arch/s390/kernel/setup.c

   1 /*
   2  *  S390 version
   3  *    Copyright IBM Corp. 1999, 2012
   4  *    Author(s): Hartmut Penner (hp@de.ibm.com),
   5  *               Martin Schwidefsky (schwidefsky@de.ibm.com)
   6  *
   7  *  Derived from "arch/i386/kernel/setup.c"
   8  *    Copyright (C) 1995, Linus Torvalds
   9  */
  10
  11 /*
  12  * This file handles the architecture-dependent parts of initialization
  13  */
  14
  15 #define KMSG_COMPONENT "setup"
  16 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
  17
  18 #include <linux/errno.h>
  19 #include <linux/export.h>
  20 #include <linux/sched.h>
  21 #include <linux/kernel.h>
  22 #include <linux/memblock.h>
  23 #include <linux/mm.h>
  24 #include <linux/stddef.h>
  25 #include <linux/unistd.h>
  26 #include <linux/ptrace.h>
  27 #include <linux/user.h>
  28 #include <linux/tty.h>
  29 #include <linux/ioport.h>
  30 #include <linux/delay.h>
  31 #include <linux/init.h>
  32 #include <linux/initrd.h>
  33 #include <linux/bootmem.h>
  34 #include <linux/root_dev.h>
  35 #include <linux/console.h>
  36 #include <linux/kernel_stat.h>
  37 #include <linux/device.h>
  38 #include <linux/notifier.h>
  39 #include <linux/pfn.h>
  40 #include <linux/ctype.h>
  41 #include <linux/reboot.h>
  42 #include <linux/topology.h>
  43 #include <linux/ftrace.h>
  44 #include <linux/kexec.h>
  45 #include <linux/crash_dump.h>
  46 #include <linux/memory.h>
  47 #include <linux/compat.h>
  48
  49 #include <asm/ipl.h>
  50 #include <asm/facility.h>
  51 #include <asm/smp.h>
  52 #include <asm/mmu_context.h>
  53 #include <asm/cpcmd.h>
  54 #include <asm/lowcore.h>
  55 #include <asm/irq.h>
  56 #include <asm/page.h>
  57 #include <asm/ptrace.h>
  58 #include <asm/sections.h>
  59 #include <asm/ebcdic.h>
  60 #include <asm/kvm_virtio.h>
  61 #include <asm/diag.h>
  62 #include <asm/os_info.h>
  63 #include <asm/sclp.h>
  64 #include "entry.h"
  65
  66 /*
  67  * Machine setup..
  68  */
  69 unsigned int console_mode = 0;
  70 EXPORT_SYMBOL(console_mode);
  71
  72 unsigned int console_devno = -1;
  73 EXPORT_SYMBOL(console_devno);
  74
  75 unsigned int console_irq = -1;
  76 EXPORT_SYMBOL(console_irq);
  77
  78 unsigned long elf_hwcap = 0;
  79 char elf_platform[ELF_PLATFORM_SIZE];
  80
  81 struct mem_chunk __initdata memory_chunk[MEMORY_CHUNKS];
  82
  83 int __initdata memory_end_set;
  84 unsigned long __initdata memory_end;
  85
  86 unsigned long VMALLOC_START;
  87 EXPORT_SYMBOL(VMALLOC_START);
  88
  89 unsigned long VMALLOC_END;
  90 EXPORT_SYMBOL(VMALLOC_END);
  91
  92 struct page *vmemmap;
  93 EXPORT_SYMBOL(vmemmap);
  94
  95 #ifdef CONFIG_64BIT
  96 unsigned long MODULES_VADDR;
  97 unsigned long MODULES_END;
  98 #endif
  99
 100 /* An array with a pointer to the lowcore of every CPU. */
 101 struct _lowcore *lowcore_ptr[NR_CPUS];
 102 EXPORT_SYMBOL(lowcore_ptr);
 103
 104 /*
 105  * This is set up by the setup-routine at boot-time
 106  * for S390 need to find out, what we have to setup
 107  * using address 0x10400 ...
 108  */
 109
 110 #include <asm/setup.h>
 111
 112 /*
 113  * condev= and conmode= setup parameter.
