setup_64.c revision 376af5947c0e441ccbf98f0212d4ffbf171528f6
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#define DEBUG
14
15#include <linux/export.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/memblock.h>
38#include <linux/hugetlb.h>
39#include <linux/memory.h>
40
41#include <asm/io.h>
42#include <asm/kdump.h>
43#include <asm/prom.h>
44#include <asm/processor.h>
45#include <asm/pgtable.h>
46#include <asm/smp.h>
47#include <asm/elf.h>
48#include <asm/machdep.h>
49#include <asm/paca.h>
50#include <asm/time.h>
51#include <asm/cputable.h>
52#include <asm/sections.h>
53#include <asm/btext.h>
54#include <asm/nvram.h>
55#include <asm/setup.h>
56#include <asm/rtas.h>
57#include <asm/iommu.h>
58#include <asm/serial.h>
59#include <asm/cache.h>
60#include <asm/page.h>
61#include <asm/mmu.h>
62#include <asm/firmware.h>
63#include <asm/xmon.h>
64#include <asm/udbg.h>
65#include <asm/kexec.h>
66#include <asm/mmu_context.h>
67#include <asm/code-patching.h>
68#include <asm/kvm_ppc.h>
69#include <asm/hugetlb.h>
70#include <asm/epapr_hcalls.h>
71
72#ifdef DEBUG
73#define DBG(fmt...) udbg_printf(fmt)
74#else
75#define DBG(fmt...)
76#endif
77
78int spinning_secondaries;
79u64 ppc64_pft_size;
80
81/* Pick defaults since we might want to patch instructions
82 * before we've read this from the device tree.
83 */
84struct ppc64_caches ppc64_caches = {
85	.dline_size = 0x40,
86	.log_dline_size = 6,
87	.iline_size = 0x40,
88	.log_iline_size = 6
89};
90EXPORT_SYMBOL_GPL(ppc64_caches);
91
92/*
93 * These are used in binfmt_elf.c to put aux entries on the stack
94 * for each elf executable being started.
95 */
96int dcache_bsize;
97int icache_bsize;
98int ucache_bsize;
99
100#if defined(CONFIG_PPC_BOOK3E) && defined(CONFIG_SMP)
101static void setup_tlb_core_data(void)
102{
103	int cpu;
104
105	BUILD_BUG_ON(offsetof(struct tlb_core_data, lock) != 0);
106
107	for_each_possible_cpu(cpu) {
108		int first = cpu_first_thread_sibling(cpu);
109
110		paca[cpu].tcd_ptr = &paca[first].tcd;
111
112		/*
113		 * If we have threads, we need either tlbsrx.
114		 * or e6500 tablewalk mode, or else TLB handlers
115		 * will be racy and could produce duplicate entries.
