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/uaccess.h>
  51 #include <asm/facility.h>
  52 #include <asm/smp.h>
  53 #include <asm/mmu_context.h>
  54 #include <asm/cpcmd.h>
  55 #include <asm/lowcore.h>
  56 #include <asm/irq.h>
  57 #include <asm/page.h>
  58 #include <asm/ptrace.h>
  59 #include <asm/sections.h>
  60 #include <asm/ebcdic.h>
  61 #include <asm/kvm_virtio.h>
  62 #include <asm/diag.h>
  63 #include <asm/os_info.h>
  64 #include <asm/sclp.h>
  65 #include "entry.h"
  66
  67 long psw_kernel_bits    = PSW_DEFAULT_KEY | PSW_MASK_BASE | PSW_ASC_PRIMARY |
  68                           PSW_MASK_EA | PSW_MASK_BA;
  69 long psw_user_bits      = PSW_MASK_DAT | PSW_MASK_IO | PSW_MASK_EXT |
  70                           PSW_DEFAULT_KEY | PSW_MASK_BASE | PSW_MASK_MCHECK |
  71                           PSW_MASK_PSTATE | PSW_ASC_HOME;
  72
  73 /*
  74  * User copy operations.
  75  */
  76 struct uaccess_ops uaccess;
  77 EXPORT_SYMBOL(uaccess);
  78
  79 /*
  80  * Machine setup..
  81  */
  82 unsigned int console_mode = 0;
  83 EXPORT_SYMBOL(console_mode);
  84
  85 unsigned int console_devno = -1;
  86 EXPORT_SYMBOL(console_devno);
  87
  88 unsigned int console_irq = -1;
  89 EXPORT_SYMBOL(console_irq);
  90
  91 unsigned long elf_hwcap = 0;
  92 char elf_platform[ELF_PLATFORM_SIZE];
  93
  94 struct mem_chunk __initdata memory_chunk[MEMORY_CHUNKS];
  95
  96 int __initdata memory_end_set;
  97 unsigned long __initdata memory_end;
  98
  99 unsigned long VMALLOC_START;
 100 EXPORT_SYMBOL(VMALLOC_START);
 101
 102 unsigned long VMALLOC_END;
 103 EXPORT_SYMBOL(VMALLOC_END);
 104
 105 struct page *vmemmap;
 106 EXPORT_SYMBOL(vmemmap);
 107
 108 #ifdef CONFIG_64BIT
 109 unsigned long MODULES_VADDR;
 110 unsigned long MODULES_END;
 111 #endif
 112
 113 /* An array with a pointer to the lowcore of every CPU. */
 114 struct _lowcore *lowcore_ptr[NR_CPUS];
 115 EXPORT_SYMBOL(lowcore_ptr);
 116
 117 /*
 118  * This is set up by the setup-routine at boot-time
 119  * for S390 need to find out, what we have to setup
 120  * using address 0x10400 ...
 121  */
 122
 123 #include <asm/setup.h>
 124
 125 /*
 126  * condev= and conmode= setup parameter.
 127  */
 128
 129 static int __init condev_setup(char *str)
 130 {
 131         int vdev;
 132
 133         vdev = simple_strtoul(str, &str, 0);
 134         if (vdev >= 0 && vdev < 65536) {
 135                 console_devno = vdev;
 136                 console_irq = -1;
 137         }
 138         return 1;
 139 }
 140
 141 __setup("condev=", condev_setup);
 142
 143 static void __init set_preferred_console(void)
 144 {
 145         if (MACHINE_IS_KVM) {
 146                 if (sclp_has_vt220())
 147                         add_preferred_console("ttyS", 1, NULL);
 148                 else if (sclp_has_linemode())
 149                         add_preferred_console("ttyS", 0, NULL);
 150                 else
 151                         add_preferred_console("hvc", 0, NULL);
 152         } else if (CONSOLE_IS_3215 || CONSOLE_IS_SCLP)
 153                 add_preferred_console("ttyS", 0, NULL);
 154         else if (CONSOLE_IS_3270)
 155                 add_preferred_console("tty3270", 0, NULL);
 156 }
 157
 158 static int __init conmode_setup(char *str)
 159 {
 160 #if defined(CONFIG_SCLP_CONSOLE) || defined(CONFIG_SCLP_VT220_CONSOLE)
 161         if (strncmp(str, "hwc", 4) == 0 || strncmp(str, "sclp", 5) == 0)
 162                 SET_CONSOLE_SCLP;
 163 #endif
 164 #if defined(CONFIG_TN3215_CONSOLE)
 165         if (strncmp(str, "3215", 5) == 0)
 166                 SET_CONSOLE_3215;
 167 #endif
 168 #if defined(CONFIG_TN3270_CONSOLE)
 169         if (strncmp(str, "3270", 5) == 0)
 170                 SET_CONSOLE_3270;
 171 #endif
 172         set_preferred_console();
 173         return 1;
 174 }
 175
 176 __setup("conmode=", conmode_setup);
 177
 178 static void __init conmode_default(void)
 179 {
 180         char query_buffer[1024];
 181         char *ptr;
 182
 183         if (MACHINE_IS_VM) {
 184                 cpcmd("QUERY CONSOLE", query_buffer, 1024, NULL);
 185                 console_devno = simple_strtoul(query_buffer + 5, NULL, 16);
 186                 ptr = strstr(query_buffer, "SUBCHANNEL =");
 187                 console_irq = simple_strtoul(ptr + 13, NULL, 16);
 188                 cpcmd("QUERY TERM", query_buffer, 1024, NULL);
 189                 ptr = strstr(query_buffer, "CONMODE");
 190                 /*
 191                  * Set the conmode to 3215 so that the device recognition
 192                  * will set the cu_type of the console to 3215. If the
 193                  * conmode is 3270 and we don't set it back then both
 194                  * 3215 and the 3270 driver will try to access the console
 195                  * device (3215 as console and 3270 as normal tty).
