arch/s390/kernel/crash_dump.c

   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 <linux/memblock.h>
  17 #include <asm/os_info.h>
  18 #include <asm/elf.h>
  19 #include <asm/ipl.h>
  20 #include <asm/sclp.h>
  21
  22 #define PTR_ADD(x, y) (((char *) (x)) + ((unsigned long) (y)))
  23 #define PTR_SUB(x, y) (((char *) (x)) - ((unsigned long) (y)))
  24 #define PTR_DIFF(x, y) ((unsigned long)(((char *) (x)) - ((unsigned long) (y))))
  25
  26 static struct memblock_region oldmem_region;
  27
  28 static struct memblock_type oldmem_type = {
  29         .cnt = 1,
  30         .max = 1,
  31         .total_size = 0,
  32         .regions = &oldmem_region,
  33 };
  34
  35 struct dump_save_areas dump_save_areas;
  36
  37 /*
  38  * Return physical address for virtual address
  39  */
  40 static inline void *load_real_addr(void *addr)
  41 {
  42         unsigned long real_addr;
  43
  44         asm volatile(
  45                    "    lra     %0,0(%1)\n"
  46                    "    jz      0f\n"
  47                    "    la      %0,0\n"
  48                    "0:"
  49                    : "=a" (real_addr) : "a" (addr) : "cc");
  50         return (void *)real_addr;
  51 }
  52
  53 /*
  54  * Copy memory of the old, dumped system to a kernel space virtual address
  55  */
  56 int copy_oldmem_kernel(void *dst, void *src, size_t count)
  57 {
  58         unsigned long from, len;
  59         void *ra;
  60         int rc;
  61
  62         while (count) {
  63                 from = __pa(src);
  64                 if (!OLDMEM_BASE && from < sclp.hsa_size) {
  65                         /* Copy from zfcpdump HSA area */
  66                         len = min(count, sclp.hsa_size - from);
  67                         rc = memcpy_hsa_kernel(dst, from, len);
  68                         if (rc)
  69                                 return rc;
  70                 } else {
  71                         /* Check for swapped kdump oldmem areas */
  72                         if (OLDMEM_BASE && from - OLDMEM_BASE < OLDMEM_SIZE) {
  73                                 from -= OLDMEM_BASE;
  74                                 len = min(count, OLDMEM_SIZE - from);
  75                         } else if (OLDMEM_BASE && from < OLDMEM_SIZE) {
  76                                 len = min(count, OLDMEM_SIZE - from);
  77                                 from += OLDMEM_BASE;
  78                         } else {
  79                                 len = count;
  80                         }
  81                         if (is_vmalloc_or_module_addr(dst)) {
  82                                 ra = load_real_addr(dst);
  83                                 len = min(PAGE_SIZE - offset_in_page(ra), len);
  84                         } else {
  85                                 ra = dst;
  86                         }
  87                         if (memcpy_real(ra, (void *) from, len))
  88                                 return -EFAULT;
  89                 }
  90                 dst += len;
  91                 src += len;
  92                 count -= len;
  93         }
  94         return 0;
  95 }
  96
  97 /*
  98  * Copy memory of the old, dumped system to a user space virtual address
  99  */
 100 int copy_oldmem_user(void __user *dst, void *src, size_t count)
 101 {
 102         unsigned long from, len;
 103         int rc;
 104
 105         while (count) {
 106                 from = __pa(src);
 107                 if (!OLDMEM_BASE && from < sclp.hsa_size) {
 108                         /* Copy from zfcpdump HSA area */
