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 <linux/memblock.h>
17 #include <asm/os_info.h>
22 #define PTR_ADD(x, y) (((char *) (x)) + ((unsigned long) (y)))
23 #define PTR_SUB(x, y) (((char *) (x)) - ((unsigned long) (y)))
24 #define PTR_DIFF(x, y) ((unsigned long)(((char *) (x)) - ((unsigned long) (y))))
26 static struct memblock_region oldmem_region
;
28 static struct memblock_type oldmem_type
= {
32 .regions
= &oldmem_region
,
35 #define for_each_dump_mem_range(i, nid, p_start, p_end, p_nid) \
36 for (i = 0, __next_mem_range(&i, nid, MEMBLOCK_NONE, \
38 &oldmem_type, p_start, \
40 i != (u64)ULLONG_MAX; \
41 __next_mem_range(&i, nid, MEMBLOCK_NONE, &memblock.physmem,\
43 p_start, p_end, p_nid))
45 struct dump_save_areas dump_save_areas
;
48 * Return physical address for virtual address
50 static inline void *load_real_addr(void *addr
)
52 unsigned long real_addr
;
59 : "=a" (real_addr
) : "a" (addr
) : "cc");
60 return (void *)real_addr
;
64 * Copy real to virtual or real memory
66 static int copy_from_realmem(void *dest
, void *src
, size_t count
)
72 if (!is_vmalloc_or_module_addr(dest
))
73 return memcpy_real(dest
, src
, count
);
75 size
= min(count
, PAGE_SIZE
- (__pa(dest
) & ~PAGE_MASK
));
76 if (memcpy_real(load_real_addr(dest
), src
, size
))
86 * Pointer to ELF header in new kernel
88 static void *elfcorehdr_newmem
;
91 * Copy one page from zfcpdump "oldmem"
93 * For pages below HSA size memory from the HSA is copied. Otherwise
94 * real memory copy is used.
96 static ssize_t
copy_oldmem_page_zfcpdump(char *buf
, size_t csize
,
97 unsigned long src
, int userbuf
)
101 if (src
< sclp
.hsa_size
) {
102 rc
= memcpy_hsa(buf
, src
, csize
, userbuf
);
105 rc
= copy_to_user_real((void __force __user
*) buf
,
106 (void *) src
, csize
);
108 rc
= memcpy_real(buf
, (void *) src
, csize
);
110 return rc
? rc
: csize
;
114 * Copy one page from kdump "oldmem"
116 * For the kdump reserved memory this functions performs a swap operation:
117 * - [OLDMEM_BASE - OLDMEM_BASE + OLDMEM_SIZE] is mapped to [0 - OLDMEM_SIZE].
118 * - [0 - OLDMEM_SIZE] is mapped to [OLDMEM_BASE - OLDMEM_BASE + OLDMEM_SIZE]
120 static ssize_t
copy_oldmem_page_kdump(char *buf
, size_t csize
,
121 unsigned long src
, int userbuf
)
126 if (src
< OLDMEM_SIZE
)
128 else if (src
> OLDMEM_BASE
&&
129 src
< OLDMEM_BASE
+ OLDMEM_SIZE
)
132 rc
= copy_to_user_real((void __force __user
*) buf
,
133 (void *) src
, csize
);
135 rc
= copy_from_realmem(buf
, (void *) src
, csize
);
136 return (rc
== 0) ? rc
: csize
;
140 * Copy one page from "oldmem"
142 ssize_t
copy_oldmem_page(unsigned long pfn
, char *buf
, size_t csize
,
143 unsigned long offset
, int userbuf
)
149 src
= (pfn
<< PAGE_SHIFT
) + offset
;
151 return copy_oldmem_page_kdump(buf
, csize
, src
, userbuf
);
153 return copy_oldmem_page_zfcpdump(buf
, csize
, src
, userbuf
);
157 * Remap "oldmem" for kdump
159 * For the kdump reserved memory this functions performs a swap operation:
160 * [0 - OLDMEM_SIZE] is mapped to [OLDMEM_BASE - OLDMEM_BASE + OLDMEM_SIZE]
162 static int remap_oldmem_pfn_range_kdump(struct vm_area_struct
*vma
,
163 unsigned long from
, unsigned long pfn
,
164 unsigned long size
, pgprot_t prot
)
166 unsigned long size_old
;
169 if (pfn
< OLDMEM_SIZE
>> PAGE_SHIFT
) {
170 size_old
= min(size
, OLDMEM_SIZE
- (pfn
<< PAGE_SHIFT
));
171 rc
= remap_pfn_range(vma
, from
,
172 pfn
+ (OLDMEM_BASE
>> PAGE_SHIFT
),
174 if (rc
|| size
== size_old
)
178 pfn
+= size_old
>> PAGE_SHIFT
;
180 return remap_pfn_range(vma
, from
, pfn
, size
, prot
);
184 * Remap "oldmem" for zfcpdump
186 * We only map available memory above HSA size. Memory below HSA size
187 * is read on demand using the copy_oldmem_page() function.
