1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * S390 kdump implementation 4 * 5 * Copyright IBM Corp. 2011 6 * Author(s): Michael Holzheu <holzheu@linux.vnet.ibm.com> 7 */ 8 9 #include <linux/crash_dump.h> 10 #include <asm/lowcore.h> 11 #include <linux/kernel.h> 12 #include <linux/init.h> 13 #include <linux/mm.h> 14 #include <linux/gfp.h> 15 #include <linux/slab.h> 16 #include <linux/bootmem.h> 17 #include <linux/elf.h> 18 #include <asm/asm-offsets.h> 19 #include <linux/memblock.h> 20 #include <asm/os_info.h> 21 #include <asm/elf.h> 22 #include <asm/ipl.h> 23 #include <asm/sclp.h> 24 25 #define PTR_ADD(x, y) (((char *) (x)) + ((unsigned long) (y))) 26 #define PTR_SUB(x, y) (((char *) (x)) - ((unsigned long) (y))) 27 #define PTR_DIFF(x, y) ((unsigned long)(((char *) (x)) - ((unsigned long) (y)))) 28 29 static struct memblock_region oldmem_region; 30 31 static struct memblock_type oldmem_type = { 32 .cnt = 1, 33 .max = 1, 34 .total_size = 0, 35 .regions = &oldmem_region, 36 .name = "oldmem", 37 }; 38 39 struct save_area { 40 struct list_head list; 41 u64 psw[2]; 42 u64 ctrs[16]; 43 u64 gprs[16]; 44 u32 acrs[16]; 45 u64 fprs[16]; 46 u32 fpc; 47 u32 prefix; 48 u64 todpreg; 49 u64 timer; 50 u64 todcmp; 51 u64 vxrs_low[16]; 52 __vector128 vxrs_high[16]; 53 }; 54 55 static LIST_HEAD(dump_save_areas); 56 57 /* 58 * Allocate a save area 59 */ 60 struct save_area * __init save_area_alloc(bool is_boot_cpu) 61 { 62 struct save_area *sa; 63 64 sa = (void *) memblock_alloc(sizeof(*sa), 8); 65 if (is_boot_cpu) 66 list_add(&sa->list, &dump_save_areas); 67 else 68 list_add_tail(&sa->list, &dump_save_areas); 69 return sa; 70 } 71 72 /* 73 * Return the address of the save area for the boot CPU 74 */ 75 struct save_area * __init save_area_boot_cpu(void) 76 { 77 return list_first_entry_or_null(&dump_save_areas, struct save_area, list); 78 } 79 80 /* 81 * Copy CPU registers into the save area 82 */ 83 void __init save_area_add_regs(struct save_area *sa, void *regs) 84 { 85 struct lowcore *lc; 86 87 lc = (struct lowcore *)(regs - __LC_FPREGS_SAVE_AREA); 88 memcpy(&sa->psw, &lc->psw_save_area, sizeof(sa->psw)); 89 memcpy(&sa->ctrs, &lc->cregs_save_area, sizeof(sa->ctrs)); 90 memcpy(&sa->gprs, &lc->gpregs_save_area, sizeof(sa->gprs)); 91 memcpy(&sa->acrs, &lc->access_regs_save_area, sizeof(sa->acrs)); 92 memcpy(&sa->fprs, &lc->floating_pt_save_area, sizeof(sa->fprs)); 93 memcpy(&sa->fpc, &lc->fpt_creg_save_area, sizeof(sa->fpc)); 94 memcpy(&sa->prefix, &lc->prefixreg_save_area, sizeof(sa->prefix)); 95 memcpy(&sa->todpreg, &lc->tod_progreg_save_area, sizeof(sa->todpreg)); 96 memcpy(&sa->timer, &lc->cpu_timer_save_area, sizeof(sa->timer)); 97 memcpy(&sa->todcmp, &lc->clock_comp_save_area, sizeof(sa->todcmp)); 98 } 99 100 /* 101 * Copy vector registers into the save area 102 */ 103 void __init