 196                  */
 197                 cpcmd("TERM CONMODE 3215", NULL, 0, NULL);
 198                 if (ptr == NULL) {
 199 #if defined(CONFIG_SCLP_CONSOLE) || defined(CONFIG_SCLP_VT220_CONSOLE)
 200                         SET_CONSOLE_SCLP;
 201 #endif
 202                         return;
 203                 }
 204                 if (strncmp(ptr + 8, "3270", 4) == 0) {
 205 #if defined(CONFIG_TN3270_CONSOLE)
 206                         SET_CONSOLE_3270;
 207 #elif defined(CONFIG_TN3215_CONSOLE)
 208                         SET_CONSOLE_3215;
 209 #elif defined(CONFIG_SCLP_CONSOLE) || defined(CONFIG_SCLP_VT220_CONSOLE)
 210                         SET_CONSOLE_SCLP;
 211 #endif
 212                 } else if (strncmp(ptr + 8, "3215", 4) == 0) {
 213 #if defined(CONFIG_TN3215_CONSOLE)
 214                         SET_CONSOLE_3215;
 215 #elif defined(CONFIG_TN3270_CONSOLE)
 216                         SET_CONSOLE_3270;
 217 #elif defined(CONFIG_SCLP_CONSOLE) || defined(CONFIG_SCLP_VT220_CONSOLE)
 218                         SET_CONSOLE_SCLP;
 219 #endif
 220                 }
 221         } else {
 222 #if defined(CONFIG_SCLP_CONSOLE) || defined(CONFIG_SCLP_VT220_CONSOLE)
 223                 SET_CONSOLE_SCLP;
 224 #endif
 225         }
 226 }
 227
 228 #ifdef CONFIG_ZFCPDUMP
 229 static void __init setup_zfcpdump(void)
 230 {
 231         if (ipl_info.type != IPL_TYPE_FCP_DUMP)
 232                 return;
 233         if (OLDMEM_BASE)
 234                 return;
 235         strcat(boot_command_line, " cio_ignore=all,!ipldev,!condev");
 236         console_loglevel = 2;
 237 }
 238 #else
 239 static inline void setup_zfcpdump(void) {}
 240 #endif /* CONFIG_ZFCPDUMP */
 241
 242  /*
 243  * Reboot, halt and power_off stubs. They just call _machine_restart,
 244  * _machine_halt or _machine_power_off.
 245  */
 246
 247 void machine_restart(char *command)
 248 {
 249         if ((!in_interrupt() && !in_atomic()) || oops_in_progress)
 250                 /*
 251                  * Only unblank the console if we are called in enabled
 252                  * context or a bust_spinlocks cleared the way for us.
 253                  */
 254                 console_unblank();
 255         _machine_restart(command);
 256 }
 257
 258 void machine_halt(void)
 259 {
 260         if (!in_interrupt() || oops_in_progress)
 261                 /*
 262                  * Only unblank the console if we are called in enabled
 263                  * context or a bust_spinlocks cleared the way for us.
 264                  */
 265                 console_unblank();
 266         _machine_halt();
 267 }
 268
 269 void machine_power_off(void)
 270 {
 271         if (!in_interrupt() || oops_in_progress)
 272                 /*
 273                  * Only unblank the console if we are called in enabled
 274                  * context or a bust_spinlocks cleared the way for us.
 275                  */
 276                 console_unblank();
 277         _machine_power_off();
 278 }
 279
 280 /*
 281  * Dummy power off function.
 282  */
 283 void (*pm_power_off)(void) = machine_power_off;
 284 EXPORT_SYMBOL_GPL(pm_power_off);
 285
 286 static int __init early_parse_mem(char *p)
 287 {
 288         memory_end = memparse(p, &p);
 289         memory_end_set = 1;
 290         return 0;
 291 }
 292 early_param("mem", early_parse_mem);
 293
 294 static int __init parse_vmalloc(char *arg)
 295 {
 296         if (!arg)
 297                 return -EINVAL;
 298         VMALLOC_END = (memparse(arg, &arg) + PAGE_SIZE - 1) & PAGE_MASK;
 299         return 0;
 300 }
 301 early_param("vmalloc", parse_vmalloc);
 302
 303 unsigned int s390_user_mode = PRIMARY_SPACE_MODE;
 304 EXPORT_SYMBOL_GPL(s390_user_mode);
 305
 306 static void __init set_user_mode_primary(void)
 307 {
 308         psw_kernel_bits = (psw_kernel_bits & ~PSW_MASK_ASC) | PSW_ASC_HOME;
 309         psw_user_bits = (psw_user_bits & ~PSW_MASK_ASC) | PSW_ASC_PRIMARY;
 310 #ifdef CONFIG_COMPAT
 311         psw32_user_bits =
 312                 (psw32_user_bits & ~PSW32_MASK_ASC) | PSW32_ASC_PRIMARY;
 313 #endif
 314         uaccess = MACHINE_HAS_MVCOS ? uaccess_mvcos_switch : uaccess_pt;
 315 }
 316
 317 static int __init early_parse_user_mode(char *p)
 318 {
 319         if (p && strcmp(p, "primary") == 0)
 320                 s390_user_mode = PRIMARY_SPACE_MODE;
 321         else if (!p || strcmp(p, "home") == 0)
