setup_64.c revision fd6e7d2d6a0231ebfa08e1f9a323497ea548da7d 
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/ide.h>
24#include <linux/seq_file.h>
25#include <linux/ioport.h>
26#include <linux/console.h>
27#include <linux/utsname.h>
28#include <linux/tty.h>
29#include <linux/root_dev.h>
30#include <linux/notifier.h>
31#include <linux/cpu.h>
32#include <linux/unistd.h>
33#include <linux/serial.h>
34#include <linux/serial_8250.h>
35#include <linux/bootmem.h>
36#include <asm/io.h>
37#include <asm/kdump.h>
38#include <asm/prom.h>
39#include <asm/processor.h>
40#include <asm/pgtable.h>
41#include <asm/smp.h>
42#include <asm/elf.h>
43#include <asm/machdep.h>
44#include <asm/paca.h>
45#include <asm/time.h>
46#include <asm/cputable.h>
47#include <asm/sections.h>
48#include <asm/btext.h>
49#include <asm/nvram.h>
50#include <asm/setup.h>
51#include <asm/system.h>
52#include <asm/rtas.h>
53#include <asm/iommu.h>
54#include <asm/serial.h>
55#include <asm/cache.h>
56#include <asm/page.h>
57#include <asm/mmu.h>
58#include <asm/lmb.h>
59#include <asm/firmware.h>
60#include <asm/xmon.h>
61#include <asm/udbg.h>
62#include <asm/kexec.h>
63
64#include "setup.h"
65
66#ifdef DEBUG
67#define DBG(fmt...) udbg_printf(fmt)
68#else
69#define DBG(fmt...)
70#endif
71
72int have_of = 1;
73int boot_cpuid = 0;
74u64 ppc64_pft_size;
75
76/* Pick defaults since we might want to patch instructions
77 * before we've read this from the device tree.
78 */
79struct ppc64_caches ppc64_caches = {
80	.dline_size = 0x40,
81	.log_dline_size = 6,
82	.iline_size = 0x40,
83	.log_iline_size = 6
84};
85EXPORT_SYMBOL_GPL(ppc64_caches);
86
87/*
88 * These are used in binfmt_elf.c to put aux entries on the stack
89 * for each elf executable being started.
90 */
91int dcache_bsize;
92int icache_bsize;
93int ucache_bsize;
94
95#ifdef CONFIG_SMP
96
97static int smt_enabled_cmdline;
98
99/* Look for ibm,smt-enabled OF option */
100static void check_smt_enabled(void)
101{
102	struct device_node *dn;
103	const char *smt_option;
104
105	/* Allow the command line to overrule the OF option */
106	if (smt_enabled_cmdline)
107		return;
108
109	dn = of_find_node_by_path("/options");
110
111	if (dn) {
112		smt_option = get_property(dn, "ibm,smt-enabled", NULL);
113
114                if (smt_option) {
115			if (!strcmp(smt_option, "on"))
116				smt_enabled_at_boot = 1;
117			else if (!strcmp(smt_option, "off"))
118				smt_enabled_at_boot = 0;
119                }
120        }
121}
122
123/* Look for smt-enabled= cmdline option */
124static int __init early_smt_enabled(char *p)
125{
126	smt_enabled_cmdline = 1;
127
128	if (!p)
129		return 0;
130
131	if (!strcmp(p, "on") || !strcmp(p, "1"))
132		smt_enabled_at_boot = 1;
133	else if (!strcmp(p, "off") || !strcmp(p, "0"))
134		smt_enabled_at_boot = 0;
135
136	return 0;
137}
138early_param("smt-enabled", early_smt_enabled);
139
140#else
141#define check_smt_enabled()
142#endif /* CONFIG_SMP */
143
144/* Put the paca pointer into r13 and SPRG3 */
145void __init setup_paca(int cpu)
146{
147	local_paca = &paca[cpu];
148	mtspr(SPRN_SPRG3, local_paca);
149}
150
151/*
152 * Early initialization entry point. This is called by head.S
153 * with MMU translation disabled. We rely on the "feature" of
154 * the CPU that ignores the top 2 bits of the address in real
155 * mode so we can access kernel globals normally provided we
156 * only toy with things in the RMO region. From here, we do
157 * some early parsing of the device-tree to setup out LMB
158 * data structures, and allocate & initialize the hash table
159 * and segment tables so we can start running with translation
160 * enabled.
161 *
162 * It is this function which will call the probe() callback of
163 * the various platform types and copy the matching one to the
164 * global ppc_md structure. Your platform can eventually do
165 * some very early initializations from the probe() routine, but
166 * this is not recommended, be very careful as, for example, the
167 * device-tree is not accessible via normal means at this point.
