setup_64.c revision 7e990266c845d7f712c96013891aaf74baef198f 
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/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>
37#include <asm/io.h>
38#include <asm/kdump.h>
39#include <asm/prom.h>
40#include <asm/processor.h>
41#include <asm/pgtable.h>
42#include <asm/smp.h>
43#include <asm/elf.h>
44#include <asm/machdep.h>
45#include <asm/paca.h>
46#include <asm/time.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>
53#include <asm/rtas.h>
54#include <asm/iommu.h>
55#include <asm/serial.h>
56#include <asm/cache.h>
57#include <asm/page.h>
58#include <asm/mmu.h>
59#include <asm/lmb.h>
60#include <asm/iseries/it_lp_naca.h>
61#include <asm/firmware.h>
62#include <asm/xmon.h>
63#include <asm/udbg.h>
64#include <asm/kexec.h>
65
66#include "setup.h"
67
68#ifdef DEBUG
69#define DBG(fmt...) udbg_printf(fmt)
70#else
71#define DBG(fmt...)
72#endif
73
74int have_of = 1;
75int boot_cpuid = 0;
76dev_t boot_dev;
77u64 ppc64_pft_size;
78
79/* Pick defaults since we might want to patch instructions
80 * before we've read this from the device tree.
81 */
82struct ppc64_caches ppc64_caches = {
83	.dline_size = 0x80,
84	.log_dline_size = 7,
85	.iline_size = 0x80,
86	.log_iline_size = 7
87};
88EXPORT_SYMBOL_GPL(ppc64_caches);
89
90/*
91 * These are used in binfmt_elf.c to put aux entries on the stack
92 * for each elf executable being started.
93 */
94int dcache_bsize;
95int icache_bsize;
96int ucache_bsize;
97
98#ifdef CONFIG_MAGIC_SYSRQ
99unsigned long SYSRQ_KEY;
100#endif /* CONFIG_MAGIC_SYSRQ */
101
102
103#ifdef CONFIG_SMP
104
105static int smt_enabled_cmdline;
106
107/* Look for ibm,smt-enabled OF option */
108static void check_smt_enabled(void)
109{
110	struct device_node *dn;
111	char *smt_option;
112
113	/* Allow the command line to overrule the OF option */
114	if (smt_enabled_cmdline)
115		return;
116
117	dn = of_find_node_by_path("/options");
118
119	if (dn) {
120		smt_option = (char *)get_property(dn, "ibm,smt-enabled", NULL);
121
122                if (smt_option) {
123			if (!strcmp(smt_option, "on"))
124				smt_enabled_at_boot = 1;
125			else if (!strcmp(smt_option, "off"))
126				smt_enabled_at_boot = 0;
127                }
128        }
129}
130
131/* Look for smt-enabled= cmdline option */
132static int __init early_smt_enabled(char *p)
133{
134	smt_enabled_cmdline = 1;
135
136	if (!p)
137		return 0;
138
139	if (!strcmp(p, "on") || !strcmp(p, "1"))
140		smt_enabled_at_boot = 1;
141	else if (!strcmp(p, "off") || !strcmp(p, "0"))
142		smt_enabled_at_boot = 0;
143
144	return 0;
145}
146early_param("smt-enabled", early_smt_enabled);
147
148#else
149#define check_smt_enabled()
150#endif /* CONFIG_SMP */
151
152/*
153 * Early initialization entry point. This is called by head.S
154 * with MMU translation disabled. We rely on the "feature" of
155 * the CPU that ignores the top 2 bits of the address in real
156 * mode so we can access kernel globals normally provided we
157 * only toy with things in the RMO region. From here, we do
158 * some early parsing of the device-tree to setup out LMB
159 * data structures, and allocate & initialize the hash table
160 * and segment tables so we can start running with translation
161 * enabled.
162 *
163 * It is this function which will call the probe() callback of
164 * the various platform types and copy the matching one to the
165 * global ppc_md structure. Your platform can eventually do
166 * some very early initializations from the probe() routine, but
167 * this is not recommended, be very careful as, for example, the
168 * device-tree is not accessible via normal means at this point.
