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/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>
38 #include <asm/kdump.h>
40 #include <asm/processor.h>
41 #include <asm/pgtable.h>
44 #include <asm/machdep.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>
54 #include <asm/iommu.h>
55 #include <asm/serial.h>
56 #include <asm/cache.h>
60 #include <asm/iseries/it_lp_naca.h>
61 #include <asm/firmware.h>
64 #include <asm/kexec.h>
69 #define DBG(fmt...) udbg_printf(fmt)
76 int boot_cpuid_phys = 0;
80 /* Pick defaults since we might want to patch instructions
81 * before we've read this from the device tree.
83 struct ppc64_caches ppc64_caches = {
89 EXPORT_SYMBOL_GPL(ppc64_caches);
92 * These are used in binfmt_elf.c to put aux entries on the stack
93 * for each elf executable being started.
99 /* The main machine-dep calls structure
101 struct machdep_calls ppc_md;
102 EXPORT_SYMBOL(ppc_md);
104 #ifdef CONFIG_MAGIC_SYSRQ
105 unsigned long SYSRQ_KEY;
106 #endif /* CONFIG_MAGIC_SYSRQ */
109 static int ppc64_panic_event(struct notifier_block *, unsigned long, void *);
110 static struct notifier_block ppc64_panic_block = {
111 .notifier_call = ppc64_panic_event,
112 .priority = INT_MIN /* may not return; must be done last */
117 static int smt_enabled_cmdline;
119 /* Look for ibm,smt-enabled OF option */
120 static void check_smt_enabled(void)
122 struct device_node *dn;
125 /* Allow the command line to overrule the OF option */
126 if (smt_enabled_cmdline)
129 dn = of_find_node_by_path("/options");
132 smt_option = (char *)get_property(dn, "ibm,smt-enabled", NULL);
135 if (!strcmp(smt_option, "on"))
136 smt_enabled_at_boot = 1;
137 else if (!strcmp(smt_option, "off"))
138 smt_enabled_at_boot = 0;
143 /* Look for smt-enabled= cmdline option */
144 static int __init early_smt_enabled(char *p)
146 smt_enabled_cmdline = 1;
151 if (!strcmp(p, "on") || !strcmp(p, "1"))
152 smt_enabled_at_boot = 1;
153 else if (!strcmp(p, "off") || !strcmp(p, "0"))
154 smt_enabled_at_boot = 0;
158 early_param("smt-enabled", early_smt_enabled);
161 #define check_smt_enabled()
162 #endif /* CONFIG_SMP */
164 extern struct machdep_calls pSeries_md;
165 extern struct machdep_calls pmac_md;
166 extern struct machdep_calls maple_md;
167 extern struct machdep_calls cell_md;
168 extern struct machdep_calls iseries_md;
170 /* Ultimately, stuff them in an elf section like initcalls... */
171 static struct machdep_calls __initdata *machines[] = {
172 #ifdef CONFIG_PPC_PSERIES
174 #endif /* CONFIG_PPC_PSERIES */
175 #ifdef CONFIG_PPC_PMAC
177 #endif /* CONFIG_PPC_PMAC */
178 #ifdef CONFIG_PPC_MAPLE
180 #endif /* CONFIG_PPC_MAPLE */
181 #ifdef CONFIG_PPC_CELL
184 #ifdef CONFIG_PPC_ISERIES
191 * Early initialization entry point. This is called by head.S
192 * with MMU translation disabled. We rely on the "feature" of
193 * the CPU that ignores the top 2 bits of the address in real
194 * mode so we can access kernel globals normally provided we
195 * only toy with things in the RMO region. From here, we do
196 * some early parsing of the device-tree to setup out LMB
197 * data structures, and allocate & initialize the hash table
198 * and segment tables so we can start running with translation
201 * It is this function which will call the probe() callback of
202 * the various platform types and copy the matching one to the
203 * global ppc_md structure. Your platform can eventually do
204 * some very early initializations from the probe() routine, but
205 * this is not recommended, be very careful as, for example, the
206 * device-tree is not accessible via normal means at this point.
