setup_64.c revision 4e491d14f2506b218d678935c25a7027b79178b1 
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
65#include "setup.h"
66
67#ifdef DEBUG
68#define DBG(fmt...) udbg_printf(fmt)
69#else
70#define DBG(fmt...)
71#endif
72
73int have_of = 1;
74int boot_cpuid = 0;
75u64 ppc64_pft_size;
76
77/* Pick defaults since we might want to patch instructions
78 * before we've read this from the device tree.
79 */
80struct ppc64_caches ppc64_caches = {
81	.dline_size = 0x40,
82	.log_dline_size = 6,
83	.iline_size = 0x40,
84	.log_iline_size = 6
85};
86EXPORT_SYMBOL_GPL(ppc64_caches);
87
88#ifdef CONFIG_FTRACE
89extern void _mcount(void);
90EXPORT_SYMBOL(_mcount);
91#endif
92
93/*
94 * These are used in binfmt_elf.c to put aux entries on the stack
95 * for each elf executable being started.
96 */
97int dcache_bsize;
98int icache_bsize;
99int ucache_bsize;
100
101#ifdef CONFIG_SMP
102
103static int smt_enabled_cmdline;
104
105/* Look for ibm,smt-enabled OF option */
106static void check_smt_enabled(void)
107{
108	struct device_node *dn;
109	const char *smt_option;
110
111	/* Allow the command line to overrule the OF option */
112	if (smt_enabled_cmdline)
113		return;
114
115	dn = of_find_node_by_path("/options");
116
117	if (dn) {
118		smt_option = of_get_property(dn, "ibm,smt-enabled", NULL);
119
120                if (smt_option) {
121			if (!strcmp(smt_option, "on"))
122				smt_enabled_at_boot = 1;
123			else if (!strcmp(smt_option, "off"))
124				smt_enabled_at_boot = 0;
125                }
126        }
127}
128
129/* Look for smt-enabled= cmdline option */
130static int __init early_smt_enabled(char *p)
131{
132	smt_enabled_cmdline = 1;
133
134	if (!p)
135		return 0;
136
137	if (!strcmp(p, "on") || !strcmp(p, "1"))
138		smt_enabled_at_boot = 1;
139	else if (!strcmp(p, "off") || !strcmp(p, "0"))
140		smt_enabled_at_boot = 0;
141
142	return 0;
143}
144early_param("smt-enabled", early_smt_enabled);
145
146#else
147#define check_smt_enabled()
148#endif /* CONFIG_SMP */
149
150/* Put the paca pointer into r13 and SPRG3 */
151void __init setup_paca(int cpu)
152{
153	local_paca = &paca[cpu];
154	mtspr(SPRN_SPRG3, local_paca);
155}
156
157/*
158 * Early initialization entry point. This is called by head.S
159 * with MMU translation disabled. We rely on the "feature" of
160 * the CPU that ignores the top 2 bits of the address in real
161 * mode so we can access kernel globals normally provided we
162 * only toy with things in the RMO region. From here, we do
163 * some early parsing of the device-tree to setup out LMB
164 * data structures, and allocate & initialize the hash table
165 * and segment tables so we can start running with translation
166 * enabled.
167 *
168 * It is this function which will call the probe() callback of
169 * the various platform types and copy the matching one to the
170 * global ppc_md structure. Your platform can eventually do
171 * some very early initializations from the probe() routine, but
172 * this is not recommended, be very careful as, for example, the
173 * device-tree is not accessible via normal means at this point.
174 */
175
176void __init early_setup(unsigned long dt_ptr)
177{
178	/* -------- printk is _NOT_ safe to use here ! ------- */
179
180	/* Fill in any unititialised pacas */
181	initialise_pacas();
182
183	/* Identify CPU type */
184	identify_cpu(0, mfspr(SPRN_PVR));
185
186	/* Assume we're on cpu 0 for now. Don't write to the paca yet! */
187	setup_paca(0);
188
189	/* Initialize lockdep early or else spinlocks will blow */
190	lockdep_init();
191
192	/* -------- printk is now safe to use ------- */
193
194	/* Enable early debugging if any specified (see udbg.h) */
195	udbg_early_init();
196
197 	DBG(" -> early_setup(), dt_ptr: 0x%lx\n", dt_ptr);
198
199	/*
200	 * Do early initialization using the flattened device
201	 * tree, such as retrieving the physical memory map or
202	 * calculating/retrieving the hash table size.
