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/init.h> 12 #include <linux/mm.h> 13 #include <linux/gfp.h> 14 #include <linux/slab.h> 15 #include <linux/bootmem.h> 16 #include <linux/elf.h> 17 #include <asm/asm-offsets.h> 18 #include <linux/memblock.h> 19 #include <asm/os_info.h> 20 #include <asm/elf.h> 21 #include <asm/ipl.h> 22 #include <asm/sclp.h> 23 24 #define PTR_ADD(x, y) (((char *) (x)) + ((unsigned long) (y))) 25 #define PTR_SUB(x, y) (((char *) (x)) - ((unsigned long) (y))) 26 #define PTR_DIFF(x, y) ((unsigned long)(((char *) (x)) - ((unsigned long) (y)))) 27 28 static struct memblock_region oldmem_region; 29 30 static struct memblock_type oldmem_type = { 31 .cnt = 1, 32 .max = 1, 33 .total_size = 0, 34 .regions = &oldmem_region, 35 }; 36 37 struct save_area { 38 struct list_head list; 39 u64 psw[2]; 40 u64 ctrs[16]; 41 u64 gprs[16]; 42 u32 acrs[16]; 43 u64 fprs[16]; 44 u32 fpc; 45 u32 prefix; 46 u64 todpreg; 47 u64 timer; 48 u64 todcmp; 49 u64 vxrs_low[16]; 50 __vector128 vxrs_high[16]; 51 }; 52 53 static LIST_HEAD(dump_save_areas); 54 55 /* 56 * Allocate a save area 57 */ 58 struct save_area * __init save_area_alloc(bool is_boot_cpu) 59 { 60 struct save_area *sa; 61 62 sa = (void *) memblock_alloc(sizeof(*sa), 8); 63 if (is_boot_cpu) 64 list_add(&sa->list, &dump_save_areas); 65 else 66 list_add_tail(&sa->list, &dump_save_areas); 67 return sa; 68 } 69 70 /* 71 * Return the address of the save area for the boot CPU 72 */ 73 struct save_area * __init save_area_boot_cpu(void) 74 { 75 return list_first_entry_or_null(&dump_save_areas, struct save_area, list); 76 } 77 78 /* 79 * Copy CPU registers into the save area 80 */ 81 void __init save_area_add_regs(struct save_area *sa, void *regs) 82 { 83 struct lowcore *lc; 84 85 lc = (struct lowcore *)(regs - __LC_FPREGS_SAVE_AREA); 86 memcpy(&sa->psw, &lc->psw_save_area, sizeof(sa->psw)); 87 memcpy(&sa->ctrs, &lc->cregs_save_area, sizeof(sa->ctrs)); 88 memcpy(&sa->gprs, &lc->gpregs_save_area, sizeof(sa->gprs)); 89 memcpy(&sa->acrs, &lc->access_regs_save_area, sizeof(sa->acrs)); 90 memcpy(&sa->fprs, &lc->floating_pt_save_area, sizeof(sa->fprs)); 91 memcpy(&sa->fpc, &lc->fpt_creg_save_area, sizeof(sa->fpc)); 92 memcpy(&sa->prefix, &lc->prefixreg_save_area, sizeof(sa->prefix)); 93 memcpy(&sa->todpreg, &lc->tod_progreg_save_area, sizeof(sa->todpreg)); 94 memcpy(&sa->timer, &lc->cpu_timer_save_area, sizeof(sa->timer)); 95 memcpy(&sa->todcmp, &lc->clock_comp_save_area, sizeof(sa->todcmp)); 96 } 97 98 /* 99 * Copy vector registers into the save area 100 */ 101 void __init save_area_add_vxrs(struct save_area *sa, __vector128 *vxrs) 102 { 103 int i; 104 105 /* Copy lower halves of vector registers 0-15 */ 106 for (i = 0; i < 16; i++) 107 