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 <asm/io.h>
  37 #include <asm/kdump.h>
  38 #include <asm/prom.h>
  39 #include <asm/processor.h>
  40 #include <asm/pgtable.h>
  41 #include <asm/smp.h>
  42 #include <asm/elf.h>
  43 #include <asm/machdep.h>
  44 #include <asm/paca.h>
  45 #include <asm/time.h>
  46 #include <asm/cputable.h>
  47 #include <asm/sections.h>
  48 #include <asm/btext.h>
  49 #include <asm/nvram.h>
  50 #include <asm/setup.h>
  51 #include <asm/system.h>
  52 #include <asm/rtas.h>
  53 #include <asm/iommu.h>
  54 #include <asm/serial.h>
  55 #include <asm/cache.h>
  56 #include <asm/page.h>
  57 #include <asm/mmu.h>
  58 #include <asm/lmb.h>
  59 #include <asm/iseries/it_lp_naca.h>
  60 #include <asm/firmware.h>
  61 #include <asm/xmon.h>
  62 #include <asm/udbg.h>
  63 #include <asm/kexec.h>
  64
  65 #include "setup.h"
  66
  67 #ifdef DEBUG
  68 #define DBG(fmt...) udbg_printf(fmt)
  69 #else
  70 #define DBG(fmt...)
  71 #endif
  72
  73 /*
  74  * Here are some early debugging facilities. You can enable one
  75  * but your kernel will not boot on anything else if you do so
  76  */
  77
  78 /* This one is for use on LPAR machines that support an HVC console
  79  * on vterm 0
  80  */
  81 extern void udbg_init_debug_lpar(void);
  82 /* This one is for use on Apple G5 machines
  83  */
  84 extern void udbg_init_pmac_realmode(void);
  85 /* That's RTAS panel debug */
  86 extern void call_rtas_display_status_delay(unsigned char c);
  87 /* Here's maple real mode debug */
  88 extern void udbg_init_maple_realmode(void);
  89
  90 #define EARLY_DEBUG_INIT() do {} while(0)
  91
  92 #if 0
  93 #define EARLY_DEBUG_INIT() udbg_init_debug_lpar()
  94 #define EARLY_DEBUG_INIT() udbg_init_maple_realmode()
  95 #define EARLY_DEBUG_INIT() udbg_init_pmac_realmode()
  96 #define EARLY_DEBUG_INIT()                                              \
  97         do { udbg_putc = call_rtas_display_status_delay; } while(0)
  98 #endif
  99
 100 int have_of = 1;
 101 int boot_cpuid = 0;
 102 int boot_cpuid_phys = 0;
 103 dev_t boot_dev;
 104 u64 ppc64_pft_size;
 105
 106 /* Pick defaults since we might want to patch instructions
 107  * before we've read this from the device tree.
 108  */
 109 struct ppc64_caches ppc64_caches = {
 110         .dline_size = 0x80,
 111         .log_dline_size = 7,
 112         .iline_size = 0x80,
 113         .log_iline_size = 7
 114 };
 115 EXPORT_SYMBOL_GPL(ppc64_caches);
 116
 117 /*
 118  * These are used in binfmt_elf.c to put aux entries on the stack
 119  * for each elf executable being started.
