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/export.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/seq_file.h>
24 #include <linux/ioport.h>
25 #include <linux/console.h>
26 #include <linux/utsname.h>
27 #include <linux/tty.h>
28 #include <linux/root_dev.h>
29 #include <linux/notifier.h>
30 #include <linux/cpu.h>
31 #include <linux/unistd.h>
32 #include <linux/serial.h>
33 #include <linux/serial_8250.h>
34 #include <linux/bootmem.h>
35 #include <linux/pci.h>
36 #include <linux/lockdep.h>
37 #include <linux/memblock.h>
38 #include <linux/hugetlb.h>
41 #include <asm/kdump.h>
43 #include <asm/processor.h>
44 #include <asm/pgtable.h>
47 #include <asm/machdep.h>
50 #include <asm/cputable.h>
51 #include <asm/sections.h>
52 #include <asm/btext.h>
53 #include <asm/nvram.h>
54 #include <asm/setup.h>
56 #include <asm/iommu.h>
57 #include <asm/serial.h>
58 #include <asm/cache.h>
61 #include <asm/firmware.h>
64 #include <asm/kexec.h>
65 #include <asm/mmu_context.h>
66 #include <asm/code-patching.h>
67 #include <asm/kvm_ppc.h>
68 #include <asm/hugetlb.h>
73 #define DBG(fmt...) udbg_printf(fmt)
79 int __initdata spinning_secondaries
;
82 /* Pick defaults since we might want to patch instructions
83 * before we've read this from the device tree.
85 struct ppc64_caches ppc64_caches
= {
91 EXPORT_SYMBOL_GPL(ppc64_caches
);
94 * These are used in binfmt_elf.c to put aux entries on the stack
95 * for each elf executable being started.
103 static char *smt_enabled_cmdline
;
105 /* Look for ibm,smt-enabled OF option */
106 static void check_smt_enabled(void)
108 struct device_node
*dn
;
109 const char *smt_option
;
111 /* Default to enabling all threads */
112 smt_enabled_at_boot
= threads_per_core
;
114 /* Allow the command line to overrule the OF option */
115 if (smt_enabled_cmdline
) {
116 if (!strcmp(smt_enabled_cmdline
, "on"))
117 smt_enabled_at_boot
= threads_per_core
;
118 else if (!strcmp(smt_enabled_cmdline
, "off"))
119 smt_enabled_at_boot
= 0;
124 rc
= strict_strtol(smt_enabled_cmdline
, 10, &smt
);
126 smt_enabled_at_boot
=
127 min(threads_per_core
, (int)smt
);
130 dn
= of_find_node_by_path("/options");
132 smt_option
= of_get_property(dn
, "ibm,smt-enabled",
136 if (!strcmp(smt_option
, "on"))
137 smt_enabled_at_boot
= threads_per_core
;
138 else if (!strcmp(smt_option
, "off"))
139 smt_enabled_at_boot
= 0;
147 /* Look for smt-enabled= cmdline option */
148 static int __init
early_smt_enabled(char *p
)
150 smt_enabled_cmdline
= p
;
153 early_param("smt-enabled", early_smt_enabled
);
156 #define check_smt_enabled()
157 #endif /* CONFIG_SMP */
159 /** Fix up paca fields required for the boot cpu */
160 static void fixup_boot_paca(void)
162 /* The boot cpu is started */
163 get_paca()->cpu_start
= 1;
164 /* Allow percpu accesses to work until we setup percpu data */
165 get_paca()->data_offset
= 0;
169 * Early initialization entry point. This is called by head.S
170 * with MMU translation disabled. We rely on the "feature" of
171 * the CPU that ignores the top 2 bits of the address in real
172 * mode so we can access kernel globals normally provided we
173 * only toy with things in the RMO region. From here, we do
174 * some early parsing of the device-tree to setup out MEMBLOCK
175 * data structures, and allocate & initialize the hash table
176 * and segment tables so we can start running with translation
179 * It is this function which will call the probe() callback of
180 * the various platform types and copy the matching one to the
181 * global ppc_md structure. Your platform can eventually do
182 * some very early initializations from the probe() routine, but
183 * this is not recommended, be very careful as, for example, the
184 * device-tree is not accessible via normal means at this point.
187 void __init
early_setup(unsigned long dt_ptr
)
189 static __initdata
struct paca_struct boot_paca
;
191 /* -------- printk is _NOT_ safe to use here ! ------- */
193 /* Identify CPU type */
194 identify_cpu(0, mfspr(SPRN_PVR
));
196 /* Assume we're on cpu 0 for now. Don't write to the paca yet! */
197 initialise_paca(&boot_paca
, 0);
198 setup_paca(&boot_paca
);
201 /* Initialize lockdep early or else spinlocks will blow */
204 /* -------- printk is now safe to use ------- */
206 /* Enable early debugging if any specified (see udbg.h) */
209 DBG(" -> early_setup(), dt_ptr: 0x%lx\n", dt_ptr
);
212 * Do early initialization using the flattened device
213 * tree, such as retrieving the physical memory map or
214 * calculating/retrieving the hash table size.