203	 */
204	early_init_devtree(__va(dt_ptr));
205
206	/* Now we know the logical id of our boot cpu, setup the paca. */
207	setup_paca(boot_cpuid);
208
209	/* Fix up paca fields required for the boot cpu */
210	get_paca()->cpu_start = 1;
211	get_paca()->stab_real = __pa((u64)&initial_stab);
212	get_paca()->stab_addr = (u64)&initial_stab;
213
214	/* Probe the machine type */
215	probe_machine();
216
217	setup_kdump_trampoline();
218
219	DBG("Found, Initializing memory management...\n");
220
221	/*
222	 * Initialize the MMU Hash table and create the linear mapping
223	 * of memory. Has to be done before stab/slb initialization as
224	 * this is currently where the page size encoding is obtained
225	 */
226	htab_initialize();
227
228	/*
229	 * Initialize stab / SLB management except on iSeries
230	 */
231	if (cpu_has_feature(CPU_FTR_SLB))
232		slb_initialize();
233	else if (!firmware_has_feature(FW_FEATURE_ISERIES))
234		stab_initialize(get_paca()->stab_real);
235
236	DBG(" <- early_setup()\n");
237}
238
239#ifdef CONFIG_SMP
240void early_setup_secondary(void)
241{
242	struct paca_struct *lpaca = get_paca();
243
244	/* Mark interrupts enabled in PACA */
245	lpaca->soft_enabled = 0;
246
247	/* Initialize hash table for that CPU */
248	htab_initialize_secondary();
249
250	/* Initialize STAB/SLB. We use a virtual address as it works
251	 * in real mode on pSeries and we want a virutal address on
252	 * iSeries anyway
253	 */
254	if (cpu_has_feature(CPU_FTR_SLB))
255		slb_initialize();
256	else
257		stab_initialize(lpaca->stab_addr);
258}
259
260#endif /* CONFIG_SMP */
261
262#if defined(CONFIG_SMP) || defined(CONFIG_KEXEC)
263void smp_release_cpus(void)
264{
265	extern unsigned long __secondary_hold_spinloop;
266	unsigned long *ptr;
267
268	DBG(" -> smp_release_cpus()\n");
269
270	/* All secondary cpus are spinning on a common spinloop, release them
271	 * all now so they can start to spin on their individual paca
272	 * spinloops. For non SMP kernels, the secondary cpus never get out
273	 * of the common spinloop.
274	 * This is useless but harmless on iSeries, secondaries are already
275	 * waiting on their paca spinloops. */
276
277	ptr  = (unsigned long *)((unsigned long)&__secondary_hold_spinloop
278			- PHYSICAL_START);
279	*ptr = 1;
280	mb();
281
282	DBG(" <- smp_release_cpus()\n");
283}
284#endif /* CONFIG_SMP || CONFIG_KEXEC */
285
286/*
287 * Initialize some remaining members of the ppc64_caches and systemcfg
288 * structures
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
292 */
293static void __init initialize_cache_info(void)
294{
295	struct device_node *np;
296	unsigned long num_cpus = 0;
297
298	DBG(" -> initialize_cache_info()\n");
299
300	for (np = NULL; (np = of_find_node_by_type(np, "cpu"));) {
301		num_cpus += 1;
302
303		/* We're assuming *all* of the CPUs have the same
304		 * d-cache and i-cache sizes... -Peter
305		 */
306
307		if ( num_cpus == 1 ) {
308			const u32 *sizep, *lsizep;
309			u32 size, lsize;
310
311			size = 0;
312			lsize = cur_cpu_spec->dcache_bsize;
313			sizep = of_get_property(np, "d-cache-size", NULL);
314			if (sizep != NULL)
315				size = *sizep;
316			lsizep = of_get_property(np, "d-cache-block-size", NULL);
317			/* fallback if block size missing */
318			if (lsizep == NULL)
319				lsizep = of_get_property(np, "d-cache-line-size", NULL);
320			if (lsizep != NULL)
321				lsize = *lsizep;
322			if (sizep == 0 || lsizep == 0)
323				DBG("Argh, can't find dcache properties ! "
324				    "sizep: %p, lsizep: %p\n", sizep, lsizep);
325
326			ppc64_caches.dsize = size;
327			ppc64_caches.dline_size = lsize;
328			ppc64_caches.log_dline_size = __ilog2(lsize);
329			ppc64_caches.dlines_per_page = PAGE_SIZE / lsize;
330
331			size = 0;
332			lsize = cur_cpu_spec->icache_bsize;
333			sizep = of_get_property(np, "i-cache-size", NULL);
334			if (sizep != NULL)
335				size = *sizep;
336			lsizep = of_get_property(np, "i-cache-block-size", NULL);
337			if (lsizep == NULL)
338				lsizep = of_get_property(np, "i-cache-line-size", NULL);
339			if (lsizep != NULL)
340				lsize = *lsizep;
341			if (sizep == 0 || lsizep == 0)
342				DBG("Argh, can't find icache properties ! "
343				    "sizep: %p, lsizep: %p\n", sizep, lsizep);
344
345			ppc64_caches.isize = size;
346			ppc64_caches.iline_size = lsize;
347			ppc64_caches.log_iline_size = __ilog2(lsize);
348			ppc64_caches.ilines_per_page = PAGE_SIZE / lsize;
349		}
350	}
351
352	DBG(" <- initialize_cache_info()\n");
353}
354
355
356/*
357 * Do some initial setup of the system.  The parameters are those which
358 * were passed in from the bootloader.
