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