 322                 s390_user_mode = HOME_SPACE_MODE;
 323         else
 324                 return 1;
 325         return 0;
 326 }
 327 early_param("user_mode", early_parse_user_mode);
 328
 329 static void __init setup_addressing_mode(void)
 330 {
 331         if (s390_user_mode != PRIMARY_SPACE_MODE)
 332                 return;
 333         set_user_mode_primary();
 334         if (MACHINE_HAS_MVCOS)
 335                 pr_info("Address spaces switched, mvcos available\n");
 336         else
 337                 pr_info("Address spaces switched, mvcos not available\n");
 338 }
 339
 340 void *restart_stack __attribute__((__section__(".data")));
 341
 342 static void __init setup_lowcore(void)
 343 {
 344         struct _lowcore *lc;
 345
 346         /*
 347          * Setup lowcore for boot cpu
 348          */
 349         BUILD_BUG_ON(sizeof(struct _lowcore) != LC_PAGES * 4096);
 350         lc = __alloc_bootmem_low(LC_PAGES * PAGE_SIZE, LC_PAGES * PAGE_SIZE, 0);
 351         lc->restart_psw.mask = psw_kernel_bits;
 352         lc->restart_psw.addr =
 353                 PSW_ADDR_AMODE | (unsigned long) restart_int_handler;
 354         lc->external_new_psw.mask = psw_kernel_bits |
 355                 PSW_MASK_DAT | PSW_MASK_MCHECK;
 356         lc->external_new_psw.addr =
 357                 PSW_ADDR_AMODE | (unsigned long) ext_int_handler;
 358         lc->svc_new_psw.mask = psw_kernel_bits |
 359                 PSW_MASK_DAT | PSW_MASK_IO | PSW_MASK_EXT | PSW_MASK_MCHECK;
 360         lc->svc_new_psw.addr = PSW_ADDR_AMODE | (unsigned long) system_call;
 361         lc->program_new_psw.mask = psw_kernel_bits |
 362                 PSW_MASK_DAT | PSW_MASK_MCHECK;
 363         lc->program_new_psw.addr =
 364                 PSW_ADDR_AMODE | (unsigned long) pgm_check_handler;
 365         lc->mcck_new_psw.mask = psw_kernel_bits;
 366         lc->mcck_new_psw.addr =
 367                 PSW_ADDR_AMODE | (unsigned long) mcck_int_handler;
 368         lc->io_new_psw.mask = psw_kernel_bits |
 369                 PSW_MASK_DAT | PSW_MASK_MCHECK;
 370         lc->io_new_psw.addr = PSW_ADDR_AMODE | (unsigned long) io_int_handler;
 371         lc->clock_comparator = -1ULL;
 372         lc->kernel_stack = ((unsigned long) &init_thread_union)
 373                 + THREAD_SIZE - STACK_FRAME_OVERHEAD - sizeof(struct pt_regs);
 374         lc->async_stack = (unsigned long)
 375                 __alloc_bootmem(ASYNC_SIZE, ASYNC_SIZE, 0)
 376                 + ASYNC_SIZE - STACK_FRAME_OVERHEAD - sizeof(struct pt_regs);
 377         lc->panic_stack = (unsigned long)
 378                 __alloc_bootmem(PAGE_SIZE, PAGE_SIZE, 0)
 379                 + PAGE_SIZE - STACK_FRAME_OVERHEAD - sizeof(struct pt_regs);
 380         lc->current_task = (unsigned long) init_thread_union.thread_info.task;
 381         lc->thread_info = (unsigned long) &init_thread_union;
 382         lc->machine_flags = S390_lowcore.machine_flags;
 383         lc->stfl_fac_list = S390_lowcore.stfl_fac_list;
 384         memcpy(lc->stfle_fac_list, S390_lowcore.stfle_fac_list,
 385                MAX_FACILITY_BIT/8);
 386 #ifndef CONFIG_64BIT
 387         if (MACHINE_HAS_IEEE) {
 388                 lc->extended_save_area_addr = (__u32)
 389                         __alloc_bootmem_low(PAGE_SIZE, PAGE_SIZE, 0);
 390                 /* enable extended save area */
 391                 __ctl_set_bit(14, 29);
 392         }
 393 #else
 394         lc->vdso_per_cpu_data = (unsigned long) &lc->paste[0];
 395 #endif
 396         lc->sync_enter_timer = S390_lowcore.sync_enter_timer;
 397         lc->async_enter_timer = S390_lowcore.async_enter_timer;
 398         lc->exit_timer = S390_lowcore.exit_timer;
 399         lc->user_timer = S390_lowcore.user_timer;
 400         lc->system_timer = S390_lowcore.system_timer;
 401         lc->steal_timer = S390_lowcore.steal_timer;
 402         lc->last_update_timer = S390_lowcore.last_update_timer;
 403         lc->last_update_clock = S390_lowcore.last_update_clock;
 404         lc->ftrace_func = S390_lowcore.ftrace_func;
 405
 406         restart_stack = __alloc_bootmem(ASYNC_SIZE, ASYNC_SIZE, 0);
 407         restart_stack += ASYNC_SIZE;
 408
 409         /*
 410          * Set up PSW restart to call ipl.c:do_restart(). Copy the relevant
 411          * restart data to the absolute zero lowcore. This is necesary if
 412          * PSW restart is done on an offline CPU that has lowcore zero.
