arch/powerpc/kernel/setup_64.c

   1 /*
   2  *
   3  * Common boot and setup code.
   4  *
   5  * Copyright (C) 2001 PPC64 Team, IBM Corp
   6  *
   7  *      This program is free software; you can redistribute it and/or
   8  *      modify it under the terms of the GNU General Public License
   9  *      as published by the Free Software Foundation; either version
  10  *      2 of the License, or (at your option) any later version.
  11  */
  12
  13 #undef DEBUG
  14
  15 #include <linux/config.h>
  16 #include <linux/module.h>
  17 #include <linux/string.h>
  18 #include <linux/sched.h>
  19 #include <linux/init.h>
  20 #include <linux/kernel.h>
  21 #include <linux/reboot.h>
  22 #include <linux/delay.h>
  23 #include <linux/initrd.h>
  24 #include <linux/ide.h>
  25 #include <linux/seq_file.h>
  26 #include <linux/ioport.h>
  27 #include <linux/console.h>
  28 #include <linux/utsname.h>
  29 #include <linux/tty.h>
  30 #include <linux/root_dev.h>
  31 #include <linux/notifier.h>
  32 #include <linux/cpu.h>
  33 #include <linux/unistd.h>
  34 #include <linux/serial.h>
  35 #include <linux/serial_8250.h>
  36 #include <linux/bootmem.h>
  37 #include <asm/io.h>
  38 #include <asm/kdump.h>
  39 #include <asm/prom.h>
  40 #include <asm/processor.h>
  41 #include <asm/pgtable.h>
  42 #include <asm/smp.h>
  43 #include <asm/elf.h>
  44 #include <asm/machdep.h>
  45 #include <asm/paca.h>
  46 #include <asm/time.h>
  47 #include <asm/cputable.h>
  48 #include <asm/sections.h>
  49 #include <asm/btext.h>
  50 #include <asm/nvram.h>
  51 #include <asm/setup.h>
  52 #include <asm/system.h>
  53 #include <asm/rtas.h>
  54 #include <asm/iommu.h>
  55 #include <asm/serial.h>
  56 #include <asm/cache.h>
  57 #include <asm/page.h>
  58 #include <asm/mmu.h>
  59 #include <asm/lmb.h>
  60 #include <asm/iseries/it_lp_naca.h>
  61 #include <asm/firmware.h>
  62 #include <asm/xmon.h>
  63 #include <asm/udbg.h>
  64 #include <asm/kexec.h>
  65
  66 #include "setup.h"
  67
  68 #ifdef DEBUG
  69 #define DBG(fmt...) udbg_printf(fmt)
  70 #else
  71 #define DBG(fmt...)
  72 #endif
  73
  74 int have_of = 1;
  75 int boot_cpuid = 0;
  76 dev_t boot_dev;
  77 u64 ppc64_pft_size;
  78
  79 /* Pick defaults since we might want to patch instructions
  80  * before we've read this from the device tree.
  81  */
  82 struct ppc64_caches ppc64_caches = {
  83         .dline_size = 0x80,
  84         .log_dline_size = 7,
  85         .iline_size = 0x80,
  86         .log_iline_size = 7
  87 };
  88 EXPORT_SYMBOL_GPL(ppc64_caches);
  89
  90 /*
  91  * These are used in binfmt_elf.c to put aux entries on the stack
  92  * for each elf executable being started.
