1/*
2 * Procedures for interfacing to Open Firmware.
3 *
4 * Paul Mackerras	August 1996.
5 * Copyright (C) 1996-2005 Paul Mackerras.
6 *
7 *  Adapted for 64bit PowerPC by Dave Engebretsen and Peter Bergner.
8 *    {engebret|bergner}@us.ibm.com
9 *
10 *      This program is free software; you can redistribute it and/or
11 *      modify it under the terms of the GNU General Public License
12 *      as published by the Free Software Foundation; either version
13 *      2 of the License, or (at your option) any later version.
14 */
15
16#undef DEBUG_PROM
17
18#include <stdarg.h>
19#include <linux/kernel.h>
20#include <linux/string.h>
21#include <linux/init.h>
22#include <linux/threads.h>
23#include <linux/spinlock.h>
24#include <linux/types.h>
25#include <linux/pci.h>
26#include <linux/proc_fs.h>
27#include <linux/stringify.h>
28#include <linux/delay.h>
29#include <linux/initrd.h>
30#include <linux/bitops.h>
31#include <asm/prom.h>
32#include <asm/rtas.h>
33#include <asm/page.h>
34#include <asm/processor.h>
35#include <asm/irq.h>
36#include <asm/io.h>
37#include <asm/smp.h>
38#include <asm/mmu.h>
39#include <asm/pgtable.h>
40#include <asm/iommu.h>
41#include <asm/btext.h>
42#include <asm/sections.h>
43#include <asm/machdep.h>
44#include <asm/opal.h>
45
46#include <linux/linux_logo.h>
47
48/*
49 * Eventually bump that one up
50 */
51#define DEVTREE_CHUNK_SIZE	0x100000
52
53/*
54 * This is the size of the local memory reserve map that gets copied
55 * into the boot params passed to the kernel. That size is totally
56 * flexible as the kernel just reads the list until it encounters an
57 * entry with size 0, so it can be changed without breaking binary
58 * compatibility
59 */
60#define MEM_RESERVE_MAP_SIZE	8
61
62/*
63 * prom_init() is called very early on, before the kernel text
64 * and data have been mapped to KERNELBASE.  At this point the code
65 * is running at whatever address it has been loaded at.
66 * On ppc32 we compile with -mrelocatable, which means that references
67 * to extern and static variables get relocated automatically.
68 * ppc64 objects are always relocatable, we just need to relocate the
69 * TOC.
70 *
71 * Because OF may have mapped I/O devices into the area starting at
72 * KERNELBASE, particularly on CHRP machines, we can't safely call
73 * OF once the kernel has been mapped to KERNELBASE.  Therefore all
74 * OF calls must be done within prom_init().
75 *
76 * ADDR is used in calls to call_prom.  The 4th and following
77 * arguments to call_prom should be 32-bit values.
78 * On ppc64, 64 bit values are truncated to 32 bits (and
79 * fortunately don't get interpreted as two arguments).
80 */
81#define ADDR(x)		(u32)(unsigned long)(x)
82
83#ifdef CONFIG_PPC64
84#define OF_WORKAROUNDS	0
85#else
86#define OF_WORKAROUNDS	of_workarounds
87int of_workarounds;
88#endif
89
90#define OF_WA_CLAIM	1	/* do phys/virt claim separately, then map */
91#define OF_WA_LONGTRAIL	2	/* work around longtrail bugs */
92
93#define PROM_BUG() do {						\
94        prom_printf("kernel BUG at %s line 0x%x!\n",		\
95		    __FILE__, __LINE__);			\
96        __asm__ __volatile__(".long " BUG_ILLEGAL_INSTR);	\
97} while (0)
98
99#ifdef DEBUG_PROM
100#define prom_debug(x...)	prom_printf(x)
101#else
102#define prom_debug(x...)
103#endif
104
105
106typedef u32 prom_arg_t;
107
108struct prom_args {
109        __be32 service;
110        __be32 nargs;
111        __be32 nret;
112        __be32 args[10];
113};
114
115struct prom_t {
116	ihandle root;
117	phandle chosen;
118	int cpu;
119	ihandle stdout;
120	ihandle mmumap;
121	ihandle memory;
122};
123
124struct mem_map_entry {
125	__be64	base;
126	__be64	size;
127};
128
129typedef __be32 cell_t;
130
131extern void __start(unsigned long r3, unsigned long r4, unsigned long r5,
132		    unsigned long r6, unsigned long r7, unsigned long r8,
133		    unsigned long r9);
134
135#ifdef CONFIG_PPC64
136extern int enter_prom(struct prom_args *args, unsigned long entry);
137#else
138static inline int enter_prom(struct prom_args *args, unsigned long entry)
139{
140	return ((int (*)(struct prom_args *))entry)(args);
141}
142#endif
143
144extern void copy_and_flush(unsigned long dest, unsigned long src,
145			   unsigned long size, unsigned long offset);
146
147/* prom structure */
148static struct prom_t __initdata prom;
149
150static unsigned long prom_entry __initdata;
151
152#define PROM_SCRATCH_SIZE 256
153
154static char __initdata of_stdout_device[256];
155static char __initdata prom_scratch[PROM_SCRATCH_SIZE];
156
157static unsigned long __initdata dt_header_start;
158static unsigned long __initdata dt_struct_start, dt_struct_end;
159static unsigned long __initdata dt_string_start, dt_string_end;
160
161static unsigned long __initdata prom_initrd_start, prom_initrd_end;
162
163#ifdef CONFIG_PPC64
164static int __initdata prom_iommu_force_on;
165static int __initdata prom_iommu_off;
166static unsigned long __initdata prom_tce_alloc_start;
167static unsigned long __initdata prom_tce_alloc_end;
168#endif
169
170/* Platforms codes are now obsolete in the kernel. Now only used within this
171 * file and ultimately gone too. Feel free to change them if you need, they
172 * are not shared with anything outside of this file anymore
173 */
174#define PLATFORM_PSERIES	0x0100
175#define PLATFORM_PSERIES_LPAR	0x0101
176#define PLATFORM_LPAR		0x0001
177#define PLATFORM_POWERMAC	0x0400
178#define PLATFORM_GENERIC	0x0500
179#define PLATFORM_OPAL		0x0600
180
181static int __initdata of_platform;
182
183static char __initdata prom_cmd_line[COMMAND_LINE_SIZE];
184
185static unsigned long __initdata prom_memory_limit;
186
187static unsigned long __initdata alloc_top;
188static unsigned long __initdata alloc_top_high;
189static unsigned long __initdata alloc_bottom;
190static unsigned long __initdata rmo_top;
191static unsigned long __initdata ram_top;
192
193static struct mem_map_entry __initdata mem_reserve_map[MEM_RESERVE_MAP_SIZE];
194static int __initdata mem_reserve_cnt;
195
196static cell_t __initdata regbuf[1024];
197
198static bool rtas_has_query_cpu_stopped;
199
200
201/*
202 * Error results ... some OF calls will return "-1" on error, some
203 * will return 0, some will return either. To simplify, here are
204 * macros to use with any ihandle or phandle return value to check if
205 * it is valid
206 */
207
208#define PROM_ERROR		(-1u)
209#define PHANDLE_VALID(p)	((p) != 0 && (p) != PROM_ERROR)
210#define IHANDLE_VALID(i)	((i) != 0 && (i) != PROM_ERROR)
211
212
213/* This is the one and *ONLY* place where we actually call open
214 * firmware.
215 */
216
217static int __init call_prom(const char *service, int nargs, int nret, ...)
218{
219	int i;
220	struct prom_args args;
221	va_list list;
222
223	args.service = cpu_to_be32(ADDR(service));
224	args.nargs = cpu_to_be32(nargs);
225	args.nret = cpu_to_be32(nret);
226
227	va_start(list, nret);
228	for (i = 0; i < nargs; i++)
229		args.args[i] = cpu_to_be32(va_arg(list, prom_arg_t));
230	va_end(list);
231
232	for (i = 0; i < nret; i++)
233		args.args[nargs+i] = 0;
234
235	if (enter_prom(&args, prom_entry) < 0)
236		return PROM_ERROR;
237
238	return (nret > 0) ? be32_to_cpu(args.args[nargs]) : 0;
239}
240
241static int __init call_prom_ret(const char *service, int nargs, int nret,
242				prom_arg_t *rets, ...)
243{
244	int i;
245	struct prom_args args;
246	va_list list;
247
248	args.service = cpu_to_be32(ADDR(service));
249	args.nargs = cpu_to_be32(nargs);
250	args.nret = cpu_to_be32(nret);
251
252	va_start(list, rets);
253	for (i = 0; i < nargs; i++)
254		args.args[i] = cpu_to_be32(va_arg(list, prom_arg_t));
255	va_end(list);
256
257	for (i = 0; i < nret; i++)
258		args.args[nargs+i] = 0;
259
260	if (enter_prom(&args, prom_entry) < 0)
261		return PROM_ERROR;
262
263	if (rets != NULL)
264		for (i = 1; i < nret; ++i)
265			rets[i-1] = be32_to_cpu(args.args[nargs+i]);
266
267	return (nret > 0) ? be32_to_cpu(args.args[nargs]) : 0;
268}
269
270
271static void __init prom_print(const char *msg)
272{
273	const char *p, *q;
274
275	if (prom.stdout == 0)
276		return;
277
278	for (p = msg; *p != 0; p = q) {
279		for (q = p; *q != 0 && *q != '\n'; ++q)
280			;
281		if (q > p)
282			call_prom("write", 3, 1, prom.stdout, p, q - p);
283		if (*q == 0)
284			break;
285		++q;
286		call_prom("write", 3, 1, prom.stdout, ADDR("\r\n"), 2);
287	}
288}
289
290
291static void __init prom_print_hex(unsigned long val)
292{
293	int i, nibbles = sizeof(val)*2;
294	char buf[sizeof(val)*2+1];
295
296	for (i = nibbles-1;  i >= 0;  i--) {
297		buf[i] = (val & 0xf) + '0';
298		if (buf[i] > '9')
299			buf[i] += ('a'-'0'-10);
300		val >>= 4;
301	}
302	buf[nibbles] = '\0';
303	call_prom("write", 3, 1, prom.stdout, buf, nibbles);
304}
305
306/* max number of decimal digits in an unsigned long */
307#define UL_DIGITS 21
308static void __init prom_print_dec(unsigned long val)
309{
310	int i, size;
311	char buf[UL_DIGITS+1];
312
313	for (i = UL_DIGITS-1; i >= 0;  i--) {
314		buf[i] = (val % 10) + '0';
315		val = val/10;
316		if (val == 0)
317			break;
318	}
319	/* shift stuff down */
320	size = UL_DIGITS - i;
321	call_prom("write", 3, 1, prom.stdout, buf+i, size);
322}
323
324static void __init prom_printf(const char *format, ...)