 120  */
 121 int dcache_bsize;
 122 int icache_bsize;
 123 int ucache_bsize;
 124
 125 /* The main machine-dep calls structure
 126  */
 127 struct machdep_calls ppc_md;
 128 EXPORT_SYMBOL(ppc_md);
 129
 130 #ifdef CONFIG_MAGIC_SYSRQ
 131 unsigned long SYSRQ_KEY;
 132 #endif /* CONFIG_MAGIC_SYSRQ */
 133
 134
 135 static int ppc64_panic_event(struct notifier_block *, unsigned long, void *);
 136 static struct notifier_block ppc64_panic_block = {
 137         .notifier_call = ppc64_panic_event,
 138         .priority = INT_MIN /* may not return; must be done last */
 139 };
 140
 141 #ifdef CONFIG_SMP
 142
 143 static int smt_enabled_cmdline;
 144
 145 /* Look for ibm,smt-enabled OF option */
 146 static void check_smt_enabled(void)
 147 {
 148         struct device_node *dn;
 149         char *smt_option;
 150
 151         /* Allow the command line to overrule the OF option */
 152         if (smt_enabled_cmdline)
 153                 return;
 154
 155         dn = of_find_node_by_path("/options");
 156
 157         if (dn) {
 158                 smt_option = (char *)get_property(dn, "ibm,smt-enabled", NULL);
 159
 160                 if (smt_option) {
 161                         if (!strcmp(smt_option, "on"))
 162                                 smt_enabled_at_boot = 1;
 163                         else if (!strcmp(smt_option, "off"))
 164                                 smt_enabled_at_boot = 0;
 165                 }
 166         }
 167 }
 168
 169 /* Look for smt-enabled= cmdline option */
 170 static int __init early_smt_enabled(char *p)
 171 {
 172         smt_enabled_cmdline = 1;
 173
 174         if (!p)
 175                 return 0;
 176
 177         if (!strcmp(p, "on") || !strcmp(p, "1"))
 178                 smt_enabled_at_boot = 1;
 179         else if (!strcmp(p, "off") || !strcmp(p, "0"))
 180                 smt_enabled_at_boot = 0;
 181
 182         return 0;
 183 }
 184 early_param("smt-enabled", early_smt_enabled);
 185
 186 #else
 187 #define check_smt_enabled()
 188 #endif /* CONFIG_SMP */
 189
 190 extern struct machdep_calls pSeries_md;
 191 extern struct machdep_calls pmac_md;
 192 extern struct machdep_calls maple_md;
 193 extern struct machdep_calls cell_md;
 194 extern struct machdep_calls iseries_md;
 195
 196 /* Ultimately, stuff them in an elf section like initcalls... */
 197 static struct machdep_calls __initdata *machines[] = {
 198 #ifdef CONFIG_PPC_PSERIES
 199         &pSeries_md,
 200 #endif /* CONFIG_PPC_PSERIES */
 201 #ifdef CONFIG_PPC_PMAC
 202         &pmac_md,
 203 #endif /* CONFIG_PPC_PMAC */
 204 #ifdef CONFIG_PPC_MAPLE
 205         &maple_md,
 206 #endif /* CONFIG_PPC_MAPLE */
 207 #ifdef CONFIG_PPC_CELL
 208         &cell_md,
 209 #endif
 210 #ifdef CONFIG_PPC_ISERIES
 211         &iseries_md,
 212 #endif
 213         NULL
 214 };
 215
 216 /*
 217  * Early initialization entry point. This is called by head.S
 218  * with MMU translation disabled. We rely on the "feature" of
 219  * the CPU that ignores the top 2 bits of the address in real
 220  * mode so we can access kernel globals normally provided we
 221  * only toy with things in the RMO region. From here, we do
 222  * some early parsing of the device-tree to setup out LMB
 223  * data structures, and allocate & initialize the hash table
 224  * and segment tables so we can start running with translation
 225  * enabled.
 226  *
 227  * It is this function which will call the probe() callback of
 228  * the various platform types and copy the matching one to the
 229  * global ppc_md structure. Your platform can eventually do
 230  * some very early initializations from the probe() routine, but
 231  * this is not recommended, be very careful as, for example, the
 232  * device-tree is not accessible via normal means at this point.