116		 */
117		if (smt_enabled_at_boot >= 2 &&
118		    !mmu_has_feature(MMU_FTR_USE_TLBRSRV) &&
119		    book3e_htw_mode != PPC_HTW_E6500) {
120			/* Should we panic instead? */
121			WARN_ONCE("%s: unsupported MMU configuration -- expect problems\n",
122				  __func__);
123		}
124	}
125}
126#else
127static void setup_tlb_core_data(void)
128{
129}
130#endif
131
132#ifdef CONFIG_SMP
133
134static char *smt_enabled_cmdline;
135
136/* Look for ibm,smt-enabled OF option */
137static void check_smt_enabled(void)
138{
139	struct device_node *dn;
140	const char *smt_option;
141
142	/* Default to enabling all threads */
143	smt_enabled_at_boot = threads_per_core;
144
145	/* Allow the command line to overrule the OF option */
146	if (smt_enabled_cmdline) {
147		if (!strcmp(smt_enabled_cmdline, "on"))
148			smt_enabled_at_boot = threads_per_core;
149		else if (!strcmp(smt_enabled_cmdline, "off"))
150			smt_enabled_at_boot = 0;
151		else {
152			long smt;
153			int rc;
154
155			rc = strict_strtol(smt_enabled_cmdline, 10, &smt);
156			if (!rc)
157				smt_enabled_at_boot =
158					min(threads_per_core, (int)smt);
159		}
160	} else {
161		dn = of_find_node_by_path("/options");
162		if (dn) {
163			smt_option = of_get_property(dn, "ibm,smt-enabled",
164						     NULL);
165
166			if (smt_option) {
167				if (!strcmp(smt_option, "on"))
168					smt_enabled_at_boot = threads_per_core;
169				else if (!strcmp(smt_option, "off"))
170					smt_enabled_at_boot = 0;
171			}
172
173			of_node_put(dn);
174		}
175	}
176}
177
178/* Look for smt-enabled= cmdline option */
179static int __init early_smt_enabled(char *p)
180{
181	smt_enabled_cmdline = p;
182	return 0;
183}
184early_param("smt-enabled", early_smt_enabled);
185
186#else
187#define check_smt_enabled()
188#endif /* CONFIG_SMP */
189
190/** Fix up paca fields required for the boot cpu */
191static void fixup_boot_paca(void)
192{
193	/* The boot cpu is started */
194	get_paca()->cpu_start = 1;
195	/* Allow percpu accesses to work until we setup percpu data */
196	get_paca()->data_offset = 0;
197}
198
199static void cpu_ready_for_interrupts(void)
200{
201	/* Set IR and DR in PACA MSR */
202	get_paca()->kernel_msr = MSR_KERNEL;
203
204	/* Enable AIL if supported */
205	if (cpu_has_feature(CPU_FTR_HVMODE) &&
206	    cpu_has_feature(CPU_FTR_ARCH_207S)) {
207		unsigned long lpcr = mfspr(SPRN_LPCR);
208		mtspr(SPRN_LPCR, lpcr | LPCR_AIL_3);
209	}
210}
211
212/*
213 * Early initialization entry point. This is called by head.S
214 * with MMU translation disabled. We rely on the "feature" of
215 * the CPU that ignores the top 2 bits of the address in real
216 * mode so we can access kernel globals normally provided we
217 * only toy with things in the RMO region. From here, we do
218 * some early parsing of the device-tree to setup out MEMBLOCK
219 * data structures, and allocate & initialize the hash table
220 * and segment tables so we can start running with translation
221 * enabled.
222 *
223 * It is this function which will call the probe() callback of
224 * the various platform types and copy the matching one to the
225 * global ppc_md structure. Your platform can eventually do
226 * some very early initializations from the probe() routine, but
227 * this is not recommended, be very careful as, for example, the
228 * device-tree is not accessible via normal means at this point.
229 */
230
231void __init early_setup(unsigned long dt_ptr)
232{
233	static __initdata struct paca_struct boot_paca;
234
235	/* -------- printk is _NOT_ safe to use here ! ------- */
236
237	/* Identify CPU type */
238	identify_cpu(0, mfspr(SPRN_PVR));
239
240	/* Assume we're on cpu 0 for now. Don't write to the paca yet! */
241	initialise_paca(&boot_paca, 0);
242	setup_paca(&boot_paca);
243	fixup_boot_paca();
244
245	/* Initialize lockdep early or else spinlocks will blow */
246	lockdep_init();
247
248	/* -------- printk is now safe to use ------- */
249
250	/* Enable early debugging if any specified (see udbg.h) */
251	udbg_early_init();
252
253 	DBG(" -> early_setup(), dt_ptr: 0x%lx\n", dt_ptr);
254
255	/*
256	 * Do early initialization using the flattened device
257	 * tree, such as retrieving the physical memory map or
258	 * calculating/retrieving the hash table size.