memcpy(&sa->vxrs_low[i], &vxrs[i].u[2], 8); 108 /* Copy vector registers 16-31 */ 109 memcpy(sa->vxrs_high, vxrs + 16, 16 * sizeof(__vector128)); 110 } 111 112 /* 113 * Return physical address for virtual address 114 */ 115 static inline void *load_real_addr(void *addr) 116 { 117 unsigned long real_addr; 118 119 asm volatile( 120 " lra %0,0(%1)\n" 121 " jz 0f\n" 122 " la %0,0\n" 123 "0:" 124 : "=a" (real_addr) : "a" (addr) : "cc"); 125 return (void *)real_addr; 126 } 127 128 /* 129 * Copy memory of the old, dumped system to a kernel space virtual address 130 */ 131 int copy_oldmem_kernel(void *dst, void *src, size_t count) 132 { 133 unsigned long from, len; 134 void *ra; 135 int rc; 136 137 while (count) { 138 from = __pa(src); 139 if (!OLDMEM_BASE && from < sclp.hsa_size) { 140 /* Copy from zfcpdump HSA area */ 141 len = min(count, sclp.hsa_size - from); 142 rc = memcpy_hsa_kernel(dst, from, len); 143 if (rc) 144 return rc; 145 } else { 146 /* Check for swapped kdump oldmem areas */ 147 if (OLDMEM_BASE && from - OLDMEM_BASE < OLDMEM_SIZE) { 148 from -= OLDMEM_BASE; 149 len = min(count, OLDMEM_SIZE - from); 150 } else if (OLDMEM_BASE && from < OLDMEM_SIZE) { 151 len = min(count, OLDMEM_SIZE - from); 152 from += OLDMEM_BASE; 153 } else { 154 len = count; 155 } 156 if (is_vmalloc_or_module_addr(dst)) { 157 ra = load_real_addr(dst); 158 len = min(PAGE_SIZE - offset_in_page(ra), len); 159 } else { 160 ra = dst; 161 } 162 if (memcpy_real(ra, (void *) from, len)) 163 return -EFAULT; 164 } 165 dst += len; 166 src += len; 167 count -= len; 168 } 169 return 0; 170 } 171 172 /* 173 * Copy memory of the old, dumped system to a user space virtual address 174 */ 175 static int copy_oldmem_user(void __user *dst, void *src, size_t count) 176 { 177 unsigned long from, len; 178 int rc; 179 180 while (count) { 181 from = __pa(src); 182 if (!OLDMEM_BASE && from < sclp.hsa_size) { 183 /* Copy from zfcpdump HSA area */ 184 len = min(count, sclp.hsa_size - from); 185 rc = memcpy_hsa_user(dst, from, len); 186 if (rc) 187 return rc; 188 } else { 189 /* Check for swapped kdump oldmem areas */ 190 if (OLDMEM_BASE && from - OLDMEM_BASE < OLDMEM_SIZE) { 191 from -= OLDMEM_BASE; 192 len = min(count, OLDMEM_SIZE - from); 193 } else if (OLDMEM_BASE && from < OLDMEM_SIZE) { 194 len = min(count, OLDMEM_SIZE - from); 195 from += OLDMEM_BASE; 196 } else { 197 len = count; 198 } 199 rc = copy_to_user_real(dst, (void *) from, count); 200 if (rc) 201 return rc; 202 } 203 dst += len; 204 src += len; 205 count -= len; 206 } 207 return 0; 208 } 209 210 /* 211 * Copy one page from "oldmem" 212 */ 213 ssize_t copy_oldmem_page(unsigned long pfn, char *buf, size_t csize, 214 unsigned long offset, int userbuf) 215 { 216 void *src; 217 