setup_64.c revision 1426d5a3bd07589534286375998c0c8c6fdc5260 
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/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/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/lmb.h>
38#include <asm/io.h>
39#include <asm/kdump.h>
40#include <asm/prom.h>
41#include <asm/processor.h>
42#include <asm/pgtable.h>
43#include <asm/smp.h>
44#include <asm/elf.h>
45#include <asm/machdep.h>
46#include <asm/paca.h>
47#include <asm/time.h>
48#include <asm/cputable.h>
49#include <asm/sections.h>
50#include <asm/btext.h>
51#include <asm/nvram.h>
52#include <asm/setup.h>
53#include <asm/system.h>
54#include <asm/rtas.h>
55#include <asm/iommu.h>
56#include <asm/serial.h>
57#include <asm/cache.h>
58#include <asm/page.h>
59#include <asm/mmu.h>
60#include <asm/firmware.h>
61#include <asm/xmon.h>
62#include <asm/udbg.h>
63#include <asm/kexec.h>
64#include <asm/swiotlb.h>
65#include <asm/mmu_context.h>
66
67#include "setup.h"
68
69#ifdef DEBUG
70#define DBG(fmt...) udbg_printf(fmt)
71#else
72#define DBG(fmt...)
73#endif
74
75int boot_cpuid = 0;
76u64 ppc64_pft_size;
77
78/* Pick defaults since we might want to patch instructions
79 * before we've read this from the device tree.
80 */
81struct ppc64_caches ppc64_caches = {
82	.dline_size = 0x40,
83	.log_dline_size = 6,
84	.iline_size = 0x40,
85	.log_iline_size = 6
86};
87EXPORT_SYMBOL_GPL(ppc64_caches);
88
89/*
90 * These are used in binfmt_elf.c to put aux entries on the stack
91 * for each elf executable being started.
92 */
93int dcache_bsize;
94int icache_bsize;
95int ucache_bsize;
96
97#ifdef CONFIG_SMP
98
99static int smt_enabled_cmdline;
100
101/* Look for ibm,smt-enabled OF option */
102static void check_smt_enabled(void)
103{
104	struct device_node *dn;
105	const char *smt_option;
106
107	/* Allow the command line to overrule the OF option */
108	if (smt_enabled_cmdline)
109		return;
110
111	dn = of_find_node_by_path("/options");
112
113	if (dn) {
114		smt_option = of_get_property(dn, "ibm,smt-enabled", NULL);
115
116                if (smt_option) {
117			if (!strcmp(smt_option, "on"))
118				smt_enabled_at_boot = 1;
119			else if (!strcmp(smt_option, "off"))
120				smt_enabled_at_boot = 0;
121                }
122        }
123}
124
125/* Look for smt-enabled= cmdline option */
126static int __init early_smt_enabled(char *p)
127{
128	smt_enabled_cmdline = 1;
129
130	if (!p)
131		return 0;
132
133	if (!strcmp(p, "on") || !strcmp(p, "1"))
134		smt_enabled_at_boot = 1;
135	else if (!strcmp(p, "off") || !strcmp(p, "0"))
136		smt_enabled_at_boot = 0;
137
138	return 0;
139}
140early_param("smt-enabled", early_smt_enabled);
141
142#else
143#define check_smt_enabled()
144#endif /* CONFIG_SMP */
145
146/* Put the paca pointer into r13 and SPRG_PACA */
147static void __init setup_paca(struct paca_struct *new_paca)
148{
149	local_paca = new_paca;
150	mtspr(SPRN_SPRG_PACA, local_paca);
151#ifdef CONFIG_PPC_BOOK3E
152	mtspr(SPRN_SPRG_TLB_EXFRAME, local_paca->extlb);
153#endif
154}
155
156/*
157 * Early initialization entry point. This is called by head.S
158 * with MMU translation disabled. We rely on the "feature" of
159 * the CPU that ignores the top 2 bits of the address in real
160 * mode so we can access kernel globals normally provided we
161 * only toy with things in the RMO region. From here, we do
162 * some early parsing of the device-tree to setup out LMB
163 * data structures, and allocate & initialize the hash table
164 * and segment tables so we can start running with translation
165 * enabled.