 233  */
 234
 235 void __init early_setup(unsigned long dt_ptr)
 236 {
 237         struct paca_struct *lpaca = get_paca();
 238         static struct machdep_calls **mach;
 239
 240         /*
 241          * Enable early debugging if any specified (see top of
 242          * this file)
 243          */
 244         EARLY_DEBUG_INIT();
 245
 246         DBG(" -> early_setup()\n");
 247
 248         /*
 249          * Do early initializations using the flattened device
 250          * tree, like retreiving the physical memory map or
 251          * calculating/retreiving the hash table size
 252          */
 253         early_init_devtree(__va(dt_ptr));
 254
 255         /*
 256          * Iterate all ppc_md structures until we find the proper
 257          * one for the current machine type
 258          */
 259         DBG("Probing machine type for platform %x...\n", _machine);
 260
 261         for (mach = machines; *mach; mach++) {
 262                 if ((*mach)->probe(_machine))
 263                         break;
 264         }
 265         /* What can we do if we didn't find ? */
 266         if (*mach == NULL) {
 267                 DBG("No suitable machine found !\n");
 268                 for (;;);
 269         }
 270         ppc_md = **mach;
 271
 272 #ifdef CONFIG_CRASH_DUMP
 273         kdump_setup();
 274 #endif
 275
 276         DBG("Found, Initializing memory management...\n");
 277
 278         /*
 279          * Initialize the MMU Hash table and create the linear mapping
 280          * of memory. Has to be done before stab/slb initialization as
 281          * this is currently where the page size encoding is obtained
 282          */
 283         htab_initialize();
 284
 285         /*
 286          * Initialize stab / SLB management except on iSeries
 287          */
 288         if (!firmware_has_feature(FW_FEATURE_ISERIES)) {
 289                 if (cpu_has_feature(CPU_FTR_SLB))
 290                         slb_initialize();
 291                 else
 292                         stab_initialize(lpaca->stab_real);
 293         }
 294
 295         DBG(" <- early_setup()\n");
 296 }
 297
 298 #ifdef CONFIG_SMP
 299 void early_setup_secondary(void)
 300 {
 301         struct paca_struct *lpaca = get_paca();
 302
 303         /* Mark enabled in PACA */
 304         lpaca->proc_enabled = 0;
 305
 306         /* Initialize hash table for that CPU */
 307         htab_initialize_secondary();
 308
 309         /* Initialize STAB/SLB. We use a virtual address as it works
 310          * in real mode on pSeries and we want a virutal address on
 311          * iSeries anyway
 312          */
 313         if (cpu_has_feature(CPU_FTR_SLB))
 314                 slb_initialize();
 315         else
 316                 stab_initialize(lpaca->stab_addr);
 317 }
 318
 319 #endif /* CONFIG_SMP */
 320
 321 #if defined(CONFIG_SMP) || defined(CONFIG_KEXEC)
 322 void smp_release_cpus(void)
 323 {
 324         extern unsigned long __secondary_hold_spinloop;
 325         unsigned long *ptr;
 326
 327         DBG(" -> smp_release_cpus()\n");
 328
 329         /* All secondary cpus are spinning on a common spinloop, release them
 330          * all now so they can start to spin on their individual paca
 331          * spinloops. For non SMP kernels, the secondary cpus never get out
 332          * of the common spinloop.
 333          * This is useless but harmless on iSeries, secondaries are already
 334          * waiting on their paca spinloops. */
 335
 336         ptr  = (unsigned long *)((unsigned long)&__secondary_hold_spinloop
 337                         - PHYSICAL_START);
 338         *ptr = 1;
 339         mb();
 340
 341         DBG(" <- smp_release_cpus()\n");
 342 }
 343 #else
 344 #define smp_release_cpus()
 345 #endif /* CONFIG_SMP || CONFIG_KEXEC */
 346
 347 /*
 348  * Initialize some remaining members of the ppc64_caches and systemcfg
 349  * structures
 350  * (at least until we get rid of them completely). This is mostly some
 351  * cache informations about the CPU that will be used by cache flush
 352  * routines and/or provided to userland
 353  */