  93  */
  94 int dcache_bsize;
  95 int icache_bsize;
  96 int ucache_bsize;
  97
  98 #ifdef CONFIG_MAGIC_SYSRQ
  99 unsigned long SYSRQ_KEY;
 100 #endif /* CONFIG_MAGIC_SYSRQ */
 101
 102
 103 static int ppc64_panic_event(struct notifier_block *, unsigned long, void *);
 104 static struct notifier_block ppc64_panic_block = {
 105         .notifier_call = ppc64_panic_event,
 106         .priority = INT_MIN /* may not return; must be done last */
 107 };
 108
 109 #ifdef CONFIG_SMP
 110
 111 static int smt_enabled_cmdline;
 112
 113 /* Look for ibm,smt-enabled OF option */
 114 static void check_smt_enabled(void)
 115 {
 116         struct device_node *dn;
 117         char *smt_option;
 118
 119         /* Allow the command line to overrule the OF option */
 120         if (smt_enabled_cmdline)
 121                 return;
 122
 123         dn = of_find_node_by_path("/options");
 124
 125         if (dn) {
 126                 smt_option = (char *)get_property(dn, "ibm,smt-enabled", NULL);
 127
 128                 if (smt_option) {
 129                         if (!strcmp(smt_option, "on"))
 130                                 smt_enabled_at_boot = 1;
 131                         else if (!strcmp(smt_option, "off"))
 132                                 smt_enabled_at_boot = 0;
 133                 }
 134         }
 135 }
 136
 137 /* Look for smt-enabled= cmdline option */
 138 static int __init early_smt_enabled(char *p)
 139 {
 140         smt_enabled_cmdline = 1;
 141
 142         if (!p)
 143                 return 0;
 144
 145         if (!strcmp(p, "on") || !strcmp(p, "1"))
 146                 smt_enabled_at_boot = 1;
 147         else if (!strcmp(p, "off") || !strcmp(p, "0"))
 148                 smt_enabled_at_boot = 0;
 149
 150         return 0;
 151 }
 152 early_param("smt-enabled", early_smt_enabled);
 153
 154 #else
 155 #define check_smt_enabled()
 156 #endif /* CONFIG_SMP */
 157
 158 /*
 159  * Early initialization entry point. This is called by head.S
 160  * with MMU translation disabled. We rely on the "feature" of
 161  * the CPU that ignores the top 2 bits of the address in real
 162  * mode so we can access kernel globals normally provided we
 163  * only toy with things in the RMO region. From here, we do
 164  * some early parsing of the device-tree to setup out LMB
 165  * data structures, and allocate & initialize the hash table
 166  * and segment tables so we can start running with translation
 167  * enabled.
 168  *
 169  * It is this function which will call the probe() callback of
 170  * the various platform types and copy the matching one to the
 171  * global ppc_md structure. Your platform can eventually do
 172  * some very early initializations from the probe() routine, but
 173  * this is not recommended, be very careful as, for example, the
 174  * device-tree is not accessible via normal means at this point.
 175  */
 176
 177 void __init early_setup(unsigned long dt_ptr)
 178 {
 179         /* Enable early debugging if any specified (see udbg.h) */
 180         udbg_early_init();
 181
 182         DBG(" -> early_setup(), dt_ptr: 0x%lx\n", dt_ptr);
 183
 184         /*
 185          * Do early initializations using the flattened device
 186          * tree, like retreiving the physical memory map or
 187          * calculating/retreiving the hash table size
 188          */
 189         early_init_devtree(__va(dt_ptr));
 190
 191         /* Now we know the logical id of our boot cpu, setup the paca. */
 192         setup_boot_paca();
 193
 194         /* Fix up paca fields required for the boot cpu */
 195         get_paca()->cpu_start = 1;
 196         get_paca()->stab_real = __pa((u64)&initial_stab);
 197         get_paca()->stab_addr = (u64)&initial_stab;
 198
 199         /* Probe the machine type */
 200         probe_machine();
 201
 202 #ifdef CONFIG_CRASH_DUMP
 203         kdump_setup();
 204 #endif
 205
 206         DBG("Found, Initializing memory management...\n");
 207
 208         /*
 209          * Initialize the MMU Hash table and create the linear mapping