325{
326	const char *p, *q, *s;
327	va_list args;
328	unsigned long v;
329	long vs;
330
331	va_start(args, format);
332	for (p = format; *p != 0; p = q) {
333		for (q = p; *q != 0 && *q != '\n' && *q != '%'; ++q)
334			;
335		if (q > p)
336			call_prom("write", 3, 1, prom.stdout, p, q - p);
337		if (*q == 0)
338			break;
339		if (*q == '\n') {
340			++q;
341			call_prom("write", 3, 1, prom.stdout,
342				  ADDR("\r\n"), 2);
343			continue;
344		}
345		++q;
346		if (*q == 0)
347			break;
348		switch (*q) {
349		case 's':
350			++q;
351			s = va_arg(args, const char *);
352			prom_print(s);
353			break;
354		case 'x':
355			++q;
356			v = va_arg(args, unsigned long);
357			prom_print_hex(v);
358			break;
359		case 'd':
360			++q;
361			vs = va_arg(args, int);
362			if (vs < 0) {
363				prom_print("-");
364				vs = -vs;
365			}
366			prom_print_dec(vs);
367			break;
368		case 'l':
369			++q;
370			if (*q == 0)
371				break;
372			else if (*q == 'x') {
373				++q;
374				v = va_arg(args, unsigned long);
375				prom_print_hex(v);
376			} else if (*q == 'u') { /* '%lu' */
377				++q;
378				v = va_arg(args, unsigned long);
379				prom_print_dec(v);
380			} else if (*q == 'd') { /* %ld */
381				++q;
382				vs = va_arg(args, long);
383				if (vs < 0) {
384					prom_print("-");
385					vs = -vs;
386				}
387				prom_print_dec(vs);
388			}
389			break;
390		}
391	}
392}
393
394
395static unsigned int __init prom_claim(unsigned long virt, unsigned long size,
396				unsigned long align)
397{
398
399	if (align == 0 && (OF_WORKAROUNDS & OF_WA_CLAIM)) {
400		/*
401		 * Old OF requires we claim physical and virtual separately
402		 * and then map explicitly (assuming virtual mode)
403		 */
404		int ret;
405		prom_arg_t result;
406
407		ret = call_prom_ret("call-method", 5, 2, &result,
408				    ADDR("claim"), prom.memory,
409				    align, size, virt);
410		if (ret != 0 || result == -1)
411			return -1;
412		ret = call_prom_ret("call-method", 5, 2, &result,
413				    ADDR("claim"), prom.mmumap,
414				    align, size, virt);
415		if (ret != 0) {
416			call_prom("call-method", 4, 1, ADDR("release"),
417				  prom.memory, size, virt);
418			return -1;
419		}
420		/* the 0x12 is M (coherence) + PP == read/write */
421		call_prom("call-method", 6, 1,
422			  ADDR("map"), prom.mmumap, 0x12, size, virt, virt);
423		return virt;
424	}
425	return call_prom("claim", 3, 1, (prom_arg_t)virt, (prom_arg_t)size,
426			 (prom_arg_t)align);
427}
428
429static void __init __attribute__((noreturn)) prom_panic(const char *reason)
430{
431	prom_print(reason);
432	/* Do not call exit because it clears the screen on pmac
433	 * it also causes some sort of double-fault on early pmacs */
434	if (of_platform == PLATFORM_POWERMAC)
435		asm("trap\n");
436
437	/* ToDo: should put up an SRC here on pSeries */
438	call_prom("exit", 0, 0);
439
440	for (;;)			/* should never get here */
441		;
442}
443
444
445static int __init prom_next_node(phandle *nodep)
446{
447	phandle node;
448
449	if ((node = *nodep) != 0
450	    && (*nodep = call_prom("child", 1, 1, node)) != 0)
451		return 1;
452	if ((*nodep = call_prom("peer", 1, 1, node)) != 0)
453		return 1;
454	for (;;) {
455		if ((node = call_prom("parent", 1, 1, node)) == 0)
456			return 0;
457		if ((*nodep = call_prom("peer", 1, 1, node)) != 0)
458			return 1;
459	}
460}
461
462static int inline prom_getprop(phandle node, const char *pname,
463			       void *value, size_t valuelen)
464{
465	return call_prom("getprop", 4, 1, node, ADDR(pname),
466			 (u32)(unsigned long) value, (u32) valuelen);
467}
468
469static int inline prom_getproplen(phandle node, const char *pname)
470{
471	return call_prom("getproplen", 2, 1, node, ADDR(pname));
472}
473
474static void add_string(char **str, const char *q)
475{
476	char *p = *str;
477
478	while (*q)
479		*p++ = *q++;
480	*p++ = ' ';
481	*str = p;
482}
483
484static char *tohex(unsigned int x)
485{
486	static char digits[] = "0123456789abcdef";
487	static char result[9];
488	int i;
489
490	result[8] = 0;
491	i = 8;
492	do {
493		--i;
494		result[i] = digits[x & 0xf];
495		x >>= 4;
496	} while (x != 0 && i > 0);
497	return &result[i];
498}
499
500static int __init prom_setprop(phandle node, const char *nodename,
501			       const char *pname, void *value, size_t valuelen)
502{
503	char cmd[256], *p;
504
505	if (!(OF_WORKAROUNDS & OF_WA_LONGTRAIL))
506		return call_prom("setprop", 4, 1, node, ADDR(pname),
507				 (u32)(unsigned long) value, (u32) valuelen);
508
509	/* gah... setprop doesn't work on longtrail, have to use interpret */
510	p = cmd;
511	add_string(&p, "dev");
512	add_string(&p, nodename);
513	add_string(&p, tohex((u32)(unsigned long) value));
514	add_string(&p, tohex(valuelen));
515	add_string(&p, tohex(ADDR(pname)));
516	add_string(&p, tohex(strlen(pname)));
517	add_string(&p, "property");
518	*p = 0;
519	return call_prom("interpret", 1, 1, (u32)(unsigned long) cmd);
520}
521
522/* We can't use the standard versions because of relocation headaches. */
523#define isxdigit(c)	(('0' <= (c) && (c) <= '9') \
524			 || ('a' <= (c) && (c) <= 'f') \
525			 || ('A' <= (c) && (c) <= 'F'))
526
527#define isdigit(c)	('0' <= (c) && (c) <= '9')
528#define islower(c)	('a' <= (c) && (c) <= 'z')
529#define toupper(c)	(islower(c) ? ((c) - 'a' + 'A') : (c))
530
531static unsigned long prom_strtoul(const char *cp, const char **endp)
532{
533	unsigned long result = 0, base = 10, value;
534
535	if (*cp == '0') {
536		base = 8;
537		cp++;
538		if (toupper(*cp) == 'X') {
539			cp++;
540			base = 16;
541		}
542	}
543
544	while (isxdigit(*cp) &&
545	       (value = isdigit(*cp) ? *cp - '0' : toupper(*cp) - 'A' + 10) < base) {
546		result = result * base + value;
547		cp++;
548	}
549
550	if (endp)
551		*endp = cp;
552
553	return result;
554}
555
556static unsigned long prom_memparse(const char *ptr, const char **retptr)
557{
558	unsigned long ret = prom_strtoul(ptr, retptr);
559	int shift = 0;
560
561	/*
562	 * We can't use a switch here because GCC *may* generate a
563	 * jump table which won't work, because we're not running at
564	 * the address we're linked at.
565	 */
566	if ('G' == **retptr || 'g' == **retptr)
567		shift = 30;
568
569	if ('M' == **retptr || 'm' == **retptr)
570		shift = 20;
571
572	if ('K' == **retptr || 'k' == **retptr)
573		shift = 10;
574
575	if (shift) {
576		ret <<= shift;
577		(*retptr)++;
578	}
579
580	return ret;
581}
582
583/*
584 * Early parsing of the command line passed to the kernel, used for
585 * "mem=x" and the options that affect the iommu
586 */
587static void __init early_cmdline_parse(void)
588{
589	const char *opt;
590
591	char *p;
592	int l = 0;
593
594	prom_cmd_line[0] = 0;
595	p = prom_cmd_line;
596	if ((long)prom.chosen > 0)
597		l = prom_getprop(prom.chosen, "bootargs", p, COMMAND_LINE_SIZE-1);
598#ifdef CONFIG_CMDLINE
599	if (l <= 0 || p[0] == '\0') /* dbl check */
600		strlcpy(prom_cmd_line,
601			CONFIG_CMDLINE, sizeof(prom_cmd_line));
602#endif /* CONFIG_CMDLINE */
603	prom_printf("command line: %s\n", prom_cmd_line);
604
605#ifdef CONFIG_PPC64
606	opt = strstr(prom_cmd_line, "iommu=");
607	if (opt) {
608		prom_printf("iommu opt is: %s\n", opt);
609		opt += 6;
610		while (*opt && *opt == ' ')
611			opt++;
612		if (!strncmp(opt, "off", 3))
613			prom_iommu_off = 1;
614		else if (!strncmp(opt, "force", 5))
615			prom_iommu_force_on = 1;
616	}
617#endif
618	opt = strstr(prom_cmd_line, "mem=");
619	if (opt) {
620		opt += 4;
621		prom_memory_limit = prom_memparse(opt, (const char **)&opt);
622#ifdef CONFIG_PPC64
623		/* Align to 16 MB == size of ppc64 large page */
624		prom_memory_limit = ALIGN(prom_memory_limit, 0x1000000);
625#endif
626	}
627}
628
629#if defined(CONFIG_PPC_PSERIES) || defined(CONFIG_PPC_POWERNV)
630/*
631 * The architecture vector has an array of PVR mask/value pairs,
632 * followed by # option vectors - 1, followed by the option vectors.
633 *
634 * See prom.h for the definition of the bits specified in the
635 * architecture vector.
636 *
637 * Because the description vector contains a mix of byte and word
638 * values, we declare it as an unsigned char array, and use this
639 * macro to put word values in.
640 */
641#define W(x)	((x) >> 24) & 0xff, ((x) >> 16) & 0xff, \
642		((x) >> 8) & 0xff, (x) & 0xff
643
644/* Firmware expects the value to be n - 1, where n is the # of vectors */
645#define NUM_VECTORS(n)		((n) - 1)
646
647/*
648 * Firmware expects 1 + n - 2, where n is the length of the option vector in
649 * bytes. The 1 accounts for the length byte itself, the - 2 .. ?
650 */
651#define VECTOR_LENGTH(n)	(1 + (n) - 2)
652
653unsigned char ibm_architecture_vec[] = {
654	W(0xfffe0000), W(0x003a0000),	/* POWER5/POWER5+ */
655	W(0xffff0000), W(0x003e0000),	/* POWER6 */
656	W(0xffff0000), W(0x003f0000),	/* POWER7 */
657	W(0xffff0000), W(0x004b0000),	/* POWER8E */
658	W(0xffff0000), W(0x004c0000),   /* POWER8NVL */
659	W(0xffff0000), W(0x004d0000),	/* POWER8 */
660	W(0xffffffff), W(0x0f000004),	/* all 2.07-compliant */
661	W(0xffffffff), W(0x0f000003),	/* all 2.06-compliant */
662	W(0xffffffff), W(0x0f000002),	/* all 2.05-compliant */
663	W(0xfffffffe), W(0x0f000001),	/* all 2.04-compliant and earlier */
664	NUM_VECTORS(6),			/* 6 option vectors */
665
666	/* option vector 1: processor architectures supported */
667	VECTOR_LENGTH(2),		/* length */
668	0,				/* don't ignore, don't halt */
669	OV1_PPC_2_00 | OV1_PPC_2_01 | OV1_PPC_2_02 | OV1_PPC_2_03 |
670	OV1_PPC_2_04 | OV1_PPC_2_05 | OV1_PPC_2_06 | OV1_PPC_2_07,
671
672	/* option vector 2: Open Firmware options supported */
673	VECTOR_LENGTH(33),		/* length */
674	OV2_REAL_MODE,
675	0, 0,
676	W(0xffffffff),			/* real_base */
677	W(0xffffffff),			/* real_size */
678	W(0xffffffff),			/* virt_base */
679	W(0xffffffff),			/* virt_size */
680	W(0xffffffff),			/* load_base */
681	W(256),				/* 256MB min RMA */
682	W(0xffffffff),			/* full client load */
683	0,				/* min RMA percentage of total RAM */
684	48,				/* max log_2(hash table size) */
685
686	/* option vector 3: processor options supported */
687	VECTOR_LENGTH(2),		/* length */
688	0,				/* don't ignore, don't halt */
689	OV3_FP | OV3_VMX | OV3_DFP,
690
691	/* option vector 4: IBM PAPR implementation */
692	VECTOR_LENGTH(2),		/* length */
693	0,				/* don't halt */
694	OV4_MIN_ENT_CAP,		/* minimum VP entitled capacity */
695
696	/* option vector 5: PAPR/OF options */
697	VECTOR_LENGTH(18),		/* length */
698	0,				/* don't ignore, don't halt */
699	OV5_FEAT(OV5_LPAR) | OV5_FEAT(OV5_SPLPAR) | OV5_FEAT(OV5_LARGE_PAGES) |
700	OV5_FEAT(OV5_DRCONF_MEMORY) | OV5_FEAT(OV5_DONATE_DEDICATE_CPU) |
701#ifdef CONFIG_PCI_MSI
702	/* PCIe/MSI support.  Without MSI full PCIe is not supported */
703	OV5_FEAT(OV5_MSI),
704#else
705	0,
706#endif
707	0,
708#ifdef CONFIG_PPC_SMLPAR
709	OV5_FEAT(OV5_CMO) | OV5_FEAT(OV5_XCMO),
710#else
711	0,
712#endif
713	OV5_FEAT(OV5_TYPE1_AFFINITY) | OV5_FEAT(OV5_PRRN),
714	0,
715	0,
716	0,
717	/* WARNING: The offset of the "number of cores" field below
718	 * must match by the macro below. Update the definition if
719	 * the structure layout changes.
720	 */
721#define IBM_ARCH_VEC_NRCORES_OFFSET	125
722	W(NR_CPUS),			/* number of cores supported */
723	0,
724	0,
725	0,
726	0,
727	OV5_FEAT(OV5_PFO_HW_RNG) | OV5_FEAT(OV5_PFO_HW_ENCR) |
728	OV5_FEAT(OV5_PFO_HW_842),
729	OV5_FEAT(OV5_SUB_PROCESSORS),
730
731	/* option vector 6: IBM PAPR hints */
732	VECTOR_LENGTH(3),		/* length */
733	0,
734	0,
735	OV6_LINUX,
736};
737
738/* Old method - ELF header with PT_NOTE sections only works on BE */
739#ifdef __BIG_ENDIAN__
740static struct fake_elf {
741	Elf32_Ehdr	elfhdr;
742	Elf32_Phdr	phdr[2];
743	struct chrpnote {
744		u32	namesz;
745		u32	descsz;
746		u32	type;
747		char	name[8];	/* "PowerPC" */
748		struct chrpdesc {
749			u32	real_mode;
750			u32	real_base;
751			u32	real_size;
752			u32	virt_base;
753			u32	virt_size;
754			u32	load_base;
755		} chrpdesc;
756	} chrpnote;
757	struct rpanote {
758		u32	namesz;
759		u32	descsz;
760		u32	type;
761		char	name[24];	/* "IBM,RPA-Client-Config" */
762		struct rpadesc {
763			u32	lpar_affinity;
764			u32	min_rmo_size;
765			u32	min_rmo_percent;
766			u32	max_pft_size;
767			u32	splpar;
768			u32	min_load;
769			u32	new_mem_def;
770			u32	ignore_me;
771		} rpadesc;
772	} rpanote;
773} fake_elf = {
774	.elfhdr = {
775		.e_ident = { 0x7f, 'E', 'L', 'F',
776			     ELFCLASS32, ELFDATA2MSB, EV_CURRENT },
777		.e_type = ET_EXEC,	/* yeah right */
778		.e_machine = EM_PPC,
779		.e_version = EV_CURRENT,
780		.e_phoff = offsetof(struct fake_elf, phdr),
781		.e_phentsize = sizeof(Elf32_Phdr),
782		.e_phnum = 2
783	},
784	.phdr = {
785		[0] = {
786			.p_type = PT_NOTE,
787			.p_offset = offsetof(struct fake_elf, chrpnote),
788			.p_filesz = sizeof(struct chrpnote)
789		}, [1] = {
790			.p_type = PT_NOTE,
791			.p_offset = offsetof(struct fake_elf, rpanote),
792			.p_filesz = sizeof(struct rpanote)
793		}
794	},
795	.chrpnote = {
796		.namesz = sizeof("PowerPC"),
797		.descsz = sizeof(struct chrpdesc),
798		.type = 0x1275,
799		.name = "PowerPC",
800		.chrpdesc = {
801			.real_mode = ~0U,	/* ~0 means "don't care" */
802			.real_base = ~0U,
803			.real_size = ~0U,
804			.virt_base = ~0U,
805			.virt_size = ~0U,
806			.load_base = ~0U
807		},
808	},
809	.rpanote = {
810		.namesz = sizeof("IBM,RPA-Client-Config"),
811		.descsz = sizeof(struct rpadesc),
812		.type = 0x12759999,
813		.name = "IBM,RPA-Client-Config",
814		.rpadesc = {
815			.lpar_affinity = 0,
816			.min_rmo_size = 64,	/* in megabytes */
817			.min_rmo_percent = 0,
818			.max_pft_size = 48,	/* 2^48 bytes max PFT size */
819			.splpar = 1,
820			.min_load = ~0U,
821			.new_mem_def = 0
822		}
823	}
824};
825#endif /* __BIG_ENDIAN__ */
826
827static int __init prom_count_smt_threads(void)
828{
829	phandle node;
830	char type[64];
831	unsigned int plen;
832
833	/* Pick up th first CPU node we can find */
834	for (node = 0; prom_next_node(&node); ) {
835		type[0] = 0;
836		prom_getprop(node, "device_type", type, sizeof(type));
837
838		if (strcmp(type, "cpu"))
839			continue;
840		/*
841		 * There is an entry for each smt thread, each entry being
842		 * 4 bytes long.  All cpus should have the same number of
843		 * smt threads, so return after finding the first.