 109                         len = min(count, sclp.hsa_size - from);
 110                         rc = memcpy_hsa_user(dst, from, len);
 111                         if (rc)
 112                                 return rc;
 113                 } else {
 114                         /* Check for swapped kdump oldmem areas */
 115                         if (OLDMEM_BASE && from - OLDMEM_BASE < OLDMEM_SIZE) {
 116                                 from -= OLDMEM_BASE;
 117                                 len = min(count, OLDMEM_SIZE - from);
 118                         } else if (OLDMEM_BASE && from < OLDMEM_SIZE) {
 119                                 len = min(count, OLDMEM_SIZE - from);
 120                                 from += OLDMEM_BASE;
 121                         } else {
 122                                 len = count;
 123                         }
 124                         rc = copy_to_user_real(dst, (void *) from, count);
 125                         if (rc)
 126                                 return rc;
 127                 }
 128                 dst += len;
 129                 src += len;
 130                 count -= len;
 131         }
 132         return 0;
 133 }
 134
 135 /*
 136  * Copy one page from "oldmem"
 137  */
 138 ssize_t copy_oldmem_page(unsigned long pfn, char *buf, size_t csize,
 139                          unsigned long offset, int userbuf)
 140 {
 141         void *src;
 142         int rc;
 143
 144         if (!csize)
 145                 return 0;
 146         src = (void *) (pfn << PAGE_SHIFT) + offset;
 147         if (userbuf)
 148                 rc = copy_oldmem_user((void __force __user *) buf, src, csize);
 149         else
 150                 rc = copy_oldmem_kernel((void *) buf, src, csize);
 151         return rc;
 152 }
 153
 154 /*
 155  * Remap "oldmem" for kdump
 156  *
 157  * For the kdump reserved memory this functions performs a swap operation:
 158  * [0 - OLDMEM_SIZE] is mapped to [OLDMEM_BASE - OLDMEM_BASE + OLDMEM_SIZE]
 159  */
 160 static int remap_oldmem_pfn_range_kdump(struct vm_area_struct *vma,
 161                                         unsigned long from, unsigned long pfn,
 162                                         unsigned long size, pgprot_t prot)
 163 {
 164         unsigned long size_old;
 165         int rc;
 166
 167         if (pfn < OLDMEM_SIZE >> PAGE_SHIFT) {
 168                 size_old = min(size, OLDMEM_SIZE - (pfn << PAGE_SHIFT));
 169                 rc = remap_pfn_range(vma, from,
 170                                      pfn + (OLDMEM_BASE >> PAGE_SHIFT),
 171                                      size_old, prot);
 172                 if (rc || size == size_old)
 173                         return rc;
 174                 size -= size_old;
 175                 from += size_old;
 176                 pfn += size_old >> PAGE_SHIFT;
 177         }
 178         return remap_pfn_range(vma, from, pfn, size, prot);
 179 }
 180
 181 /*
 182  * Remap "oldmem" for zfcpdump
 183  *
 184  * We only map available memory above HSA size. Memory below HSA size
 185  * is read on demand using the copy_oldmem_page() function.
 186  */
 187 static int remap_oldmem_pfn_range_zfcpdump(struct vm_area_struct *vma,
 188                                            unsigned long from,
 189                                            unsigned long pfn,
 190                                            unsigned long size, pgprot_t prot)
 191 {
 192         unsigned long hsa_end = sclp.hsa_size;
 193         unsigned long size_hsa;
 194
 195         if (pfn < hsa_end >> PAGE_SHIFT) {