216 early_init_devtree(__va(dt_ptr
));
218 /* Now we know the logical id of our boot cpu, setup the paca. */
219 setup_paca(&paca
[boot_cpuid
]);
222 /* Probe the machine type */
225 setup_kdump_trampoline();
227 DBG("Found, Initializing memory management...\n");
229 /* Initialize the hash table or TLB handling */
233 * Reserve any gigantic pages requested on the command line.
234 * memblock needs to have been initialized by the time this is
235 * called since this will reserve memory.
237 reserve_hugetlb_gpages();
239 DBG(" <- early_setup()\n");
243 void early_setup_secondary(void)
245 /* Mark interrupts enabled in PACA */
246 get_paca()->soft_enabled
= 0;
248 /* Initialize the hash table or TLB handling */
249 early_init_mmu_secondary();
252 #endif /* CONFIG_SMP */
254 #if defined(CONFIG_SMP) || defined(CONFIG_KEXEC)
255 void smp_release_cpus(void)
260 DBG(" -> smp_release_cpus()\n");
262 /* All secondary cpus are spinning on a common spinloop, release them
263 * all now so they can start to spin on their individual paca
264 * spinloops. For non SMP kernels, the secondary cpus never get out
265 * of the common spinloop.
268 ptr
= (unsigned long *)((unsigned long)&__secondary_hold_spinloop
270 *ptr
= __pa(generic_secondary_smp_init
);
272 /* And wait a bit for them to catch up */
273 for (i
= 0; i
< 100000; i
++) {
276 if (spinning_secondaries
== 0)
280 DBG("spinning_secondaries = %d\n", spinning_secondaries
);
282 DBG(" <- smp_release_cpus()\n");
284 #endif /* CONFIG_SMP || CONFIG_KEXEC */
287 * Initialize some remaining members of the ppc64_caches and systemcfg
289 * (at least until we get rid of them completely). This is mostly some
290 * cache informations about the CPU that will be used by cache flush
291 * routines and/or provided to userland
293 static void __init
initialize_cache_info(void)
295 struct device_node
*np
;
296 unsigned long num_cpus
= 0;
298 DBG(" -> initialize_cache_info()\n");
300 for_each_node_by_type(np
, "cpu") {
304 * We're assuming *all* of the CPUs have the same
305 * d-cache and i-cache sizes... -Peter
308 const u32
*sizep
, *lsizep
;
312 lsize
= cur_cpu_spec
->dcache_bsize
;
313 sizep
= of_get_property(np
, "d-cache-size", NULL
);
316 lsizep
= of_get_property(np
, "d-cache-block-size",
318 /* fallback if block size missing */
320 lsizep
= of_get_property(np
,
325 if (sizep
== 0 || lsizep
== 0)
326 DBG("Argh, can't find dcache properties ! "
327 "sizep: %p, lsizep: %p\n", sizep
, lsizep
);
329 ppc64_caches
.dsize
= size
;
330 ppc64_caches
.dline_size
= lsize
;
331 ppc64_caches
.log_dline_size
= __ilog2(lsize
);
332 ppc64_caches
.dlines_per_page
= PAGE_SIZE
/ lsize
;
335 lsize
= cur_cpu_spec
->icache_bsize
;
336 sizep
= of_get_property(np
, "i-cache-size", NULL
);
339 lsizep
= of_get_property(np
, "i-cache-block-size",
342 lsizep
= of_get_property(np
,
347 if (sizep
== 0 || lsizep
== 0)
348 DBG("Argh, can't find icache properties ! "
349 "sizep: %p, lsizep: %p\n", sizep
, lsizep
);
351 ppc64_caches
.isize
= size
;
352 ppc64_caches
.iline_size
= lsize
;
353 ppc64_caches
.log_iline_size
= __ilog2(lsize
);
354 ppc64_caches
.ilines_per_page
= PAGE_SIZE
/ lsize
;
358 DBG(" <- initialize_cache_info()\n");
363 * Do some initial setup of the system. The parameters are those which
364 * were passed in from the bootloader.
