setup_64.c revision aa98c50dcb5d5b85d2a4c26d54fa1e3c31c11e4b
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 */ 101 102 103#ifdef CONFIG_SMP 104 105static int smt_enabled_cmdline; 106 107/* Look for ibm,smt-enabled OF option */ 108static void check_smt_enabled(void) 109{ 110 struct device_node *dn; 111 char *smt_option; 112 113 /* Allow the command line to overrule the OF option */ 114 if (smt_enabled_cmdline) 115 return; 116 117 dn = of_find_node_by_path("/options"); 118 119 if (dn) { 120 smt_option = (char *)get_property(dn, "ibm,smt-enabled", NULL); 121 122 if (smt_option) { 123 if (!strcmp(smt_option, "on")) 124 smt_enabled_at_boot = 1; 125 else if (!strcmp(smt_option, "off")) 126 smt_enabled_at_boot = 0; 127 } 128 } 129} 130 131/* Look for smt-enabled= cmdline option */ 132static int __init early_smt_enabled(char *p) 133{ 134 smt_enabled_cmdline = 1; 135 136 if (!p) 137 return 0; 138 139 if (!strcmp(p, "on") || !strcmp(p, "1")) 140 smt_enabled_at_boot = 1; 141 else if (!strcmp(p, "off") || !strcmp(p, "0")) 142 smt_enabled_at_boot = 0; 143 144 return 0; 145} 146early_param("smt-enabled", early_smt_enabled); 147 148#else 149#define check_smt_enabled() 150#endif /* CONFIG_SMP */ 151 152/* 153 * Early initialization entry point. This is called by head.S 154 * with MMU translation disabled. We rely on the "feature" of 155 * the CPU that ignores the top 2 bits of the address in real 156 * mode so we can access kernel globals normally provided we 157 * only toy with things in the RMO region. From here, we do 158 * some early parsing of the device-tree to setup out LMB 159 * data structures, and allocate & initialize the hash table 160 * and segment tables so we can start running with translation 161 * enabled. 162 * 163 * It is this function which will call the probe() callback of 164 * the various platform types and copy the matching one to the 165 * global ppc_md structure. Your platform can eventually do 166 * some very early initializations from the probe() routine, but 167 * this is not recommended, be very careful as, for example, the 168 * device-tree is not accessible via normal means at this point. 169 */ 170 171void __init early_setup(unsigned long dt_ptr) 172{ 173 /* Enable early debugging if any specified (see udbg.h) */ 174 udbg_early_init(); 175 176 DBG(" -> early_setup(), dt_ptr: 0x%lx\n", dt_ptr); 177 178 /* 179 * Do early initializations using the flattened device 180 * tree, like retreiving the physical memory map or 181 * calculating/retreiving the hash table size 182 */ 183 early_init_devtree(__va(dt_ptr)); 184 185 /* Now we know the logical id of our boot cpu, setup the paca. */ 186 setup_boot_paca(); 187 188 /* Fix up paca fields required for the boot cpu */ 189 get_paca()->cpu_start = 1; 190 get_paca()->stab_real = __pa((u64)&initial_stab); 191 get_paca()->stab_addr = (u64)&initial_stab; 192 193 /* Probe the machine type */ 194 probe_machine(); 195 196 setup_kdump_trampoline(); 197 198 DBG("Found, Initializing memory management...