xref: /openbmc/linux/arch/s390/kernel/crash_dump.c (revision a06c488d)
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/asm-offsets.h>
17 #include <linux/memblock.h>
18 #include <asm/os_info.h>
19 #include <asm/elf.h>
20 #include <asm/ipl.h>
21 #include <asm/sclp.h>
22 
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))))
26 
27 static struct memblock_region oldmem_region;
28 
29 static struct memblock_type oldmem_type = {
30 	.cnt = 1,
31 	.max = 1,
32 	.total_size = 0,
33 	.regions = &oldmem_region,
34 };
35 
36 struct save_area {
37 	struct list_head list;
38 	u64 psw[2];
39 	u64 ctrs[16];
40 	u64 gprs[16];
41 	u32 acrs[16];
42 	u64 fprs[16];
43 	u32 fpc;
44 	u32 prefix;
45 	u64 todpreg;
46 	u64 timer;
47 	u64 todcmp;
48 	u64 vxrs_low[16];
49 	__vector128 vxrs_high[16];
50 };
51 
52 static LIST_HEAD(dump_save_areas);
53 
54 /*
55  * Allocate a save area
56  */
57 struct save_area * __init save_area_alloc(bool is_boot_cpu)
58 {
59 	struct save_area *sa;
60 
61 	sa = (void *) memblock_alloc(sizeof(*sa), 8);
62 	if (!sa)
63 		return NULL;
64 	if (is_boot_cpu)
65 		list_add(&sa->list, &dump_save_areas);
66 	else
67 		list_add_tail(&sa->list, &dump_save_areas);
68 	return sa;
69 }
70 
71 /*
72  * Return the address of the save area for the boot CPU
73  */
74 struct save_area * __init save_area_boot_cpu(void)
75 {
76 	if (list_empty(&dump_save_areas))
77 		return NULL;
78 	return list_first_entry(&dump_save_areas, struct save_area, list);
79 }
80 
81 /*
82  * Copy CPU registers into the save area
83  */
84 void __init save_area_add_regs(struct save_area *sa, void *regs)
85 {
86 	struct lowcore *lc;
87 
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));
99 }
100 
101 /*
102  * Copy vector registers into the save area
103  */
104 void __init save_area_add_vxrs(struct save_area *sa, __vector128 *vxrs)
105 {
106 	int i;
107 
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));
113 }
114 
115 /*
116  * Return physical address for virtual address
117  */
118 static inline void *load_real_addr(void *addr)
119 {
120 	unsigned long real_addr;
121 
122 	asm volatile(
123 		   "	lra     %0,0(%1)\n"
124 		   "	jz	0f\n"
125 		   "	la	%0,0\n"
126 		   "0:"
127 		   : "=a" (real_addr) : "a" (addr) : "cc");
128 	return (void *)real_addr;
129 }
130 
131 /*
132  * Copy memory of the old, dumped system to a kernel space virtual address
133  */
134 int copy_oldmem_kernel(void *dst, void *src, size_t count)
135 {
136 	unsigned long from, len;
137 	void *ra;
138 	int rc;
139 
140 	while (count) {
141 		from = __pa(src);
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);
146 			if (rc)
147 				return rc;
148 		} else {
149 			/* Check for swapped kdump oldmem areas */
150 			if (OLDMEM_BASE && from - OLDMEM_BASE < OLDMEM_SIZE) {
151 				from -= OLDMEM_BASE;
152 				len = min(count, OLDMEM_SIZE - from);
153 			} else if (OLDMEM_BASE && from < OLDMEM_SIZE) {
154 				len = min(count, OLDMEM_SIZE - from);
155 				from += OLDMEM_BASE;
156 			} else {
157 				len = count;
158 			}
159 			if (is_vmalloc_or_module_addr(dst)) {
160 				ra = load_real_addr(dst);
161 				len = min(PAGE_SIZE - offset_in_page(ra), len);