int rc; 218 219 if (!csize) 220 return 0; 221 src = (void *) (pfn << PAGE_SHIFT) + offset; 222 if (userbuf) 223 rc = copy_oldmem_user((void __force __user *) buf, src, csize); 224 else 225 rc = copy_oldmem_kernel((void *) buf, src, csize); 226 return rc; 227 } 228 229 /* 230 * Remap "oldmem" for kdump 231 * 232 * For the kdump reserved memory this functions performs a swap operation: 233 * [0 - OLDMEM_SIZE] is mapped to [OLDMEM_BASE - OLDMEM_BASE + OLDMEM_SIZE] 234 */ 235 static int remap_oldmem_pfn_range_kdump(struct vm_area_struct *vma, 236 unsigned long from, unsigned long pfn, 237 unsigned long size, pgprot_t prot) 238 { 239 unsigned long size_old; 240 int rc; 241 242 if (pfn < OLDMEM_SIZE >> PAGE_SHIFT) { 243 size_old = min(size, OLDMEM_SIZE - (pfn << PAGE_SHIFT)); 244 rc = remap_pfn_range(vma, from, 245 pfn + (OLDMEM_BASE >> PAGE_SHIFT), 246 size_old, prot); 247 if (rc || size == size_old) 248 return rc; 249 size -= size_old; 250 from += size_old; 251 pfn += size_old >> PAGE_SHIFT; 252 } 253 return remap_pfn_range(vma, from, pfn, size, prot); 254 } 255 256 /* 257 * Remap "oldmem" for zfcpdump 258 * 259 * We only map available memory above HSA size. Memory below HSA size 260 * is read on demand using the copy_oldmem_page() function. 261 */ 262 static int remap_oldmem_pfn_range_zfcpdump(struct vm_area_struct *vma, 263 unsigned long from, 264 unsigned long pfn, 265 unsigned long size, pgprot_t prot) 266 { 267 unsigned long hsa_end = sclp.hsa_size; 268 unsigned long size_hsa; 269 270 if (pfn < hsa_end >> PAGE_SHIFT) { 271 size_hsa = min(size, hsa_end - (pfn << PAGE_SHIFT)); 272 if (size == size_hsa) 273 return 0; 274 size -= size_hsa; 275 from += size_hsa; 276 pfn += size_hsa >> PAGE_SHIFT; 277 } 278 return remap_pfn_range(vma, from, pfn, size, prot); 279 } 280 281 /* 282 * Remap "oldmem" for kdump or zfcpdump 283 */ 284 int remap_oldmem_pfn_range(struct vm_area_struct *vma, unsigned long from, 285 unsigned long pfn, unsigned long size, pgprot_t prot) 286 { 287 if (OLDMEM_BASE) 288 return remap_oldmem_pfn_range_kdump(vma, from, pfn, size, prot); 289 else 290 return remap_oldmem_pfn_range_zfcpdump(vma, from, pfn, size, 291 prot); 292 } 293 294 /* 295 * Alloc memory and panic in case of ENOMEM 296 */ 297 static void *kzalloc_panic(int len) 298 { 299 void *rc; 300 301 rc = kzalloc(len, GFP_KERNEL); 302 if (!rc) 303 panic("s390 kdump kzalloc (%d) failed", len); 304 return rc; 305 } 306 307 /* 308 * Initialize ELF note 309 */ 310 static void *nt_init_name(void *buf, Elf64_Word type, void *desc, int d_len, 311 const char *name) 312 { 313 Elf64_Nhdr *note; 314 u64 len; 315 316 note = (Elf64_Nhdr *)buf; 317 note->n_namesz = strlen(name) + 1; 318 note->n_descsz = d_len; 319 note->n_type = type; 