209 void __init early_setup(unsigned long dt_ptr)
211 struct paca_struct *lpaca = get_paca();
212 static struct machdep_calls **mach;
214 /* Enable early debugging if any specified (see udbg.h) */
217 DBG(" -> early_setup()\n");
220 * Do early initializations using the flattened device
221 * tree, like retreiving the physical memory map or
222 * calculating/retreiving the hash table size
224 early_init_devtree(__va(dt_ptr));
227 * Iterate all ppc_md structures until we find the proper
228 * one for the current machine type
230 DBG("Probing machine type for platform %x...\n", _machine);
232 for (mach = machines; *mach; mach++) {
233 if ((*mach)->probe(_machine))
236 /* What can we do if we didn't find ? */
238 DBG("No suitable machine found !\n");
243 #ifdef CONFIG_CRASH_DUMP
247 DBG("Found, Initializing memory management...\n");
250 * Initialize the MMU Hash table and create the linear mapping
251 * of memory. Has to be done before stab/slb initialization as
252 * this is currently where the page size encoding is obtained
257 * Initialize stab / SLB management except on iSeries
259 if (!firmware_has_feature(FW_FEATURE_ISERIES)) {
260 if (cpu_has_feature(CPU_FTR_SLB))
263 stab_initialize(lpaca->stab_real);
266 DBG(" <- early_setup()\n");
270 void early_setup_secondary(void)
272 struct paca_struct *lpaca = get_paca();
274 /* Mark enabled in PACA */
275 lpaca->proc_enabled = 0;
277 /* Initialize hash table for that CPU */
278 htab_initialize_secondary();
280 /* Initialize STAB/SLB. We use a virtual address as it works
281 * in real mode on pSeries and we want a virutal address on
284 if (cpu_has_feature(CPU_FTR_SLB))
287 stab_initialize(lpaca->stab_addr);
290 #endif /* CONFIG_SMP */
292 #if defined(CONFIG_SMP) || defined(CONFIG_KEXEC)
293 void smp_release_cpus(void)
295 extern unsigned long __secondary_hold_spinloop;
298 DBG(" -> smp_release_cpus()\n");
300 /* All secondary cpus are spinning on a common spinloop, release them
301 * all now so they can start to spin on their individual paca
302 * spinloops. For non SMP kernels, the secondary cpus never get out
303 * of the common spinloop.
304 * This is useless but harmless on iSeries, secondaries are already
305 * waiting on their paca spinloops. */
307 ptr = (unsigned long *)((unsigned long)&__secondary_hold_spinloop
312 DBG(" <- smp_release_cpus()\n");
314 #endif /* CONFIG_SMP || CONFIG_KEXEC */
317 * Initialize some remaining members of the ppc64_caches and systemcfg
319 * (at least until we get rid of them completely). This is mostly some
320 * cache informations about the CPU that will be used by cache flush
321 * routines and/or provided to userland
323 static void __init initialize_cache_info(void)
325 struct device_node *np;
326 unsigned long num_cpus = 0;
328 DBG(" -> initialize_cache_info()\n");
330 for (np = NULL; (np = of_find_node_by_type(np, "cpu"));) {
333 /* We're assuming *all* of the CPUs have the same
334 * d-cache and i-cache sizes... -Peter
337 if ( num_cpus == 1 ) {
342 /* Then read cache informations */
343 if (_machine == PLATFORM_POWERMAC) {
344 dc = "d-cache-block-size";
345 ic = "i-cache-block-size";
347 dc = "d-cache-line-size";
348 ic = "i-cache-line-size";
352 lsize = cur_cpu_spec->dcache_bsize;
353 sizep = (u32 *)get_property(np, "d-cache-size", NULL);
356 lsizep = (u32 *) get_property(np, dc, NULL);
359 if (sizep == 0 || lsizep == 0)
360 DBG("Argh, can't find dcache properties ! "
361 "sizep: %p, lsizep: %p\n", sizep, lsizep);
363 ppc64_caches.dsize = size;
364 ppc64_caches.dline_size = lsize;
365 ppc64_caches.log_dline_size = __ilog2(lsize);
366 ppc64_caches.dlines_per_page = PAGE_SIZE / lsize;
369 lsize = cur_cpu_spec->icache_bsize;
370 sizep = (u32 *)get_property(np, "i-cache-size", NULL);
373 lsizep = (u32 *)get_property(np, ic, NULL);
376 if (sizep == 0 || lsizep == 0)
377 DBG("Argh, can't find icache properties ! "
378 "sizep: %p, lsizep: %p\n", sizep, lsizep);
380 ppc64_caches.isize = size;
381 ppc64_caches.iline_size = lsize;
382 ppc64_caches.log_iline_size = __ilog2(lsize);
383 ppc64_caches.ilines_per_page = PAGE_SIZE / lsize;
387 DBG(" <- initialize_cache_info()\n");
392 * Do some initial setup of the system. The parameters are those which
393 * were passed in from the bootloader.