162 			} else {
163 				ra = dst;
164 			}
165 			if (memcpy_real(ra, (void *) from, len))
166 				return -EFAULT;
167 		}
168 		dst += len;
169 		src += len;
170 		count -= len;
171 	}
172 	return 0;
173 }
174 
175 /*
176  * Copy memory of the old, dumped system to a user space virtual address
177  */
178 int copy_oldmem_user(void __user *dst, void *src, size_t count)
179 {
180 	unsigned long from, len;
181 	int rc;
182 
183 	while (count) {
184 		from = __pa(src);
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);
189 			if (rc)
190 				return rc;
191 		} else {
192 			/* Check for swapped kdump oldmem areas */
193 			if (OLDMEM_BASE && from - OLDMEM_BASE < OLDMEM_SIZE) {
194 				from -= OLDMEM_BASE;
195 				len = min(count, OLDMEM_SIZE - from);
196 			} else if (OLDMEM_BASE && from < OLDMEM_SIZE) {
197 				len = min(count, OLDMEM_SIZE - from);
198 				from += OLDMEM_BASE;
199 			} else {
200 				len = count;
201 			}
202 			rc = copy_to_user_real(dst, (void *) from, count);
203 			if (rc)
204 				return rc;
205 		}
206 		dst += len;
207 		src += len;
208 		count -= len;
209 	}
210 	return 0;
211 }
212 
213 /*
214  * Copy one page from "oldmem"
215  */
216 ssize_t copy_oldmem_page(unsigned long pfn, char *buf, size_t csize,
217 			 unsigned long offset, int userbuf)
218 {
219 	void *src;
220 	int rc;
221 
222 	if (!csize)
223 		return 0;
224 	src = (void *) (pfn << PAGE_SHIFT) + offset;
225 	if (userbuf)
226 		rc = copy_oldmem_user((void __force __user *) buf, src, csize);
227 	else
228 		rc = copy_oldmem_kernel((void *) buf, src, csize);
229 	return rc;
230 }
231 
232 /*
233  * Remap "oldmem" for kdump
234  *
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]
237  */
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)
241 {
242 	unsigned long size_old;
243 	int rc;
244 
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),
249 				     size_old, prot);
250 		if (rc || size == size_old)
251 			return rc;
252 		size -= size_old;
253 		from += size_old;
254 		pfn += size_old >> PAGE_SHIFT;
255 	}
256 	return remap_pfn_range(vma, from, pfn, size, prot);
257 }
258 
259 /*
260  * Remap "oldmem" for zfcpdump
261  *
262  * We only map available memory above HSA size. Memory below HSA size
263  * is read on demand using the copy_oldmem_page() function.
264  */
265 static int remap_oldmem_pfn_range_zfcpdump(struct vm_area_struct *vma,
266 					   unsigned long from,
267 					   unsigned long pfn,
268 					   unsigned long size, pgprot_t prot)
269 {
270 	unsigned long hsa_end = sclp.hsa_size;
271 	unsigned long size_hsa;
272 
273 	if (pfn < hsa_end >> PAGE_SHIFT) {
274 		size_hsa = min(size, hsa_end - (pfn << PAGE_SHIFT));
275 		if (size == size_hsa)
276 			return 0;
277 		size -= size_hsa;
278 		from += size_hsa;
279 		pfn += size_hsa >> PAGE_SHIFT;
280 	}
281 	return remap_pfn_range(vma, from, pfn, size, prot);
282 }
283 
284 /*
285  * Remap "oldmem" for kdump or zfcpdump
286  */
287 int remap_oldmem_pfn_range(struct vm_area_struct *vma, unsigned long from,
288 			   unsigned long pfn, unsigned long size, pgprot_t prot)
289 {
290 	if (OLDMEM_BASE)
291 		return remap_oldmem_pfn_range_kdump(vma, from, pfn, size, prot);