\n"); 199 200 /* 201 * Initialize the MMU Hash table and create the linear mapping 202 * of memory. Has to be done before stab/slb initialization as 203 * this is currently where the page size encoding is obtained 204 */ 205 htab_initialize(); 206 207 /* 208 * Initialize stab / SLB management except on iSeries 209 */ 210 if (cpu_has_feature(CPU_FTR_SLB)) 211 slb_initialize(); 212 else if (!firmware_has_feature(FW_FEATURE_ISERIES)) 213 stab_initialize(get_paca()->stab_real); 214 215 DBG(" <- early_setup()\n"); 216} 217 218#ifdef CONFIG_SMP 219void early_setup_secondary(void) 220{ 221 struct paca_struct *lpaca = get_paca(); 222 223 /* Mark enabled in PACA */ 224 lpaca->proc_enabled = 0; 225 226 /* Initialize hash table for that CPU */ 227 htab_initialize_secondary(); 228 229 /* Initialize STAB/SLB. We use a virtual address as it works 230 * in real mode on pSeries and we want a virutal address on 231 * iSeries anyway 232 */ 233 if (cpu_has_feature(CPU_FTR_SLB)) 234 slb_initialize(); 235 else 236 stab_initialize(lpaca->stab_addr); 237} 238 239#endif /* CONFIG_SMP */ 240 241#if defined(CONFIG_SMP) || defined(CONFIG_KEXEC) 242void smp_release_cpus(void) 243{ 244 extern unsigned long __secondary_hold_spinloop; 245 unsigned long *ptr; 246 247 DBG(" -> smp_release_cpus()\n"); 248 249 /* All secondary cpus are spinning on a common spinloop, release them 250 * all now so they can start to spin on their individual paca 251 * spinloops. For non SMP kernels, the secondary cpus never get out 252 * of the common spinloop. 253 * This is useless but harmless on iSeries, secondaries are already 254 * waiting on their paca spinloops. */ 255 256 ptr = (unsigned long *)((unsigned long)&__secondary_hold_spinloop 257 - PHYSICAL_START); 258 *ptr = 1; 259 mb(); 260 261 DBG(" <- smp_release_cpus()\n"); 262} 263#endif /* CONFIG_SMP || CONFIG_KEXEC */ 264 265/* 266 * Initialize some remaining members of the ppc64_caches and systemcfg 267 * structures 268 * (at least until we get rid of them completely). This is mostly some 269 * cache informations about the CPU that will be used by cache flush 270 * routines and/or provided to userland 271 */ 272static void __init initialize_cache_info(void) 273{ 274 struct device_node *np; 275 unsigned long num_cpus = 0; 276 277 DBG(" -> initialize_cache_info()\n"); 278 279 for (np = NULL; (np = of_find_node_by_type(np, "cpu"));) { 280 num_cpus += 1; 281 282 /* We're assuming *all* of the CPUs have the same 283 * d-cache and i-cache sizes... -Peter 284 */ 285 286 if ( num_cpus == 1 ) { 287 u32 *sizep, *lsizep; 288 u32 size, lsize; 289 const char *dc, *ic; 290 291 /* Then read cache informations */ 292 if (machine_is(powermac)) { 293 dc = "d-cache-block-size"; 294 ic = "i-cache-block-size"; 295 } else { 296 dc = "d-cache-line-size"; 297 ic = "i-cache-line-size"; 298 } 299 300 size = 0; 301 lsize = cur_cpu_spec->dcache_bsize; 302 sizep = (u32 *)get_property(np, "d-cache-size", NULL); 303 if (sizep != NULL) 304 size = *sizep; 305 lsizep = (u32 *) get_property(np, dc, NULL); 306 if (lsizep != NULL) 307 lsize = *lsizep; 308 if (sizep == 0 || lsizep == 0) 309 DBG("Argh, can't find dcache properties ! " 310 "sizep: %p, lsizep: %p\n", sizep, lsizep); 311 312 ppc64_caches.dsize = size; 313 ppc64_caches.dline_size = lsize; 314 ppc64_caches.log_dline_size = __ilog2(lsize); 315 ppc64_caches.dlines_per_page = PAGE_SIZE / lsize; 316 317 size = 0; 318 lsize = cur_cpu_spec->icache_bsize; 319 sizep = (u32 *)get_property(np, "i-cache-size", NULL); 320 if (sizep != NULL) 321 size = *sizep; 322 lsizep = (u32 *)get_property(np, ic, NULL); 323 if (lsizep != NULL) 324 lsize = *lsizep; 325 if (sizep == 0 || lsizep == 0) 326 DBG("Argh, can't find icache properties ! " 327 "sizep: %p, lsizep: %p\n", sizep, lsizep); 328 329 ppc64_caches.isize = size; 330 ppc64_caches.iline_size = lsize; 331 ppc64_caches.log_iline_size = __ilog2(lsize); 332 ppc64_caches.ilines_per_page = PAGE_SIZE / lsize; 333 } 334 } 335 336 DBG(" <- initialize_cache_info()\n"); 337} 338 339 340/* 341 * Do some initial setup of the system. The parameters are those which 342 * were passed in from the bootloader. 