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 26 /* 27 * Return physical address for virtual address 28 */ 29 static inline void *load_real_addr(void *addr) 30 { 31 unsigned long real_addr; 32 33 asm volatile( 34 " lra %0,0(%1)\n" 35 " jz 0f\n" 36 " la %0,0\n" 37 "0:" 38 : "=a" (real_addr) : "a" (addr) : "cc"); 39 return (void *)real_addr; 40 } 41 42 /* 43 * Copy real to virtual or real memory 44 */ 45 static int copy_from_realmem(void *dest, void *src, size_t count) 46 { 47 unsigned long size; 48 int rc; 49 50 if (!count) 51 return 0; 52 if (!is_vmalloc_or_module_addr(dest)) 53 return memcpy_real(dest, src, count); 54 do { 55 size = min(count, PAGE_SIZE - (__pa(dest) & ~PAGE_MASK)); 56 if (memcpy_real(load_real_addr(dest), src, size)) 57 return -EFAULT; 58 count -= size; 59 dest += size; 60 src += size; 61 } while (count); 62 return 0; 63 } 64 65 /* 66 * Pointer to ELF header in new kernel 67 */ 68 static void *elfcorehdr_newmem; 69 70 /* 71 * Copy one page from zfcpdump "oldmem" 72 * 73 * For pages below ZFCPDUMP_HSA_SIZE memory from the HSA is copied. Otherwise 74 * real memory copy is used. 75 */ 76 static ssize_t copy_oldmem_page_zfcpdump(char *buf, size_t csize, 77 unsigned long src, int userbuf) 78 { 79 int rc; 80 81 if (src < ZFCPDUMP_HSA_SIZE) { 82 rc = memcpy_hsa(buf, src, csize, userbuf); 83 } else { 84 if (userbuf) 85 rc = copy_to_user_real((void __force __user *) buf, 86 (void *) src, csize); 87 else 88 rc = memcpy_real(buf, (void *) src, csize); 89 } 90 return rc ? rc : csize; 91 } 92 93 /* 94 * Copy one page from kdump "oldmem" 95 * 96 * For the kdump reserved memory this functions performs a swap operation: 97 * - [OLDMEM_BASE - OLDMEM_BASE + OLDMEM_SIZE] is mapped to [0 - OLDMEM_SIZE]. 98 * - [0 - OLDMEM_SIZE] is mapped to [OLDMEM_BASE - OLDMEM_BASE + OLDMEM_SIZE] 99 */ 100 static ssize_t copy_oldmem_page_kdump(char *buf, size_t csize, 101 unsigned long src, int userbuf) 102 103 { 104 int rc; 105 106 if (src < OLDMEM_SIZE) 107 src += OLDMEM_BASE; 108 else if (src > OLDMEM_BASE && 109 src < OLDMEM_BASE + OLDMEM_SIZE) 110 src -= OLDMEM_BASE; 111 if (userbuf) 112 rc = copy_to_user_real((void __force __user *) buf, 113 (void *) src, csize); 114 else 115 rc = copy_from_realmem(buf, (void *) src, csize); 116 return (rc == 0) ? rc : csize; 117 } 118 119 /* 120 * Copy one page from "oldmem" 121 */ 122 ssize_t copy_oldmem_page(unsigned long pfn, char *buf, size_t csize, 123 unsigned long offset, int userbuf) 124 { 125 unsigned long src; 126 127 if (!csize) 128 return 0; 129 src = (pfn << PAGE_SHIFT) + offset; 130 if (OLDMEM_BASE) 131 return copy_oldmem_page_kdump(buf, csize, src, userbuf); 132 else 133 return copy_oldmem_page_zfcpdump(buf, csize, src, userbuf); 134 } 135 136 /* 137 * Remap "oldmem" for kdump 138 * 139 * For the kdump reserved memory this functions performs a swap operation: 140 * [0 - OLDMEM_SIZE] is mapped to [OLDMEM_BASE - OLDMEM_BASE + OLDMEM_SIZE] 141 */ 142 static int remap_oldmem_pfn_range_kdump(struct vm_area_struct *vma, 143 unsigned long from, unsigned long pfn, 144 unsigned long size, pgprot_t prot) 145 { 146 unsigned long size_old; 147 int rc; 148 149 if (pfn < OLDMEM_SIZE >> PAGE_SHIFT) { 150 size_old = min(size, OLDMEM_SIZE - (pfn << PAGE_SHIFT)); 151 rc = remap_pfn_range(vma, from, 152 pfn + (OLDMEM_BASE >> PAGE_SHIFT), 153 size_old, prot); 154 if (rc || size == size_old) 155 return rc; 156 size -= size_old; 157 from += size_old; 158 pfn += size_old >> PAGE_SHIFT; 159 } 160 return remap_pfn_range(vma, from, pfn, size, prot); 161 } 162 163 /* 164 * Remap "oldmem" for zfcpdump 165 * 166 * We only map available memory above ZFCPDUMP_HSA_SIZE. Memory below 167 * ZFCPDUMP_HSA_SIZE is read on demand using the copy_oldmem_page() function. 168 */ 169 static int remap_oldmem_pfn_range_zfcpdump(struct vm_area_struct *vma, 170 unsigned long from, 171 unsigned long pfn, 172 unsigned long size, pgprot_t prot) 173 { 174 unsigned long size_hsa; 175 176 if (pfn < ZFCPDUMP_HSA_SIZE >> PAGE_SHIFT) { 177 size_hsa = min(size, ZFCPDUMP_HSA_SIZE - (pfn << PAGE_SHIFT)); 178 if (size == size_hsa) 179 return 0; 180 size -= size_hsa; 181 from += size_hsa; 182 pfn += size_hsa >> PAGE_SHIFT; 183 } 184 return remap_pfn_range(vma, from, pfn, size, prot); 185 } 186 187 /* 188 * Remap "oldmem" for kdump or zfcpdump 189 */ 190 int remap_oldmem_pfn_range(struct vm_area_struct *vma, unsigned long from, 191 unsigned long pfn, unsigned long size, pgprot_t prot) 192 { 193 if (OLDMEM_BASE) 194 return remap_oldmem_pfn_range_kdump(vma, from, pfn, size, prot); 195 else 196 return remap_oldmem_pfn_range_zfcpdump(vma, from, pfn, size, 197 prot); 198 } 199 200 /* 201 * Copy memory from old kernel 202 */ 203 int copy_from_oldmem(void *dest, void *src, size_t count) 204 { 205 unsigned long copied = 0; 206 int rc; 207 208 if (OLDMEM_BASE) { 209 if ((unsigned long) src < OLDMEM_SIZE) { 210 copied = min(count, OLDMEM_SIZE - (unsigned long) src); 211 rc = copy_from_realmem(dest, src + OLDMEM_BASE, copied); 212 if (rc) 213 return rc; 214 } 215 } else { 216 if ((unsigned long) src < ZFCPDUMP_HSA_SIZE) { 217 copied = min(count, 218 ZFCPDUMP_HSA_SIZE - (unsigned long) src); 219 rc = memcpy_hsa(dest, (unsigned long) src, copied, 0); 220 if (rc) 221 return rc; 222 } 223 } 224 return copy_from_realmem(dest + copied, src + copied, count - copied); 225 } 226 227 /* 228 * Alloc memory and panic in case of ENOMEM 229 */ 230 static void *kzalloc_panic(int len) 231 { 232 void *rc; 233 234 rc = kzalloc(len, GFP_KERNEL); 235 if (!rc) 236 panic("s390 kdump kzalloc (%d) failed", len); 237 return rc; 238 } 239 240 /* 241 * Get memory layout and create hole for oldmem 242 */ 243 static struct mem_chunk *get_memory_layout(void) 244 { 245 struct mem_chunk *chunk_array; 246 247 chunk_array = kzalloc_panic(MEMORY_CHUNKS * sizeof(struct mem_chunk)); 248 detect_memory_layout(chunk_array, 0); 249 create_mem_hole(chunk_array, OLDMEM_BASE, OLDMEM_SIZE); 250 return chunk_array; 251 } 252 253 /* 254 * Initialize ELF note 255 */ 256 static void *nt_init(void *buf, Elf64_Word type, void *desc, int d_len, 257 const char *name) 258 { 259 Elf64_Nhdr *note; 260 u64 len; 261 262 note = (Elf64_Nhdr *)buf; 263 note->n_namesz = strlen(name) + 1; 264 note->n_descsz = d_len; 265 note->n_type = type; 266 len = sizeof(Elf64_Nhdr); 267 268 memcpy(buf + len, name, note->n_namesz); 269 len = roundup(len + note->n_namesz, 4); 270 271 memcpy(buf + len, desc, note->n_descsz); 272 len = roundup(len + note->n_descsz, 4); 273 274 return PTR_ADD(buf, len); 275 } 276 277 /* 278 * Initialize prstatus note 279 */ 280 static void *nt_prstatus(void *ptr, struct save_area *sa) 281 { 282 struct elf_prstatus nt_prstatus; 283 static int cpu_nr = 1; 284 285 memset(&nt_prstatus, 0, sizeof(nt_prstatus)); 286 memcpy(&nt_prstatus.pr_reg.gprs, sa->gp_regs, sizeof(sa->gp_regs)); 287 memcpy(&nt_prstatus.pr_reg.psw, sa->psw, sizeof(sa->psw)); 288 memcpy(&nt_prstatus.pr_reg.acrs, sa->acc_regs, sizeof(sa->acc_regs)); 289 nt_prstatus.pr_pid = cpu_nr; 290 cpu_nr++; 291 292 return nt_init(ptr, NT_PRSTATUS, &nt_prstatus, sizeof(nt_prstatus), 293 "CORE"); 294 } 295 296 /* 297 * Initialize fpregset (floating point) note 298 */ 299 static void *nt_fpregset(void *ptr, struct save_area *sa) 300 { 301 elf_fpregset_t nt_fpregset; 302 303 memset(&nt_fpregset, 0, sizeof(nt_fpregset)); 304 memcpy(&nt_fpregset.fpc, &sa->fp_ctrl_reg, sizeof(sa->fp_ctrl_reg)); 305 memcpy(&nt_fpregset.fprs, &sa->fp_regs, sizeof(sa->fp_regs)); 306 307 return nt_init(ptr, NT_PRFPREG, &nt_fpregset, sizeof(nt_fpregset), 308 "CORE"); 309 } 310 311 /* 312 * Initialize timer note 313 */ 314 static void *nt_s390_timer(void *ptr, struct save_area *sa) 315 { 316 return nt_init(ptr, NT_S390_TIMER, &sa->timer, sizeof(sa->timer), 317 KEXEC_CORE_NOTE_NAME); 318 } 319 320 /* 321 * Initialize TOD clock comparator note 322 */ 323 static void *nt_s390_tod_cmp(void *ptr, struct save_area *sa) 324 { 325 return nt_init(ptr, NT_S390_TODCMP, &sa->clk_cmp, 326 sizeof(sa->clk_cmp), KEXEC_CORE_NOTE_NAME); 327 } 328 329 /* 330 * Initialize TOD programmable register note 331 */ 332 static void *nt_s390_tod_preg(void *ptr, struct save_area *sa) 333 { 334 return nt_init(ptr, NT_S390_TODPREG, &sa->tod_reg, 335 sizeof(sa->tod_reg), KEXEC_CORE_NOTE_NAME); 336 } 337 338 /* 339 * Initialize control register note 340 */ 341 static void *nt_s390_ctrs(void *ptr, struct save_area *sa) 342 { 343 return nt_init(ptr, NT_S390_CTRS, &sa->ctrl_regs, 344 sizeof(sa->ctrl_regs), KEXEC_CORE_NOTE_NAME); 345 } 346 347 /* 348 * Initialize prefix register note 349 */ 350 static void *nt_s390_prefix(void *ptr, struct save_area *sa) 351 { 352 return nt_init(ptr, NT_S390_PREFIX, &sa->pref_reg, 353 sizeof(sa->pref_reg), KEXEC_CORE_NOTE_NAME); 354 } 355 356 /* 357 * Fill ELF notes for one CPU with save area registers 358 */ 359 void *fill_cpu_elf_notes(void *ptr, struct save_area *sa) 360 { 361 ptr = nt_prstatus(ptr, sa); 362 ptr = nt_fpregset(ptr, sa); 363 ptr = nt_s390_timer(ptr, sa); 364 ptr = nt_s390_tod_cmp(ptr, sa); 365 ptr = nt_s390_tod_preg(ptr, sa); 366 ptr = nt_s390_ctrs(ptr, sa); 367 ptr = nt_s390_prefix(ptr, sa); 368 return ptr; 369 } 370 371 /* 372 * Initialize prpsinfo note (new kernel) 373 */ 374 static void *nt_prpsinfo(void *ptr) 375 { 376 struct elf_prpsinfo prpsinfo; 377 378 memset(&prpsinfo, 0, sizeof(prpsinfo)); 379 prpsinfo.pr_sname = 'R'; 380 strcpy(prpsinfo.pr_fname, "vmlinux"); 381 return nt_init(ptr, NT_PRPSINFO, &prpsinfo, sizeof(prpsinfo), 382 