 210          * of memory. Has to be done before stab/slb initialization as
 211          * this is currently where the page size encoding is obtained
 212          */
 213         htab_initialize();
 214
 215         /*
 216          * Initialize stab / SLB management except on iSeries
 217          */
 218         if (cpu_has_feature(CPU_FTR_SLB))
 219                 slb_initialize();
 220         else if (!firmware_has_feature(FW_FEATURE_ISERIES))
 221                 stab_initialize(get_paca()->stab_real);
 222
 223         DBG(" <- early_setup()\n");
 224 }
 225
 226 #ifdef CONFIG_SMP
 227 void early_setup_secondary(void)
 228 {
 229         struct paca_struct *lpaca = get_paca();
 230
 231         /* Mark enabled in PACA */
 232         lpaca->proc_enabled = 0;
 233
 234         /* Initialize hash table for that CPU */
 235         htab_initialize_secondary();
 236
 237         /* Initialize STAB/SLB. We use a virtual address as it works
 238          * in real mode on pSeries and we want a virutal address on
 239          * iSeries anyway
 240          */
 241         if (cpu_has_feature(CPU_FTR_SLB))
 242                 slb_initialize();
 243         else
 244                 stab_initialize(lpaca->stab_addr);
 245 }
 246
 247 #endif /* CONFIG_SMP */
 248
 249 #if defined(CONFIG_SMP) || defined(CONFIG_KEXEC)
 250 void smp_release_cpus(void)
 251 {
 252         extern unsigned long __secondary_hold_spinloop;
 253         unsigned long *ptr;
 254
 255         DBG(" -> smp_release_cpus()\n");
 256
 257         /* All secondary cpus are spinning on a common spinloop, release them
 258          * all now so they can start to spin on their individual paca
 259          * spinloops. For non SMP kernels, the secondary cpus never get out
 260          * of the common spinloop.
 261          * This is useless but harmless on iSeries, secondaries are already
 262          * waiting on their paca spinloops. */
 263
 264         ptr  = (unsigned long *)((unsigned long)&__secondary_hold_spinloop
 265                         - PHYSICAL_START);
 266         *ptr = 1;
 267         mb();
 268
 269         DBG(" <- smp_release_cpus()\n");
 270 }
 271 #endif /* CONFIG_SMP || CONFIG_KEXEC */
 272
 273 /*
 274  * Initialize some remaining members of the ppc64_caches and systemcfg
 275  * structures
 276  * (at least until we get rid of them completely). This is mostly some
 277  * cache informations about the CPU that will be used by cache flush
 278  * routines and/or provided to userland
 279  */
 280 static void __init initialize_cache_info(void)
 281 {
 282         struct device_node *np;
 283         unsigned long num_cpus = 0;
 284
 285         DBG(" -> initialize_cache_info()\n");
 286
 287         for (np = NULL; (np = of_find_node_by_type(np, "cpu"));) {
 288                 num_cpus += 1;
 289
 290                 /* We're assuming *all* of the CPUs have the same
 291                  * d-cache and i-cache sizes... -Peter
 292                  */
 293
 294                 if ( num_cpus == 1 ) {
 295                         u32 *sizep, *lsizep;
 296                         u32 size, lsize;
 297                         const char *dc, *ic;
 298
 299                         /* Then read cache informations */
 300                         if (machine_is(powermac)) {
 301                                 dc = "d-cache-block-size";
 302                                 ic = "i-cache-block-size";
 303                         } else {
 304                                 dc = "d-cache-line-size";
 305                                 ic = "i-cache-line-size";
 306                         }
 307
 308                         size = 0;
 309                         lsize = cur_cpu_spec->dcache_bsize;
 310                         sizep = (u32 *)get_property(np, "d-cache-size", NULL);
 311                         if (sizep != NULL)
 312                                 size = *sizep;
 313                         lsizep = (u32 *) get_property(np, dc, NULL);
 314                         if (lsizep != NULL)
 315                                 lsize = *lsizep;