 114  */
 115
 116 static int __init condev_setup(char *str)
 117 {
 118         int vdev;
 119
 120         vdev = simple_strtoul(str, &str, 0);
 121         if (vdev >= 0 && vdev < 65536) {
 122                 console_devno = vdev;
 123                 console_irq = -1;
 124         }
 125         return 1;
 126 }
 127
 128 __setup("condev=", condev_setup);
 129
 130 static void __init set_preferred_console(void)
 131 {
 132         if (MACHINE_IS_KVM) {
 133                 if (sclp_has_vt220())
 134                         add_preferred_console("ttyS", 1, NULL);
 135                 else if (sclp_has_linemode())
 136                         add_preferred_console("ttyS", 0, NULL);
 137                 else
 138                         add_preferred_console("hvc", 0, NULL);
 139         } else if (CONSOLE_IS_3215 || CONSOLE_IS_SCLP)
 140                 add_preferred_console("ttyS", 0, NULL);
 141         else if (CONSOLE_IS_3270)
 142                 add_preferred_console("tty3270", 0, NULL);
 143 }
 144
 145 static int __init conmode_setup(char *str)
 146 {
 147 #if defined(CONFIG_SCLP_CONSOLE) || defined(CONFIG_SCLP_VT220_CONSOLE)
 148         if (strncmp(str, "hwc", 4) == 0 || strncmp(str, "sclp", 5) == 0)
 149                 SET_CONSOLE_SCLP;
 150 #endif
 151 #if defined(CONFIG_TN3215_CONSOLE)
 152         if (strncmp(str, "3215", 5) == 0)
 153                 SET_CONSOLE_3215;
 154 #endif
 155 #if defined(CONFIG_TN3270_CONSOLE)
 156         if (strncmp(str, "3270", 5) == 0)
 157                 SET_CONSOLE_3270;
 158 #endif
 159         set_preferred_console();
 160         return 1;
 161 }
 162
 163 __setup("conmode=", conmode_setup);
 164
 165 static void __init conmode_default(void)
 166 {
 167         char query_buffer[1024];
 168         char *ptr;
 169
 170         if (MACHINE_IS_VM) {
 171                 cpcmd("QUERY CONSOLE", query_buffer, 1024, NULL);
 172                 console_devno = simple_strtoul(query_buffer + 5, NULL, 16);
 173                 ptr = strstr(query_buffer, "SUBCHANNEL =");
 174                 console_irq = simple_strtoul(ptr + 13, NULL, 16);
 175                 cpcmd("QUERY TERM", query_buffer, 1024, NULL);
 176                 ptr = strstr(query_buffer, "CONMODE");
 177                 /*
 178                  * Set the conmode to 3215 so that the device recognition
 179                  * will set the cu_type of the console to 3215. If the
 180                  * conmode is 3270 and we don't set it back then both
 181                  * 3215 and the 3270 driver will try to access the console
 182                  * device (3215 as console and 3270 as normal tty).
 183                  */
 184                 cpcmd("TERM CONMODE 3215", NULL, 0, NULL);
 185                 if (ptr == NULL) {
 186 #if defined(CONFIG_SCLP_CONSOLE) || defined(CONFIG_SCLP_VT220_CONSOLE)
 187                         SET_CONSOLE_SCLP;
 188 #endif
 189                         return;
 190                 }
 191                 if (strncmp(ptr + 8, "3270", 4) == 0) {
 192 #if defined(CONFIG_TN3270_CONSOLE)
 193                         SET_CONSOLE_3270;
 194 #elif defined(CONFIG_TN3215_CONSOLE)
 195                         SET_CONSOLE_3215;
 196 #elif defined(CONFIG_SCLP_CONSOLE) || defined(CONFIG_SCLP_VT220_CONSOLE)
 197                         SET_CONSOLE_SCLP;
 198 #endif
 199                 } else if (strncmp(ptr + 8, "3215", 4) == 0) {
 200 #if defined(CONFIG_TN3215_CONSOLE)
 201                         SET_CONSOLE_3215;
 202 #elif defined(CONFIG_TN3270_CONSOLE)
 203                         SET_CONSOLE_3270;
 204 #elif defined(CONFIG_SCLP_CONSOLE) || defined(CONFIG_SCLP_VT220_CONSOLE)
 205                         SET_CONSOLE_SCLP;
 206 #endif
 207                 }
 208         } else {
 209 #if defined(CONFIG_SCLP_CONSOLE) || defined(CONFIG_SCLP_VT220_CONSOLE)
 210                 SET_CONSOLE_SCLP;
 211 #endif
 212         }
 213 }
 214
 215 #ifdef CONFIG_ZFCPDUMP
 216 static void __init setup_zfcpdump(void)
 217 {
 218         if (ipl_info.type != IPL_TYPE_FCP_DUMP)
 219                 return;
 220         if (OLDMEM_BASE)
 221                 return;
 222         strcat(boot_command_line, " cio_ignore=all,!ipldev,!condev");
 223         console_loglevel = 2;
 224 }
 225 #else
 226 static inline void setup_zfcpdump(void) {}
 227 #endif /* CONFIG_ZFCPDUMP */
 228
 229  /*
 230  * Reboot, halt and power_off stubs. They just call _machine_restart,
 231  * _machine_halt or _machine_power_off.
 232  */
 233
 234 void machine_restart(char *command)
 235 {
 236         if ((!in_interrupt() && !in_atomic()) || oops_in_progress)
 237                 /*
 238                  * Only unblank the console if we are called in enabled
 239                  * context or a bust_spinlocks cleared the way for us.
 240                  */
 241                 console_unblank();
 242         _machine_restart(command);
 243 }
 244
 245 void machine_halt(void)
 246 {
 247         if (!in_interrupt() || oops_in_progress)
 248                 /*
 249                  * Only unblank the console if we are called in enabled
 250                  * context or a bust_spinlocks cleared the way for us.
 251                  */
 252                 console_unblank();
 253         _machine_halt();
 254 }
 255
 256 void machine_power_off(void)
 257 {
 258         if (!in_interrupt() || oops_in_progress)
 259                 /*
 260                  * Only unblank the console if we are called in enabled
 261                  * context or a bust_spinlocks cleared the way for us.
 262                  */
 263                 console_unblank();
 264         _machine_power_off();
 265 }
 266
 267 /*
 268  * Dummy power off function.