844		 */
845		plen = prom_getproplen(node, "ibm,ppc-interrupt-server#s");
846		if (plen == PROM_ERROR)
847			break;
848		plen >>= 2;
849		prom_debug("Found %lu smt threads per core\n", (unsigned long)plen);
850
851		/* Sanity check */
852		if (plen < 1 || plen > 64) {
853			prom_printf("Threads per core %lu out of bounds, assuming 1\n",
854				    (unsigned long)plen);
855			return 1;
856		}
857		return plen;
858	}
859	prom_debug("No threads found, assuming 1 per core\n");
860
861	return 1;
862
863}
864
865
866static void __init prom_send_capabilities(void)
867{
868	ihandle root;
869	prom_arg_t ret;
870	u32 cores;
871	unsigned char *ptcores;
872
873	root = call_prom("open", 1, 1, ADDR("/"));
874	if (root != 0) {
875		/* We need to tell the FW about the number of cores we support.
876		 *
877		 * To do that, we count the number of threads on the first core
878		 * (we assume this is the same for all cores) and use it to
879		 * divide NR_CPUS.
880		 */
881
882		/* The core value may start at an odd address. If such a word
883		 * access is made at a cache line boundary, this leads to an
884		 * exception which may not be handled at this time.
885		 * Forcing a per byte access to avoid exception.
886		 */
887		ptcores = &ibm_architecture_vec[IBM_ARCH_VEC_NRCORES_OFFSET];
888		cores = 0;
889		cores |= ptcores[0] << 24;
890		cores |= ptcores[1] << 16;
891		cores |= ptcores[2] << 8;
892		cores |= ptcores[3];
893		if (cores != NR_CPUS) {
894			prom_printf("WARNING ! "
895				    "ibm_architecture_vec structure inconsistent: %lu!\n",
896				    cores);
897		} else {
898			cores = DIV_ROUND_UP(NR_CPUS, prom_count_smt_threads());
899			prom_printf("Max number of cores passed to firmware: %lu (NR_CPUS = %lu)\n",
900				    cores, NR_CPUS);
901			ptcores[0] = (cores >> 24) & 0xff;
902			ptcores[1] = (cores >> 16) & 0xff;
903			ptcores[2] = (cores >> 8) & 0xff;
904			ptcores[3] = cores & 0xff;
905		}
906
907		/* try calling the ibm,client-architecture-support method */
908		prom_printf("Calling ibm,client-architecture-support...");
909		if (call_prom_ret("call-method", 3, 2, &ret,
910				  ADDR("ibm,client-architecture-support"),
911				  root,
912				  ADDR(ibm_architecture_vec)) == 0) {
913			/* the call exists... */
914			if (ret)
915				prom_printf("\nWARNING: ibm,client-architecture"
916					    "-support call FAILED!\n");
917			call_prom("close", 1, 0, root);
918			prom_printf(" done\n");
919			return;
920		}
921		call_prom("close", 1, 0, root);
922		prom_printf(" not implemented\n");
923	}
924
925#ifdef __BIG_ENDIAN__
926	{
927		ihandle elfloader;
928
929		/* no ibm,client-architecture-support call, try the old way */
930		elfloader = call_prom("open", 1, 1,
931				      ADDR("/packages/elf-loader"));
932		if (elfloader == 0) {
933			prom_printf("couldn't open /packages/elf-loader\n");
934			return;
935		}
936		call_prom("call-method", 3, 1, ADDR("process-elf-header"),
937			  elfloader, ADDR(&fake_elf));
938		call_prom("close", 1, 0, elfloader);
939	}
940#endif /* __BIG_ENDIAN__ */
941}
942#endif /* #if defined(CONFIG_PPC_PSERIES) || defined(CONFIG_PPC_POWERNV) */
943
944/*
945 * Memory allocation strategy... our layout is normally:
946 *
947 *  at 14Mb or more we have vmlinux, then a gap and initrd.  In some
948 *  rare cases, initrd might end up being before the kernel though.
949 *  We assume this won't override the final kernel at 0, we have no
950 *  provision to handle that in this version, but it should hopefully
951 *  never happen.
952 *
953 *  alloc_top is set to the top of RMO, eventually shrink down if the
954 *  TCEs overlap
955 *
956 *  alloc_bottom is set to the top of kernel/initrd
957 *
958 *  from there, allocations are done this way : rtas is allocated
959 *  topmost, and the device-tree is allocated from the bottom. We try
960 *  to grow the device-tree allocation as we progress. If we can't,
961 *  then we fail, we don't currently have a facility to restart
962 *  elsewhere, but that shouldn't be necessary.
963 *
964 *  Note that calls to reserve_mem have to be done explicitly, memory
965 *  allocated with either alloc_up or alloc_down isn't automatically
966 *  reserved.
967 */
968
969
970/*
971 * Allocates memory in the RMO upward from the kernel/initrd
972 *
973 * When align is 0, this is a special case, it means to allocate in place
974 * at the current location of alloc_bottom or fail (that is basically
975 * extending the previous allocation). Used for the device-tree flattening
976 */
977static unsigned long __init alloc_up(unsigned long size, unsigned long align)
978{
979	unsigned long base = alloc_bottom;
980	unsigned long addr = 0;
981
982	if (align)
983		base = _ALIGN_UP(base, align);
984	prom_debug("alloc_up(%x, %x)\n", size, align);
985	if (ram_top == 0)
986		prom_panic("alloc_up() called with mem not initialized\n");
987
988	if (align)
989		base = _ALIGN_UP(alloc_bottom, align);
990	else
991		base = alloc_bottom;
992
993	for(; (base + size) <= alloc_top;
994	    base = _ALIGN_UP(base + 0x100000, align)) {
995		prom_debug("    trying: 0x%x\n\r", base);
996		addr = (unsigned long)prom_claim(base, size, 0);
997		if (addr != PROM_ERROR && addr != 0)
998			break;
999		addr = 0;
1000		if (align == 0)
1001			break;
1002	}
1003	if (addr == 0)
1004		return 0;
1005	alloc_bottom = addr + size;
1006
1007	prom_debug(" -> %x\n", addr);
1008	prom_debug("  alloc_bottom : %x\n", alloc_bottom);
1009	prom_debug("  alloc_top    : %x\n", alloc_top);
1010	prom_debug("  alloc_top_hi : %x\n", alloc_top_high);
1011	prom_debug("  rmo_top      : %x\n", rmo_top);
1012	prom_debug("  ram_top      : %x\n", ram_top);
1013
1014	return addr;
1015}
1016
1017/*
1018 * Allocates memory downward, either from top of RMO, or if highmem
1019 * is set, from the top of RAM.  Note that this one doesn't handle
1020 * failures.  It does claim memory if highmem is not set.
1021 */
1022static unsigned long __init alloc_down(unsigned long size, unsigned long align,
1023				       int highmem)
1024{
1025	unsigned long base, addr = 0;
1026
1027	prom_debug("alloc_down(%x, %x, %s)\n", size, align,
1028		   highmem ? "(high)" : "(low)");
1029	if (ram_top == 0)
1030		prom_panic("alloc_down() called with mem not initialized\n");
1031
1032	if (highmem) {
1033		/* Carve out storage for the TCE table. */
1034		addr = _ALIGN_DOWN(alloc_top_high - size, align);
1035		if (addr <= alloc_bottom)
1036			return 0;
1037		/* Will we bump into the RMO ? If yes, check out that we
1038		 * didn't overlap existing allocations there, if we did,
1039		 * we are dead, we must be the first in town !
1040		 */
1041		if (addr < rmo_top) {
1042			/* Good, we are first */
1043			if (alloc_top == rmo_top)
1044				alloc_top = rmo_top = addr;
1045			else
1046				return 0;
1047		}
1048		alloc_top_high = addr;
1049		goto bail;
1050	}
1051
1052	base = _ALIGN_DOWN(alloc_top - size, align);
1053	for (; base > alloc_bottom;
1054	     base = _ALIGN_DOWN(base - 0x100000, align))  {
1055		prom_debug("    trying: 0x%x\n\r", base);
1056		addr = (unsigned long)prom_claim(base, size, 0);
1057		if (addr != PROM_ERROR && addr != 0)
1058			break;
1059		addr = 0;
1060	}
1061	if (addr == 0)
1062		return 0;
1063	alloc_top = addr;
1064
1065 bail:
1066	prom_debug(" -> %x\n", addr);
1067	prom_debug("  alloc_bottom : %x\n", alloc_bottom);
1068	prom_debug("  alloc_top    : %x\n", alloc_top);
1069	prom_debug("  alloc_top_hi : %x\n", alloc_top_high);
1070	prom_debug("  rmo_top      : %x\n", rmo_top);
1071	prom_debug("  ram_top      : %x\n", ram_top);
1072
1073	return addr;
1074}
1075
1076/*
1077 * Parse a "reg" cell
1078 */
1079static unsigned long __init prom_next_cell(int s, cell_t **cellp)
1080{
1081	cell_t *p = *cellp;
1082	unsigned long r = 0;
1083
1084	/* Ignore more than 2 cells */
1085	while (s > sizeof(unsigned long) / 4) {
1086		p++;
1087		s--;
1088	}
1089	r = be32_to_cpu(*p++);
1090#ifdef CONFIG_PPC64
1091	if (s > 1) {
1092		r <<= 32;
1093		r |= be32_to_cpu(*(p++));
1094	}
1095#endif
1096	*cellp = p;
1097	return r;
1098}
1099
1100/*
1101 * Very dumb function for adding to the memory reserve list, but
1102 * we don't need anything smarter at this point
1103 *
1104 * XXX Eventually check for collisions.  They should NEVER happen.
1105 * If problems seem to show up, it would be a good start to track
1106 * them down.
1107 */
1108static void __init reserve_mem(u64 base, u64 size)
1109{
1110	u64 top = base + size;
1111	unsigned long cnt = mem_reserve_cnt;
1112
1113	if (size == 0)
1114		return;
1115
1116	/* We need to always keep one empty entry so that we
1117	 * have our terminator with "size" set to 0 since we are
1118	 * dumb and just copy this entire array to the boot params
1119	 */
1120	base = _ALIGN_DOWN(base, PAGE_SIZE);
1121	top = _ALIGN_UP(top, PAGE_SIZE);
1122	size = top - base;
1123
1124	if (cnt >= (MEM_RESERVE_MAP_SIZE - 1))
1125		prom_panic("Memory reserve map exhausted !\n");
1126	mem_reserve_map[cnt].base = cpu_to_be64(base);
1127	mem_reserve_map[cnt].size = cpu_to_be64(size);
1128	mem_reserve_cnt = cnt + 1;
1129}
1130
1131/*
1132 * Initialize memory allocation mechanism, parse "memory" nodes and
1133 * obtain that way the top of memory and RMO to setup out local allocator
1134 */
1135static void __init prom_init_mem(void)
1136{
1137	phandle node;
1138	char *path, type[64];
1139	unsigned int plen;
1140	cell_t *p, *endp;
1141	__be32 val;
1142	u32 rac, rsc;
1143
1144	/*
1145	 * We iterate the memory nodes to find
1146	 * 1) top of RMO (first node)
1147	 * 2) top of memory
1148	 */
1149	val = cpu_to_be32(2);
1150	prom_getprop(prom.root, "#address-cells", &val, sizeof(val));
1151	rac = be32_to_cpu(val);
1152	val = cpu_to_be32(1);
1153	prom_getprop(prom.root, "#size-cells", &val, sizeof(rsc));
1154	rsc = be32_to_cpu(val);
1155	prom_debug("root_addr_cells: %x\n", rac);
1156	prom_debug("root_size_cells: %x\n", rsc);
1157
1158	prom_debug("scanning memory:\n");
1159	path = prom_scratch;
1160
1161	for (node = 0; prom_next_node(&node); ) {
1162		type[0] = 0;
1163		prom_getprop(node, "device_type", type, sizeof(type));
1164
1165		if (type[0] == 0) {
1166			/*
1167			 * CHRP Longtrail machines have no device_type
1168			 * on the memory node, so check the name instead...
1169			 */
1170			prom_getprop(node, "name", type, sizeof(type));
1171		}
1172		if (strcmp(type, "memory"))
1173			continue;
1174
1175		plen = prom_getprop(node, "reg", regbuf, sizeof(regbuf));
1176		if (plen > sizeof(regbuf)) {
1177			prom_printf("memory node too large for buffer !\n");
1178			plen = sizeof(regbuf);
1179		}
1180		p = regbuf;
1181		endp = p + (plen / sizeof(cell_t));
1182
1183#ifdef DEBUG_PROM
1184		memset(path, 0, PROM_SCRATCH_SIZE);
1185		call_prom("package-to-path", 3, 1, node, path, PROM_SCRATCH_SIZE-1);
1186		prom_debug("  node %s :\n", path);
1187#endif /* DEBUG_PROM */
1188
1189		while ((endp - p) >= (rac + rsc)) {
1190			unsigned long base, size;
1191
1192			base = prom_next_cell(rac, &p);
1193			size = prom_next_cell(rsc, &p);
1194
1195			if (size == 0)
1196				continue;
1197			prom_debug("    %x %x\n", base, size);
1198			if (base == 0 && (of_platform & PLATFORM_LPAR))
1199				rmo_top = size;
1200			if ((base + size) > ram_top)
1201				ram_top = base + size;
1202		}
1203	}
1204
1205	alloc_bottom = PAGE_ALIGN((unsigned long)&_end + 0x4000);
1206
1207	/*
1208	 * If prom_memory_limit is set we reduce the upper limits *except* for
1209	 * alloc_top_high. This must be the real top of RAM so we can put
1210	 * TCE's up there.