 316                         if (sizep == 0 || lsizep == 0)
 317                                 DBG("Argh, can't find dcache properties ! "
 318                                     "sizep: %p, lsizep: %p\n", sizep, lsizep);
 319
 320                         ppc64_caches.dsize = size;
 321                         ppc64_caches.dline_size = lsize;
 322                         ppc64_caches.log_dline_size = __ilog2(lsize);
 323                         ppc64_caches.dlines_per_page = PAGE_SIZE / lsize;
 324
 325                         size = 0;
 326                         lsize = cur_cpu_spec->icache_bsize;
 327                         sizep = (u32 *)get_property(np, "i-cache-size", NULL);
 328                         if (sizep != NULL)
 329                                 size = *sizep;
 330                         lsizep = (u32 *)get_property(np, ic, NULL);
 331                         if (lsizep != NULL)
 332                                 lsize = *lsizep;
 333                         if (sizep == 0 || lsizep == 0)
 334                                 DBG("Argh, can't find icache properties ! "
 335                                     "sizep: %p, lsizep: %p\n", sizep, lsizep);
 336
 337                         ppc64_caches.isize = size;
 338                         ppc64_caches.iline_size = lsize;
 339                         ppc64_caches.log_iline_size = __ilog2(lsize);
 340                         ppc64_caches.ilines_per_page = PAGE_SIZE / lsize;
 341                 }
 342         }
 343
 344         DBG(" <- initialize_cache_info()\n");
 345 }
 346
 347
 348 /*
 349  * Do some initial setup of the system.  The parameters are those which
 350  * were passed in from the bootloader.
 351  */
 352 void __init setup_system(void)
 353 {
 354         DBG(" -> setup_system()\n");
 355
 356 #ifdef CONFIG_KEXEC
 357         kdump_move_device_tree();
 358 #endif
 359         /*
 360          * Unflatten the device-tree passed by prom_init or kexec
 361          */
 362         unflatten_device_tree();
 363
 364 #ifdef CONFIG_KEXEC
 365         kexec_setup();  /* requires unflattened device tree. */
 366 #endif
 367
 368         /*
 369          * Fill the ppc64_caches & systemcfg structures with informations
 370          * retrieved from the device-tree. Need to be called before
 371          * finish_device_tree() since the later requires some of the
 372          * informations filled up here to properly parse the interrupt
 373          * tree.
 374          * It also sets up the cache line sizes which allows to call
 375          * routines like flush_icache_range (used by the hash init
 376          * later on).
 377          */
 378         initialize_cache_info();
 379
 380 #ifdef CONFIG_PPC_RTAS
 381         /*
 382          * Initialize RTAS if available
 383          */
 384         rtas_initialize();
 385 #endif /* CONFIG_PPC_RTAS */
 386
 387         /*
 388          * Check if we have an initrd provided via the device-tree
 389          */
 390         check_for_initrd();
 391
 392         /*
 393          * Do some platform specific early initializations, that includes
 394          * setting up the hash table pointers. It also sets up some interrupt-mapping
 395          * related options that will be used by finish_device_tree()
 396          */
 397         ppc_md.init_early();
 398
 399         /*
 400          * We can discover serial ports now since the above did setup the
 401          * hash table management for us, thus ioremap works. We do that early
 402          * so that further code can be debugged
 403          */
 404         find_legacy_serial_ports();
 405
 406         /*
 407          * "Finish" the device-tree, that is do the actual parsing of
 408          * some of the properties like the interrupt map
 409          */
 410         finish_device_tree();
 411
 412         /*
 413          * Initialize xmon
 414          */
 415 #ifdef CONFIG_XMON_DEFAULT
 416         xmon_init(1);
 417 #endif
 418         /*
 419          * Register early console
 420          */
 421         register_early_udbg_console();
 422