189 static int remap_oldmem_pfn_range_zfcpdump(struct vm_area_struct
*vma
,
192 unsigned long size
, pgprot_t prot
)
194 unsigned long hsa_end
= sclp
.hsa_size
;
195 unsigned long size_hsa
;
197 if (pfn
< hsa_end
>> PAGE_SHIFT
) {
198 size_hsa
= min(size
, hsa_end
- (pfn
<< PAGE_SHIFT
));
199 if (size
== size_hsa
)
203 pfn
+= size_hsa
>> PAGE_SHIFT
;
205 return remap_pfn_range(vma
, from
, pfn
, size
, prot
);
209 * Remap "oldmem" for kdump or zfcpdump
211 int remap_oldmem_pfn_range(struct vm_area_struct
*vma
, unsigned long from
,
212 unsigned long pfn
, unsigned long size
, pgprot_t prot
)
215 return remap_oldmem_pfn_range_kdump(vma
, from
, pfn
, size
, prot
);
217 return remap_oldmem_pfn_range_zfcpdump(vma
, from
, pfn
, size
,
222 * Copy memory from old kernel
224 int copy_from_oldmem(void *dest
, void *src
, size_t count
)
226 unsigned long copied
= 0;
230 if ((unsigned long) src
< OLDMEM_SIZE
) {
231 copied
= min(count
, OLDMEM_SIZE
- (unsigned long) src
);
232 rc
= copy_from_realmem(dest
, src
+ OLDMEM_BASE
, copied
);
237 unsigned long hsa_end
= sclp
.hsa_size
;
238 if ((unsigned long) src
< hsa_end
) {
239 copied
= min(count
, hsa_end
- (unsigned long) src
);
240 rc
= memcpy_hsa(dest
, (unsigned long) src
, copied
, 0);
245 return copy_from_realmem(dest
+ copied
, src
+ copied
, count
- copied
);
249 * Alloc memory and panic in case of ENOMEM
251 static void *kzalloc_panic(int len
)
255 rc
= kzalloc(len
, GFP_KERNEL
);
257 panic("s390 kdump kzalloc (%d) failed", len
);
262 * Initialize ELF note
264 static void *nt_init(void *buf
, Elf64_Word type
, void *desc
, int d_len
,
270 note
= (Elf64_Nhdr
*)buf
;
271 note
->n_namesz
= strlen(name
) + 1;
272 note
->n_descsz
= d_len
;
274 len
= sizeof(Elf64_Nhdr
);
276 memcpy(buf
+ len
, name
, note
->n_namesz
);
277 len
= roundup(len
+ note
->n_namesz
, 4);
279 memcpy(buf
+ len
, desc
, note
->n_descsz
);
280 len
= roundup(len
+ note
->n_descsz
, 4);
282 return PTR_ADD(buf
, len
);
286 * Initialize prstatus note
288 static void *nt_prstatus(void *ptr
, struct save_area
*sa
)
290 struct elf_prstatus nt_prstatus
;
291 static int cpu_nr
= 1;
293 memset(&nt_prstatus
, 0, sizeof(nt_prstatus
));
294 memcpy(&nt_prstatus
.pr_reg
.gprs
, sa
->gp_regs
, sizeof(sa
->gp_regs
));
295 memcpy(&nt_prstatus
.pr_reg
.psw
, sa
->psw
, sizeof(sa
->psw
));
296 memcpy(&nt_prstatus
.pr_reg
.acrs
, sa
->acc_regs
, sizeof(sa
->acc_regs
));
297 nt_prstatus
.pr_pid
= cpu_nr
;
300 return nt_init(ptr
, NT_PRSTATUS
, &nt_prstatus
, sizeof(nt_prstatus
),
305 * Initialize fpregset (floating point) note
307 static void *nt_fpregset(void *ptr
, struct save_area
*sa
)
309 elf_fpregset_t nt_fpregset
;
311 memset(&nt_fpregset