169 */
170
171void __init early_setup(unsigned long dt_ptr)
172{
173	/* Enable early debugging if any specified (see udbg.h) */
174	udbg_early_init();
175
176 	DBG(" -> early_setup(), dt_ptr: 0x%lx\n", dt_ptr);
177
178	/*
179	 * Do early initializations using the flattened device
180	 * tree, like retreiving the physical memory map or
181	 * calculating/retreiving the hash table size
182	 */
183	early_init_devtree(__va(dt_ptr));
184
185	/* Now we know the logical id of our boot cpu, setup the paca. */
186	setup_boot_paca();
187
188	/* Fix up paca fields required for the boot cpu */
189	get_paca()->cpu_start = 1;
190	get_paca()->stab_real = __pa((u64)&initial_stab);
191	get_paca()->stab_addr = (u64)&initial_stab;
192
193	/* Probe the machine type */
194	probe_machine();
195
196#ifdef CONFIG_CRASH_DUMP
197	kdump_setup();
198#endif
199
200	DBG("Found, Initializing memory management...\n");
201
202	/*
203	 * Initialize the MMU Hash table and create the linear mapping
204	 * of memory. Has to be done before stab/slb initialization as
205	 * this is currently where the page size encoding is obtained
206	 */
207	htab_initialize();
208
209	/*
210	 * Initialize stab / SLB management except on iSeries
211	 */
212	if (cpu_has_feature(CPU_FTR_SLB))
213		slb_initialize();
214	else if (!firmware_has_feature(FW_FEATURE_ISERIES))
215		stab_initialize(get_paca()->stab_real);
216
217	DBG(" <- early_setup()\n");
218}
219
220#ifdef CONFIG_SMP
221void early_setup_secondary(void)
222{
223	struct paca_struct *lpaca = get_paca();
224
225	/* Mark enabled in PACA */
226	lpaca->proc_enabled = 0;
227
228	/* Initialize hash table for that CPU */
229	htab_initialize_secondary();
230
231	/* Initialize STAB/SLB. We use a virtual address as it works
232	 * in real mode on pSeries and we want a virutal address on
233	 * iSeries anyway
234	 */
235	if (cpu_has_feature(CPU_FTR_SLB))
236		slb_initialize();
237	else
238		stab_initialize(lpaca->stab_addr);
239}
240
241#endif /* CONFIG_SMP */
242
243#if defined(CONFIG_SMP) || defined(CONFIG_KEXEC)
244void smp_release_cpus(void)
245{
246	extern unsigned long __secondary_hold_spinloop;
247	unsigned long *ptr;
248
249	DBG(" -> smp_release_cpus()\n");
250
251	/* All secondary cpus are spinning on a common spinloop, release them
252	 * all now so they can start to spin on their individual paca
253	 * spinloops. For non SMP kernels, the secondary cpus never get out
254	 * of the common spinloop.
255	 * This is useless but harmless on iSeries, secondaries are already
256	 * waiting on their paca spinloops. */
257
258	ptr  = (unsigned long *)((unsigned long)&__secondary_hold_spinloop
259			- PHYSICAL_START);
260	*ptr = 1;
261	mb();
262
263	DBG(" <- smp_release_cpus()\n");
264}
265#endif /* CONFIG_SMP || CONFIG_KEXEC */
266
267/*
268 * Initialize some remaining members of the ppc64_caches and systemcfg
269 * structures
270 * (at least until we get rid of them completely). This is mostly some
271 * cache informations about the CPU that will be used by cache flush
272 * routines and/or provided to userland
273 */
274static void __init initialize_cache_info(void)
275{
276	struct device_node *np;
277	unsigned long num_cpus = 0;
278
279	DBG(" -> initialize_cache_info()\n");
280
281	for (np = NULL; (np = of_find_node_by_type(np, "cpu"));) {
282		num_cpus += 1;
283
284		/* We're assuming *all* of the CPUs have the same
285		 * d-cache and i-cache sizes... -Peter
286		 */
287
288		if ( num_cpus == 1 ) {
289			u32 *sizep, *lsizep;
290			u32 size, lsize;
291			const char *dc, *ic;
292
293			/* Then read cache informations */
294			if (machine_is(powermac)) {
295				dc = "d-cache-block-size";
296				ic = "i-cache-block-size";
297			} else {
298				dc = "d-cache-line-size";
299				ic = "i-cache-line-size";
300			}
301
302			size = 0;
303			lsize = cur_cpu_spec->dcache_bsize;
304			sizep = (u32 *)get_property(np, "d-cache-size", NULL);
305			if (sizep != NULL)
306				size = *sizep;
307			lsizep = (u32 *) get_property(np, dc, NULL);
308			if (lsizep != NULL)
309				lsize = *lsizep;
310			if (sizep == 0 || lsizep == 0)
311				DBG("Argh, can't find dcache properties ! "
312				    "sizep: %p, lsizep: %p\n", sizep, lsizep);
313
314			ppc64_caches.dsize = size;
315			ppc64_caches.dline_size = lsize;
316			ppc64_caches.log_dline_size = __ilog2(lsize);
317			ppc64_caches.dlines_per_page = PAGE_SIZE / lsize;
318
319			size = 0;
320			lsize = cur_cpu_spec->icache_bsize;
321			sizep = (u32 *)get_property(np, "i-cache-size", NULL);
322			if (sizep != NULL)
323				size = *sizep;
324			lsizep = (u32 *)get_property(np, ic, NULL);
325			if (lsizep != NULL)
326				lsize = *lsizep;
327			if (sizep == 0 || lsizep == 0)
328				DBG("Argh, can't find icache properties ! "
329				    "sizep: %p, lsizep: %p\n", sizep, lsizep);
330
331			ppc64_caches.isize = size;
332			ppc64_caches.iline_size = lsize;
333			ppc64_caches.log_iline_size = __ilog2(lsize);
334			ppc64_caches.ilines_per_page = PAGE_SIZE / lsize;
335		}
336	}
337
338	DBG(" <- initialize_cache_info()\n");
339}
340
341
342/*
343 * Do some initial setup of the system.  The parameters are those which
344 * were passed in from the bootloader.