395 void __init setup_system(void)
397 DBG(" -> setup_system()\n");
400 kdump_move_device_tree();
403 * Unflatten the device-tree passed by prom_init or kexec
405 unflatten_device_tree();
408 kexec_setup(); /* requires unflattened device tree. */
412 * Fill the ppc64_caches & systemcfg structures with informations
413 * retrieved from the device-tree. Need to be called before
414 * finish_device_tree() since the later requires some of the
415 * informations filled up here to properly parse the interrupt
417 * It also sets up the cache line sizes which allows to call
418 * routines like flush_icache_range (used by the hash init
421 initialize_cache_info();
423 #ifdef CONFIG_PPC_RTAS
425 * Initialize RTAS if available
428 #endif /* CONFIG_PPC_RTAS */
431 * Check if we have an initrd provided via the device-tree
436 * Do some platform specific early initializations, that includes
437 * setting up the hash table pointers. It also sets up some interrupt-mapping
438 * related options that will be used by finish_device_tree()
443 * We can discover serial ports now since the above did setup the
444 * hash table management for us, thus ioremap works. We do that early
445 * so that further code can be debugged
447 find_legacy_serial_ports();
450 * "Finish" the device-tree, that is do the actual parsing of
451 * some of the properties like the interrupt map
453 finish_device_tree();
458 #ifdef CONFIG_XMON_DEFAULT
462 * Register early console
464 register_early_udbg_console();
466 /* Save unparsed command line copy for /proc/cmdline */
467 strlcpy(saved_command_line, cmd_line, COMMAND_LINE_SIZE);
472 smp_setup_cpu_maps();
475 /* Release secondary cpus out of their spinloops at 0x60 now that
476 * we can map physical -> logical CPU ids
481 printk("Starting Linux PPC64 %s\n", system_utsname.version);
483 printk("-----------------------------------------------------\n");
484 printk("ppc64_pft_size = 0x%lx\n", ppc64_pft_size);
485 printk("ppc64_interrupt_controller = 0x%ld\n",
486 ppc64_interrupt_controller);
487 printk("platform = 0x%x\n", _machine);
488 printk("physicalMemorySize = 0x%lx\n", lmb_phys_mem_size());
489 printk("ppc64_caches.dcache_line_size = 0x%x\n",
490 ppc64_caches.dline_size);
491 printk("ppc64_caches.icache_line_size = 0x%x\n",
492 ppc64_caches.iline_size);
493 printk("htab_address = 0x%p\n", htab_address);
494 printk("htab_hash_mask = 0x%lx\n", htab_hash_mask);
495 #if PHYSICAL_START > 0
496 printk("physical_start = 0x%x\n", PHYSICAL_START);
498 printk("-----------------------------------------------------\n");
502 DBG(" <- setup_system()\n");
505 static int ppc64_panic_event(struct notifier_block *this,
506 unsigned long event, void *ptr)
508 ppc_md.panic((char *)ptr); /* May not return */
512 #ifdef CONFIG_IRQSTACKS
513 static void __init irqstack_early_init(void)
518 * interrupt stacks must be under 256MB, we cannot afford to take
519 * SLB misses on them.
522 softirq_ctx[i] = (struct thread_info *)
523 __va(lmb_alloc_base(THREAD_SIZE,
524 THREAD_SIZE, 0x10000000));
525 hardirq_ctx[i] = (struct thread_info *)
526 __va(lmb_alloc_base(THREAD_SIZE,
527 THREAD_SIZE, 0x10000000));
531 #define irqstack_early_init()
535 * Stack space used when we detect a bad kernel stack pointer, and
536 * early in SMP boots before relocation is enabled.