 269  */
 270 void (*pm_power_off)(void) = machine_power_off;
 271 EXPORT_SYMBOL_GPL(pm_power_off);
 272
 273 static int __init early_parse_mem(char *p)
 274 {
 275         memory_end = memparse(p, &p);
 276         memory_end_set = 1;
 277         return 0;
 278 }
 279 early_param("mem", early_parse_mem);
 280
 281 static int __init parse_vmalloc(char *arg)
 282 {
 283         if (!arg)
 284                 return -EINVAL;
 285         VMALLOC_END = (memparse(arg, &arg) + PAGE_SIZE - 1) & PAGE_MASK;
 286         return 0;
 287 }
 288 early_param("vmalloc", parse_vmalloc);
 289
 290 void *restart_stack __attribute__((__section__(".data")));
 291
 292 static void __init setup_lowcore(void)
 293 {
 294         struct _lowcore *lc;
 295
 296         /*
 297          * Setup lowcore for boot cpu
 298          */
 299         BUILD_BUG_ON(sizeof(struct _lowcore) != LC_PAGES * 4096);
 300         lc = __alloc_bootmem_low(LC_PAGES * PAGE_SIZE, LC_PAGES * PAGE_SIZE, 0);
 301         lc->restart_psw.mask = PSW_KERNEL_BITS;
 302         lc->restart_psw.addr =
 303                 PSW_ADDR_AMODE | (unsigned long) restart_int_handler;
 304         lc->external_new_psw.mask = PSW_KERNEL_BITS |
 305                 PSW_MASK_DAT | PSW_MASK_MCHECK;
 306         lc->external_new_psw.addr =
 307                 PSW_ADDR_AMODE | (unsigned long) ext_int_handler;
 308         lc->svc_new_psw.mask = PSW_KERNEL_BITS |
 309                 PSW_MASK_DAT | PSW_MASK_IO | PSW_MASK_EXT | PSW_MASK_MCHECK;
 310         lc->svc_new_psw.addr = PSW_ADDR_AMODE | (unsigned long) system_call;
 311         lc->program_new_psw.mask = PSW_KERNEL_BITS |
 312                 PSW_MASK_DAT | PSW_MASK_MCHECK;
 313         lc->program_new_psw.addr =
 314                 PSW_ADDR_AMODE | (unsigned long) pgm_check_handler;
 315         lc->mcck_new_psw.mask = PSW_KERNEL_BITS;
 316         lc->mcck_new_psw.addr =
 317                 PSW_ADDR_AMODE | (unsigned long) mcck_int_handler;
 318         lc->io_new_psw.mask = PSW_KERNEL_BITS |
 319                 PSW_MASK_DAT | PSW_MASK_MCHECK;
 320         lc->io_new_psw.addr = PSW_ADDR_AMODE | (unsigned long) io_int_handler;
 321         lc->clock_comparator = -1ULL;
 322         lc->kernel_stack = ((unsigned long) &init_thread_union)
 323                 + THREAD_SIZE - STACK_FRAME_OVERHEAD - sizeof(struct pt_regs);
 324         lc->async_stack = (unsigned long)
 325                 __alloc_bootmem(ASYNC_SIZE, ASYNC_SIZE, 0)
 326                 + ASYNC_SIZE - STACK_FRAME_OVERHEAD - sizeof(struct pt_regs);
 327         lc->panic_stack = (unsigned long)
 328                 __alloc_bootmem(PAGE_SIZE, PAGE_SIZE, 0)
 329                 + PAGE_SIZE - STACK_FRAME_OVERHEAD - sizeof(struct pt_regs);
 330         lc->current_task = (unsigned long) init_thread_union.thread_info.task;
 331         lc->thread_info = (unsigned long) &init_thread_union;
 332         lc->machine_flags = S390_lowcore.machine_flags;
 333         lc->stfl_fac_list = S390_lowcore.stfl_fac_list;
 334         memcpy(lc->stfle_fac_list, S390_lowcore.stfle_fac_list,
 335                MAX_FACILITY_BIT/8);
 336 #ifndef CONFIG_64BIT
 337         if (MACHINE_HAS_IEEE) {
 338                 lc->extended_save_area_addr = (__u32)
 339                         __alloc_bootmem_low(PAGE_SIZE, PAGE_SIZE, 0);
 340                 /* enable extended save area */
 341                 __ctl_set_bit(14, 29);
 342         }
 343 #else
 344         lc->vdso_per_cpu_data = (unsigned long) &lc->paste[0];
 345 #endif
 346         lc->sync_enter_timer = S390_lowcore.sync_enter_timer;
 347         lc->async_enter_timer = S390_lowcore.async_enter_timer;
 348         lc->exit_timer = S390_lowcore.exit_timer;
 349         lc->user_timer = S390_lowcore.user_timer;
 350         lc->system_timer = S390_lowcore.system_timer;
 351         lc->steal_timer = S390_lowcore.steal_timer;
 352         lc->last_update_timer = S390_lowcore.last_update_timer;
 353         lc->last_update_clock = S390_lowcore.last_update_clock;
 354         lc->ftrace_func = S390_lowcore.ftrace_func;
 355
 356         restart_stack = __alloc_bootmem(ASYNC_SIZE, ASYNC_SIZE, 0);
 357         restart_stack += ASYNC_SIZE;
 358
 359         /*
 360          * Set up PSW restart to call ipl.c:do_restart(). Copy the relevant
 361          * restart data to the absolute zero lowcore. This is necessary if
 362          * PSW restart is done on an offline CPU that has lowcore zero.