KEXEC_CORE_NOTE_NAME); 383 } 384 385 /* 386 * Get vmcoreinfo using lowcore->vmcore_info (new kernel) 387 */ 388 static void *get_vmcoreinfo_old(unsigned long *size) 389 { 390 char nt_name[11], *vmcoreinfo; 391 Elf64_Nhdr note; 392 void *addr; 393 394 if (copy_from_oldmem(&addr, &S390_lowcore.vmcore_info, sizeof(addr))) 395 return NULL; 396 memset(nt_name, 0, sizeof(nt_name)); 397 if (copy_from_oldmem(¬e, addr, sizeof(note))) 398 return NULL; 399 if (copy_from_oldmem(nt_name, addr + sizeof(note), sizeof(nt_name) - 1)) 400 return NULL; 401 if (strcmp(nt_name, "VMCOREINFO") != 0) 402 return NULL; 403 vmcoreinfo = kzalloc_panic(note.n_descsz); 404 if (copy_from_oldmem(vmcoreinfo, addr + 24, note.n_descsz)) 405 return NULL; 406 *size = note.n_descsz; 407 return vmcoreinfo; 408 } 409 410 /* 411 * Initialize vmcoreinfo note (new kernel) 412 */ 413 static void *nt_vmcoreinfo(void *ptr) 414 { 415 unsigned long size; 416 void *vmcoreinfo; 417 418 vmcoreinfo = os_info_old_entry(OS_INFO_VMCOREINFO, &size); 419 if (!vmcoreinfo) 420 vmcoreinfo = get_vmcoreinfo_old(&size); 421 if (!vmcoreinfo) 422 return ptr; 423 return nt_init(ptr, 0, vmcoreinfo, size, "VMCOREINFO"); 424 } 425 426 /* 427 * Initialize ELF header (new kernel) 428 */ 429 static void *ehdr_init(Elf64_Ehdr *ehdr, int mem_chunk_cnt) 430 { 431 memset(ehdr, 0, sizeof(*ehdr)); 432 memcpy(ehdr->e_ident, ELFMAG, SELFMAG); 433 ehdr->e_ident[EI_CLASS] = ELFCLASS64; 434 ehdr->e_ident[EI_DATA] = ELFDATA2MSB; 435 ehdr->e_ident[EI_VERSION] = EV_CURRENT; 436 memset(ehdr->e_ident + EI_PAD, 0, EI_NIDENT - EI_PAD); 437 ehdr->e_type = ET_CORE; 438 ehdr->e_machine = EM_S390; 439 ehdr->e_version = EV_CURRENT; 440 ehdr->e_phoff = sizeof(Elf64_Ehdr); 441 ehdr->e_ehsize = sizeof(Elf64_Ehdr); 442 ehdr->e_phentsize = sizeof(Elf64_Phdr); 443 ehdr->e_phnum = mem_chunk_cnt + 1; 444 return ehdr + 1; 445 } 446 447 /* 448 * Return CPU count for ELF header (new kernel) 449 */ 450 static int get_cpu_cnt(void) 451 { 452 int i, cpus = 0; 453 454 for (i = 0; zfcpdump_save_areas[i]; i++) { 455 if (zfcpdump_save_areas[i]->pref_reg == 0) 456 continue; 457 cpus++; 458 } 459 return cpus; 460 } 461 462 /* 463 * Return memory chunk count for ELF header (new kernel) 464 */ 465 static int get_mem_chunk_cnt(void) 466 { 467 struct mem_chunk *chunk_array, *mem_chunk; 468 int i, cnt = 0; 469 470 chunk_array = get_memory_layout(); 471 for (i = 0; i < MEMORY_CHUNKS; i++) { 472 mem_chunk = &chunk_array[i]; 473 if (chunk_array[i].type != CHUNK_READ_WRITE && 474 chunk_array[i].type != CHUNK_READ_ONLY) 475 continue; 476 if (mem_chunk->size == 0) 477 continue; 478 cnt++; 479 } 480 kfree(chunk_array); 481 return cnt; 482 } 483 484 /* 485 * Initialize ELF loads (new kernel) 486 */ 487 static int loads_init(Elf64_Phdr *phdr, u64 loads_offset) 488 { 489 struct mem_chunk *chunk_array, *mem_chunk; 490 int i; 491 492 chunk_array = get_memory_layout(); 493 for (i = 0; i < MEMORY_CHUNKS; i++) { 494 mem_chunk = &chunk_array[i]; 495 if (mem_chunk->size == 0) 496 continue; 497 if (chunk_array[i].type != CHUNK_READ_WRITE && 498 chunk_array[i].type != CHUNK_READ_ONLY) 499 continue; 500 else 501 phdr->p_filesz = mem_chunk->size; 502 