpt_creg_save_area, sizeof(sa->fpc));
91 memcpy(&sa->prefix, &lc->prefixreg_save_area, sizeof(sa->prefix));
92 memcpy(&sa->todpreg, &lc->tod_progreg_save_area, sizeof(sa->todpreg));
93 memcpy(&sa->timer, &lc->cpu_timer_save_area, sizeof(sa->timer));
94 memcpy(&sa->todcmp, &lc->clock_comp_save_area, sizeof(sa->todcmp));
98 * Copy vector registers into the save area
100 void __init save_area_add_vxrs(struct save_area *sa, __vector128 *vxrs)
104 /* Copy lower halves of vector registers 0-15 */
105 for (i = 0; i < 16; i++)
106 memcpy(&sa->vxrs_low[i], &vxrs[i].u[2], 8);
107 /* Copy vector registers 16-31 */
108 memcpy(sa->vxrs_high, vxrs + 16, 16 * sizeof(__vector128));
112 * Return physical address for virtual address
114 static inline void *load_real_addr(void *addr)
116 unsigned long real_addr;
123 : "=a" (real_addr) : "a" (addr) : "cc");
124 return (void *)real_addr;
128 * Copy memory of the old, dumped system to a kernel space virtual address
130 int copy_oldmem_kernel(void *dst, void *src, size_t count)
132 unsigned long from, len;
138 if (!OLDMEM_BASE && from < sclp.hsa_size) {
139 /* Copy from zfcpdump HSA area */
140 len = min(count, sclp.hsa_size - from);
141 rc = memcpy_hsa_kernel(dst, from, len);
145 /* Check for swapped kdump oldmem areas */
146 if (OLDMEM_BASE && from - OLDMEM_BASE < OLDMEM_SIZE) {
148 len = min(count, OLDMEM_SIZE - from);
149 } else if (OLDMEM_BASE && from < OLDMEM_SIZE) {
150 len = min(count, OLDMEM_SIZE - from);
155 if (is_vmalloc_or_module_addr(dst)) {
156 ra = load_real_addr(dst);
157 len = min(PAGE_SIZE - offset_in_page(ra), len);
161 if (memcpy_real(ra, (void *) from, len))
172 * Copy memory of the old, dumped system to a user space virtual address
174 static int copy_oldmem_user(void __user *dst, void *src, size_t count)
176 unsigned long from, len;
181 if (!OLDMEM_BASE && from < sclp.hsa_size) {
182 /* Copy from zfcpdump HSA area */
183 len = min(count, sclp.hsa_size - from);
184 rc = memcpy_hsa_user(dst, from, len);
188 /* Check for swapped kdump oldmem areas */
189 if (OLDMEM_BASE && from - OLDMEM_BASE < OLDMEM_SIZE) {
191 len = min(count, OLDMEM_SIZE - from);
192 } else if (OLDMEM_BASE && from < OLDMEM_SIZE) {
193 len = min(count, OLDMEM_SIZE - from);
198 rc = copy_to_user_real(dst, (void *) from, count);
210 * Copy one page from "oldmem"
212 ssize_t copy_oldmem_page(unsigned long pfn, char *buf, size_t csize,
213 unsigned long offset, int userbuf)
220 src = (void *) (pfn << PAGE_SHIFT) + offset;
222 rc = copy_oldmem_user((void __force __user *) buf, src, csize);
224 rc = copy_oldmem_kernel((void *) buf, src, csize);
229 * Remap "oldmem" for kdump
231 * For the kdump reserved memory this functions performs a swap operation:
232 * [0 - OLDMEM_SIZE] is mapped to [OLDMEM_BASE - OLDMEM_BASE + OLDMEM_SIZE]
234 static int remap_oldmem_pfn_range_kdump(struct vm_area_struct *vma,
235 unsigned long from, unsigned long pfn,
236 unsigned long size, pgprot_t prot)
238 unsigned long size_old;
241 if (pfn < OLDMEM_SIZE >> PAGE_SHIFT) {
242 size_old = min(size, OLDMEM_SIZE - (pfn << PAGE_SHIFT));
243 rc = remap_pfn_range(vma, from,
244 pfn + (OLDMEM_BASE >> PAGE_SHIFT),
246 if (rc || size == size_old)
250 pfn += size_old >> PAGE_SHIFT;
252 return remap_pfn_range(vma, from, pfn, size, prot);
256 * Remap "oldmem" for zfcpdump
258 * We only map available memory above HSA size. Memory below HSA size
259 * is read on demand using the copy_oldmem_page() function.