1211	 */
1212
1213	alloc_top_high = ram_top;
1214
1215	if (prom_memory_limit) {
1216		if (prom_memory_limit <= alloc_bottom) {
1217			prom_printf("Ignoring mem=%x <= alloc_bottom.\n",
1218				prom_memory_limit);
1219			prom_memory_limit = 0;
1220		} else if (prom_memory_limit >= ram_top) {
1221			prom_printf("Ignoring mem=%x >= ram_top.\n",
1222				prom_memory_limit);
1223			prom_memory_limit = 0;
1224		} else {
1225			ram_top = prom_memory_limit;
1226			rmo_top = min(rmo_top, prom_memory_limit);
1227		}
1228	}
1229
1230	/*
1231	 * Setup our top alloc point, that is top of RMO or top of
1232	 * segment 0 when running non-LPAR.
1233	 * Some RS64 machines have buggy firmware where claims up at
1234	 * 1GB fail.  Cap at 768MB as a workaround.
1235	 * Since 768MB is plenty of room, and we need to cap to something
1236	 * reasonable on 32-bit, cap at 768MB on all machines.
1237	 */
1238	if (!rmo_top)
1239		rmo_top = ram_top;
1240	rmo_top = min(0x30000000ul, rmo_top);
1241	alloc_top = rmo_top;
1242	alloc_top_high = ram_top;
1243
1244	/*
1245	 * Check if we have an initrd after the kernel but still inside
1246	 * the RMO.  If we do move our bottom point to after it.
1247	 */
1248	if (prom_initrd_start &&
1249	    prom_initrd_start < rmo_top &&
1250	    prom_initrd_end > alloc_bottom)
1251		alloc_bottom = PAGE_ALIGN(prom_initrd_end);
1252
1253	prom_printf("memory layout at init:\n");
1254	prom_printf("  memory_limit : %x (16 MB aligned)\n", prom_memory_limit);
1255	prom_printf("  alloc_bottom : %x\n", alloc_bottom);
1256	prom_printf("  alloc_top    : %x\n", alloc_top);
1257	prom_printf("  alloc_top_hi : %x\n", alloc_top_high);
1258	prom_printf("  rmo_top      : %x\n", rmo_top);
1259	prom_printf("  ram_top      : %x\n", ram_top);
1260}
1261
1262static void __init prom_close_stdin(void)
1263{
1264	__be32 val;
1265	ihandle stdin;
1266
1267	if (prom_getprop(prom.chosen, "stdin", &val, sizeof(val)) > 0) {
1268		stdin = be32_to_cpu(val);
1269		call_prom("close", 1, 0, stdin);
1270	}
1271}
1272
1273#ifdef CONFIG_PPC_POWERNV
1274
1275#ifdef CONFIG_PPC_EARLY_DEBUG_OPAL
1276static u64 __initdata prom_opal_base;
1277static u64 __initdata prom_opal_entry;
1278#endif
1279
1280/*
1281 * Allocate room for and instantiate OPAL
1282 */
1283static void __init prom_instantiate_opal(void)
1284{
1285	phandle opal_node;
1286	ihandle opal_inst;
1287	u64 base, entry;
1288	u64 size = 0, align = 0x10000;
1289	__be64 val64;
1290	u32 rets[2];
1291
1292	prom_debug("prom_instantiate_opal: start...\n");
1293
1294	opal_node = call_prom("finddevice", 1, 1, ADDR("/ibm,opal"));
1295	prom_debug("opal_node: %x\n", opal_node);
1296	if (!PHANDLE_VALID(opal_node))
1297		return;
1298
1299	val64 = 0;
1300	prom_getprop(opal_node, "opal-runtime-size", &val64, sizeof(val64));
1301	size = be64_to_cpu(val64);
1302	if (size == 0)
1303		return;
1304	val64 = 0;
1305	prom_getprop(opal_node, "opal-runtime-alignment", &val64,sizeof(val64));
1306	align = be64_to_cpu(val64);
1307
1308	base = alloc_down(size, align, 0);
1309	if (base == 0) {
1310		prom_printf("OPAL allocation failed !\n");
1311		return;
1312	}
1313
1314	opal_inst = call_prom("open", 1, 1, ADDR("/ibm,opal"));
1315	if (!IHANDLE_VALID(opal_inst)) {
1316		prom_printf("opening opal package failed (%x)\n", opal_inst);
1317		return;
1318	}
1319
1320	prom_printf("instantiating opal at 0x%x...", base);
1321
1322	if (call_prom_ret("call-method", 4, 3, rets,
1323			  ADDR("load-opal-runtime"),
1324			  opal_inst,
1325			  base >> 32, base & 0xffffffff) != 0
1326	    || (rets[0] == 0 && rets[1] == 0)) {
1327		prom_printf(" failed\n");
1328		return;
1329	}
1330	entry = (((u64)rets[0]) << 32) | rets[1];
1331
1332	prom_printf(" done\n");
1333
1334	reserve_mem(base, size);
1335
1336	prom_debug("opal base     = 0x%x\n", base);
1337	prom_debug("opal align    = 0x%x\n", align);
1338	prom_debug("opal entry    = 0x%x\n", entry);
1339	prom_debug("opal size     = 0x%x\n", (long)size);
1340
1341	prom_setprop(opal_node, "/ibm,opal", "opal-base-address",
1342		     &base, sizeof(base));
1343	prom_setprop(opal_node, "/ibm,opal", "opal-entry-address",
1344		     &entry, sizeof(entry));
1345
1346#ifdef CONFIG_PPC_EARLY_DEBUG_OPAL
1347	prom_opal_base = base;
1348	prom_opal_entry = entry;
1349#endif
1350	prom_debug("prom_instantiate_opal: end...\n");
1351}
1352
1353#endif /* CONFIG_PPC_POWERNV */
1354
1355/*
1356 * Allocate room for and instantiate RTAS
1357 */
1358static void __init prom_instantiate_rtas(void)
1359{
1360	phandle rtas_node;
1361	ihandle rtas_inst;
1362	u32 base, entry = 0;
1363	__be32 val;
1364	u32 size = 0;
1365
1366	prom_debug("prom_instantiate_rtas: start...\n");
1367
1368	rtas_node = call_prom("finddevice", 1, 1, ADDR("/rtas"));
1369	prom_debug("rtas_node: %x\n", rtas_node);
1370	if (!PHANDLE_VALID(rtas_node))
1371		return;
1372
1373	val = 0;
1374	prom_getprop(rtas_node, "rtas-size", &val, sizeof(size));
1375	size = be32_to_cpu(val);
1376	if (size == 0)
1377		return;
1378
1379	base = alloc_down(size, PAGE_SIZE, 0);
1380	if (base == 0)
1381		prom_panic("Could not allocate memory for RTAS\n");
1382
1383	rtas_inst = call_prom("open", 1, 1, ADDR("/rtas"));
1384	if (!IHANDLE_VALID(rtas_inst)) {
1385		prom_printf("opening rtas package failed (%x)\n", rtas_inst);
1386		return;
1387	}
1388
1389	prom_printf("instantiating rtas at 0x%x...", base);
1390
1391	if (call_prom_ret("call-method", 3, 2, &entry,
1392			  ADDR("instantiate-rtas"),
1393			  rtas_inst, base) != 0
1394	    || entry == 0) {
1395		prom_printf(" failed\n");
1396		return;
1397	}
1398	prom_printf(" done\n");
1399
1400	reserve_mem(base, size);
1401
1402	val = cpu_to_be32(base);
1403	prom_setprop(rtas_node, "/rtas", "linux,rtas-base",
1404		     &val, sizeof(val));
1405	val = cpu_to_be32(entry);
1406	prom_setprop(rtas_node, "/rtas", "linux,rtas-entry",
1407		     &val, sizeof(val));
1408
1409	/* Check if it supports "query-cpu-stopped-state" */
1410	if (prom_getprop(rtas_node, "query-cpu-stopped-state",
1411			 &val, sizeof(val)) != PROM_ERROR)
1412		rtas_has_query_cpu_stopped = true;
1413
1414	prom_debug("rtas base     = 0x%x\n", base);
1415	prom_debug("rtas entry    = 0x%x\n", entry);
1416	prom_debug("rtas size     = 0x%x\n", (long)size);
1417
1418	prom_debug("prom_instantiate_rtas: end...\n");
1419}
1420
1421#ifdef CONFIG_PPC64
1422/*
1423 * Allocate room for and instantiate Stored Measurement Log (SML)
1424 */
1425static void __init prom_instantiate_sml(void)
1426{
1427	phandle ibmvtpm_node;
1428	ihandle ibmvtpm_inst;
1429	u32 entry = 0, size = 0, succ = 0;
1430	u64 base;
1431	__be32 val;
1432
1433	prom_debug("prom_instantiate_sml: start...\n");
1434
1435	ibmvtpm_node = call_prom("finddevice", 1, 1, ADDR("/vdevice/vtpm"));
1436	prom_debug("ibmvtpm_node: %x\n", ibmvtpm_node);
1437	if (!PHANDLE_VALID(ibmvtpm_node))
1438		return;
1439
1440	ibmvtpm_inst = call_prom("open", 1, 1, ADDR("/vdevice/vtpm"));
1441	if (!IHANDLE_VALID(ibmvtpm_inst)) {
1442		prom_printf("opening vtpm package failed (%x)\n", ibmvtpm_inst);
1443		return;
1444	}
1445
1446	if (prom_getprop(ibmvtpm_node, "ibm,sml-efi-reformat-supported",
1447			 &val, sizeof(val)) != PROM_ERROR) {
1448		if (call_prom_ret("call-method", 2, 2, &succ,
1449				  ADDR("reformat-sml-to-efi-alignment"),
1450				  ibmvtpm_inst) != 0 || succ == 0) {
1451			prom_printf("Reformat SML to EFI alignment failed\n");
1452			return;
1453		}
1454
1455		if (call_prom_ret("call-method", 2, 2, &size,
1456				  ADDR("sml-get-allocated-size"),
1457				  ibmvtpm_inst) != 0 || size == 0) {
1458			prom_printf("SML get allocated size failed\n");
1459			return;
1460		}
1461	} else {
1462		if (call_prom_ret("call-method", 2, 2, &size,
1463				  ADDR("sml-get-handover-size"),
1464				  ibmvtpm_inst) != 0 || size == 0) {
1465			prom_printf("SML get handover size failed\n");
1466			return;
1467		}
1468	}
1469
1470	base = alloc_down(size, PAGE_SIZE, 0);
1471	if (base == 0)
1472		prom_panic("Could not allocate memory for sml\n");
1473
1474	prom_printf("instantiating sml at 0x%x...", base);
1475
1476	memset((void *)base, 0, size);
1477
1478	if (call_prom_ret("call-method", 4, 2, &entry,
1479			  ADDR("sml-handover"),
1480			  ibmvtpm_inst, size, base) != 0 || entry == 0) {
1481		prom_printf("SML handover failed\n");
1482		return;
1483	}
1484	prom_printf(" done\n");
1485
1486	reserve_mem(base, size);
1487
1488	prom_setprop(ibmvtpm_node, "/vdevice/vtpm", "linux,sml-base",
1489		     &base, sizeof(base));
1490	prom_setprop(ibmvtpm_node, "/vdevice/vtpm", "linux,sml-size",
1491		     &size, sizeof(size));
1492
1493	prom_debug("sml base     = 0x%x\n", base);
1494	prom_debug("sml size     = 0x%x\n", (long)size);
1495
1496	prom_debug("prom_instantiate_sml: end...\n");
1497}
1498
1499/*
1500 * Allocate room for and initialize TCE tables
1501 */
1502#ifdef __BIG_ENDIAN__
1503static void __init prom_initialize_tce_table(void)
1504{
1505	phandle node;
1506	ihandle phb_node;
1507	char compatible[64], type[64], model[64];
1508	char *path = prom_scratch;
1509	u64 base, align;
1510	u32 minalign, minsize;
1511	u64 tce_entry, *tce_entryp;
1512	u64 local_alloc_top, local_alloc_bottom;
1513	u64 i;
1514
1515	if (prom_iommu_off)
1516		return;
1517
1518	prom_debug("starting prom_initialize_tce_table\n");
1519
1520	/* Cache current top of allocs so we reserve a single block */
1521	local_alloc_top = alloc_top_high;
1522	local_alloc_bottom = local_alloc_top;
1523
1524	/* Search all nodes looking for PHBs. */
1525	for (node = 0; prom_next_node(&node); ) {
1526		compatible[0] = 0;
1527		type[0] = 0;
1528		model[0] = 0;
1529		prom_getprop(node, "compatible",
1530			     compatible, sizeof(compatible));
1531		prom_getprop(node, "device_type", type, sizeof(type));
1532		prom_getprop(node, "model", model, sizeof(model));
1533
1534		if ((type[0] == 0) || (strstr(type, "pci") == NULL))
1535			continue;
1536
1537		/* Keep the old logic intact to avoid regression. */
1538		if (compatible[0] != 0) {
1539			if ((strstr(compatible, "python") == NULL) &&
1540			    (strstr(compatible, "Speedwagon") == NULL) &&
1541			    (strstr(compatible, "Winnipeg") == NULL))
1542				continue;
1543		} else if (model[0] != 0) {
1544			if ((strstr(model, "ython") == NULL) &&
1545			    (strstr(model, "peedwagon") == NULL) &&
1546			    (strstr(model, "innipeg") == NULL))
1547				continue;
1548		}
1549
1550		if (prom_getprop(node, "tce-table-minalign", &minalign,
1551				 sizeof(minalign)) == PROM_ERROR)
1552			minalign = 0;
1553		if (prom_getprop(node, "tce-table-minsize", &minsize,
1554				 sizeof(minsize)) == PROM_ERROR)
1555			minsize = 4UL << 20;
1556
1557		/*
1558		 * Even though we read what OF wants, we just set the table
1559		 * size to 4 MB.  This is enough to map 2GB of PCI DMA space.