 423         /* Save unparsed command line copy for /proc/cmdline */
 424         strlcpy(saved_command_line, cmd_line, COMMAND_LINE_SIZE);
 425
 426         parse_early_param();
 427
 428         check_smt_enabled();
 429         smp_setup_cpu_maps();
 430
 431 #ifdef CONFIG_SMP
 432         /* Release secondary cpus out of their spinloops at 0x60 now that
 433          * we can map physical -> logical CPU ids
 434          */
 435         smp_release_cpus();
 436 #endif
 437
 438         printk("Starting Linux PPC64 %s\n", system_utsname.version);
 439
 440         printk("-----------------------------------------------------\n");
 441         printk("ppc64_pft_size                = 0x%lx\n", ppc64_pft_size);
 442         printk("ppc64_interrupt_controller    = 0x%ld\n",
 443                ppc64_interrupt_controller);
 444         printk("physicalMemorySize            = 0x%lx\n", lmb_phys_mem_size());
 445         printk("ppc64_caches.dcache_line_size = 0x%x\n",
 446                ppc64_caches.dline_size);
 447         printk("ppc64_caches.icache_line_size = 0x%x\n",
 448                ppc64_caches.iline_size);
 449         printk("htab_address                  = 0x%p\n", htab_address);
 450         printk("htab_hash_mask                = 0x%lx\n", htab_hash_mask);
 451 #if PHYSICAL_START > 0
 452         printk("physical_start                = 0x%x\n", PHYSICAL_START);
 453 #endif
 454         printk("-----------------------------------------------------\n");
 455
 456         DBG(" <- setup_system()\n");
 457 }
 458
 459 static int ppc64_panic_event(struct notifier_block *this,
 460                              unsigned long event, void *ptr)
 461 {
 462         ppc_md.panic((char *)ptr);  /* May not return */
 463         return NOTIFY_DONE;
 464 }
 465
 466 #ifdef CONFIG_IRQSTACKS
 467 static void __init irqstack_early_init(void)
 468 {
 469         unsigned int i;
 470
 471         /*
 472          * interrupt stacks must be under 256MB, we cannot afford to take
 473          * SLB misses on them.
 474          */
 475         for_each_possible_cpu(i) {
 476                 softirq_ctx[i] = (struct thread_info *)
 477                         __va(lmb_alloc_base(THREAD_SIZE,
 478                                             THREAD_SIZE, 0x10000000));
 479                 hardirq_ctx[i] = (struct thread_info *)
 480                         __va(lmb_alloc_base(THREAD_SIZE,
 481                                             THREAD_SIZE, 0x10000000));
 482         }
 483 }
 484 #else
 485 #define irqstack_early_init()
 486 #endif
 487
 488 /*
 489  * Stack space used when we detect a bad kernel stack pointer, and
 490  * early in SMP boots before relocation is enabled.
 491  */
 492 static void __init emergency_stack_init(void)
 493 {
 494         unsigned long limit;
 495         unsigned int i;
 496
 497         /*
 498          * Emergency stacks must be under 256MB, we cannot afford to take
 499          * SLB misses on them. The ABI also requires them to be 128-byte
 500          * aligned.
 501          *
 502          * Since we use these as temporary stacks during secondary CPU
 503          * bringup, we need to get at them in real mode. This means they
 504          * must also be within the RMO region.
 505          */
 506         limit = min(0x10000000UL, lmb.rmo_size);
 507
 508         for_each_possible_cpu(i)
 509                 paca[i].emergency_sp =
 510                 __va(lmb_alloc_base(HW_PAGE_SIZE, 128, limit)) + HW_PAGE_SIZE;
 511 }
 512
 513 /*
 514  * Called into from start_kernel, after lock_kernel has been called.
 515  * Initializes bootmem, which is unsed to manage page allocation until
 516  * mem_init is called.
 517  */
 518 void __init setup_arch(char **cmdline_p)
 519 {
 520         extern void do_init_bootmem(void);
 521
 522         ppc64_boot_msg(0x12, "Setup Arch");
 523
 524         *cmdline_p = cmd_line;
 525
 526         /*
 527          * Set cache line size based on type of cpu as a default.
 528          * Systems with OF can look in the properties on the cpu node(s)
 529          * for a possibly more accurate value.