292 	else
293 		return remap_oldmem_pfn_range_zfcpdump(vma, from, pfn, size,
294 						       prot);
295 }
296 
297 /*
298  * Alloc memory and panic in case of ENOMEM
299  */
300 static void *kzalloc_panic(int len)
301 {
302 	void *rc;
303 
304 	rc = kzalloc(len, GFP_KERNEL);
305 	if (!rc)
306 		panic("s390 kdump kzalloc (%d) failed", len);
307 	return rc;
308 }
309 
310 /*
311  * Initialize ELF note
312  */
313 static void *nt_init_name(void *buf, Elf64_Word type, void *desc, int d_len,
314 			  const char *name)
315 {
316 	Elf64_Nhdr *note;
317 	u64 len;
318 
319 	note = (Elf64_Nhdr *)buf;
320 	note->n_namesz = strlen(name) + 1;
321 	note->n_descsz = d_len;
322 	note->n_type = type;
323 	len = sizeof(Elf64_Nhdr);
324 
325 	memcpy(buf + len, name, note->n_namesz);
326 	len = roundup(len + note->n_namesz, 4);
327 
328 	memcpy(buf + len, desc, note->n_descsz);
329 	len = roundup(len + note->n_descsz, 4);
330 
331 	return PTR_ADD(buf, len);
332 }
333 
334 static inline void *nt_init(void *buf, Elf64_Word type, void *desc, int d_len)
335 {
336 	return nt_init_name(buf, type, desc, d_len, KEXEC_CORE_NOTE_NAME);
337 }
338 
339 /*
340  * Fill ELF notes for one CPU with save area registers
341  */
342 static void *fill_cpu_elf_notes(void *ptr, int cpu, struct save_area *sa)
343 {
344 	struct elf_prstatus nt_prstatus;
345 	elf_fpregset_t nt_fpregset;
346 
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));
370 	}
371 	return ptr;
372 }
373 
374 /*
375  * Initialize prpsinfo note (new kernel)
376  */
377 static void *nt_prpsinfo(void *ptr)
378 {
379 	struct elf_prpsinfo prpsinfo;
380 
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));
385 }
386 
387 /*
388  * Get vmcoreinfo using lowcore->vmcore_info (new kernel)
389  */
390 static void *get_vmcoreinfo_old(unsigned long *size)
391 {
392 	char nt_name[11], *vmcoreinfo;
393 	Elf64_Nhdr note;
394 	void *addr;
395 
396 	if (copy_oldmem_kernel(&addr, &S390_lowcore.vmcore_info, sizeof(addr)))
397 		return NULL;
398 	memset(nt_name, 0, sizeof(nt_name));
399 	if (copy_oldmem_kernel(&note, addr, sizeof(note)))
400 		return NULL;
401 	if (copy_oldmem_kernel(nt_name, addr + sizeof(note),
402 			       sizeof(nt_name) - 1))
403 		return NULL;
404 	if (strcmp(nt_name, "VMCOREINFO") != 0)
405 		return NULL;
406 	vmcoreinfo = kzalloc_panic(note.n_descsz);
407 	if (copy_oldmem_kernel(vmcoreinfo, addr + 24, note.n_descsz))
408 		return NULL;
409 	*size = note.n_descsz;
410 	return vmcoreinfo;
411 }
412 
413 /*
414  * Initialize vmcoreinfo note (new kernel)
415  */
416 static void *nt_vmcoreinfo(void *ptr)
417 {
418 	unsigned long size;
419 	void *vmcoreinfo;
420 
421 	vmcoreinfo = os_info_old_entry(OS_INFO_VMCOREINFO, &size);
422 	if (!vmcoreinfo)
423 		vmcoreinfo = get_vmcoreinfo_old(&size);
424 	if (!vmcoreinfo)
425 		return ptr;
426 	return nt_init_name(ptr, 0, vmcoreinfo, size, "VMCOREINFO");
427 }
428 
429 /*
430  * Initialize ELF header (new kernel)
431  */
432 static void *ehdr_init(Elf64_Ehdr *ehdr, int mem_chunk_cnt)
433 {
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;
447 	return ehdr + 1;
448 }
449 
450 /*
451  * Return CPU count for ELF header (new kernel)
