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/module.h>
  16 #include <linux/string.h>
  17 #include <linux/sched.h>
  18 #include <linux/init.h>
  19 #include <linux/kernel.h>
  20 #include <linux/reboot.h>
  21 #include <linux/delay.h>
  22 #include <linux/initrd.h>
  23 #include <linux/seq_file.h>
  24 #include <linux/ioport.h>
  25 #include <linux/console.h>
  26 #include <linux/utsname.h>
  27 #include <linux/tty.h>
  28 #include <linux/root_dev.h>
  29 #include <linux/notifier.h>
  30 #include <linux/cpu.h>
  31 #include <linux/unistd.h>
  32 #include <linux/serial.h>
  33 #include <linux/serial_8250.h>
  34 #include <linux/bootmem.h>
  35 #include <linux/pci.h>
  36 #include <linux/lockdep.h>
  37 #include <linux/lmb.h>
  38 #include <asm/io.h>
  39 #include <asm/kdump.h>
  40 #include <asm/prom.h>
  41 #include <asm/processor.h>
  42 #include <asm/pgtable.h>
  43 #include <asm/smp.h>
  44 #include <asm/elf.h>
  45 #include <asm/machdep.h>
  46 #include <asm/paca.h>
  47 #include <asm/time.h>
  48 #include <asm/cputable.h>
  49 #include <asm/sections.h>
  50 #include <asm/btext.h>
  51 #include <asm/nvram.h>
  52 #include <asm/setup.h>
  53 #include <asm/system.h>
  54 #include <asm/rtas.h>
  55 #include <asm/iommu.h>
  56 #include <asm/serial.h>
  57 #include <asm/cache.h>
  58 #include <asm/page.h>
  59 #include <asm/mmu.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 int boot_cpuid = 0;
  74 u64 ppc64_pft_size;
  75
  76 /* Pick defaults since we might want to patch instructions
  77  * before we've read this from the device tree.
  78  */
  79 struct ppc64_caches ppc64_caches = {
  80         .dline_size = 0x40,
  81         .log_dline_size = 6,
  82         .iline_size = 0x40,
  83         .log_iline_size = 6
  84 };
  85 EXPORT_SYMBOL_GPL(ppc64_caches);
  86
  87 /*
  88  * These are used in binfmt_elf.c to put aux entries on the stack
  89  * for each elf executable being started.
  90  */
  91 int dcache_bsize;
  92 int icache_bsize;
  93 int ucache_bsize;
  94
  95 #ifdef CONFIG_SMP
  96
  97 static int smt_enabled_cmdline;
  98
  99 /* Look for ibm,smt-enabled OF option */
 100 static void check_smt_enabled(void)
 101 {
 102         struct device_node *dn;
 103         const char *smt_option;
 104
 105         /* Allow the command line to overrule the OF option */
 106         if (smt_enabled_cmdline)
 107                 return;
 108
 109         dn = of_find_node_by_path("/options");
 110
 111         if (dn) {
 112                 smt_option = of_get_property(dn, "ibm,smt-enabled", NULL);
 113
 114                 if (smt_option) {
 115                         if (!strcmp(smt_option, "on"))
 116                                 smt_enabled_at_boot = 1;
 117                         else if (!strcmp(smt_option, "off"))
 118                                 smt_enabled_at_boot = 0;
 119                 }
 120         }
 121 }
 122
 123 /* Look for smt-enabled= cmdline option */
 124 static int __init early_smt_enabled(char *p)
 125 {
 126         smt_enabled_cmdline = 1;
 127
 128         if (!p)
 129                 return 0;
 130
 131         if (!strcmp(p, "on") || !strcmp(p, "1"))
 132                 smt_enabled_at_boot = 1;
 133         else if (!strcmp(p, "off") || !strcmp(p, "0"))
 134                 smt_enabled_at_boot = 0;
 135
 136         return 0;
 137 }
 138 early_param("smt-enabled", early_smt_enabled);
 139
 140 #else
 141 #define check_smt_enabled()
 142 #endif /* CONFIG_SMP */
 143
 144 /* Put the paca pointer into r13 and SPRG3 */
 145 void __init setup_paca(int cpu)
 146 {
 147         local_paca = &paca[cpu];
 148         mtspr(SPRN_SPRG3, local_paca);
 149 }
 150
 151 /*
 152  * Early initialization entry point. This is called by head.S
 153  * with MMU translation disabled. We rely on the "feature" of
 154  * the CPU that ignores the top 2 bits of the address in real
 155  * mode so we can access kernel globals normally provided we
 156  * only toy with things in the RMO region. From here, we do
 157  * some early parsing of the device-tree to setup out LMB
 158  * data structures, and allocate & initialize the hash table
 159  * and segment tables so we can start running with translation
 160  * enabled.