168 */
169
170void __init early_setup(unsigned long dt_ptr)
171{
172	/* Assume we're on cpu 0 for now. Don't write to the paca yet! */
173	setup_paca(0);
174
175	/* Enable early debugging if any specified (see udbg.h) */
176	udbg_early_init();
177
178 	DBG(" -> early_setup(), dt_ptr: 0x%lx\n", dt_ptr);
179
180	/*
181	 * Do early initializations using the flattened device
182	 * tree, like retreiving the physical memory map or
183	 * calculating/retreiving the hash table size
184	 */
185	early_init_devtree(__va(dt_ptr));
186
187	/* Now we know the logical id of our boot cpu, setup the paca. */
188	setup_paca(boot_cpuid);
189
190	/* Fix up paca fields required for the boot cpu */
191	get_paca()->cpu_start = 1;
192	get_paca()->stab_real = __pa((u64)&initial_stab);
193	get_paca()->stab_addr = (u64)&initial_stab;
194
195	/* Probe the machine type */
196	probe_machine();
197
198	setup_kdump_trampoline();
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 interrupts enabled in PACA */
226	lpaca->soft_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			const 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 = get_property(np, "d-cache-size", NULL);
305			if (sizep != NULL)
306				size = *sizep;
307			lsizep = 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 = get_property(np, "i-cache-size", NULL);
322			if (sizep != NULL)
323				size = *sizep;
324			lsizep = 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	/*
351	 * Unflatten the device-tree passed by prom_init or kexec
352	 */
353	unflatten_device_tree();
354
355	/*
356	 * Fill the ppc64_caches & systemcfg structures with informations
357 	 * retrieved from the device-tree.
358	 */
359	initialize_cache_info();
360
361	/*
362	 * Initialize irq remapping subsystem
363	 */
364	irq_early_init();
365
366#ifdef CONFIG_PPC_RTAS
367	/*
368	 * Initialize RTAS if available
369	 */
370	rtas_initialize();
371#endif /* CONFIG_PPC_RTAS */
372
373	/*
374	 * Check if we have an initrd provided via the device-tree
375	 */
376	check_for_initrd();
377
378	/*
379	 * Do some platform specific early initializations, that includes
380	 * setting up the hash table pointers. It also sets up some interrupt-mapping
381	 * related options that will be used by finish_device_tree()
382	 */
383	ppc_md.init_early();
384
385 	/*
386	 * We can discover serial ports now since the above did setup the
387	 * hash table management for us, thus ioremap works. We do that early
388	 * so that further code can be debugged
389	 */
390	find_legacy_serial_ports();
391
392	/*
393	 * Register early console
394	 */
395	register_early_udbg_console();
396
397	/*
398	 * Initialize xmon
399	 */
400	xmon_setup();
401
402	check_smt_enabled();
403	smp_setup_cpu_maps();
404
405#ifdef CONFIG_SMP
406	/* Release secondary cpus out of their spinloops at 0x60 now that
407	 * we can map physical -> logical CPU ids
408	 */
409	smp_release_cpus();
410#endif
411
412	printk("Starting Linux PPC64 %s\n", init_utsname()->version);
413
414	printk("-----------------------------------------------------\n");
415	printk("ppc64_pft_size                = 0x%lx\n", ppc64_pft_size);
416	printk("physicalMemorySize            = 0x%lx\n", lmb_phys_mem_size());
417	printk("ppc64_caches.dcache_line_size = 0x%x\n",
418	       ppc64_caches.dline_size);
419	printk("ppc64_caches.icache_line_size = 0x%x\n",
420	       ppc64_caches.iline_size);
421	printk("htab_address                  = 0x%p\n", htab_address);
422	printk("htab_hash_mask                = 0x%lx\n", htab_hash_mask);
423#if PHYSICAL_START > 0
424	printk("physical_start                = 0x%x\n", PHYSICAL_START);
425#endif
426	printk("-----------------------------------------------------\n");
427
428	DBG(" <- setup_system()\n");
429}
430
431#ifdef CONFIG_IRQSTACKS
432static void __init irqstack_early_init(void)
433{
434	unsigned int i;
435
436	/*
437	 * interrupt stacks must be under 256MB, we cannot afford to take
438	 * SLB misses on them.
439	 */
440	for_each_possible_cpu(i) {
441		softirq_ctx[i] = (struct thread_info *)
442			__va(lmb_alloc_base(THREAD_SIZE,
443					    THREAD_SIZE, 0x10000000));
444		hardirq_ctx[i] = (struct thread_info *)
445			__va(lmb_alloc_base(THREAD_SIZE,
446					    THREAD_SIZE, 0x10000000));
447	}
448}
449#else
450#define irqstack_early_init()
451#endif
452
453/*
454 * Stack space used when we detect a bad kernel stack pointer, and
455 * early in SMP boots before relocation is enabled.
