setup_64.c revision cd015707176820b86d07b5dffdecfefdd539a497 
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 */
147void __init setup_paca(int cpu)
148{
149	local_paca = &paca[cpu];
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	/* Fill in any unititialised pacas */
180	initialise_pacas();
181
182	/* Identify CPU type */
183	identify_cpu(0, mfspr(SPRN_PVR));
184
185	/* Assume we're on cpu 0 for now. Don't write to the paca yet! */
186	setup_paca(0);
187
188	/* Initialize lockdep early or else spinlocks will blow */
189	lockdep_init();
190
191	/* -------- printk is now safe to use ------- */
192
193	/* Enable early debugging if any specified (see udbg.h) */
194	udbg_early_init();
195
196 	DBG(" -> early_setup(), dt_ptr: 0x%lx\n", dt_ptr);
197
198	/*
199	 * Do early initialization using the flattened device
200	 * tree, such as retrieving the physical memory map or
201	 * calculating/retrieving the hash table size.
202	 */
203	early_init_devtree(__va(dt_ptr));
204
205	/* Now we know the logical id of our boot cpu, setup the paca. */
206	setup_paca(boot_cpuid);
207
208	/* Fix up paca fields required for the boot cpu */
209	get_paca()->cpu_start = 1;
210
211	/* Probe the machine type */
212	probe_machine();
213
214	setup_kdump_trampoline();
215
216	DBG("Found, Initializing memory management...\n");
217
218	/* Initialize the hash table or TLB handling */
219	early_init_mmu();
220
221	DBG(" <- early_setup()\n");
222}
223
224#ifdef CONFIG_SMP
225void early_setup_secondary(void)
226{
227	/* Mark interrupts enabled in PACA */
228	get_paca()->soft_enabled = 0;
229
230	/* Initialize the hash table or TLB handling */
231	early_init_mmu_secondary();
232}
233
234#endif /* CONFIG_SMP */
235
236#if defined(CONFIG_SMP) || defined(CONFIG_KEXEC)
237void smp_release_cpus(void)
238{
239	unsigned long *ptr;
240
241	DBG(" -> smp_release_cpus()\n");
242
243	/* All secondary cpus are spinning on a common spinloop, release them
244	 * all now so they can start to spin on their individual paca
245	 * spinloops. For non SMP kernels, the secondary cpus never get out
246	 * of the common spinloop.
247	 */
248
249	ptr  = (unsigned long *)((unsigned long)&__secondary_hold_spinloop
250			- PHYSICAL_START);
251	*ptr = __pa(generic_secondary_smp_init);
252	mb();
253
254	DBG(" <- smp_release_cpus()\n");
255}
256#endif /* CONFIG_SMP || CONFIG_KEXEC */
257
258/*
259 * Initialize some remaining members of the ppc64_caches and systemcfg
260 * structures
261 * (at least until we get rid of them completely). This is mostly some
262 * cache informations about the CPU that will be used by cache flush
263 * routines and/or provided to userland
264 */
265static void __init initialize_cache_info(void)
266{
267	struct device_node *np;
268	unsigned long num_cpus = 0;
269
270	DBG(" -> initialize_cache_info()\n");
271
272	for (np = NULL; (np = of_find_node_by_type(np, "cpu"));) {
273		num_cpus += 1;
274
275		/* We're assuming *all* of the CPUs have the same
276		 * d-cache and i-cache sizes... -Peter
277		 */
278
279		if ( num_cpus == 1 ) {
280			const u32 *sizep, *lsizep;
281			u32 size, lsize;
282
283			size = 0;
284			lsize = cur_cpu_spec->dcache_bsize;
285			sizep = of_get_property(np, "d-cache-size", NULL);
286			if (sizep != NULL)
287				size = *sizep;
288			lsizep = of_get_property(np, "d-cache-block-size", NULL);
289			/* fallback if block size missing */
290			if (lsizep == NULL)
291				lsizep = of_get_property(np, "d-cache-line-size", NULL);
292			if (lsizep != NULL)
293				lsize = *lsizep;
294			if (sizep == 0 || lsizep == 0)
295				DBG("Argh, can't find dcache properties ! "
296				    "sizep: %p, lsizep: %p\n", sizep, lsizep);
297
298			ppc64_caches.dsize = size;
299			ppc64_caches.dline_size = lsize;
300			ppc64_caches.log_dline_size = __ilog2(lsize);
301			ppc64_caches.dlines_per_page = PAGE_SIZE / lsize;
302
303			size = 0;
304			lsize = cur_cpu_spec->icache_bsize;
305			sizep = of_get_property(np, "i-cache-size", NULL);
306			if (sizep != NULL)
307				size = *sizep;
