xref: /openbmc/linux/arch/s390/kernel/crash_dump.c (revision 161f4089)
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 
25 struct dump_save_areas dump_save_areas;
26 
27 /*
28  * Allocate and add a save area for a CPU
29  */
30 struct 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  */
55 static 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  */
71 static 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  */
93 static void *elfcorehdr_newmem;
94 
95 /*
96  * Copy one page from zfcpdump "oldmem"
97  *
98  * For pages below ZFCPDUMP_HSA_SIZE memory from the HSA is copied. Otherwise
99  * real memory copy is used.
100  */
101 static 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 < ZFCPDUMP_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  */
125 static 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  */
147 ssize_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  */
167 static 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 ZFCPDUMP_HSA_SIZE. Memory below
192  * ZFCPDUMP_HSA_SIZE is read on demand using the copy_oldmem_page() function.
193  */
194 static 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 size_hsa;
200 
201 	if (pfn < ZFCPDUMP_HSA_SIZE >> PAGE_SHIFT) {
202 		size_hsa = min(size, ZFCPDUMP_HSA_SIZE - (pfn << PAGE_SHIFT));
203 		if (size == size_hsa)
204 			return 0;
205 		size -= size_hsa;
206 		from += size_hsa;
207 		pfn += size_hsa >> PAGE_SHIFT;
208 	}
209 	return remap_pfn_range(vma, from, pfn, size, prot);
210 }
211 
212 /*
213  * Remap "oldmem" for kdump or zfcpdump
214  */
215 int remap_oldmem_pfn_range(struct vm_area_struct *vma, unsigned long from,
216 			   unsigned long pfn, unsigned long size, pgprot_t prot)
217 {
218 	if (OLDMEM_BASE)
219 		return remap_oldmem_pfn_range_kdump(vma, from, pfn, size, prot);
220 	else
221 		return remap_oldmem_pfn_range_zfcpdump(vma, from, pfn, size,
222 						       prot);
223 }
224 
225 /*
226  * Copy memory from old kernel
227  */
228 int copy_from_oldmem(void *dest, void *src, size_t count)
229 {
230 	unsigned long copied = 0;
231 	int rc;
232 
233 	if (OLDMEM_BASE) {
234 		if ((unsigned long) src < OLDMEM_SIZE) {
235 			copied = min(count, OLDMEM_SIZE - (unsigned long) src);
236 			rc = copy_from_realmem(dest, src + OLDMEM_BASE, copied);
237 			if (rc)
238 				return rc;
239 		}
240 	} else {
241 		if ((unsigned long) src < ZFCPDUMP_HSA_SIZE) {
242 			copied = min(count,
243 				     ZFCPDUMP_HSA_SIZE - (unsigned long) src);
244 			rc = memcpy_hsa(dest, (unsigned long) src, copied, 0);
245 			if (rc)
246 				return rc;
247 		}
248 	}
249 	return copy_from_realmem(dest + copied, src + copied, count - copied);
250 }
251 
252 /*
253  * Alloc memory and panic in case of ENOMEM
254  */
255 static void *kzalloc_panic(int len)
256 {
257 	void *rc;
258 
259 	rc = kzalloc(len, GFP_KERNEL);
260 	if (!rc)
261 		panic("s390 kdump kzalloc (%d) failed", len);
262 	return rc;
263 }
264 
265 /*
266  * Get memory layout and create hole for oldmem
267  */
268 static struct mem_chunk *get_memory_layout(void)
269 {
270 	struct mem_chunk *chunk_array;
271 
272 	chunk_array = kzalloc_panic(MEMORY_CHUNKS * sizeof(struct mem_chunk));
273 	detect_memory_layout(chunk_array, 0);