 161  *
 162  * It is this function which will call the probe() callback of
 163  * the various platform types and copy the matching one to the
 164  * global ppc_md structure. Your platform can eventually do
 165  * some very early initializations from the probe() routine, but
 166  * this is not recommended, be very careful as, for example, the
 167  * device-tree is not accessible via normal means at this point.
 168  */
 169
 170 void __init early_setup(unsigned long dt_ptr)
 171 {
 172         /* -------- printk is _NOT_ safe to use here ! ------- */
 173
 174         /* Fill in any unititialised pacas */
 175         initialise_pacas();
 176
 177         /* Identify CPU type */
 178         identify_cpu(0, mfspr(SPRN_PVR));
 179
 180         /* Assume we're on cpu 0 for now. Don't write to the paca yet! */
 181         setup_paca(0);
 182
 183         /* Initialize lockdep early or else spinlocks will blow */
 184         lockdep_init();
 185
 186         /* -------- printk is now safe to use ------- */
 187
 188         /* Enable early debugging if any specified (see udbg.h) */
 189         udbg_early_init();
 190
 191         DBG(" -> early_setup(), dt_ptr: 0x%lx\n", dt_ptr);
 192
 193         /*
 194          * Do early initialization using the flattened device
 195          * tree, such as retrieving the physical memory map or
 196          * calculating/retrieving the hash table size.
 197          */
 198         early_init_devtree(__va(dt_ptr));
 199
 200         /* Now we know the logical id of our boot cpu, setup the paca. */
 201         setup_paca(boot_cpuid);
 202
 203         /* Fix up paca fields required for the boot cpu */
 204         get_paca()->cpu_start = 1;
 205         get_paca()->stab_real = __pa((u64)&initial_stab);
 206         get_paca()->stab_addr = (u64)&initial_stab;
 207
 208         /* Probe the machine type */
 209         probe_machine();
 210
 211         setup_kdump_trampoline();
 212
 213         DBG("Found, Initializing memory management...\n");
 214
 215         /*
 216          * Initialize the MMU Hash table and create the linear mapping
 217          * of memory. Has to be done before stab/slb initialization as
 218          * this is currently where the page size encoding is obtained
 219          */
 220         htab_initialize();
 221
 222         /*
 223          * Initialize stab / SLB management except on iSeries
 224          */
 225         if (cpu_has_feature(CPU_FTR_SLB))
 226                 slb_initialize();
 227         else if (!firmware_has_feature(FW_FEATURE_ISERIES))
 228                 stab_initialize(get_paca()->stab_real);
 229
 230         DBG(" <- early_setup()\n");
 231 }
 232
 233 #ifdef CONFIG_SMP
 234 void early_setup_secondary(void)
 235 {
 236         struct paca_struct *lpaca = get_paca();
 237
 238         /* Mark interrupts enabled in PACA */
 239         lpaca->soft_enabled = 0;
 240
 241         /* Initialize hash table for that CPU */
 242         htab_initialize_secondary();
 243
 244         /* Initialize STAB/SLB. We use a virtual address as it works
 245          * in real mode on pSeries and we want a virutal address on
 246          * iSeries anyway
 247          */
 248         if (cpu_has_feature(CPU_FTR_SLB))
 249                 slb_initialize();
 250         else
 251                 stab_initialize(lpaca->stab_addr);
 252 }
 253
 254 #endif /* CONFIG_SMP */
 255
 256 #if defined(CONFIG_SMP) || defined(CONFIG_KEXEC)
 257 extern unsigned long __secondary_hold_spinloop;
 258 extern void generic_secondary_smp_init(void);
 259
 260 void smp_release_cpus(void)
 261 {
 262         unsigned long *ptr;
 263
 264         DBG(" -> smp_release_cpus()\n");
 265
 266         /* All secondary cpus are spinning on a common spinloop, release them
 267          * all now so they can start to spin on their individual paca
 268          * spinloops. For non SMP kernels, the secondary cpus never get out
 269          * of the common spinloop.