 413          */
 414         lc->restart_stack = (unsigned long) restart_stack;
 415         lc->restart_fn = (unsigned long) do_restart;
 416         lc->restart_data = 0;
 417         lc->restart_source = -1UL;
 418
 419         /* Setup absolute zero lowcore */
 420         mem_assign_absolute(S390_lowcore.restart_stack, lc->restart_stack);
 421         mem_assign_absolute(S390_lowcore.restart_fn, lc->restart_fn);
 422         mem_assign_absolute(S390_lowcore.restart_data, lc->restart_data);
 423         mem_assign_absolute(S390_lowcore.restart_source, lc->restart_source);
 424         mem_assign_absolute(S390_lowcore.restart_psw, lc->restart_psw);
 425
 426         set_prefix((u32)(unsigned long) lc);
 427         lowcore_ptr[0] = lc;
 428 }
 429
 430 static struct resource code_resource = {
 431         .name  = "Kernel code",
 432         .flags = IORESOURCE_BUSY | IORESOURCE_MEM,
 433 };
 434
 435 static struct resource data_resource = {
 436         .name = "Kernel data",
 437         .flags = IORESOURCE_BUSY | IORESOURCE_MEM,
 438 };
 439
 440 static struct resource bss_resource = {
 441         .name = "Kernel bss",
 442         .flags = IORESOURCE_BUSY | IORESOURCE_MEM,
 443 };
 444
 445 static struct resource __initdata *standard_resources[] = {
 446         &code_resource,
 447         &data_resource,
 448         &bss_resource,
 449 };
 450
 451 static void __init setup_resources(void)
 452 {
 453         struct resource *res, *std_res, *sub_res;
 454         int i, j;
 455
 456         code_resource.start = (unsigned long) &_text;
 457         code_resource.end = (unsigned long) &_etext - 1;
 458         data_resource.start = (unsigned long) &_etext;
 459         data_resource.end = (unsigned long) &_edata - 1;
 460         bss_resource.start = (unsigned long) &__bss_start;
 461         bss_resource.end = (unsigned long) &__bss_stop - 1;
 462
 463         for (i = 0; i < MEMORY_CHUNKS; i++) {
 464                 if (!memory_chunk[i].size)
 465                         continue;
 466                 res = alloc_bootmem_low(sizeof(*res));
 467                 res->flags = IORESOURCE_BUSY | IORESOURCE_MEM;
 468                 switch (memory_chunk[i].type) {
 469                 case CHUNK_READ_WRITE:
 470                         res->name = "System RAM";
 471                         break;
 472                 case CHUNK_READ_ONLY:
 473                         res->name = "System ROM";
 474                         res->flags |= IORESOURCE_READONLY;
 475                         break;
 476                 default:
 477                         res->name = "reserved";
 478                 }
 479                 res->start = memory_chunk[i].addr;
 480                 res->end = res->start + memory_chunk[i].size - 1;
 481                 request_resource(&iomem_resource, res);
 482
 483                 for (j = 0; j < ARRAY_SIZE(standard_resources); j++) {
 484                         std_res = standard_resources[j];
 485                         if (std_res->start < res->start ||
 486                             std_res->start > res->end)
 487                                 continue;
 488                         if (std_res->end > res->end) {
 489                                 sub_res = alloc_bootmem_low(sizeof(*sub_res));
 490                                 *sub_res = *std_res;
 491                                 sub_res->end = res->end;
 492                                 std_res->start = res->end + 1;
 493                                 request_resource(res, sub_res);
 494                         } else {
 495                                 request_resource(res, std_res);
 496                         }
 497                 }
 498         }
 499 }
 500
 501 static void __init setup_memory_end(void)
 502 {
 503         unsigned long vmax, vmalloc_size, tmp;
 504         unsigned long real_memory_size = 0;
 505         int i;
 506
 507
 508 #ifdef CONFIG_ZFCPDUMP
 509         if (ipl_info.type == IPL_TYPE_FCP_DUMP && !OLDMEM_BASE) {
 510                 memory_end = ZFCPDUMP_HSA_SIZE;
 511                 memory_end_set = 1;
 512         }
 513 #endif
 514         memory_end &= PAGE_MASK;
 515
 516         /*
 517          * Make sure all chunks are MAX_ORDER aligned so we don't need the
 518          * extra checks that HOLES_IN_ZONE would require.