1560		 * By doing this, we avoid the pitfalls of trying to DMA to
1561		 * MMIO space and the DMA alias hole.
1562		 *
1563		 * On POWER4, firmware sets the TCE region by assuming
1564		 * each TCE table is 8MB. Using this memory for anything
1565		 * else will impact performance, so we always allocate 8MB.
1566		 * Anton
1567		 */
1568		if (pvr_version_is(PVR_POWER4) || pvr_version_is(PVR_POWER4p))
1569			minsize = 8UL << 20;
1570		else
1571			minsize = 4UL << 20;
1572
1573		/* Align to the greater of the align or size */
1574		align = max(minalign, minsize);
1575		base = alloc_down(minsize, align, 1);
1576		if (base == 0)
1577			prom_panic("ERROR, cannot find space for TCE table.\n");
1578		if (base < local_alloc_bottom)
1579			local_alloc_bottom = base;
1580
1581		/* It seems OF doesn't null-terminate the path :-( */
1582		memset(path, 0, PROM_SCRATCH_SIZE);
1583		/* Call OF to setup the TCE hardware */
1584		if (call_prom("package-to-path", 3, 1, node,
1585			      path, PROM_SCRATCH_SIZE-1) == PROM_ERROR) {
1586			prom_printf("package-to-path failed\n");
1587		}
1588
1589		/* Save away the TCE table attributes for later use. */
1590		prom_setprop(node, path, "linux,tce-base", &base, sizeof(base));
1591		prom_setprop(node, path, "linux,tce-size", &minsize, sizeof(minsize));
1592
1593		prom_debug("TCE table: %s\n", path);
1594		prom_debug("\tnode = 0x%x\n", node);
1595		prom_debug("\tbase = 0x%x\n", base);
1596		prom_debug("\tsize = 0x%x\n", minsize);
1597
1598		/* Initialize the table to have a one-to-one mapping
1599		 * over the allocated size.
1600		 */
1601		tce_entryp = (u64 *)base;
1602		for (i = 0; i < (minsize >> 3) ;tce_entryp++, i++) {
1603			tce_entry = (i << PAGE_SHIFT);
1604			tce_entry |= 0x3;
1605			*tce_entryp = tce_entry;
1606		}
1607
1608		prom_printf("opening PHB %s", path);
1609		phb_node = call_prom("open", 1, 1, path);
1610		if (phb_node == 0)
1611			prom_printf("... failed\n");
1612		else
1613			prom_printf("... done\n");
1614
1615		call_prom("call-method", 6, 0, ADDR("set-64-bit-addressing"),
1616			  phb_node, -1, minsize,
1617			  (u32) base, (u32) (base >> 32));
1618		call_prom("close", 1, 0, phb_node);
1619	}
1620
1621	reserve_mem(local_alloc_bottom, local_alloc_top - local_alloc_bottom);
1622
1623	/* These are only really needed if there is a memory limit in
1624	 * effect, but we don't know so export them always. */
1625	prom_tce_alloc_start = local_alloc_bottom;
1626	prom_tce_alloc_end = local_alloc_top;
1627
1628	/* Flag the first invalid entry */
1629	prom_debug("ending prom_initialize_tce_table\n");
1630}
1631#endif /* __BIG_ENDIAN__ */
1632#endif /* CONFIG_PPC64 */
1633
1634/*
1635 * With CHRP SMP we need to use the OF to start the other processors.
1636 * We can't wait until smp_boot_cpus (the OF is trashed by then)
1637 * so we have to put the processors into a holding pattern controlled
1638 * by the kernel (not OF) before we destroy the OF.
1639 *
1640 * This uses a chunk of low memory, puts some holding pattern
1641 * code there and sends the other processors off to there until
1642 * smp_boot_cpus tells them to do something.  The holding pattern
1643 * checks that address until its cpu # is there, when it is that
1644 * cpu jumps to __secondary_start().  smp_boot_cpus() takes care
1645 * of setting those values.
1646 *
1647 * We also use physical address 0x4 here to tell when a cpu
1648 * is in its holding pattern code.
1649 *
1650 * -- Cort
1651 */
1652/*
1653 * We want to reference the copy of __secondary_hold_* in the
1654 * 0 - 0x100 address range
1655 */
1656#define LOW_ADDR(x)	(((unsigned long) &(x)) & 0xff)
1657
1658static void __init prom_hold_cpus(void)
1659{
1660	unsigned long i;
1661	phandle node;
1662	char type[64];
1663	unsigned long *spinloop
1664		= (void *) LOW_ADDR(__secondary_hold_spinloop);
1665	unsigned long *acknowledge
1666		= (void *) LOW_ADDR(__secondary_hold_acknowledge);
1667	unsigned long secondary_hold = LOW_ADDR(__secondary_hold);
1668
1669	/*
1670	 * On pseries, if RTAS supports "query-cpu-stopped-state",
1671	 * we skip this stage, the CPUs will be started by the
1672	 * kernel using RTAS.
1673	 */
1674	if ((of_platform == PLATFORM_PSERIES ||
1675	     of_platform == PLATFORM_PSERIES_LPAR) &&
1676	    rtas_has_query_cpu_stopped) {
1677		prom_printf("prom_hold_cpus: skipped\n");
1678		return;
1679	}
1680
1681	prom_debug("prom_hold_cpus: start...\n");
1682	prom_debug("    1) spinloop       = 0x%x\n", (unsigned long)spinloop);
1683	prom_debug("    1) *spinloop      = 0x%x\n", *spinloop);
1684	prom_debug("    1) acknowledge    = 0x%x\n",
1685		   (unsigned long)acknowledge);
1686	prom_debug("    1) *acknowledge   = 0x%x\n", *acknowledge);
1687	prom_debug("    1) secondary_hold = 0x%x\n", secondary_hold);
1688
1689	/* Set the common spinloop variable, so all of the secondary cpus
1690	 * will block when they are awakened from their OF spinloop.
1691	 * This must occur for both SMP and non SMP kernels, since OF will
1692	 * be trashed when we move the kernel.
1693	 */
1694	*spinloop = 0;
1695
1696	/* look for cpus */
1697	for (node = 0; prom_next_node(&node); ) {
1698		unsigned int cpu_no;
1699		__be32 reg;
1700
1701		type[0] = 0;
1702		prom_getprop(node, "device_type", type, sizeof(type));
1703		if (strcmp(type, "cpu") != 0)
1704			continue;
1705
1706		/* Skip non-configured cpus. */
1707		if (prom_getprop(node, "status", type, sizeof(type)) > 0)
1708			if (strcmp(type, "okay") != 0)
1709				continue;
1710
1711		reg = cpu_to_be32(-1); /* make sparse happy */
1712		prom_getprop(node, "reg", &reg, sizeof(reg));
1713		cpu_no = be32_to_cpu(reg);
1714
1715		prom_debug("cpu hw idx   = %lu\n", cpu_no);
1716
1717		/* Init the acknowledge var which will be reset by
1718		 * the secondary cpu when it awakens from its OF
1719		 * spinloop.
1720		 */
1721		*acknowledge = (unsigned long)-1;
1722
1723		if (cpu_no != prom.cpu) {
1724			/* Primary Thread of non-boot cpu or any thread */
1725			prom_printf("starting cpu hw idx %lu... ", cpu_no);
1726			call_prom("start-cpu", 3, 0, node,
1727				  secondary_hold, cpu_no);
1728
1729			for (i = 0; (i < 100000000) &&
1730			     (*acknowledge == ((unsigned long)-1)); i++ )
1731				mb();
1732
1733			if (*acknowledge == cpu_no)
1734				prom_printf("done\n");
1735			else
1736				prom_printf("failed: %x\n", *acknowledge);
1737		}
1738#ifdef CONFIG_SMP
1739		else
1740			prom_printf("boot cpu hw idx %lu\n", cpu_no);
1741#endif /* CONFIG_SMP */
1742	}
1743
1744	prom_debug("prom_hold_cpus: end...\n");
1745}
1746
1747
1748static void __init prom_init_client_services(unsigned long pp)
1749{
1750	/* Get a handle to the prom entry point before anything else */
1751	prom_entry = pp;
1752
1753	/* get a handle for the stdout device */
1754	prom.chosen = call_prom("finddevice", 1, 1, ADDR("/chosen"));
1755	if (!PHANDLE_VALID(prom.chosen))
1756		prom_panic("cannot find chosen"); /* msg won't be printed :( */
1757
1758	/* get device tree root */
1759	prom.root = call_prom("finddevice", 1, 1, ADDR("/"));
1760	if (!PHANDLE_VALID(prom.root))
1761		prom_panic("cannot find device tree root"); /* msg won't be printed :( */
1762
1763	prom.mmumap = 0;
1764}
1765
1766#ifdef CONFIG_PPC32
1767/*
1768 * For really old powermacs, we need to map things we claim.
1769 * For that, we need the ihandle of the mmu.
1770 * Also, on the longtrail, we need to work around other bugs.
1771 */
1772static void __init prom_find_mmu(void)
1773{
1774	phandle oprom;
1775	char version[64];
1776
1777	oprom = call_prom("finddevice", 1, 1, ADDR("/openprom"));
1778	if (!PHANDLE_VALID(oprom))
1779		return;
1780	if (prom_getprop(oprom, "model", version, sizeof(version)) <= 0)
1781		return;
1782	version[sizeof(version) - 1] = 0;
1783	/* XXX might need to add other versions here */
1784	if (strcmp(version, "Open Firmware, 1.0.5") == 0)
1785		of_workarounds = OF_WA_CLAIM;
1786	else if (strncmp(version, "FirmWorks,3.", 12) == 0) {
1787		of_workarounds = OF_WA_CLAIM | OF_WA_LONGTRAIL;
1788		call_prom("interpret", 1, 1, "dev /memory 0 to allow-reclaim");
1789	} else
1790		return;
1791	prom.memory = call_prom("open", 1, 1, ADDR("/memory"));
1792	prom_getprop(prom.chosen, "mmu", &prom.mmumap,
1793		     sizeof(prom.mmumap));
1794	prom.mmumap = be32_to_cpu(prom.mmumap);
1795	if (!IHANDLE_VALID(prom.memory) || !IHANDLE_VALID(prom.mmumap))
1796		of_workarounds &= ~OF_WA_CLAIM;		/* hmmm */
1797}
1798#else
1799#define prom_find_mmu()
1800#endif
1801
1802static void __init prom_init_stdout(void)
1803{
1804	char *path = of_stdout_device;
1805	char type[16];
1806	phandle stdout_node;
1807	__be32 val;
1808
1809	if (prom_getprop(prom.chosen, "stdout", &val, sizeof(val)) <= 0)
1810		prom_panic("cannot find stdout");
1811
1812	prom.stdout = be32_to_cpu(val);
1813
1814	/* Get the full OF pathname of the stdout device */
1815	memset(path, 0, 256);
1816	call_prom("instance-to-path", 3, 1, prom.stdout, path, 255);
1817	prom_printf("OF stdout device is: %s\n", of_stdout_device);
1818	prom_setprop(prom.chosen, "/chosen", "linux,stdout-path",
1819		     path, strlen(path) + 1);
1820
1821	/* instance-to-package fails on PA-Semi */
1822	stdout_node = call_prom("instance-to-package", 1, 1, prom.stdout);
1823	if (stdout_node != PROM_ERROR) {
1824		val = cpu_to_be32(stdout_node);
1825		prom_setprop(prom.chosen, "/chosen", "linux,stdout-package",
1826			     &val, sizeof(val));
1827
1828		/* If it's a display, note it */
1829		memset(type, 0, sizeof(type));
1830		prom_getprop(stdout_node, "device_type", type, sizeof(type));
1831		if (strcmp(type, "display") == 0)
1832			prom_setprop(stdout_node, path, "linux,boot-display", NULL, 0);
1833	}
1834}
1835
1836static int __init prom_find_machine_type(void)
1837{
1838	char compat[256];
1839	int len, i = 0;
1840#ifdef CONFIG_PPC64
1841	phandle rtas;
1842	int x;
1843#endif
1844
1845	/* Look for a PowerMac or a Cell */
1846	len = prom_getprop(prom.root, "compatible",
1847			   compat, sizeof(compat)-1);
1848	if (len > 0) {
1849		compat[len] = 0;
1850		while (i < len) {
1851			char *p = &compat[i];
1852			int sl = strlen(p);
1853			if (sl == 0)
1854				break;
1855			if (strstr(p, "Power Macintosh") ||
1856			    strstr(p, "MacRISC"))
1857				return PLATFORM_POWERMAC;
1858#ifdef CONFIG_PPC64
1859			/* We must make sure we don't detect the IBM Cell
1860			 * blades as pSeries due to some firmware issues,
1861			 * so we do it here.
1862			 */
1863			if (strstr(p, "IBM,CBEA") ||
1864			    strstr(p, "IBM,CPBW-1.0"))
1865				return PLATFORM_GENERIC;
1866#endif /* CONFIG_PPC64 */
1867			i += sl + 1;
1868		}
1869	}
1870#ifdef CONFIG_PPC64
1871	/* Try to detect OPAL */
1872	if (PHANDLE_VALID(call_prom("finddevice", 1, 1, ADDR("/ibm,opal"))))
1873		return PLATFORM_OPAL;
1874
1875	/* Try to figure out if it's an IBM pSeries or any other
1876	 * PAPR compliant platform. We assume it is if :
1877	 *  - /device_type is "chrp" (please, do NOT use that for future
1878	 *    non-IBM designs !