 270          */
 271
 272         ptr  = (unsigned long *)((unsigned long)&__secondary_hold_spinloop
 273                         - PHYSICAL_START);
 274         *ptr = __pa(generic_secondary_smp_init);
 275         mb();
 276
 277         DBG(" <- smp_release_cpus()\n");
 278 }
 279 #endif /* CONFIG_SMP || CONFIG_KEXEC */
 280
 281 /*
 282  * Initialize some remaining members of the ppc64_caches and systemcfg
 283  * structures
 284  * (at least until we get rid of them completely). This is mostly some
 285  * cache informations about the CPU that will be used by cache flush
 286  * routines and/or provided to userland
 287  */
 288 static void __init initialize_cache_info(void)
 289 {
 290         struct device_node *np;
 291         unsigned long num_cpus = 0;
 292
 293         DBG(" -> initialize_cache_info()\n");
 294
 295         for (np = NULL; (np = of_find_node_by_type(np, "cpu"));) {
 296                 num_cpus += 1;
 297
 298                 /* We're assuming *all* of the CPUs have the same
 299                  * d-cache and i-cache sizes... -Peter
 300                  */
 301
 302                 if ( num_cpus == 1 ) {
 303                         const u32 *sizep, *lsizep;
 304                         u32 size, lsize;
 305
 306                         size = 0;
 307                         lsize = cur_cpu_spec->dcache_bsize;
 308                         sizep = of_get_property(np, "d-cache-size", NULL);
 309                         if (sizep != NULL)
 310                                 size = *sizep;
 311                         lsizep = of_get_property(np, "d-cache-block-size", NULL);
 312                         /* fallback if block size missing */
 313                         if (lsizep == NULL)
 314                                 lsizep = of_get_property(np, "d-cache-line-size", NULL);
 315                         if (lsizep != NULL)
 316                                 lsize = *lsizep;
 317                         if (sizep == 0 || lsizep == 0)
 318                                 DBG("Argh, can't find dcache properties ! "
 319                                     "sizep: %p, lsizep: %p\n", sizep, lsizep);
 320
 321                         ppc64_caches.dsize = size;
 322                         ppc64_caches.dline_size = lsize;
 323                         ppc64_caches.log_dline_size = __ilog2(lsize);
 324                         ppc64_caches.dlines_per_page = PAGE_SIZE / lsize;
 325
 326                         size = 0;
 327                         lsize = cur_cpu_spec->icache_bsize;
 328                         sizep = of_get_property(np, "i-cache-size", NULL);
 329                         if (sizep != NULL)
 330                                 size = *sizep;
 331                         lsizep = of_get_property(np, "i-cache-block-size", NULL);
 332                         if (lsizep == NULL)
 333                                 lsizep = of_get_property(np, "i-cache-line-size", NULL);
 334                         if (lsizep != NULL)
 335                                 lsize = *lsizep;
 336                         if (sizep == 0 || lsizep == 0)
 337                                 DBG("Argh, can't find icache properties ! "
 338                                     "sizep: %p, lsizep: %p\n", sizep, lsizep);
 339
 340                         ppc64_caches.isize = size;
 341                         ppc64_caches.iline_size = lsize;
 342                         ppc64_caches.log_iline_size = __ilog2(lsize);
 343                         ppc64_caches.ilines_per_page = PAGE_SIZE / lsize;
 344                 }
 345         }
 346
 347         DBG(" <- initialize_cache_info()\n");
 348 }
 349
 350
 351 /*
 352  * Do some initial setup of the system.  The parameters are those which
 353  * were passed in from the bootloader.
 354  */
 355 void __init setup_system(void)
 356 {
 357         DBG(" -> setup_system()\n");
 358
 359         /* Apply the CPUs-specific and firmware specific fixups to kernel
 360          * text (nop out sections not relevant to this CPU or this firmware)
 361          */
 362         do_feature_fixups(cur_cpu_spec->cpu_features,
 363                           &__start___ftr_fixup, &__stop___ftr_fixup);
 364         do_feature_fixups(cur_cpu_spec->mmu_features,
 365                           &__start___mmu_ftr_fixup, &__stop___mmu_ftr_fixup);
 366         do_feature_fixups(powerpc_firmware_features,
 367                           &__start___fw_ftr_fixup, &__stop___fw_ftr_fixup);
 368         do_lwsync_fixups(cur_cpu_spec->cpu_features,
 369                          &__start___lwsync_fixup, &__stop___lwsync_fixup);
 370
 371         /*
 372          * Unflatten the device-tree passed by prom_init or kexec
 373          */
 374         unflatten_device_tree();
 375
 376         /*
 377          * Fill the ppc64_caches & systemcfg structures with informations
 378          * retrieved from the device-tree.
