3 * Common boot and setup code.
5 * Copyright (C) 2001 PPC64 Team, IBM Corp
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.
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/ide.h>
24 #include <linux/seq_file.h>
25 #include <linux/ioport.h>
26 #include <linux/console.h>
27 #include <linux/utsname.h>
28 #include <linux/tty.h>
29 #include <linux/root_dev.h>
30 #include <linux/notifier.h>
31 #include <linux/cpu.h>
32 #include <linux/unistd.h>
33 #include <linux/serial.h>
34 #include <linux/serial_8250.h>
35 #include <linux/bootmem.h>
37 #include <asm/kdump.h>
39 #include <asm/processor.h>
40 #include <asm/pgtable.h>
43 #include <asm/machdep.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>
53 #include <asm/iommu.h>
54 #include <asm/serial.h>
55 #include <asm/cache.h>
59 #include <asm/firmware.h>
62 #include <asm/kexec.h>
67 #define DBG(fmt...) udbg_printf(fmt)
77 /* Pick defaults since we might want to patch instructions
78 * before we've read this from the device tree.
80 struct ppc64_caches ppc64_caches = {
86 EXPORT_SYMBOL_GPL(ppc64_caches);
89 * These are used in binfmt_elf.c to put aux entries on the stack
90 * for each elf executable being started.
98 static int smt_enabled_cmdline;
100 /* Look for ibm,smt-enabled OF option */
101 static void check_smt_enabled(void)
103 struct device_node *dn;
104 const char *smt_option;
106 /* Allow the command line to overrule the OF option */
107 if (smt_enabled_cmdline)
110 dn = of_find_node_by_path("/options");
113 smt_option = get_property(dn, "ibm,smt-enabled", NULL);
116 if (!strcmp(smt_option, "on"))
117 smt_enabled_at_boot = 1;
118 else if (!strcmp(smt_option, "off"))
119 smt_enabled_at_boot = 0;
124 /* Look for smt-enabled= cmdline option */
125 static int __init early_smt_enabled(char *p)
127 smt_enabled_cmdline = 1;
132 if (!strcmp(p, "on") || !strcmp(p, "1"))
133 smt_enabled_at_boot = 1;
134 else if (!strcmp(p, "off") || !strcmp(p, "0"))
135 smt_enabled_at_boot = 0;
139 early_param("smt-enabled", early_smt_enabled);
142 #define check_smt_enabled()
143 #endif /* CONFIG_SMP */
145 /* Put the paca pointer into r13 and SPRG3 */
146 void __init setup_paca(int cpu)
148 local_paca = &paca[cpu];
149 mtspr(SPRN_SPRG3, local_paca);
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
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.
171 void __init early_setup(unsigned long dt_ptr)
173 /* Identify CPU type */
176 /* Assume we're on cpu 0 for now. Don't write to the paca yet! */
179 /* Enable early debugging if any specified (see udbg.h) */
182 DBG(" -> early_setup(), dt_ptr: 0x%lx\n", dt_ptr);
185 * Do early initializations using the flattened device
186 * tree, like retreiving the physical memory map or
187 * calculating/retreiving the hash table size
189 early_init_devtree(__va(dt_ptr));
191 /* Now we know the logical id of our boot cpu, setup the paca. */
192 setup_paca(boot_cpuid);
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;
199 /* Probe the machine type */
202 setup_kdump_trampoline();
204 DBG("Found, Initializing memory management...\n");
207 * Initialize the MMU Hash table and create the linear mapping
208 * of memory. Has to be done before stab/slb initialization as
209 * this is currently where the page size encoding is obtained
214 * Initialize stab / SLB management except on iSeries
216 if (cpu_has_feature(CPU_FTR_SLB))
218 else if (!firmware_has_feature(FW_FEATURE_ISERIES))
219 stab_initialize(get_paca()->stab_real);
221 DBG(" <- early_setup()\n");
225 void early_setup_secondary(void)
227 struct paca_struct *lpaca = get_paca();
229 /* Mark enabled in PACA */
230 lpaca->proc_enabled = 0;
232 /* Initialize hash table for that CPU */
233 htab_initialize_secondary();
235 /* Initialize STAB/SLB. We use a virtual address as it works
236 * in real mode on pSeries and we want a virutal address on
239 if (cpu_has_feature(CPU_FTR_SLB))
242 stab_initialize(lpaca->stab_addr);
245 #endif /* CONFIG_SMP */
247 #if defined(CONFIG_SMP) || defined(CONFIG_KEXEC)
248 void smp_release_cpus(void)
250 extern unsigned long __secondary_hold_spinloop;
253 DBG(" -> smp_release_cpus()\n");
255 /* All secondary cpus are spinning on a common spinloop, release them
256 * all now so they can start to spin on their individual paca
257 * spinloops. For non SMP kernels, the secondary cpus never get out
258 * of the common spinloop.