343 */ 344void __init setup_system(void) 345{ 346 DBG(" -> setup_system()\n"); 347 348 /* 349 * Unflatten the device-tree passed by prom_init or kexec 350 */ 351 unflatten_device_tree(); 352 353 /* 354 * Fill the ppc64_caches & systemcfg structures with informations 355 * retrieved from the device-tree. Need to be called before 356 * finish_device_tree() since the later requires some of the 357 * informations filled up here to properly parse the interrupt tree. 358 */ 359 initialize_cache_info(); 360 361#ifdef CONFIG_PPC_RTAS 362 /* 363 * Initialize RTAS if available 364 */ 365 rtas_initialize(); 366#endif /* CONFIG_PPC_RTAS */ 367 368 /* 369 * Check if we have an initrd provided via the device-tree 370 */ 371 check_for_initrd(); 372 373 /* 374 * Do some platform specific early initializations, that includes 375 * setting up the hash table pointers. It also sets up some interrupt-mapping 376 * related options that will be used by finish_device_tree() 377 */ 378 ppc_md.init_early(); 379 380 /* 381 * We can discover serial ports now since the above did setup the 382 * hash table management for us, thus ioremap works. We do that early 383 * so that further code can be debugged 384 */ 385 find_legacy_serial_ports(); 386 387 /* 388 * "Finish" the device-tree, that is do the actual parsing of 389 * some of the properties like the interrupt map 390 */ 391 finish_device_tree(); 392 393 /* 394 * Initialize xmon 395 */ 396#ifdef CONFIG_XMON_DEFAULT 397 xmon_init(1); 398#endif 399 /* 400 * Register early console 401 */ 402 register_early_udbg_console(); 403 404 if (do_early_xmon) 405 debugger(NULL); 406 407 check_smt_enabled(); 408 smp_setup_cpu_maps(); 409 410#ifdef CONFIG_SMP 411 /* Release secondary cpus out of their spinloops at 0x60 now that 412 * we can map physical -> logical CPU ids 413 */ 414 smp_release_cpus(); 415#endif 416 417 printk("Starting Linux PPC64 %s\n", system_utsname.version); 418 419 printk("-----------------------------------------------------\n"); 420 printk("ppc64_pft_size = 0x%lx\n", ppc64_pft_size); 421 printk("ppc64_interrupt_controller = 0x%ld\n", 422 ppc64_interrupt_controller); 423 printk("physicalMemorySize = 0x%lx\n", lmb_phys_mem_size()); 424 printk("ppc64_caches.dcache_line_size = 0x%x\n", 425 ppc64_caches.dline_size); 426 printk("ppc64_caches.icache_line_size = 0x%x\n", 427 ppc64_caches.iline_size); 428 printk("htab_address = 0x%p\n", htab_address); 429 printk("htab_hash_mask = 0x%lx\n", htab_hash_mask); 430#if PHYSICAL_START > 0 431 printk("physical_start = 0x%x\n", PHYSICAL_START); 432#endif 433 printk("-----------------------------------------------------\n"); 434 435 DBG(" <- setup_system()\n"); 436} 437 438#ifdef CONFIG_IRQSTACKS 439static void __init irqstack_early_init(void) 440{ 441 unsigned int i; 442 443 /* 444 * interrupt stacks must be under 256MB, we cannot afford to take 445 * SLB misses on them. 446 */ 447 for_each_possible_cpu(i) { 448 softirq_ctx[i] = (struct thread_info *) 449 __va(lmb_alloc_base(THREAD_SIZE, 450 THREAD_SIZE, 0x10000000)); 451 hardirq_ctx[i] = (struct thread_info *) 452 __va(lmb_alloc_base(THREAD_SIZE, 453 THREAD_SIZE, 0x10000000)); 454 } 455} 456#else 457#define irqstack_early_init() 458#endif 459 460/* 461 * Stack space used when we detect a bad kernel stack pointer, and 462 * early in SMP boots before relocation is enabled. 