274 	create_mem_hole(chunk_array, OLDMEM_BASE, OLDMEM_SIZE);
275 	return chunk_array;
276 }
277 
278 /*
279  * Initialize ELF note
280  */
281 static void *nt_init(void *buf, Elf64_Word type, void *desc, int d_len,
282 		     const char *name)
283 {
284 	Elf64_Nhdr *note;
285 	u64 len;
286 
287 	note = (Elf64_Nhdr *)buf;
288 	note->n_namesz = strlen(name) + 1;
289 	note->n_descsz = d_len;
290 	note->n_type = type;
291 	len = sizeof(Elf64_Nhdr);
292 
293 	memcpy(buf + len, name, note->n_namesz);
294 	len = roundup(len + note->n_namesz, 4);
295 
296 	memcpy(buf + len, desc, note->n_descsz);
297 	len = roundup(len + note->n_descsz, 4);
298 
299 	return PTR_ADD(buf, len);
300 }
301 
302 /*
303  * Initialize prstatus note
304  */
305 static void *nt_prstatus(void *ptr, struct save_area *sa)
306 {
307 	struct elf_prstatus nt_prstatus;
308 	static int cpu_nr = 1;
309 
310 	memset(&nt_prstatus, 0, sizeof(nt_prstatus));
311 	memcpy(&nt_prstatus.pr_reg.gprs, sa->gp_regs, sizeof(sa->gp_regs));
312 	memcpy(&nt_prstatus.pr_reg.psw, sa->psw, sizeof(sa->psw));
313 	memcpy(&nt_prstatus.pr_reg.acrs, sa->acc_regs, sizeof(sa->acc_regs));
314 	nt_prstatus.pr_pid = cpu_nr;
315 	cpu_nr++;
316 
317 	return nt_init(ptr, NT_PRSTATUS, &nt_prstatus, sizeof(nt_prstatus),
318 			 "CORE");
319 }
320 
321 /*
322  * Initialize fpregset (floating point) note
323  */
324 static void *nt_fpregset(void *ptr, struct save_area *sa)
325 {
326 	elf_fpregset_t nt_fpregset;
327 
328 	memset(&nt_fpregset, 0, sizeof(nt_fpregset));
329 	memcpy(&nt_fpregset.fpc, &sa->fp_ctrl_reg, sizeof(sa->fp_ctrl_reg));
330 	memcpy(&nt_fpregset.fprs, &sa->fp_regs, sizeof(sa->fp_regs));
331 
332 	return nt_init(ptr, NT_PRFPREG, &nt_fpregset, sizeof(nt_fpregset),
333 		       "CORE");
334 }
335 
336 /*
337  * Initialize timer note
338  */
339 static void *nt_s390_timer(void *ptr, struct save_area *sa)
340 {
341 	return nt_init(ptr, NT_S390_TIMER, &sa->timer, sizeof(sa->timer),
342 			 KEXEC_CORE_NOTE_NAME);
343 }
344 
345 /*
346  * Initialize TOD clock comparator note
347  */
348 static void *nt_s390_tod_cmp(void *ptr, struct save_area *sa)
349 {
350 	return nt_init(ptr, NT_S390_TODCMP, &sa->clk_cmp,
351 		       sizeof(sa->clk_cmp), KEXEC_CORE_NOTE_NAME);
352 }
353 
354 /*
355  * Initialize TOD programmable register note
356  */
357 static void *nt_s390_tod_preg(void *ptr, struct save_area *sa)
358 {
359 	return nt_init(ptr, NT_S390_TODPREG, &sa->tod_reg,
360 		       sizeof(sa->tod_reg), KEXEC_CORE_NOTE_NAME);
361 }
362 
363 /*
364  * Initialize control register note
365  */
366 static void *nt_s390_ctrs(void *ptr, struct save_area *sa)
367 {
368 	return nt_init(ptr, NT_S390_CTRS, &sa->ctrl_regs,
369 		       sizeof(sa->ctrl_regs), KEXEC_CORE_NOTE_NAME);
370 }
371 
372 /*
373  * Initialize prefix register note
374  */
375 static void *nt_s390_prefix(void *ptr, struct save_area *sa)
376 {
377 	return nt_init(ptr, NT_S390_PREFIX, &sa->pref_reg,
378 			 sizeof(sa->pref_reg), KEXEC_CORE_NOTE_NAME);
379 }
380 
381 /*
382  * Fill ELF notes for one CPU with save area registers
383  */
384 void *fill_cpu_elf_notes(void *ptr, struct save_area *sa)
385 {
386 	ptr = nt_prstatus(ptr, sa);
387 	ptr = nt_fpregset(ptr, sa);
388 	ptr = nt_s390_timer(ptr, sa);