 519          */
 520         for (i = 0; i < MEMORY_CHUNKS; i++) {
 521                 unsigned long start, end;
 522                 struct mem_chunk *chunk;
 523                 unsigned long align;
 524
 525                 chunk = &memory_chunk[i];
 526                 if (!chunk->size)
 527                         continue;
 528                 align = 1UL << (MAX_ORDER + PAGE_SHIFT - 1);
 529                 start = (chunk->addr + align - 1) & ~(align - 1);
 530                 end = (chunk->addr + chunk->size) & ~(align - 1);
 531                 if (start >= end)
 532                         memset(chunk, 0, sizeof(*chunk));
 533                 else {
 534                         chunk->addr = start;
 535                         chunk->size = end - start;
 536                 }
 537                 real_memory_size = max(real_memory_size,
 538                                        chunk->addr + chunk->size);
 539         }
 540
 541         /* Choose kernel address space layout: 2, 3, or 4 levels. */
 542 #ifdef CONFIG_64BIT
 543         vmalloc_size = VMALLOC_END ?: (128UL << 30) - MODULES_LEN;
 544         tmp = (memory_end ?: real_memory_size) / PAGE_SIZE;
 545         tmp = tmp * (sizeof(struct page) + PAGE_SIZE) + vmalloc_size;
 546         if (tmp <= (1UL << 42))
 547                 vmax = 1UL << 42;       /* 3-level kernel page table */
 548         else
 549                 vmax = 1UL << 53;       /* 4-level kernel page table */
 550         /* module area is at the end of the kernel address space. */
 551         MODULES_END = vmax;
 552         MODULES_VADDR = MODULES_END - MODULES_LEN;
 553         VMALLOC_END = MODULES_VADDR;
 554 #else
 555         vmalloc_size = VMALLOC_END ?: 96UL << 20;
 556         vmax = 1UL << 31;               /* 2-level kernel page table */
 557         /* vmalloc area is at the end of the kernel address space. */
 558         VMALLOC_END = vmax;
 559 #endif
 560         VMALLOC_START = vmax - vmalloc_size;
 561
 562         /* Split remaining virtual space between 1:1 mapping & vmemmap array */
 563         tmp = VMALLOC_START / (PAGE_SIZE + sizeof(struct page));
 564         /* vmemmap contains a multiple of PAGES_PER_SECTION struct pages */
 565         tmp = SECTION_ALIGN_UP(tmp);
 566         tmp = VMALLOC_START - tmp * sizeof(struct page);
 567         tmp &= ~((vmax >> 11) - 1);     /* align to page table level */
 568         tmp = min(tmp, 1UL << MAX_PHYSMEM_BITS);
 569         vmemmap = (struct page *) tmp;
 570
 571         /* Take care that memory_end is set and <= vmemmap */
 572         memory_end = min(memory_end ?: real_memory_size, tmp);
 573
 574         /* Fixup memory chunk array to fit into 0..memory_end */
 575         for (i = 0; i < MEMORY_CHUNKS; i++) {
 576                 struct mem_chunk *chunk = &memory_chunk[i];
 577
 578                 if (!chunk->size)
 579                         continue;
 580                 if (chunk->addr >= memory_end) {
 581                         memset(chunk, 0, sizeof(*chunk));
 582                         continue;
 583                 }
 584                 if (chunk->addr + chunk->size > memory_end)
 585                         chunk->size = memory_end - chunk->addr;
 586         }
 587 }
 588
 589 static void __init setup_vmcoreinfo(void)
 590 {
 591         mem_assign_absolute(S390_lowcore.vmcore_info, paddr_vmcoreinfo_note());
 592 }
 593
 594 #ifdef CONFIG_CRASH_DUMP
 595
 596 /*
 597  * Find suitable location for crashkernel memory
 598  */
 599 static unsigned long __init find_crash_base(unsigned long crash_size,
 600                                             char **msg)
 601 {
 602         unsigned long crash_base;
 603         struct mem_chunk *chunk;
 604         int i;
 605
 606         if (memory_chunk[0].size < crash_size) {
 607                 *msg = "first memory chunk must be at least crashkernel size";
 608                 return 0;
 609         }
 610         if (OLDMEM_BASE && crash_size == OLDMEM_SIZE)
 611                 return OLDMEM_BASE;
 612
 613         for (i = MEMORY_CHUNKS - 1; i >= 0; i--) {
 614                 chunk = &memory_chunk[i];
 615                 if (chunk->size == 0)
 616                         continue;
 617                 if (chunk->type != CHUNK_READ_WRITE)
 618                         continue;
 619                 if (chunk->size < crash_size)
 620                         continue;
 621                 crash_base = (chunk->addr + chunk->size) - crash_size;
 622                 if (crash_base < crash_size)
 623                         continue;
 624                 if (crash_base < ZFCPDUMP_HSA_SIZE_MAX)
 625                         continue;
 626                 if (crash_base < (unsigned long) INITRD_START + INITRD_SIZE)
 627                         continue;
 628                 return crash_base;
 629         }
 630         *msg = "no suitable area found";
 631         return 0;
 632 }
 633
 634 /*
 635  * Check if crash_base and crash_size is valid
 636  */
 637 static int __init verify_crash_base(unsigned long crash_base,
 638                                     unsigned long crash_size,
 639                                     char **msg)
 640 {
 641         struct mem_chunk *chunk;
 642         int i;
 643
 644         /*
 645          * Because we do the swap to zero, we must have at least 'crash_size'
 646          * bytes free space before crash_base
 647          */
 648         if (crash_size > crash_base) {
 649                 *msg = "crashkernel offset must be greater than size";
 650                 return -EINVAL;
 651         }
 652
 653         /* First memory chunk must be at least crash_size */
 654         if (memory_chunk[0].size < crash_size) {
 655                 *msg = "first memory chunk must be at least crashkernel size";
 656                 return -EINVAL;
 657         }
 658         /* Check if we fit into the respective memory chunk */
 659         for (i = 0; i < MEMORY_CHUNKS; i++) {
 660                 chunk = &memory_chunk[i];
 661                 if (chunk->size == 0)
 662                         continue;
 663                 if (crash_base < chunk->addr)
 664                         continue;
 665                 if (crash_base >= chunk->addr + chunk->size)
 666                         continue;
 667                 /* we have found the memory chunk */
 668                 if (crash_base + crash_size > chunk->addr + chunk->size) {
 669                         *msg = "selected memory chunk is too small for "
 670                                 "crashkernel memory";
 671                         return -EINVAL;
 672                 }
 673                 return 0;
 674         }
 675         *msg = "invalid memory range specified";
 676         return -EINVAL;
 677 }
 678
 679 /*
 680  * When kdump is enabled, we have to ensure that no memory from
 681  * the area [0 - crashkernel memory size] and
 682  * [crashk_res.start - crashk_res.end] is set offline.