 196                 size_hsa = min(size, hsa_end - (pfn << PAGE_SHIFT));
 197                 if (size == size_hsa)
 198                         return 0;
 199                 size -= size_hsa;
 200                 from += size_hsa;
 201                 pfn += size_hsa >> PAGE_SHIFT;
 202         }
 203         return remap_pfn_range(vma, from, pfn, size, prot);
 204 }
 205
 206 /*
 207  * Remap "oldmem" for kdump or zfcpdump
 208  */
 209 int remap_oldmem_pfn_range(struct vm_area_struct *vma, unsigned long from,
 210                            unsigned long pfn, unsigned long size, pgprot_t prot)
 211 {
 212         if (OLDMEM_BASE)
 213                 return remap_oldmem_pfn_range_kdump(vma, from, pfn, size, prot);
 214         else
 215                 return remap_oldmem_pfn_range_zfcpdump(vma, from, pfn, size,
 216                                                        prot);
 217 }
 218
 219 /*
 220  * Alloc memory and panic in case of ENOMEM
 221  */
 222 static void *kzalloc_panic(int len)
 223 {
 224         void *rc;
 225
 226         rc = kzalloc(len, GFP_KERNEL);
 227         if (!rc)
 228                 panic("s390 kdump kzalloc (%d) failed", len);
 229         return rc;
 230 }
 231
 232 /*
 233  * Initialize ELF note
 234  */
 235 static void *nt_init(void *buf, Elf64_Word type, void *desc, int d_len,
 236                      const char *name)
 237 {
 238         Elf64_Nhdr *note;
 239         u64 len;
 240
 241         note = (Elf64_Nhdr *)buf;
 242         note->n_namesz = strlen(name) + 1;
 243         note->n_descsz = d_len;
 244         note->n_type = type;
 245         len = sizeof(Elf64_Nhdr);
 246
 247         memcpy(buf + len, name, note->n_namesz);
 248         len = roundup(len + note->n_namesz, 4);
 249
 250         memcpy(buf + len, desc, note->n_descsz);
 251         len = roundup(len + note->n_descsz, 4);
 252
 253         return PTR_ADD(buf, len);
 254 }
 255
 256 /*
 257  * Initialize prstatus note
 258  */
 259 static void *nt_prstatus(void *ptr, struct save_area *sa)
 260 {
 261         struct elf_prstatus nt_prstatus;
 262         static int cpu_nr = 1;
 263
 264         memset(&nt_prstatus, 0, sizeof(nt_prstatus));
 265         memcpy(&nt_prstatus.pr_reg.gprs, sa->gp_regs, sizeof(sa->gp_regs));
 266         memcpy(&nt_prstatus.pr_reg.psw, sa->psw, sizeof(sa->psw));
 267         memcpy(&nt_prstatus.pr_reg.acrs, sa->acc_regs, sizeof(sa->acc_regs));
 268         nt_prstatus.pr_pid = cpu_nr;
 269         cpu_nr++;
 270
 271         return nt_init(ptr, NT_PRSTATUS, &nt_prstatus, sizeof(nt_prstatus),
 272                          "CORE");
 273 }
 274
 275 /*
 276  * Initialize fpregset (floating point) note
 277  */
 278 static void *nt_fpregset(void *ptr, struct save_area *sa)
 279 {
 280         elf_fpregset_t nt_fpregset;
 281
 282         memset(&nt_fpregset, 0, sizeof(nt_fpregset));
 283         memcpy(&nt_fpregset.fpc, &sa->fp_ctrl_reg, sizeof(sa->fp_ctrl_reg));
 284         memcpy(&nt_fpregset.fprs, &sa->fp_regs, sizeof(sa->fp_regs));
 285
 286         return nt_init(ptr, NT_PRFPREG, &nt_fpregset, sizeof(nt_fpregset),
 287                        "CORE");
 288 }
 289
 290 /*
 291  * Initialize timer note
 292  */
 293 static void *nt_s390_timer(void *ptr, struct save_area *sa)
 294 {
 295         return nt_init(ptr, NT_S390_TIMER, &sa->timer, sizeof(sa->timer),
 296                          KEXEC_CORE_NOTE_NAME);
 297 }
 298
 299 /*
 300  * Initialize TOD clock comparator note
 301  */
 302 static void *nt_s390_tod_cmp(void *ptr, struct save_area *sa)
 303 {
