1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * Procedures for interfacing to Open Firmware.
4 *
5 * Paul Mackerras	August 1996.
6 * Copyright (C) 1996-2005 Paul Mackerras.
7 *
8 *  Adapted for 64bit PowerPC by Dave Engebretsen and Peter Bergner.
9 *    {engebret|bergner}@us.ibm.com
10 */
11
12#undef DEBUG_PROM
13
14/* we cannot use FORTIFY as it brings in new symbols */
15#define __NO_FORTIFY
16
17#include <linux/stdarg.h>
18#include <linux/kernel.h>
19#include <linux/string.h>
20#include <linux/init.h>
21#include <linux/threads.h>
22#include <linux/spinlock.h>
23#include <linux/types.h>
24#include <linux/pci.h>
25#include <linux/proc_fs.h>
26#include <linux/delay.h>
27#include <linux/initrd.h>
28#include <linux/bitops.h>
29#include <linux/pgtable.h>
30#include <linux/printk.h>
31#include <linux/of.h>
32#include <linux/of_fdt.h>
33#include <asm/prom.h>
34#include <asm/rtas.h>
35#include <asm/page.h>
36#include <asm/processor.h>
37#include <asm/interrupt.h>
38#include <asm/irq.h>
39#include <asm/io.h>
40#include <asm/smp.h>
41#include <asm/mmu.h>
42#include <asm/iommu.h>
43#include <asm/btext.h>
44#include <asm/sections.h>
45#include <asm/setup.h>
46#include <asm/asm-prototypes.h>
47#include <asm/ultravisor-api.h>
48
49#include <linux/linux_logo.h>
50
51/* All of prom_init bss lives here */
52#define __prombss __section(".bss.prominit")
53
54/*
55 * Eventually bump that one up
56 */
57#define DEVTREE_CHUNK_SIZE	0x100000
58
59/*
60 * This is the size of the local memory reserve map that gets copied
61 * into the boot params passed to the kernel. That size is totally
62 * flexible as the kernel just reads the list until it encounters an
63 * entry with size 0, so it can be changed without breaking binary
64 * compatibility
65 */
66#define MEM_RESERVE_MAP_SIZE	8
67
68/*
69 * prom_init() is called very early on, before the kernel text
70 * and data have been mapped to KERNELBASE.  At this point the code
71 * is running at whatever address it has been loaded at.
72 * On ppc32 we compile with -mrelocatable, which means that references
73 * to extern and static variables get relocated automatically.
74 * ppc64 objects are always relocatable, we just need to relocate the
75 * TOC.
76 *
77 * Because OF may have mapped I/O devices into the area starting at
78 * KERNELBASE, particularly on CHRP machines, we can't safely call
79 * OF once the kernel has been mapped to KERNELBASE.  Therefore all
80 * OF calls must be done within prom_init().
81 *
82 * ADDR is used in calls to call_prom.  The 4th and following
83 * arguments to call_prom should be 32-bit values.
84 * On ppc64, 64 bit values are truncated to 32 bits (and
85 * fortunately don't get interpreted as two arguments).
86 */
87#define ADDR(x)		(u32)(unsigned long)(x)
88
89#ifdef CONFIG_PPC64
90#define OF_WORKAROUNDS	0
91#else
92#define OF_WORKAROUNDS	of_workarounds
93static int of_workarounds __prombss;
94#endif
95
96#define OF_WA_CLAIM	1	/* do phys/virt claim separately, then map */
97#define OF_WA_LONGTRAIL	2	/* work around longtrail bugs */
98
99#ifdef DEBUG_PROM
100#define prom_debug(x...)	prom_printf(x)
101#else
102#define prom_debug(x...)	do { } while (0)
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 __prombss prom;
149
150static unsigned long __prombss prom_entry;
151
152static char __prombss of_stdout_device[256];
153static char __prombss prom_scratch[256];
154
155static unsigned long __prombss dt_header_start;
156static unsigned long __prombss dt_struct_start, dt_struct_end;
157static unsigned long __prombss dt_string_start, dt_string_end;
158
159static unsigned long __prombss prom_initrd_start, prom_initrd_end;
160
161#ifdef CONFIG_PPC64
162static int __prombss prom_iommu_force_on;
163static int __prombss prom_iommu_off;
164static unsigned long __prombss prom_tce_alloc_start;
165static unsigned long __prombss prom_tce_alloc_end;
166#endif
167
168#ifdef CONFIG_PPC_PSERIES
169static bool __prombss prom_radix_disable;
170static bool __prombss prom_radix_gtse_disable;
171static bool __prombss prom_xive_disable;
172#endif
173
174#ifdef CONFIG_PPC_SVM
175static bool __prombss prom_svm_enable;
176#endif
177
178struct platform_support {
179	bool hash_mmu;
180	bool radix_mmu;
181	bool radix_gtse;
182	bool xive;
183};
184
185/* Platforms codes are now obsolete in the kernel. Now only used within this
186 * file and ultimately gone too. Feel free to change them if you need, they
187 * are not shared with anything outside of this file anymore
188 */
189#define PLATFORM_PSERIES	0x0100
190#define PLATFORM_PSERIES_LPAR	0x0101
191#define PLATFORM_LPAR		0x0001
192#define PLATFORM_POWERMAC	0x0400
193#define PLATFORM_GENERIC	0x0500
194
195static int __prombss of_platform;
196