259 * This is useless but harmless on iSeries, secondaries are already
260 * waiting on their paca spinloops. */
262 ptr = (unsigned long *)((unsigned long)&__secondary_hold_spinloop
267 DBG(" <- smp_release_cpus()\n");
269 #endif /* CONFIG_SMP || CONFIG_KEXEC */
272 * Initialize some remaining members of the ppc64_caches and systemcfg
274 * (at least until we get rid of them completely). This is mostly some
275 * cache informations about the CPU that will be used by cache flush
276 * routines and/or provided to userland
278 static void __init initialize_cache_info(void)
280 struct device_node *np;
281 unsigned long num_cpus = 0;
283 DBG(" -> initialize_cache_info()\n");
285 for (np = NULL; (np = of_find_node_by_type(np, "cpu"));) {
288 /* We're assuming *all* of the CPUs have the same
289 * d-cache and i-cache sizes... -Peter
292 if ( num_cpus == 1 ) {
293 const u32 *sizep, *lsizep;
297 /* Then read cache informations */
298 if (machine_is(powermac)) {
299 dc = "d-cache-block-size";
300 ic = "i-cache-block-size";
302 dc = "d-cache-line-size";
303 ic = "i-cache-line-size";
307 lsize = cur_cpu_spec->dcache_bsize;
308 sizep = get_property(np, "d-cache-size", NULL);
311 lsizep = get_property(np, dc, NULL);
314 if (sizep == 0 || lsizep == 0)
315 DBG("Argh, can't find dcache properties ! "
316 "sizep: %p, lsizep: %p\n", sizep, lsizep);
318 ppc64_caches.dsize = size;
319 ppc64_caches.dline_size = lsize;
320 ppc64_caches.log_dline_size = __ilog2(lsize);
321 ppc64_caches.dlines_per_page = PAGE_SIZE / lsize;
324 lsize = cur_cpu_spec->icache_bsize;
325 sizep = get_property(np, "i-cache-size", NULL);
328 lsizep = get_property(np, ic, NULL);
331 if (sizep == 0 || lsizep == 0)
332 DBG("Argh, can't find icache properties ! "
333 "sizep: %p, lsizep: %p\n", sizep, lsizep);
335 ppc64_caches.isize = size;
336 ppc64_caches.iline_size = lsize;
337 ppc64_caches.log_iline_size = __ilog2(lsize);
338 ppc64_caches.ilines_per_page = PAGE_SIZE / lsize;
342 DBG(" <- initialize_cache_info()\n");
347 * Do some initial setup of the system. The parameters are those which
348 * were passed in from the bootloader.
350 void __init setup_system(void)
352 DBG(" -> setup_system()\n");
354 /* Apply the CPUs-specific and firmware specific fixups to kernel
355 * text (nop out sections not relevant to this CPU or this firmware)
357 do_feature_fixups(cur_cpu_spec->cpu_features,
358 &__start___ftr_fixup, &__stop___ftr_fixup);
359 do_feature_fixups(powerpc_firmware_features,
360 &__start___fw_ftr_fixup, &__stop___fw_ftr_fixup);
363 * Unflatten the device-tree passed by prom_init or kexec
365 unflatten_device_tree();
368 * Fill the ppc64_caches & systemcfg structures with informations
369 * retrieved from the device-tree.
371 initialize_cache_info();
374 * Initialize irq remapping subsystem
378 #ifdef CONFIG_PPC_RTAS
380 * Initialize RTAS if available
383 #endif /* CONFIG_PPC_RTAS */
386 * Check if we have an initrd provided via the device-tree
391 * Do some platform specific early initializations, that includes
392 * setting up the hash table pointers. It also sets up some interrupt-mapping
393 * related options that will be used by finish_device_tree()
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
402 find_legacy_serial_ports();
405 * Register early console
407 register_early_udbg_console();
415 smp_setup_cpu_maps();
418 /* Release secondary cpus out of their spinloops at 0x60 now that
419 * we can map physical -> logical CPU ids
424 printk("Starting Linux PPC64 %s\n", init_utsname()->version);
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 printk("ppc64_caches.dcache_line_size = 0x%x\n",
430 ppc64_caches.dline_size);
431 printk("ppc64_caches.icache_line_size = 0x%x\n",
432 ppc64_caches.iline_size);
433 printk("htab_address = 0x%p\n", htab_address);
434 printk("htab_hash_mask = 0x%lx\n", htab_hash_mask);
435 #if PHYSICAL_START > 0
436 printk("physical_start = 0x%x\n", PHYSICAL_START);
438 printk("-----------------------------------------------------\n");
440 DBG(" <- setup_system()\n");
443 #ifdef CONFIG_IRQSTACKS
444 static void __init irqstack_early_init(void)
449 * interrupt stacks must be under 256MB, we cannot afford to take
450 * SLB misses on them.