345 */
346void __init setup_system(void)
347{
348	DBG(" -> setup_system()\n");
349
350#ifdef CONFIG_KEXEC
351	kdump_move_device_tree();
352#endif
353	/*
354	 * Unflatten the device-tree passed by prom_init or kexec
355	 */
356	unflatten_device_tree();
357
358#ifdef CONFIG_KEXEC
359	kexec_setup();	/* requires unflattened device tree. */
360#endif
361
362	/*
363	 * Fill the ppc64_caches & systemcfg structures with informations
364	 * retrieved from the device-tree. Need to be called before
365	 * finish_device_tree() since the later requires some of the
366	 * informations filled up here to properly parse the interrupt
367	 * tree.
368	 * It also sets up the cache line sizes which allows to call
369	 * routines like flush_icache_range (used by the hash init
370	 * later on).
371	 */
372	initialize_cache_info();
373
374#ifdef CONFIG_PPC_RTAS
375	/*
376	 * Initialize RTAS if available
377	 */
378	rtas_initialize();
379#endif /* CONFIG_PPC_RTAS */
380
381	/*
382	 * Check if we have an initrd provided via the device-tree
383	 */
384	check_for_initrd();
385
386	/*
387	 * Do some platform specific early initializations, that includes
388	 * setting up the hash table pointers. It also sets up some interrupt-mapping
389	 * related options that will be used by finish_device_tree()
390	 */
391	ppc_md.init_early();
392
393 	/*
394	 * We can discover serial ports now since the above did setup the
395	 * hash table management for us, thus ioremap works. We do that early
396	 * so that further code can be debugged
397	 */
398	find_legacy_serial_ports();
399
400	/*
401	 * "Finish" the device-tree, that is do the actual parsing of
402	 * some of the properties like the interrupt map
403	 */
404	finish_device_tree();
405
406	/*
407	 * Initialize xmon
408	 */
409#ifdef CONFIG_XMON_DEFAULT
410	xmon_init(1);
411#endif
412	/*
413	 * Register early console
414	 */
415	register_early_udbg_console();
416
417	/* Save unparsed command line copy for /proc/cmdline */
418	strlcpy(saved_command_line, cmd_line, COMMAND_LINE_SIZE);
419
420	parse_early_param();
421
422	check_smt_enabled();
423	smp_setup_cpu_maps();
424
425#ifdef CONFIG_SMP
426	/* Release secondary cpus out of their spinloops at 0x60 now that
427	 * we can map physical -> logical CPU ids
428	 */
429	smp_release_cpus();
430#endif
431
432	printk("Starting Linux PPC64 %s\n", system_utsname.version);
433
434	printk("-----------------------------------------------------\n");
435	printk("ppc64_pft_size                = 0x%lx\n", ppc64_pft_size);
436	printk("ppc64_interrupt_controller    = 0x%ld\n",
437	       ppc64_interrupt_controller);
438	printk("physicalMemorySize            = 0x%lx\n", lmb_phys_mem_size());
439	printk("ppc64_caches.dcache_line_size = 0x%x\n",
440	       ppc64_caches.dline_size);
441	printk("ppc64_caches.icache_line_size = 0x%x\n",
442	       ppc64_caches.iline_size);
443	printk("htab_address                  = 0x%p\n", htab_address);
444	printk("htab_hash_mask                = 0x%lx\n", htab_hash_mask);
445#if PHYSICAL_START > 0
446	printk("physical_start                = 0x%x\n", PHYSICAL_START);
447#endif
448	printk("-----------------------------------------------------\n");
449
450	DBG(" <- setup_system()\n");
451}
452
453#ifdef CONFIG_IRQSTACKS
454static void __init irqstack_early_init(void)
455{
456	unsigned int i;
457
458	/*
459	 * interrupt stacks must be under 256MB, we cannot afford to take
460	 * SLB misses on them.