197static char __prombss prom_cmd_line[COMMAND_LINE_SIZE];
198
199static unsigned long __prombss prom_memory_limit;
200
201static unsigned long __prombss alloc_top;
202static unsigned long __prombss alloc_top_high;
203static unsigned long __prombss alloc_bottom;
204static unsigned long __prombss rmo_top;
205static unsigned long __prombss ram_top;
206
207static struct mem_map_entry __prombss mem_reserve_map[MEM_RESERVE_MAP_SIZE];
208static int __prombss mem_reserve_cnt;
209
210static cell_t __prombss regbuf[1024];
211
212static bool  __prombss rtas_has_query_cpu_stopped;
213
214
215/*
216 * Error results ... some OF calls will return "-1" on error, some
217 * will return 0, some will return either. To simplify, here are
218 * macros to use with any ihandle or phandle return value to check if
219 * it is valid
220 */
221
222#define PROM_ERROR		(-1u)
223#define PHANDLE_VALID(p)	((p) != 0 && (p) != PROM_ERROR)
224#define IHANDLE_VALID(i)	((i) != 0 && (i) != PROM_ERROR)
225
226/* Copied from lib/string.c and lib/kstrtox.c */
227
228static int __init prom_strcmp(const char *cs, const char *ct)
229{
230	unsigned char c1, c2;
231
232	while (1) {
233		c1 = *cs++;
234		c2 = *ct++;
235		if (c1 != c2)
236			return c1 < c2 ? -1 : 1;
237		if (!c1)
238			break;
239	}
240	return 0;
241}
242
243static ssize_t __init prom_strscpy_pad(char *dest, const char *src, size_t n)
244{
245	ssize_t rc;
246	size_t i;
247
248	if (n == 0 || n > INT_MAX)
249		return -E2BIG;
250
251	// Copy up to n bytes
252	for (i = 0; i < n && src[i] != '\0'; i++)
253		dest[i] = src[i];
254
255	rc = i;
256
257	// If we copied all n then we have run out of space for the nul
258	if (rc == n) {
259		// Rewind by one character to ensure nul termination
260		i--;
261		rc = -E2BIG;
262	}
263
264	for (; i < n; i++)
265		dest[i] = '\0';
266
267	return rc;
268}
269
270static int __init prom_strncmp(const char *cs, const char *ct, size_t count)
271{
272	unsigned char c1, c2;
273
274	while (count) {
275		c1 = *cs++;
276		c2 = *ct++;
277		if (c1 != c2)
278			return c1 < c2 ? -1 : 1;
279		if (!c1)
280			break;
281		count--;
282	}
283	return 0;
284}
285
286static size_t __init prom_strlen(const char *s)
287{
288	const char *sc;
289
290	for (sc = s; *sc != '\0'; ++sc)
291		/* nothing */;
292	return sc - s;
293}
294
295static int __init prom_memcmp(const void *cs, const void *ct, size_t count)
296{
297	const unsigned char *su1, *su2;
298	int res = 0;
299
300	for (su1 = cs, su2 = ct; 0 < count; ++su1, ++su2, count--)
301		if ((res = *su1 - *su2) != 0)
302			break;
303	return res;
304}
305
306static char __init *prom_strstr(const char *s1, const char *s2)
307{
308	size_t l1, l2;
309
310	l2 = prom_strlen(s2);
311	if (!l2)
312		return (char *)s1;
313	l1 = prom_strlen(s1);
314	while (l1 >= l2) {
315		l1--;
316		if (!prom_memcmp(s1, s2, l2))
317			return (char *)s1;
318		s1++;
319	}
320	return NULL;
321}
322
323static size_t __init prom_strlcat(char *dest, const char *src, size_t count)
324{
325	size_t dsize = prom_strlen(dest);
326	size_t len = prom_strlen(src);
327	size_t res = dsize + len;
328
329	/* This would be a bug */
330	if (dsize >= count)
331		return count;
332
333	dest += dsize;
334	count -= dsize;
335	if (len >= count)
336		len = count-1;
337	memcpy(dest, src, len);
338	dest[len] = 0;
339	return res;
340
341}
342
343#ifdef CONFIG_PPC_PSERIES
344static int __init prom_strtobool(const char *s, bool *res)
345{
346	if (!s)
347		return -EINVAL;
348
349	switch (s[0]) {
350	case 'y':
351	case 'Y':
352	case '1':
353		*res = true;
354		return 0;
355	case 'n':
356	case 'N':
357	case '0':
358		*res = false;
359		return 0;
360	case 'o':
361	case 'O':
362		switch (s[1]) {
363		case 'n':
364		case 'N':
365			*res = true;
366			return 0;
367		case 'f':
368		case 'F':
369			*res = false;
370			return 0;
371		default:
372			break;
373		}
374		break;
375	default:
376		break;
377	}
378
379	return -EINVAL;
380}
381#endif
382
383/* This is the one and *ONLY* place where we actually call open
384 * firmware.
385 */
386
387static int __init call_prom(const char *service, int nargs, int nret, ...)
388{
389	int i;
390	struct prom_args args;
391	va_list list;
392
393	args.service = cpu_to_be32(ADDR(service));
394	args.nargs = cpu_to_be32(nargs);
395	args.nret = cpu_to_be32(nret);
396
397	va_start(list, nret);
398	for (i = 0; i < nargs; i++)
399		args.args[i] = cpu_to_be32(va_arg(list, prom_arg_t));
400	va_end(list);
401
402	for (i = 0; i < nret; i++)
403		args.args[nargs+i] = 0;
404
405	if (enter_prom(&args, prom_entry) < 0)
406		return PROM_ERROR;
407
408	return (nret > 0) ? be32_to_cpu(args.args[nargs]) : 0;
409}
410
411static int __init call_prom_ret(const char *service, int nargs, int nret,
412				prom_arg_t *rets, ...)