308			lsizep = of_get_property(np, "i-cache-block-size", NULL);
309			if (lsizep == NULL)
310				lsizep = of_get_property(np, "i-cache-line-size", NULL);
311			if (lsizep != NULL)
312				lsize = *lsizep;
313			if (sizep == 0 || lsizep == 0)
314				DBG("Argh, can't find icache properties ! "
315				    "sizep: %p, lsizep: %p\n", sizep, lsizep);
316
317			ppc64_caches.isize = size;
318			ppc64_caches.iline_size = lsize;
319			ppc64_caches.log_iline_size = __ilog2(lsize);
320			ppc64_caches.ilines_per_page = PAGE_SIZE / lsize;
321		}
322	}
323
324	DBG(" <- initialize_cache_info()\n");
325}
326
327
328/*
329 * Do some initial setup of the system.  The parameters are those which
330 * were passed in from the bootloader.
331 */
332void __init setup_system(void)
333{
334	DBG(" -> setup_system()\n");
335
336	/* Apply the CPUs-specific and firmware specific fixups to kernel
337	 * text (nop out sections not relevant to this CPU or this firmware)
338	 */
339	do_feature_fixups(cur_cpu_spec->cpu_features,
340			  &__start___ftr_fixup, &__stop___ftr_fixup);
341	do_feature_fixups(cur_cpu_spec->mmu_features,
342			  &__start___mmu_ftr_fixup, &__stop___mmu_ftr_fixup);
343	do_feature_fixups(powerpc_firmware_features,
344			  &__start___fw_ftr_fixup, &__stop___fw_ftr_fixup);
345	do_lwsync_fixups(cur_cpu_spec->cpu_features,
346			 &__start___lwsync_fixup, &__stop___lwsync_fixup);
347
348	/*
349	 * Unflatten the device-tree passed by prom_init or kexec
350	 */
351	unflatten_device_tree();
352
353	/*
354	 * Fill the ppc64_caches & systemcfg structures with informations
355 	 * retrieved from the device-tree.
356	 */
357	initialize_cache_info();
358
359#ifdef CONFIG_PPC_RTAS
360	/*
361	 * Initialize RTAS if available
362	 */
363	rtas_initialize();
364#endif /* CONFIG_PPC_RTAS */
365
366	/*
367	 * Check if we have an initrd provided via the device-tree
368	 */
369	check_for_initrd();
370
371	/*
372	 * Do some platform specific early initializations, that includes
373	 * setting up the hash table pointers. It also sets up some interrupt-mapping
374	 * related options that will be used by finish_device_tree()
375	 */
376	if (ppc_md.init_early)
377		ppc_md.init_early();
378
379 	/*
380	 * We can discover serial ports now since the above did setup the
381	 * hash table management for us, thus ioremap works. We do that early
382	 * so that further code can be debugged
383	 */
384	find_legacy_serial_ports();
385
386	/*
387	 * Register early console
388	 */
389	register_early_udbg_console();
390
391	/*
392	 * Initialize xmon
393	 */
394	xmon_setup();
395
396	check_smt_enabled();
397	smp_setup_cpu_maps();
398
399#ifdef CONFIG_SMP
400	/* Release secondary cpus out of their spinloops at 0x60 now that
401	 * we can map physical -> logical CPU ids
402	 */
403	smp_release_cpus();
404#endif
405
406	printk("Starting Linux PPC64 %s\n", init_utsname()->version);
407
408	printk("-----------------------------------------------------\n");
409	printk("ppc64_pft_size                = 0x%llx\n", ppc64_pft_size);
410	printk("physicalMemorySize            = 0x%llx\n", lmb_phys_mem_size());
411	if (ppc64_caches.dline_size != 0x80)
412		printk("ppc64_caches.dcache_line_size = 0x%x\n",
413		       ppc64_caches.dline_size);
414	if (ppc64_caches.iline_size != 0x80)
415		printk("ppc64_caches.icache_line_size = 0x%x\n",
416		       ppc64_caches.iline_size);
417#ifdef CONFIG_PPC_STD_MMU_64
418	if (htab_address)
419		printk("htab_address                  = 0x%p\n", htab_address);
420	printk("htab_hash_mask                = 0x%lx\n", htab_hash_mask);
421#endif /* CONFIG_PPC_STD_MMU_64 */
422	if (PHYSICAL_START > 0)
423		printk("physical_start                = 0x%llx\n",
424		       (unsigned long long)PHYSICAL_START);
425	printk("-----------------------------------------------------\n");
426
427	DBG(" <- setup_system()\n");
428}
429
430#ifdef CONFIG_IRQSTACKS
431static void __init irqstack_early_init(void)
432{
433	unsigned int i;
434
435	/*
436	 * interrupt stacks must be under 256MB, we cannot afford to take
437	 * SLB misses on them.