259	 */
260	early_init_devtree(__va(dt_ptr));
261
262	epapr_paravirt_early_init();
263
264	/* Now we know the logical id of our boot cpu, setup the paca. */
265	setup_paca(&paca[boot_cpuid]);
266	fixup_boot_paca();
267
268	/* Probe the machine type */
269	probe_machine();
270
271	setup_kdump_trampoline();
272
273	DBG("Found, Initializing memory management...\n");
274
275	/* Initialize the hash table or TLB handling */
276	early_init_mmu();
277
278	/*
279	 * At this point, we can let interrupts switch to virtual mode
280	 * (the MMU has been setup), so adjust the MSR in the PACA to
281	 * have IR and DR set and enable AIL if it exists
282	 */
283	cpu_ready_for_interrupts();
284
285	/* Reserve large chunks of memory for use by CMA for KVM */
286	kvm_cma_reserve();
287
288	/*
289	 * Reserve any gigantic pages requested on the command line.
290	 * memblock needs to have been initialized by the time this is
291	 * called since this will reserve memory.
292	 */
293	reserve_hugetlb_gpages();
294
295	DBG(" <- early_setup()\n");
296
297#ifdef CONFIG_PPC_EARLY_DEBUG_BOOTX
298	/*
299	 * This needs to be done *last* (after the above DBG() even)
300	 *
301	 * Right after we return from this function, we turn on the MMU
302	 * which means the real-mode access trick that btext does will
303	 * no longer work, it needs to switch to using a real MMU
304	 * mapping. This call will ensure that it does
305	 */
306	btext_map();
307#endif /* CONFIG_PPC_EARLY_DEBUG_BOOTX */
308}
309
310#ifdef CONFIG_SMP
311void early_setup_secondary(void)
312{
313	/* Mark interrupts enabled in PACA */
314	get_paca()->soft_enabled = 0;
315
316	/* Initialize the hash table or TLB handling */
317	early_init_mmu_secondary();
318
319	/*
320	 * At this point, we can let interrupts switch to virtual mode
321	 * (the MMU has been setup), so adjust the MSR in the PACA to
322	 * have IR and DR set.
323	 */
324	cpu_ready_for_interrupts();
325}
326
327#endif /* CONFIG_SMP */
328
329#if defined(CONFIG_SMP) || defined(CONFIG_KEXEC)
330void smp_release_cpus(void)
331{
332	unsigned long *ptr;
333	int i;
334
335	DBG(" -> smp_release_cpus()\n");
336
337	/* All secondary cpus are spinning on a common spinloop, release them
338	 * all now so they can start to spin on their individual paca
339	 * spinloops. For non SMP kernels, the secondary cpus never get out
340	 * of the common spinloop.
341	 */
342
343	ptr  = (unsigned long *)((unsigned long)&__secondary_hold_spinloop
344			- PHYSICAL_START);
345	*ptr = ppc_function_entry(generic_secondary_smp_init);
346
347	/* And wait a bit for them to catch up */
348	for (i = 0; i < 100000; i++) {
349		mb();
350		HMT_low();
351		if (spinning_secondaries == 0)
352			break;
353		udelay(1);
354	}
355	DBG("spinning_secondaries = %d\n", spinning_secondaries);
356
357	DBG(" <- smp_release_cpus()\n");
358}
359#endif /* CONFIG_SMP || CONFIG_KEXEC */
360
361/*
362 * Initialize some remaining members of the ppc64_caches and systemcfg
363 * structures
364 * (at least until we get rid of them completely). This is mostly some
365 * cache informations about the CPU that will be used by cache flush