413{
414	int i;
415	struct prom_args args;
416	va_list list;
417
418	args.service = cpu_to_be32(ADDR(service));
419	args.nargs = cpu_to_be32(nargs);
420	args.nret = cpu_to_be32(nret);
421
422	va_start(list, rets);
423	for (i = 0; i < nargs; i++)
424		args.args[i] = cpu_to_be32(va_arg(list, prom_arg_t));
425	va_end(list);
426
427	for (i = 0; i < nret; i++)
428		args.args[nargs+i] = 0;
429
430	if (enter_prom(&args, prom_entry) < 0)
431		return PROM_ERROR;
432
433	if (rets != NULL)
434		for (i = 1; i < nret; ++i)
435			rets[i-1] = be32_to_cpu(args.args[nargs+i]);
436
437	return (nret > 0) ? be32_to_cpu(args.args[nargs]) : 0;
438}
439
440
441static void __init prom_print(const char *msg)
442{
443	const char *p, *q;
444
445	if (prom.stdout == 0)
446		return;
447
448	for (p = msg; *p != 0; p = q) {
449		for (q = p; *q != 0 && *q != '\n'; ++q)
450			;
451		if (q > p)
452			call_prom("write", 3, 1, prom.stdout, p, q - p);
453		if (*q == 0)
454			break;
455		++q;
456		call_prom("write", 3, 1, prom.stdout, ADDR("\r\n"), 2);
457	}
458}
459
460
461/*
462 * Both prom_print_hex & prom_print_dec takes an unsigned long as input so that
463 * we do not need __udivdi3 or __umoddi3 on 32bits.
464 */
465static void __init prom_print_hex(unsigned long val)
466{
467	int i, nibbles = sizeof(val)*2;
468	char buf[sizeof(val)*2+1];
469
470	for (i = nibbles-1;  i >= 0;  i--) {
471		buf[i] = (val & 0xf) + '0';
472		if (buf[i] > '9')
473			buf[i] += ('a'-'0'-10);
474		val >>= 4;
475	}
476	buf[nibbles] = '\0';
477	call_prom("write", 3, 1, prom.stdout, buf, nibbles);
478}
479
480/* max number of decimal digits in an unsigned long */
481#define UL_DIGITS 21
482static void __init prom_print_dec(unsigned long val)
483{
484	int i, size;
485	char buf[UL_DIGITS+1];
486
487	for (i = UL_DIGITS-1; i >= 0;  i--) {
488		buf[i] = (val % 10) + '0';
489		val = val/10;
490		if (val == 0)
491			break;
492	}
493	/* shift stuff down */
494	size = UL_DIGITS - i;
495	call_prom("write", 3, 1, prom.stdout, buf+i, size);
496}
497
498__printf(1, 2)
499static void __init prom_printf(const char *format, ...)
500{
501	const char *p, *q, *s;
502	va_list args;
503	unsigned long v;
504	long vs;
505	int n = 0;
506
507	va_start(args, format);
508	for (p = format; *p != 0; p = q) {
509		for (q = p; *q != 0 && *q != '\n' && *q != '%'; ++q)
510			;
511		if (q > p)
512			call_prom("write", 3, 1, prom.stdout, p, q - p);
513		if (*q == 0)
514			break;
515		if (*q == '\n') {
516			++q;
517			call_prom("write", 3, 1, prom.stdout,
518				  ADDR("\r\n"), 2);
519			continue;
520		}
521		++q;
522		if (*q == 0)
523			break;
524		while (*q == 'l') {
525			++q;
526			++n;
527		}
528		switch (*q) {
529		case 's':
530			++q;
531			s = va_arg(args, const char *);
532			prom_print(s);
533			break;
534		case 'x':
535			++q;
536			switch (n) {
537			case 0:
538				v = va_arg(args, unsigned int);
539				break;
540			case 1:
541				v = va_arg(args, unsigned long);
542				break;
543			case 2:
544			default:
545				v = va_arg(args, unsigned long long);
546				break;
547			}
548			prom_print_hex(v);
549			break;
550		case 'u':
551			++q;
552			switch (n) {
553			case 0:
554				v = va_arg(args, unsigned int);
555				break;
556			case 1:
557				v = va_arg(args, unsigned long);
558				break;
559			case 2:
560			default:
561				v = va_arg(args, unsigned long long);
562				break;
563			}
564			prom_print_dec(v);
565			break;
566		case 'd':
567			++q;
568			switch (n) {
569			case 0:
570				vs = va_arg(args, int);
571				break;
572			case 1:
573				vs = va_arg(args, long);
574				break;
575			case 2:
576			default:
577				vs = va_arg(args, long long);
578				break;
579			}
580			if (vs < 0) {
581				prom_print("-");
582				vs = -vs;
583			}
584			prom_print_dec(vs);
585			break;
586		}
587	}
588	va_end(args);
589}
590
591
592static unsigned int __init prom_claim(unsigned long virt, unsigned long size,
593				unsigned long align)
594{
595
596	if (align == 0 && (OF_WORKAROUNDS & OF_WA_CLAIM)) {
597		/*
598		 * Old OF requires we claim physical and virtual separately
599		 * and then map explicitly (assuming virtual mode)
600		 */
601		int ret;
602		prom_arg_t result;
603
604		ret = call_prom_ret("call-method", 5, 2, &result,
605				    ADDR("claim"), prom.memory,
606				    align, size, virt);
607		if (ret != 0 || result == -1)