166 *
167 * It is this function which will call the probe() callback of
168 * the various platform types and copy the matching one to the
169 * global ppc_md structure. Your platform can eventually do
170 * some very early initializations from the probe() routine, but
171 * this is not recommended, be very careful as, for example, the
172 * device-tree is not accessible via normal means at this point.
173 */
174
175void __init early_setup(unsigned long dt_ptr)
176{
177	/* -------- printk is _NOT_ safe to use here ! ------- */
178
179	/* Identify CPU type */
180	identify_cpu(0, mfspr(SPRN_PVR));
181
182	/* Assume we're on cpu 0 for now. Don't write to the paca yet! */
183	initialise_paca(&boot_paca, 0);
184	setup_paca(&boot_paca);
185
186	/* Initialize lockdep early or else spinlocks will blow */
187	lockdep_init();
188
189	/* -------- printk is now safe to use ------- */
190
191	/* Enable early debugging if any specified (see udbg.h) */
192	udbg_early_init();
193
194 	DBG(" -> early_setup(), dt_ptr: 0x%lx\n", dt_ptr);
195
196	/*
197	 * Do early initialization using the flattened device
198	 * tree, such as retrieving the physical memory map or
199	 * calculating/retrieving the hash table size.
200	 */
201	early_init_devtree(__va(dt_ptr));
202
203	/* Now we know the logical id of our boot cpu, setup the paca. */
204	setup_paca(&paca[boot_cpuid]);
205
206	/* Fix up paca fields required for the boot cpu */
207	get_paca()->cpu_start = 1;
208
209	/* Probe the machine type */
210	probe_machine();
211
212	setup_kdump_trampoline();
213
214	DBG("Found, Initializing memory management...\n");
215
216	/* Initialize the hash table or TLB handling */
217	early_init_mmu();
218
219	DBG(" <- early_setup()\n");
220}
221
222#ifdef CONFIG_SMP
223void early_setup_secondary(void)
224{
225	/* Mark interrupts enabled in PACA */
226	get_paca()->soft_enabled = 0;
227
228	/* Initialize the hash table or TLB handling */
229	early_init_mmu_secondary();
230}
231
232#endif /* CONFIG_SMP */
233
234#if defined(CONFIG_SMP) || defined(CONFIG_KEXEC)
235void smp_release_cpus(void)
236{
237	unsigned long *ptr;
238
239	DBG(" -> smp_release_cpus()\n");
240
241	/* All secondary cpus are spinning on a common spinloop, release them
242	 * all now so they can start to spin on their individual paca
243	 * spinloops. For non SMP kernels, the secondary cpus never get out
244	 * of the common spinloop.
245	 */
246
247	ptr  = (unsigned long *)((unsigned long)&__secondary_hold_spinloop
248			- PHYSICAL_START);
249	*ptr = __pa(generic_secondary_smp_init);
250	mb();
251
252	DBG(" <- smp_release_cpus()\n");
253}
254#endif /* CONFIG_SMP || CONFIG_KEXEC */
255
256/*
257 * Initialize some remaining members of the ppc64_caches and systemcfg
258 * structures
259 * (at least until we get rid of them completely). This is mostly some
260 * cache informations about the CPU that will be used by cache flush
261 * routines and/or provided to userland
262 */
263static void __init initialize_cache_info(void)
264{
265	struct device_node *np;
266	unsigned long num_cpus = 0;
267
268	DBG(" -> initialize_cache_info()\n");
269
270	for (np = NULL; (np = of_find_node_by_type(np, "cpu"));) {
271		num_cpus += 1;
272
273		/* We're assuming *all* of the CPUs have the same
274		 * d-cache and i-cache sizes... -Peter
275		 */
276
277		if ( num_cpus == 1 ) {
278			const u32 *sizep, *lsizep;
279			u32 size, lsize;
280
281			size = 0;
282			lsize = cur_cpu_spec->dcache_bsize;
283			sizep = of_get_property(np, "d-cache-size", NULL);
284			if (sizep != NULL)
285				size = *sizep;
286			lsizep = of_get_property(np, "d-cache-block-size", NULL);
287			/* fallback if block size missing */
288			if (lsizep == NULL)
289				lsizep = of_get_property(np, "d-cache-line-size", NULL);
290			if (lsizep != NULL)
291				lsize = *lsizep;
292			if (sizep == 0 || lsizep == 0)
293				DBG("Argh, can't find dcache properties ! "
294				    "sizep: %p, lsizep: %p\n", sizep, lsizep);
295
296			ppc64_caches.dsize = size;
297			ppc64_caches.dline_size = lsize;
298			ppc64_caches.log_dline_size = __ilog2(lsize);
299			ppc64_caches.dlines_per_page = PAGE_SIZE / lsize;
300
301			size = 0;
302			lsize = cur_cpu_spec->icache_bsize;