438	 */
439	for_each_possible_cpu(i) {
440		softirq_ctx[i] = (struct thread_info *)
441			__va(lmb_alloc_base(THREAD_SIZE,
442					    THREAD_SIZE, 0x10000000));
443		hardirq_ctx[i] = (struct thread_info *)
444			__va(lmb_alloc_base(THREAD_SIZE,
445					    THREAD_SIZE, 0x10000000));
446	}
447}
448#else
449#define irqstack_early_init()
450#endif
451
452#ifdef CONFIG_PPC_BOOK3E
453static void __init exc_lvl_early_init(void)
454{
455	unsigned int i;
456
457	for_each_possible_cpu(i) {
458		critirq_ctx[i] = (struct thread_info *)
459			__va(lmb_alloc(THREAD_SIZE, THREAD_SIZE));
460		dbgirq_ctx[i] = (struct thread_info *)
461			__va(lmb_alloc(THREAD_SIZE, THREAD_SIZE));
462		mcheckirq_ctx[i] = (struct thread_info *)
463			__va(lmb_alloc(THREAD_SIZE, THREAD_SIZE));
464	}
465}
466#else
467#define exc_lvl_early_init()
468#endif
469
470/*
471 * Stack space used when we detect a bad kernel stack pointer, and
472 * early in SMP boots before relocation is enabled.
473 */
474static void __init emergency_stack_init(void)
475{
476	unsigned long limit;
477	unsigned int i;
478
479	/*
480	 * Emergency stacks must be under 256MB, we cannot afford to take
481	 * SLB misses on them. The ABI also requires them to be 128-byte
482	 * aligned.
483	 *
484	 * Since we use these as temporary stacks during secondary CPU
485	 * bringup, we need to get at them in real mode. This means they
486	 * must also be within the RMO region.
487	 */
488	limit = min(0x10000000ULL, lmb.rmo_size);
489
490	for_each_possible_cpu(i) {
491		unsigned long sp;
492		sp  = lmb_alloc_base(THREAD_SIZE, THREAD_SIZE, limit);
493		sp += THREAD_SIZE;
494		paca[i].emergency_sp = __va(sp);
495	}
496}
497
498/*
499 * Called into from start_kernel, after lock_kernel has been called.
500 * Initializes bootmem, which is unsed to manage page allocation until
501 * mem_init is called.
502 */
503void __init setup_arch(char **cmdline_p)
504{
505	ppc64_boot_msg(0x12, "Setup Arch");
506
507	*cmdline_p = cmd_line;
508
509	/*
510	 * Set cache line size based on type of cpu as a default.
511	 * Systems with OF can look in the properties on the cpu node(s)
512	 * for a possibly more accurate value.