389 	ptr = nt_s390_tod_cmp(ptr, sa);
390 	ptr = nt_s390_tod_preg(ptr, sa);
391 	ptr = nt_s390_ctrs(ptr, sa);
392 	ptr = nt_s390_prefix(ptr, sa);
393 	return ptr;
394 }
395 
396 /*
397  * Initialize prpsinfo note (new kernel)
398  */
399 static void *nt_prpsinfo(void *ptr)
400 {
401 	struct elf_prpsinfo prpsinfo;
402 
403 	memset(&prpsinfo, 0, sizeof(prpsinfo));
404 	prpsinfo.pr_sname = 'R';
405 	strcpy(prpsinfo.pr_fname, "vmlinux");
406 	return nt_init(ptr, NT_PRPSINFO, &prpsinfo, sizeof(prpsinfo),
407 		       KEXEC_CORE_NOTE_NAME);
408 }
409 
410 /*
411  * Get vmcoreinfo using lowcore->vmcore_info (new kernel)
412  */
413 static void *get_vmcoreinfo_old(unsigned long *size)
414 {
415 	char nt_name[11], *vmcoreinfo;
416 	Elf64_Nhdr note;
417 	void *addr;
418 
419 	if (copy_from_oldmem(&addr, &S390_lowcore.vmcore_info, sizeof(addr)))
420 		return NULL;
421 	memset(nt_name, 0, sizeof(nt_name));
422 	if (copy_from_oldmem(&note, addr, sizeof(note)))
423 		return NULL;
424 	if (copy_from_oldmem(nt_name, addr + sizeof(note), sizeof(nt_name) - 1))
425 		return NULL;
426 	if (strcmp(nt_name, "VMCOREINFO") != 0)
427 		return NULL;
428 	vmcoreinfo = kzalloc_panic(note.n_descsz);
429 	if (copy_from_oldmem(vmcoreinfo, addr + 24, note.n_descsz))
430 		return NULL;
431 	*size = note.n_descsz;
432 	return vmcoreinfo;
433 }
434 
435 /*
436  * Initialize vmcoreinfo note (new kernel)
437  */
438 static void *nt_vmcoreinfo(void *ptr)
439 {
440 	unsigned long size;
441 	void *vmcoreinfo;
442 
443 	vmcoreinfo = os_info_old_entry(OS_INFO_VMCOREINFO, &size);
444 	if (!vmcoreinfo)
445 		vmcoreinfo = get_vmcoreinfo_old(&size);
446 	if (!vmcoreinfo)
447 		return ptr;
448 	return nt_init(ptr, 0, vmcoreinfo, size, "VMCOREINFO");
449 }
450 
451 /*
452  * Initialize ELF header (new kernel)
453  */
454 static 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  */
475 static int get_cpu_cnt(void)
476 {
477 	int i, cpus = 0;
478 
479 	for (i = 0; i < dump_save_areas.count; i++) {
480 		if (dump_save_areas.areas[i]->pref_reg == 0)
481 			continue;
482 		cpus++;
483 	}
484 	return cpus;
485 }
486 
487 /*
488  * Return memory chunk count for ELF header (new kernel)
489  */
490 static int get_mem_chunk_cnt(void)
491 {
492 	struct mem_chunk *chunk_array, *mem_chunk;
493 	int i, cnt = 0;
494 
495 	chunk_array = get_memory_layout();
496 	for (i = 0; i < MEMORY_CHUNKS; i++) {
497 		mem_chunk = &chunk_array[i];
498 		if (chunk_array[i].type != CHUNK_READ_WRITE &&
499 		    chunk_array[i].type != CHUNK_READ_ONLY)
500 			continue;
501 		if (mem_chunk->size == 0)
502 			continue;
503 		cnt++;
504 	}
505 	kfree(chunk_array);
506 	return cnt;
507 }
508 
509 /*
510  * Initialize ELF loads (new kernel)
511  */
512 static int loads_init(Elf64_Phdr *phdr, u64 loads_offset)
513 {
514 	struct mem_chunk *chunk_array, *mem_chunk;
515 	int i;
516 
517 	chunk_array = get_memory_layout();
518 	for (i = 0; i < MEMORY_CHUNKS; i++) {
519 		mem_chunk = &chunk_array[i];
520 		if (mem_chunk->size == 0)
521 			continue;
522 		if (chunk_array[i].type != CHUNK_READ_WRITE &&
523 		    chunk_array[i].type != CHUNK_READ_ONLY)
524 			continue;
525 		else
526 			phdr->p_filesz = mem_chunk->size;
527 		phdr->p_type = PT_LOAD;
528 		phdr->p_offset = mem_chunk->addr;
