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)
79 /* Pick defaults since we might want to patch instructions
80 * before we've read this from the device tree.
82 struct ppc64_caches ppc64_caches
= {
88 EXPORT_SYMBOL_GPL(ppc64_caches
);
91 * These are used in binfmt_elf.c to put aux entries on the stack
92 * for each elf executable being started.
98 /* The main machine-dep calls structure
100 struct machdep_calls ppc_md
;
101 EXPORT_SYMBOL(ppc_md
);
103 #ifdef CONFIG_MAGIC_SYSRQ
104 unsigned long SYSRQ_KEY
;
105 #endif /* CONFIG_MAGIC_SYSRQ */
108 static int ppc64_panic_event(struct notifier_block
*, unsigned long, void *);
109 static struct notifier_block ppc64_panic_block
= {
110 .notifier_call
= ppc64_panic_event
,
111 .priority
= INT_MIN
/* may not return; must be done last */
116 static int smt_enabled_cmdline
;
118 /* Look for ibm,smt-enabled OF option */
119 static void check_smt_enabled(void)
121 struct device_node
*dn
;
124 /* Allow the command line to overrule the OF option */
125 if (smt_enabled_cmdline
)
128 dn
= of_find_node_by_path("/options");
131 smt_option
= (char *)get_property(dn
, "ibm,smt-enabled", NULL
);
134 if (!strcmp(smt_option
, "on"))
135 smt_enabled_at_boot
= 1;
136 else if (!strcmp(smt_option
, "off"))
137 smt_enabled_at_boot
= 0;
142 /* Look for smt-enabled= cmdline option */
143 static int __init
early_smt_enabled(char *p
)
145 smt_enabled_cmdline
= 1;
150 if (!strcmp(p
, "on") || !strcmp(p
, "1"))
151 smt_enabled_at_boot
= 1;
152 else if (!strcmp(p
, "off") || !strcmp(p
, "0"))
153 smt_enabled_at_boot
= 0;
157 early_param("smt-enabled", early_smt_enabled
);
160 #define check_smt_enabled()
161 #endif /* CONFIG_SMP */
163 extern struct machdep_calls pSeries_md
;
164 extern struct machdep_calls pmac_md
;
165 extern struct machdep_calls maple_md
;
166 extern struct machdep_calls cell_md
;
167 extern struct machdep_calls iseries_md
;
169 /* Ultimately, stuff them in an elf section like initcalls... */
170 static struct machdep_calls __initdata
*machines
[] = {
171 #ifdef CONFIG_PPC_PSERIES
173 #endif /* CONFIG_PPC_PSERIES */
174 #ifdef CONFIG_PPC_PMAC
176 #endif /* CONFIG_PPC_PMAC */
177 #ifdef CONFIG_PPC_MAPLE
179 #endif /* CONFIG_PPC_MAPLE */
180 #ifdef CONFIG_PPC_CELL
183 #ifdef CONFIG_PPC_ISERIES
190 * Early initialization entry point. This is called by head.S
191 * with MMU translation disabled. We rely on the "feature" of
192 * the CPU that ignores the top 2 bits of the address in real
193 * mode so we can access kernel globals normally provided we
194 * only toy with things in the RMO region. From here, we do
195 * some early parsing of the device-tree to setup out LMB
196 * data structures, and allocate & initialize the hash table
197 * and segment tables so we can start running with translation
200 * It is this function which will call the probe() callback of
201 * the various platform types and copy the matching one to the
202 * global ppc_md structure. Your platform can eventually do
203 * some very early initializations from the probe() routine, but
204 * this is not recommended, be very careful as, for example, the
205 * device-tree is not accessible via normal means at this point.