 304         return nt_init(ptr, NT_S390_TODCMP, &sa->clk_cmp,
 305                        sizeof(sa->clk_cmp), KEXEC_CORE_NOTE_NAME);
 306 }
 307
 308 /*
 309  * Initialize TOD programmable register note
 310  */
 311 static void *nt_s390_tod_preg(void *ptr, struct save_area *sa)
 312 {
 313         return nt_init(ptr, NT_S390_TODPREG, &sa->tod_reg,
 314                        sizeof(sa->tod_reg), KEXEC_CORE_NOTE_NAME);
 315 }
 316
 317 /*
 318  * Initialize control register note
 319  */
 320 static void *nt_s390_ctrs(void *ptr, struct save_area *sa)
 321 {
 322         return nt_init(ptr, NT_S390_CTRS, &sa->ctrl_regs,
 323                        sizeof(sa->ctrl_regs), KEXEC_CORE_NOTE_NAME);
 324 }
 325
 326 /*
 327  * Initialize prefix register note
 328  */
 329 static void *nt_s390_prefix(void *ptr, struct save_area *sa)
 330 {
 331         return nt_init(ptr, NT_S390_PREFIX, &sa->pref_reg,
 332                          sizeof(sa->pref_reg), KEXEC_CORE_NOTE_NAME);
 333 }
 334
 335 /*
 336  * Initialize vxrs high note (full 128 bit VX registers 16-31)
 337  */
 338 static void *nt_s390_vx_high(void *ptr, __vector128 *vx_regs)
 339 {
 340         return nt_init(ptr, NT_S390_VXRS_HIGH, &vx_regs[16],
 341                        16 * sizeof(__vector128), KEXEC_CORE_NOTE_NAME);
 342 }
 343
 344 /*
 345  * Initialize vxrs low note (lower halves of VX registers 0-15)
 346  */
 347 static void *nt_s390_vx_low(void *ptr, __vector128 *vx_regs)
 348 {
 349         Elf64_Nhdr *note;
 350         u64 len;
 351         int i;
 352
 353         note = (Elf64_Nhdr *)ptr;
 354         note->n_namesz = strlen(KEXEC_CORE_NOTE_NAME) + 1;
 355         note->n_descsz = 16 * 8;
 356         note->n_type = NT_S390_VXRS_LOW;
 357         len = sizeof(Elf64_Nhdr);
 358
 359         memcpy(ptr + len, KEXEC_CORE_NOTE_NAME, note->n_namesz);
 360         len = roundup(len + note->n_namesz, 4);
 361
 362         ptr += len;
 363         /* Copy lower halves of SIMD registers 0-15 */
 364         for (i = 0; i < 16; i++) {
 365                 memcpy(ptr, &vx_regs[i].u[2], 8);
 366                 ptr += 8;
 367         }
 368         return ptr;
 369 }
 370
 371 /*
 372  * Fill ELF notes for one CPU with save area registers
 373  */
 374 static void *fill_cpu_elf_notes(void *ptr, struct save_area *sa,
 375                                 __vector128 *vx_regs)
 376 {
 377         ptr = nt_prstatus(ptr, sa);
 378         ptr = nt_fpregset(ptr, sa);
 379         ptr = nt_s390_timer(ptr, sa);
 380         ptr = nt_s390_tod_cmp(ptr, sa);
 381         ptr = nt_s390_tod_preg(ptr, sa);
 382         ptr = nt_s390_ctrs(ptr, sa);
 383         ptr = nt_s390_prefix(ptr, sa);
 384         if (MACHINE_HAS_VX && vx_regs) {
 385                 ptr = nt_s390_vx_low(ptr, vx_regs);
 386                 ptr = nt_s390_vx_high(ptr, vx_regs);
 387         }
 388         return ptr;
 389 }
 390
 391 /*
 392  * Initialize prpsinfo note (new kernel)
 393  */
 394 static void *nt_prpsinfo(void *ptr)
 395 {
 396         struct elf_prpsinfo prpsinfo;
 397
 398         memset(&prpsinfo, 0, sizeof(prpsinfo));
 399         prpsinfo.pr_sname = 'R';
 400         strcpy(prpsinfo.pr_fname, "vmlinux");
 401         return nt_init(ptr, NT_PRPSINFO, &prpsinfo, sizeof(prpsinfo),
 402                        KEXEC_CORE_NOTE_NAME);
 403 }
 404
 405 /*
 406  * Get vmcoreinfo using lowcore->vmcore_info (new kernel)
 407  */
 408 static void *get_vmcoreinfo_old(unsigned long *size)
 409 {
 410         char nt_name[11], *vmcoreinfo;
 411         Elf64_Nhdr note;
 412         void *addr;
 413
