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
 25struct dump_save_areas dump_save_areas;
 26
 27/*
 28 * Allocate and add a save area for a CPU
 29 */
 30struct 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 */
 55static 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 */
 71static 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 */
 93static void *elfcorehdr_newmem;
 94
 95/*
 96 * Copy one page from zfcpdump "oldmem"
 97 *
 98 * For pages below HSA size memory from the HSA is copied. Otherwise
 99 * real memory copy is used.
100 */
101static 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 < sclp_get_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 */
125static 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 */
147ssize_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 */
167static 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 HSA size. Memory below HSA size
192 * is read on demand using the copy_oldmem_page() function.
193 */
194static 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 hsa_end = sclp_get_hsa_size();
200	unsigned long size_hsa;
201
202	if (pfn < hsa_end >> PAGE_SHIFT) {
203		size_hsa = min(size, hsa_end - (pfn << PAGE_SHIFT));
204		if (size == size_hsa)
205			return 0;
206		size -= size_hsa;
207		from += size_hsa;
208		pfn += size_hsa >> PAGE_SHIFT;
209	}
210	return remap_pfn_range(vma, from, pfn, size, prot);
211}
212
213/*
214 * Remap "oldmem" for kdump or zfcpdump
215 */
216int remap_oldmem_pfn_range(struct vm_area_struct *vma, unsigned long from,
217			   unsigned long pfn, unsigned long size, pgprot_t prot)
218{
219	if (OLDMEM_BASE)
220		return remap_oldmem_pfn_range_kdump(vma, from, pfn, size, prot);
221	else
222		return remap_oldmem_pfn_range_zfcpdump(vma, from, pfn, size,
223						       prot);
224}
225
226/*
227 * Copy memory from old kernel
228 */
229int copy_from_oldmem(void *dest, void *src, size_t count)
230{
231	unsigned long copied = 0;
232	int rc;
233
234	if (OLDMEM_BASE) {
235		if ((unsigned long) src < OLDMEM_SIZE) {
236			copied = min(count, OLDMEM_SIZE - (unsigned long) src);
237			rc = copy_from_realmem(dest, src + OLDMEM_BASE, copied);
238			if (rc)
239				return rc;
240		}
241	} else {
242		unsigned long hsa_end = sclp_get_hsa_size();
243		if ((unsigned long) src < hsa_end) {
244			copied = min(count, hsa_end - (unsigned long) src);
245			rc = memcpy_hsa(dest, (unsigned long) src, copied, 0);
246			if (rc)
247				return rc;
248		}
249	}
250	return copy_from_realmem(dest + copied, src + copied, count - copied);
251}
252
253/*
254 * Alloc memory and panic in case of ENOMEM
255 */
256static void *kzalloc_panic(int len)
257{
258	void *rc;
259
260	rc = kzalloc(len, GFP_KERNEL);
261	if (!rc)
262		panic("s390 kdump kzalloc (%d) failed", len);
263	return rc;
264}
265
266/*
267 * Get memory layout and create hole for oldmem
268 */
269static struct mem_chunk *get_memory_layout(void)
270{
271	struct mem_chunk *chunk_array;
272
273	chunk_array = kzalloc_panic(MEMORY_CHUNKS * sizeof(struct mem_chunk));
274	detect_memory_layout(chunk_array, 0);
275	create_mem_hole(chunk_array, OLDMEM_BASE, OLDMEM_SIZE);
276	return chunk_array;
277}
278
279/*
280 * Initialize ELF note
281 */
282static void *nt_init(void *buf, Elf64_Word type, void *desc, int d_len,
283		     const char *name)
