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 .name = "oldmem", 36 }; 37 38 struct save_area { 39 struct list_head list; 40 u64 psw[2]; 41 u64 ctrs[16]; 42 u64 gprs[16]; 43 u32 acrs[16]; 44 u64 fprs[16]; 45 u32 fpc; 46 u32 prefix; 47 u64 todpreg; 48 u64 timer; 49 u64 todcmp; 50 u64 vxrs_low[16]; 51 __vector128 vxrs_high[16]; 52 }; 53 54 static LIST_HEAD(dump_save_areas); 55 56 /* 57 * Allocate a save area 58 */ 59 struct save_area * __init save_area_alloc(bool is_boot_cpu) 60 { 61 struct save_area *sa; 62 63 sa = (void *) memblock_alloc(sizeof(*sa), 8); 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 return list_first_entry_or_null(&dump_save_areas, struct save_area, list); 77 } 78 79 /* 80 * Copy CPU registers into the save area 81 */ 82 void __init save_area_add_regs(struct save_area *sa, void *regs) 83 { 84 struct lowcore *lc; 85 86 lc = (struct lowcore *)(regs - __LC_FPREGS_SAVE_AREA); 87 memcpy(&sa->psw, &lc->psw_save_area, sizeof(sa->psw)); 88 memcpy(&sa->ctrs, &lc->cregs_save_area, sizeof(sa->ctrs)); 89 memcpy(&sa->gprs, &lc->gpregs_save_area, sizeof(sa->gprs)); 90 memcpy(&sa->acrs, &lc->access_regs_save_area, sizeof(sa->acrs)); 91 memcpy(&sa->fprs, &lc->floating_pt_save_area, sizeof(sa->fprs)); 92 memcpy(&sa->fpc, &lc->fpt_creg_save_area, sizeof(sa->fpc)); 93 memcpy(&sa->prefix, &lc->prefixreg_save_area, sizeof(sa->prefix)); 94 memcpy(&sa->todpreg, &lc->tod_progreg_save_area, sizeof(sa->todpreg)); 95 memcpy(&sa->timer, &lc->cpu_timer_save_area, sizeof(sa->timer)); 96 memcpy(&sa->todcmp, &lc->clock_comp_save_area, sizeof(sa->todcmp)); 97 } 98 99 /* 100 * Copy vector registers into the save area 101 */ 102 void __init save_area_add_vxrs(struct save_area *sa, __vector128 *vxrs) 103 { 104 int i; 105 106 /* Copy lower halves of vector registers 0-15 */ 107 for (i = 0; i < 16; i++) 108 memcpy(&sa->vxrs_low[i], &vxrs[i].u[2], 8); 109 /* Copy vector registers 16-31 */ 110 memcpy(sa->vxrs_high, vxrs + 16, 16 * sizeof(__vector128)); 111 } 112 113 /* 114 * Return physical address for virtual address 115 */ 116 static inline void *load_real_addr(void *addr) 117 { 118 unsigned long real_addr; 119 120 asm volatile( 121 " lra %0,0(%1)\n" 122 " jz 0f\n" 123 " la %0,0\n" 124 "0:" 125 : "=a" (real_addr) : "a" (addr) : "cc"); 126 return (void *)real_addr; 127 } 128 129 /* 130 * Copy memory of the old, dumped system to a kernel space virtual address 131 */ 132 int copy_oldmem_kernel(void *dst, void *src, size_t count) 133 { 134 unsigned long from, len; 135 void *ra; 136 int rc; 137 138 while (count) { 139 from = __pa(src); 140 if (!OLDMEM_BASE && from < sclp.hsa_size) { 141 /* Copy from zfcpdump HSA area */ 142 len = min(count, sclp.hsa_size - from); 143 rc = memcpy_hsa_kernel(dst, from, len); 144 if (rc) 145 return rc; 146 } else { 147 /* Check for swapped kdump oldmem areas */ 148 if (OLDMEM_BASE && from - OLDMEM_BASE < OLDMEM_SIZE) { 149 from -= OLDMEM_BASE; 150 len = min(count, OLDMEM_SIZE - from); 151 } else if (OLDMEM_BASE && from < OLDMEM_SIZE) { 152 len = min(count, OLDMEM_SIZE - from); 