save_area_add_vxrs(struct save_area *sa, __vector128 *vxrs) 104 { 105 int i; 106 107 /* Copy lower halves of vector registers 0-15 */ 108 for (i = 0; i < 16; i++) 109 memcpy(&sa->vxrs_low[i], &vxrs[i].u[2], 8); 110 /* Copy vector registers 16-31 */ 111 memcpy(sa->vxrs_high, vxrs + 16, 16 * sizeof(__vector128)); 112 } 113 114 /* 115 * Return physical address for virtual address 116 */ 117 static inline void *load_real_addr(void *addr) 118 { 119 unsigned long real_addr; 120 121 asm volatile( 122 " lra %0,0(%1)\n" 123 " jz 0f\n" 124 " la %0,0\n" 125 "0:" 126 : "=a" (real_addr) : "a" (addr) : "cc"); 127 return (void *)real_addr; 128 } 129 130 /* 131 * Copy memory of the old, dumped system to a kernel space virtual address 132 */ 133 int copy_oldmem_kernel(void *dst, void *src, size_t count) 134 { 135 unsigned long from, len; 136 void *ra; 137 int rc; 138 139 while (count) { 140 from = __pa(src); 141 if (!OLDMEM_BASE && from < sclp.hsa_size) { 142 /* Copy from zfcpdump HSA area */ 143 len = min(count, sclp.hsa_size - from); 144 rc = memcpy_hsa_kernel(dst, from, len); 145 if (rc) 146 return rc; 147 } else { 148 /* Check for swapped kdump oldmem areas */ 149 if (OLDMEM_BASE && from - OLDMEM_BASE < OLDMEM_SIZE) { 150 from -= OLDMEM_BASE; 151 len = min(count, OLDMEM_SIZE - from); 152 } else if (OLDMEM_BASE && from < OLDMEM_SIZE) { 153 len = min(count, OLDMEM_SIZE - from); 154 from += OLDMEM_BASE; 155 } else { 156 len = count; 157 } 158 if (is_vmalloc_or_module_addr(dst)) { 159 ra = load_real_addr(dst); 160 len = min(PAGE_SIZE - offset_in_page(ra), len); 161 } else { 162 ra = dst; 163 } 164 if (memcpy_real(ra, (void *) from, len)) 165 return -EFAULT; 166 } 167 dst += len; 168 src += len; 169 count -= len; 170 } 171 return 0; 172 } 173 174 /* 175 * Copy memory of the old, dumped system to a user space virtual address 176 */ 177 static int copy_oldmem_user(void __user *dst, void *src, size_t count) 178 { 179 unsigned long from, len; 180 int rc; 181 182 while (count) { 183 from = __pa(src); 184 if (!OLDMEM_BASE && from < sclp.hsa_size) { 185 /* Copy from zfcpdump HSA area */ 186 len = min(count, sclp.hsa_size - from); 187 rc = memcpy_hsa_user(dst, from, len); 188 if (rc) 189 return rc; 190 } else { 191 /* Check for swapped kdump oldmem areas */ 192 if (OLDMEM_BASE && from - OLDMEM_BASE < OLDMEM_SIZE) { 193 from -= OLDMEM_BASE; 194 len = min(count, OLDMEM_SIZE - from); 195 } else if (OLDMEM_BASE && from < OLDMEM_SIZE) { 196 len = min(count, OLDMEM_SIZE - from); 197 from += OLDMEM_BASE; 198 } else { 199 len = count; 200 } 201 rc = copy_to_user_real(dst, (void *) from, count); 202 if (rc) 203 return rc; 204 } 205 dst += len; 206 src += len; 207 count -= len; 208 } 209 return 0; 210 } 211 212 /* 213 * Copy one page from "oldmem" 214 */ 215 ssize_t copy_oldmem_page(unsigned long pfn, char *buf, size_t csize, 