setup_64.c revision e8222502ee6157e2713da9e0792c21f4ad458d50
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 74int have_of = 1; 75int boot_cpuid = 0; 76dev_t boot_dev; 77u64 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 */ 82struct ppc64_caches ppc64_caches = { 83 .dline_size = 0x80, 84 .log_dline_size = 7, 85 .iline_size = 0x80, 86 .log_iline_size = 7 87}; 88EXPORT_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 */ 94int dcache_bsize; 95int icache_bsize; 96int ucache_bsize; 97 98#ifdef CONFIG_MAGIC_SYSRQ 99unsigned long SYSRQ_KEY; 100#endif /* CONFIG_MAGIC_SYSRQ */ 101 102 103static int ppc64_panic_event(struct notifier_block *, unsigned long, void *); 104static 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 111static int smt_enabled_cmdline; 112 113/* Look for ibm,smt-enabled OF option */ 114static 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 */ 138static 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} 152early_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 177void __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 (!firmware_has_feature(FW_FEATURE_ISERIES)) { 219 if (cpu_has_feature(CPU_FTR_SLB)) 220 slb_initialize(); 221 else 222 stab_initialize(get_paca()->stab_real); 223 } 224 225 DBG(" <- early_setup()\n"); 226} 227 228#ifdef CONFIG_SMP 229void early_setup_secondary(void) 230{ 231 struct paca_struct *lpaca = get_paca(); 232 233 /* Mark enabled in PACA */ 234 lpaca->proc_enabled = 0; 235 236 /* Initialize hash table for that CPU */ 237 htab_initialize_secondary(); 238 239 /* Initialize STAB/SLB. We use a virtual address as it works 240 * in real mode on pSeries and we want a virutal address on 241 * iSeries anyway 242 */ 243 if (cpu_has_feature(CPU_FTR_SLB)) 244 slb_initialize(); 245 else 246 stab_initialize(lpaca->stab_addr); 247} 248 249#endif /* CONFIG_SMP */ 250 251#if defined(CONFIG_SMP) || defined(CONFIG_KEXEC) 252void smp_release_cpus(void) 253{ 254 extern unsigned long __secondary_hold_spinloop; 255 unsigned long *ptr; 256 257 DBG(" -> smp_release_cpus()\n"); 258 259 /* All secondary cpus are spinning on a common spinloop, release them 260 * all now so they can start to spin on their individual paca 261 * spinloops. For non SMP kernels, the secondary cpus never get out 262 * of the common spinloop. 263 * This is useless but harmless on iSeries, secondaries are already 264 * waiting on their paca spinloops. */ 265 266 ptr = (unsigned long *)((unsigned long)&__secondary_hold_spinloop 267 - PHYSICAL_START); 268 *ptr = 1; 269 mb(); 270 271 DBG(" <- smp_release_cpus()\n"); 272} 273#endif /* CONFIG_SMP || CONFIG_KEXEC */ 274 275/* 276 * Initialize some remaining members of the ppc64_caches and systemcfg 277 * structures 278 * (at least until we get rid of them completely). This is mostly some 279 * cache informations about the CPU that will be used by cache flush 280 * routines and/or provided to userland 281 */ 282static void __init initialize_cache_info(void) 283{ 284 struct device_node *np; 285 unsigned long num_cpus = 0; 286 287 DBG(" -> initialize_cache_info()\n"); 288 289 for (np = NULL; (np = of_find_node_by_type(np, "cpu"));) { 290 num_cpus += 1; 291 292 /* We're assuming *all* of the CPUs have the same 293 * d-cache and i-cache sizes... -Peter 294 */ 295 296 if ( num_cpus == 1 ) { 297 u32 *sizep, *lsizep; 298 u32 size, lsize; 299 const char *dc, *ic; 300 301 /* Then read cache informations */ 302 if (machine_is(powermac)) { 303 dc = "d-cache-block-size"; 304 ic = "i-cache-block-size"; 305 } else { 306 dc = "d-cache-line-size"; 307 ic = "i-cache-line-size"; 308 } 309 310 size = 0; 311 lsize = cur_cpu_spec->dcache_bsize; 312 sizep = (u32 *)get_property(np, "d-cache-size", NULL); 313 if (sizep != NULL) 314 size = *sizep; 315 lsizep = (u32 *) get_property(np, dc, NULL); 316 if (lsizep != NULL) 317 lsize = *lsizep; 318 if (sizep == 0 || lsizep == 0) 319 DBG("Argh, can't find dcache properties ! " 320 "sizep: %p, lsizep: %p\n", sizep, lsizep); 321 322 ppc64_caches.dsize = size; 323 ppc64_caches.dline_size = lsize; 324 ppc64_caches.log_dline_size = __ilog2(lsize); 325 ppc64_caches.dlines_per_page = PAGE_SIZE / lsize; 326 327 size = 0; 328 lsize = cur_cpu_spec->icache_bsize; 329 sizep = (u32 *)get_property(np, "i-cache-size", NULL); 330 if (sizep != NULL) 331 size = *sizep; 332 lsizep = (u32 *)get_property(np, ic, NULL); 333 if (lsizep != NULL) 334 lsize = *lsizep; 335 if (sizep == 0 || lsizep == 0) 336 DBG("Argh, can't find icache properties ! " 337 "sizep: %p, lsizep: %p\n", sizep, lsizep); 338 339 ppc64_caches.isize = size; 340 ppc64_caches.iline_size = lsize; 341 ppc64_caches.log_iline_size = __ilog2(lsize); 342 ppc64_caches.ilines_per_page = PAGE_SIZE / lsize; 343 } 344 } 345 346 DBG(" <- initialize_cache_info()\n"); 347} 348 349 350/* 351 * Do some initial setup of the system. The parameters are those which 352 * were passed in from the bootloader. 353 */ 354void __init setup_system(void) 355{ 356 DBG(" -> setup_system()\n"); 357 358#ifdef CONFIG_KEXEC 359 kdump_move_device_tree(); 360#endif 361 /* 362 * Unflatten the device-tree passed by prom_init or kexec 363 */ 364 unflatten_device_tree(); 365 366#ifdef CONFIG_KEXEC 367 kexec_setup(); /* requires unflattened device tree. */ 368#endif 369 370 /* 371 * Fill the ppc64_caches & systemcfg structures with informations 372 * retrieved from the device-tree. Need to be called before 373 * finish_device_tree() since the later requires some of the 374 * informations filled up here to properly parse the interrupt 375 * tree. 376 * It also sets up the cache line sizes which allows to call 377 * routines like flush_icache_range (used by the hash init 378 * later on). 379 */ 380 initialize_cache_info(); 381 382#ifdef CONFIG_PPC_RTAS 383 /* 384 * Initialize RTAS if available 385 */ 386 rtas_initialize(); 387#endif /* CONFIG_PPC_RTAS */ 388 389 /* 390 * Check if we have an initrd provided via the device-tree 391 */ 392 check_for_initrd(); 393 394 /* 395 * Do some platform specific early initializations, that includes 396 * setting up the hash table pointers. It also sets up some interrupt-mapping 397 * related options that will be used by finish_device_tree() 398 */ 399 ppc_md.init_early(); 400 401 /* 402 * We can discover serial ports now since the above did setup the 403 * hash table management for us, thus