452 for_each_possible_cpu(i) {
453 softirq_ctx[i] = (struct thread_info *)
454 __va(lmb_alloc_base(THREAD_SIZE,
455 THREAD_SIZE, 0x10000000));
456 hardirq_ctx[i] = (struct thread_info *)
457 __va(lmb_alloc_base(THREAD_SIZE,
458 THREAD_SIZE, 0x10000000));
462 #define irqstack_early_init()
466 * Stack space used when we detect a bad kernel stack pointer, and
467 * early in SMP boots before relocation is enabled.
469 static void __init emergency_stack_init(void)
475 * Emergency stacks must be under 256MB, we cannot afford to take
476 * SLB misses on them. The ABI also requires them to be 128-byte
479 * Since we use these as temporary stacks during secondary CPU
480 * bringup, we need to get at them in real mode. This means they
481 * must also be within the RMO region.
483 limit = min(0x10000000UL, lmb.rmo_size);
485 for_each_possible_cpu(i)
486 paca[i].emergency_sp =
487 __va(lmb_alloc_base(HW_PAGE_SIZE, 128, limit)) + HW_PAGE_SIZE;
491 * Called into from start_kernel, after lock_kernel has been called.
492 * Initializes bootmem, which is unsed to manage page allocation until
493 * mem_init is called.
495 void __init setup_arch(char **cmdline_p)
497 ppc64_boot_msg(0x12, "Setup Arch");
499 *cmdline_p = cmd_line;
502 * Set cache line size based on type of cpu as a default.
503 * Systems with OF can look in the properties on the cpu node(s)
504 * for a possibly more accurate value.
506 dcache_bsize = ppc64_caches.dline_size;
507 icache_bsize = ppc64_caches.iline_size;
509 /* reboot on panic */
515 init_mm.start_code = PAGE_OFFSET;
516 init_mm.end_code = (unsigned long) _etext;
517 init_mm.end_data = (unsigned long) _edata;
518 init_mm.brk = klimit;
520 irqstack_early_init();
521 emergency_stack_init();
525 /* set up the bootmem stuff with available memory */
529 #ifdef CONFIG_DUMMY_CONSOLE
530 conswitchp = &dummy_con;
536 ppc64_boot_msg(0x15, "Setup Done");
540 /* ToDo: do something useful if ppc_md is not yet setup. */
541 #define PPC64_LINUX_FUNCTION 0x0f000000
542 #define PPC64_IPL_MESSAGE 0xc0000000
543 #define PPC64_TERM_MESSAGE 0xb0000000
545 static void ppc64_do_msg(unsigned int src, const char *msg)
547 if (ppc_md.progress) {
550 sprintf(buf, "%08X\n", src);
551 ppc_md.progress(buf, 0);
552 snprintf(buf, 128, "%s", msg);
553 ppc_md.progress(buf, 0);
557 /* Print a boot progress message. */
558 void ppc64_boot_msg(unsigned int src, const char *msg)
560 ppc64_do_msg(PPC64_LINUX_FUNCTION|PPC64_IPL_MESSAGE|src, msg);
561 printk("[boot]%04x %s\n", src, msg);
564 /* Print a termination message (print only -- does not stop the kernel) */
565 void ppc64_terminate_msg(unsigned int src, const char *msg)
567 ppc64_do_msg(PPC64_LINUX_FUNCTION|PPC64_TERM_MESSAGE|src, msg);
568 printk("[terminate]%04x %s\n", src, msg);
578 void __init setup_per_cpu_areas(void)
584 /* Copy section for each CPU (we discard the original) */
585 size = ALIGN(__per_cpu_end - __per_cpu_start, SMP_CACHE_BYTES);
586 #ifdef CONFIG_MODULES
587 if (size < PERCPU_ENOUGH_ROOM)
588 size = PERCPU_ENOUGH_ROOM;
591 for_each_possible_cpu(i) {
592 ptr = alloc_bootmem_node(NODE_DATA(cpu_to_node(i)), size);
594 panic("Cannot allocate cpu data for CPU %d\n", i);
596 paca[i].data_offset = ptr - __per_cpu_start;
597 memcpy(ptr, __per_cpu_start, __per_cpu_end - __per_cpu_start);