452  */
453 static int get_cpu_cnt(void)
454 {
455 	struct save_area *sa;
456 	int cpus = 0;
457 
458 	list_for_each_entry(sa, &dump_save_areas, list)
459 		if (sa->prefix != 0)
460 			cpus++;
461 	return cpus;
462 }
463 
464 /*
465  * Return memory chunk count for ELF header (new kernel)
466  */
467 static int get_mem_chunk_cnt(void)
468 {
469 	int cnt = 0;
470 	u64 idx;
471 
472 	for_each_mem_range(idx, &memblock.physmem, &oldmem_type, NUMA_NO_NODE,
473 			   MEMBLOCK_NONE, NULL, NULL, NULL)
474 		cnt++;
475 	return cnt;
476 }
477 
478 /*
479  * Initialize ELF loads (new kernel)
480  */
481 static void loads_init(Elf64_Phdr *phdr, u64 loads_offset)
482 {
483 	phys_addr_t start, end;
484 	u64 idx;
485 
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;
496 		phdr++;
497 	}
498 }
499 
500 /*
501  * Initialize notes (new kernel)
502  */
503 static void *notes_init(Elf64_Phdr *phdr, void *ptr, u64 notes_offset)
504 {
505 	struct save_area *sa;
506 	void *ptr_start = ptr;
507 	int cpu;
508 
509 	ptr = nt_prpsinfo(ptr);
510 
511 	cpu = 1;
512 	list_for_each_entry(sa, &dump_save_areas, list)
513 		if (sa->prefix != 0)
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;
521 	return ptr;
522 }
523 
524 /*
525  * Create ELF core header (new kernel)
526  */
527 int elfcorehdr_alloc(unsigned long long *addr, unsigned long long *size)
528 {
529 	Elf64_Phdr *phdr_notes, *phdr_loads;
530 	int mem_chunk_cnt;
531 	void *ptr, *hdr;
532 	u32 alloc_size;
533 	u64 hdr_off;
534 
535 	/* If we are not in kdump or zfcpdump mode return */
536 	if (!OLDMEM_BASE && ipl_info.type != IPL_TYPE_FCP_DUMP)
537 		return 0;
538 	/* If we cannot get HSA size for zfcpdump return error */
539 	if (ipl_info.type == IPL_TYPE_FCP_DUMP && !sclp.hsa_size)
540 		return -ENODEV;
541 
542 	/* For kdump, exclude previous crashkernel memory */
543 	if (OLDMEM_BASE) {
544 		oldmem_region.base = OLDMEM_BASE;
545 		oldmem_region.size = OLDMEM_SIZE;
546 		oldmem_type.total_size = OLDMEM_SIZE;
547 	}
548 
549 	mem_chunk_cnt = get_mem_chunk_cnt();
550 
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 */
557 	phdr_notes = ptr;
558 	ptr = PTR_ADD(ptr, sizeof(Elf64_Phdr));
559 	phdr_loads = ptr;
560 	ptr = PTR_ADD(ptr, sizeof(Elf64_Phdr) * mem_chunk_cnt);
561 	/* Init notes */
562 	hdr_off = PTR_DIFF(ptr, hdr);
563 	ptr = notes_init(phdr_notes, ptr, ((unsigned long) hdr) + hdr_off);
564 	/* Init loads */
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);
570 	return 0;
571 }
572 
573 /*
574  * Free ELF core header (new kernel)
575  */
576 void elfcorehdr_free(unsigned long long addr)
577 {
578 	kfree((void *)(unsigned long)addr);
579 }
580 
581 /*
582  * Read from ELF header
583  */
584 ssize_t elfcorehdr_read(char *buf, size_t count, u64 *ppos)
585 {
586 	void *src = (void *)(unsigned long)*ppos;
587 
588 	memcpy(buf, src, count);
589 	*ppos += count;
590 	return count;
591 }
592 
593 /*
594  * Read from ELF notes data
595  */
596 ssize_t elfcorehdr_read_notes(char *buf, size_t count, u64 *ppos)
597 {
598 	void *src = (void *)(unsigned long)*ppos;
599 
600 	memcpy(buf, src, count);
601 	*ppos += count;
602 	return count;
603 }
604