366 * routines and/or provided to userland
367 */
368static void __init initialize_cache_info(void)
369{
370	struct device_node *np;
371	unsigned long num_cpus = 0;
372
373	DBG(" -> initialize_cache_info()\n");
374
375	for_each_node_by_type(np, "cpu") {
376		num_cpus += 1;
377
378		/*
379		 * We're assuming *all* of the CPUs have the same
380		 * d-cache and i-cache sizes... -Peter
381		 */
382		if (num_cpus == 1) {
383			const __be32 *sizep, *lsizep;
384			u32 size, lsize;
385
386			size = 0;
387			lsize = cur_cpu_spec->dcache_bsize;
388			sizep = of_get_property(np, "d-cache-size", NULL);
389			if (sizep != NULL)
390				size = be32_to_cpu(*sizep);
391			lsizep = of_get_property(np, "d-cache-block-size",
392						 NULL);
393			/* fallback if block size missing */
394			if (lsizep == NULL)
395				lsizep = of_get_property(np,
396							 "d-cache-line-size",
397							 NULL);
398			if (lsizep != NULL)
399				lsize = be32_to_cpu(*lsizep);
400			if (sizep == NULL || lsizep == NULL)
401				DBG("Argh, can't find dcache properties ! "
402				    "sizep: %p, lsizep: %p\n", sizep, lsizep);
403
404			ppc64_caches.dsize = size;
405			ppc64_caches.dline_size = lsize;
406			ppc64_caches.log_dline_size = __ilog2(lsize);
407			ppc64_caches.dlines_per_page = PAGE_SIZE / lsize;
408
409			size = 0;
410			lsize = cur_cpu_spec->icache_bsize;
411			sizep = of_get_property(np, "i-cache-size", NULL);
412			if (sizep != NULL)
413				size = be32_to_cpu(*sizep);
414			lsizep = of_get_property(np, "i-cache-block-size",
415						 NULL);
416			if (lsizep == NULL)
417				lsizep = of_get_property(np,
418							 "i-cache-line-size",
419							 NULL);
420			if (lsizep != NULL)
421				lsize = be32_to_cpu(*lsizep);
422			if (sizep == NULL || lsizep == NULL)
423				DBG("Argh, can't find icache properties ! "
424				    "sizep: %p, lsizep: %p\n", sizep, lsizep);
425
426			ppc64_caches.isize = size;
427			ppc64_caches.iline_size = lsize;
428			ppc64_caches.log_iline_size = __ilog2(lsize);
429			ppc64_caches.ilines_per_page = PAGE_SIZE / lsize;
430		}
431	}
432
433	DBG(" <- initialize_cache_info()\n");
434}
435
436
437/*
438 * Do some initial setup of the system.  The parameters are those which
439 * were passed in from the bootloader.
440 */
441void __init setup_system(void)
442{
443	DBG(" -> setup_system()\n");
444
445	/* Apply the CPUs-specific and firmware specific fixups to kernel
446	 * text (nop out sections not relevant to this CPU or this firmware)
447	 */
448	do_feature_fixups(cur_cpu_spec->cpu_features,
449			  &__start___ftr_fixup, &__stop___ftr_fixup);
450	do_feature_fixups(cur_cpu_spec->mmu_features,
451			  &__start___mmu_ftr_fixup, &__stop___mmu_ftr_fixup);
452	do_feature_fixups(powerpc_firmware_features,
453			  &__start___fw_ftr_fixup, &__stop___fw_ftr_fixup);
454	do_lwsync_fixups(cur_cpu_spec->cpu_features,
455			 &__start___lwsync_fixup, &__stop___lwsync_fixup);
456	do_final_fixups();
457
458	/*
459	 * Unflatten the device-tree passed by prom_init or kexec
460	 */
461	unflatten_device_tree();
462
463	/*
464	 * Fill the ppc64_caches & systemcfg structures with informations
465 	 * retrieved from the device-tree.