366 void __init
setup_system(void)
368 DBG(" -> setup_system()\n");
370 /* Apply the CPUs-specific and firmware specific fixups to kernel
371 * text (nop out sections not relevant to this CPU or this firmware)
373 do_feature_fixups(cur_cpu_spec
->cpu_features
,
374 &__start___ftr_fixup
, &__stop___ftr_fixup
);
375 do_feature_fixups(cur_cpu_spec
->mmu_features
,
376 &__start___mmu_ftr_fixup
, &__stop___mmu_ftr_fixup
);
377 do_feature_fixups(powerpc_firmware_features
,
378 &__start___fw_ftr_fixup
, &__stop___fw_ftr_fixup
);
379 do_lwsync_fixups(cur_cpu_spec
->cpu_features
,
380 &__start___lwsync_fixup
, &__stop___lwsync_fixup
);
384 * Unflatten the device-tree passed by prom_init or kexec
386 unflatten_device_tree();
389 * Fill the ppc64_caches & systemcfg structures with informations
390 * retrieved from the device-tree.
392 initialize_cache_info();
394 #ifdef CONFIG_PPC_RTAS
396 * Initialize RTAS if available
399 #endif /* CONFIG_PPC_RTAS */
402 * Check if we have an initrd provided via the device-tree
407 * Do some platform specific early initializations, that includes
408 * setting up the hash table pointers. It also sets up some interrupt-mapping
409 * related options that will be used by finish_device_tree()
411 if (ppc_md
.init_early
)
415 * We can discover serial ports now since the above did setup the
416 * hash table management for us, thus ioremap works. We do that early
417 * so that further code can be debugged
419 find_legacy_serial_ports();
422 * Register early console
424 register_early_udbg_console();
431 smp_setup_cpu_maps();
435 /* Release secondary cpus out of their spinloops at 0x60 now that
436 * we can map physical -> logical CPU ids
441 printk("Starting Linux PPC64 %s\n", init_utsname()->version
);
443 printk("-----------------------------------------------------\n");
444 printk("ppc64_pft_size = 0x%llx\n", ppc64_pft_size
);
445 printk("physicalMemorySize = 0x%llx\n", memblock_phys_mem_size());
446 if (ppc64_caches
.dline_size
!= 0x80)
447 printk("ppc64_caches.dcache_line_size = 0x%x\n",
448 ppc64_caches
.dline_size
);
449 if (ppc64_caches
.iline_size
!= 0x80)
450 printk("ppc64_caches.icache_line_size = 0x%x\n",
451 ppc64_caches
.iline_size
);
452 #ifdef CONFIG_PPC_STD_MMU_64
454 printk("htab_address = 0x%p\n", htab_address
);
455 printk("htab_hash_mask = 0x%lx\n", htab_hash_mask
);
456 #endif /* CONFIG_PPC_STD_MMU_64 */
457 if (PHYSICAL_START
> 0)
458 printk("physical_start = 0x%llx\n",
459 (unsigned long long)PHYSICAL_START
);
460 printk("-----------------------------------------------------\n");
462 DBG(" <- setup_system()\n");
465 /* This returns the limit below which memory accesses to the linear
466 * mapping are guarnateed not to cause a TLB or SLB miss. This is
467 * used to allocate interrupt or emergency stacks for which our
468 * exception entry path doesn't deal with being interrupted.
470 static u64
safe_stack_limit(void)
472 #ifdef CONFIG_PPC_BOOK3E
473 /* Freescale BookE bolts the entire linear mapping */
474 if (mmu_has_feature(MMU_FTR_TYPE_FSL_E
))
475 return linear_map_top
;
476 /* Other BookE, we assume the first GB is bolted */
479 /* BookS, the first segment is bolted */
480 if (mmu_has_feature(MMU_FTR_1T_SEGMENT
))
481 return 1UL << SID_SHIFT_1T
;
482 return 1UL << SID_SHIFT
;
486 static void __init
irqstack_early_init(void)
488 u64 limit
= safe_stack_limit();
492 * Interrupt stacks must be in the first segment since we
493 * cannot afford to take SLB misses on them.
495 for_each_possible_cpu(i
) {
496 softirq_ctx
[i
] = (struct thread_info
*)
497 __va(memblock_alloc_base(THREAD_SIZE
,
498 THREAD_SIZE
, limit
));
499 hardirq_ctx
[i
] = (struct thread_info
*)
500 __va(memblock_alloc_base(THREAD_SIZE
,
501 THREAD_SIZE
, limit
));
505 #ifdef CONFIG_PPC_BOOK3E
506 static void __init
exc_lvl_early_init(void)
508 extern unsigned int interrupt_base_book3e
;
509 extern unsigned int exc_debug_debug_book3e
;
513 for_each_possible_cpu(i
) {
514 critirq_ctx
[i
] = (struct thread_info
*)
515 __va(memblock_alloc(THREAD_SIZE
, THREAD_SIZE
));
516 dbgirq_ctx
[i
] = (struct thread_info
*)
517 __va(memblock_alloc(THREAD_SIZE
, THREAD_SIZE
));
518 mcheckirq_ctx
[i
] = (struct thread_info
*)
519 __va(memblock_alloc(THREAD_SIZE
, THREAD_SIZE
));
522 if (cpu_has_feature(CPU_FTR_DEBUG_LVL_EXC
))
523 patch_branch(&interrupt_base_book3e
+ (0x040 / 4) + 1,
524 (unsigned long)&exc_debug_debug_book3e
, 0);
527 #define exc_lvl_early_init()
531 * Stack space used when we detect a bad kernel stack pointer, and
532 * early in SMP boots before relocation is enabled.