208 void __init
early_setup(unsigned long dt_ptr
)
210 static struct machdep_calls
**mach
;
212 /* Enable early debugging if any specified (see udbg.h) */
215 DBG(" -> early_setup()\n");
218 * Do early initializations using the flattened device
219 * tree, like retreiving the physical memory map or
220 * calculating/retreiving the hash table size
222 early_init_devtree(__va(dt_ptr
));
224 /* Now we know the logical id of our boot cpu, setup the paca. */
227 /* Fix up paca fields required for the boot cpu */
228 get_paca()->cpu_start
= 1;
229 get_paca()->stab_real
= __pa((u64
)&initial_stab
);
230 get_paca()->stab_addr
= (u64
)&initial_stab
;
233 * Iterate all ppc_md structures until we find the proper
234 * one for the current machine type
236 DBG("Probing machine type for platform %x...\n", _machine
);
238 for (mach
= machines
; *mach
; mach
++) {
239 if ((*mach
)->probe(_machine
))
242 /* What can we do if we didn't find ? */
244 DBG("No suitable machine found !\n");
249 #ifdef CONFIG_CRASH_DUMP
253 DBG("Found, Initializing memory management...\n");
256 * Initialize the MMU Hash table and create the linear mapping
257 * of memory. Has to be done before stab/slb initialization as
258 * this is currently where the page size encoding is obtained
263 * Initialize stab / SLB management except on iSeries
265 if (!firmware_has_feature(FW_FEATURE_ISERIES
)) {
266 if (cpu_has_feature(CPU_FTR_SLB
))
269 stab_initialize(get_paca()->stab_real
);
272 DBG(" <- early_setup()\n");
276 void early_setup_secondary(void)
278 struct paca_struct
*lpaca
= get_paca();
280 /* Mark enabled in PACA */
281 lpaca
->proc_enabled
= 0;
283 /* Initialize hash table for that CPU */
284 htab_initialize_secondary();
286 /* Initialize STAB/SLB. We use a virtual address as it works
287 * in real mode on pSeries and we want a virutal address on
290 if (cpu_has_feature(CPU_FTR_SLB
))
293 stab_initialize(lpaca
->stab_addr
);
296 #endif /* CONFIG_SMP */
298 #if defined(CONFIG_SMP) || defined(CONFIG_KEXEC)
299 void smp_release_cpus(void)
301 extern unsigned long __secondary_hold_spinloop
;
304 DBG(" -> smp_release_cpus()\n");
306 /* All secondary cpus are spinning on a common spinloop, release them
307 * all now so they can start to spin on their individual paca
308 * spinloops. For non SMP kernels, the secondary cpus never get out
309 * of the common spinloop.
310 * This is useless but harmless on iSeries, secondaries are already
311 * waiting on their paca spinloops. */
313 ptr
= (unsigned long *)((unsigned long)&__secondary_hold_spinloop
318 DBG(" <- smp_release_cpus()\n");
320 #endif /* CONFIG_SMP || CONFIG_KEXEC */
323 * Initialize some remaining members of the ppc64_caches and systemcfg
325 * (at least until we get rid of them completely). This is mostly some
326 * cache informations about the CPU that will be used by cache flush
327 * routines and/or provided to userland
329 static void __init
initialize_cache_info(void)
331 struct device_node
*np
;
332 unsigned long num_cpus
= 0;
334 DBG(" -> initialize_cache_info()\n");
336 for (np
= NULL
; (np
= of_find_node_by_type(np
, "cpu"));) {
339 /* We're assuming *all* of the CPUs have the same
340 * d-cache and i-cache sizes... -Peter
343 if ( num_cpus
== 1 ) {
348 /* Then read cache informations */
349 if (_machine
== PLATFORM_POWERMAC
) {
350 dc
= "d-cache-block-size";
351 ic
= "i-cache-block-size";
353 dc
= "d-cache-line-size";
354 ic
= "i-cache-line-size";
358 lsize
= cur_cpu_spec
->dcache_bsize
;
359 sizep
= (u32
*)get_property(np
, "d-cache-size", NULL
);
362 lsizep
= (u32
*) get_property(np
, dc
, NULL
);
365 if (sizep
== 0 || lsizep
== 0)
366 DBG("Argh, can't find dcache properties ! "
367 "sizep: %p, lsizep: %p\n", sizep
, lsizep
);
369 ppc64_caches
.dsize
= size
;
370 ppc64_caches
.dline_size
= lsize
;
371 ppc64_caches
.log_dline_size
= __ilog2(lsize
);
372 ppc64_caches
.dlines_per_page
= PAGE_SIZE
/ lsize
;
375 lsize
= cur_cpu_spec
->icache_bsize
;
376 sizep
= (u32
*)get_property(np
, "i-cache-size", NULL
);
379 lsizep
= (u32
*)get_property(np
, ic
, NULL
);
382 if (sizep
== 0 || lsizep
== 0)
383 DBG("Argh, can't find icache properties ! "
384 "sizep: %p, lsizep: %p\n", sizep
, lsizep
);
386 ppc64_caches
.isize
= size
;
387 ppc64_caches
.iline_size
= lsize
;
388 ppc64_caches
.log_iline_size
= __ilog2(lsize
);
389 ppc64_caches
.ilines_per_page
= PAGE_SIZE
/ lsize
;
393 DBG(" <- initialize_cache_info()\n");
398 * Do some initial setup of the system. The parameters are those which
399 * were passed in from the bootloader.