320 len = sizeof(Elf64_Nhdr); 321 322 memcpy(buf + len, name, note->n_namesz); 323 len = roundup(len + note->n_namesz, 4); 324 325 memcpy(buf + len, desc, note->n_descsz); 326 len = roundup(len + note->n_descsz, 4); 327 328 return PTR_ADD(buf, len); 329 } 330 331 static inline void *nt_init(void *buf, Elf64_Word type, void *desc, int d_len) 332 { 333 const char *note_name = "LINUX"; 334 335 if (type == NT_PRPSINFO || type == NT_PRSTATUS || type == NT_PRFPREG) 336 note_name = KEXEC_CORE_NOTE_NAME; 337 return nt_init_name(buf, type, desc, d_len, note_name); 338 } 339 340 /* 341 * Fill ELF notes for one CPU with save area registers 342 */ 343 static void *fill_cpu_elf_notes(void *ptr, int cpu, struct save_area *sa) 344 { 345 struct elf_prstatus nt_prstatus; 346 elf_fpregset_t nt_fpregset; 347 348 /* Prepare prstatus note */ 349 memset(&nt_prstatus, 0, sizeof(nt_prstatus)); 350 memcpy(&nt_prstatus.pr_reg.gprs, sa->gprs, sizeof(sa->gprs)); 351 memcpy(&nt_prstatus.pr_reg.psw, sa->psw, sizeof(sa->psw)); 352 memcpy(&nt_prstatus.pr_reg.acrs, sa->acrs, sizeof(sa->acrs)); 353 nt_prstatus.pr_pid = cpu; 354 /* Prepare fpregset (floating point) note */ 355 memset(&nt_fpregset, 0, sizeof(nt_fpregset)); 356 memcpy(&nt_fpregset.fpc, &sa->fpc, sizeof(sa->fpc)); 357 memcpy(&nt_fpregset.fprs, &sa->fprs, sizeof(sa->fprs)); 358 /* Create ELF notes for the CPU */ 359 ptr = nt_init(ptr, NT_PRSTATUS, &nt_prstatus, sizeof(nt_prstatus)); 360 ptr = nt_init(ptr, NT_PRFPREG, &nt_fpregset, sizeof(nt_fpregset)); 361 ptr = nt_init(ptr, NT_S390_TIMER, &sa->timer, sizeof(sa->timer)); 362 ptr = nt_init(ptr, NT_S390_TODCMP, &sa->todcmp, sizeof(sa->todcmp)); 363 ptr = nt_init(ptr, NT_S390_TODPREG, &sa->todpreg, sizeof(sa->todpreg)); 364 ptr = nt_init(ptr, NT_S390_CTRS, &sa->ctrs, sizeof(sa->ctrs)); 365 ptr = nt_init(ptr, NT_S390_PREFIX, &sa->prefix, sizeof(sa->prefix)); 366 if (MACHINE_HAS_VX) { 367 ptr = nt_init(ptr, NT_S390_VXRS_HIGH, 368 &sa->vxrs_high, sizeof(sa->vxrs_high)); 369 ptr = nt_init(ptr, NT_S390_VXRS_LOW, 370 &sa->vxrs_low, sizeof(sa->vxrs_low)); 371 } 372 return ptr; 373 } 374 375 /* 376 * Initialize prpsinfo note (new kernel) 377 */ 378 static void *nt_prpsinfo(void *ptr) 379 { 380 struct elf_prpsinfo prpsinfo; 381 382 memset(&prpsinfo, 0, sizeof(prpsinfo)); 383 prpsinfo.pr_sname = 'R'; 384 strcpy(prpsinfo.pr_fname, "vmlinux"); 385 return nt_init(ptr, NT_PRPSINFO, &prpsinfo, sizeof(prpsinfo)); 386 } 387 388 /* 389 * Get vmcoreinfo using lowcore->vmcore_info (new kernel) 390 */ 391 static void *get_vmcoreinfo_old(unsigned long *size) 392 { 393 char nt_name[11], *vmcoreinfo; 394 Elf64_Nhdr note; 395 void *addr; 396 397 if (copy_oldmem_kernel(&addr, &S390_lowcore.vmcore_info, sizeof(addr))) 398 return NULL; 