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 <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/firmware.h>
  60 #include <asm/xmon.h>
  61 #include <asm/udbg.h>
  62 #include <asm/kexec.h>
  63
  64 #include "setup.h"
  65
  66 #ifdef DEBUG
  67 #define DBG(fmt...) udbg_printf(fmt)
  68 #else
  69 #define DBG(fmt...)
  70 #endif
  71
  72 int have_of = 1;
  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         /* Identify CPU type */
 173         identify_cpu(0, mfspr(SPRN_PVR));
 174
 175         /* Assume we're on cpu 0 for now. Don't write to the paca yet! */
 176         setup_paca(0);
 177
 178         /* Enable early debugging if any specified (see udbg.h) */
 179         udbg_early_init();
 180
 181         DBG(" -> early_setup(), dt_ptr: 0x%lx\n", dt_ptr);
 182
 183         /*
 184          * Do early initialization using the flattened device
 185          * tree, such as retrieving the physical memory map or
 186          * calculating/retrieving the hash table size.
 187          */
 188         early_init_devtree(__va(dt_ptr));
 189
 190         /* Now we know the logical id of our boot cpu, setup the paca. */
 191         setup_paca(boot_cpuid);
 192
 193         /* Fix up paca fields required for the boot cpu */
 194         get_paca()->cpu_start = 1;
 195         get_paca()->stab_real = __pa((u64)&initial_stab);
 196         get_paca()->stab_addr = (u64)&initial_stab;
 197
 198         /* Probe the machine type */
 199         probe_machine();
 200
 201         setup_kdump_trampoline();
 202
 203         DBG("Found, Initializing memory management...\n");
 204
 205         /*
 206          * Initialize the MMU Hash table and create the linear mapping
 207          * of memory. Has to be done before stab/slb initialization as
 208          * this is currently where the page size encoding is obtained
 209          */
 210         htab_initialize();
 211
 212         /*
 213          * Initialize stab / SLB management except on iSeries
 214          */
 215         if (cpu_has_feature(CPU_FTR_SLB))
 216                 slb_initialize();
 217         else if (!firmware_has_feature(FW_FEATURE_ISERIES))
 218                 stab_initialize(get_paca()->stab_real);
 219
 220         DBG(" <- early_setup()\n");
 221 }
 222
 223 #ifdef CONFIG_SMP
 224 void early_setup_secondary(void)
 225 {
 226         struct paca_struct *lpaca = get_paca();
 227
 228         /* Mark interrupts enabled in PACA */
 229         lpaca->soft_enabled = 0;
 230
 231         /* Initialize hash table for that CPU */
 232         htab_initialize_secondary();
 233
 234         /* Initialize STAB/SLB. We use a virtual address as it works
 235          * in real mode on pSeries and we want a virutal address on
 236          * iSeries anyway
 237          */
 238         if (cpu_has_feature(CPU_FTR_SLB))
 239                 slb_initialize();
 240         else
 241                 stab_initialize(lpaca->stab_addr);
 242 }
 243
 244 #endif /* CONFIG_SMP */
 245
 246 #if defined(CONFIG_SMP) || defined(CONFIG_KEXEC)
 247 void smp_release_cpus(void)
 248 {
 249         extern unsigned long __secondary_hold_spinloop;
 250         unsigned long *ptr;
 251
 252         DBG(" -> smp_release_cpus()\n");
 253
 254         /* All secondary cpus are spinning on a common spinloop, release them
 255          * all now so they can start to spin on their individual paca
 256          * spinloops. For non SMP kernels, the secondary cpus never get out
 257          * of the common spinloop.