153 from += OLDMEM_BASE; 154 } else { 155 len = count; 156 } 157 if (is_vmalloc_or_module_addr(dst)) { 158 ra = load_real_addr(dst); 159 len = min(PAGE_SIZE - offset_in_page(ra), len); 160 } else { 161 ra = dst; 162 } 163 if (memcpy_real(ra, (void *) from, len)) 164 return -EFAULT; 165 } 166 dst += len; 167 src += len; 168 count -= len; 169 } 170 return 0; 171 } 172 173 /* 174 * Copy memory of the old, dumped system to a user space virtual address 175 */ 176 static int copy_oldmem_user(void __user *dst, void *src, size_t count) 177 { 178 unsigned long from, len; 179 int rc; 180 181 while (count) { 182 from = __pa(src); 183 if (!OLDMEM_BASE && from < sclp.hsa_size) { 184 /* Copy from zfcpdump HSA area */ 185 len = min(count, sclp.hsa_size - from); 186 rc = memcpy_hsa_user(dst, from, len); 187 if (rc) 188 return rc; 189 } else { 190 /* Check for swapped kdump oldmem areas */ 191 if (OLDMEM_BASE && from - OLDMEM_BASE < OLDMEM_SIZE) { 192 from -= OLDMEM_BASE; 193 len = min(count, OLDMEM_SIZE - from); 194 } else if (OLDMEM_BASE && from < OLDMEM_SIZE) { 195 len = min(count, OLDMEM_SIZE - from); 196 from += OLDMEM_BASE; 197 } else { 198 len = count; 199 } 200 rc = copy_to_user_real(dst, (void *) from, count); 201 if (rc) 202 return rc; 203 } 204 dst += len; 205 src += len; 206 count -= len; 207 } 208 return 0; 209 } 210 211 /* 212 * Copy one page from "oldmem" 213 */ 214 ssize_t copy_oldmem_page(unsigned long pfn, char *buf, size_t csize, 215 unsigned long offset, int userbuf) 216 { 217 void *src; 218 int rc; 219 220 if (!csize) 221 return 0; 222 src = (void *) (pfn << PAGE_SHIFT) + offset; 223 if (userbuf) 224 rc = copy_oldmem_user((void __force __user *) buf, src, csize); 225 else 226 rc = copy_oldmem_kernel((void *) buf, src, csize); 227 return rc; 228 } 229 230 /* 231 * Remap "oldmem" for kdump 232 * 233 * For the kdump reserved memory this functions performs a swap operation: 234 * [0 - OLDMEM_SIZE] is mapped to [OLDMEM_BASE - OLDMEM_BASE + OLDMEM_SIZE] 235 */ 236 static int remap_oldmem_pfn_range_kdump(struct vm_area_struct *vma, 237 unsigned long from, unsigned long pfn, 238 unsigned long size, pgprot_t prot) 239 { 240 unsigned long size_old; 241 int rc; 242 243 if (pfn < OLDMEM_SIZE >> PAGE_SHIFT) { 244 size_old = min(size, OLDMEM_SIZE - (pfn << PAGE_SHIFT)); 245 rc = remap_pfn_range(vma, from, 246 pfn + (OLDMEM_BASE >> PAGE_SHIFT), 247 size_old, prot); 248 if (rc || size == size_old) 249 return rc; 250 size -= size_old; 251 from += size_old; 252 pfn += size_old >> PAGE_SHIFT; 253 } 254 return remap_pfn_range(vma, from, pfn, size, prot); 255 } 256 257 /* 258 * Remap "oldmem" for zfcpdump 259 * 260 * We only map available memory above HSA size. Memory below HSA size 261 * is read on demand using the copy_oldmem_page() function. 