 683  */
 684 static int kdump_mem_notifier(struct notifier_block *nb,
 685                               unsigned long action, void *data)
 686 {
 687         struct memory_notify *arg = data;
 688
 689         if (arg->start_pfn < PFN_DOWN(resource_size(&crashk_res)))
 690                 return NOTIFY_BAD;
 691         if (arg->start_pfn > PFN_DOWN(crashk_res.end))
 692                 return NOTIFY_OK;
 693         if (arg->start_pfn + arg->nr_pages - 1 < PFN_DOWN(crashk_res.start))
 694                 return NOTIFY_OK;
 695         return NOTIFY_BAD;
 696 }
 697
 698 static struct notifier_block kdump_mem_nb = {
 699         .notifier_call = kdump_mem_notifier,
 700 };
 701
 702 #endif
 703
 704 /*
 705  * Make sure that oldmem, where the dump is stored, is protected
 706  */
 707 static void reserve_oldmem(void)
 708 {
 709 #ifdef CONFIG_CRASH_DUMP
 710         unsigned long real_size = 0;
 711         int i;
 712
 713         if (!OLDMEM_BASE)
 714                 return;
 715         for (i = 0; i < MEMORY_CHUNKS; i++) {
 716                 struct mem_chunk *chunk = &memory_chunk[i];
 717
 718                 real_size = max(real_size, chunk->addr + chunk->size);
 719         }
 720         create_mem_hole(memory_chunk, OLDMEM_BASE, OLDMEM_SIZE);
 721         create_mem_hole(memory_chunk, OLDMEM_SIZE, real_size - OLDMEM_SIZE);
 722 #endif
 723 }
 724
 725 /*
 726  * Reserve memory for kdump kernel to be loaded with kexec
 727  */
 728 static void __init reserve_crashkernel(void)
 729 {
 730 #ifdef CONFIG_CRASH_DUMP
 731         unsigned long long crash_base, crash_size;
 732         char *msg = NULL;
 733         int rc;
 734
 735         rc = parse_crashkernel(boot_command_line, memory_end, &crash_size,
 736                                &crash_base);
 737         if (rc || crash_size == 0)
 738                 return;
 739         crash_base = ALIGN(crash_base, KEXEC_CRASH_MEM_ALIGN);
 740         crash_size = ALIGN(crash_size, KEXEC_CRASH_MEM_ALIGN);
 741         if (register_memory_notifier(&kdump_mem_nb))
 742                 return;
 743         if (!crash_base)
 744                 crash_base = find_crash_base(crash_size, &msg);
 745         if (!crash_base) {
 746                 pr_info("crashkernel reservation failed: %s\n", msg);
 747                 unregister_memory_notifier(&kdump_mem_nb);
 748                 return;
 749         }
 750         if (verify_crash_base(crash_base, crash_size, &msg)) {
 751                 pr_info("crashkernel reservation failed: %s\n", msg);
 752                 unregister_memory_notifier(&kdump_mem_nb);
 753                 return;
 754         }
 755         if (!OLDMEM_BASE && MACHINE_IS_VM)
 756                 diag10_range(PFN_DOWN(crash_base), PFN_DOWN(crash_size));
 757         crashk_res.start = crash_base;
 758         crashk_res.end = crash_base + crash_size - 1;
 759         insert_resource(&iomem_resource, &crashk_res);
 760         create_mem_hole(memory_chunk, crash_base, crash_size);
 761         pr_info("Reserving %lluMB of memory at %lluMB "
 762                 "for crashkernel (System RAM: %luMB)\n",
 763                 crash_size >> 20, crash_base >> 20, memory_end >> 20);
 764         os_info_crashkernel_add(crash_base, crash_size);
 765 #endif
 766 }
 767
 768 static void __init setup_memory(void)
 769 {
 770         unsigned long bootmap_size;
 771         unsigned long start_pfn, end_pfn;
 772         int i;
 773
 774         /*
 775          * partially used pages are not usable - thus
 776          * we are rounding upwards:
 777          */
 778         start_pfn = PFN_UP(__pa(&_end));
 779         end_pfn = max_pfn = PFN_DOWN(memory_end);
 780
 781 #ifdef CONFIG_BLK_DEV_INITRD
 782         /*
 783          * Move the initrd in case the bitmap of the bootmem allocater
 784          * would overwrite it.
 785          */
 786
 787         if (INITRD_START && INITRD_SIZE) {
 788                 unsigned long bmap_size;
 789                 unsigned long start;
 790
 791                 bmap_size = bootmem_bootmap_pages(end_pfn - start_pfn + 1);
 792                 bmap_size = PFN_PHYS(bmap_size);
 793
 794                 if (PFN_PHYS(start_pfn) + bmap_size > INITRD_START) {
 795                         start = PFN_PHYS(start_pfn) + bmap_size + PAGE_SIZE;
 796
 797 #ifdef CONFIG_CRASH_DUMP
 798                         if (OLDMEM_BASE) {
 799                                 /* Move initrd behind kdump oldmem */
 800                                 if (start + INITRD_SIZE > OLDMEM_BASE &&
 801                                     start < OLDMEM_BASE + OLDMEM_SIZE)
 802                                         start = OLDMEM_BASE + OLDMEM_SIZE;
 803                         }
 804 #endif
 805                         if (start + INITRD_SIZE > memory_end) {
 806                                 pr_err("initrd extends beyond end of "
 807                                        "memory (0x%08lx > 0x%08lx) "
 808                                        "disabling initrd\n",
 809                                        start + INITRD_SIZE, memory_end);
 810                                 INITRD_START = INITRD_SIZE = 0;
 811                         } else {
 812                                 pr_info("Moving initrd (0x%08lx -> "
 813                                         "0x%08lx, size: %ld)\n",
 814                                         INITRD_START, start, INITRD_SIZE);
 815                                 memmove((void *) start, (void *) INITRD_START,
 816                                         INITRD_SIZE);
 817                                 INITRD_START = start;
 818                         }
 819                 }
 820         }
 821 #endif
 822
 823         /*
 824          * Initialize the boot-time allocator
 825          */
 826         bootmap_size = init_bootmem(start_pfn, end_pfn);
 827
 828         /*
 829          * Register RAM areas with the bootmem allocator.