phdr->p_type = PT_LOAD; 503 phdr->p_offset = mem_chunk->addr; 504 phdr->p_vaddr = mem_chunk->addr; 505 phdr->p_paddr = mem_chunk->addr; 506 phdr->p_memsz = mem_chunk->size; 507 phdr->p_flags = PF_R | PF_W | PF_X; 508 phdr->p_align = PAGE_SIZE; 509 phdr++; 510 } 511 kfree(chunk_array); 512 return i; 513 } 514 515 /* 516 * Initialize notes (new kernel) 517 */ 518 static void *notes_init(Elf64_Phdr *phdr, void *ptr, u64 notes_offset) 519 { 520 struct save_area *sa; 521 void *ptr_start = ptr; 522 int i; 523 524 ptr = nt_prpsinfo(ptr); 525 526 for (i = 0; zfcpdump_save_areas[i]; i++) { 527 sa = zfcpdump_save_areas[i]; 528 if (sa->pref_reg == 0) 529 continue; 530 ptr = fill_cpu_elf_notes(ptr, sa); 531 } 532 ptr = nt_vmcoreinfo(ptr); 533 memset(phdr, 0, sizeof(*phdr)); 534 phdr->p_type = PT_NOTE; 535 phdr->p_offset = notes_offset; 536 phdr->p_filesz = (unsigned long) PTR_SUB(ptr, ptr_start); 537 phdr->p_memsz = phdr->p_filesz; 538 return ptr; 539 } 540 541 /* 542 * Create ELF core header (new kernel) 543 */ 544 int elfcorehdr_alloc(unsigned long long *addr, unsigned long long *size) 545 { 546 Elf64_Phdr *phdr_notes, *phdr_loads; 547 int mem_chunk_cnt; 548 void *ptr, *hdr; 549 u32 alloc_size; 550 u64 hdr_off; 551 552 /* If we are not in kdump or zfcpdump mode return */ 553 if (!OLDMEM_BASE && ipl_info.type != IPL_TYPE_FCP_DUMP) 554 return 0; 555 /* If elfcorehdr= has been passed via cmdline, we use that one */ 556 if (elfcorehdr_addr != ELFCORE_ADDR_MAX) 557 return 0; 558 mem_chunk_cnt = get_mem_chunk_cnt(); 559 560 alloc_size = 0x1000 + get_cpu_cnt() * 0x300 + 561 mem_chunk_cnt * sizeof(Elf64_Phdr); 562 hdr = kzalloc_panic(alloc_size); 563 /* Init elf header */ 564 ptr = ehdr_init(hdr, mem_chunk_cnt); 565 /* Init program headers */ 566 phdr_notes = ptr; 567 ptr = PTR_ADD(ptr, sizeof(Elf64_Phdr)); 568 phdr_loads = ptr; 569 ptr = PTR_ADD(ptr, sizeof(Elf64_Phdr) * mem_chunk_cnt); 570 /* Init notes */ 571 hdr_off = PTR_DIFF(ptr, hdr); 572 ptr = notes_init(phdr_notes, ptr, ((unsigned long) hdr) + hdr_off); 573 /* Init loads */ 574 hdr_off = PTR_DIFF(ptr, hdr); 575 loads_init(phdr_loads, hdr_off); 576 *addr = (unsigned long long) hdr; 577 elfcorehdr_newmem = hdr; 578 *size = (unsigned long long) hdr_off; 579 BUG_ON(elfcorehdr_size > alloc_size); 580 return 0; 581 } 582 583 /* 584 * Free ELF core header (new kernel) 585 */ 586 void elfcorehdr_free(unsigned long long addr) 587 { 588 if (!elfcorehdr_newmem) 589 return; 590 kfree((void *)(unsigned long)addr); 591 } 592 593 /* 594 * Read from ELF header 595 */ 596 ssize_t elfcorehdr_read(char *buf, size_t count, u64 *ppos) 597 { 598 void *src = (void *)(unsigned long)*ppos; 599 600 src = elfcorehdr_newmem ? src : src - OLDMEM_BASE; 601 memcpy(buf, src, count); 602 *ppos += count; 603 return count; 604 } 605 606 /* 607 * Read from ELF notes data 608 */ 609 ssize_t elfcorehdr_read_notes(char *buf, size_t count, u64 *ppos) 610 { 611 void *src = (void *)(unsigned long)*ppos; 612 int rc; 613 614 if (elfcorehdr_newmem) { 615 memcpy(buf, src, count); 616 } else { 617 rc = copy_from_oldmem(buf, src, count); 618 if (rc) 619 return rc; 620 } 621 *ppos += count; 622 return count; 623 } 624