1879	 *  - it has /rtas
1880	 */
1881	len = prom_getprop(prom.root, "device_type",
1882			   compat, sizeof(compat)-1);
1883	if (len <= 0)
1884		return PLATFORM_GENERIC;
1885	if (strcmp(compat, "chrp"))
1886		return PLATFORM_GENERIC;
1887
1888	/* Default to pSeries. We need to know if we are running LPAR */
1889	rtas = call_prom("finddevice", 1, 1, ADDR("/rtas"));
1890	if (!PHANDLE_VALID(rtas))
1891		return PLATFORM_GENERIC;
1892	x = prom_getproplen(rtas, "ibm,hypertas-functions");
1893	if (x != PROM_ERROR) {
1894		prom_debug("Hypertas detected, assuming LPAR !\n");
1895		return PLATFORM_PSERIES_LPAR;
1896	}
1897	return PLATFORM_PSERIES;
1898#else
1899	return PLATFORM_GENERIC;
1900#endif
1901}
1902
1903static int __init prom_set_color(ihandle ih, int i, int r, int g, int b)
1904{
1905	return call_prom("call-method", 6, 1, ADDR("color!"), ih, i, b, g, r);
1906}
1907
1908/*
1909 * If we have a display that we don't know how to drive,
1910 * we will want to try to execute OF's open method for it
1911 * later.  However, OF will probably fall over if we do that
1912 * we've taken over the MMU.
1913 * So we check whether we will need to open the display,
1914 * and if so, open it now.
1915 */
1916static void __init prom_check_displays(void)
1917{
1918	char type[16], *path;
1919	phandle node;
1920	ihandle ih;
1921	int i;
1922
1923	static unsigned char default_colors[] = {
1924		0x00, 0x00, 0x00,
1925		0x00, 0x00, 0xaa,
1926		0x00, 0xaa, 0x00,
1927		0x00, 0xaa, 0xaa,
1928		0xaa, 0x00, 0x00,
1929		0xaa, 0x00, 0xaa,
1930		0xaa, 0xaa, 0x00,
1931		0xaa, 0xaa, 0xaa,
1932		0x55, 0x55, 0x55,
1933		0x55, 0x55, 0xff,
1934		0x55, 0xff, 0x55,
1935		0x55, 0xff, 0xff,
1936		0xff, 0x55, 0x55,
1937		0xff, 0x55, 0xff,
1938		0xff, 0xff, 0x55,
1939		0xff, 0xff, 0xff
1940	};
1941	const unsigned char *clut;
1942
1943	prom_debug("Looking for displays\n");
1944	for (node = 0; prom_next_node(&node); ) {
1945		memset(type, 0, sizeof(type));
1946		prom_getprop(node, "device_type", type, sizeof(type));
1947		if (strcmp(type, "display") != 0)
1948			continue;
1949
1950		/* It seems OF doesn't null-terminate the path :-( */
1951		path = prom_scratch;
1952		memset(path, 0, PROM_SCRATCH_SIZE);
1953
1954		/*
1955		 * leave some room at the end of the path for appending extra
1956		 * arguments
1957		 */
1958		if (call_prom("package-to-path", 3, 1, node, path,
1959			      PROM_SCRATCH_SIZE-10) == PROM_ERROR)
1960			continue;
1961		prom_printf("found display   : %s, opening... ", path);
1962
1963		ih = call_prom("open", 1, 1, path);
1964		if (ih == 0) {
1965			prom_printf("failed\n");
1966			continue;
1967		}
1968
1969		/* Success */
1970		prom_printf("done\n");
1971		prom_setprop(node, path, "linux,opened", NULL, 0);
1972
1973		/* Setup a usable color table when the appropriate
1974		 * method is available. Should update this to set-colors */
1975		clut = default_colors;
1976		for (i = 0; i < 16; i++, clut += 3)
1977			if (prom_set_color(ih, i, clut[0], clut[1],
1978					   clut[2]) != 0)
1979				break;
1980
1981#ifdef CONFIG_LOGO_LINUX_CLUT224
1982		clut = PTRRELOC(logo_linux_clut224.clut);
1983		for (i = 0; i < logo_linux_clut224.clutsize; i++, clut += 3)
1984			if (prom_set_color(ih, i + 32, clut[0], clut[1],
1985					   clut[2]) != 0)
1986				break;
1987#endif /* CONFIG_LOGO_LINUX_CLUT224 */
1988
1989#ifdef CONFIG_PPC_EARLY_DEBUG_BOOTX
1990		if (prom_getprop(node, "linux,boot-display", NULL, 0) !=
1991		    PROM_ERROR) {
1992			u32 width, height, pitch, addr;
1993
1994			prom_printf("Setting btext !\n");
1995			prom_getprop(node, "width", &width, 4);
1996			prom_getprop(node, "height", &height, 4);
1997			prom_getprop(node, "linebytes", &pitch, 4);
1998			prom_getprop(node, "address", &addr, 4);
1999			prom_printf("W=%d H=%d LB=%d addr=0x%x\n",
2000				    width, height, pitch, addr);
2001			btext_setup_display(width, height, 8, pitch, addr);
2002		}
2003#endif /* CONFIG_PPC_EARLY_DEBUG_BOOTX */
2004	}
2005}
2006
2007
2008/* Return (relocated) pointer to this much memory: moves initrd if reqd. */
2009static void __init *make_room(unsigned long *mem_start, unsigned long *mem_end,
2010			      unsigned long needed, unsigned long align)
2011{
2012	void *ret;
2013
2014	*mem_start = _ALIGN(*mem_start, align);
2015	while ((*mem_start + needed) > *mem_end) {
2016		unsigned long room, chunk;
2017
2018		prom_debug("Chunk exhausted, claiming more at %x...\n",
2019			   alloc_bottom);
2020		room = alloc_top - alloc_bottom;
2021		if (room > DEVTREE_CHUNK_SIZE)
2022			room = DEVTREE_CHUNK_SIZE;
2023		if (room < PAGE_SIZE)
2024			prom_panic("No memory for flatten_device_tree "
2025				   "(no room)\n");
2026		chunk = alloc_up(room, 0);
2027		if (chunk == 0)
2028			prom_panic("No memory for flatten_device_tree "
2029				   "(claim failed)\n");
2030		*mem_end = chunk + room;
2031	}
2032
2033	ret = (void *)*mem_start;
2034	*mem_start += needed;
2035
2036	return ret;
2037}
2038
2039#define dt_push_token(token, mem_start, mem_end) do { 			\
2040		void *room = make_room(mem_start, mem_end, 4, 4);	\
2041		*(__be32 *)room = cpu_to_be32(token);			\
2042	} while(0)
2043
2044static unsigned long __init dt_find_string(char *str)
2045{
2046	char *s, *os;
2047
2048	s = os = (char *)dt_string_start;
2049	s += 4;
2050	while (s <  (char *)dt_string_end) {
2051		if (strcmp(s, str) == 0)
2052			return s - os;
2053		s += strlen(s) + 1;
2054	}
2055	return 0;
2056}
2057
2058/*
2059 * The Open Firmware 1275 specification states properties must be 31 bytes or
2060 * less, however not all firmwares obey this. Make it 64 bytes to be safe.
2061 */
2062#define MAX_PROPERTY_NAME 64
2063
2064static void __init scan_dt_build_strings(phandle node,
2065					 unsigned long *mem_start,
2066					 unsigned long *mem_end)
2067{
2068	char *prev_name, *namep, *sstart;
2069	unsigned long soff;
2070	phandle child;
2071
2072	sstart =  (char *)dt_string_start;
2073
2074	/* get and store all property names */
2075	prev_name = "";
2076	for (;;) {
2077		/* 64 is max len of name including nul. */
2078		namep = make_room(mem_start, mem_end, MAX_PROPERTY_NAME, 1);
2079		if (call_prom("nextprop", 3, 1, node, prev_name, namep) != 1) {
2080			/* No more nodes: unwind alloc */
2081			*mem_start = (unsigned long)namep;
2082			break;
2083		}
2084
2085 		/* skip "name" */
2086 		if (strcmp(namep, "name") == 0) {
2087 			*mem_start = (unsigned long)namep;
2088 			prev_name = "name";
2089 			continue;
2090 		}
2091		/* get/create string entry */
2092		soff = dt_find_string(namep);
2093		if (soff != 0) {
2094			*mem_start = (unsigned long)namep;
2095			namep = sstart + soff;
2096		} else {
2097			/* Trim off some if we can */
2098			*mem_start = (unsigned long)namep + strlen(namep) + 1;
2099			dt_string_end = *mem_start;
2100		}
2101		prev_name = namep;
2102	}
2103
2104	/* do all our children */
2105	child = call_prom("child", 1, 1, node);
2106	while (child != 0) {
2107		scan_dt_build_strings(child, mem_start, mem_end);
2108		child = call_prom("peer", 1, 1, child);
2109	}
2110}
2111
2112static void __init scan_dt_build_struct(phandle node, unsigned long *mem_start,
2113					unsigned long *mem_end)
2114{
2115	phandle child;
2116	char *namep, *prev_name, *sstart, *p, *ep, *lp, *path;
2117	unsigned long soff;
2118	unsigned char *valp;
2119	static char pname[MAX_PROPERTY_NAME];
2120	int l, room, has_phandle = 0;
2121
2122	dt_push_token(OF_DT_BEGIN_NODE, mem_start, mem_end);
2123
2124	/* get the node's full name */
2125	namep = (char *)*mem_start;
2126	room = *mem_end - *mem_start;
2127	if (room > 255)
2128		room = 255;
2129	l = call_prom("package-to-path", 3, 1, node, namep, room);
2130	if (l >= 0) {
2131		/* Didn't fit?  Get more room. */
2132		if (l >= room) {
2133			if (l >= *mem_end - *mem_start)
2134				namep = make_room(mem_start, mem_end, l+1, 1);
2135			call_prom("package-to-path", 3, 1, node, namep, l);
2136		}
2137		namep[l] = '\0';
2138
2139		/* Fixup an Apple bug where they have bogus \0 chars in the
2140		 * middle of the path in some properties, and extract
2141		 * the unit name (everything after the last '/').
2142		 */
2143		for (lp = p = namep, ep = namep + l; p < ep; p++) {
2144			if (*p == '/')
2145				lp = namep;
2146			else if (*p != 0)
2147				*lp++ = *p;
2148		}
2149		*lp = 0;
2150		*mem_start = _ALIGN((unsigned long)lp + 1, 4);
2151	}
2152
2153	/* get it again for debugging */
2154	path = prom_scratch;
2155	memset(path, 0, PROM_SCRATCH_SIZE);
2156	call_prom("package-to-path", 3, 1, node, path, PROM_SCRATCH_SIZE-1);
2157
2158	/* get and store all properties */
2159	prev_name = "";
2160	sstart = (char *)dt_string_start;
2161	for (;;) {
2162		if (call_prom("nextprop", 3, 1, node, prev_name,
2163			      pname) != 1)
2164			break;
2165
2166 		/* skip "name" */
2167 		if (strcmp(pname, "name") == 0) {
2168 			prev_name = "name";
2169 			continue;
2170 		}
2171
2172		/* find string offset */
2173		soff = dt_find_string(pname);
2174		if (soff == 0) {
2175			prom_printf("WARNING: Can't find string index for"
2176				    " <%s>, node %s\n", pname, path);
2177			break;
2178		}
2179		prev_name = sstart + soff;
2180
2181		/* get length */
2182		l = call_prom("getproplen", 2, 1, node, pname);
2183
2184		/* sanity checks */
2185		if (l == PROM_ERROR)
2186			continue;
2187
2188		/* push property head */
2189		dt_push_token(OF_DT_PROP, mem_start, mem_end);
2190		dt_push_token(l, mem_start, mem_end);
2191		dt_push_token(soff, mem_start, mem_end);
2192
2193		/* push property content */
2194		valp = make_room(mem_start, mem_end, l, 4);
2195		call_prom("getprop", 4, 1, node, pname, valp, l);
2196		*mem_start = _ALIGN(*mem_start, 4);
2197
2198		if (!strcmp(pname, "phandle"))
2199			has_phandle = 1;
2200	}
2201
2202	/* Add a "linux,phandle" property if no "phandle" property already
2203	 * existed (can happen with OPAL)
2204	 */
2205	if (!has_phandle) {
2206		soff = dt_find_string("linux,phandle");
2207		if (soff == 0)
2208			prom_printf("WARNING: Can't find string index for"
2209				    " <linux-phandle> node %s\n", path);
2210		else {
2211			dt_push_token(OF_DT_PROP, mem_start, mem_end);
2212			dt_push_token(4, mem_start, mem_end);
2213			dt_push_token(soff, mem_start, mem_end);
2214			valp = make_room(mem_start, mem_end, 4, 4);
2215			*(__be32 *)valp = cpu_to_be32(node);
2216		}
2217	}
2218
2219	/* do all our children */
2220	child = call_prom("child", 1, 1, node);
2221	while (child != 0) {
2222		scan_dt_build_struct(child, mem_start, mem_end);
2223		child = call_prom("peer", 1, 1, child);
2224	}
2225
2226	dt_push_token(OF_DT_END_NODE, mem_start, mem_end);
2227}
2228
2229static void __init flatten_device_tree(void)
2230{
2231	phandle root;
2232	unsigned long mem_start, mem_end, room;
2233	struct boot_param_header *hdr;
2234	char *namep;
2235	u64 *rsvmap;
2236
2237	/*
2238	 * Check how much room we have between alloc top & bottom (+/- a
2239	 * few pages), crop to 1MB, as this is our "chunk" size
2240	 */
2241	room = alloc_top - alloc_bottom - 0x4000;
2242	if (room > DEVTREE_CHUNK_SIZE)
2243		room = DEVTREE_CHUNK_SIZE;
2244	prom_debug("starting device tree allocs at %x\n", alloc_bottom);
2245
2246	/* Now try to claim that */
2247	mem_start = (unsigned long)alloc_up(room, PAGE_SIZE);
2248	if (mem_start == 0)
2249		prom_panic("Can't allocate initial device-tree chunk\n");
2250	mem_end = mem_start + room;
2251
2252	/* Get root of tree */
2253	root = call_prom("peer", 1, 1, (phandle)0);
2254	if (root == (phandle)0)
2255		prom_panic ("couldn't get device tree root\n");
2256
2257	/* Build header and make room for mem rsv map */
2258	mem_start = _ALIGN(mem_start, 4);
2259	hdr = make_room(&mem_start, &mem_end,
2260			sizeof(struct boot_param_header), 4);
2261	dt_header_start = (unsigned long)hdr;
2262	rsvmap = make_room(&mem_start, &mem_end, sizeof(mem_reserve_map), 8);
2263
2264	/* Start of strings */
2265	mem_start = PAGE_ALIGN(mem_start);
2266	dt_string_start = mem_start;
2267	mem_start += 4; /* hole */
2268
2269	/* Add "linux,phandle" in there, we'll need it */
2270	namep = make_room(&mem_start, &mem_end, 16, 1);
2271	strcpy(namep, "linux,phandle");
2272	mem_start = (unsigned long)namep + strlen(namep) + 1;
2273
2274	/* Build string array */
2275	prom_printf("Building dt strings...\n");
2276	scan_dt_build_strings(root, &mem_start, &mem_end);
2277	dt_string_end = mem_start;
2278
2279	/* Build structure */
2280	mem_start = PAGE_ALIGN(mem_start);
2281	dt_struct_start = mem_start;
2282	prom_printf("Building dt structure...\n");
2283	scan_dt_build_struct(root, &mem_start, &mem_end);
2284	dt_push_token(OF_DT_END, &mem_start, &mem_end);
2285	dt_struct_end = PAGE_ALIGN(mem_start);
2286
2287	/* Finish header */
2288	hdr->boot_cpuid_phys = cpu_to_be32(prom.cpu);
2289	hdr->magic = cpu_to_be32(OF_DT_HEADER);
2290	hdr->totalsize = cpu_to_be32(dt_struct_end - dt_header_start);
2291	hdr->off_dt_struct = cpu_to_be32(dt_struct_start - dt_header_start);
2292	hdr->off_dt_strings = cpu_to_be32(dt_string_start - dt_header_start);
2293	hdr->dt_strings_size = cpu_to_be32(dt_string_end - dt_string_start);
2294	hdr->off_mem_rsvmap = cpu_to_be32(((unsigned long)rsvmap) - dt_header_start);
2295	hdr->version = cpu_to_be32(OF_DT_VERSION);
2296	/* Version 16 is not backward compatible */
2297	hdr->last_comp_version = cpu_to_be32(0x10);
2298
2299	/* Copy the reserve map in */
2300	memcpy(rsvmap, mem_reserve_map, sizeof(mem_reserve_map));
2301
2302#ifdef DEBUG_PROM
2303	{
2304		int i;
2305		prom_printf("reserved memory map:\n");
2306		for (i = 0; i < mem_reserve_cnt; i++)
2307			prom_printf("  %x - %x\n",
2308				    be64_to_cpu(mem_reserve_map[i].base),
2309				    be64_to_cpu(mem_reserve_map[i].size));
2310	}
2311#endif
2312	/* Bump mem_reserve_cnt to cause further reservations to fail
2313	 * since it's too late.