284{
285	Elf64_Nhdr *note;
286	u64 len;
287
288	note = (Elf64_Nhdr *)buf;
289	note->n_namesz = strlen(name) + 1;
290	note->n_descsz = d_len;
291	note->n_type = type;
292	len = sizeof(Elf64_Nhdr);
293
294	memcpy(buf + len, name, note->n_namesz);
295	len = roundup(len + note->n_namesz, 4);
296
297	memcpy(buf + len, desc, note->n_descsz);
298	len = roundup(len + note->n_descsz, 4);
299
300	return PTR_ADD(buf, len);
301}
302
303/*
304 * Initialize prstatus note
305 */
306static void *nt_prstatus(void *ptr, struct save_area *sa)
307{
308	struct elf_prstatus nt_prstatus;
309	static int cpu_nr = 1;
310
311	memset(&nt_prstatus, 0, sizeof(nt_prstatus));
312	memcpy(&nt_prstatus.pr_reg.gprs, sa->gp_regs, sizeof(sa->gp_regs));
313	memcpy(&nt_prstatus.pr_reg.psw, sa->psw, sizeof(sa->psw));
314	memcpy(&nt_prstatus.pr_reg.acrs, sa->acc_regs, sizeof(sa->acc_regs));
315	nt_prstatus.pr_pid = cpu_nr;
316	cpu_nr++;
317
318	return nt_init(ptr, NT_PRSTATUS, &nt_prstatus, sizeof(nt_prstatus),
319			 "CORE");
320}
321
322/*
323 * Initialize fpregset (floating point) note
324 */
325static void *nt_fpregset(void *ptr, struct save_area *sa)
326{
327	elf_fpregset_t nt_fpregset;
328
329	memset(&nt_fpregset, 0, sizeof(nt_fpregset));
330	memcpy(&nt_fpregset.fpc, &sa->fp_ctrl_reg, sizeof(sa->fp_ctrl_reg));
331	memcpy(&nt_fpregset.fprs, &sa->fp_regs, sizeof(sa->fp_regs));
332
333	return nt_init(ptr, NT_PRFPREG, &nt_fpregset, sizeof(nt_fpregset),
334		       "CORE");
335}
336
337/*
338 * Initialize timer note
339 */
340static void *nt_s390_timer(void *ptr, struct save_area *sa)
341{
342	return nt_init(ptr, NT_S390_TIMER, &sa->timer, sizeof(sa->timer),
343			 KEXEC_CORE_NOTE_NAME);
344}
345
346/*
347 * Initialize TOD clock comparator note
348 */
349static void *nt_s390_tod_cmp(void *ptr, struct save_area *sa)
350{
351	return nt_init(ptr, NT_S390_TODCMP, &sa->clk_cmp,
352		       sizeof(sa->clk_cmp), KEXEC_CORE_NOTE_NAME);
353}
354
355/*
356 * Initialize TOD programmable register note
357 */
358static void *nt_s390_tod_preg(void *ptr, struct save_area *sa)
359{
360	return nt_init(ptr, NT_S390_TODPREG, &sa->tod_reg,
361		       sizeof(sa->tod_reg), KEXEC_CORE_NOTE_NAME);
362}
363
364/*
365 * Initialize control register note
366 */
367static void *nt_s390_ctrs(void *ptr, struct save_area *sa)
368{
369	return nt_init(ptr, NT_S390_CTRS, &sa->ctrl_regs,
370		       sizeof(sa->ctrl_regs), KEXEC_CORE_NOTE_NAME);
371}
372
373/*
374 * Initialize prefix register note
375 */
376static void *nt_s390_prefix(void *ptr, struct save_area *sa)
377{
378	return nt_init(ptr, NT_S390_PREFIX, &sa->pref_reg,
379			 sizeof(sa->pref_reg), KEXEC_CORE_NOTE_NAME);
380}
381
382/*
383 * Fill ELF notes for one CPU with save area registers
384 */
385void *fill_cpu_elf_notes(void *ptr, struct save_area *sa)
386{
387	ptr = nt_prstatus(ptr, sa);
388	ptr = nt_fpregset(ptr, sa);
389	ptr = nt_s390_timer(ptr, sa);
390	ptr = nt_s390_tod_cmp(ptr, sa);
391	ptr = nt_s390_tod_preg(ptr, sa);
392	ptr = nt_s390_ctrs(ptr, sa);
393	ptr = nt_s390_prefix(ptr, sa);
394	return ptr;
395}
396
397/*
398 * Initialize prpsinfo note (new kernel)
399 */
400static void *nt_prpsinfo(void *ptr)
401{
402	struct elf_prpsinfo prpsinfo;
403
404	memset(&prpsinfo, 0, sizeof(prpsinfo));
405	prpsinfo.pr_sname = 'R';
406	strcpy(prpsinfo.pr_fname, "vmlinux");