216 unsigned long offset, int userbuf) 217 { 218 void *src; 219 int rc; 220 221 if (!csize) 222 return 0; 223 src = (void *) (pfn << PAGE_SHIFT) + offset; 224 if (userbuf) 225 rc = copy_oldmem_user((void __force __user *) buf, src, csize); 226 else 227 rc = copy_oldmem_kernel((void *) buf, src, csize); 228 return rc; 229 } 230 231 /* 232 * Remap "oldmem" for kdump 233 * 234 * For the kdump reserved memory this functions performs a swap operation: 235 * [0 - OLDMEM_SIZE] is mapped to [OLDMEM_BASE - OLDMEM_BASE + OLDMEM_SIZE] 236 */ 237 static int remap_oldmem_pfn_range_kdump(struct vm_area_struct *vma, 238 unsigned long from, unsigned long pfn, 239 unsigned long size, pgprot_t prot) 240 { 241 unsigned long size_old; 242 int rc; 243 244 if (pfn < OLDMEM_SIZE >> PAGE_SHIFT) { 245 size_old = min(size, OLDMEM_SIZE - (pfn << PAGE_SHIFT)); 246 rc = remap_pfn_range(vma, from, 247 pfn + (OLDMEM_BASE >> PAGE_SHIFT), 248 size_old, prot); 249 if (rc || size == size_old) 250 return rc; 251 size -= size_old; 252 from += size_old; 253 pfn += size_old >> PAGE_SHIFT; 254 } 255 return remap_pfn_range(vma, from, pfn, size, prot); 256 } 257 258 /* 259 * Remap "oldmem" for zfcpdump 260 * 261 * We only map available memory above HSA size. Memory below HSA size 262 * is read on demand using the copy_oldmem_page() function. 263 */ 264 static int remap_oldmem_pfn_range_zfcpdump(struct vm_area_struct *vma, 265 unsigned long from, 266 unsigned long pfn, 267 unsigned long size, pgprot_t prot) 268 { 269 unsigned long hsa_end = sclp.hsa_size; 270 unsigned long size_hsa; 271 272 if (pfn < hsa_end >> PAGE_SHIFT) { 273 size_hsa = min(size, hsa_end - (pfn << PAGE_SHIFT)); 274 if (size == size_hsa) 275 return 0; 276 size -= size_hsa; 277 from += size_hsa; 278 pfn += size_hsa >> PAGE_SHIFT; 279 } 280 return remap_pfn_range(vma, from, pfn, size, prot); 281 } 282 283 /* 284 * Remap "oldmem" for kdump or zfcpdump 285 */ 286 int remap_oldmem_pfn_range(struct vm_area_struct *vma, unsigned long from, 287 unsigned long pfn, unsigned long size, pgprot_t prot) 288 { 289 if (OLDMEM_BASE) 290 return remap_oldmem_pfn_range_kdump(vma, from, pfn, size, prot); 291 else 292 return remap_oldmem_pfn_range_zfcpdump(vma, from, pfn, size, 293 prot); 294 } 295 296 /* 297 * Alloc memory and panic in case of ENOMEM 298 */ 299 static void *kzalloc_panic(int len) 300 { 301 void *rc; 302 303 rc = kzalloc(len, GFP_KERNEL); 304 if (!rc) 305 panic("s390 kdump kzalloc (%d) failed", len); 306 return rc; 307 } 308 309 /* 310 * Initialize ELF note 311 */ 312 static void *nt_init_name(void *buf, Elf64_Word type, void *desc, int d_len, 313 const char *name) 314 { 315 Elf64_Nhdr *note; 316 u64 len; 317 318 note = (Elf64_Nhdr *)buf; 319 note->n_namesz = strlen(name) + 1; 320 note->n_descsz = d_len; 321 note->n_type = type; 322 len = sizeof(Elf64_Nhdr); 