399 memset(nt_name, 0, sizeof(nt_name)); 400 if (copy_oldmem_kernel(¬e, addr, sizeof(note))) 401 return NULL; 402 if (copy_oldmem_kernel(nt_name, addr + sizeof(note), 403 sizeof(nt_name) - 1)) 404 return NULL; 405 if (strcmp(nt_name, "VMCOREINFO") != 0) 406 return NULL; 407 vmcoreinfo = kzalloc_panic(note.n_descsz); 408 if (copy_oldmem_kernel(vmcoreinfo, addr + 24, note.n_descsz)) 409 return NULL; 410 *size = note.n_descsz; 411 return vmcoreinfo; 412 } 413 414 /* 415 * Initialize vmcoreinfo note (new kernel) 416 */ 417 static void *nt_vmcoreinfo(void *ptr) 418 { 419 unsigned long size; 420 void *vmcoreinfo; 421 422 vmcoreinfo = os_info_old_entry(OS_INFO_VMCOREINFO, &size); 423 if (!vmcoreinfo) 424 vmcoreinfo = get_vmcoreinfo_old(&size); 425 if (!vmcoreinfo) 426 return ptr; 427 return nt_init_name(ptr, 0, vmcoreinfo, size, "VMCOREINFO"); 428 } 429 430 /* 431 * Initialize ELF header (new kernel) 432 */ 433 static void *ehdr_init(Elf64_Ehdr *ehdr, int mem_chunk_cnt) 434 { 435 memset(ehdr, 0, sizeof(*ehdr)); 436 memcpy(ehdr->e_ident, ELFMAG, SELFMAG); 437 ehdr->e_ident[EI_CLASS] = ELFCLASS64; 438 ehdr->e_ident[EI_DATA] = ELFDATA2MSB; 439 ehdr->e_ident[EI_VERSION] = EV_CURRENT; 440 memset(ehdr->e_ident + EI_PAD, 0, EI_NIDENT - EI_PAD); 441 ehdr->e_type = ET_CORE; 442 ehdr->e_machine = EM_S390; 443 ehdr->e_version = EV_CURRENT; 444 ehdr->e_phoff = sizeof(Elf64_Ehdr); 445 ehdr->e_ehsize = sizeof(Elf64_Ehdr); 446 ehdr->e_phentsize = sizeof(Elf64_Phdr); 447 ehdr->e_phnum = mem_chunk_cnt + 1; 448 return ehdr + 1; 449 } 450 451 /* 452 * Return CPU count for ELF header (new kernel) 453 */ 454 static int get_cpu_cnt(void) 455 { 456 struct save_area *sa; 457 int cpus = 0; 458 459 list_for_each_entry(sa, &dump_save_areas, list) 460 if (sa->prefix != 0) 461 cpus++; 462 return cpus; 463 } 464 465 /* 466 * Return memory chunk count for ELF header (new kernel) 467 */ 468 static int get_mem_chunk_cnt(void) 469 { 470 int cnt = 0; 471 u64 idx; 472 473 for_each_mem_range(idx, &memblock.physmem, &oldmem_type, NUMA_NO_NODE, 474 MEMBLOCK_NONE, NULL, NULL, NULL) 475 cnt++; 476 return cnt; 477 } 478 479 /* 480 * Initialize ELF loads (new kernel) 481 */ 482 static void loads_init(Elf64_Phdr *phdr, u64 loads_offset) 483 { 484 phys_addr_t start, end; 485 u64 idx; 486 487 for_each_mem_range(idx, &memblock.physmem, &oldmem_type, NUMA_NO_NODE, 488 MEMBLOCK_NONE, &start, &end, NULL) { 489 phdr->p_filesz = end - start; 490 phdr->p_type = PT_LOAD; 491 phdr->p_offset = start; 492 phdr->p_vaddr = start; 493 phdr->p_paddr = start; 494 phdr->p_memsz = end - start; 495 phdr->p_flags = PF_R | PF_W | PF_X; 496 phdr->p_align = PAGE_SIZE; 497 phdr++; 498 } 499 } 500 501 /* 502 * Initialize notes (new kernel) 503 */ 504 static void *notes_init(Elf64_Phdr *phdr, void *ptr, u64 notes_offset) 505 { 506 struct save_area *sa; 507 void *ptr_start = ptr; 508 int cpu; 509 510 ptr = nt_prpsinfo(ptr); 511 512 cpu = 1; 513 list_for_each_entry(sa, &dump_save_areas, list) 514 if (sa->prefix != 0) 515 ptr = fill_cpu_elf_notes(ptr, cpu++, sa); 516 ptr = nt_vmcoreinfo(ptr); 517 memset(phdr, 0, sizeof(*phdr)); 518 phdr->p_type = PT_NOTE; 519 phdr->p_offset = notes_offset; 520 phdr->p_filesz = (unsigned long) PTR_SUB(ptr, ptr_start); 521 phdr->p_memsz = phdr->p_filesz; 522 return ptr; 523 } 524 525 /* 526 * Create ELF core header (new kernel) 527 */ 528 int elfcorehdr_alloc(unsigned long long *addr, unsigned long long *size) 529 { 530 Elf64_Phdr *phdr_notes, *phdr_loads; 531 int mem_chunk_cnt; 532 void *ptr, *hdr; 533 u32 alloc_size; 534 u64 hdr_off; 535 536 /* If we are not in kdump or zfcpdump mode return */ 537 if (!OLDMEM_BASE && ipl_info.type != IPL_TYPE_FCP_DUMP) 538 return 0; 539 /* If we cannot get HSA size for zfcpdump return error */ 540 if (ipl_info.type == IPL_TYPE_FCP_DUMP && !sclp.hsa_size) 541 return -ENODEV; 542 543 /* For kdump, exclude previous crashkernel memory */ 544 if (OLDMEM_BASE) { 545 oldmem_region.base = OLDMEM_BASE; 546 oldmem_region.size = OLDMEM_SIZE; 547 oldmem_type.total_size = OLDMEM_SIZE; 548 } 549 550 mem_chunk_cnt = get_mem_chunk_cnt(); 551 552 alloc_size = 0x1000 + get_cpu_cnt() * 0x4a0 + 553 mem_chunk_cnt * sizeof(Elf64_Phdr); 554 hdr = kzalloc_panic(alloc_size); 555 /* Init elf header */ 556 ptr = ehdr_init(hdr, mem_chunk_cnt); 557 /* Init program headers */ 558 phdr_notes = ptr; 559 ptr = PTR_ADD(ptr, sizeof(Elf64_Phdr)); 560 phdr_loads = ptr; 561 ptr = PTR_ADD(ptr, sizeof(Elf64_Phdr) * mem_chunk_cnt); 562 /* Init notes */ 563 hdr_off = PTR_DIFF(ptr, hdr); 564 ptr = notes_init(phdr_notes, ptr, ((unsigned long) hdr) + hdr_off); 565 /* Init loads */ 566 hdr_off = PTR_DIFF(ptr, hdr); 567 loads_init(phdr_loads, hdr_off); 568 *addr = (unsigned long long) hdr; 569 *size = (unsigned long long) hdr_off; 570 BUG_ON(elfcorehdr_size > alloc_size); 571 return 0; 572 } 573 574 /* 575 * Free ELF core header (new kernel) 576 */ 577 void elfcorehdr_free(unsigned long long addr) 578 { 579 kfree((void *)(unsigned long)addr); 580 } 581 582 /* 583 * Read from ELF header 584 */ 585 ssize_t elfcorehdr_read(char *buf, size_t count, u64 *ppos) 586 { 587 void *src = (void *)(unsigned long)*ppos; 588 589 memcpy(buf, src, count); 590 *ppos += count; 591 return count; 592 } 593 594 /* 595 * Read from ELF notes data 596 */ 597 ssize_t elfcorehdr_read_notes(char *buf, size_t count, u64 *ppos) 598 { 599 void *src = (void *)(unsigned long)*ppos; 600 601 memcpy(buf, src, count); 602 *ppos += count; 603 return count; 604 } 605