466	 */
467	initialize_cache_info();
468
469#ifdef CONFIG_PPC_RTAS
470	/*
471	 * Initialize RTAS if available
472	 */
473	rtas_initialize();
474#endif /* CONFIG_PPC_RTAS */
475
476	/*
477	 * Check if we have an initrd provided via the device-tree
478	 */
479	check_for_initrd();
480
481	/*
482	 * Do some platform specific early initializations, that includes
483	 * setting up the hash table pointers. It also sets up some interrupt-mapping
484	 * related options that will be used by finish_device_tree()
485	 */
486	if (ppc_md.init_early)
487		ppc_md.init_early();
488
489 	/*
490	 * We can discover serial ports now since the above did setup the
491	 * hash table management for us, thus ioremap works. We do that early
492	 * so that further code can be debugged
493	 */
494	find_legacy_serial_ports();
495
496	/*
497	 * Register early console
498	 */
499	register_early_udbg_console();
500
501	/*
502	 * Initialize xmon
503	 */
504	xmon_setup();
505
506	smp_setup_cpu_maps();
507	check_smt_enabled();
508	setup_tlb_core_data();
509
510#ifdef CONFIG_SMP
511	/* Release secondary cpus out of their spinloops at 0x60 now that
512	 * we can map physical -> logical CPU ids
513	 */
514	smp_release_cpus();
515#endif
516
517	printk("Starting Linux PPC64 %s\n", init_utsname()->version);
518
519	printk("-----------------------------------------------------\n");
520	printk("ppc64_pft_size                = 0x%llx\n", ppc64_pft_size);
521	printk("physicalMemorySize            = 0x%llx\n", memblock_phys_mem_size());
522	if (ppc64_caches.dline_size != 0x80)
523		printk("ppc64_caches.dcache_line_size = 0x%x\n",
524		       ppc64_caches.dline_size);
525	if (ppc64_caches.iline_size != 0x80)
526		printk("ppc64_caches.icache_line_size = 0x%x\n",
527		       ppc64_caches.iline_size);
528#ifdef CONFIG_PPC_STD_MMU_64
529	if (htab_address)
530		printk("htab_address                  = 0x%p\n", htab_address);
531	printk("htab_hash_mask                = 0x%lx\n", htab_hash_mask);
532#endif /* CONFIG_PPC_STD_MMU_64 */
533	if (PHYSICAL_START > 0)
534		printk("physical_start                = 0x%llx\n",
535		       (unsigned long long)PHYSICAL_START);
536	printk("-----------------------------------------------------\n");
537
538	DBG(" <- setup_system()\n");
539}
540
541/* This returns the limit below which memory accesses to the linear
542 * mapping are guarnateed not to cause a TLB or SLB miss. This is
543 * used to allocate interrupt or emergency stacks for which our
544 * exception entry path doesn't deal with being interrupted.
545 */
546static u64 safe_stack_limit(void)
547{
548#ifdef CONFIG_PPC_BOOK3E
549	/* Freescale BookE bolts the entire linear mapping */
550	if (mmu_has_feature(MMU_FTR_TYPE_FSL_E))
551		return linear_map_top;
552	/* Other BookE, we assume the first GB is bolted */
553	return 1ul << 30;
554#else
555	/* BookS, the first segment is bolted */
556	if (mmu_has_feature(MMU_FTR_1T_SEGMENT))
557		return 1UL << SID_SHIFT_1T;
558	return 1UL << SID_SHIFT;
559#endif
560}
561
562static void __init irqstack_early_init(void)
563{
564	u64 limit = safe_stack_limit();
565	unsigned int i;
566
567	/*
568	 * Interrupt stacks must be in the first segment since we
569	 * cannot afford to take SLB misses on them.