261 static int remap_oldmem_pfn_range_zfcpdump(struct vm_area_struct *vma,
264 unsigned long size, pgprot_t prot)
266 unsigned long hsa_end = sclp.hsa_size;
267 unsigned long size_hsa;
269 if (pfn < hsa_end >> PAGE_SHIFT) {
270 size_hsa = min(size, hsa_end - (pfn << PAGE_SHIFT));
271 if (size == size_hsa)
275 pfn += size_hsa >> PAGE_SHIFT;
277 return remap_pfn_range(vma, from, pfn, size, prot);
281 * Remap "oldmem" for kdump or zfcpdump
283 int remap_oldmem_pfn_range(struct vm_area_struct *vma, unsigned long from,
284 unsigned long pfn, unsigned long size, pgprot_t prot)
287 return remap_oldmem_pfn_range_kdump(vma, from, pfn, size, prot);
289 return remap_oldmem_pfn_range_zfcpdump(vma, from, pfn, size,
294 * Alloc memory and panic in case of ENOMEM
296 static void *kzalloc_panic(int len)
300 rc = kzalloc(len, GFP_KERNEL);
302 panic("s390 kdump kzalloc (%d) failed", len);
307 * Initialize ELF note
309 static void *nt_init_name(void *buf, Elf64_Word type, void *desc, int d_len,
315 note = (Elf64_Nhdr *)buf;
316 note->n_namesz = strlen(name) + 1;
317 note->n_descsz = d_len;
319 len = sizeof(Elf64_Nhdr);
321 memcpy(buf + len, name, note->n_namesz);
322 len = roundup(len + note->n_namesz, 4);
324 memcpy(buf + len, desc, note->n_descsz);
325 len = roundup(len + note->n_descsz, 4);
327 return PTR_ADD(buf, len);
330 static inline void *nt_init(void *buf, Elf64_Word type, void *desc, int d_len)
332 const char *note_name = "LINUX";
334 if (type == NT_PRPSINFO || type == NT_PRSTATUS || type == NT_PRFPREG)
335 note_name = KEXEC_CORE_NOTE_NAME;
336 return nt_init_name(buf, type, desc, d_len, 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_NOTE_NAME) != 0)
406 vmcoreinfo = kzalloc_panic(note.n_descsz);
407 if (copy_oldmem_kernel(vmcoreinfo, addr + 24, note.n_descsz)) {
411 *size = note.n_descsz;
416 * Initialize vmcoreinfo note (new kernel)
418 static void *nt_vmcoreinfo(void *ptr)
420 const char *name = VMCOREINFO_NOTE_NAME;
424 vmcoreinfo = os_info_old_entry(OS_INFO_VMCOREINFO, &size);
426 return nt_init_name(ptr, 0, vmcoreinfo, size, name);
428 vmcoreinfo = get_vmcoreinfo_old(&size);
431 ptr = nt_init_name(ptr, 0, vmcoreinfo, size, name);
437 * Initialize final note (needed for /proc/vmcore code)
439 static void *nt_final(void *ptr)
443 note = (Elf64_Nhdr *) ptr;
447 return PTR_ADD(ptr, sizeof(Elf64_Nhdr));
451 * Initialize ELF header (new kernel)
453 static void *ehdr_init(Elf64_Ehdr *ehdr, int mem_chunk_cnt)
455 memset(ehdr, 0, sizeof(*ehdr));
456 memcpy(ehdr->e_ident, ELFMAG, SELFMAG);
457 ehdr->e_ident[EI_CLASS] = ELFCLASS64;
458 ehdr->e_ident[EI_DATA] = ELFDATA2MSB;
459 ehdr->e_ident[EI_VERSION] = EV_CURRENT;
460 memset(ehdr->e_ident + EI_PAD, 0, EI_NIDENT - EI_PAD);
461 ehdr->e_type = ET_CORE;
462 ehdr->e_machine = EM_S390;
463 ehdr->e_version = EV_CURRENT;
464 ehdr->e_phoff = sizeof(Elf64_Ehdr);
465 ehdr->e_ehsize = sizeof(Elf64_Ehdr);
466 ehdr->e_phentsize = sizeof(Elf64_Phdr);
467 ehdr->e_phnum = mem_chunk_cnt + 1;
472 * Return CPU count for ELF header (new kernel)