534 static void __init
emergency_stack_init(void)
540 * Emergency stacks must be under 256MB, we cannot afford to take
541 * SLB misses on them. The ABI also requires them to be 128-byte
544 * Since we use these as temporary stacks during secondary CPU
545 * bringup, we need to get at them in real mode. This means they
546 * must also be within the RMO region.
548 limit
= min(safe_stack_limit(), ppc64_rma_size
);
550 for_each_possible_cpu(i
) {
552 sp
= memblock_alloc_base(THREAD_SIZE
, THREAD_SIZE
, limit
);
554 paca
[i
].emergency_sp
= __va(sp
);
559 * Called into from start_kernel this initializes bootmem, which is used
560 * to manage page allocation until mem_init is called.
562 void __init
setup_arch(char **cmdline_p
)
564 ppc64_boot_msg(0x12, "Setup Arch");
566 *cmdline_p
= cmd_line
;
569 * Set cache line size based on type of cpu as a default.
570 * Systems with OF can look in the properties on the cpu node(s)
571 * for a possibly more accurate value.
573 dcache_bsize
= ppc64_caches
.dline_size
;
574 icache_bsize
= ppc64_caches
.iline_size
;
576 /* reboot on panic */
582 init_mm
.start_code
= (unsigned long)_stext
;
583 init_mm
.end_code
= (unsigned long) _etext
;
584 init_mm
.end_data
= (unsigned long) _edata
;
585 init_mm
.brk
= klimit
;
587 irqstack_early_init();
588 exc_lvl_early_init();
589 emergency_stack_init();
591 #ifdef CONFIG_PPC_STD_MMU_64
594 /* set up the bootmem stuff with available memory */
598 #ifdef CONFIG_DUMMY_CONSOLE
599 conswitchp
= &dummy_con
;
602 if (ppc_md
.setup_arch
)
607 /* Initialize the MMU context management stuff */
612 /* Interrupt code needs to be 64K-aligned */
613 if ((unsigned long)_stext
& 0xffff)
614 panic("Kernelbase not 64K-aligned (0x%lx)!\n",
615 (unsigned long)_stext
);
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
);
646 #define PCPU_DYN_SIZE ()
648 static void * __init
pcpu_fc_alloc(unsigned int cpu
, size_t size
, size_t align
)
650 return __alloc_bootmem_node(NODE_DATA(cpu_to_node(cpu
)), size
, align
,
651 __pa(MAX_DMA_ADDRESS
));
654 static void __init
pcpu_fc_free(void *ptr
, size_t size
)
656 free_bootmem(__pa(ptr
), size
);
659 static int pcpu_cpu_distance(unsigned int from
, unsigned int to
)
661 if (cpu_to_node(from
) == cpu_to_node(to
))
662 return LOCAL_DISTANCE
;
664 return REMOTE_DISTANCE
;
667 unsigned long __per_cpu_offset
[NR_CPUS
] __read_mostly
;
668 EXPORT_SYMBOL(__per_cpu_offset
);
670 void __init
setup_per_cpu_areas(void)
672 const size_t dyn_size
= PERCPU_MODULE_RESERVE
+ PERCPU_DYNAMIC_RESERVE
;
679 * Linear mapping is one of 4K, 1M and 16M. For 4K, no need
680 * to group units. For larger mappings, use 1M atom which
681 * should be large enough to contain a number of units.
683 if (mmu_linear_psize
== MMU_PAGE_4K
)
684 atom_size
= PAGE_SIZE
;
688 rc
= pcpu_embed_first_chunk(0, dyn_size
, atom_size
, pcpu_cpu_distance
,
689 pcpu_fc_alloc
, pcpu_fc_free
);
691 panic("cannot initialize percpu area (err=%d)", rc
);
693 delta
= (unsigned long)pcpu_base_addr
- (unsigned long)__per_cpu_start
;
694 for_each_possible_cpu(cpu
) {
695 __per_cpu_offset
[cpu
] = delta
+ pcpu_unit_offsets
[cpu
];
696 paca
[cpu
].data_offset
= __per_cpu_offset
[cpu
];
702 #ifdef CONFIG_PPC_INDIRECT_IO
703 struct ppc_pci_io ppc_pci_io
;
704 EXPORT_SYMBOL(ppc_pci_io
);
705 #endif /* CONFIG_PPC_INDIRECT_IO */