2314	 */
2315	mem_reserve_cnt = MEM_RESERVE_MAP_SIZE;
2316
2317	prom_printf("Device tree strings 0x%x -> 0x%x\n",
2318		    dt_string_start, dt_string_end);
2319	prom_printf("Device tree struct  0x%x -> 0x%x\n",
2320		    dt_struct_start, dt_struct_end);
2321}
2322
2323#ifdef CONFIG_PPC_MAPLE
2324/* PIBS Version 1.05.0000 04/26/2005 has an incorrect /ht/isa/ranges property.
2325 * The values are bad, and it doesn't even have the right number of cells. */
2326static void __init fixup_device_tree_maple(void)
2327{
2328	phandle isa;
2329	u32 rloc = 0x01002000; /* IO space; PCI device = 4 */
2330	u32 isa_ranges[6];
2331	char *name;
2332
2333	name = "/ht@0/isa@4";
2334	isa = call_prom("finddevice", 1, 1, ADDR(name));
2335	if (!PHANDLE_VALID(isa)) {
2336		name = "/ht@0/isa@6";
2337		isa = call_prom("finddevice", 1, 1, ADDR(name));
2338		rloc = 0x01003000; /* IO space; PCI device = 6 */
2339	}
2340	if (!PHANDLE_VALID(isa))
2341		return;
2342
2343	if (prom_getproplen(isa, "ranges") != 12)
2344		return;
2345	if (prom_getprop(isa, "ranges", isa_ranges, sizeof(isa_ranges))
2346		== PROM_ERROR)
2347		return;
2348
2349	if (isa_ranges[0] != 0x1 ||
2350		isa_ranges[1] != 0xf4000000 ||
2351		isa_ranges[2] != 0x00010000)
2352		return;
2353
2354	prom_printf("Fixing up bogus ISA range on Maple/Apache...\n");
2355
2356	isa_ranges[0] = 0x1;
2357	isa_ranges[1] = 0x0;
2358	isa_ranges[2] = rloc;
2359	isa_ranges[3] = 0x0;
2360	isa_ranges[4] = 0x0;
2361	isa_ranges[5] = 0x00010000;
2362	prom_setprop(isa, name, "ranges",
2363			isa_ranges, sizeof(isa_ranges));
2364}
2365
2366#define CPC925_MC_START		0xf8000000
2367#define CPC925_MC_LENGTH	0x1000000
2368/* The values for memory-controller don't have right number of cells */
2369static void __init fixup_device_tree_maple_memory_controller(void)
2370{
2371	phandle mc;
2372	u32 mc_reg[4];
2373	char *name = "/hostbridge@f8000000";
2374	u32 ac, sc;
2375
2376	mc = call_prom("finddevice", 1, 1, ADDR(name));
2377	if (!PHANDLE_VALID(mc))
2378		return;
2379
2380	if (prom_getproplen(mc, "reg") != 8)
2381		return;
2382
2383	prom_getprop(prom.root, "#address-cells", &ac, sizeof(ac));
2384	prom_getprop(prom.root, "#size-cells", &sc, sizeof(sc));
2385	if ((ac != 2) || (sc != 2))
2386		return;
2387
2388	if (prom_getprop(mc, "reg", mc_reg, sizeof(mc_reg)) == PROM_ERROR)
2389		return;
2390
2391	if (mc_reg[0] != CPC925_MC_START || mc_reg[1] != CPC925_MC_LENGTH)
2392		return;
2393
2394	prom_printf("Fixing up bogus hostbridge on Maple...\n");
2395
2396	mc_reg[0] = 0x0;
2397	mc_reg[1] = CPC925_MC_START;
2398	mc_reg[2] = 0x0;
2399	mc_reg[3] = CPC925_MC_LENGTH;
2400	prom_setprop(mc, name, "reg", mc_reg, sizeof(mc_reg));
2401}
2402#else
2403#define fixup_device_tree_maple()
2404#define fixup_device_tree_maple_memory_controller()
2405#endif
2406
2407#ifdef CONFIG_PPC_CHRP
2408/*
2409 * Pegasos and BriQ lacks the "ranges" property in the isa node
2410 * Pegasos needs decimal IRQ 14/15, not hexadecimal
2411 * Pegasos has the IDE configured in legacy mode, but advertised as native
2412 */
2413static void __init fixup_device_tree_chrp(void)
2414{
2415	phandle ph;
2416	u32 prop[6];
2417	u32 rloc = 0x01006000; /* IO space; PCI device = 12 */
2418	char *name;
2419	int rc;
2420
2421	name = "/pci@80000000/isa@c";
2422	ph = call_prom("finddevice", 1, 1, ADDR(name));
2423	if (!PHANDLE_VALID(ph)) {
2424		name = "/pci@ff500000/isa@6";
2425		ph = call_prom("finddevice", 1, 1, ADDR(name));
2426		rloc = 0x01003000; /* IO space; PCI device = 6 */
2427	}
2428	if (PHANDLE_VALID(ph)) {
2429		rc = prom_getproplen(ph, "ranges");
2430		if (rc == 0 || rc == PROM_ERROR) {
2431			prom_printf("Fixing up missing ISA range on Pegasos...\n");
2432
2433			prop[0] = 0x1;
2434			prop[1] = 0x0;
2435			prop[2] = rloc;
2436			prop[3] = 0x0;
2437			prop[4] = 0x0;
2438			prop[5] = 0x00010000;
2439			prom_setprop(ph, name, "ranges", prop, sizeof(prop));
2440		}
2441	}
2442
2443	name = "/pci@80000000/ide@C,1";
2444	ph = call_prom("finddevice", 1, 1, ADDR(name));
2445	if (PHANDLE_VALID(ph)) {
2446		prom_printf("Fixing up IDE interrupt on Pegasos...\n");
2447		prop[0] = 14;
2448		prop[1] = 0x0;
2449		prom_setprop(ph, name, "interrupts", prop, 2*sizeof(u32));
2450		prom_printf("Fixing up IDE class-code on Pegasos...\n");
2451		rc = prom_getprop(ph, "class-code", prop, sizeof(u32));
2452		if (rc == sizeof(u32)) {
2453			prop[0] &= ~0x5;
2454			prom_setprop(ph, name, "class-code", prop, sizeof(u32));
2455		}
2456	}
2457}
2458#else
2459#define fixup_device_tree_chrp()
2460#endif
2461
2462#if defined(CONFIG_PPC64) && defined(CONFIG_PPC_PMAC)
2463static void __init fixup_device_tree_pmac(void)
2464{
2465	phandle u3, i2c, mpic;
2466	u32 u3_rev;
2467	u32 interrupts[2];
2468	u32 parent;
2469
2470	/* Some G5s have a missing interrupt definition, fix it up here */
2471	u3 = call_prom("finddevice", 1, 1, ADDR("/u3@0,f8000000"));
2472	if (!PHANDLE_VALID(u3))
2473		return;
2474	i2c = call_prom("finddevice", 1, 1, ADDR("/u3@0,f8000000/i2c@f8001000"));
2475	if (!PHANDLE_VALID(i2c))
2476		return;
2477	mpic = call_prom("finddevice", 1, 1, ADDR("/u3@0,f8000000/mpic@f8040000"));
2478	if (!PHANDLE_VALID(mpic))
2479		return;
2480
2481	/* check if proper rev of u3 */
2482	if (prom_getprop(u3, "device-rev", &u3_rev, sizeof(u3_rev))
2483	    == PROM_ERROR)
2484		return;
2485	if (u3_rev < 0x35 || u3_rev > 0x39)
2486		return;
2487	/* does it need fixup ? */
2488	if (prom_getproplen(i2c, "interrupts") > 0)
2489		return;
2490
2491	prom_printf("fixing up bogus interrupts for u3 i2c...\n");
2492
2493	/* interrupt on this revision of u3 is number 0 and level */
2494	interrupts[0] = 0;
2495	interrupts[1] = 1;
2496	prom_setprop(i2c, "/u3@0,f8000000/i2c@f8001000", "interrupts",
2497		     &interrupts, sizeof(interrupts));
2498	parent = (u32)mpic;
2499	prom_setprop(i2c, "/u3@0,f8000000/i2c@f8001000", "interrupt-parent",
2500		     &parent, sizeof(parent));
2501}
2502#else
2503#define fixup_device_tree_pmac()
2504#endif
2505
2506#ifdef CONFIG_PPC_EFIKA
2507/*
2508 * The MPC5200 FEC driver requires an phy-handle property to tell it how
2509 * to talk to the phy.  If the phy-handle property is missing, then this
2510 * function is called to add the appropriate nodes and link it to the
2511 * ethernet node.
2512 */
2513static void __init fixup_device_tree_efika_add_phy(void)
2514{
2515	u32 node;
2516	char prop[64];
2517	int rv;
2518
2519	/* Check if /builtin/ethernet exists - bail if it doesn't */
2520	node = call_prom("finddevice", 1, 1, ADDR("/builtin/ethernet"));
2521	if (!PHANDLE_VALID(node))
2522		return;
2523
2524	/* Check if the phy-handle property exists - bail if it does */
2525	rv = prom_getprop(node, "phy-handle", prop, sizeof(prop));
2526	if (!rv)
2527		return;
2528
2529	/*
2530	 * At this point the ethernet device doesn't have a phy described.