 354 static void __init initialize_cache_info(void)
 355 {
 356         struct device_node *np;
 357         unsigned long num_cpus = 0;
 358
 359         DBG(" -> initialize_cache_info()\n");
 360
 361         for (np = NULL; (np = of_find_node_by_type(np, "cpu"));) {
 362                 num_cpus += 1;
 363
 364                 /* We're assuming *all* of the CPUs have the same
 365                  * d-cache and i-cache sizes... -Peter
 366                  */
 367
 368                 if ( num_cpus == 1 ) {
 369                         u32 *sizep, *lsizep;
 370                         u32 size, lsize;
 371                         const char *dc, *ic;
 372
 373                         /* Then read cache informations */
 374                         if (_machine == PLATFORM_POWERMAC) {
 375                                 dc = "d-cache-block-size";
 376                                 ic = "i-cache-block-size";
 377                         } else {
 378                                 dc = "d-cache-line-size";
 379                                 ic = "i-cache-line-size";
 380                         }
 381
 382                         size = 0;
 383                         lsize = cur_cpu_spec->dcache_bsize;
 384                         sizep = (u32 *)get_property(np, "d-cache-size", NULL);
 385                         if (sizep != NULL)
 386                                 size = *sizep;
 387                         lsizep = (u32 *) get_property(np, dc, NULL);
 388                         if (lsizep != NULL)
 389                                 lsize = *lsizep;
 390                         if (sizep == 0 || lsizep == 0)
 391                                 DBG("Argh, can't find dcache properties ! "
 392                                     "sizep: %p, lsizep: %p\n", sizep, lsizep);
 393
 394                         ppc64_caches.dsize = size;
 395                         ppc64_caches.dline_size = lsize;
 396                         ppc64_caches.log_dline_size = __ilog2(lsize);
 397                         ppc64_caches.dlines_per_page = PAGE_SIZE / lsize;
 398
 399                         size = 0;
 400                         lsize = cur_cpu_spec->icache_bsize;
 401                         sizep = (u32 *)get_property(np, "i-cache-size", NULL);
 402                         if (sizep != NULL)
 403                                 size = *sizep;
 404                         lsizep = (u32 *)get_property(np, ic, NULL);
 405                         if (lsizep != NULL)
 406                                 lsize = *lsizep;
 407                         if (sizep == 0 || lsizep == 0)
 408                                 DBG("Argh, can't find icache properties ! "
 409                                     "sizep: %p, lsizep: %p\n", sizep, lsizep);
 410
 411                         ppc64_caches.isize = size;
 412                         ppc64_caches.iline_size = lsize;
 413                         ppc64_caches.log_iline_size = __ilog2(lsize);
 414                         ppc64_caches.ilines_per_page = PAGE_SIZE / lsize;
 415                 }
 416         }
 417
 418         DBG(" <- initialize_cache_info()\n");
 419 }
 420
 421
 422 /*
 423  * Do some initial setup of the system.  The parameters are those which
 424  * were passed in from the bootloader.
 425  */
 426 void __init setup_system(void)
 427 {
 428         DBG(" -> setup_system()\n");
 429
 430         /*
 431          * Unflatten the device-tree passed by prom_init or kexec
 432          */
 433         unflatten_device_tree();
 434
 435 #ifdef CONFIG_KEXEC
 436         kexec_setup();  /* requires unflattened device tree. */
 437 #endif
 438
 439         /*
 440          * Fill the ppc64_caches & systemcfg structures with informations
 441          * retrieved from the device-tree. Need to be called before
 442          * finish_device_tree() since the later requires some of the
 443          * informations filled up here to properly parse the interrupt
 444          * tree.
 445          * It also sets up the cache line sizes which allows to call
 446          * routines like flush_icache_range (used by the hash init
 447          * later on).