401 void __init
setup_system(void)
403 DBG(" -> setup_system()\n");
406 kdump_move_device_tree();
409 * Unflatten the device-tree passed by prom_init or kexec
411 unflatten_device_tree();
414 kexec_setup(); /* requires unflattened device tree. */
418 * Fill the ppc64_caches & systemcfg structures with informations
419 * retrieved from the device-tree. Need to be called before
420 * finish_device_tree() since the later requires some of the
421 * informations filled up here to properly parse the interrupt
423 * It also sets up the cache line sizes which allows to call
424 * routines like flush_icache_range (used by the hash init
427 initialize_cache_info();
429 #ifdef CONFIG_PPC_RTAS
431 * Initialize RTAS if available
434 #endif /* CONFIG_PPC_RTAS */
437 * Check if we have an initrd provided via the device-tree
442 * Do some platform specific early initializations, that includes
443 * setting up the hash table pointers. It also sets up some interrupt-mapping
444 * related options that will be used by finish_device_tree()
449 * We can discover serial ports now since the above did setup the
450 * hash table management for us, thus ioremap works. We do that early
451 * so that further code can be debugged
453 find_legacy_serial_ports();
456 * "Finish" the device-tree, that is do the actual parsing of
457 * some of the properties like the interrupt map
459 finish_device_tree();
464 #ifdef CONFIG_XMON_DEFAULT
468 * Register early console
470 register_early_udbg_console();
472 /* Save unparsed command line copy for /proc/cmdline */
473 strlcpy(saved_command_line
, cmd_line
, COMMAND_LINE_SIZE
);
478 smp_setup_cpu_maps();
481 /* Release secondary cpus out of their spinloops at 0x60 now that
482 * we can map physical -> logical CPU ids
487 printk("Starting Linux PPC64 %s\n", system_utsname
.version
);
489 printk("-----------------------------------------------------\n");
490 printk("ppc64_pft_size = 0x%lx\n", ppc64_pft_size
);
491 printk("ppc64_interrupt_controller = 0x%ld\n",
492 ppc64_interrupt_controller
);
493 printk("platform = 0x%x\n", _machine
);
494 printk("physicalMemorySize = 0x%lx\n", lmb_phys_mem_size());
495 printk("ppc64_caches.dcache_line_size = 0x%x\n",
496 ppc64_caches
.dline_size
);
497 printk("ppc64_caches.icache_line_size = 0x%x\n",
498 ppc64_caches
.iline_size
);
499 printk("htab_address = 0x%p\n", htab_address
);
500 printk("htab_hash_mask = 0x%lx\n", htab_hash_mask
);
501 #if PHYSICAL_START > 0
502 printk("physical_start = 0x%x\n", PHYSICAL_START
);
504 printk("-----------------------------------------------------\n");
506 DBG(" <- setup_system()\n");
509 static int ppc64_panic_event(struct notifier_block
*this,
510 unsigned long event
, void *ptr
)
512 ppc_md
.panic((char *)ptr
); /* May not return */
516 #ifdef CONFIG_IRQSTACKS
517 static void __init
irqstack_early_init(void)
522 * interrupt stacks must be under 256MB, we cannot afford to take
523 * SLB misses on them.
526 softirq_ctx
[i
] = (struct thread_info
*)
527 __va(lmb_alloc_base(THREAD_SIZE
,
528 THREAD_SIZE
, 0x10000000));
529 hardirq_ctx
[i
] = (struct thread_info
*)
530 __va(lmb_alloc_base(THREAD_SIZE
,
531 THREAD_SIZE
, 0x10000000));
535 #define irqstack_early_init()
539 * Stack space used when we detect a bad kernel stack pointer, and
540 * early in SMP boots before relocation is enabled.