 379          */
 380         initialize_cache_info();
 381
 382         /*
 383          * Initialize irq remapping subsystem
 384          */
 385         irq_early_init();
 386
 387 #ifdef CONFIG_PPC_RTAS
 388         /*
 389          * Initialize RTAS if available
 390          */
 391         rtas_initialize();
 392 #endif /* CONFIG_PPC_RTAS */
 393
 394         /*
 395          * Check if we have an initrd provided via the device-tree
 396          */
 397         check_for_initrd();
 398
 399         /*
 400          * Do some platform specific early initializations, that includes
 401          * setting up the hash table pointers. It also sets up some interrupt-mapping
 402          * related options that will be used by finish_device_tree()
 403          */
 404         if (ppc_md.init_early)
 405                 ppc_md.init_early();
 406
 407         /*
 408          * We can discover serial ports now since the above did setup the
 409          * hash table management for us, thus ioremap works. We do that early
 410          * so that further code can be debugged
 411          */
 412         find_legacy_serial_ports();
 413
 414         /*
 415          * Register early console
 416          */
 417         register_early_udbg_console();
 418
 419         /*
 420          * Initialize xmon
 421          */
 422         xmon_setup();
 423
 424         check_smt_enabled();
 425         smp_setup_cpu_maps();
 426
 427 #ifdef CONFIG_SMP
 428         /* Release secondary cpus out of their spinloops at 0x60 now that
 429          * we can map physical -> logical CPU ids
 430          */
 431         smp_release_cpus();
 432 #endif
 433
 434         printk("Starting Linux PPC64 %s\n", init_utsname()->version);
 435
 436         printk("-----------------------------------------------------\n");
 437         printk("ppc64_pft_size                = 0x%lx\n", ppc64_pft_size);
 438         printk("physicalMemorySize            = 0x%lx\n", lmb_phys_mem_size());
 439         if (ppc64_caches.dline_size != 0x80)
 440                 printk("ppc64_caches.dcache_line_size = 0x%x\n",
 441                        ppc64_caches.dline_size);
 442         if (ppc64_caches.iline_size != 0x80)
 443                 printk("ppc64_caches.icache_line_size = 0x%x\n",
 444                        ppc64_caches.iline_size);
 445         if (htab_address)
 446                 printk("htab_address                  = 0x%p\n", htab_address);
 447         printk("htab_hash_mask                = 0x%lx\n", htab_hash_mask);
 448         if (PHYSICAL_START > 0)
 449                 printk("physical_start                = 0x%lx\n",
 450                        PHYSICAL_START);
 451         printk("-----------------------------------------------------\n");
 452
 453         DBG(" <- setup_system()\n");
 454 }
 455
 456 #ifdef CONFIG_IRQSTACKS
 457 static void __init irqstack_early_init(void)
 458 {
 459         unsigned int i;
 460
 461         /*
 462          * interrupt stacks must be under 256MB, we cannot afford to take
 463          * SLB misses on them.
 464          */
 465         for_each_possible_cpu(i) {
 466                 softirq_ctx[i] = (struct thread_info *)
 467                         __va(lmb_alloc_base(THREAD_SIZE,
 468                                             THREAD_SIZE, 0x10000000));
 469                 hardirq_ctx[i] = (struct thread_info *)
 470                         __va(lmb_alloc_base(THREAD_SIZE,
 471                                             THREAD_SIZE, 0x10000000));
 472         }
 473 }
 474 #else
 475 #define irqstack_early_init()
 476 #endif
 477
 478 /*
 479  * Stack space used when we detect a bad kernel stack pointer, and
 480  * early in SMP boots before relocation is enabled.
 481  */
 482 static void __init emergency_stack_init(void)
 483 {
 484         unsigned long limit;
 485         unsigned int i;
 486
 487         /*
 488          * Emergency stacks must be under 256MB, we cannot afford to take
 489          * SLB misses on them. The ABI also requires them to be 128-byte
 490          * aligned.
 491          *
 492          * Since we use these as temporary stacks during secondary CPU
 493          * bringup, we need to get at them in real mode. This means they
 494          * must also be within the RMO region.
 495          */
 496         limit = min(0x10000000UL, lmb.rmo_size);
 497
 498         for_each_possible_cpu(i) {
 499                 unsigned long sp;
 500                 sp  = lmb_alloc_base(THREAD_SIZE, THREAD_SIZE, limit);
 501                 sp += THREAD_SIZE;
 502                 paca[i].emergency_sp = __va(sp);
 503         }
 504 }
 505
 506 /*
 507  * Called into from start_kernel, after lock_kernel has been called.