359 */
360void __init setup_system(void)
361{
362	DBG(" -> setup_system()\n");
363
364	/* Apply the CPUs-specific and firmware specific fixups to kernel
365	 * text (nop out sections not relevant to this CPU or this firmware)
366	 */
367	do_feature_fixups(cur_cpu_spec->cpu_features,
368			  &__start___ftr_fixup, &__stop___ftr_fixup);
369	do_feature_fixups(powerpc_firmware_features,
370			  &__start___fw_ftr_fixup, &__stop___fw_ftr_fixup);
371
372	/*
373	 * Unflatten the device-tree passed by prom_init or kexec
374	 */
375	unflatten_device_tree();
376
377	/*
378	 * Fill the ppc64_caches & systemcfg structures with informations
379 	 * retrieved from the device-tree.
380	 */
381	initialize_cache_info();
382
383	/*
384	 * Initialize irq remapping subsystem
385	 */
386	irq_early_init();
387
388#ifdef CONFIG_PPC_RTAS
389	/*
390	 * Initialize RTAS if available
391	 */
392	rtas_initialize();
393#endif /* CONFIG_PPC_RTAS */
394
395	/*
396	 * Check if we have an initrd provided via the device-tree
397	 */
398	check_for_initrd();
399
400	/*
401	 * Do some platform specific early initializations, that includes
402	 * setting up the hash table pointers. It also sets up some interrupt-mapping
403	 * related options that will be used by finish_device_tree()
404	 */
405	if (ppc_md.init_early)
406		ppc_md.init_early();
407
408 	/*
409	 * We can discover serial ports now since the above did setup the
410	 * hash table management for us, thus ioremap works. We do that early
411	 * so that further code can be debugged
412	 */
413	find_legacy_serial_ports();
414
415	/*
416	 * Register early console
417	 */
418	register_early_udbg_console();
419
420	/*
421	 * Initialize xmon
422	 */
423	xmon_setup();
424
425	check_smt_enabled();
426	smp_setup_cpu_maps();
427
428#ifdef CONFIG_SMP
429	/* Release secondary cpus out of their spinloops at 0x60 now that
430	 * we can map physical -> logical CPU ids
431	 */
432	smp_release_cpus();
433#endif
434
435	printk("Starting Linux PPC64 %s\n", init_utsname()->version);
436
437	printk("-----------------------------------------------------\n");
438	printk("ppc64_pft_size                = 0x%lx\n", ppc64_pft_size);
439	printk("physicalMemorySize            = 0x%lx\n", lmb_phys_mem_size());
440	if (ppc64_caches.dline_size != 0x80)
441		printk("ppc64_caches.dcache_line_size = 0x%x\n",
442		       ppc64_caches.dline_size);
443	if (ppc64_caches.iline_size != 0x80)
444		printk("ppc64_caches.icache_line_size = 0x%x\n",
445		       ppc64_caches.iline_size);
446	if (htab_address)
447		printk("htab_address                  = 0x%p\n", htab_address);
448	printk("htab_hash_mask                = 0x%lx\n", htab_hash_mask);
449#if PHYSICAL_START > 0
450	printk("physical_start                = 0x%lx\n", PHYSICAL_START);
451#endif
452	printk("-----------------------------------------------------\n");
453
454	DBG(" <- setup_system()\n");
455}
456
457#ifdef CONFIG_IRQSTACKS
458static void __init irqstack_early_init(void)
459{
460	unsigned int i;
461
462	/*
463	 * interrupt stacks must be under 256MB, we cannot afford to take
464	 * SLB misses on them.