570	 */
571	for_each_possible_cpu(i) {
572		softirq_ctx[i] = (struct thread_info *)
573			__va(memblock_alloc_base(THREAD_SIZE,
574					    THREAD_SIZE, limit));
575		hardirq_ctx[i] = (struct thread_info *)
576			__va(memblock_alloc_base(THREAD_SIZE,
577					    THREAD_SIZE, limit));
578	}
579}
580
581#ifdef CONFIG_PPC_BOOK3E
582static void __init exc_lvl_early_init(void)
583{
584	unsigned int i;
585	unsigned long sp;
586
587	for_each_possible_cpu(i) {
588		sp = memblock_alloc(THREAD_SIZE, THREAD_SIZE);
589		critirq_ctx[i] = (struct thread_info *)__va(sp);
590		paca[i].crit_kstack = __va(sp + THREAD_SIZE);
591
592		sp = memblock_alloc(THREAD_SIZE, THREAD_SIZE);
593		dbgirq_ctx[i] = (struct thread_info *)__va(sp);
594		paca[i].dbg_kstack = __va(sp + THREAD_SIZE);
595
596		sp = memblock_alloc(THREAD_SIZE, THREAD_SIZE);
597		mcheckirq_ctx[i] = (struct thread_info *)__va(sp);
598		paca[i].mc_kstack = __va(sp + THREAD_SIZE);
599	}
600
601	if (cpu_has_feature(CPU_FTR_DEBUG_LVL_EXC))
602		patch_exception(0x040, exc_debug_debug_book3e);
603}
604#else
605#define exc_lvl_early_init()
606#endif
607
608/*
609 * Stack space used when we detect a bad kernel stack pointer, and
610 * early in SMP boots before relocation is enabled. Exclusive emergency
611 * stack for machine checks.
612 */
613static void __init emergency_stack_init(void)
614{
615	u64 limit;
616	unsigned int i;
617
618	/*
619	 * Emergency stacks must be under 256MB, we cannot afford to take
620	 * SLB misses on them. The ABI also requires them to be 128-byte
621	 * aligned.
622	 *
623	 * Since we use these as temporary stacks during secondary CPU
624	 * bringup, we need to get at them in real mode. This means they
625	 * must also be within the RMO region.
626	 */
627	limit = min(safe_stack_limit(), ppc64_rma_size);
628
629	for_each_possible_cpu(i) {
630		unsigned long sp;
631		sp  = memblock_alloc_base(THREAD_SIZE, THREAD_SIZE, limit);
632		sp += THREAD_SIZE;
633		paca[i].emergency_sp = __va(sp);
634
635#ifdef CONFIG_PPC_BOOK3S_64
636		/* emergency stack for machine check exception handling. */
637		sp  = memblock_alloc_base(THREAD_SIZE, THREAD_SIZE, limit);
638		sp += THREAD_SIZE;
639		paca[i].mc_emergency_sp = __va(sp);
640#endif
641	}
642}
643
644/*
645 * Called into from start_kernel this initializes bootmem, which is used
646 * to manage page allocation until mem_init is called.
647 */
648void __init setup_arch(char **cmdline_p)
649{
650	ppc64_boot_msg(0x12, "Setup Arch");
651
652	*cmdline_p = cmd_line;
653
654	/*
655	 * Set cache line size based on type of cpu as a default.
656	 * Systems with OF can look in the properties on the cpu node(s)
657	 * for a possibly more accurate value.