456 */
457static void __init emergency_stack_init(void)
458{
459	unsigned long limit;
460	unsigned int i;
461
462	/*
463	 * Emergency stacks must be under 256MB, we cannot afford to take
464	 * SLB misses on them. The ABI also requires them to be 128-byte
465	 * aligned.
466	 *
467	 * Since we use these as temporary stacks during secondary CPU
468	 * bringup, we need to get at them in real mode. This means they
469	 * must also be within the RMO region.
470	 */
471	limit = min(0x10000000UL, lmb.rmo_size);
472
473	for_each_possible_cpu(i)
474		paca[i].emergency_sp =
475		__va(lmb_alloc_base(HW_PAGE_SIZE, 128, limit)) + HW_PAGE_SIZE;
476}
477
478/*
479 * Called into from start_kernel, after lock_kernel has been called.
480 * Initializes bootmem, which is unsed to manage page allocation until
481 * mem_init is called.
482 */
483void __init setup_arch(char **cmdline_p)
484{
485	ppc64_boot_msg(0x12, "Setup Arch");
486
487	*cmdline_p = cmd_line;
488
489	/*
490	 * Set cache line size based on type of cpu as a default.
491	 * Systems with OF can look in the properties on the cpu node(s)
492	 * for a possibly more accurate value.
493	 */
494	dcache_bsize = ppc64_caches.dline_size;
495	icache_bsize = ppc64_caches.iline_size;
496
497	/* reboot on panic */
498	panic_timeout = 180;
499
500	if (ppc_md.panic)
501		setup_panic();
502
503	init_mm.start_code = PAGE_OFFSET;
504	init_mm.end_code = (unsigned long) _etext;
505	init_mm.end_data = (unsigned long) _edata;
506	init_mm.brk = klimit;
507
508	irqstack_early_init();
509	emergency_stack_init();
510
511	stabs_alloc();
512
513	/* set up the bootmem stuff with available memory */
514	do_init_bootmem();
515	sparse_init();
516
517#ifdef CONFIG_DUMMY_CONSOLE
518	conswitchp = &dummy_con;
519#endif
520
521	ppc_md.setup_arch();
522
523	paging_init();
524	ppc64_boot_msg(0x15, "Setup Done");
525}
526
527
528/* ToDo: do something useful if ppc_md is not yet setup. */
529#define PPC64_LINUX_FUNCTION 0x0f000000
530#define PPC64_IPL_MESSAGE 0xc0000000
531#define PPC64_TERM_MESSAGE 0xb0000000
532
533static void ppc64_do_msg(unsigned int src, const char *msg)
534{
535	if (ppc_md.progress) {
536		char buf[128];
537
538		sprintf(buf, "%08X\n", src);
539		ppc_md.progress(buf, 0);
540		snprintf(buf, 128, "%s", msg);
541		ppc_md.progress(buf, 0);
542	}
543}
544
545/* Print a boot progress message. */
546void ppc64_boot_msg(unsigned int src, const char *msg)
547{
548	ppc64_do_msg(PPC64_LINUX_FUNCTION|PPC64_IPL_MESSAGE|src, msg);
549	printk("[boot]%04x %s\n", src, msg);
550}
551
552/* Print a termination message (print only -- does not stop the kernel) */
553void ppc64_terminate_msg(unsigned int src, const char *msg)
554{
555	ppc64_do_msg(PPC64_LINUX_FUNCTION|PPC64_TERM_MESSAGE|src, msg);
556	printk("[terminate]%04x %s\n", src, msg);
557}
558
559void cpu_die(void)
560{
561	if (ppc_md.cpu_die)
562		ppc_md.cpu_die();
563}
564
565#ifdef CONFIG_SMP
566void __init setup_per_cpu_areas(void)
567{
568	int i;
569	unsigned long size;
570	char *ptr;
571
572	/* Copy section for each CPU (we discard the original) */
573	size = ALIGN(__per_cpu_end - __per_cpu_start, SMP_CACHE_BYTES);
574#ifdef CONFIG_MODULES
575	if (size < PERCPU_ENOUGH_ROOM)
576		size = PERCPU_ENOUGH_ROOM;
577#endif
578
579	for_each_possible_cpu(i) {
580		ptr = alloc_bootmem_node(NODE_DATA(cpu_to_node(i)), size);
581		if (!ptr)
582			panic("Cannot allocate cpu data for CPU %d\n", i);
583
584		paca[i].data_offset = ptr - __per_cpu_start;
585		memcpy(ptr, __per_cpu_start, __per_cpu_end - __per_cpu_start);
586	}
587}
588#endif
589