323 324 memcpy(buf + len, name, note->n_namesz); 325 len = roundup(len + note->n_namesz, 4); 326 327 memcpy(buf + len, desc, note->n_descsz); 328 len = roundup(len + note->n_descsz, 4); 329 330 return PTR_ADD(buf, len); 331 } 332 333 static inline void *nt_init(void *buf, Elf64_Word type, void *desc, int d_len) 334 { 335 const char *note_name = "LINUX"; 336 337 if (type == NT_PRPSINFO || type == NT_PRSTATUS || type == NT_PRFPREG) 338 note_name = KEXEC_CORE_NOTE_NAME; 339 return nt_init_name(buf, type, desc, d_len, note_name); 340 } 341 342 /* 343 * Fill ELF notes for one CPU with save area registers 344 */ 345 static void *fill_cpu_elf_notes(void *ptr, int cpu, struct save_area *sa) 346 { 347 struct elf_prstatus nt_prstatus; 348 elf_fpregset_t nt_fpregset; 349 350 /* Prepare prstatus note */ 351 memset(&nt_prstatus, 0, sizeof(nt_prstatus)); 352 memcpy(&nt_prstatus.pr_reg.gprs, sa->gprs, sizeof(sa->gprs)); 353 memcpy(&nt_prstatus.pr_reg.psw, sa->psw, sizeof(sa->psw)); 354 memcpy(&nt_prstatus.pr_reg.acrs, sa->acrs, sizeof(sa->acrs)); 355 nt_prstatus.pr_pid = cpu; 356 /* Prepare fpregset (floating point) note */ 357 memset(&nt_fpregset, 0, sizeof(nt_fpregset)); 358 memcpy(&nt_fpregset.fpc, &sa->fpc, sizeof(sa->fpc)); 359 memcpy(&nt_fpregset.fprs, &sa->fprs, sizeof(sa->fprs)); 360 /* Create ELF notes for the CPU */ 361 ptr = nt_init(ptr, NT_PRSTATUS, &nt_prstatus, sizeof(nt_prstatus)); 362 ptr = nt_init(ptr, NT_PRFPREG, &nt_fpregset, sizeof(nt_fpregset)); 363 ptr = nt_init(ptr, NT_S390_TIMER, &sa->timer, sizeof(sa->timer)); 364 ptr = nt_init(ptr, NT_S390_TODCMP, &sa->todcmp, sizeof(sa->todcmp)); 365 ptr = nt_init(ptr, NT_S390_TODPREG, &sa->todpreg, sizeof(sa->todpreg)); 366 ptr = nt_init(ptr, NT_S390_CTRS, &sa->ctrs, sizeof(sa->ctrs)); 367 ptr = nt_init(ptr, NT_S390_PREFIX, &sa->prefix, sizeof(sa->prefix)); 368 if (MACHINE_HAS_VX) { 369 ptr = nt_init(ptr, NT_S390_VXRS_HIGH, 370 &sa->vxrs_high, sizeof(sa->vxrs_high)); 371 ptr = nt_init(ptr, NT_S390_VXRS_LOW, 372 &sa->vxrs_low, sizeof(sa->vxrs_low)); 373 } 374 return ptr; 375 } 376 377 /* 378 * Initialize prpsinfo note (new kernel) 379 */ 380 static void *nt_prpsinfo(void *ptr) 381 { 382 struct elf_prpsinfo prpsinfo; 383 384 memset(&prpsinfo, 0, sizeof(prpsinfo)); 385 prpsinfo.pr_sname = 'R'; 386 strcpy(prpsinfo.pr_fname, "vmlinux"); 387 return nt_init(ptr, NT_PRPSINFO, &prpsinfo, sizeof(prpsinfo)); 388 } 389 390 /* 391 * Get vmcoreinfo using lowcore->vmcore_info (new kernel) 392 */ 393 static void *get_vmcoreinfo_old(unsigned long *size) 394 { 395 char nt_name[11], *vmcoreinfo; 396 Elf64_Nhdr note; 397 void *addr; 398 399 if (copy_oldmem_kernel(&addr, &S390_lowcore.vmcore_info, sizeof(addr))) 400 return NULL; 401 memset(nt_name, 0, sizeof(nt_name)); 402 if (copy_oldmem_kernel(¬e, addr, sizeof(note))) 403 return NULL; 404 if (copy_oldmem_kernel(nt_name, addr + sizeof(note), 405 sizeof(nt_name) - 1)) 406 return NULL; 407 if (strcmp(nt_name, "VMCOREINFO") != 0) 408 return NULL; 409 vmcoreinfo = kzalloc_panic(note.n_descsz); 410 if (copy_oldmem_kernel(vmcoreinfo, addr + 24, note.n_descsz)) 411 return NULL; 412 *size = note.n_descsz; 413 return vmcoreinfo; 414 } 415 416 /* 417 * Initialize vmcoreinfo note (new kernel) 418 */ 419 static void *nt_vmcoreinfo(void *ptr) 420 { 421 unsigned long size; 422 void *vmcoreinfo; 423 424 vmcoreinfo = os_info_old_entry(OS_INFO_VMCOREINFO, &size); 425 if (!vmcoreinfo) 426 vmcoreinfo = get_vmcoreinfo_old(&size); 427 if (!vmcoreinfo) 428 return ptr; 429 return nt_init_name(ptr, 0, vmcoreinfo, size, "VMCOREINFO"); 430 } 431 432 /* 433 * Initialize final note (needed for /proc/vmcore code) 434 */ 435 static void *nt_final(void *ptr) 436 { 437 Elf64_Nhdr *note; 438 439 note = (Elf64_Nhdr *) ptr; 440 note->n_namesz = 0; 441 note->n_descsz = 0; 442 note->n_type = 0; 443 return PTR_ADD(ptr, sizeof(Elf64_Nhdr)); 444 } 445 446 /* 447 * Initialize ELF header (new kernel) 448 */ 449 static void *ehdr_init(Elf64_Ehdr *ehdr, int mem_chunk_cnt) 450 { 451 memset(ehdr, 0, sizeof(*ehdr)); 452 memcpy(ehdr->e_ident, ELFMAG, SELFMAG); 453 ehdr->e_ident[EI_CLASS] = ELFCLASS64; 454 ehdr->e_ident[EI_DATA] = ELFDATA2MSB; 455 ehdr->e_ident[EI_VERSION] = EV_CURRENT; 456 memset(ehdr->e_ident + EI_PAD, 0, EI_NIDENT - EI_PAD); 457 ehdr->e_type = ET_CORE; 458 ehdr->e_machine = EM_S390; 459 ehdr->e_version = EV_CURRENT; 460 ehdr->e_phoff = sizeof(Elf64_Ehdr); 461 ehdr->e_ehsize = sizeof(Elf64_Ehdr); 462 ehdr->e_phentsize = sizeof(Elf64_Phdr); 463 ehdr->e_phnum = mem_chunk_cnt + 1; 464 return ehdr + 1; 465 } 466 467 /* 468 * Return CPU count for ELF header (new kernel) 469 */ 470 static int get_cpu_cnt(void) 471 { 472 struct save_area *sa; 473 int cpus = 0; 474 475 list_for_each_entry(sa, &dump_save_areas, list) 476 if (sa->prefix != 0) 477 cpus++; 478 return cpus; 479 } 480 481 /* 482 * Return memory chunk count for ELF header (new kernel) 483 */ 484 static int get_mem_chunk_cnt(void) 485 { 486 int cnt = 0; 487 u64 idx; 488 489 for_each_mem_range(idx, &memblock.physmem, &oldmem_type, NUMA_NO_NODE, 490 MEMBLOCK_NONE, NULL, NULL, NULL) 491 cnt++; 492 return cnt; 493 } 494 495 /* 496 * Initialize ELF loads (new kernel) 497 */ 498 static void loads_init(Elf64_Phdr *phdr, u64 loads_offset) 499 { 500 phys_addr_t start, end; 501 u64 idx; 502 503 for_each_mem_range(idx, &memblock.physmem, &oldmem_type, NUMA_NO_NODE, 504 MEMBLOCK_NONE, &start, &end, NULL) { 505 phdr->p_filesz = end - start; 506 phdr->p_type = PT_LOAD; 507 phdr->p_offset = start; 508 phdr->p_vaddr = start; 509 phdr->p_paddr = start; 510 phdr->p_memsz = end - start; 511 phdr->p_flags = PF_R | PF_W | PF_X; 512 phdr->p_align = PAGE_SIZE; 513 phdr++; 514 } 515 } 516 517 /* 518 * Initialize notes (new kernel) 519 */ 520 static void *notes_init(Elf64_Phdr *phdr, void *ptr, u64 notes_offset) 521 { 522 struct save_area *sa; 523 void *ptr_start = ptr; 524 int cpu; 525 526 ptr = nt_prpsinfo(ptr); 527 528 cpu = 1; 529 list_for_each_entry(sa, &dump_save_areas, list) 530 if (sa->prefix != 0) 531 ptr = fill_cpu_elf_notes(ptr, cpu++, sa); 532 ptr = nt_vmcoreinfo(ptr); 533 ptr = nt_final(ptr); 534 memset(phdr, 0, sizeof(*phdr)); 535 phdr->p_type = PT_NOTE; 536 phdr->p_offset = notes_offset; 537 phdr->p_filesz = (unsigned long) PTR_SUB(ptr, ptr_start); 538 phdr->p_memsz = phdr->p_filesz; 539 return ptr; 540 } 541 542 /* 543 * Create ELF core header (new kernel) 544 */ 545 int elfcorehdr_alloc(unsigned long long *addr, unsigned long long *size) 546 { 547 Elf64_Phdr *phdr_notes, *phdr_loads; 548 int mem_chunk_cnt; 549 void *ptr, *hdr; 550 u32 alloc_size; 551 u64 hdr_off; 552 553 /* If we are not in kdump or zfcpdump mode return */ 554 if (!OLDMEM_BASE && ipl_info.type != IPL_TYPE_FCP_DUMP) 555 return 0; 556 /* If we cannot get HSA size for zfcpdump return error */ 557 if (ipl_info.type == IPL_TYPE_FCP_DUMP && !sclp.hsa_size) 558 return -ENODEV; 559 560 /* For kdump, exclude previous crashkernel memory */ 561 if (OLDMEM_BASE) { 562 oldmem_region.base = OLDMEM_BASE; 563 oldmem_region.size = OLDMEM_SIZE; 564 oldmem_type.total_size = OLDMEM_SIZE; 565 } 566 567 mem_chunk_cnt = get_mem_chunk_cnt(); 568 569 alloc_size = 0x1000 + get_cpu_cnt() * 0x4a0 + 570 mem_chunk_cnt * sizeof(Elf64_Phdr); 571 hdr = kzalloc_panic(alloc_size); 572 /* Init elf header */ 573 ptr = ehdr_init(hdr, mem_chunk_cnt); 574 /* Init program headers */ 575 phdr_notes = ptr; 576 ptr = PTR_ADD(ptr, sizeof(Elf64_Phdr)); 577 phdr_loads = ptr; 578 ptr = PTR_ADD(ptr, sizeof(Elf64_Phdr) * mem_chunk_cnt); 579 /* Init notes */ 580 hdr_off = PTR_DIFF(ptr, hdr); 581 ptr = notes_init(phdr_notes, ptr, ((unsigned long) hdr) + hdr_off); 582 /* Init loads */ 583 hdr_off = PTR_DIFF(ptr, hdr); 584 loads_init(phdr_loads, hdr_off); 585 *addr = (unsigned long long) hdr; 586 *size = (unsigned long long) hdr_off; 587 BUG_ON(elfcorehdr_size > alloc_size); 588 return 0; 589 } 590 591 /* 592 * Free ELF core header (new kernel) 593 */ 594 void elfcorehdr_free(unsigned long long addr) 595 { 596 kfree((void *)(unsigned long)addr); 597 } 598 599 /* 600 * Read from ELF header 601 */ 602 ssize_t elfcorehdr_read(char *buf, size_t count, u64 *ppos) 603 { 604 void *src = (void *)(unsigned long)*ppos; 605 606 memcpy(buf, src, count); 607 *ppos += count; 608 return count; 609 } 610 611 /* 612 * Read from ELF notes data 613 */ 614 ssize_t elfcorehdr_read_notes(char *buf, size_t count, u64 *ppos) 615 { 616 void *src = (void *)(unsigned long)*ppos; 617 618 memcpy(buf, src, count); 619 *ppos += count; 620 return count; 621 } 622