, 0, sizeof(nt_fpregset
));
312 memcpy(&nt_fpregset
.fpc
, &sa
->fp_ctrl_reg
, sizeof(sa
->fp_ctrl_reg
));
313 memcpy(&nt_fpregset
.fprs
, &sa
->fp_regs
, sizeof(sa
->fp_regs
));
315 return nt_init(ptr
, NT_PRFPREG
, &nt_fpregset
, sizeof(nt_fpregset
),
320 * Initialize timer note
322 static void *nt_s390_timer(void *ptr
, struct save_area
*sa
)
324 return nt_init(ptr
, NT_S390_TIMER
, &sa
->timer
, sizeof(sa
->timer
),
325 KEXEC_CORE_NOTE_NAME
);
329 * Initialize TOD clock comparator note
331 static void *nt_s390_tod_cmp(void *ptr
, struct save_area
*sa
)
333 return nt_init(ptr
, NT_S390_TODCMP
, &sa
->clk_cmp
,
334 sizeof(sa
->clk_cmp
), KEXEC_CORE_NOTE_NAME
);
338 * Initialize TOD programmable register note
340 static void *nt_s390_tod_preg(void *ptr
, struct save_area
*sa
)
342 return nt_init(ptr
, NT_S390_TODPREG
, &sa
->tod_reg
,
343 sizeof(sa
->tod_reg
), KEXEC_CORE_NOTE_NAME
);
347 * Initialize control register note
349 static void *nt_s390_ctrs(void *ptr
, struct save_area
*sa
)
351 return nt_init(ptr
, NT_S390_CTRS
, &sa
->ctrl_regs
,
352 sizeof(sa
->ctrl_regs
), KEXEC_CORE_NOTE_NAME
);
356 * Initialize prefix register note
358 static void *nt_s390_prefix(void *ptr
, struct save_area
*sa
)
360 return nt_init(ptr
, NT_S390_PREFIX
, &sa
->pref_reg
,
361 sizeof(sa
->pref_reg
), KEXEC_CORE_NOTE_NAME
);
365 * Initialize vxrs high note (full 128 bit VX registers 16-31)
367 static void *nt_s390_vx_high(void *ptr
, __vector128
*vx_regs
)
369 return nt_init(ptr
, NT_S390_VXRS_HIGH
, &vx_regs
[16],
370 16 * sizeof(__vector128
), KEXEC_CORE_NOTE_NAME
);
374 * Initialize vxrs low note (lower halves of VX registers 0-15)
376 static void *nt_s390_vx_low(void *ptr
, __vector128
*vx_regs
)
382 note
= (Elf64_Nhdr
*)ptr
;
383 note
->n_namesz
= strlen(KEXEC_CORE_NOTE_NAME
) + 1;
384 note
->n_descsz
= 16 * 8;
385 note
->n_type
= NT_S390_VXRS_LOW
;
386 len
= sizeof(Elf64_Nhdr
);
388 memcpy(ptr
+ len
, KEXEC_CORE_NOTE_NAME
, note
->n_namesz
);
389 len
= roundup(len
+ note
->n_namesz
, 4);
392 /* Copy lower halves of SIMD registers 0-15 */
393 for (i
= 0; i
< 16; i
++) {
394 memcpy(ptr
, &vx_regs
[i
].u
[2], 8);
401 * Fill ELF notes for one CPU with save area registers
403 void *fill_cpu_elf_notes(void *ptr
, struct save_area
*sa
, __vector128
*vx_regs
)
405 ptr
= nt_prstatus(ptr
, sa
);
406 ptr
= nt_fpregset(ptr
, sa
);
407 ptr
= nt_s390_timer(ptr
, sa
);
408 ptr
= nt_s390_tod_cmp(ptr
, sa
);
409 ptr
= nt_s390_tod_preg(ptr
, sa
);
410 ptr
= nt_s390_ctrs(ptr
, sa
);
411 ptr
= nt_s390_prefix(ptr
, sa
);
412 if (MACHINE_HAS_VX
&& vx_regs
) {
413 ptr
= nt_s390_vx_low(ptr
, vx_regs
);
414 ptr
= nt_s390_vx_high(ptr
, vx_regs
);
420 * Initialize prpsinfo note (new kernel)