407	return nt_init(ptr, NT_PRPSINFO, &prpsinfo, sizeof(prpsinfo),
408		       KEXEC_CORE_NOTE_NAME);
409}
410
411/*
412 * Get vmcoreinfo using lowcore->vmcore_info (new kernel)
413 */
414static void *get_vmcoreinfo_old(unsigned long *size)
415{
416	char nt_name[11], *vmcoreinfo;
417	Elf64_Nhdr note;
418	void *addr;
419
420	if (copy_from_oldmem(&addr, &S390_lowcore.vmcore_info, sizeof(addr)))
421		return NULL;
422	memset(nt_name, 0, sizeof(nt_name));
423	if (copy_from_oldmem(&note, addr, sizeof(note)))
424		return NULL;
425	if (copy_from_oldmem(nt_name, addr + sizeof(note), sizeof(nt_name) - 1))
426		return NULL;
427	if (strcmp(nt_name, "VMCOREINFO") != 0)
428		return NULL;
429	vmcoreinfo = kzalloc_panic(note.n_descsz);
430	if (copy_from_oldmem(vmcoreinfo, addr + 24, note.n_descsz))
431		return NULL;
432	*size = note.n_descsz;
433	return vmcoreinfo;
434}
435
436/*
437 * Initialize vmcoreinfo note (new kernel)
438 */
439static void *nt_vmcoreinfo(void *ptr)
440{
441	unsigned long size;
442	void *vmcoreinfo;
443
444	vmcoreinfo = os_info_old_entry(OS_INFO_VMCOREINFO, &size);
445	if (!vmcoreinfo)
446		vmcoreinfo = get_vmcoreinfo_old(&size);
447	if (!vmcoreinfo)
448		return ptr;
449	return nt_init(ptr, 0, vmcoreinfo, size, "VMCOREINFO");
450}
451
452/*
453 * Initialize ELF header (new kernel)
454 */
455static void *ehdr_init(Elf64_Ehdr *ehdr, int mem_chunk_cnt)
456{
457	memset(ehdr, 0, sizeof(*ehdr));
458	memcpy(ehdr->e_ident, ELFMAG, SELFMAG);
459	ehdr->e_ident[EI_CLASS] = ELFCLASS64;
460	ehdr->e_ident[EI_DATA] = ELFDATA2MSB;
461	ehdr->e_ident[EI_VERSION] = EV_CURRENT;
462	memset(ehdr->e_ident + EI_PAD, 0, EI_NIDENT - EI_PAD);
463	ehdr->e_type = ET_CORE;
464	ehdr->e_machine = EM_S390;
465	ehdr->e_version = EV_CURRENT;
466	ehdr->e_phoff = sizeof(Elf64_Ehdr);
467	ehdr->e_ehsize = sizeof(Elf64_Ehdr);
468	ehdr->e_phentsize = sizeof(Elf64_Phdr);
469	ehdr->e_phnum = mem_chunk_cnt + 1;
470	return ehdr + 1;
471}
472
473/*
474 * Return CPU count for ELF header (new kernel)
475 */
476static int get_cpu_cnt(void)
477{
478	int i, cpus = 0;
479
480	for (i = 0; i < dump_save_areas.count; i++) {
481		if (dump_save_areas.areas[i]->pref_reg == 0)
482			continue;
483		cpus++;
484	}
485	return cpus;
486}
487
488/*
489 * Return memory chunk count for ELF header (new kernel)
490 */
491static int get_mem_chunk_cnt(void)
492{
493	struct mem_chunk *chunk_array, *mem_chunk;
494	int i, cnt = 0;
495
496	chunk_array = get_memory_layout();
497	for (i = 0; i < MEMORY_CHUNKS; i++) {
498		mem_chunk = &chunk_array[i];
499		if (chunk_array[i].type != CHUNK_READ_WRITE &&
500		    chunk_array[i].type != CHUNK_READ_ONLY)
501			continue;
502		if (mem_chunk->size == 0)
503			continue;
504		cnt++;
505	}
506	kfree(chunk_array);
507	return cnt;
508}
509
510/*
511 * Initialize ELF loads (new kernel)
512 */
513static int loads_init(Elf64_Phdr *phdr, u64 loads_offset)
514{
515	struct mem_chunk *chunk_array, *mem_chunk;
516	int i;
517
518	chunk_array = get_memory_layout();
519	for (i = 0; i < MEMORY_CHUNKS; i++) {
520		mem_chunk = &chunk_array[i];
521		if (mem_chunk->size == 0)
522			continue;
523		if (chunk_array[i].type != CHUNK_READ_WRITE &&
524		    chunk_array[i].type != CHUNK_READ_ONLY)
525			continue;
526		else
527			phdr->p_filesz = mem_chunk->size;
528		phdr->p_type = PT_LOAD;