474 static int get_cpu_cnt(void)
476 struct save_area *sa;
479 list_for_each_entry(sa, &dump_save_areas, list)
486 * Return memory chunk count for ELF header (new kernel)
488 static int get_mem_chunk_cnt(void)
493 for_each_mem_range(idx, &memblock.physmem, &oldmem_type, NUMA_NO_NODE,
494 MEMBLOCK_NONE, NULL, NULL, NULL)
500 * Initialize ELF loads (new kernel)
502 static void loads_init(Elf64_Phdr *phdr, u64 loads_offset)
504 phys_addr_t start, end;
507 for_each_mem_range(idx, &memblock.physmem, &oldmem_type, NUMA_NO_NODE,
508 MEMBLOCK_NONE, &start, &end, NULL) {
509 phdr->p_filesz = end - start;
510 phdr->p_type = PT_LOAD;
511 phdr->p_offset = start;
512 phdr->p_vaddr = start;
513 phdr->p_paddr = start;
514 phdr->p_memsz = end - start;
515 phdr->p_flags = PF_R | PF_W | PF_X;
516 phdr->p_align = PAGE_SIZE;
522 * Initialize notes (new kernel)
524 static void *notes_init(Elf64_Phdr *phdr, void *ptr, u64 notes_offset)
526 struct save_area *sa;
527 void *ptr_start = ptr;
530 ptr = nt_prpsinfo(ptr);
533 list_for_each_entry(sa, &dump_save_areas, list)
535 ptr = fill_cpu_elf_notes(ptr, cpu++, sa);
536 ptr = nt_vmcoreinfo(ptr);
538 memset(phdr, 0, sizeof(*phdr));
539 phdr->p_type = PT_NOTE;
540 phdr->p_offset = notes_offset;
541 phdr->p_filesz = (unsigned long) PTR_SUB(ptr, ptr_start);
542 phdr->p_memsz = phdr->p_filesz;
547 * Create ELF core header (new kernel)
549 int elfcorehdr_alloc(unsigned long long *addr, unsigned long long *size)
551 Elf64_Phdr *phdr_notes, *phdr_loads;
557 /* If we are not in kdump or zfcpdump mode return */
558 if (!OLDMEM_BASE && ipl_info.type != IPL_TYPE_FCP_DUMP)
560 /* If we cannot get HSA size for zfcpdump return error */
561 if (ipl_info.type == IPL_TYPE_FCP_DUMP && !sclp.hsa_size)
564 /* For kdump, exclude previous crashkernel memory */
566 oldmem_region.base = OLDMEM_BASE;
567 oldmem_region.size = OLDMEM_SIZE;
568 oldmem_type.total_size = OLDMEM_SIZE;
571 mem_chunk_cnt = get_mem_chunk_cnt();
573 alloc_size = 0x1000 + get_cpu_cnt() * 0x4a0 +
574 mem_chunk_cnt * sizeof(Elf64_Phdr);
575 hdr = kzalloc_panic(alloc_size);
576 /* Init elf header */
577 ptr = ehdr_init(hdr, mem_chunk_cnt);
578 /* Init program headers */
580 ptr = PTR_ADD(ptr, sizeof(Elf64_Phdr));
582 ptr = PTR_ADD(ptr, sizeof(Elf64_Phdr) * mem_chunk_cnt);
584 hdr_off = PTR_DIFF(ptr, hdr);
585 ptr = notes_init(phdr_notes, ptr, ((unsigned long) hdr) + hdr_off);
587 hdr_off = PTR_DIFF(ptr, hdr);
588 loads_init(phdr_loads, hdr_off);
589 *addr = (unsigned long long) hdr;
590 *size = (unsigned long long) hdr_off;
591 BUG_ON(elfcorehdr_size > alloc_size);
596 * Free ELF core header (new kernel)
598 void elfcorehdr_free(unsigned long long addr)
600 kfree((void *)(unsigned long)addr);
604 * Read from ELF header
606 ssize_t elfcorehdr_read(char *buf, size_t count, u64 *ppos)
608 void *src = (void *)(unsigned long)*ppos;
610 memcpy(buf, src, count);
616 * Read from ELF notes data
618 ssize_t elfcorehdr_read_notes(char *buf, size_t count, u64 *ppos)
620 void *src = (void *)(unsigned long)*ppos;
622 memcpy(buf, src, count);