542 static void __init
emergency_stack_init(void)
548 * Emergency stacks must be under 256MB, we cannot afford to take
549 * SLB misses on them. The ABI also requires them to be 128-byte
552 * Since we use these as temporary stacks during secondary CPU
553 * bringup, we need to get at them in real mode. This means they
554 * must also be within the RMO region.
556 limit
= min(0x10000000UL
, lmb
.rmo_size
);
559 paca
[i
].emergency_sp
=
560 __va(lmb_alloc_base(HW_PAGE_SIZE
, 128, limit
)) + HW_PAGE_SIZE
;
564 * Called into from start_kernel, after lock_kernel has been called.
565 * Initializes bootmem, which is unsed to manage page allocation until
566 * mem_init is called.
568 void __init
setup_arch(char **cmdline_p
)
570 extern void do_init_bootmem(void);
572 ppc64_boot_msg(0x12, "Setup Arch");
574 *cmdline_p
= cmd_line
;
577 * Set cache line size based on type of cpu as a default.
578 * Systems with OF can look in the properties on the cpu node(s)
579 * for a possibly more accurate value.
581 dcache_bsize
= ppc64_caches
.dline_size
;
582 icache_bsize
= ppc64_caches
.iline_size
;
584 /* reboot on panic */
588 notifier_chain_register(&panic_notifier_list
, &ppc64_panic_block
);
590 init_mm
.start_code
= PAGE_OFFSET
;
591 init_mm
.end_code
= (unsigned long) _etext
;
592 init_mm
.end_data
= (unsigned long) _edata
;
593 init_mm
.brk
= klimit
;
595 irqstack_early_init();
596 emergency_stack_init();
600 /* set up the bootmem stuff with available memory */
604 #ifdef CONFIG_DUMMY_CONSOLE
605 conswitchp
= &dummy_con
;
610 /* Use the default idle loop if the platform hasn't provided one. */
611 if (NULL
== ppc_md
.idle_loop
) {
612 ppc_md
.idle_loop
= default_idle
;
613 printk(KERN_INFO
"Using default idle loop\n");
617 ppc64_boot_msg(0x15, "Setup Done");
621 /* ToDo: do something useful if ppc_md is not yet setup. */
622 #define PPC64_LINUX_FUNCTION 0x0f000000
623 #define PPC64_IPL_MESSAGE 0xc0000000
624 #define PPC64_TERM_MESSAGE 0xb0000000
626 static void ppc64_do_msg(unsigned int src
, const char *msg
)
628 if (ppc_md
.progress
) {
631 sprintf(buf
, "%08X\n", src
);
632 ppc_md
.progress(buf
, 0);
633 snprintf(buf
, 128, "%s", msg
);
634 ppc_md
.progress(buf
, 0);
638 /* Print a boot progress message. */
639 void ppc64_boot_msg(unsigned int src
, const char *msg
)
641 ppc64_do_msg(PPC64_LINUX_FUNCTION
|PPC64_IPL_MESSAGE
|src
, msg
);
642 printk("[boot]%04x %s\n", src
, msg
);
645 /* Print a termination message (print only -- does not stop the kernel) */
646 void ppc64_terminate_msg(unsigned int src
, const char *msg
)
648 ppc64_do_msg(PPC64_LINUX_FUNCTION
|PPC64_TERM_MESSAGE
|src
, msg
);
649 printk("[terminate]%04x %s\n", src
, msg
);
652 int check_legacy_ioport(unsigned long base_port
)
654 if (ppc_md
.check_legacy_ioport
== NULL
)
656 return ppc_md
.check_legacy_ioport(base_port
);
658 EXPORT_SYMBOL(check_legacy_ioport
);
667 void __init
setup_per_cpu_areas(void)
673 /* Copy section for each CPU (we discard the original) */
674 size
= ALIGN(__per_cpu_end
- __per_cpu_start
, SMP_CACHE_BYTES
);
675 #ifdef CONFIG_MODULES
676 if (size
< PERCPU_ENOUGH_ROOM
)
677 size
= PERCPU_ENOUGH_ROOM
;
681 ptr
= alloc_bootmem_node(NODE_DATA(cpu_to_node(i
)), size
);
683 panic("Cannot allocate cpu data for CPU %d\n", i
);
685 paca
[i
].data_offset
= ptr
- __per_cpu_start
;
686 memcpy(ptr
, __per_cpu_start
, __per_cpu_end
- __per_cpu_start
);