303			sizep = of_get_property(np, "i-cache-size", NULL);
304			if (sizep != NULL)
305				size = *sizep;
306			lsizep = of_get_property(np, "i-cache-block-size", NULL);
307			if (lsizep == NULL)
308				lsizep = of_get_property(np, "i-cache-line-size", NULL);
309			if (lsizep != NULL)
310				lsize = *lsizep;
311			if (sizep == 0 || lsizep == 0)
312				DBG("Argh, can't find icache properties ! "
313				    "sizep: %p, lsizep: %p\n", sizep, lsizep);
314
315			ppc64_caches.isize = size;
316			ppc64_caches.iline_size = lsize;
317			ppc64_caches.log_iline_size = __ilog2(lsize);
318			ppc64_caches.ilines_per_page = PAGE_SIZE / lsize;
319		}
320	}
321
322	DBG(" <- initialize_cache_info()\n");
323}
324
325
326/*
327 * Do some initial setup of the system.  The parameters are those which
328 * were passed in from the bootloader.
329 */
330void __init setup_system(void)
331{
332	DBG(" -> setup_system()\n");
333
334	/* Apply the CPUs-specific and firmware specific fixups to kernel
335	 * text (nop out sections not relevant to this CPU or this firmware)
336	 */
337	do_feature_fixups(cur_cpu_spec->cpu_features,
338			  &__start___ftr_fixup, &__stop___ftr_fixup);
339	do_feature_fixups(cur_cpu_spec->mmu_features,
340			  &__start___mmu_ftr_fixup, &__stop___mmu_ftr_fixup);
341	do_feature_fixups(powerpc_firmware_features,
342			  &__start___fw_ftr_fixup, &__stop___fw_ftr_fixup);
343	do_lwsync_fixups(cur_cpu_spec->cpu_features,
344			 &__start___lwsync_fixup, &__stop___lwsync_fixup);
345
346	/*
347	 * Unflatten the device-tree passed by prom_init or kexec
348	 */
349	unflatten_device_tree();
350
351	/*
352	 * Fill the ppc64_caches & systemcfg structures with informations
353 	 * retrieved from the device-tree.
354	 */
355	initialize_cache_info();
356
357#ifdef CONFIG_PPC_RTAS
358	/*
359	 * Initialize RTAS if available
360	 */
361	rtas_initialize();
362#endif /* CONFIG_PPC_RTAS */
363
364	/*
365	 * Check if we have an initrd provided via the device-tree
366	 */
367	check_for_initrd();
368
369	/*
370	 * Do some platform specific early initializations, that includes
371	 * setting up the hash table pointers. It also sets up some interrupt-mapping
372	 * related options that will be used by finish_device_tree()
373	 */
374	if (ppc_md.init_early)
375		ppc_md.init_early();
376
377 	/*
378	 * We can discover serial ports now since the above did setup the
379	 * hash table management for us, thus ioremap works. We do that early
380	 * so that further code can be debugged
381	 */
382	find_legacy_serial_ports();
383
384	/*
385	 * Register early console
386	 */
387	register_early_udbg_console();
388
389	/*
390	 * Initialize xmon
391	 */
392	xmon_setup();
393
394	check_smt_enabled();
395	smp_setup_cpu_maps();
396
397#ifdef CONFIG_SMP
398	/* Release secondary cpus out of their spinloops at 0x60 now that
399	 * we can map physical -> logical CPU ids
400	 */
401	smp_release_cpus();
402#endif
403
404	printk("Starting Linux PPC64 %s\n", init_utsname()->version);
405
406	printk("-----------------------------------------------------\n");
407	printk("ppc64_pft_size                = 0x%llx\n", ppc64_pft_size);
408	printk("physicalMemorySize            = 0x%llx\n", lmb_phys_mem_size());
409	if (ppc64_caches.dline_size != 0x80)
410		printk("ppc64_caches.dcache_line_size = 0x%x\n",
411		       ppc64_caches.dline_size);
412	if (ppc64_caches.iline_size != 0x80)
413		printk("ppc64_caches.icache_line_size = 0x%x\n",
414		       ppc64_caches.iline_size);
415#ifdef CONFIG_PPC_STD_MMU_64
416	if (htab_address)
417		printk("htab_address                  = 0x%p\n", htab_address);
418	printk("htab_hash_mask                = 0x%lx\n", htab_hash_mask);
419#endif /* CONFIG_PPC_STD_MMU_64 */
420	if (PHYSICAL_START > 0)
421		printk("physical_start                = 0x%llx\n",
422		       (unsigned long long)PHYSICAL_START);
423	printk("-----------------------------------------------------\n");
424
425	DBG(" <- setup_system()\n");
426}
427
428#ifdef CONFIG_IRQSTACKS
429static void __init irqstack_early_init(void)
430{
431	unsigned int i;
432
433	/*
434	 * interrupt stacks must be under 256MB, we cannot afford to take
435	 * SLB misses on them.