 448          */
 449         initialize_cache_info();
 450
 451 #ifdef CONFIG_PPC_RTAS
 452         /*
 453          * Initialize RTAS if available
 454          */
 455         rtas_initialize();
 456 #endif /* CONFIG_PPC_RTAS */
 457
 458         /*
 459          * Check if we have an initrd provided via the device-tree
 460          */
 461         check_for_initrd();
 462
 463         /*
 464          * Do some platform specific early initializations, that includes
 465          * setting up the hash table pointers. It also sets up some interrupt-mapping
 466          * related options that will be used by finish_device_tree()
 467          */
 468         ppc_md.init_early();
 469
 470         /*
 471          * We can discover serial ports now since the above did setup the
 472          * hash table management for us, thus ioremap works. We do that early
 473          * so that further code can be debugged
 474          */
 475 #ifdef CONFIG_SERIAL_8250
 476         find_legacy_serial_ports();
 477 #endif
 478
 479         /*
 480          * "Finish" the device-tree, that is do the actual parsing of
 481          * some of the properties like the interrupt map
 482          */
 483         finish_device_tree();
 484
 485         /*
 486          * Initialize xmon
 487          */
 488 #ifdef CONFIG_XMON_DEFAULT
 489         xmon_init(1);
 490 #endif
 491         /*
 492          * Register early console
 493          */
 494         register_early_udbg_console();
 495
 496         /* Save unparsed command line copy for /proc/cmdline */
 497         strlcpy(saved_command_line, cmd_line, COMMAND_LINE_SIZE);
 498
 499         parse_early_param();
 500
 501         check_smt_enabled();
 502         smp_setup_cpu_maps();
 503
 504         /* Release secondary cpus out of their spinloops at 0x60 now that
 505          * we can map physical -> logical CPU ids
 506          */
 507         smp_release_cpus();
 508
 509         printk("Starting Linux PPC64 %s\n", system_utsname.version);
 510
 511         printk("-----------------------------------------------------\n");
 512         printk("ppc64_pft_size                = 0x%lx\n", ppc64_pft_size);
 513         printk("ppc64_interrupt_controller    = 0x%ld\n",
 514                ppc64_interrupt_controller);
 515         printk("platform                      = 0x%x\n", _machine);
 516         printk("physicalMemorySize            = 0x%lx\n", lmb_phys_mem_size());
 517         printk("ppc64_caches.dcache_line_size = 0x%x\n",
 518                ppc64_caches.dline_size);
 519         printk("ppc64_caches.icache_line_size = 0x%x\n",
 520                ppc64_caches.iline_size);
 521         printk("htab_address                  = 0x%p\n", htab_address);
 522         printk("htab_hash_mask                = 0x%lx\n", htab_hash_mask);
 523 #if PHYSICAL_START > 0
 524         printk("physical_start                = 0x%x\n", PHYSICAL_START);
 525 #endif
 526         printk("-----------------------------------------------------\n");
 527
 528         mm_init_ppc64();
 529
 530         DBG(" <- setup_system()\n");
 531 }
 532
 533 static int ppc64_panic_event(struct notifier_block *this,
 534                              unsigned long event, void *ptr)
 535 {
 536         ppc_md.panic((char *)ptr);  /* May not return */
 537         return NOTIFY_DONE;
 538 }
 539
 540 #ifdef CONFIG_IRQSTACKS
 541 static void __init irqstack_early_init(void)
 542 {
 543         unsigned int i;
 544
 545         /*
 546          * interrupt stacks must be under 256MB, we cannot afford to take
 547          * SLB misses on them.
 548          */
 549         for_each_cpu(i) {
 550                 softirq_ctx[i] = (struct thread_info *)
 551                         __va(lmb_alloc_base(THREAD_SIZE,
 552                                             THREAD_SIZE, 0x10000000));
 553                 hardirq_ctx[i] = (struct thread_info *)
 554                         __va(lmb_alloc_base(THREAD_SIZE,
 555                                             THREAD_SIZE, 0x10000000));
 556         }
 557 }
 558 #else
 559 #define irqstack_early_init()
 560 #endif
 561
 562 /*
 563  * Stack space used when we detect a bad kernel stack pointer, and
 564  * early in SMP boots before relocation is enabled.
 565  */
 566 static void __init emergency_stack_init(void)
 567 {
 568         unsigned long limit;
 569         unsigned int i;
 570
 571         /*
 572          * Emergency stacks must be under 256MB, we cannot afford to take
 573          * SLB misses on them. The ABI also requires them to be 128-byte
 574          * aligned.
 575          *
 576          * Since we use these as temporary stacks during secondary CPU
 577          * bringup, we need to get at them in real mode. This means they
 578          * must also be within the RMO region.