529		phdr->p_offset = mem_chunk->addr;
530		phdr->p_vaddr = mem_chunk->addr;
531		phdr->p_paddr = mem_chunk->addr;
532		phdr->p_memsz = mem_chunk->size;
533		phdr->p_flags = PF_R | PF_W | PF_X;
534		phdr->p_align = PAGE_SIZE;
535		phdr++;
536	}
537	kfree(chunk_array);
538	return i;
539}
540
541/*
542 * Initialize notes (new kernel)
543 */
544static void *notes_init(Elf64_Phdr *phdr, void *ptr, u64 notes_offset)
545{
546	struct save_area *sa;
547	void *ptr_start = ptr;
548	int i;
549
550	ptr = nt_prpsinfo(ptr);
551
552	for (i = 0; i < dump_save_areas.count; i++) {
553		sa = dump_save_areas.areas[i];
554		if (sa->pref_reg == 0)
555			continue;
556		ptr = fill_cpu_elf_notes(ptr, sa);
557	}
558	ptr = nt_vmcoreinfo(ptr);
559	memset(phdr, 0, sizeof(*phdr));
560	phdr->p_type = PT_NOTE;
561	phdr->p_offset = notes_offset;
562	phdr->p_filesz = (unsigned long) PTR_SUB(ptr, ptr_start);
563	phdr->p_memsz = phdr->p_filesz;
564	return ptr;
565}
566
567/*
568 * Create ELF core header (new kernel)
569 */
570int elfcorehdr_alloc(unsigned long long *addr, unsigned long long *size)
571{
572	Elf64_Phdr *phdr_notes, *phdr_loads;
573	int mem_chunk_cnt;
574	void *ptr, *hdr;
575	u32 alloc_size;
576	u64 hdr_off;
577
578	/* If we are not in kdump or zfcpdump mode return */
579	if (!OLDMEM_BASE && ipl_info.type != IPL_TYPE_FCP_DUMP)
580		return 0;
581	/* If elfcorehdr= has been passed via cmdline, we use that one */
582	if (elfcorehdr_addr != ELFCORE_ADDR_MAX)
583		return 0;
584	/* If we cannot get HSA size for zfcpdump return error */
585	if (ipl_info.type == IPL_TYPE_FCP_DUMP && !sclp_get_hsa_size())
586		return -ENODEV;
587	mem_chunk_cnt = get_mem_chunk_cnt();
588
589	alloc_size = 0x1000 + get_cpu_cnt() * 0x300 +
590		mem_chunk_cnt * sizeof(Elf64_Phdr);
591	hdr = kzalloc_panic(alloc_size);
592	/* Init elf header */
593	ptr = ehdr_init(hdr, mem_chunk_cnt);
594	/* Init program headers */
595	phdr_notes = ptr;
596	ptr = PTR_ADD(ptr, sizeof(Elf64_Phdr));
597	phdr_loads = ptr;
598	ptr = PTR_ADD(ptr, sizeof(Elf64_Phdr) * mem_chunk_cnt);
599	/* Init notes */
600	hdr_off = PTR_DIFF(ptr, hdr);
601	ptr = notes_init(phdr_notes, ptr, ((unsigned long) hdr) + hdr_off);
602	/* Init loads */
603	hdr_off = PTR_DIFF(ptr, hdr);
604	loads_init(phdr_loads, hdr_off);
605	*addr = (unsigned long long) hdr;
606	elfcorehdr_newmem = hdr;
607	*size = (unsigned long long) hdr_off;
608	BUG_ON(elfcorehdr_size > alloc_size);
609	return 0;
610}
611
612/*
613 * Free ELF core header (new kernel)
614 */
615void elfcorehdr_free(unsigned long long addr)
616{
617	if (!elfcorehdr_newmem)
618		return;
619	kfree((void *)(unsigned long)addr);
620}
621
622/*
623 * Read from ELF header
624 */
625ssize_t elfcorehdr_read(char *buf, size_t count, u64 *ppos)
626{
627	void *src = (void *)(unsigned long)*ppos;
628
629	src = elfcorehdr_newmem ? src : src - OLDMEM_BASE;
630	memcpy(buf, src, count);
631	*ppos += count;
632	return count;
633}
634
635/*
636 * Read from ELF notes data
637 */
638ssize_t elfcorehdr_read_notes(char *buf, size_t count, u64 *ppos)
639{
640	void *src = (void *)(unsigned long)*ppos;
641	int rc;
642
643	if (elfcorehdr_newmem) {
644		memcpy(buf, src, count);
645	} else {
646		rc = copy_from_oldmem(buf, src, count);
647		if (rc)
648			return rc;
649	}
650	*ppos += count;
651	return count;
652}