436	 */
437	for_each_possible_cpu(i) {
438		softirq_ctx[i] = (struct thread_info *)
439			__va(lmb_alloc_base(THREAD_SIZE,
440					    THREAD_SIZE, 0x10000000));
441		hardirq_ctx[i] = (struct thread_info *)
442			__va(lmb_alloc_base(THREAD_SIZE,
443					    THREAD_SIZE, 0x10000000));
444	}
445}
446#else
447#define irqstack_early_init()
448#endif
449
450#ifdef CONFIG_PPC_BOOK3E
451static void __init exc_lvl_early_init(void)
452{
453	unsigned int i;
454
455	for_each_possible_cpu(i) {
456		critirq_ctx[i] = (struct thread_info *)
457			__va(lmb_alloc(THREAD_SIZE, THREAD_SIZE));
458		dbgirq_ctx[i] = (struct thread_info *)
459			__va(lmb_alloc(THREAD_SIZE, THREAD_SIZE));
460		mcheckirq_ctx[i] = (struct thread_info *)
461			__va(lmb_alloc(THREAD_SIZE, THREAD_SIZE));
462	}
463}
464#else
465#define exc_lvl_early_init()
466#endif
467
468/*
469 * Stack space used when we detect a bad kernel stack pointer, and
470 * early in SMP boots before relocation is enabled.
471 */
472static void __init emergency_stack_init(void)
473{
474	unsigned long limit;
475	unsigned int i;
476
477	/*
478	 * Emergency stacks must be under 256MB, we cannot afford to take
479	 * SLB misses on them. The ABI also requires them to be 128-byte
480	 * aligned.
481	 *
482	 * Since we use these as temporary stacks during secondary CPU
483	 * bringup, we need to get at them in real mode. This means they
484	 * must also be within the RMO region.
485	 */
486	limit = min(0x10000000ULL, lmb.rmo_size);
487
488	for_each_possible_cpu(i) {
489		unsigned long sp;
490		sp  = lmb_alloc_base(THREAD_SIZE, THREAD_SIZE, limit);
491		sp += THREAD_SIZE;
492		paca[i].emergency_sp = __va(sp);
493	}
494}
495
496/*
497 * Called into from start_kernel, after lock_kernel has been called.
498 * Initializes bootmem, which is unsed to manage page allocation until
499 * mem_init is called.
500 */
501void __init setup_arch(char **cmdline_p)
502{
503	ppc64_boot_msg(0x12, "Setup Arch");
504
505	*cmdline_p = cmd_line;
506
507	/*
508	 * Set cache line size based on type of cpu as a default.
509	 * Systems with OF can look in the properties on the cpu node(s)
510	 * for a possibly more accurate value.