 508  * Initializes bootmem, which is unsed to manage page allocation until
 509  * mem_init is called.
 510  */
 511 void __init setup_arch(char **cmdline_p)
 512 {
 513         ppc64_boot_msg(0x12, "Setup Arch");
 514
 515         *cmdline_p = cmd_line;
 516
 517         /*
 518          * Set cache line size based on type of cpu as a default.
 519          * Systems with OF can look in the properties on the cpu node(s)
 520          * for a possibly more accurate value.
 521          */
 522         dcache_bsize = ppc64_caches.dline_size;
 523         icache_bsize = ppc64_caches.iline_size;
 524
 525         /* reboot on panic */
 526         panic_timeout = 180;
 527
 528         if (ppc_md.panic)
 529                 setup_panic();
 530
 531         init_mm.start_code = (unsigned long)_stext;
 532         init_mm.end_code = (unsigned long) _etext;
 533         init_mm.end_data = (unsigned long) _edata;
 534         init_mm.brk = klimit;
 535
 536         irqstack_early_init();
 537         emergency_stack_init();
 538
 539         stabs_alloc();
 540
 541         /* set up the bootmem stuff with available memory */
 542         do_init_bootmem();
 543         sparse_init();
 544
 545 #ifdef CONFIG_DUMMY_CONSOLE
 546         conswitchp = &dummy_con;
 547 #endif
 548
 549         if (ppc_md.setup_arch)
 550                 ppc_md.setup_arch();
 551
 552         paging_init();
 553         ppc64_boot_msg(0x15, "Setup Done");
 554 }
 555
 556
 557 /* ToDo: do something useful if ppc_md is not yet setup. */
 558 #define PPC64_LINUX_FUNCTION 0x0f000000
 559 #define PPC64_IPL_MESSAGE 0xc0000000
 560 #define PPC64_TERM_MESSAGE 0xb0000000
 561
 562 static void ppc64_do_msg(unsigned int src, const char *msg)
 563 {
 564         if (ppc_md.progress) {
 565                 char buf[128];
 566
 567                 sprintf(buf, "%08X\n", src);
 568                 ppc_md.progress(buf, 0);
 569                 snprintf(buf, 128, "%s", msg);
 570                 ppc_md.progress(buf, 0);
 571         }
 572 }
 573
 574 /* Print a boot progress message. */
 575 void ppc64_boot_msg(unsigned int src, const char *msg)
 576 {
 577         ppc64_do_msg(PPC64_LINUX_FUNCTION|PPC64_IPL_MESSAGE|src, msg);
 578         printk("[boot]%04x %s\n", src, msg);
 579 }
 580
 581 /* Print a termination message (print only -- does not stop the kernel) */
 582 void ppc64_terminate_msg(unsigned int src, const char *msg)
 583 {
 584         ppc64_do_msg(PPC64_LINUX_FUNCTION|PPC64_TERM_MESSAGE|src, msg);
 585         printk("[terminate]%04x %s\n", src, msg);
 586 }
 587
 588 void cpu_die(void)
 589 {
 590         if (ppc_md.cpu_die)
 591                 ppc_md.cpu_die();
 592 }
 593
 594 #ifdef CONFIG_SMP
 595 void __init setup_per_cpu_areas(void)
 596 {
 597         int i;
 598         unsigned long size;
 599         char *ptr;
 600
 601         /* Copy section for each CPU (we discard the original) */
 602         size = ALIGN(__per_cpu_end - __per_cpu_start, PAGE_SIZE);
 603 #ifdef CONFIG_MODULES
 604         if (size < PERCPU_ENOUGH_ROOM)
 605                 size = PERCPU_ENOUGH_ROOM;
 606 #endif
 607
 608         for_each_possible_cpu(i) {
 609                 ptr = alloc_bootmem_pages_node(NODE_DATA(cpu_to_node(i)), size);
 610
 611                 paca[i].data_offset = ptr - __per_cpu_start;
 612                 memcpy(ptr, __per_cpu_start, __per_cpu_end - __per_cpu_start);
 613         }
 614 }
 615 #endif
 616
 617
 618 #ifdef CONFIG_PPC_INDIRECT_IO
 619 struct ppc_pci_io ppc_pci_io;
 620 EXPORT_SYMBOL(ppc_pci_io);
 621 #endif /* CONFIG_PPC_INDIRECT_IO */
 622