 530          */
 531         dcache_bsize = ppc64_caches.dline_size;
 532         icache_bsize = ppc64_caches.iline_size;
 533
 534         /* reboot on panic */
 535         panic_timeout = 180;
 536
 537         if (ppc_md.panic)
 538                 atomic_notifier_chain_register(&panic_notifier_list,
 539                                 &ppc64_panic_block);
 540
 541         init_mm.start_code = PAGE_OFFSET;
 542         init_mm.end_code = (unsigned long) _etext;
 543         init_mm.end_data = (unsigned long) _edata;
 544         init_mm.brk = klimit;
 545
 546         irqstack_early_init();
 547         emergency_stack_init();
 548
 549         stabs_alloc();
 550
 551         /* set up the bootmem stuff with available memory */
 552         do_init_bootmem();
 553         sparse_init();
 554
 555 #ifdef CONFIG_DUMMY_CONSOLE
 556         conswitchp = &dummy_con;
 557 #endif
 558
 559         ppc_md.setup_arch();
 560
 561         paging_init();
 562         ppc64_boot_msg(0x15, "Setup Done");
 563 }
 564
 565
 566 /* ToDo: do something useful if ppc_md is not yet setup. */
 567 #define PPC64_LINUX_FUNCTION 0x0f000000
 568 #define PPC64_IPL_MESSAGE 0xc0000000
 569 #define PPC64_TERM_MESSAGE 0xb0000000
 570
 571 static void ppc64_do_msg(unsigned int src, const char *msg)
 572 {
 573         if (ppc_md.progress) {
 574                 char buf[128];
 575
 576                 sprintf(buf, "%08X\n", src);
 577                 ppc_md.progress(buf, 0);
 578                 snprintf(buf, 128, "%s", msg);
 579                 ppc_md.progress(buf, 0);
 580         }
 581 }
 582
 583 /* Print a boot progress message. */
 584 void ppc64_boot_msg(unsigned int src, const char *msg)
 585 {
 586         ppc64_do_msg(PPC64_LINUX_FUNCTION|PPC64_IPL_MESSAGE|src, msg);
 587         printk("[boot]%04x %s\n", src, msg);
 588 }
 589
 590 /* Print a termination message (print only -- does not stop the kernel) */
 591 void ppc64_terminate_msg(unsigned int src, const char *msg)
 592 {
 593         ppc64_do_msg(PPC64_LINUX_FUNCTION|PPC64_TERM_MESSAGE|src, msg);
 594         printk("[terminate]%04x %s\n", src, msg);
 595 }
 596
 597 int check_legacy_ioport(unsigned long base_port)
 598 {
 599         if (ppc_md.check_legacy_ioport == NULL)
 600                 return 0;
 601         return ppc_md.check_legacy_ioport(base_port);
 602 }
 603 EXPORT_SYMBOL(check_legacy_ioport);
 604
 605 void cpu_die(void)
 606 {
 607         if (ppc_md.cpu_die)
 608                 ppc_md.cpu_die();
 609 }
 610
 611 #ifdef CONFIG_SMP
 612 void __init setup_per_cpu_areas(void)
 613 {
 614         int i;
 615         unsigned long size;
 616         char *ptr;
 617
 618         /* Copy section for each CPU (we discard the original) */
 619         size = ALIGN(__per_cpu_end - __per_cpu_start, SMP_CACHE_BYTES);
 620 #ifdef CONFIG_MODULES
 621         if (size < PERCPU_ENOUGH_ROOM)
 622                 size = PERCPU_ENOUGH_ROOM;
 623 #endif
 624
 625         for_each_possible_cpu(i) {
 626                 ptr = alloc_bootmem_node(NODE_DATA(cpu_to_node(i)), size);
 627                 if (!ptr)
 628                         panic("Cannot allocate cpu data for CPU %d\n", i);
 629
 630                 paca[i].data_offset = ptr - __per_cpu_start;
 631                 memcpy(ptr, __per_cpu_start, __per_cpu_end - __per_cpu_start);
 632         }
 633 }
 634 #endif