262 */ 263 static int remap_oldmem_pfn_range_zfcpdump(struct vm_area_struct *vma, 264 unsigned long from, 265 unsigned long pfn, 266 unsigned long size, pgprot_t prot) 267 { 268 unsigned long hsa_end = sclp.hsa_size; 269 unsigned long size_hsa; 270 271 if (pfn < hsa_end >> PAGE_SHIFT) { 272 size_hsa = min(size, hsa_end - (pfn << PAGE_SHIFT)); 273 if (size == size_hsa) 274 return 0; 275 size -= size_hsa; 276 from += size_hsa; 277 pfn += size_hsa >> PAGE_SHIFT; 278 } 279 return remap_pfn_range(vma, from, pfn, size, prot); 280 } 281 282 /* 283 * Remap "oldmem" for kdump or zfcpdump 284 */ 285 int remap_oldmem_pfn_range(struct vm_area_struct *vma, unsigned long from, 286 unsigned long pfn, unsigned long size, pgprot_t prot) 287 { 288 if (OLDMEM_BASE) 289 return remap_oldmem_pfn_range_kdump(vma, from, pfn, size, prot); 290 else 291 return remap_oldmem_pfn_range_zfcpdump(vma, from, pfn, size, 292 prot); 293 } 294 295 /* 296 * Alloc memory and panic in case of ENOMEM 297 */ 298 static void *kzalloc_panic(int len) 299 { 300 void *rc; 301 302 rc = kzalloc(len, GFP_KERNEL); 303 if (!rc) 304 panic("s390 kdump kzalloc (%d) failed", len); 305 return rc; 306 } 307 308 /* 309 * Initialize ELF note 310 */ 311 static void *nt_init_name(void *buf, Elf64_Word type, void *desc, int d_len, 312 const char *name) 313 { 314 Elf64_Nhdr *note; 315 u64 len; 316 317 note = (Elf64_Nhdr *)buf; 318 note->n_namesz = strlen(name) + 1; 319 note->n_descsz = d_len; 320 note->n_type = type; 321 len = sizeof(Elf64_Nhdr); 322 323 memcpy(buf + len, name, note->n_namesz); 324 len = roundup(len + note->n_namesz, 4); 325 326 memcpy(buf + len, desc, note->n_descsz); 327 len = roundup(len + note->n_descsz, 4); 328 329 return PTR_ADD(buf, len); 330 } 331 332 static inline void *nt_init(void *buf, Elf64_Word type, void *desc, int d_len) 333 { 334 const char *note_name = "LINUX"; 335 336 if (type == NT_PRPSINFO || type == NT_PRSTATUS || type == NT_PRFPREG) 337 note_name = KEXEC_CORE_NOTE_NAME; 338 return nt_init_name(buf, type, desc, d_len, note_name); 339 } 340 341 /* 342 * Fill ELF notes for one CPU with save area registers 343 */ 344 static void *fill_cpu_elf_notes(void *ptr, int cpu, struct save_area *sa) 345 { 346 struct elf_prstatus nt_prstatus; 347 elf_fpregset_t nt_fpregset; 348 349 /* Prepare prstatus note */ 350 memset(&nt_prstatus, 0, sizeof(nt_prstatus)); 351 memcpy(&nt_prstatus.pr_reg.gprs, sa->gprs, sizeof(sa->gprs)); 352 memcpy(&nt_prstatus.pr_reg.psw, sa->psw, sizeof(sa->psw)); 353 memcpy(&nt_prstatus.pr_reg.acrs, sa->acrs, sizeof(sa->acrs)); 354 nt_prstatus.pr_pid = cpu; 355 /* Prepare fpregset (floating point) note */ 356 memset(&nt_fpregset, 0, sizeof(nt_fpregset)); 357 memcpy(&nt_fpregset.fpc, &sa->fpc, sizeof(sa->fpc)); 358 memcpy(&nt_fpregset.fprs, &sa->fprs, sizeof(sa->fprs)); 359 /* Create ELF notes for the CPU */ 360 ptr = nt_init(ptr, NT_PRSTATUS, &nt_prstatus, sizeof(nt_prstatus)); 361 ptr = nt_init(ptr, NT_PRFPREG, &nt_fpregset, sizeof(nt_fpregset)); 362 ptr = nt_init(ptr, NT_S390_TIMER, &sa->timer, sizeof(sa->timer)); 363 ptr = nt_init(ptr, NT_S390_TODCMP, &sa->todcmp, sizeof(sa->todcmp)); 364 ptr = nt_init(ptr, NT_S390_TODPREG, &sa->todpreg, sizeof(sa->todpreg)); 365 ptr = nt_init(ptr, NT_S390_CTRS, &sa->ctrs, sizeof(sa->ctrs)); 366 ptr = nt_init(ptr, NT_S390_PREFIX, &sa->prefix, sizeof(sa->prefix)); 