422 static void *nt_prpsinfo(void *ptr
)
424 struct elf_prpsinfo prpsinfo
;
426 memset(&prpsinfo
, 0, sizeof(prpsinfo
));
427 prpsinfo
.pr_sname
= 'R';
428 strcpy(prpsinfo
.pr_fname
, "vmlinux");
429 return nt_init(ptr
, NT_PRPSINFO
, &prpsinfo
, sizeof(prpsinfo
),
430 KEXEC_CORE_NOTE_NAME
);
434 * Get vmcoreinfo using lowcore->vmcore_info (new kernel)
436 static void *get_vmcoreinfo_old(unsigned long *size
)
438 char nt_name
[11], *vmcoreinfo
;
442 if (copy_from_oldmem(&addr
, &S390_lowcore
.vmcore_info
, sizeof(addr
)))
444 memset(nt_name
, 0, sizeof(nt_name
));
445 if (copy_from_oldmem(¬e
, addr
, sizeof(note
)))
447 if (copy_from_oldmem(nt_name
, addr
+ sizeof(note
), sizeof(nt_name
) - 1))
449 if (strcmp(nt_name
, "VMCOREINFO") != 0)
451 vmcoreinfo
= kzalloc_panic(note
.n_descsz
);
452 if (copy_from_oldmem(vmcoreinfo
, addr
+ 24, note
.n_descsz
))
454 *size
= note
.n_descsz
;
459 * Initialize vmcoreinfo note (new kernel)
461 static void *nt_vmcoreinfo(void *ptr
)
466 vmcoreinfo
= os_info_old_entry(OS_INFO_VMCOREINFO
, &size
);
468 vmcoreinfo
= get_vmcoreinfo_old(&size
);
471 return nt_init(ptr
, 0, vmcoreinfo
, size
, "VMCOREINFO");
475 * Initialize ELF header (new kernel)
477 static void *ehdr_init(Elf64_Ehdr
*ehdr
, int mem_chunk_cnt
)
479 memset(ehdr
, 0, sizeof(*ehdr
));
480 memcpy(ehdr
->e_ident
, ELFMAG
, SELFMAG
);
481 ehdr
->e_ident
[EI_CLASS
] = ELFCLASS64
;
482 ehdr
->e_ident
[EI_DATA
] = ELFDATA2MSB
;
483 ehdr
->e_ident
[EI_VERSION
] = EV_CURRENT
;
484 memset(ehdr
->e_ident
+ EI_PAD
, 0, EI_NIDENT
- EI_PAD
);
485 ehdr
->e_type
= ET_CORE
;
486 ehdr
->e_machine
= EM_S390
;
487 ehdr
->e_version
= EV_CURRENT
;
488 ehdr
->e_phoff
= sizeof(Elf64_Ehdr
);
489 ehdr
->e_ehsize
= sizeof(Elf64_Ehdr
);
490 ehdr
->e_phentsize
= sizeof(Elf64_Phdr
);
491 ehdr
->e_phnum
= mem_chunk_cnt
+ 1;
496 * Return CPU count for ELF header (new kernel)
498 static int get_cpu_cnt(void)
502 for (i
= 0; i
< dump_save_areas
.count
; i
++) {
503 if (dump_save_areas
.areas
[i
]->sa
.pref_reg
== 0)
511 * Return memory chunk count for ELF header (new kernel)
513 static int get_mem_chunk_cnt(void)
518 for_each_dump_mem_range(idx
, NUMA_NO_NODE
, NULL
, NULL
, NULL
)
524 * Initialize ELF loads (new kernel)
526 static void loads_init(Elf64_Phdr
*phdr
, u64 loads_offset
)
528 phys_addr_t start
, end
;
531 for_each_dump_mem_range(idx
, NUMA_NO_NODE
, &start
, &end
, NULL
) {
532 phdr
->p_filesz
= end
- start
;
533 phdr
->p_type
= PT_LOAD
;
534 phdr
->p_offset
= start
;
535 phdr
->p_vaddr
= start
;
536 phdr
->p_paddr
= start
;
537 phdr
->p_memsz
= end
- start
;
538 phdr
->p_flags
= PF_R
| PF_W
| PF_X
;
539 phdr
->p_align
= PAGE_SIZE
;
545 * Initialize notes (new kernel)