608			return -1;
609		ret = call_prom_ret("call-method", 5, 2, &result,
610				    ADDR("claim"), prom.mmumap,
611				    align, size, virt);
612		if (ret != 0) {
613			call_prom("call-method", 4, 1, ADDR("release"),
614				  prom.memory, size, virt);
615			return -1;
616		}
617		/* the 0x12 is M (coherence) + PP == read/write */
618		call_prom("call-method", 6, 1,
619			  ADDR("map"), prom.mmumap, 0x12, size, virt, virt);
620		return virt;
621	}
622	return call_prom("claim", 3, 1, (prom_arg_t)virt, (prom_arg_t)size,
623			 (prom_arg_t)align);
624}
625
626static void __init __attribute__((noreturn)) prom_panic(const char *reason)
627{
628	prom_print(reason);
629	/* Do not call exit because it clears the screen on pmac
630	 * it also causes some sort of double-fault on early pmacs */
631	if (of_platform == PLATFORM_POWERMAC)
632		asm("trap\n");
633
634	/* ToDo: should put up an SRC here on pSeries */
635	call_prom("exit", 0, 0);
636
637	for (;;)			/* should never get here */
638		;
639}
640
641
642static int __init prom_next_node(phandle *nodep)
643{
644	phandle node;
645
646	if ((node = *nodep) != 0
647	    && (*nodep = call_prom("child", 1, 1, node)) != 0)
648		return 1;
649	if ((*nodep = call_prom("peer", 1, 1, node)) != 0)
650		return 1;
651	for (;;) {
652		if ((node = call_prom("parent", 1, 1, node)) == 0)
653			return 0;
654		if ((*nodep = call_prom("peer", 1, 1, node)) != 0)
655			return 1;
656	}
657}
658
659static inline int __init prom_getprop(phandle node, const char *pname,
660				      void *value, size_t valuelen)
661{
662	return call_prom("getprop", 4, 1, node, ADDR(pname),
663			 (u32)(unsigned long) value, (u32) valuelen);
664}
665
666static inline int __init prom_getproplen(phandle node, const char *pname)
667{
668	return call_prom("getproplen", 2, 1, node, ADDR(pname));
669}
670
671static void __init add_string(char **str, const char *q)
672{
673	char *p = *str;
674
675	while (*q)
676		*p++ = *q++;
677	*p++ = ' ';
678	*str = p;
679}
680
681static char *__init tohex(unsigned int x)
682{
683	static const char digits[] __initconst = "0123456789abcdef";
684	static char result[9] __prombss;
685	int i;
686
687	result[8] = 0;
688	i = 8;
689	do {
690		--i;
691		result[i] = digits[x & 0xf];
692		x >>= 4;
693	} while (x != 0 && i > 0);
694	return &result[i];
695}
696
697static int __init prom_setprop(phandle node, const char *nodename,
698			       const char *pname, void *value, size_t valuelen)
699{
700	char cmd[256], *p;
701
702	if (!(OF_WORKAROUNDS & OF_WA_LONGTRAIL))
703		return call_prom("setprop", 4, 1, node, ADDR(pname),
704				 (u32)(unsigned long) value, (u32) valuelen);
705
706	/* gah... setprop doesn't work on longtrail, have to use interpret */
707	p = cmd;
708	add_string(&p, "dev");
709	add_string(&p, nodename);
710	add_string(&p, tohex((u32)(unsigned long) value));
711	add_string(&p, tohex(valuelen));
712	add_string(&p, tohex(ADDR(pname)));
713	add_string(&p, tohex(prom_strlen(pname)));
714	add_string(&p, "property");
715	*p = 0;
716	return call_prom("interpret", 1, 1, (u32)(unsigned long) cmd);
717}
718
719/* We can't use the standard versions because of relocation headaches. */
720#define prom_isxdigit(c) \
721	(('0' <= (c) && (c) <= '9') || ('a' <= (c) && (c) <= 'f') || ('A' <= (c) && (c) <= 'F'))
722
723#define prom_isdigit(c)	('0' <= (c) && (c) <= '9')
724#define prom_islower(c)	('a' <= (c) && (c) <= 'z')
725#define prom_toupper(c)	(prom_islower(c) ? ((c) - 'a' + 'A') : (c))
726
727static unsigned long __init prom_strtoul(const char *cp, const char **endp)
728{
729	unsigned long result = 0, base = 10, value;
730
731	if (*cp == '0') {
732		base = 8;
733		cp++;
734		if (prom_toupper(*cp) == 'X') {
735			cp++;
736			base = 16;
737		}
738	}
739
740	while (prom_isxdigit(*cp) &&
741	       (value = prom_isdigit(*cp) ? *cp - '0' : prom_toupper(*cp) - 'A' + 10) < base) {
742		result = result * base + value;
743		cp++;
744	}
745
746	if (endp)
747		*endp = cp;
748
749	return result;
750}
751
752static unsigned long __init prom_memparse(const char *ptr, const char **retptr)
753{
754	unsigned long ret = prom_strtoul(ptr, retptr);
755	int shift = 0;
756
757	/*
758	 * We can't use a switch here because GCC *may* generate a
759	 * jump table which won't work, because we're not running at
760	 * the address we're linked at.