 258          * This is useless but harmless on iSeries, secondaries are already
 259          * waiting on their paca spinloops. */
 260
 261         ptr  = (unsigned long *)((unsigned long)&__secondary_hold_spinloop
 262                         - PHYSICAL_START);
 263         *ptr = 1;
 264         mb();
 265
 266         DBG(" <- smp_release_cpus()\n");
 267 }
 268 #endif /* CONFIG_SMP || CONFIG_KEXEC */
 269
 270 /*
 271  * Initialize some remaining members of the ppc64_caches and systemcfg
 272  * structures
 273  * (at least until we get rid of them completely). This is mostly some
 274  * cache informations about the CPU that will be used by cache flush
 275  * routines and/or provided to userland
 276  */
 277 static void __init initialize_cache_info(void)
 278 {
 279         struct device_node *np;
 280         unsigned long num_cpus = 0;
 281
 282         DBG(" -> initialize_cache_info()\n");
 283
 284         for (np = NULL; (np = of_find_node_by_type(np, "cpu"));) {
 285                 num_cpus += 1;
 286
 287                 /* We're assuming *all* of the CPUs have the same
 288                  * d-cache and i-cache sizes... -Peter
 289                  */
 290
 291                 if ( num_cpus == 1 ) {
 292                         const u32 *sizep, *lsizep;
 293                         u32 size, lsize;
 294                         const char *dc, *ic;
 295
 296                         /* Then read cache informations */
 297                         if (machine_is(powermac)) {
 298                                 dc = "d-cache-block-size";
 299                                 ic = "i-cache-block-size";
 300                         } else {
 301                                 dc = "d-cache-line-size";
 302                                 ic = "i-cache-line-size";
 303                         }
 304
 305                         size = 0;
 306                         lsize = cur_cpu_spec->dcache_bsize;
 307                         sizep = of_get_property(np, "d-cache-size", NULL);
 308                         if (sizep != NULL)
 309                                 size = *sizep;
 310                         lsizep = of_get_property(np, dc, NULL);
 311                         if (lsizep != NULL)
 312                                 lsize = *lsizep;
 313                         if (sizep == 0 || lsizep == 0)
 314                                 DBG("Argh, can't find dcache properties ! "
 315                                     "sizep: %p, lsizep: %p\n", sizep, lsizep);
 316
 317                         ppc64_caches.dsize = size;
 318                         ppc64_caches.dline_size = lsize;
 319                         ppc64_caches.log_dline_size = __ilog2(lsize);
 320                         ppc64_caches.dlines_per_page = PAGE_SIZE / lsize;
 321
 322                         size = 0;
 323                         lsize = cur_cpu_spec->icache_bsize;
 324                         sizep = of_get_property(np, "i-cache-size", NULL);
 325                         if (sizep != NULL)
 326                                 size = *sizep;
 327                         lsizep = of_get_property(np, ic, NULL);
 328                         if (lsizep != NULL)
 329                                 lsize = *lsizep;
 330                         if (sizep == 0 || lsizep == 0)
 331                                 DBG("Argh, can't find icache properties ! "
 332                                     "sizep: %p, lsizep: %p\n", sizep, lsizep);
 333
 334                         ppc64_caches.isize = size;
 335                         ppc64_caches.iline_size = lsize;
 336                         ppc64_caches.log_iline_size = __ilog2(lsize);
 337                         ppc64_caches.ilines_per_page = PAGE_SIZE / lsize;
 338                 }
 339         }
 340
 341         DBG(" <- initialize_cache_info()\n");
 342 }
 343
 344
 345 /*
 346  * Do some initial setup of the system.  The parameters are those which
 347  * were passed in from the bootloader.
 348  */
 349 void __init setup_system(void)
 350 {
 351         DBG(" -> setup_system()\n");
 352
 353         /* Apply the CPUs-specific and firmware specific fixups to kernel
 354          * text (nop out sections not relevant to this CPU or this firmware)
 355          */
 356         do_feature_fixups(cur_cpu_spec->cpu_features,
 357                           &__start___ftr_fixup, &__stop___ftr_fixup);
 358         do_feature_fixups(powerpc_firmware_features,
 359                           &__start___fw_ftr_fixup, &__stop___fw_ftr_fixup);
 360
 361         /*
 362          * Unflatten the device-tree passed by prom_init or kexec
 363          */
 364         unflatten_device_tree();
 365
 366         /*
 367          * Fill the ppc64_caches & systemcfg structures with informations
 368          * retrieved from the device-tree.