 363          */
 364         lc->restart_stack = (unsigned long) restart_stack;
 365         lc->restart_fn = (unsigned long) do_restart;
 366         lc->restart_data = 0;
 367         lc->restart_source = -1UL;
 368
 369         /* Setup absolute zero lowcore */
 370         mem_assign_absolute(S390_lowcore.restart_stack, lc->restart_stack);
 371         mem_assign_absolute(S390_lowcore.restart_fn, lc->restart_fn);
 372         mem_assign_absolute(S390_lowcore.restart_data, lc->restart_data);
 373         mem_assign_absolute(S390_lowcore.restart_source, lc->restart_source);
 374         mem_assign_absolute(S390_lowcore.restart_psw, lc->restart_psw);
 375
 376         set_prefix((u32)(unsigned long) lc);
 377         lowcore_ptr[0] = lc;
 378 }
 379
 380 static struct resource code_resource = {
 381         .name  = "Kernel code",
 382         .flags = IORESOURCE_BUSY | IORESOURCE_MEM,
 383 };
 384
 385 static struct resource data_resource = {
 386         .name = "Kernel data",
 387         .flags = IORESOURCE_BUSY | IORESOURCE_MEM,
 388 };
 389
 390 static struct resource bss_resource = {
 391         .name = "Kernel bss",
 392         .flags = IORESOURCE_BUSY | IORESOURCE_MEM,
 393 };
 394
 395 static struct resource __initdata *standard_resources[] = {
 396         &code_resource,
 397         &data_resource,
 398         &bss_resource,
 399 };
 400
 401 static void __init setup_resources(void)
 402 {
 403         struct resource *res, *std_res, *sub_res;
 404         int i, j;
 405
 406         code_resource.start = (unsigned long) &_text;
 407         code_resource.end = (unsigned long) &_etext - 1;
 408         data_resource.start = (unsigned long) &_etext;
 409         data_resource.end = (unsigned long) &_edata - 1;
 410         bss_resource.start = (unsigned long) &__bss_start;
 411         bss_resource.end = (unsigned long) &__bss_stop - 1;
 412
 413         for (i = 0; i < MEMORY_CHUNKS; i++) {
 414                 if (!memory_chunk[i].size)
 415                         continue;
 416                 res = alloc_bootmem_low(sizeof(*res));
 417                 res->flags = IORESOURCE_BUSY | IORESOURCE_MEM;
 418                 switch (memory_chunk[i].type) {
 419                 case CHUNK_READ_WRITE:
 420                         res->name = "System RAM";
 421                         break;
 422                 case CHUNK_READ_ONLY:
 423                         res->name = "System ROM";
 424                         res->flags |= IORESOURCE_READONLY;
 425                         break;
 426                 default:
 427                         res->name = "reserved";
 428                 }
 429                 res->start = memory_chunk[i].addr;
 430                 res->end = res->start + memory_chunk[i].size - 1;
 431                 request_resource(&iomem_resource, res);
 432
 433                 for (j = 0; j < ARRAY_SIZE(standard_resources); j++) {
 434                         std_res = standard_resources[j];
 435                         if (std_res->start < res->start ||
 436                             std_res->start > res->end)
 437                                 continue;
 438                         if (std_res->end > res->end) {
 439                                 sub_res = alloc_bootmem_low(sizeof(*sub_res));
 440                                 *sub_res = *std_res;
 441                                 sub_res->end = res->end;
 442                                 std_res->start = res->end + 1;
 443                                 request_resource(res, sub_res);
 444                         } else {
 445                                 request_resource(res, std_res);
 446                         }
 447                 }
 448         }
 449 }
 450
 451 static void __init setup_memory_end(void)
 452 {
 453         unsigned long vmax, vmalloc_size, tmp;
 454         unsigned long real_memory_size = 0;
 455         int i;
 456
 457
 458 #ifdef CONFIG_ZFCPDUMP
 459         if (ipl_info.type == IPL_TYPE_FCP_DUMP &&
 460             !OLDMEM_BASE && sclp_get_hsa_size()) {
 461                 memory_end = sclp_get_hsa_size();
 462                 memory_end_set = 1;
 463         }
 464 #endif
 465         memory_end &= PAGE_MASK;
 466
 467         /*
 468          * Make sure all chunks are MAX_ORDER aligned so we don't need the
 469          * extra checks that HOLES_IN_ZONE would require.