658	 */
659	dcache_bsize = ppc64_caches.dline_size;
660	icache_bsize = ppc64_caches.iline_size;
661
662	if (ppc_md.panic)
663		setup_panic();
664
665	init_mm.start_code = (unsigned long)_stext;
666	init_mm.end_code = (unsigned long) _etext;
667	init_mm.end_data = (unsigned long) _edata;
668	init_mm.brk = klimit;
669#ifdef CONFIG_PPC_64K_PAGES
670	init_mm.context.pte_frag = NULL;
671#endif
672	irqstack_early_init();
673	exc_lvl_early_init();
674	emergency_stack_init();
675
676	/* set up the bootmem stuff with available memory */
677	do_init_bootmem();
678	sparse_init();
679
680#ifdef CONFIG_DUMMY_CONSOLE
681	conswitchp = &dummy_con;
682#endif
683
684	if (ppc_md.setup_arch)
685		ppc_md.setup_arch();
686
687	paging_init();
688
689	/* Initialize the MMU context management stuff */
690	mmu_context_init();
691
692	/* Interrupt code needs to be 64K-aligned */
693	if ((unsigned long)_stext & 0xffff)
694		panic("Kernelbase not 64K-aligned (0x%lx)!\n",
695		      (unsigned long)_stext);
696
697	ppc64_boot_msg(0x15, "Setup Done");
698}
699
700
701/* ToDo: do something useful if ppc_md is not yet setup. */
702#define PPC64_LINUX_FUNCTION 0x0f000000
703#define PPC64_IPL_MESSAGE 0xc0000000
704#define PPC64_TERM_MESSAGE 0xb0000000
705
706static void ppc64_do_msg(unsigned int src, const char *msg)
707{
708	if (ppc_md.progress) {
709		char buf[128];
710
711		sprintf(buf, "%08X\n", src);
712		ppc_md.progress(buf, 0);
713		snprintf(buf, 128, "%s", msg);
714		ppc_md.progress(buf, 0);
715	}
716}
717
718/* Print a boot progress message. */
719void ppc64_boot_msg(unsigned int src, const char *msg)
720{
721	ppc64_do_msg(PPC64_LINUX_FUNCTION|PPC64_IPL_MESSAGE|src, msg);
722	printk("[boot]%04x %s\n", src, msg);
723}
724
725#ifdef CONFIG_SMP
726#define PCPU_DYN_SIZE		()
727
728static void * __init pcpu_fc_alloc(unsigned int cpu, size_t size, size_t align)
729{
730	return __alloc_bootmem_node(NODE_DATA(cpu_to_node(cpu)), size, align,
731				    __pa(MAX_DMA_ADDRESS));
732}
733
734static void __init pcpu_fc_free(void *ptr, size_t size)
735{
736	free_bootmem(__pa(ptr), size);
737}
738
739static int pcpu_cpu_distance(unsigned int from, unsigned int to)
740{
741	if (cpu_to_node(from) == cpu_to_node(to))
742		return LOCAL_DISTANCE;
743	else
744		return REMOTE_DISTANCE;
745}
746
747unsigned long __per_cpu_offset[NR_CPUS] __read_mostly;
748EXPORT_SYMBOL(__per_cpu_offset);
749
750void __init setup_per_cpu_areas(void)
751{
752	const size_t dyn_size = PERCPU_MODULE_RESERVE + PERCPU_DYNAMIC_RESERVE;
753	size_t atom_size;
754	unsigned long delta;
755	unsigned int cpu;
756	int rc;
757
758	/*
759	 * Linear mapping is one of 4K, 1M and 16M.  For 4K, no need
760	 * to group units.  For larger mappings, use 1M atom which
761	 * should be large enough to contain a number of units.
762	 */
763	if (mmu_linear_psize == MMU_PAGE_4K)
764		atom_size = PAGE_SIZE;
765	else
766		atom_size = 1 << 20;
767
768	rc = pcpu_embed_first_chunk(0, dyn_size, atom_size, pcpu_cpu_distance,
769				    pcpu_fc_alloc, pcpu_fc_free);
770	if (rc < 0)
771		panic("cannot initialize percpu area (err=%d)", rc);
772
773	delta = (unsigned long)pcpu_base_addr - (unsigned long)__per_cpu_start;
774	for_each_possible_cpu(cpu) {
775                __per_cpu_offset[cpu] = delta + pcpu_unit_offsets[cpu];
776		paca[cpu].data_offset = __per_cpu_offset[cpu];
777	}
778}
779#endif
780
781#ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
782unsigned long memory_block_size_bytes(void)
783{
784	if (ppc_md.memory_block_size)
785		return ppc_md.memory_block_size();
786
787	return MIN_MEMORY_BLOCK_SIZE;
788}
789#endif
790
791#if defined(CONFIG_PPC_INDIRECT_PIO) || defined(CONFIG_PPC_INDIRECT_MMIO)
792struct ppc_pci_io ppc_pci_io;
793EXPORT_SYMBOL(ppc_pci_io);
794#endif
795