529 		phdr->p_vaddr = mem_chunk->addr;
530 		phdr->p_paddr = mem_chunk->addr;
531 		phdr->p_memsz = mem_chunk->size;
532 		phdr->p_flags = PF_R | PF_W | PF_X;
533 		phdr->p_align = PAGE_SIZE;
534 		phdr++;
535 	}
536 	kfree(chunk_array);
537 	return i;
538 }
539 
540 /*
541  * Initialize notes (new kernel)
542  */
543 static void *notes_init(Elf64_Phdr *phdr, void *ptr, u64 notes_offset)
544 {
545 	struct save_area *sa;
546 	void *ptr_start = ptr;
547 	int i;
548 
549 	ptr = nt_prpsinfo(ptr);
550 
551 	for (i = 0; i < dump_save_areas.count; i++) {
552 		sa = dump_save_areas.areas[i];
553 		if (sa->pref_reg == 0)
554 			continue;
555 		ptr = fill_cpu_elf_notes(ptr, sa);
556 	}
557 	ptr = nt_vmcoreinfo(ptr);
558 	memset(phdr, 0, sizeof(*phdr));
559 	phdr->p_type = PT_NOTE;
560 	phdr->p_offset = notes_offset;
561 	phdr->p_filesz = (unsigned long) PTR_SUB(ptr, ptr_start);
562 	phdr->p_memsz = phdr->p_filesz;
563 	return ptr;
564 }
565 
566 /*
567  * Create ELF core header (new kernel)
568  */
569 int elfcorehdr_alloc(unsigned long long *addr, unsigned long long *size)
570 {
571 	Elf64_Phdr *phdr_notes, *phdr_loads;
572 	int mem_chunk_cnt;
573 	void *ptr, *hdr;
574 	u32 alloc_size;
575 	u64 hdr_off;
576 
577 	/* If we are not in kdump or zfcpdump mode return */
578 	if (!OLDMEM_BASE && ipl_info.type != IPL_TYPE_FCP_DUMP)
579 		return 0;
580 	/* If elfcorehdr= has been passed via cmdline, we use that one */
581 	if (elfcorehdr_addr != ELFCORE_ADDR_MAX)
582 		return 0;
583 	mem_chunk_cnt = get_mem_chunk_cnt();
584 
585 	alloc_size = 0x1000 + get_cpu_cnt() * 0x300 +
586 		mem_chunk_cnt * sizeof(Elf64_Phdr);
587 	hdr = kzalloc_panic(alloc_size);
588 	/* Init elf header */
589 	ptr = ehdr_init(hdr, mem_chunk_cnt);
590 	/* Init program headers */
591 	phdr_notes = ptr;
592 	ptr = PTR_ADD(ptr, sizeof(Elf64_Phdr));
593 	phdr_loads = ptr;
594 	ptr = PTR_ADD(ptr, sizeof(Elf64_Phdr) * mem_chunk_cnt);
595 	/* Init notes */
596 	hdr_off = PTR_DIFF(ptr, hdr);
597 	ptr = notes_init(phdr_notes, ptr, ((unsigned long) hdr) + hdr_off);
598 	/* Init loads */
599 	hdr_off = PTR_DIFF(ptr, hdr);
600 	loads_init(phdr_loads, hdr_off);
601 	*addr = (unsigned long long) hdr;
602 	elfcorehdr_newmem = hdr;
603 	*size = (unsigned long long) hdr_off;
604 	BUG_ON(elfcorehdr_size > alloc_size);
605 	return 0;
606 }
607 
608 /*
609  * Free ELF core header (new kernel)
610  */
611 void elfcorehdr_free(unsigned long long addr)
612 {
613 	if (!elfcorehdr_newmem)
614 		return;
615 	kfree((void *)(unsigned long)addr);
616 }
617 
618 /*
619  * Read from ELF header
620  */
621 ssize_t elfcorehdr_read(char *buf, size_t count, u64 *ppos)
622 {
623 	void *src = (void *)(unsigned long)*ppos;
624 
625 	src = elfcorehdr_newmem ? src : src - OLDMEM_BASE;
626 	memcpy(buf, src, count);
627 	*ppos += count;
628 	return count;
629 }
630 
631 /*
632  * Read from ELF notes data
633  */
634 ssize_t elfcorehdr_read_notes(char *buf, size_t count, u64 *ppos)
635 {
636 	void *src = (void *)(unsigned long)*ppos;
637 	int rc;
638 
639 	if (elfcorehdr_newmem) {
640 		memcpy(buf, src, count);
641 	} else {
642 		rc = copy_from_oldmem(buf, src, count);
643 		if (rc)
644 			return rc;
645 	}
646 	*ppos += count;
647 	return count;
648 }
649