 470          */
 471         for (i = 0; i < MEMORY_CHUNKS; i++) {
 472                 unsigned long start, end;
 473                 struct mem_chunk *chunk;
 474                 unsigned long align;
 475
 476                 chunk = &memory_chunk[i];
 477                 if (!chunk->size)
 478                         continue;
 479                 align = 1UL << (MAX_ORDER + PAGE_SHIFT - 1);
 480                 start = (chunk->addr + align - 1) & ~(align - 1);
 481                 end = (chunk->addr + chunk->size) & ~(align - 1);
 482                 if (start >= end)
 483                         memset(chunk, 0, sizeof(*chunk));
 484                 else {
 485                         chunk->addr = start;
 486                         chunk->size = end - start;
 487                 }
 488                 real_memory_size = max(real_memory_size,
 489                                        chunk->addr + chunk->size);
 490         }
 491
 492         /* Choose kernel address space layout: 2, 3, or 4 levels. */
 493 #ifdef CONFIG_64BIT
 494         vmalloc_size = VMALLOC_END ?: (128UL << 30) - MODULES_LEN;
 495         tmp = (memory_end ?: real_memory_size) / PAGE_SIZE;
 496         tmp = tmp * (sizeof(struct page) + PAGE_SIZE) + vmalloc_size;
 497         if (tmp <= (1UL << 42))
 498                 vmax = 1UL << 42;       /* 3-level kernel page table */
 499         else
 500                 vmax = 1UL << 53;       /* 4-level kernel page table */
 501         /* module area is at the end of the kernel address space. */
 502         MODULES_END = vmax;
 503         MODULES_VADDR = MODULES_END - MODULES_LEN;
 504         VMALLOC_END = MODULES_VADDR;
 505 #else
 506         vmalloc_size = VMALLOC_END ?: 96UL << 20;
 507         vmax = 1UL << 31;               /* 2-level kernel page table */
 508         /* vmalloc area is at the end of the kernel address space. */
 509         VMALLOC_END = vmax;
 510 #endif
 511         VMALLOC_START = vmax - vmalloc_size;
 512
 513         /* Split remaining virtual space between 1:1 mapping & vmemmap array */
 514         tmp = VMALLOC_START / (PAGE_SIZE + sizeof(struct page));
 515         /* vmemmap contains a multiple of PAGES_PER_SECTION struct pages */
 516         tmp = SECTION_ALIGN_UP(tmp);
 517         tmp = VMALLOC_START - tmp * sizeof(struct page);
 518         tmp &= ~((vmax >> 11) - 1);     /* align to page table level */
 519         tmp = min(tmp, 1UL << MAX_PHYSMEM_BITS);
 520         vmemmap = (struct page *) tmp;
 521
 522         /* Take care that memory_end is set and <= vmemmap */
 523         memory_end = min(memory_end ?: real_memory_size, tmp);
 524
 525         /* Fixup memory chunk array to fit into 0..memory_end */
 526         for (i = 0; i < MEMORY_CHUNKS; i++) {
 527                 struct mem_chunk *chunk = &memory_chunk[i];
 528
 529                 if (!chunk->size)
 530                         continue;
 531                 if (chunk->addr >= memory_end) {
 532                         memset(chunk, 0, sizeof(*chunk));
 533                         continue;
 534                 }
 535                 if (chunk->addr + chunk->size > memory_end)
 536                         chunk->size = memory_end - chunk->addr;
 537         }
 538 }
 539
 540 static void __init setup_vmcoreinfo(void)
 541 {
 542         mem_assign_absolute(S390_lowcore.vmcore_info, paddr_vmcoreinfo_note());
 543 }
 544
 545 #ifdef CONFIG_CRASH_DUMP
 546
 547 /*
 548  * Find suitable location for crashkernel memory
 549  */
 550 static unsigned long __init find_crash_base(unsigned long crash_size,
 551                                             char **msg)
 552 {
 553         unsigned long crash_base;
 554         struct mem_chunk *chunk;
 555         int i;
 556
 557         if (memory_chunk[0].size < crash_size) {
 558                 *msg = "first memory chunk must be at least crashkernel size";
 559                 return 0;
 560         }
 561         if (OLDMEM_BASE && crash_size == OLDMEM_SIZE)
 562                 return OLDMEM_BASE;
 563
 564         for (i = MEMORY_CHUNKS - 1; i >= 0; i--) {
 565                 chunk = &memory_chunk[i];
 566                 if (chunk->size == 0)
 567                         continue;
 568                 if (chunk->type != CHUNK_READ_WRITE)
 569                         continue;
 570                 if (chunk->size < crash_size)
 571                         continue;
 572                 crash_base = (chunk->addr + chunk->size) - crash_size;
 573                 if (crash_base < crash_size)
 574                         continue;
 575                 if (crash_base < sclp_get_hsa_size())
 576                         continue;
 577                 if (crash_base < (unsigned long) INITRD_START + INITRD_SIZE)
 578                         continue;
 579                 return crash_base;
 580         }
 581         *msg = "no suitable area found";
 582         return 0;
 583 }
 584
 585 /*
 586  * Check if crash_base and crash_size is valid
 587  */
 588 static int __init verify_crash_base(unsigned long crash_base,
 589                                     unsigned long crash_size,
 590                                     char **msg)
 591 {
 592         struct mem_chunk *chunk;
 593         int i;
 594
 595         /*
 596          * Because we do the swap to zero, we must have at least 'crash_size'
 597          * bytes free space before crash_base
 598          */
 599         if (crash_size > crash_base) {
 600                 *msg = "crashkernel offset must be greater than size";
 601                 return -EINVAL;
 602         }
 603
 604         /* First memory chunk must be at least crash_size */
 605         if (memory_chunk[0].size < crash_size) {
 606                 *msg = "first memory chunk must be at least crashkernel size";
 607                 return -EINVAL;
 608         }
 609         /* Check if we fit into the respective memory chunk */
 610         for (i = 0; i < MEMORY_CHUNKS; i++) {
 611                 chunk = &memory_chunk[i];
 612                 if (chunk->size == 0)
 613                         continue;
 614                 if (crash_base < chunk->addr)
 615                         continue;
 616                 if (crash_base >= chunk->addr + chunk->size)
 617                         continue;
 618                 /* we have found the memory chunk */
 619                 if (crash_base + crash_size > chunk->addr + chunk->size) {
 620                         *msg = "selected memory chunk is too small for "
 621                                 "crashkernel memory";
 622                         return -EINVAL;
 623                 }
 624                 return 0;
 625         }
 626         *msg = "invalid memory range specified";
 627         return -EINVAL;
 628 }
 629
 630 /*
 631  * When kdump is enabled, we have to ensure that no memory from
 632  * the area [0 - crashkernel memory size] and
 633  * [crashk_res.start - crashk_res.end] is set offline.