465	 */
466	for_each_possible_cpu(i) {
467		softirq_ctx[i] = (struct thread_info *)
468			__va(lmb_alloc_base(THREAD_SIZE,
469					    THREAD_SIZE, 0x10000000));
470		hardirq_ctx[i] = (struct thread_info *)
471			__va(lmb_alloc_base(THREAD_SIZE,
472					    THREAD_SIZE, 0x10000000));
473	}
474}
475#else
476#define irqstack_early_init()
477#endif
478
479/*
480 * Stack space used when we detect a bad kernel stack pointer, and
481 * early in SMP boots before relocation is enabled.
482 */
483static void __init emergency_stack_init(void)
484{
485	unsigned long limit;
486	unsigned int i;
487
488	/*
489	 * Emergency stacks must be under 256MB, we cannot afford to take
490	 * SLB misses on them. The ABI also requires them to be 128-byte
491	 * aligned.
492	 *
493	 * Since we use these as temporary stacks during secondary CPU
494	 * bringup, we need to get at them in real mode. This means they
495	 * must also be within the RMO region.
496	 */
497	limit = min(0x10000000UL, lmb.rmo_size);
498
499	for_each_possible_cpu(i) {
500		unsigned long sp;
501		sp  = lmb_alloc_base(THREAD_SIZE, THREAD_SIZE, limit);
502		sp += THREAD_SIZE;
503		paca[i].emergency_sp = __va(sp);
504	}
505}
506
507/*
508 * Called into from start_kernel, after lock_kernel has been called.
509 * Initializes bootmem, which is unsed to manage page allocation until
510 * mem_init is called.
511 */
512void __init setup_arch(char **cmdline_p)
513{
514	ppc64_boot_msg(0x12, "Setup Arch");
515
516	*cmdline_p = cmd_line;
517
518	/*
519	 * Set cache line size based on type of cpu as a default.
520	 * Systems with OF can look in the properties on the cpu node(s)
521	 * for a possibly more accurate value.
522	 */
523	dcache_bsize = ppc64_caches.dline_size;
524	icache_bsize = ppc64_caches.iline_size;
525
526	/* reboot on panic */
527	panic_timeout = 180;
528
529	if (ppc_md.panic)
530		setup_panic();
531
532	init_mm.start_code = (unsigned long)_stext;
533	init_mm.end_code = (unsigned long) _etext;
534	init_mm.end_data = (unsigned long) _edata;
535	init_mm.brk = klimit;
536
537	irqstack_early_init();
538	emergency_stack_init();
539
540	stabs_alloc();
541
542	/* set up the bootmem stuff with available memory */
543	do_init_bootmem();
544	sparse_init();
545
546#ifdef CONFIG_DUMMY_CONSOLE
547	conswitchp = &dummy_con;
548#endif
549
550	if (ppc_md.setup_arch)
551		ppc_md.setup_arch();
552
553	paging_init();
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
582/* Print a termination message (print only -- does not stop the kernel) */
583void ppc64_terminate_msg(unsigned int src, const char *msg)
584{
585	ppc64_do_msg(PPC64_LINUX_FUNCTION|PPC64_TERM_MESSAGE|src, msg);
586	printk("[terminate]%04x %s\n", src, msg);
587}
588
589void cpu_die(void)
590{
591	if (ppc_md.cpu_die)
592		ppc_md.cpu_die();
593}
594
595#ifdef CONFIG_SMP
596void __init setup_per_cpu_areas(void)
597{
598	int i;
599	unsigned long size;
600	char *ptr;
601
602	/* Copy section for each CPU (we discard the original) */
603	size = ALIGN(__per_cpu_end - __per_cpu_start, PAGE_SIZE);
604#ifdef CONFIG_MODULES
605	if (size < PERCPU_ENOUGH_ROOM)
606		size = PERCPU_ENOUGH_ROOM;
607#endif
608
609	for_each_possible_cpu(i) {
610		ptr = alloc_bootmem_pages_node(NODE_DATA(cpu_to_node(i)), size);
611		if (!ptr)
612			panic("Cannot allocate cpu data for CPU %d\n", i);
613
614		paca[i].data_offset = ptr - __per_cpu_start;
615		memcpy(ptr, __per_cpu_start, __per_cpu_end - __per_cpu_start);
616	}
617
618	/* Now that per_cpu is setup, initialize cpu_sibling_map */
619	smp_setup_cpu_sibling_map();
620}
621#endif
622
623
624#ifdef CONFIG_PPC_INDIRECT_IO
625struct ppc_pci_io ppc_pci_io;
626EXPORT_SYMBOL(ppc_pci_io);
627#endif /* CONFIG_PPC_INDIRECT_IO */
628
629