2531	 * Now we need to add the missing phy node and linkage
2532	 */
2533
2534	/* Check for an MDIO bus node - if missing then create one */
2535	node = call_prom("finddevice", 1, 1, ADDR("/builtin/mdio"));
2536	if (!PHANDLE_VALID(node)) {
2537		prom_printf("Adding Ethernet MDIO node\n");
2538		call_prom("interpret", 1, 1,
2539			" s\" /builtin\" find-device"
2540			" new-device"
2541				" 1 encode-int s\" #address-cells\" property"
2542				" 0 encode-int s\" #size-cells\" property"
2543				" s\" mdio\" device-name"
2544				" s\" fsl,mpc5200b-mdio\" encode-string"
2545				" s\" compatible\" property"
2546				" 0xf0003000 0x400 reg"
2547				" 0x2 encode-int"
2548				" 0x5 encode-int encode+"
2549				" 0x3 encode-int encode+"
2550				" s\" interrupts\" property"
2551			" finish-device");
2552	};
2553
2554	/* Check for a PHY device node - if missing then create one and
2555	 * give it's phandle to the ethernet node */
2556	node = call_prom("finddevice", 1, 1,
2557			 ADDR("/builtin/mdio/ethernet-phy"));
2558	if (!PHANDLE_VALID(node)) {
2559		prom_printf("Adding Ethernet PHY node\n");
2560		call_prom("interpret", 1, 1,
2561			" s\" /builtin/mdio\" find-device"
2562			" new-device"
2563				" s\" ethernet-phy\" device-name"
2564				" 0x10 encode-int s\" reg\" property"
2565				" my-self"
2566				" ihandle>phandle"
2567			" finish-device"
2568			" s\" /builtin/ethernet\" find-device"
2569				" encode-int"
2570				" s\" phy-handle\" property"
2571			" device-end");
2572	}
2573}
2574
2575static void __init fixup_device_tree_efika(void)
2576{
2577	int sound_irq[3] = { 2, 2, 0 };
2578	int bcomm_irq[3*16] = { 3,0,0, 3,1,0, 3,2,0, 3,3,0,
2579				3,4,0, 3,5,0, 3,6,0, 3,7,0,
2580				3,8,0, 3,9,0, 3,10,0, 3,11,0,
2581				3,12,0, 3,13,0, 3,14,0, 3,15,0 };
2582	u32 node;
2583	char prop[64];
2584	int rv, len;
2585
2586	/* Check if we're really running on a EFIKA */
2587	node = call_prom("finddevice", 1, 1, ADDR("/"));
2588	if (!PHANDLE_VALID(node))
2589		return;
2590
2591	rv = prom_getprop(node, "model", prop, sizeof(prop));
2592	if (rv == PROM_ERROR)
2593		return;
2594	if (strcmp(prop, "EFIKA5K2"))
2595		return;
2596
2597	prom_printf("Applying EFIKA device tree fixups\n");
2598
2599	/* Claiming to be 'chrp' is death */
2600	node = call_prom("finddevice", 1, 1, ADDR("/"));
2601	rv = prom_getprop(node, "device_type", prop, sizeof(prop));
2602	if (rv != PROM_ERROR && (strcmp(prop, "chrp") == 0))
2603		prom_setprop(node, "/", "device_type", "efika", sizeof("efika"));
2604
2605	/* CODEGEN,description is exposed in /proc/cpuinfo so
2606	   fix that too */
2607	rv = prom_getprop(node, "CODEGEN,description", prop, sizeof(prop));
2608	if (rv != PROM_ERROR && (strstr(prop, "CHRP")))
2609		prom_setprop(node, "/", "CODEGEN,description",
2610			     "Efika 5200B PowerPC System",
2611			     sizeof("Efika 5200B PowerPC System"));
2612
2613	/* Fixup bestcomm interrupts property */
2614	node = call_prom("finddevice", 1, 1, ADDR("/builtin/bestcomm"));
2615	if (PHANDLE_VALID(node)) {
2616		len = prom_getproplen(node, "interrupts");
2617		if (len == 12) {
2618			prom_printf("Fixing bestcomm interrupts property\n");
2619			prom_setprop(node, "/builtin/bestcom", "interrupts",
2620				     bcomm_irq, sizeof(bcomm_irq));
2621		}
2622	}
2623
2624	/* Fixup sound interrupts property */
2625	node = call_prom("finddevice", 1, 1, ADDR("/builtin/sound"));
2626	if (PHANDLE_VALID(node)) {
2627		rv = prom_getprop(node, "interrupts", prop, sizeof(prop));
2628		if (rv == PROM_ERROR) {
2629			prom_printf("Adding sound interrupts property\n");
2630			prom_setprop(node, "/builtin/sound", "interrupts",
2631				     sound_irq, sizeof(sound_irq));
2632		}
2633	}
2634
2635	/* Make sure ethernet phy-handle property exists */
2636	fixup_device_tree_efika_add_phy();
2637}
2638#else
2639#define fixup_device_tree_efika()
2640#endif
2641
2642static void __init fixup_device_tree(void)
2643{
2644	fixup_device_tree_maple();
2645	fixup_device_tree_maple_memory_controller();
2646	fixup_device_tree_chrp();
2647	fixup_device_tree_pmac();
2648	fixup_device_tree_efika();
2649}
2650
2651static void __init prom_find_boot_cpu(void)
2652{
2653	__be32 rval;
2654	ihandle prom_cpu;
2655	phandle cpu_pkg;
2656
2657	rval = 0;
2658	if (prom_getprop(prom.chosen, "cpu", &rval, sizeof(rval)) <= 0)
2659		return;
2660	prom_cpu = be32_to_cpu(rval);
2661
2662	cpu_pkg = call_prom("instance-to-package", 1, 1, prom_cpu);
2663
2664	prom_getprop(cpu_pkg, "reg", &rval, sizeof(rval));
2665	prom.cpu = be32_to_cpu(rval);
2666
2667	prom_debug("Booting CPU hw index = %lu\n", prom.cpu);
2668}
2669
2670static void __init prom_check_initrd(unsigned long r3, unsigned long r4)
2671{
2672#ifdef CONFIG_BLK_DEV_INITRD
2673	if (r3 && r4 && r4 != 0xdeadbeef) {
2674		__be64 val;
2675
2676		prom_initrd_start = is_kernel_addr(r3) ? __pa(r3) : r3;
2677		prom_initrd_end = prom_initrd_start + r4;
2678
2679		val = cpu_to_be64(prom_initrd_start);
2680		prom_setprop(prom.chosen, "/chosen", "linux,initrd-start",
2681			     &val, sizeof(val));
2682		val = cpu_to_be64(prom_initrd_end);
2683		prom_setprop(prom.chosen, "/chosen", "linux,initrd-end",
2684			     &val, sizeof(val));
2685
2686		reserve_mem(prom_initrd_start,
2687			    prom_initrd_end - prom_initrd_start);
2688
2689		prom_debug("initrd_start=0x%x\n", prom_initrd_start);
2690		prom_debug("initrd_end=0x%x\n", prom_initrd_end);
2691	}
2692#endif /* CONFIG_BLK_DEV_INITRD */
2693}
2694
2695#ifdef CONFIG_PPC64
2696#ifdef CONFIG_RELOCATABLE
2697static void reloc_toc(void)
2698{
2699}
2700
2701static void unreloc_toc(void)
2702{
2703}
2704#else
2705static void __reloc_toc(unsigned long offset, unsigned long nr_entries)
2706{
2707	unsigned long i;
2708	unsigned long *toc_entry;
2709
2710	/* Get the start of the TOC by using r2 directly. */
2711	asm volatile("addi %0,2,-0x8000" : "=b" (toc_entry));
2712
2713	for (i = 0; i < nr_entries; i++) {
2714		*toc_entry = *toc_entry + offset;
2715		toc_entry++;
2716	}
2717}
2718
2719static void reloc_toc(void)
2720{
2721	unsigned long offset = reloc_offset();
2722	unsigned long nr_entries =
2723		(__prom_init_toc_end - __prom_init_toc_start) / sizeof(long);
2724
2725	__reloc_toc(offset, nr_entries);
2726
2727	mb();
2728}
2729
2730static void unreloc_toc(void)
2731{
2732	unsigned long offset = reloc_offset();
2733	unsigned long nr_entries =
2734		(__prom_init_toc_end - __prom_init_toc_start) / sizeof(long);
2735
2736	mb();
2737
2738	__reloc_toc(-offset, nr_entries);
2739}
2740#endif
2741#endif
2742
2743/*
2744 * We enter here early on, when the Open Firmware prom is still
2745 * handling exceptions and the MMU hash table for us.
2746 */
2747
2748unsigned long __init prom_init(unsigned long r3, unsigned long r4,
2749			       unsigned long pp,
2750			       unsigned long r6, unsigned long r7,
2751			       unsigned long kbase)
2752{
2753	unsigned long hdr;
2754
2755#ifdef CONFIG_PPC32
2756	unsigned long offset = reloc_offset();
2757	reloc_got2(offset);
2758#else
2759	reloc_toc();
2760#endif
2761
2762	/*
2763	 * First zero the BSS
2764	 */
2765	memset(&__bss_start, 0, __bss_stop - __bss_start);
2766
2767	/*
2768	 * Init interface to Open Firmware, get some node references,
2769	 * like /chosen
2770	 */
2771	prom_init_client_services(pp);
2772
2773	/*
2774	 * See if this OF is old enough that we need to do explicit maps
2775	 * and other workarounds
2776	 */
2777	prom_find_mmu();
2778
2779	/*
2780	 * Init prom stdout device
2781	 */
2782	prom_init_stdout();
2783
2784	prom_printf("Preparing to boot %s", linux_banner);
2785
2786	/*
2787	 * Get default machine type. At this point, we do not differentiate
2788	 * between pSeries SMP and pSeries LPAR
2789	 */
2790	of_platform = prom_find_machine_type();
2791	prom_printf("Detected machine type: %x\n", of_platform);
2792
2793#ifndef CONFIG_NONSTATIC_KERNEL
2794	/* Bail if this is a kdump kernel. */
2795	if (PHYSICAL_START > 0)
2796		prom_panic("Error: You can't boot a kdump kernel from OF!\n");
2797#endif
2798
2799	/*
2800	 * Check for an initrd
2801	 */
2802	prom_check_initrd(r3, r4);
2803
2804#if defined(CONFIG_PPC_PSERIES) || defined(CONFIG_PPC_POWERNV)
2805	/*
2806	 * On pSeries, inform the firmware about our capabilities
2807	 */
2808	if (of_platform == PLATFORM_PSERIES ||
2809	    of_platform == PLATFORM_PSERIES_LPAR)
2810		prom_send_capabilities();
2811#endif
2812
2813	/*
2814	 * Copy the CPU hold code
2815	 */
2816	if (of_platform != PLATFORM_POWERMAC)
2817		copy_and_flush(0, kbase, 0x100, 0);
2818
2819	/*
2820	 * Do early parsing of command line
2821	 */
2822	early_cmdline_parse();
2823
2824	/*
2825	 * Initialize memory management within prom_init
2826	 */
2827	prom_init_mem();
2828
2829	/*
2830	 * Determine which cpu is actually running right _now_
2831	 */
2832	prom_find_boot_cpu();
2833
2834	/*
2835	 * Initialize display devices
2836	 */
2837	prom_check_displays();
2838
2839#if defined(CONFIG_PPC64) && defined(__BIG_ENDIAN__)
2840	/*
2841	 * Initialize IOMMU (TCE tables) on pSeries. Do that before anything else
2842	 * that uses the allocator, we need to make sure we get the top of memory
2843	 * available for us here...
2844	 */
2845	if (of_platform == PLATFORM_PSERIES)
2846		prom_initialize_tce_table();
2847#endif
2848
2849	/*
2850	 * On non-powermacs, try to instantiate RTAS. PowerMacs don't
2851	 * have a usable RTAS implementation.
2852	 */
2853	if (of_platform != PLATFORM_POWERMAC &&
2854	    of_platform != PLATFORM_OPAL)
2855		prom_instantiate_rtas();
2856
2857#ifdef CONFIG_PPC_POWERNV
2858	if (of_platform == PLATFORM_OPAL)
2859		prom_instantiate_opal();
2860#endif /* CONFIG_PPC_POWERNV */
2861
2862#ifdef CONFIG_PPC64
2863	/* instantiate sml */
2864	prom_instantiate_sml();
2865#endif
2866
2867	/*
2868	 * On non-powermacs, put all CPUs in spin-loops.
2869	 *
2870	 * PowerMacs use a different mechanism to spin CPUs
2871	 *
2872	 * (This must be done after instanciating RTAS)
2873	 */
2874	if (of_platform != PLATFORM_POWERMAC &&
2875	    of_platform != PLATFORM_OPAL)
2876		prom_hold_cpus();
2877
2878	/*
2879	 * Fill in some infos for use by the kernel later on
2880	 */
2881	if (prom_memory_limit) {
2882		__be64 val = cpu_to_be64(prom_memory_limit);
2883		prom_setprop(prom.chosen, "/chosen", "linux,memory-limit",
2884			     &val, sizeof(val));
2885	}
2886#ifdef CONFIG_PPC64
2887	if (prom_iommu_off)
2888		prom_setprop(prom.chosen, "/chosen", "linux,iommu-off",
2889			     NULL, 0);
2890
2891	if (prom_iommu_force_on)
2892		prom_setprop(prom.chosen, "/chosen", "linux,iommu-force-on",
2893			     NULL, 0);
2894
2895	if (prom_tce_alloc_start) {
2896		prom_setprop(prom.chosen, "/chosen", "linux,tce-alloc-start",
2897			     &prom_tce_alloc_start,
2898			     sizeof(prom_tce_alloc_start));
2899		prom_setprop(prom.chosen, "/chosen", "linux,tce-alloc-end",
2900			     &prom_tce_alloc_end,
2901			     sizeof(prom_tce_alloc_end));
2902	}
2903#endif
2904
2905	/*
2906	 * Fixup any known bugs in the device-tree
2907	 */
2908	fixup_device_tree();
2909
2910	/*
2911	 * Now finally create the flattened device-tree
2912	 */
2913	prom_printf("copying OF device tree...\n");
2914	flatten_device_tree();
2915
2916	/*
2917	 * in case stdin is USB and still active on IBM machines...
2918	 * Unfortunately quiesce crashes on some powermacs if we have
2919	 * closed stdin already (in particular the powerbook 101). It
2920	 * appears that the OPAL version of OFW doesn't like it either.
2921	 */
2922	if (of_platform != PLATFORM_POWERMAC &&
2923	    of_platform != PLATFORM_OPAL)
2924		prom_close_stdin();
2925
2926	/*
2927	 * Call OF "quiesce" method to shut down pending DMA's from
2928	 * devices etc...
2929	 */
2930	prom_printf("Quiescing Open Firmware ...\n");
2931	call_prom("quiesce", 0, 0);
2932
2933	/*
2934	 * And finally, call the kernel passing it the flattened device
2935	 * tree and NULL as r5, thus triggering the new entry point which
2936	 * is common to us and kexec
2937	 */
2938	hdr = dt_header_start;
2939
2940	/* Don't print anything after quiesce under OPAL, it crashes OFW */
2941	if (of_platform != PLATFORM_OPAL) {
2942		prom_printf("Booting Linux via __start() ...\n");
2943		prom_debug("->dt_header_start=0x%x\n", hdr);
2944	}
2945
2946#ifdef CONFIG_PPC32
2947	reloc_got2(-offset);
2948#else
2949	unreloc_toc();
2950#endif
2951
2952#ifdef CONFIG_PPC_EARLY_DEBUG_OPAL
2953	/* OPAL early debug gets the OPAL base & entry in r8 and r9 */
2954	__start(hdr, kbase, 0, 0, 0,
2955		prom_opal_base, prom_opal_entry);
2956#else
2957	__start(hdr, kbase, 0, 0, 0, 0, 0);
2958#endif
2959
2960	return 0;
2961}
2962