 634  */
 635 static int kdump_mem_notifier(struct notifier_block *nb,
 636                               unsigned long action, void *data)
 637 {
 638         struct memory_notify *arg = data;
 639
 640         if (arg->start_pfn < PFN_DOWN(resource_size(&crashk_res)))
 641                 return NOTIFY_BAD;
 642         if (arg->start_pfn > PFN_DOWN(crashk_res.end))
 643                 return NOTIFY_OK;
 644         if (arg->start_pfn + arg->nr_pages - 1 < PFN_DOWN(crashk_res.start))
 645                 return NOTIFY_OK;
 646         return NOTIFY_BAD;
 647 }
 648
 649 static struct notifier_block kdump_mem_nb = {
 650         .notifier_call = kdump_mem_notifier,
 651 };
 652
 653 #endif
 654
 655 /*
 656  * Make sure that oldmem, where the dump is stored, is protected
 657  */
 658 static void reserve_oldmem(void)
 659 {
 660 #ifdef CONFIG_CRASH_DUMP
 661         unsigned long real_size = 0;
 662         int i;
 663
 664         if (!OLDMEM_BASE)
 665                 return;
 666         for (i = 0; i < MEMORY_CHUNKS; i++) {
 667                 struct mem_chunk *chunk = &memory_chunk[i];
 668
 669                 real_size = max(real_size, chunk->addr + chunk->size);
 670         }
 671         create_mem_hole(memory_chunk, OLDMEM_BASE, OLDMEM_SIZE);
 672         create_mem_hole(memory_chunk, OLDMEM_SIZE, real_size - OLDMEM_SIZE);
 673 #endif
 674 }
 675
 676 /*
 677  * Reserve memory for kdump kernel to be loaded with kexec
 678  */
 679 static void __init reserve_crashkernel(void)
 680 {
 681 #ifdef CONFIG_CRASH_DUMP
 682         unsigned long long crash_base, crash_size;
 683         char *msg = NULL;
 684         int rc;
 685
 686         rc = parse_crashkernel(boot_command_line, memory_end, &crash_size,
 687                                &crash_base);
 688         if (rc || crash_size == 0)
 689                 return;
 690         crash_base = ALIGN(crash_base, KEXEC_CRASH_MEM_ALIGN);
 691         crash_size = ALIGN(crash_size, KEXEC_CRASH_MEM_ALIGN);
 692         if (register_memory_notifier(&kdump_mem_nb))
 693                 return;
 694         if (!crash_base)
 695                 crash_base = find_crash_base(crash_size, &msg);
 696         if (!crash_base) {
 697                 pr_info("crashkernel reservation failed: %s\n", msg);
 698                 unregister_memory_notifier(&kdump_mem_nb);
 699                 return;
 700         }
 701         if (verify_crash_base(crash_base, crash_size, &msg)) {
 702                 pr_info("crashkernel reservation failed: %s\n", msg);
 703                 unregister_memory_notifier(&kdump_mem_nb);
 704                 return;
 705         }
 706         if (!OLDMEM_BASE && MACHINE_IS_VM)
 707                 diag10_range(PFN_DOWN(crash_base), PFN_DOWN(crash_size));
 708         crashk_res.start = crash_base;
 709         crashk_res.end = crash_base + crash_size - 1;
 710         insert_resource(&iomem_resource, &crashk_res);
 711         create_mem_hole(memory_chunk, crash_base, crash_size);
 712         pr_info("Reserving %lluMB of memory at %lluMB "
 713                 "for crashkernel (System RAM: %luMB)\n",
 714                 crash_size >> 20, crash_base >> 20, memory_end >> 20);
 715         os_info_crashkernel_add(crash_base, crash_size);
 716 #endif
 717 }
 718
 719 static void __init setup_memory(void)
 720 {
 721         unsigned long bootmap_size;
 722         unsigned long start_pfn, end_pfn;
 723         int i;
 724
 725         /*
 726          * partially used pages are not usable - thus
 727          * we are rounding upwards:
 728          */
 729         start_pfn = PFN_UP(__pa(&_end));
 730         end_pfn = max_pfn = PFN_DOWN(memory_end);
 731
 732 #ifdef CONFIG_BLK_DEV_INITRD
 733         /*
 734          * Move the initrd in case the bitmap of the bootmem allocater
 735          * would overwrite it.
 736          */
 737
 738         if (INITRD_START && INITRD_SIZE) {
 739                 unsigned long bmap_size;
 740                 unsigned long start;
 741
 742                 bmap_size = bootmem_bootmap_pages(end_pfn - start_pfn + 1);
 743                 bmap_size = PFN_PHYS(bmap_size);
 744
 745                 if (PFN_PHYS(start_pfn) + bmap_size > INITRD_START) {
 746                         start = PFN_PHYS(start_pfn) + bmap_size + PAGE_SIZE;
 747
 748 #ifdef CONFIG_CRASH_DUMP
 749                         if (OLDMEM_BASE) {
 750                                 /* Move initrd behind kdump oldmem */
 751                                 if (start + INITRD_SIZE > OLDMEM_BASE &&
 752                                     start < OLDMEM_BASE + OLDMEM_SIZE)
 753                                         start = OLDMEM_BASE + OLDMEM_SIZE;
 754                         }
 755 #endif
 756                         if (start + INITRD_SIZE > memory_end) {
 757                                 pr_err("initrd extends beyond end of "
 758                                        "memory (0x%08lx > 0x%08lx) "
 759                                        "disabling initrd\n",
 760                                        start + INITRD_SIZE, memory_end);
 761                                 INITRD_START = INITRD_SIZE = 0;
 762                         } else {
 763                                 pr_info("Moving initrd (0x%08lx -> "
 764                                         "0x%08lx, size: %ld)\n",
 765                                         INITRD_START, start, INITRD_SIZE);
 766                                 memmove((void *) start, (void *) INITRD_START,
 767                                         INITRD_SIZE);
 768                                 INITRD_START = start;
 769                         }
 770                 }
 771         }
 772 #endif
 773
 774         /*
 775          * Initialize the boot-time allocator
 776          */
 777         bootmap_size = init_bootmem(start_pfn, end_pfn);
 778
 779         /*
 780          * Register RAM areas with the bootmem allocator.