538 static void __init emergency_stack_init(void)
544 * Emergency stacks must be under 256MB, we cannot afford to take
545 * SLB misses on them. The ABI also requires them to be 128-byte
548 * Since we use these as temporary stacks during secondary CPU
549 * bringup, we need to get at them in real mode. This means they
550 * must also be within the RMO region.
552 limit = min(0x10000000UL, lmb.rmo_size);
555 paca[i].emergency_sp =
556 __va(lmb_alloc_base(HW_PAGE_SIZE, 128, limit)) + HW_PAGE_SIZE;
560 * Called into from start_kernel, after lock_kernel has been called.
561 * Initializes bootmem, which is unsed to manage page allocation until
562 * mem_init is called.
564 void __init setup_arch(char **cmdline_p)
566 extern void do_init_bootmem(void);
568 ppc64_boot_msg(0x12, "Setup Arch");
570 *cmdline_p = cmd_line;
573 * Set cache line size based on type of cpu as a default.
574 * Systems with OF can look in the properties on the cpu node(s)
575 * for a possibly more accurate value.
577 dcache_bsize = ppc64_caches.dline_size;
578 icache_bsize = ppc64_caches.iline_size;
580 /* reboot on panic */
584 notifier_chain_register(&panic_notifier_list, &ppc64_panic_block);
586 init_mm.start_code = PAGE_OFFSET;
587 init_mm.end_code = (unsigned long) _etext;
588 init_mm.end_data = (unsigned long) _edata;
589 init_mm.brk = klimit;
591 irqstack_early_init();
592 emergency_stack_init();
596 /* set up the bootmem stuff with available memory */
600 #ifdef CONFIG_DUMMY_CONSOLE
601 conswitchp = &dummy_con;
606 /* Use the default idle loop if the platform hasn't provided one. */
607 if (NULL == ppc_md.idle_loop) {
608 ppc_md.idle_loop = default_idle;
609 printk(KERN_INFO "Using default idle loop\n");
613 ppc64_boot_msg(0x15, "Setup Done");
617 /* ToDo: do something useful if ppc_md is not yet setup. */
618 #define PPC64_LINUX_FUNCTION 0x0f000000
619 #define PPC64_IPL_MESSAGE 0xc0000000
620 #define PPC64_TERM_MESSAGE 0xb0000000
622 static void ppc64_do_msg(unsigned int src, const char *msg)
624 if (ppc_md.progress) {
627 sprintf(buf, "%08X\n", src);
628 ppc_md.progress(buf, 0);
629 snprintf(buf, 128, "%s", msg);
630 ppc_md.progress(buf, 0);
634 /* Print a boot progress message. */
635 void ppc64_boot_msg(unsigned int src, const char *msg)
637 ppc64_do_msg(PPC64_LINUX_FUNCTION|PPC64_IPL_MESSAGE|src, msg);
638 printk("[boot]%04x %s\n", src, msg);
641 /* Print a termination message (print only -- does not stop the kernel) */
642 void ppc64_terminate_msg(unsigned int src, const char *msg)
644 ppc64_do_msg(PPC64_LINUX_FUNCTION|PPC64_TERM_MESSAGE|src, msg);
645 printk("[terminate]%04x %s\n", src, msg);
648 int check_legacy_ioport(unsigned long base_port)
650 if (ppc_md.check_legacy_ioport == NULL)
652 return ppc_md.check_legacy_ioport(base_port);
654 EXPORT_SYMBOL(check_legacy_ioport);
663 void __init setup_per_cpu_areas(void)
669 /* Copy section for each CPU (we discard the original) */
670 size = ALIGN(__per_cpu_end - __per_cpu_start, SMP_CACHE_BYTES);
671 #ifdef CONFIG_MODULES
672 if (size < PERCPU_ENOUGH_ROOM)
673 size = PERCPU_ENOUGH_ROOM;
677 ptr = alloc_bootmem_node(NODE_DATA(cpu_to_node(i)), size);
679 panic("Cannot allocate cpu data for CPU %d\n", i);
681 paca[i].data_offset = ptr - __per_cpu_start;
682 memcpy(ptr, __per_cpu_start, __per_cpu_end - __per_cpu_start);