761	 */
762	if ('G' == **retptr || 'g' == **retptr)
763		shift = 30;
764
765	if ('M' == **retptr || 'm' == **retptr)
766		shift = 20;
767
768	if ('K' == **retptr || 'k' == **retptr)
769		shift = 10;
770
771	if (shift) {
772		ret <<= shift;
773		(*retptr)++;
774	}
775
776	return ret;
777}
778
779/*
780 * Early parsing of the command line passed to the kernel, used for
781 * "mem=x" and the options that affect the iommu
782 */
783static void __init early_cmdline_parse(void)
784{
785	const char *opt;
786
787	char *p;
788	int l = 0;
789
790	prom_cmd_line[0] = 0;
791	p = prom_cmd_line;
792
793	if (!IS_ENABLED(CONFIG_CMDLINE_FORCE) && (long)prom.chosen > 0)
794		l = prom_getprop(prom.chosen, "bootargs", p, COMMAND_LINE_SIZE-1);
795
796	if (IS_ENABLED(CONFIG_CMDLINE_EXTEND) || l <= 0 || p[0] == '\0')
797		prom_strlcat(prom_cmd_line, " " CONFIG_CMDLINE,
798			     sizeof(prom_cmd_line));
799
800	prom_printf("command line: %s\n", prom_cmd_line);
801
802#ifdef CONFIG_PPC64
803	opt = prom_strstr(prom_cmd_line, "iommu=");
804	if (opt) {
805		prom_printf("iommu opt is: %s\n", opt);
806		opt += 6;
807		while (*opt && *opt == ' ')
808			opt++;
809		if (!prom_strncmp(opt, "off", 3))
810			prom_iommu_off = 1;
811		else if (!prom_strncmp(opt, "force", 5))
812			prom_iommu_force_on = 1;
813	}
814#endif
815	opt = prom_strstr(prom_cmd_line, "mem=");
816	if (opt) {
817		opt += 4;
818		prom_memory_limit = prom_memparse(opt, (const char **)&opt);
819#ifdef CONFIG_PPC64
820		/* Align to 16 MB == size of ppc64 large page */
821		prom_memory_limit = ALIGN(prom_memory_limit, 0x1000000);
822#endif
823	}
824
825#ifdef CONFIG_PPC_PSERIES
826	prom_radix_disable = !IS_ENABLED(CONFIG_PPC_RADIX_MMU_DEFAULT);
827	opt = prom_strstr(prom_cmd_line, "disable_radix");
828	if (opt) {
829		opt += 13;
830		if (*opt && *opt == '=') {
831			bool val;
832
833			if (prom_strtobool(++opt, &val))
834				prom_radix_disable = false;
835			else
836				prom_radix_disable = val;
837		} else
838			prom_radix_disable = true;
839	}
840	if (prom_radix_disable)
841		prom_debug("Radix disabled from cmdline\n");
842
843	opt = prom_strstr(prom_cmd_line, "radix_hcall_invalidate=on");
844	if (opt) {
845		prom_radix_gtse_disable = true;
846		prom_debug("Radix GTSE disabled from cmdline\n");
847	}
848
849	opt = prom_strstr(prom_cmd_line, "xive=off");
850	if (opt) {
851		prom_xive_disable = true;
852		prom_debug("XIVE disabled from cmdline\n");
853	}
854#endif /* CONFIG_PPC_PSERIES */
855
856#ifdef CONFIG_PPC_SVM
857	opt = prom_strstr(prom_cmd_line, "svm=");
858	if (opt) {
859		bool val;
860
861		opt += sizeof("svm=") - 1;
862		if (!prom_strtobool(opt, &val))
863			prom_svm_enable = val;
864	}
865#endif /* CONFIG_PPC_SVM */
866}
867
868#ifdef CONFIG_PPC_PSERIES
869/*
870 * The architecture vector has an array of PVR mask/value pairs,
871 * followed by # option vectors - 1, followed by the option vectors.
872 *
873 * See prom.h for the definition of the bits specified in the
874 * architecture vector.
875 */
876
877/* Firmware expects the value to be n - 1, where n is the # of vectors */
878#define NUM_VECTORS(n)		((n) - 1)
879
880/*
881 * Firmware expects 1 + n - 2, where n is the length of the option vector in
882 * bytes. The 1 accounts for the length byte itself, the - 2 .. ?
883 */
884#define VECTOR_LENGTH(n)	(1 + (n) - 2)
885
886struct option_vector1 {
887	u8 byte1;
888	u8 arch_versions;
889	u8 arch_versions3;
890} __packed;
891
892struct option_vector2 {
893	u8 byte1;
894	__be16 reserved;
895	__be32 real_base;
896	__be32 real_size;
897	__be32 virt_base;
898	__be32 virt_size;
899	__be32 load_base;
900	__be32 min_rma;
901	__be32 min_load;
902	u8 min_rma_percent;
903	u8 max_pft_size;
904} __packed;
905
906struct option_vector3 {
907	u8 byte1;
908	u8 byte2;
909} __packed;
910
911struct option_vector4 {
912	u8 byte1;
913	u8 min_vp_cap;
914} __packed;
915
916struct option_vector5 {
917	u8 byte1;
918	u8 byte2;
919	u8 byte3;
920	u8 cmo;
921	u8 associativity;
922	u8 bin_opts;
923	u8 micro_checkpoint;
924	u8 reserved0;
925	__be32 max_cpus;
926	__be16 papr_level;
927	__be16 reserved1;
928	u8 platform_facilities;
929	u8 reserved2;
930	__be16 reserved3;
931	u8 subprocessors;
932	u8 byte22;
933	u8 intarch;
934	u8 mmu;
935	u8 hash_ext;
936	u8 radix_ext;
937} __packed;
938
939struct option_vector6 {
940	u8 reserved;
941	u8 secondary_pteg;