463 */ 464static void __init emergency_stack_init(void) 465{ 466 unsigned long limit; 467 unsigned int i; 468 469 /* 470 * Emergency stacks must be under 256MB, we cannot afford to take 471 * SLB misses on them. The ABI also requires them to be 128-byte 472 * aligned. 473 * 474 * Since we use these as temporary stacks during secondary CPU 475 * bringup, we need to get at them in real mode. This means they 476 * must also be within the RMO region. 477 */ 478 limit = min(0x10000000UL, lmb.rmo_size); 479 480 for_each_possible_cpu(i) 481 paca[i].emergency_sp = 482 __va(lmb_alloc_base(HW_PAGE_SIZE, 128, limit)) + HW_PAGE_SIZE; 483} 484 485/* 486 * Called into from start_kernel, after lock_kernel has been called. 487 * Initializes bootmem, which is unsed to manage page allocation until 488 * mem_init is called. 489 */ 490void __init setup_arch(char **cmdline_p) 491{ 492 ppc64_boot_msg(0x12, "Setup Arch"); 493 494 *cmdline_p = cmd_line; 495 496 /* 497 * Set cache line size based on type of cpu as a default. 498 * Systems with OF can look in the properties on the cpu node(s) 499 * for a possibly more accurate value. 500 */ 501 dcache_bsize = ppc64_caches.dline_size; 502 icache_bsize = ppc64_caches.iline_size; 503 504 /* reboot on panic */ 505 panic_timeout = 180; 506 507 if (ppc_md.panic) 508 setup_panic(); 509 510 init_mm.start_code = PAGE_OFFSET; 511 init_mm.end_code = (unsigned long) _etext; 512 init_mm.end_data = (unsigned long) _edata; 513 init_mm.brk = klimit; 514 515 irqstack_early_init(); 516 emergency_stack_init(); 517 518 stabs_alloc(); 519 520 /* set up the bootmem stuff with available memory */ 521 do_init_bootmem(); 522 sparse_init(); 523 524#ifdef CONFIG_DUMMY_CONSOLE 525 conswitchp = &dummy_con; 526#endif 527 528 ppc_md.setup_arch(); 529 530 paging_init(); 531 ppc64_boot_msg(0x15, "Setup Done"); 532} 533 534 535/* ToDo: do something useful if ppc_md is not yet setup. */ 536#define PPC64_LINUX_FUNCTION 0x0f000000 537#define PPC64_IPL_MESSAGE 0xc0000000 538#define PPC64_TERM_MESSAGE 0xb0000000 539 540static void ppc64_do_msg(unsigned int src, const char *msg) 541{ 542 if (ppc_md.progress) { 543 char buf[128]; 544 545 sprintf(buf, "%08X\n", src); 546 ppc_md.progress(buf, 0); 547 snprintf(buf, 128, "%s", msg); 548 ppc_md.progress(buf, 0); 549 } 550} 551 552/* Print a boot progress message. */ 553void ppc64_boot_msg(unsigned int src, const char *msg) 554{ 555 ppc64_do_msg(PPC64_LINUX_FUNCTION|PPC64_IPL_MESSAGE|src, msg); 556 printk("[boot]%04x %s\n", src, msg); 557} 558 559/* Print a termination message (print only -- does not stop the kernel) */ 560void ppc64_terminate_msg(unsigned int src, const char *msg) 561{ 562 ppc64_do_msg(PPC64_LINUX_FUNCTION|PPC64_TERM_MESSAGE|src, msg); 563 printk("[terminate]%04x %s\n", src, msg); 564} 565 566void cpu_die(void) 567{ 568 if (ppc_md.cpu_die) 569 ppc_md.cpu_die(); 570} 571 572#ifdef CONFIG_SMP 573void __init setup_per_cpu_areas(void) 574{ 575 int i; 576 unsigned long size; 577 char *ptr; 578 579 /* Copy section for each CPU (we discard the original) */ 580 size = ALIGN(__per_cpu_end - __per_cpu_start, SMP_CACHE_BYTES); 581#ifdef CONFIG_MODULES 582 if (size < PERCPU_ENOUGH_ROOM) 583 size = PERCPU_ENOUGH_ROOM; 584#endif 585 586 for_each_possible_cpu(i) { 587 ptr = alloc_bootmem_node(NODE_DATA(cpu_to_node(i)), size); 588 if (!ptr) 589 panic("Cannot allocate cpu data for CPU %d\n", i); 590 591 paca[i].data_offset = ptr - __per_cpu_start; 592 memcpy(ptr, __per_cpu_start, __per_cpu_end - __per_cpu_start); 593 } 594} 595#endif 596