 414         if (copy_oldmem_kernel(&addr, &S390_lowcore.vmcore_info, sizeof(addr)))
 415                 return NULL;
 416         memset(nt_name, 0, sizeof(nt_name));
 417         if (copy_oldmem_kernel(&note, addr, sizeof(note)))
 418                 return NULL;
 419         if (copy_oldmem_kernel(nt_name, addr + sizeof(note),
 420                                sizeof(nt_name) - 1))
 421                 return NULL;
 422         if (strcmp(nt_name, "VMCOREINFO") != 0)
 423                 return NULL;
 424         vmcoreinfo = kzalloc_panic(note.n_descsz);
 425         if (copy_oldmem_kernel(vmcoreinfo, addr + 24, note.n_descsz))
 426                 return NULL;
 427         *size = note.n_descsz;
 428         return vmcoreinfo;
 429 }
 430
 431 /*
 432  * Initialize vmcoreinfo note (new kernel)
 433  */
 434 static void *nt_vmcoreinfo(void *ptr)
 435 {
 436         unsigned long size;
 437         void *vmcoreinfo;
 438
 439         vmcoreinfo = os_info_old_entry(OS_INFO_VMCOREINFO, &size);
 440         if (!vmcoreinfo)
 441                 vmcoreinfo = get_vmcoreinfo_old(&size);
 442         if (!vmcoreinfo)
 443                 return ptr;
 444         return nt_init(ptr, 0, vmcoreinfo, size, "VMCOREINFO");
 445 }
 446
 447 /*
 448  * Initialize ELF header (new kernel)
 449  */
 450 static void *ehdr_init(Elf64_Ehdr *ehdr, int mem_chunk_cnt)
 451 {
 452         memset(ehdr, 0, sizeof(*ehdr));
 453         memcpy(ehdr->e_ident, ELFMAG, SELFMAG);
 454         ehdr->e_ident[EI_CLASS] = ELFCLASS64;
 455         ehdr->e_ident[EI_DATA] = ELFDATA2MSB;
 456         ehdr->e_ident[EI_VERSION] = EV_CURRENT;
 457         memset(ehdr->e_ident + EI_PAD, 0, EI_NIDENT - EI_PAD);
 458         ehdr->e_type = ET_CORE;
 459         ehdr->e_machine = EM_S390;
 460         ehdr->e_version = EV_CURRENT;
 461         ehdr->e_phoff = sizeof(Elf64_Ehdr);
 462         ehdr->e_ehsize = sizeof(Elf64_Ehdr);
 463         ehdr->e_phentsize = sizeof(Elf64_Phdr);
 464         ehdr->e_phnum = mem_chunk_cnt + 1;
 465         return ehdr + 1;
 466 }
 467
 468 /*
 469  * Return CPU count for ELF header (new kernel)
 470  */
 471 static int get_cpu_cnt(void)
 472 {
 473         int i, cpus = 0;
 474
 475         for (i = 0; i < dump_save_areas.count; i++) {
 476                 if (dump_save_areas.areas[i]->sa.pref_reg == 0)
 477                         continue;
 478                 cpus++;
 479         }
 480         return cpus;
 481 }
 482
 483 /*
 484  * Return memory chunk count for ELF header (new kernel)
 485  */
 486 static int get_mem_chunk_cnt(void)
 487 {
 488         int cnt = 0;
 489         u64 idx;
 490
 491         for_each_mem_range(idx, &memblock.physmem, &oldmem_type, NUMA_NO_NODE,
 492                            MEMBLOCK_NONE, NULL, NULL, NULL)
 493                 cnt++;
 494         return cnt;
 495 }
 496
 497 /*
 498  * Initialize ELF loads (new kernel)
 499  */
 500 static void loads_init(Elf64_Phdr *phdr, u64 loads_offset)
 501 {
 502         phys_addr_t start, end;
 503         u64 idx;
 504
 505         for_each_mem_range(idx, &memblock.physmem, &oldmem_type, NUMA_NO_NODE,
 506                            MEMBLOCK_NONE, &start, &end, NULL) {
 507                 phdr->p_filesz = end - start;
 508                 phdr->p_type = PT_LOAD;
 509                 phdr->p_offset = start;
 510                 phdr->p_vaddr = start;
 511                 phdr->p_paddr = start;
 512                 phdr->p_memsz = end - start;
 513                 phdr->p_flags = PF_R | PF_W | PF_X;
 514                 phdr->p_align = PAGE_SIZE;
 515                 phdr++;
 516         }
 517 }