ioremap works. We do that early 404 * so that further code can be debugged 405 */ 406 find_legacy_serial_ports(); 407 408 /* 409 * "Finish" the device-tree, that is do the actual parsing of 410 * some of the properties like the interrupt map 411 */ 412 finish_device_tree(); 413 414 /* 415 * Initialize xmon 416 */ 417#ifdef CONFIG_XMON_DEFAULT 418 xmon_init(1); 419#endif 420 /* 421 * Register early console 422 */ 423 register_early_udbg_console(); 424 425 /* Save unparsed command line copy for /proc/cmdline */ 426 strlcpy(saved_command_line, cmd_line, COMMAND_LINE_SIZE); 427 428 parse_early_param(); 429 430 check_smt_enabled(); 431 smp_setup_cpu_maps(); 432 433#ifdef CONFIG_SMP 434 /* Release secondary cpus out of their spinloops at 0x60 now that 435 * we can map physical -> logical CPU ids 436 */ 437 smp_release_cpus(); 438#endif 439 440 printk("Starting Linux PPC64 %s\n", system_utsname.version); 441 442 printk("-----------------------------------------------------\n"); 443 printk("ppc64_pft_size = 0x%lx\n", ppc64_pft_size); 444 printk("ppc64_interrupt_controller = 0x%ld\n", 445 ppc64_interrupt_controller); 446 printk("physicalMemorySize = 0x%lx\n", lmb_phys_mem_size()); 447 printk("ppc64_caches.dcache_line_size = 0x%x\n", 448 ppc64_caches.dline_size); 449 printk("ppc64_caches.icache_line_size = 0x%x\n", 450 ppc64_caches.iline_size); 451 printk("htab_address = 0x%p\n", htab_address); 452 printk("htab_hash_mask = 0x%lx\n", htab_hash_mask); 453#if PHYSICAL_START > 0 454 printk("physical_start = 0x%x\n", PHYSICAL_START); 455#endif 456 printk("-----------------------------------------------------\n"); 457 458 DBG(" <- setup_system()\n"); 459} 460 461static int ppc64_panic_event(struct notifier_block *this, 462 unsigned long event, void *ptr) 463{ 464 ppc_md.panic((char *)ptr); /* May not return */ 465 return NOTIFY_DONE; 466} 467 468#ifdef CONFIG_IRQSTACKS 469static void __init irqstack_early_init(void) 470{ 471 unsigned int i; 472 473 /* 474 * interrupt stacks must be under 256MB, we cannot afford to take 475 * SLB misses on them. 476 */ 477 for_each_cpu(i) { 478 softirq_ctx[i] = (struct thread_info *) 479 __va(lmb_alloc_base(THREAD_SIZE, 480 THREAD_SIZE, 0x10000000)); 481 hardirq_ctx[i] = (struct thread_info *) 482 __va(lmb_alloc_base(THREAD_SIZE, 483 THREAD_SIZE, 0x10000000)); 484 } 485} 486#else 487#define irqstack_early_init() 488#endif 489 490/* 491 * Stack space used when we detect a bad kernel stack pointer, and 492 * early in SMP boots before relocation is enabled. 493 */ 494static void __init emergency_stack_init(void) 495{ 496 unsigned long limit; 497 unsigned int i; 498 499 /* 500 * Emergency stacks must be under 256MB, we cannot afford to take 501 * SLB misses on them. The ABI also requires them to be 128-byte 502 * aligned. 503 * 504 * Since we use these as temporary stacks during secondary CPU 505 * bringup, we need to get at them in real mode. This means they 506 * must also be within the RMO region. 507 */ 508 limit = min(0x10000000UL, lmb.rmo_size); 509 510 for_each_cpu(i) 511 paca[i].emergency_sp = 512 __va(lmb_alloc_base(HW_PAGE_SIZE, 128, limit)) + HW_PAGE_SIZE; 513} 514 515/* 516 * Called into from start_kernel, after lock_kernel has been called. 