942	u8 os_name;
943} __packed;
944
945struct option_vector7 {
946	u8 os_id[256];
947} __packed;
948
949struct ibm_arch_vec {
950	struct { __be32 mask, val; } pvrs[16];
951
952	u8 num_vectors;
953
954	u8 vec1_len;
955	struct option_vector1 vec1;
956
957	u8 vec2_len;
958	struct option_vector2 vec2;
959
960	u8 vec3_len;
961	struct option_vector3 vec3;
962
963	u8 vec4_len;
964	struct option_vector4 vec4;
965
966	u8 vec5_len;
967	struct option_vector5 vec5;
968
969	u8 vec6_len;
970	struct option_vector6 vec6;
971
972	u8 vec7_len;
973	struct option_vector7 vec7;
974} __packed;
975
976static const struct ibm_arch_vec ibm_architecture_vec_template __initconst = {
977	.pvrs = {
978		{
979			.mask = cpu_to_be32(0xfffe0000), /* POWER5/POWER5+ */
980			.val  = cpu_to_be32(0x003a0000),
981		},
982		{
983			.mask = cpu_to_be32(0xffff0000), /* POWER6 */
984			.val  = cpu_to_be32(0x003e0000),
985		},
986		{
987			.mask = cpu_to_be32(0xffff0000), /* POWER7 */
988			.val  = cpu_to_be32(0x003f0000),
989		},
990		{
991			.mask = cpu_to_be32(0xffff0000), /* POWER8E */
992			.val  = cpu_to_be32(0x004b0000),
993		},
994		{
995			.mask = cpu_to_be32(0xffff0000), /* POWER8NVL */
996			.val  = cpu_to_be32(0x004c0000),
997		},
998		{
999			.mask = cpu_to_be32(0xffff0000), /* POWER8 */
1000			.val  = cpu_to_be32(0x004d0000),
1001		},
1002		{
1003			.mask = cpu_to_be32(0xffff0000), /* POWER9 */
1004			.val  = cpu_to_be32(0x004e0000),
1005		},
1006		{
1007			.mask = cpu_to_be32(0xffff0000), /* POWER10 */
1008			.val  = cpu_to_be32(0x00800000),
1009		},
1010		{
1011			.mask = cpu_to_be32(0xffff0000), /* POWER11 */
1012			.val  = cpu_to_be32(0x00820000),
1013		},
1014		{
1015			.mask = cpu_to_be32(0xffffffff), /* P11 compliant */
1016			.val  = cpu_to_be32(0x0f000007),
1017		},
1018		{
1019			.mask = cpu_to_be32(0xffffffff), /* all 3.1-compliant */
1020			.val  = cpu_to_be32(0x0f000006),
1021		},
1022		{
1023			.mask = cpu_to_be32(0xffffffff), /* all 3.00-compliant */
1024			.val  = cpu_to_be32(0x0f000005),
1025		},
1026		{
1027			.mask = cpu_to_be32(0xffffffff), /* all 2.07-compliant */
1028			.val  = cpu_to_be32(0x0f000004),
1029		},
1030		{
1031			.mask = cpu_to_be32(0xffffffff), /* all 2.06-compliant */
1032			.val  = cpu_to_be32(0x0f000003),
1033		},
1034		{
1035			.mask = cpu_to_be32(0xffffffff), /* all 2.05-compliant */
1036			.val  = cpu_to_be32(0x0f000002),
1037		},
1038		{
1039			.mask = cpu_to_be32(0xfffffffe), /* all 2.04-compliant and earlier */
1040			.val  = cpu_to_be32(0x0f000001),
1041		},
1042	},
1043
1044	.num_vectors = NUM_VECTORS(6),
1045
1046	.vec1_len = VECTOR_LENGTH(sizeof(struct option_vector1)),
1047	.vec1 = {
1048		.byte1 = 0,
1049		.arch_versions = OV1_PPC_2_00 | OV1_PPC_2_01 | OV1_PPC_2_02 | OV1_PPC_2_03 |
1050				 OV1_PPC_2_04 | OV1_PPC_2_05 | OV1_PPC_2_06 | OV1_PPC_2_07,
1051		.arch_versions3 = OV1_PPC_3_00 | OV1_PPC_3_1,
1052	},
1053
1054	.vec2_len = VECTOR_LENGTH(sizeof(struct option_vector2)),
1055	/* option vector 2: Open Firmware options supported */
1056	.vec2 = {
1057		.byte1 = OV2_REAL_MODE,
1058		.reserved = 0,
1059		.real_base = cpu_to_be32(0xffffffff),
1060		.real_size = cpu_to_be32(0xffffffff),
1061		.virt_base = cpu_to_be32(0xffffffff),
1062		.virt_size = cpu_to_be32(0xffffffff),
1063		.load_base = cpu_to_be32(0xffffffff),
1064		.min_rma = cpu_to_be32(512),		/* 512MB min RMA */
1065		.min_load = cpu_to_be32(0xffffffff),	/* full client load */
1066		.min_rma_percent = 0,	/* min RMA percentage of total RAM */
1067		.max_pft_size = 48,	/* max log_2(hash table size) */
1068	},
1069
1070	.vec3_len = VECTOR_LENGTH(sizeof(struct option_vector3)),
1071	/* option vector 3: processor options supported */
1072	.vec3 = {
1073		.byte1 = 0,			/* don't ignore, don't halt */
1074		.byte2 = OV3_FP | OV3_VMX | OV3_DFP,
1075	},
1076
1077	.vec4_len = VECTOR_LENGTH(sizeof(struct option_vector4)),
1078	/* option vector 4: IBM PAPR implementation */
1079	.vec4 = {
1080		.byte1 = 0,			/* don't halt */
1081		.min_vp_cap = OV4_MIN_ENT_CAP,	/* minimum VP entitled capacity */
1082	},
1083
1084	.vec5_len = VECTOR_LENGTH(sizeof(struct option_vector5)),
1085	/* option vector 5: PAPR/OF options */
1086	.vec5 = {
1087		.byte1 = 0,				/* don't ignore, don't halt */
1088		.byte2 = OV5_FEAT(OV5_LPAR) | OV5_FEAT(OV5_SPLPAR) | OV5_FEAT(OV5_LARGE_PAGES) |