 781          */
 782
 783         for (i = 0; i < MEMORY_CHUNKS; i++) {
 784                 unsigned long start_chunk, end_chunk, pfn;
 785
 786                 if (!memory_chunk[i].size)
 787                         continue;
 788                 start_chunk = PFN_DOWN(memory_chunk[i].addr);
 789                 end_chunk = start_chunk + PFN_DOWN(memory_chunk[i].size);
 790                 end_chunk = min(end_chunk, end_pfn);
 791                 if (start_chunk >= end_chunk)
 792                         continue;
 793                 memblock_add_node(PFN_PHYS(start_chunk),
 794                                   PFN_PHYS(end_chunk - start_chunk), 0);
 795                 pfn = max(start_chunk, start_pfn);
 796                 storage_key_init_range(PFN_PHYS(pfn), PFN_PHYS(end_chunk));
 797         }
 798
 799         psw_set_key(PAGE_DEFAULT_KEY);
 800
 801         free_bootmem_with_active_regions(0, max_pfn);
 802
 803         /*
 804          * Reserve memory used for lowcore/command line/kernel image.
 805          */
 806         reserve_bootmem(0, (unsigned long)_ehead, BOOTMEM_DEFAULT);
 807         reserve_bootmem((unsigned long)_stext,
 808                         PFN_PHYS(start_pfn) - (unsigned long)_stext,
 809                         BOOTMEM_DEFAULT);
 810         /*
 811          * Reserve the bootmem bitmap itself as well. We do this in two
 812          * steps (first step was init_bootmem()) because this catches
 813          * the (very unlikely) case of us accidentally initializing the
 814          * bootmem allocator with an invalid RAM area.
 815          */
 816         reserve_bootmem(start_pfn << PAGE_SHIFT, bootmap_size,
 817                         BOOTMEM_DEFAULT);
 818
 819 #ifdef CONFIG_CRASH_DUMP
 820         if (crashk_res.start)
 821                 reserve_bootmem(crashk_res.start,
 822                                 crashk_res.end - crashk_res.start + 1,
 823                                 BOOTMEM_DEFAULT);
 824         if (is_kdump_kernel())
 825                 reserve_bootmem(elfcorehdr_addr - OLDMEM_BASE,
 826                                 PAGE_ALIGN(elfcorehdr_size), BOOTMEM_DEFAULT);
 827 #endif
 828 #ifdef CONFIG_BLK_DEV_INITRD
 829         if (INITRD_START && INITRD_SIZE) {
 830                 if (INITRD_START + INITRD_SIZE <= memory_end) {
 831                         reserve_bootmem(INITRD_START, INITRD_SIZE,
 832                                         BOOTMEM_DEFAULT);
 833                         initrd_start = INITRD_START;
 834                         initrd_end = initrd_start + INITRD_SIZE;
 835                 } else {
 836                         pr_err("initrd extends beyond end of "
 837                                "memory (0x%08lx > 0x%08lx) "
 838                                "disabling initrd\n",
 839                                initrd_start + INITRD_SIZE, memory_end);
 840                         initrd_start = initrd_end = 0;
 841                 }
 842         }
 843 #endif
 844 }
 845
 846 /*
 847  * Setup hardware capabilities.
 848  */
 849 static void __init setup_hwcaps(void)
 850 {
 851         static const int stfl_bits[6] = { 0, 2, 7, 17, 19, 21 };
 852         struct cpuid cpu_id;
 853         int i;
 854
 855         /*
 856          * The store facility list bits numbers as found in the principles
 857          * of operation are numbered with bit 1UL<<31 as number 0 to
 858          * bit 1UL<<0 as number 31.
 859          *   Bit 0: instructions named N3, "backported" to esa-mode
 860          *   Bit 2: z/Architecture mode is active
 861          *   Bit 7: the store-facility-list-extended facility is installed
 862          *   Bit 17: the message-security assist is installed
 863          *   Bit 19: the long-displacement facility is installed
 864          *   Bit 21: the extended-immediate facility is installed
 865          *   Bit 22: extended-translation facility 3 is installed
 866          *   Bit 30: extended-translation facility 3 enhancement facility
 867          * These get translated to:
 868          *   HWCAP_S390_ESAN3 bit 0, HWCAP_S390_ZARCH bit 1,
 869          *   HWCAP_S390_STFLE bit 2, HWCAP_S390_MSA bit 3,
 870          *   HWCAP_S390_LDISP bit 4, HWCAP_S390_EIMM bit 5 and
 871          *   HWCAP_S390_ETF3EH bit 8 (22 && 30).
 872          */
 873         for (i = 0; i < 6; i++)
 874                 if (test_facility(stfl_bits[i]))
 875                         elf_hwcap |= 1UL << i;
 876
 877         if (test_facility(22) && test_facility(30))
 878                 elf_hwcap |= HWCAP_S390_ETF3EH;
 879
 880         /*
 881          * Check for additional facilities with store-facility-list-extended.