513	 */
514	dcache_bsize = ppc64_caches.dline_size;
515	icache_bsize = ppc64_caches.iline_size;
516
517	/* reboot on panic */
518	panic_timeout = 180;
519
520	if (ppc_md.panic)
521		setup_panic();
522
523	init_mm.start_code = (unsigned long)_stext;
524	init_mm.end_code = (unsigned long) _etext;
525	init_mm.end_data = (unsigned long) _edata;
526	init_mm.brk = klimit;
527
528	irqstack_early_init();
529	exc_lvl_early_init();
530	emergency_stack_init();
531
532#ifdef CONFIG_PPC_STD_MMU_64
533	stabs_alloc();
534#endif
535	/* set up the bootmem stuff with available memory */
536	do_init_bootmem();
537	sparse_init();
538
539#ifdef CONFIG_DUMMY_CONSOLE
540	conswitchp = &dummy_con;
541#endif
542
543	if (ppc_md.setup_arch)
544		ppc_md.setup_arch();
545
546#ifdef CONFIG_SWIOTLB
547	if (ppc_swiotlb_enable)
548		swiotlb_init();
549#endif
550
551	paging_init();
552
553	/* Initialize the MMU context management stuff */
554	mmu_context_init();
555
556	ppc64_boot_msg(0x15, "Setup Done");
557}
558
559
560/* ToDo: do something useful if ppc_md is not yet setup. */
561#define PPC64_LINUX_FUNCTION 0x0f000000
562#define PPC64_IPL_MESSAGE 0xc0000000
563#define PPC64_TERM_MESSAGE 0xb0000000
564
565static void ppc64_do_msg(unsigned int src, const char *msg)
566{
567	if (ppc_md.progress) {
568		char buf[128];
569
570		sprintf(buf, "%08X\n", src);
571		ppc_md.progress(buf, 0);
572		snprintf(buf, 128, "%s", msg);
573		ppc_md.progress(buf, 0);
574	}
575}
576
577/* Print a boot progress message. */
578void ppc64_boot_msg(unsigned int src, const char *msg)
579{
580	ppc64_do_msg(PPC64_LINUX_FUNCTION|PPC64_IPL_MESSAGE|src, msg);
581	printk("[boot]%04x %s\n", src, msg);
582}
583
584void cpu_die(void)
585{
586	if (ppc_md.cpu_die)
587		ppc_md.cpu_die();
588}
589
590#ifdef CONFIG_SMP
591#define PCPU_DYN_SIZE		()
592
593static void * __init pcpu_fc_alloc(unsigned int cpu, size_t size, size_t align)
594{
595	return __alloc_bootmem_node(NODE_DATA(cpu_to_node(cpu)), size, align,
596				    __pa(MAX_DMA_ADDRESS));
597}
598
599static void __init pcpu_fc_free(void *ptr, size_t size)
600{
601	free_bootmem(__pa(ptr), size);
602}
603
604static int pcpu_cpu_distance(unsigned int from, unsigned int to)
605{
606	if (cpu_to_node(from) == cpu_to_node(to))
607		return LOCAL_DISTANCE;
608	else
609		return REMOTE_DISTANCE;
610}
611
612void __init setup_per_cpu_areas(void)
613{
614	const size_t dyn_size = PERCPU_MODULE_RESERVE + PERCPU_DYNAMIC_RESERVE;
615	size_t atom_size;
616	unsigned long delta;
617	unsigned int cpu;
618	int rc;
619
620	/*
621	 * Linear mapping is one of 4K, 1M and 16M.  For 4K, no need
622	 * to group units.  For larger mappings, use 1M atom which
623	 * should be large enough to contain a number of units.
624	 */
625	if (mmu_linear_psize == MMU_PAGE_4K)
626		atom_size = PAGE_SIZE;
627	else
628		atom_size = 1 << 20;
629
630	rc = pcpu_embed_first_chunk(0, dyn_size, atom_size, pcpu_cpu_distance,
631				    pcpu_fc_alloc, pcpu_fc_free);
632	if (rc < 0)
633		panic("cannot initialize percpu area (err=%d)", rc);
634
635	delta = (unsigned long)pcpu_base_addr - (unsigned long)__per_cpu_start;
636	for_each_possible_cpu(cpu)
637		paca[cpu].data_offset = delta + pcpu_unit_offsets[cpu];
638}
639#endif
640
641
642#ifdef CONFIG_PPC_INDIRECT_IO
643struct ppc_pci_io ppc_pci_io;
644EXPORT_SYMBOL(ppc_pci_io);
645#endif /* CONFIG_PPC_INDIRECT_IO */
646
647