 579          */
 580         limit = min(0x10000000UL, lmb.rmo_size);
 581
 582         for_each_cpu(i)
 583                 paca[i].emergency_sp =
 584                 __va(lmb_alloc_base(HW_PAGE_SIZE, 128, limit)) + HW_PAGE_SIZE;
 585 }
 586
 587 /*
 588  * Called into from start_kernel, after lock_kernel has been called.
 589  * Initializes bootmem, which is unsed to manage page allocation until
 590  * mem_init is called.
 591  */
 592 void __init setup_arch(char **cmdline_p)
 593 {
 594         extern void do_init_bootmem(void);
 595
 596         ppc64_boot_msg(0x12, "Setup Arch");
 597
 598         *cmdline_p = cmd_line;
 599
 600         /*
 601          * Set cache line size based on type of cpu as a default.
 602          * Systems with OF can look in the properties on the cpu node(s)
 603          * for a possibly more accurate value.
 604          */
 605         dcache_bsize = ppc64_caches.dline_size;
 606         icache_bsize = ppc64_caches.iline_size;
 607
 608         /* reboot on panic */
 609         panic_timeout = 180;
 610
 611         if (ppc_md.panic)
 612                 notifier_chain_register(&panic_notifier_list, &ppc64_panic_block);
 613
 614         init_mm.start_code = PAGE_OFFSET;
 615         init_mm.end_code = (unsigned long) _etext;
 616         init_mm.end_data = (unsigned long) _edata;
 617         init_mm.brk = klimit;
 618
 619         irqstack_early_init();
 620         emergency_stack_init();
 621
 622         stabs_alloc();
 623
 624         /* set up the bootmem stuff with available memory */
 625         do_init_bootmem();
 626         sparse_init();
 627
 628 #ifdef CONFIG_DUMMY_CONSOLE
 629         conswitchp = &dummy_con;
 630 #endif
 631
 632         ppc_md.setup_arch();
 633
 634         /* Use the default idle loop if the platform hasn't provided one. */
 635         if (NULL == ppc_md.idle_loop) {
 636                 ppc_md.idle_loop = default_idle;
 637                 printk(KERN_INFO "Using default idle loop\n");
 638         }
 639
 640         paging_init();
 641         ppc64_boot_msg(0x15, "Setup Done");
 642 }
 643
 644
 645 /* ToDo: do something useful if ppc_md is not yet setup. */
 646 #define PPC64_LINUX_FUNCTION 0x0f000000
 647 #define PPC64_IPL_MESSAGE 0xc0000000
 648 #define PPC64_TERM_MESSAGE 0xb0000000
 649
 650 static void ppc64_do_msg(unsigned int src, const char *msg)
 651 {
 652         if (ppc_md.progress) {
 653                 char buf[128];
 654
 655                 sprintf(buf, "%08X\n", src);
 656                 ppc_md.progress(buf, 0);
 657                 snprintf(buf, 128, "%s", msg);
 658                 ppc_md.progress(buf, 0);
 659         }
 660 }
 661
 662 /* Print a boot progress message. */
 663 void ppc64_boot_msg(unsigned int src, const char *msg)
 664 {
 665         ppc64_do_msg(PPC64_LINUX_FUNCTION|PPC64_IPL_MESSAGE|src, msg);
 666         printk("[boot]%04x %s\n", src, msg);
 667 }
 668
 669 /* Print a termination message (print only -- does not stop the kernel) */
 670 void ppc64_terminate_msg(unsigned int src, const char *msg)
 671 {
 672         ppc64_do_msg(PPC64_LINUX_FUNCTION|PPC64_TERM_MESSAGE|src, msg);
 673         printk("[terminate]%04x %s\n", src, msg);
 674 }
 675
 676 int check_legacy_ioport(unsigned long base_port)
 677 {
 678         if (ppc_md.check_legacy_ioport == NULL)
 679                 return 0;
 680         return ppc_md.check_legacy_ioport(base_port);
 681 }
 682 EXPORT_SYMBOL(check_legacy_ioport);
 683
 684 void cpu_die(void)
 685 {
 686         if (ppc_md.cpu_die)
 687                 ppc_md.cpu_die();
 688 }