367 if (MACHINE_HAS_VX) { 368 ptr = nt_init(ptr, NT_S390_VXRS_HIGH, 369 &sa->vxrs_high, sizeof(sa->vxrs_high)); 370 ptr = nt_init(ptr, NT_S390_VXRS_LOW, 371 &sa->vxrs_low, sizeof(sa->vxrs_low)); 372 } 373 return ptr; 374 } 375 376 /* 377 * Initialize prpsinfo note (new kernel) 378 */ 379 static void *nt_prpsinfo(void *ptr) 380 { 381 struct elf_prpsinfo prpsinfo; 382 383 memset(&prpsinfo, 0, sizeof(prpsinfo)); 384 prpsinfo.pr_sname = 'R'; 385 strcpy(prpsinfo.pr_fname, "vmlinux"); 386 return nt_init(ptr, NT_PRPSINFO, &prpsinfo, sizeof(prpsinfo)); 387 } 388 389 /* 390 * Get vmcoreinfo using lowcore->vmcore_info (new kernel) 391 */ 392 static void *get_vmcoreinfo_old(unsigned long *size) 393 { 394 char nt_name[11], *vmcoreinfo; 395 Elf64_Nhdr note; 396 void *addr; 397 398 if (copy_oldmem_kernel(&addr, &S390_lowcore.vmcore_info, sizeof(addr))) 399 return NULL; 400 memset(nt_name, 0, sizeof(nt_name)); 401 if (copy_oldmem_kernel(¬e, addr, sizeof(note))) 402 return NULL; 403 if (copy_oldmem_kernel(nt_name, addr + sizeof(note), 404 sizeof(nt_name) - 1)) 405 return NULL; 406 if (strcmp(nt_name, "VMCOREINFO") != 0) 407 return NULL; 408 vmcoreinfo = kzalloc_panic(note.n_descsz); 409 if (copy_oldmem_kernel(vmcoreinfo, addr + 24, note.n_descsz)) 410 return NULL; 411 *size = note.n_descsz; 412 return vmcoreinfo; 413 } 414 415 /* 416 * Initialize vmcoreinfo note (new kernel) 417 */ 418 static void *nt_vmcoreinfo(void *ptr) 419 { 420 unsigned long size; 421 void *vmcoreinfo; 422 423 vmcoreinfo = os_info_old_entry(OS_INFO_VMCOREINFO, &size); 424 if (!vmcoreinfo) 425 vmcoreinfo = get_vmcoreinfo_old(&size); 426 if (!vmcoreinfo) 427 return ptr; 428 return nt_init_name(ptr, 0, vmcoreinfo, size, "VMCOREINFO"); 429 } 430 431 /* 432 * Initialize final note (needed for /proc/vmcore code) 433 */ 434 static void *nt_final(void *ptr) 435 { 436 Elf64_Nhdr *note; 437 438 note = (Elf64_Nhdr *) ptr; 439 note->n_namesz = 0; 440 note->n_descsz = 0; 441 note->n_type = 0; 442 return PTR_ADD(ptr, sizeof(Elf64_Nhdr)); 443 } 444 445 /* 446 * Initialize ELF header (new kernel) 447 */ 448 static void *ehdr_init(Elf64_Ehdr *ehdr, int mem_chunk_cnt) 449 { 450 memset(ehdr, 0, sizeof(*ehdr)); 451 memcpy(ehdr->e_ident, ELFMAG, SELFMAG); 452 ehdr->e_ident[EI_CLASS] = ELFCLASS64; 453 ehdr->e_ident[EI_DATA] = ELFDATA2MSB; 454 ehdr->e_ident[EI_VERSION] = EV_CURRENT; 455 memset(ehdr->e_ident + EI_PAD, 0, EI_NIDENT - EI_PAD); 456 ehdr->e_type = ET_CORE; 457 ehdr->e_machine = EM_S390; 458 ehdr->e_version = EV_CURRENT; 459 ehdr->e_phoff = sizeof(Elf64_Ehdr); 460 ehdr->e_ehsize = sizeof(Elf64_Ehdr); 461 ehdr->e_phentsize = sizeof(Elf64_Phdr); 462 ehdr->e_phnum = mem_chunk_cnt + 1; 463 return ehdr + 1; 464 } 465 466 /* 467 * Return CPU count for ELF header (new kernel) 468 */ 469 static int get_cpu_cnt(void) 470 { 471 struct save_area *sa; 472 int cpus = 0; 473 474 list_for_each_entry(sa, &dump_save_areas, list) 475 if (sa->prefix != 0) 476 cpus++; 477 return cpus; 478 } 479 480 /* 481 * Return memory chunk count for ELF header (new