 882          * stfle stores doublewords (8 byte) with bit 1ULL<<63 as bit 0
 883          * and 1ULL<<0 as bit 63. Bits 0-31 contain the same information
 884          * as stored by stfl, bits 32-xxx contain additional facilities.
 885          * How many facility words are stored depends on the number of
 886          * doublewords passed to the instruction. The additional facilities
 887          * are:
 888          *   Bit 42: decimal floating point facility is installed
 889          *   Bit 44: perform floating point operation facility is installed
 890          * translated to:
 891          *   HWCAP_S390_DFP bit 6 (42 && 44).
 892          */
 893         if ((elf_hwcap & (1UL << 2)) && test_facility(42) && test_facility(44))
 894                 elf_hwcap |= HWCAP_S390_DFP;
 895
 896         /*
 897          * Huge page support HWCAP_S390_HPAGE is bit 7.
 898          */
 899         if (MACHINE_HAS_HPAGE)
 900                 elf_hwcap |= HWCAP_S390_HPAGE;
 901
 902 #if defined(CONFIG_64BIT)
 903         /*
 904          * 64-bit register support for 31-bit processes
 905          * HWCAP_S390_HIGH_GPRS is bit 9.
 906          */
 907         elf_hwcap |= HWCAP_S390_HIGH_GPRS;
 908
 909         /*
 910          * Transactional execution support HWCAP_S390_TE is bit 10.
 911          */
 912         if (test_facility(50) && test_facility(73))
 913                 elf_hwcap |= HWCAP_S390_TE;
 914 #endif
 915
 916         get_cpu_id(&cpu_id);
 917         switch (cpu_id.machine) {
 918         case 0x9672:
 919 #if !defined(CONFIG_64BIT)
 920         default:        /* Use "g5" as default for 31 bit kernels. */
 921 #endif
 922                 strcpy(elf_platform, "g5");
 923                 break;
 924         case 0x2064:
 925         case 0x2066:
 926 #if defined(CONFIG_64BIT)
 927         default:        /* Use "z900" as default for 64 bit kernels. */
 928 #endif
 929                 strcpy(elf_platform, "z900");
 930                 break;
 931         case 0x2084:
 932         case 0x2086:
 933                 strcpy(elf_platform, "z990");
 934                 break;
 935         case 0x2094:
 936         case 0x2096:
 937                 strcpy(elf_platform, "z9-109");
 938                 break;
 939         case 0x2097:
 940         case 0x2098:
 941                 strcpy(elf_platform, "z10");
 942                 break;
 943         case 0x2817:
 944         case 0x2818:
 945                 strcpy(elf_platform, "z196");
 946                 break;
 947         case 0x2827:
 948         case 0x2828:
 949                 strcpy(elf_platform, "zEC12");
 950                 break;
 951         }
 952 }
 953
 954 /*
 955  * Setup function called from init/main.c just after the banner
 956  * was printed.
 957  */
 958
 959 void __init setup_arch(char **cmdline_p)
 960 {
 961         /*
 962          * print what head.S has found out about the machine
 963          */
 964 #ifndef CONFIG_64BIT
 965         if (MACHINE_IS_VM)
 966                 pr_info("Linux is running as a z/VM "
 967                         "guest operating system in 31-bit mode\n");
 968         else if (MACHINE_IS_LPAR)
 969                 pr_info("Linux is running natively in 31-bit mode\n");
 970         if (MACHINE_HAS_IEEE)
 971                 pr_info("The hardware system has IEEE compatible "
 972                         "floating point units\n");
 973         else
 974                 pr_info("The hardware system has no IEEE compatible "
 975                         "floating point units\n");
 976 #else /* CONFIG_64BIT */
 977         if (MACHINE_IS_VM)
 978                 pr_info("Linux is running as a z/VM "
 979                         "guest operating system in 64-bit mode\n");
 980         else if (MACHINE_IS_KVM)
 981                 pr_info("Linux is running under KVM in 64-bit mode\n");
 982         else if (MACHINE_IS_LPAR)
 983                 pr_info("Linux is running natively in 64-bit mode\n");
 984 #endif /* CONFIG_64BIT */
 985
 986         /* Have one command line that is parsed and saved in /proc/cmdline */
 987         /* boot_command_line has been already set up in early.c */
 988         *cmdline_p = boot_command_line;
 989
 990         ROOT_DEV = Root_RAM0;
 991
 992         init_mm.start_code = PAGE_OFFSET;
 993         init_mm.end_code = (unsigned long) &_etext;
 994         init_mm.end_data = (unsigned long) &_edata;
 995         init_mm.brk = (unsigned long) &_end;
 996
 997         parse_early_param();
 998         detect_memory_layout(memory_chunk, memory_end);
 999         os_info_init();
1000         setup_ipl();
1001         reserve_oldmem();
1002         setup_memory_end();
1003         reserve_crashkernel();
1004         setup_memory();
1005         setup_resources();
1006         setup_vmcoreinfo();
1007         setup_lowcore();
1008
1009         smp_fill_possible_mask();
1010         cpu_init();
1011         s390_init_cpu_topology();
1012
1013         /*
1014          * Setup capabilities (ELF_HWCAP & ELF_PLATFORM).
1015          */
1016         setup_hwcaps();
1017
1018         /*
1019          * Create kernel page tables and switch to virtual addressing.
1020          */
1021         paging_init();
1022
1023         /* Setup default console */
1024         conmode_default();
1025         set_preferred_console();
1026
1027         /* Setup zfcpdump support */
1028         setup_zfcpdump();
1029 }