 369          */
 370         initialize_cache_info();
 371
 372         /*
 373          * Initialize irq remapping subsystem
 374          */
 375         irq_early_init();
 376
 377 #ifdef CONFIG_PPC_RTAS
 378         /*
 379          * Initialize RTAS if available
 380          */
 381         rtas_initialize();
 382 #endif /* CONFIG_PPC_RTAS */
 383
 384         /*
 385          * Check if we have an initrd provided via the device-tree
 386          */
 387         check_for_initrd();
 388
 389         /*
 390          * Do some platform specific early initializations, that includes
 391          * setting up the hash table pointers. It also sets up some interrupt-mapping
 392          * related options that will be used by finish_device_tree()
 393          */
 394         if (ppc_md.init_early)
 395                 ppc_md.init_early();
 396
 397         /*
 398          * We can discover serial ports now since the above did setup the
 399          * hash table management for us, thus ioremap works. We do that early
 400          * so that further code can be debugged
 401          */
 402         find_legacy_serial_ports();
 403
 404         /*
 405          * Register early console
 406          */
 407         register_early_udbg_console();
 408
 409         /*
 410          * Initialize xmon
 411          */
 412         xmon_setup();
 413
 414         check_smt_enabled();
 415         smp_setup_cpu_maps();
 416
 417 #ifdef CONFIG_SMP
 418         /* Release secondary cpus out of their spinloops at 0x60 now that
 419          * we can map physical -> logical CPU ids
 420          */
 421         smp_release_cpus();
 422 #endif
 423
 424         printk("Starting Linux PPC64 %s\n", init_utsname()->version);
 425
 426         printk("-----------------------------------------------------\n");
 427         printk("ppc64_pft_size                = 0x%lx\n", ppc64_pft_size);
 428         printk("physicalMemorySize            = 0x%lx\n", lmb_phys_mem_size());
 429         if (ppc64_caches.dline_size != 0x80)
 430                 printk("ppc64_caches.dcache_line_size = 0x%x\n",
 431                        ppc64_caches.dline_size);
 432         if (ppc64_caches.iline_size != 0x80)
 433                 printk("ppc64_caches.icache_line_size = 0x%x\n",
 434                        ppc64_caches.iline_size);
 435         if (htab_address)
 436                 printk("htab_address                  = 0x%p\n", htab_address);
 437         printk("htab_hash_mask                = 0x%lx\n", htab_hash_mask);
 438 #if PHYSICAL_START > 0
 439         printk("physical_start                = 0x%x\n", PHYSICAL_START);
 440 #endif
 441         printk("-----------------------------------------------------\n");
 442
 443         DBG(" <- setup_system()\n");
 444 }
 445
 446 #ifdef CONFIG_IRQSTACKS
 447 static void __init irqstack_early_init(void)
 448 {
 449         unsigned int i;
 450
 451         /*
 452          * interrupt stacks must be under 256MB, we cannot afford to take
 453          * SLB misses on them.
 454          */
 455         for_each_possible_cpu(i) {
 456                 softirq_ctx[i] = (struct thread_info *)
 457                         __va(lmb_alloc_base(THREAD_SIZE,
 458                                             THREAD_SIZE, 0x10000000));
 459                 hardirq_ctx[i] = (struct thread_info *)
 460                         __va(lmb_alloc_base(THREAD_SIZE,
 461                                             THREAD_SIZE, 0x10000000));
 462         }
 463 }
 464 #else
 465 #define irqstack_early_init()
 466 #endif
 467
 468 /*
 469  * Stack space used when we detect a bad kernel stack pointer, and
 470  * early in SMP boots before relocation is enabled.
 471  */
 472 static void __init emergency_stack_init(void)
 473 {
 474         unsigned long limit;
 475         unsigned int i;
 476
 477         /*
 478          * Emergency stacks must be under 256MB, we cannot afford to take
 479          * SLB misses on them. The ABI also requires them to be 128-byte
 480          * aligned.
 481          *
 482          * Since we use these as temporary stacks during secondary CPU
 483          * bringup, we need to get at them in real mode. This means they
 484          * must also be within the RMO region.
 485          */
 486         limit = min(0x10000000UL, lmb.rmo_size);
 487
 488         for_each_possible_cpu(i)
 489                 paca[i].emergency_sp =
 490                 __va(lmb_alloc_base(HW_PAGE_SIZE, 128, limit)) + HW_PAGE_SIZE;
 491 }
 492
 493 /*
 494  * Called into from start_kernel, after lock_kernel has been called.
 495  * Initializes bootmem, which is unsed to manage page allocation until
 496  * mem_init is called.
 497  */
 498 void __init setup_arch(char **cmdline_p)
 499 {
 500         ppc64_boot_msg(0x12, "Setup Arch");
 501
 502         *cmdline_p = cmd_line;
 503
 504         /*
 505          * Set cache line size based on type of cpu as a default.