1089		OV5_FEAT(OV5_DRCONF_MEMORY) | OV5_FEAT(OV5_DONATE_DEDICATE_CPU) |
1090#ifdef CONFIG_PCI_MSI
1091		/* PCIe/MSI support.  Without MSI full PCIe is not supported */
1092		OV5_FEAT(OV5_MSI),
1093#else
1094		0,
1095#endif
1096		.byte3 = 0,
1097		.cmo =
1098#ifdef CONFIG_PPC_SMLPAR
1099		OV5_FEAT(OV5_CMO) | OV5_FEAT(OV5_XCMO),
1100#else
1101		0,
1102#endif
1103		.associativity = OV5_FEAT(OV5_FORM1_AFFINITY) | OV5_FEAT(OV5_PRRN) |
1104		OV5_FEAT(OV5_FORM2_AFFINITY),
1105		.bin_opts = OV5_FEAT(OV5_RESIZE_HPT) | OV5_FEAT(OV5_HP_EVT),
1106		.micro_checkpoint = 0,
1107		.reserved0 = 0,
1108		.max_cpus = cpu_to_be32(NR_CPUS),	/* number of cores supported */
1109		.papr_level = 0,
1110		.reserved1 = 0,
1111		.platform_facilities = OV5_FEAT(OV5_PFO_HW_RNG) | OV5_FEAT(OV5_PFO_HW_ENCR) | OV5_FEAT(OV5_PFO_HW_842),
1112		.reserved2 = 0,
1113		.reserved3 = 0,
1114		.subprocessors = 1,
1115		.byte22 = OV5_FEAT(OV5_DRMEM_V2) | OV5_FEAT(OV5_DRC_INFO),
1116		.intarch = 0,
1117		.mmu = 0,
1118		.hash_ext = 0,
1119		.radix_ext = 0,
1120	},
1121
1122	/* option vector 6: IBM PAPR hints */
1123	.vec6_len = VECTOR_LENGTH(sizeof(struct option_vector6)),
1124	.vec6 = {
1125		.reserved = 0,
1126		.secondary_pteg = 0,
1127		.os_name = OV6_LINUX,
1128	},
1129
1130	/* option vector 7: OS Identification */
1131	.vec7_len = VECTOR_LENGTH(sizeof(struct option_vector7)),
1132};
1133
1134static struct ibm_arch_vec __prombss ibm_architecture_vec  ____cacheline_aligned;
1135
1136/* Old method - ELF header with PT_NOTE sections only works on BE */
1137#ifdef __BIG_ENDIAN__
1138static const struct fake_elf {
1139	Elf32_Ehdr	elfhdr;
1140	Elf32_Phdr	phdr[2];
1141	struct chrpnote {
1142		u32	namesz;
1143		u32	descsz;
1144		u32	type;
1145		char	name[8];	/* "PowerPC" */
1146		struct chrpdesc {
1147			u32	real_mode;
1148			u32	real_base;
1149			u32	real_size;
1150			u32	virt_base;
1151			u32	virt_size;
1152			u32	load_base;
1153		} chrpdesc;
1154	} chrpnote;
1155	struct rpanote {
1156		u32	namesz;
1157		u32	descsz;
1158		u32	type;
1159		char	name[24];	/* "IBM,RPA-Client-Config" */
1160		struct rpadesc {
1161			u32	lpar_affinity;
1162			u32	min_rmo_size;
1163			u32	min_rmo_percent;
1164			u32	max_pft_size;
1165			u32	splpar;
1166			u32	min_load;
1167			u32	new_mem_def;
1168			u32	ignore_me;
1169		} rpadesc;
1170	} rpanote;
1171} fake_elf __initconst = {
1172	.elfhdr = {
1173		.e_ident = { 0x7f, 'E', 'L', 'F',
1174			     ELFCLASS32, ELFDATA2MSB, EV_CURRENT },
1175		.e_type = ET_EXEC,	/* yeah right */
1176		.e_machine = EM_PPC,
1177		.e_version = EV_CURRENT,
1178		.e_phoff = offsetof(struct fake_elf, phdr),
1179		.e_phentsize = sizeof(Elf32_Phdr),
1180		.e_phnum = 2
1181	},
1182	.phdr = {
1183		[0] = {
1184			.p_type = PT_NOTE,
1185			.p_offset = offsetof(struct fake_elf, chrpnote),
1186			.p_filesz = sizeof(struct chrpnote)
1187		}, [1] = {
1188			.p_type = PT_NOTE,
1189			.p_offset = offsetof(struct fake_elf, rpanote),
1190			.p_filesz = sizeof(struct rpanote)
1191		}
1192	},
1193	.chrpnote = {
1194		.namesz = sizeof("PowerPC"),
1195		.descsz = sizeof(struct chrpdesc),
1196		.type = 0x1275,
1197		.name = "PowerPC",
1198		.chrpdesc = {
1199			.real_mode = ~0U,	/* ~0 means "don't care" */
1200			.real_base = ~0U,
1201			.real_size = ~0U,
1202			.virt_base = ~0U,
1203			.virt_size = ~0U,
1204			.load_base = ~0U
1205		},
1206	},
1207	.rpanote = {
1208		.namesz = sizeof("IBM,RPA-Client-Config"),
1209		.descsz = sizeof(struct rpadesc),
1210		.type = 0x12759999,
1211		.name = "IBM,RPA-Client-Config",
1212		.rpadesc = {
1213			.lpar_affinity = 0,
1214			.min_rmo_size = 64,	/* in megabytes */
1215			.min_rmo_percent = 0,
1216			.max_pft_size = 48,	/* 2^48 bytes max PFT size */
1217			.splpar = 1,
1218			.min_load = ~0U,
1219			.new_mem_def = 0
1220		}
1221	}
1222};
1223#endif /* __BIG_ENDIAN__ */
1224
1225static int __init prom_count_smt_threads(void)
1226{
1227	phandle node;
1228	char type[64];
1229	unsigned int plen;
1230
1231	/* Pick up th first CPU node we can find */
1232	for (node = 0; prom_next_node(&node); ) {
1233		type[0] = 0;
1234		prom_getprop(node, "device_type", type, sizeof(type));
1235
1236		if (prom_strcmp(type, "cpu"))
1237			continue;
1238		/*
1239		 * There is an entry for each smt thread, each entry being
1240		 * 4 bytes long.  All cpus should have the same number of
1241		 * smt threads, so return after finding the first.