 830          */
 831
 832         for (i = 0; i < MEMORY_CHUNKS; i++) {
 833                 unsigned long start_chunk, end_chunk, pfn;
 834
 835                 if (!memory_chunk[i].size)
 836                         continue;
 837                 start_chunk = PFN_DOWN(memory_chunk[i].addr);
 838                 end_chunk = start_chunk + PFN_DOWN(memory_chunk[i].size);
 839                 end_chunk = min(end_chunk, end_pfn);
 840                 if (start_chunk >= end_chunk)
 841                         continue;
 842                 memblock_add_node(PFN_PHYS(start_chunk),
 843                                   PFN_PHYS(end_chunk - start_chunk), 0);
 844                 pfn = max(start_chunk, start_pfn);
 845                 storage_key_init_range(PFN_PHYS(pfn), PFN_PHYS(end_chunk));
 846         }
 847
 848         psw_set_key(PAGE_DEFAULT_KEY);
 849
 850         free_bootmem_with_active_regions(0, max_pfn);
 851
 852         /*
 853          * Reserve memory used for lowcore/command line/kernel image.
 854          */
 855         reserve_bootmem(0, (unsigned long)_ehead, BOOTMEM_DEFAULT);
 856         reserve_bootmem((unsigned long)_stext,
 857                         PFN_PHYS(start_pfn) - (unsigned long)_stext,
 858                         BOOTMEM_DEFAULT);
 859         /*
 860          * Reserve the bootmem bitmap itself as well. We do this in two
 861          * steps (first step was init_bootmem()) because this catches
 862          * the (very unlikely) case of us accidentally initializing the
 863          * bootmem allocator with an invalid RAM area.
 864          */
 865         reserve_bootmem(start_pfn << PAGE_SHIFT, bootmap_size,
 866                         BOOTMEM_DEFAULT);
 867
 868 #ifdef CONFIG_CRASH_DUMP
 869         if (crashk_res.start)
 870                 reserve_bootmem(crashk_res.start,
 871                                 crashk_res.end - crashk_res.start + 1,
 872                                 BOOTMEM_DEFAULT);
 873         if (is_kdump_kernel())
 874                 reserve_bootmem(elfcorehdr_addr - OLDMEM_BASE,
 875                                 PAGE_ALIGN(elfcorehdr_size), BOOTMEM_DEFAULT);
 876 #endif
 877 #ifdef CONFIG_BLK_DEV_INITRD
 878         if (INITRD_START && INITRD_SIZE) {
 879                 if (INITRD_START + INITRD_SIZE <= memory_end) {
 880                         reserve_bootmem(INITRD_START, INITRD_SIZE,
 881                                         BOOTMEM_DEFAULT);
 882                         initrd_start = INITRD_START;
 883                         initrd_end = initrd_start + INITRD_SIZE;
 884                 } else {
 885                         pr_err("initrd extends beyond end of "
 886                                "memory (0x%08lx > 0x%08lx) "
 887                                "disabling initrd\n",
 888                                initrd_start + INITRD_SIZE, memory_end);
 889                         initrd_start = initrd_end = 0;
 890                 }
 891         }
 892 #endif
 893 }
 894
 895 /*
 896  * Setup hardware capabilities.
 897  */
 898 static void __init setup_hwcaps(void)
 899 {
 900         static const int stfl_bits[6] = { 0, 2, 7, 17, 19, 21 };
 901         struct cpuid cpu_id;
 902         int i;
 903
 904         /*
 905          * The store facility list bits numbers as found in the principles
 906          * of operation are numbered with bit 1UL<<31 as number 0 to
 907          * bit 1UL<<0 as number 31.
 908          *   Bit 0: instructions named N3, "backported" to esa-mode
 909          *   Bit 2: z/Architecture mode is active
 910          *   Bit 7: the store-facility-list-extended facility is installed
 911          *   Bit 17: the message-security assist is installed
 912          *   Bit 19: the long-displacement facility is installed
 913          *   Bit 21: the extended-immediate facility is installed
 914          *   Bit 22: extended-translation facility 3 is installed
 915          *   Bit 30: extended-translation facility 3 enhancement facility
 916          * These get translated to:
 917          *   HWCAP_S390_ESAN3 bit 0, HWCAP_S390_ZARCH bit 1,
 918          *   HWCAP_S390_STFLE bit 2, HWCAP_S390_MSA bit 3,
 919          *   HWCAP_S390_LDISP bit 4, HWCAP_S390_EIMM bit 5 and
 920          *   HWCAP_S390_ETF3EH bit 8 (22 && 30).
 921          */
 922         for (i = 0; i < 6; i++)
 923                 if (test_facility(stfl_bits[i]))
 924                         elf_hwcap |= 1UL << i;
 925
 926         if (test_facility(22) && test_facility(30))
 927                 elf_hwcap |= HWCAP_S390_ETF3EH;
 928
 929         /*
 930          * Check for additional facilities with store-facility-list-extended.
 931          * stfle stores doublewords (8 byte) with bit 1ULL<<63 as bit 0
 932          * and 1ULL<<0 as bit 63. Bits 0-31 contain the same information
 933          * as stored by stfl, bits 32-xxx contain additional facilities.