461	 */
462	for_each_possible_cpu(i) {
463		softirq_ctx[i] = (struct thread_info *)
464			__va(lmb_alloc_base(THREAD_SIZE,
465					    THREAD_SIZE, 0x10000000));
466		hardirq_ctx[i] = (struct thread_info *)
467			__va(lmb_alloc_base(THREAD_SIZE,
468					    THREAD_SIZE, 0x10000000));
469	}
470}
471#else
472#define irqstack_early_init()
473#endif
474
475/*
476 * Stack space used when we detect a bad kernel stack pointer, and
477 * early in SMP boots before relocation is enabled.
478 */
479static void __init emergency_stack_init(void)
480{
481	unsigned long limit;
482	unsigned int i;
483
484	/*
485	 * Emergency stacks must be under 256MB, we cannot afford to take
486	 * SLB misses on them. The ABI also requires them to be 128-byte
487	 * aligned.
488	 *
489	 * Since we use these as temporary stacks during secondary CPU
490	 * bringup, we need to get at them in real mode. This means they
491	 * must also be within the RMO region.
492	 */
493	limit = min(0x10000000UL, lmb.rmo_size);
494
495	for_each_possible_cpu(i)
496		paca[i].emergency_sp =
497		__va(lmb_alloc_base(HW_PAGE_SIZE, 128, limit)) + HW_PAGE_SIZE;
498}
499
500/*
501 * Called into from start_kernel, after lock_kernel has been called.
502 * Initializes bootmem, which is unsed to manage page allocation until
503 * mem_init is called.
504 */
505void __init setup_arch(char **cmdline_p)
506{
507	ppc64_boot_msg(0x12, "Setup Arch");
508
509	*cmdline_p = cmd_line;
510
511	/*
512	 * Set cache line size based on type of cpu as a default.
513	 * Systems with OF can look in the properties on the cpu node(s)
514	 * for a possibly more accurate value.
515	 */
516	dcache_bsize = ppc64_caches.dline_size;
517	icache_bsize = ppc64_caches.iline_size;
518
519	/* reboot on panic */
520	panic_timeout = 180;
521
522	if (ppc_md.panic)
523		setup_panic();
524
525	init_mm.start_code = PAGE_OFFSET;
526	init_mm.end_code = (unsigned long) _etext;
527	init_mm.end_data = (unsigned long) _edata;
528	init_mm.brk = klimit;
529
530	irqstack_early_init();
531	emergency_stack_init();
532
533	stabs_alloc();
534
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	ppc_md.setup_arch();
544
545	paging_init();
546	ppc64_boot_msg(0x15, "Setup Done");
547}
548
549
550/* ToDo: do something useful if ppc_md is not yet setup. */
551#define PPC64_LINUX_FUNCTION 0x0f000000
552#define PPC64_IPL_MESSAGE 0xc0000000
553#define PPC64_TERM_MESSAGE 0xb0000000
554
555static void ppc64_do_msg(unsigned int src, const char *msg)
556{
557	if (ppc_md.progress) {
558		char buf[128];
559
560		sprintf(buf, "%08X\n", src);
561		ppc_md.progress(buf, 0);
562		snprintf(buf, 128, "%s", msg);
563		ppc_md.progress(buf, 0);
564	}
565}
566
567/* Print a boot progress message. */
568void ppc64_boot_msg(unsigned int src, const char *msg)
569{
570	ppc64_do_msg(PPC64_LINUX_FUNCTION|PPC64_IPL_MESSAGE|src, msg);
571	printk("[boot]%04x %s\n", src, msg);
572}
573
574/* Print a termination message (print only -- does not stop the kernel) */
575void ppc64_terminate_msg(unsigned int src, const char *msg)
576{
577	ppc64_do_msg(PPC64_LINUX_FUNCTION|PPC64_TERM_MESSAGE|src, msg);
578	printk("[terminate]%04x %s\n", src, msg);
579}
580
581void cpu_die(void)
582{
583	if (ppc_md.cpu_die)
584		ppc_md.cpu_die();
585}
586
587#ifdef CONFIG_SMP
588void __init setup_per_cpu_areas(void)
589{
590	int i;
591	unsigned long size;
592	char *ptr;
593
594	/* Copy section for each CPU (we discard the original) */
595	size = ALIGN(__per_cpu_end - __per_cpu_start, SMP_CACHE_BYTES);
596#ifdef CONFIG_MODULES
597	if (size < PERCPU_ENOUGH_ROOM)
598		size = PERCPU_ENOUGH_ROOM;
599#endif
600
601	for_each_possible_cpu(i) {
602		ptr = alloc_bootmem_node(NODE_DATA(cpu_to_node(i)), size);
603		if (!ptr)
604			panic("Cannot allocate cpu data for CPU %d\n", i);
605
606		paca[i].data_offset = ptr - __per_cpu_start;
607		memcpy(ptr, __per_cpu_start, __per_cpu_end - __per_cpu_start);
608	}
609}
610#endif
611