517 * Initializes bootmem, which is unsed to manage page allocation until 518 * mem_init is called. 519 */ 520void __init setup_arch(char **cmdline_p) 521{ 522 extern void do_init_bootmem(void); 523 524 ppc64_boot_msg(0x12, "Setup Arch"); 525 526 *cmdline_p = cmd_line; 527 528 /* 529 * Set cache line size based on type of cpu as a default. 530 * Systems with OF can look in the properties on the cpu node(s) 531 * for a possibly more accurate value. 532 */ 533 dcache_bsize = ppc64_caches.dline_size; 534 icache_bsize = ppc64_caches.iline_size; 535 536 /* reboot on panic */ 537 panic_timeout = 180; 538 539 if (ppc_md.panic) 540 notifier_chain_register(&panic_notifier_list, &ppc64_panic_block); 541 542 init_mm.start_code = PAGE_OFFSET; 543 init_mm.end_code = (unsigned long) _etext; 544 init_mm.end_data = (unsigned long) _edata; 545 init_mm.brk = klimit; 546 547 irqstack_early_init(); 548 emergency_stack_init(); 549 550 stabs_alloc(); 551 552 /* set up the bootmem stuff with available memory */ 553 do_init_bootmem(); 554 sparse_init(); 555 556#ifdef CONFIG_DUMMY_CONSOLE 557 conswitchp = &dummy_con; 558#endif 559 560 ppc_md.setup_arch(); 561 562 paging_init(); 563 ppc64_boot_msg(0x15, "Setup Done"); 564} 565 566 567/* ToDo: do something useful if ppc_md is not yet setup. */ 568#define PPC64_LINUX_FUNCTION 0x0f000000 569#define PPC64_IPL_MESSAGE 0xc0000000 570#define PPC64_TERM_MESSAGE 0xb0000000 571 572static void ppc64_do_msg(unsigned int src, const char *msg) 573{ 574 if (ppc_md.progress) { 575 char buf[128]; 576 577 sprintf(buf, "%08X\n", src); 578 ppc_md.progress(buf, 0); 579 snprintf(buf, 128, "%s", msg); 580 ppc_md.progress(buf, 0); 581 } 582} 583 584/* Print a boot progress message. */ 585void ppc64_boot_msg(unsigned int src, const char *msg) 586{ 587 ppc64_do_msg(PPC64_LINUX_FUNCTION|PPC64_IPL_MESSAGE|src, msg); 588 printk("[boot]%04x %s\n", src, msg); 589} 590 591/* Print a termination message (print only -- does not stop the kernel) */ 592void ppc64_terminate_msg(unsigned int src, const char *msg) 593{ 594 ppc64_do_msg(PPC64_LINUX_FUNCTION|PPC64_TERM_MESSAGE|src, msg); 595 printk("[terminate]%04x %s\n", src, msg); 596} 597 598int check_legacy_ioport(unsigned long base_port) 599{ 600 if (ppc_md.check_legacy_ioport == NULL) 601 return 0; 602 return ppc_md.check_legacy_ioport(base_port); 603} 604EXPORT_SYMBOL(check_legacy_ioport); 605 606void cpu_die(void) 607{ 608 if (ppc_md.cpu_die) 609 ppc_md.cpu_die(); 610} 611 612#ifdef CONFIG_SMP 613void __init setup_per_cpu_areas(void) 614{ 615 int i; 616 unsigned long size; 617 char *ptr; 618 619 /* Copy section for each CPU (we discard the original) */ 620 size = ALIGN(__per_cpu_end - __per_cpu_start, SMP_CACHE_BYTES); 621#ifdef CONFIG_MODULES 622 if (size < PERCPU_ENOUGH_ROOM) 623 size = PERCPU_ENOUGH_ROOM; 624#endif 625 626 for_each_cpu(i) { 627 ptr = alloc_bootmem_node(NODE_DATA(cpu_to_node(i)), size); 628 if (!ptr) 629 panic("Cannot allocate cpu data for CPU %d\n", i); 630 631 paca[i].data_offset = ptr - __per_cpu_start; 632 memcpy(ptr, __per_cpu_start, __per_cpu_end - __per_cpu_start); 633 } 634} 635#endif 636