 934          * How many facility words are stored depends on the number of
 935          * doublewords passed to the instruction. The additional facilities
 936          * are:
 937          *   Bit 42: decimal floating point facility is installed
 938          *   Bit 44: perform floating point operation facility is installed
 939          * translated to:
 940          *   HWCAP_S390_DFP bit 6 (42 && 44).
 941          */
 942         if ((elf_hwcap & (1UL << 2)) && test_facility(42) && test_facility(44))
 943                 elf_hwcap |= HWCAP_S390_DFP;
 944
 945         /*
 946          * Huge page support HWCAP_S390_HPAGE is bit 7.
 947          */
 948         if (MACHINE_HAS_HPAGE)
 949                 elf_hwcap |= HWCAP_S390_HPAGE;
 950
 951 #if defined(CONFIG_64BIT)
 952         /*
 953          * 64-bit register support for 31-bit processes
 954          * HWCAP_S390_HIGH_GPRS is bit 9.
 955          */
 956         elf_hwcap |= HWCAP_S390_HIGH_GPRS;
 957
 958         /*
 959          * Transactional execution support HWCAP_S390_TE is bit 10.
 960          */
 961         if (test_facility(50) && test_facility(73))
 962                 elf_hwcap |= HWCAP_S390_TE;
 963 #endif
 964
 965         get_cpu_id(&cpu_id);
 966         switch (cpu_id.machine) {
 967         case 0x9672:
 968 #if !defined(CONFIG_64BIT)
 969         default:        /* Use "g5" as default for 31 bit kernels. */
 970 #endif
 971                 strcpy(elf_platform, "g5");
 972                 break;
 973         case 0x2064:
 974         case 0x2066:
 975 #if defined(CONFIG_64BIT)
 976         default:        /* Use "z900" as default for 64 bit kernels. */
 977 #endif
 978                 strcpy(elf_platform, "z900");
 979                 break;
 980         case 0x2084:
 981         case 0x2086:
 982                 strcpy(elf_platform, "z990");
 983                 break;
 984         case 0x2094:
 985         case 0x2096:
 986                 strcpy(elf_platform, "z9-109");
 987                 break;
 988         case 0x2097:
 989         case 0x2098:
 990                 strcpy(elf_platform, "z10");
 991                 break;
 992         case 0x2817:
 993         case 0x2818:
 994                 strcpy(elf_platform, "z196");
 995                 break;
 996         case 0x2827:
 997         case 0x2828:
 998                 strcpy(elf_platform, "zEC12");
 999                 break;
1000         }
1001 }
1002
1003 /*
1004  * Setup function called from init/main.c just after the banner
1005  * was printed.
1006  */
1007
1008 void __init setup_arch(char **cmdline_p)
1009 {
1010         /*
1011          * print what head.S has found out about the machine
1012          */
1013 #ifndef CONFIG_64BIT
1014         if (MACHINE_IS_VM)
1015                 pr_info("Linux is running as a z/VM "
1016                         "guest operating system in 31-bit mode\n");
1017         else if (MACHINE_IS_LPAR)
1018                 pr_info("Linux is running natively in 31-bit mode\n");
1019         if (MACHINE_HAS_IEEE)
1020                 pr_info("The hardware system has IEEE compatible "
1021                         "floating point units\n");
1022         else
1023                 pr_info("The hardware system has no IEEE compatible "
1024                         "floating point units\n");
1025 #else /* CONFIG_64BIT */
1026         if (MACHINE_IS_VM)
1027                 pr_info("Linux is running as a z/VM "
1028                         "guest operating system in 64-bit mode\n");
1029         else if (MACHINE_IS_KVM)
1030                 pr_info("Linux is running under KVM in 64-bit mode\n");
1031         else if (MACHINE_IS_LPAR)
1032                 pr_info("Linux is running natively in 64-bit mode\n");
1033 #endif /* CONFIG_64BIT */
1034
1035         /* Have one command line that is parsed and saved in /proc/cmdline */
1036         /* boot_command_line has been already set up in early.c */
1037         *cmdline_p = boot_command_line;
1038
1039         ROOT_DEV = Root_RAM0;
1040
1041         init_mm.start_code = PAGE_OFFSET;
1042         init_mm.end_code = (unsigned long) &_etext;
1043         init_mm.end_data = (unsigned long) &_edata;
1044         init_mm.brk = (unsigned long) &_end;
1045
1046         if (MACHINE_HAS_MVCOS)
1047                 memcpy(&uaccess, &uaccess_mvcos, sizeof(uaccess));
1048         else
1049                 memcpy(&uaccess, &uaccess_std, sizeof(uaccess));
1050
1051         parse_early_param();
1052         detect_memory_layout(memory_chunk, memory_end);
1053         os_info_init();
1054         setup_ipl();
1055         reserve_oldmem();
1056         setup_memory_end();
1057         setup_addressing_mode();
1058         reserve_crashkernel();
1059         setup_memory();
1060         setup_resources();
1061         setup_vmcoreinfo();
1062         setup_lowcore();
1063
1064         cpu_init();
1065         s390_init_cpu_topology();
1066
1067         /*
1068          * Setup capabilities (ELF_HWCAP & ELF_PLATFORM).
1069          */
1070         setup_hwcaps();
1071
1072         /*
1073          * Create kernel page tables and switch to virtual addressing.
1074          */
1075         paging_init();
1076
1077         /* Setup default console */
1078         conmode_default();
1079         set_preferred_console();
1080
1081         /* Setup zfcpdump support */
1082         setup_zfcpdump();
1083 }