kernel) 482 */ 483 static int get_mem_chunk_cnt(void) 484 { 485 int cnt = 0; 486 u64 idx; 487 488 for_each_mem_range(idx, &memblock.physmem, &oldmem_type, NUMA_NO_NODE, 489 MEMBLOCK_NONE, NULL, NULL, NULL) 490 cnt++; 491 return cnt; 492 } 493 494 /* 495 * Initialize ELF loads (new kernel) 496 */ 497 static void loads_init(Elf64_Phdr *phdr, u64 loads_offset) 498 { 499 phys_addr_t start, end; 500 u64 idx; 501 502 for_each_mem_range(idx, &memblock.physmem, &oldmem_type, NUMA_NO_NODE, 503 MEMBLOCK_NONE, &start, &end, NULL) { 504 phdr->p_filesz = end - start; 505 phdr->p_type = PT_LOAD; 506 phdr->p_offset = start; 507 phdr->p_vaddr = start; 508 phdr->p_paddr = start; 509 phdr->p_memsz = end - start; 510 phdr->p_flags = PF_R | PF_W | PF_X; 511 phdr->p_align = PAGE_SIZE; 512 phdr++; 513 } 514 } 515 516 /* 517 * Initialize notes (new kernel) 518 */ 519 static void *notes_init(Elf64_Phdr *phdr, void *ptr, u64 notes_offset) 520 { 521 struct save_area *sa; 522 void *ptr_start = ptr; 523 int cpu; 524 525 ptr = nt_prpsinfo(ptr); 526 527 cpu = 1; 528 list_for_each_entry(sa, &dump_save_areas, list) 529 if (sa->prefix != 0) 530 ptr = fill_cpu_elf_notes(ptr, cpu++, sa); 531 ptr = nt_vmcoreinfo(ptr); 532 ptr = nt_final(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 we cannot get HSA size for zfcpdump return error */ 556 if (ipl_info.type == IPL_TYPE_FCP_DUMP && !sclp.hsa_size) 557 return -ENODEV; 558 559 /* For kdump, exclude previous crashkernel memory */ 560 if (OLDMEM_BASE) { 561 oldmem_region.base = OLDMEM_BASE; 562 oldmem_region.size = OLDMEM_SIZE; 563 oldmem_type.total_size = OLDMEM_SIZE; 564 } 565 566 mem_chunk_cnt = get_mem_chunk_cnt(); 567 568 alloc_size = 0x1000 + get_cpu_cnt() * 0x4a0 + 569 mem_chunk_cnt * sizeof(Elf64_Phdr); 570 hdr = kzalloc_panic(alloc_size); 571 /* Init elf header */ 572 ptr = ehdr_init(hdr, mem_chunk_cnt); 573 /* Init program headers */ 574 phdr_notes = ptr; 575 ptr = PTR_ADD(ptr, sizeof(Elf64_Phdr)); 576 phdr_loads = ptr; 577 ptr = PTR_ADD(ptr, sizeof(Elf64_Phdr) * mem_chunk_cnt); 578 /* Init notes */ 579 hdr_off = PTR_DIFF(ptr, hdr); 580 ptr = notes_init(phdr_notes, ptr, ((unsigned long) hdr) + hdr_off); 581 /* Init loads */ 582 hdr_off = PTR_DIFF(ptr, hdr); 583 loads_init(phdr_loads, hdr_off); 584 *addr = (unsigned long long) hdr; 585 *size = (unsigned long long) hdr_off; 586 BUG_ON(elfcorehdr_size > alloc_size); 587 return 0; 588 } 589 590 /* 591 * Free ELF core header (new kernel) 592 */ 593 void elfcorehdr_free(unsigned long long addr) 594 { 595 kfree((void *)(unsigned long)addr); 596 } 597 598 /* 599 * Read from ELF header 600 */ 601 ssize_t elfcorehdr_read(char *buf, size_t count, u64 *ppos) 602 { 603 void *src = (void *)(unsigned long)*ppos; 604 605 memcpy(buf, src, count); 606 *ppos += count; 607 return count; 608 } 609 610 /* 611 * Read from ELF notes data 612 */ 613 ssize_t elfcorehdr_read_notes(char *buf, size_t count, u64 *ppos) 614 { 615 void *src = (void *)(unsigned long)*ppos; 616 617 memcpy(buf, src, count); 618 *ppos += count; 619 return count; 620 } 621