 518
 519 /*
 520  * Initialize notes (new kernel)
 521  */
 522 static void *notes_init(Elf64_Phdr *phdr, void *ptr, u64 notes_offset)
 523 {
 524         struct save_area_ext *sa_ext;
 525         void *ptr_start = ptr;
 526         int i;
 527
 528         ptr = nt_prpsinfo(ptr);
 529
 530         for (i = 0; i < dump_save_areas.count; i++) {
 531                 sa_ext = dump_save_areas.areas[i];
 532                 if (sa_ext->sa.pref_reg == 0)
 533                         continue;
 534                 ptr = fill_cpu_elf_notes(ptr, &sa_ext->sa, sa_ext->vx_regs);
 535         }
 536         ptr = nt_vmcoreinfo(ptr);
 537         memset(phdr, 0, sizeof(*phdr));
 538         phdr->p_type = PT_NOTE;
 539         phdr->p_offset = notes_offset;
 540         phdr->p_filesz = (unsigned long) PTR_SUB(ptr, ptr_start);
 541         phdr->p_memsz = phdr->p_filesz;
 542         return ptr;
 543 }
 544
 545 /*
 546  * Create ELF core header (new kernel)
 547  */
 548 int elfcorehdr_alloc(unsigned long long *addr, unsigned long long *size)
 549 {
 550         Elf64_Phdr *phdr_notes, *phdr_loads;
 551         int mem_chunk_cnt;
 552         void *ptr, *hdr;
 553         u32 alloc_size;
 554         u64 hdr_off;
 555
 556         /* If we are not in kdump or zfcpdump mode return */
 557         if (!OLDMEM_BASE && ipl_info.type != IPL_TYPE_FCP_DUMP)
 558                 return 0;
 559         /* If we cannot get HSA size for zfcpdump return error */
 560         if (ipl_info.type == IPL_TYPE_FCP_DUMP && !sclp.hsa_size)
 561                 return -ENODEV;
 562
 563         /* For kdump, exclude previous crashkernel memory */
 564         if (OLDMEM_BASE) {
 565                 oldmem_region.base = OLDMEM_BASE;
 566                 oldmem_region.size = OLDMEM_SIZE;
 567                 oldmem_type.total_size = OLDMEM_SIZE;
 568         }
 569
 570         mem_chunk_cnt = get_mem_chunk_cnt();
 571
 572         alloc_size = 0x1000 + get_cpu_cnt() * 0x4a0 +
 573                 mem_chunk_cnt * sizeof(Elf64_Phdr);
 574         hdr = kzalloc_panic(alloc_size);
 575         /* Init elf header */
 576         ptr = ehdr_init(hdr, mem_chunk_cnt);
 577         /* Init program headers */
 578         phdr_notes = ptr;
 579         ptr = PTR_ADD(ptr, sizeof(Elf64_Phdr));
 580         phdr_loads = ptr;
 581         ptr = PTR_ADD(ptr, sizeof(Elf64_Phdr) * mem_chunk_cnt);
 582         /* Init notes */
 583         hdr_off = PTR_DIFF(ptr, hdr);
 584         ptr = notes_init(phdr_notes, ptr, ((unsigned long) hdr) + hdr_off);
 585         /* Init loads */
 586         hdr_off = PTR_DIFF(ptr, hdr);
 587         loads_init(phdr_loads, hdr_off);
 588         *addr = (unsigned long long) hdr;
 589         *size = (unsigned long long) hdr_off;
 590         BUG_ON(elfcorehdr_size > alloc_size);
 591         return 0;
 592 }
 593
 594 /*
 595  * Free ELF core header (new kernel)
 596  */
 597 void elfcorehdr_free(unsigned long long addr)
 598 {
 599         kfree((void *)(unsigned long)addr);
 600 }
 601
 602 /*
 603  * Read from ELF header
 604  */
 605 ssize_t elfcorehdr_read(char *buf, size_t count, u64 *ppos)
 606 {
 607         void *src = (void *)(unsigned long)*ppos;
 608
 609         memcpy(buf, src, count);
 610         *ppos += count;
 611         return count;
 612 }
 613
 614 /*
 615  * Read from ELF notes data
 616  */
 617 ssize_t elfcorehdr_read_notes(char *buf, size_t count, u64 *ppos)
 618 {
 619         void *src = (void *)(unsigned long)*ppos;
 620
 621         memcpy(buf, src, count);
 622         *ppos += count;
 623         return count;
 624 }