 506          * Systems with OF can look in the properties on the cpu node(s)
 507          * for a possibly more accurate value.
 508          */
 509         dcache_bsize = ppc64_caches.dline_size;
 510         icache_bsize = ppc64_caches.iline_size;
 511
 512         /* reboot on panic */
 513         panic_timeout = 180;
 514
 515         if (ppc_md.panic)
 516                 setup_panic();
 517
 518         init_mm.start_code = PAGE_OFFSET;
 519         init_mm.end_code = (unsigned long) _etext;
 520         init_mm.end_data = (unsigned long) _edata;
 521         init_mm.brk = klimit;
 522
 523         irqstack_early_init();
 524         emergency_stack_init();
 525
 526         stabs_alloc();
 527
 528         /* set up the bootmem stuff with available memory */
 529         do_init_bootmem();
 530         sparse_init();
 531
 532 #ifdef CONFIG_DUMMY_CONSOLE
 533         conswitchp = &dummy_con;
 534 #endif
 535
 536         if (ppc_md.setup_arch)
 537                 ppc_md.setup_arch();
 538
 539         paging_init();
 540         ppc64_boot_msg(0x15, "Setup Done");
 541 }
 542
 543
 544 /* ToDo: do something useful if ppc_md is not yet setup. */
 545 #define PPC64_LINUX_FUNCTION 0x0f000000
 546 #define PPC64_IPL_MESSAGE 0xc0000000
 547 #define PPC64_TERM_MESSAGE 0xb0000000
 548
 549 static void ppc64_do_msg(unsigned int src, const char *msg)
 550 {
 551         if (ppc_md.progress) {
 552                 char buf[128];
 553
 554                 sprintf(buf, "%08X\n", src);
 555                 ppc_md.progress(buf, 0);
 556                 snprintf(buf, 128, "%s", msg);
 557                 ppc_md.progress(buf, 0);
 558         }
 559 }
 560
 561 /* Print a boot progress message. */
 562 void ppc64_boot_msg(unsigned int src, const char *msg)
 563 {
 564         ppc64_do_msg(PPC64_LINUX_FUNCTION|PPC64_IPL_MESSAGE|src, msg);
 565         printk("[boot]%04x %s\n", src, msg);
 566 }
 567
 568 /* Print a termination message (print only -- does not stop the kernel) */
 569 void ppc64_terminate_msg(unsigned int src, const char *msg)
 570 {
 571         ppc64_do_msg(PPC64_LINUX_FUNCTION|PPC64_TERM_MESSAGE|src, msg);
 572         printk("[terminate]%04x %s\n", src, msg);
 573 }
 574
 575 void cpu_die(void)
 576 {
 577         if (ppc_md.cpu_die)
 578                 ppc_md.cpu_die();
 579 }
 580
 581 #ifdef CONFIG_SMP
 582 void __init setup_per_cpu_areas(void)
 583 {
 584         int i;
 585         unsigned long size;
 586         char *ptr;
 587
 588         /* Copy section for each CPU (we discard the original) */
 589         size = ALIGN(__per_cpu_end - __per_cpu_start, PAGE_SIZE);
 590 #ifdef CONFIG_MODULES
 591         if (size < PERCPU_ENOUGH_ROOM)
 592                 size = PERCPU_ENOUGH_ROOM;
 593 #endif
 594
 595         for_each_possible_cpu(i) {
 596                 ptr = alloc_bootmem_pages_node(NODE_DATA(cpu_to_node(i)), size);
 597                 if (!ptr)
 598                         panic("Cannot allocate cpu data for CPU %d\n", i);
 599
 600                 paca[i].data_offset = ptr - __per_cpu_start;
 601                 memcpy(ptr, __per_cpu_start, __per_cpu_end - __per_cpu_start);
 602         }
 603
 604         /* Now that per_cpu is setup, initialize cpu_sibling_map */
 605         smp_setup_cpu_sibling_map();
 606 }
 607 #endif
 608
 609
 610 #ifdef CONFIG_PPC_INDIRECT_IO
 611 struct ppc_pci_io ppc_pci_io;
 612 EXPORT_SYMBOL(ppc_pci_io);
 613 #endif /* CONFIG_PPC_INDIRECT_IO */
 614