511	 */
512	dcache_bsize = ppc64_caches.dline_size;
513	icache_bsize = ppc64_caches.iline_size;
514
515	/* reboot on panic */
516	panic_timeout = 180;
517
518	if (ppc_md.panic)
519		setup_panic();
520
521	init_mm.start_code = (unsigned long)_stext;
522	init_mm.end_code = (unsigned long) _etext;
523	init_mm.end_data = (unsigned long) _edata;
524	init_mm.brk = klimit;
525
526	irqstack_early_init();
527	exc_lvl_early_init();
528	emergency_stack_init();
529
530#ifdef CONFIG_PPC_STD_MMU_64
531	stabs_alloc();
532#endif
533	/* set up the bootmem stuff with available memory */
534	do_init_bootmem();
535	sparse_init();
536
537#ifdef CONFIG_DUMMY_CONSOLE
538	conswitchp = &dummy_con;
539#endif
540
541	if (ppc_md.setup_arch)
542		ppc_md.setup_arch();
543
544#ifdef CONFIG_SWIOTLB
545	if (ppc_swiotlb_enable)
546		swiotlb_init(1);
547#endif
548
549	paging_init();
550
551	/* Initialize the MMU context management stuff */
552	mmu_context_init();
553
554	ppc64_boot_msg(0x15, "Setup Done");
555}
556
557
558/* ToDo: do something useful if ppc_md is not yet setup. */
559#define PPC64_LINUX_FUNCTION 0x0f000000
560#define PPC64_IPL_MESSAGE 0xc0000000
561#define PPC64_TERM_MESSAGE 0xb0000000
562
563static void ppc64_do_msg(unsigned int src, const char *msg)
564{
565	if (ppc_md.progress) {
566		char buf[128];
567
568		sprintf(buf, "%08X\n", src);
569		ppc_md.progress(buf, 0);
570		snprintf(buf, 128, "%s", msg);
571		ppc_md.progress(buf, 0);
572	}
573}
574
575/* Print a boot progress message. */
576void ppc64_boot_msg(unsigned int src, const char *msg)
577{
578	ppc64_do_msg(PPC64_LINUX_FUNCTION|PPC64_IPL_MESSAGE|src, msg);
579	printk("[boot]%04x %s\n", src, msg);
580}
581
582void cpu_die(void)
583{
584	if (ppc_md.cpu_die)
585		ppc_md.cpu_die();
586}
587
588#ifdef CONFIG_SMP
589#define PCPU_DYN_SIZE		()
590
591static void * __init pcpu_fc_alloc(unsigned int cpu, size_t size, size_t align)
592{
593	return __alloc_bootmem_node(NODE_DATA(cpu_to_node(cpu)), size, align,
594				    __pa(MAX_DMA_ADDRESS));
595}
596
597static void __init pcpu_fc_free(void *ptr, size_t size)
598{
599	free_bootmem(__pa(ptr), size);
600}
601
602static int pcpu_cpu_distance(unsigned int from, unsigned int to)
603{
604	if (cpu_to_node(from) == cpu_to_node(to))
605		return LOCAL_DISTANCE;
606	else
607		return REMOTE_DISTANCE;
608}
609
610void __init setup_per_cpu_areas(void)
611{
612	const size_t dyn_size = PERCPU_MODULE_RESERVE + PERCPU_DYNAMIC_RESERVE;
613	size_t atom_size;
614	unsigned long delta;
615	unsigned int cpu;
616	int rc;
617
618	/*
619	 * Linear mapping is one of 4K, 1M and 16M.  For 4K, no need
620	 * to group units.  For larger mappings, use 1M atom which
621	 * should be large enough to contain a number of units.
622	 */
623	if (mmu_linear_psize == MMU_PAGE_4K)
624		atom_size = PAGE_SIZE;
625	else
626		atom_size = 1 << 20;
627
628	rc = pcpu_embed_first_chunk(0, dyn_size, atom_size, pcpu_cpu_distance,
629				    pcpu_fc_alloc, pcpu_fc_free);
630	if (rc < 0)
631		panic("cannot initialize percpu area (err=%d)", rc);
632
633	delta = (unsigned long)pcpu_base_addr - (unsigned long)__per_cpu_start;
634	for_each_possible_cpu(cpu)
635		paca[cpu].data_offset = delta + pcpu_unit_offsets[cpu];
636}
637#endif
638
639
640#ifdef CONFIG_PPC_INDIRECT_IO
641struct ppc_pci_io ppc_pci_io;
642EXPORT_SYMBOL(ppc_pci_io);
643#endif /* CONFIG_PPC_INDIRECT_IO */
644
645