547 static void *notes_init(Elf64_Phdr
*phdr
, void *ptr
, u64 notes_offset
)
549 struct save_area_ext
*sa_ext
;
550 void *ptr_start
= ptr
;
553 ptr
= nt_prpsinfo(ptr
);
555 for (i
= 0; i
< dump_save_areas
.count
; i
++) {
556 sa_ext
= dump_save_areas
.areas
[i
];
557 if (sa_ext
->sa
.pref_reg
== 0)
559 ptr
= fill_cpu_elf_notes(ptr
, &sa_ext
->sa
, sa_ext
->vx_regs
);
561 ptr
= nt_vmcoreinfo(ptr
);
562 memset(phdr
, 0, sizeof(*phdr
));
563 phdr
->p_type
= PT_NOTE
;
564 phdr
->p_offset
= notes_offset
;
565 phdr
->p_filesz
= (unsigned long) PTR_SUB(ptr
, ptr_start
);
566 phdr
->p_memsz
= phdr
->p_filesz
;
571 * Create ELF core header (new kernel)
573 int elfcorehdr_alloc(unsigned long long *addr
, unsigned long long *size
)
575 Elf64_Phdr
*phdr_notes
, *phdr_loads
;
581 /* If we are not in kdump or zfcpdump mode return */
582 if (!OLDMEM_BASE
&& ipl_info
.type
!= IPL_TYPE_FCP_DUMP
)
584 /* If elfcorehdr= has been passed via cmdline, we use that one */
585 if (elfcorehdr_addr
!= ELFCORE_ADDR_MAX
)
587 /* If we cannot get HSA size for zfcpdump return error */
588 if (ipl_info
.type
== IPL_TYPE_FCP_DUMP
&& !sclp
.hsa_size
)
591 /* For kdump, exclude previous crashkernel memory */
593 oldmem_region
.base
= OLDMEM_BASE
;
594 oldmem_region
.size
= OLDMEM_SIZE
;
595 oldmem_type
.total_size
= OLDMEM_SIZE
;
598 mem_chunk_cnt
= get_mem_chunk_cnt();
600 alloc_size
= 0x1000 + get_cpu_cnt() * 0x4a0 +
601 mem_chunk_cnt
* sizeof(Elf64_Phdr
);
602 hdr
= kzalloc_panic(alloc_size
);
603 /* Init elf header */
604 ptr
= ehdr_init(hdr
, mem_chunk_cnt
);
605 /* Init program headers */
607 ptr
= PTR_ADD(ptr
, sizeof(Elf64_Phdr
));
609 ptr
= PTR_ADD(ptr
, sizeof(Elf64_Phdr
) * mem_chunk_cnt
);
611 hdr_off
= PTR_DIFF(ptr
, hdr
);
612 ptr
= notes_init(phdr_notes
, ptr
, ((unsigned long) hdr
) + hdr_off
);
614 hdr_off
= PTR_DIFF(ptr
, hdr
);
615 loads_init(phdr_loads
, hdr_off
);
616 *addr
= (unsigned long long) hdr
;
617 elfcorehdr_newmem
= hdr
;
618 *size
= (unsigned long long) hdr_off
;
619 BUG_ON(elfcorehdr_size
> alloc_size
);
624 * Free ELF core header (new kernel)
626 void elfcorehdr_free(unsigned long long addr
)
628 if (!elfcorehdr_newmem
)
630 kfree((void *)(unsigned long)addr
);
634 * Read from ELF header
636 ssize_t
elfcorehdr_read(char *buf
, size_t count
, u64
*ppos
)
638 void *src
= (void *)(unsigned long)*ppos
;
640 src
= elfcorehdr_newmem
? src
: src
- OLDMEM_BASE
;
641 memcpy(buf
, src
, count
);
647 * Read from ELF notes data
649 ssize_t
elfcorehdr_read_notes(char *buf
, size_t count
, u64
*ppos
)
651 void *src
= (void *)(unsigned long)*ppos
;
654 if (elfcorehdr_newmem
) {
655 memcpy(buf
, src
, count
);
657 rc
= copy_from_oldmem(buf
, src
, count
);