1242		 */
1243		plen = prom_getproplen(node, "ibm,ppc-interrupt-server#s");
1244		if (plen == PROM_ERROR)
1245			break;
1246		plen >>= 2;
1247		prom_debug("Found %lu smt threads per core\n", (unsigned long)plen);
1248
1249		/* Sanity check */
1250		if (plen < 1 || plen > 64) {
1251			prom_printf("Threads per core %lu out of bounds, assuming 1\n",
1252				    (unsigned long)plen);
1253			return 1;
1254		}
1255		return plen;
1256	}
1257	prom_debug("No threads found, assuming 1 per core\n");
1258
1259	return 1;
1260
1261}
1262
1263static void __init prom_parse_mmu_model(u8 val,
1264					struct platform_support *support)
1265{
1266	switch (val) {
1267	case OV5_FEAT(OV5_MMU_DYNAMIC):
1268	case OV5_FEAT(OV5_MMU_EITHER): /* Either Available */
1269		prom_debug("MMU - either supported\n");
1270		support->radix_mmu = !prom_radix_disable;
1271		support->hash_mmu = true;
1272		break;
1273	case OV5_FEAT(OV5_MMU_RADIX): /* Only Radix */
1274		prom_debug("MMU - radix only\n");
1275		if (prom_radix_disable) {
1276			/*
1277			 * If we __have__ to do radix, we're better off ignoring
1278			 * the command line rather than not booting.
1279			 */
1280			prom_printf("WARNING: Ignoring cmdline option disable_radix\n");
1281		}
1282		support->radix_mmu = true;
1283		break;
1284	case OV5_FEAT(OV5_MMU_HASH):
1285		prom_debug("MMU - hash only\n");
1286		support->hash_mmu = true;
1287		break;
1288	default:
1289		prom_debug("Unknown mmu support option: 0x%x\n", val);
1290		break;
1291	}
1292}
1293
1294static void __init prom_parse_xive_model(u8 val,
1295					 struct platform_support *support)
1296{
1297	switch (val) {
1298	case OV5_FEAT(OV5_XIVE_EITHER): /* Either Available */
1299		prom_debug("XIVE - either mode supported\n");
1300		support->xive = !prom_xive_disable;
1301		break;
1302	case OV5_FEAT(OV5_XIVE_EXPLOIT): /* Only Exploitation mode */
1303		prom_debug("XIVE - exploitation mode supported\n");
1304		if (prom_xive_disable) {
1305			/*
1306			 * If we __have__ to do XIVE, we're better off ignoring
1307			 * the command line rather than not booting.
1308			 */
1309			prom_printf("WARNING: Ignoring cmdline option xive=off\n");
1310		}
1311		support->xive = true;
1312		break;
1313	case OV5_FEAT(OV5_XIVE_LEGACY): /* Only Legacy mode */
1314		prom_debug("XIVE - legacy mode supported\n");
1315		break;
1316	default:
1317		prom_debug("Unknown xive support option: 0x%x\n", val);
1318		break;
1319	}
1320}
1321
1322static void __init prom_parse_platform_support(u8 index, u8 val,
1323					       struct platform_support *support)
1324{
1325	switch (index) {
1326	case OV5_INDX(OV5_MMU_SUPPORT): /* MMU Model */
1327		prom_parse_mmu_model(val & OV5_FEAT(OV5_MMU_SUPPORT), support);
1328		break;
1329	case OV5_INDX(OV5_RADIX_GTSE): /* Radix Extensions */
1330		if (val & OV5_FEAT(OV5_RADIX_GTSE))
1331			support->radix_gtse = !prom_radix_gtse_disable;
1332		break;
1333	case OV5_INDX(OV5_XIVE_SUPPORT): /* Interrupt mode */
1334		prom_parse_xive_model(val & OV5_FEAT(OV5_XIVE_SUPPORT),
1335				      support);
1336		break;
1337	}
1338}
1339
1340static void __init prom_check_platform_support(void)
1341{
1342	struct platform_support supported = {
1343		.hash_mmu = false,
1344		.radix_mmu = false,
1345		.radix_gtse = false,
1346		.xive = false
1347	};
1348	int prop_len = prom_getproplen(prom.chosen,
1349				       "ibm,arch-vec-5-platform-support");
1350
1351	/*
1352	 * First copy the architecture vec template
1353	 *
1354	 * use memcpy() instead of *vec = *vec_template so that GCC replaces it
1355	 * by __memcpy() when KASAN is active
1356	 */
1357	memcpy(&ibm_architecture_vec, &ibm_architecture_vec_template,
1358	       sizeof(ibm_architecture_vec));
1359
1360	prom_strscpy_pad(ibm_architecture_vec.vec7.os_id, linux_banner, 256);
1361
1362	if (prop_len > 1) {
1363		int i;
1364		u8 vec[8];
1365		prom_debug("Found ibm,arch-vec-5-platform-support, len: %d\n",
1366			   prop_len);
1367		if (prop_len > sizeof(vec))
1368			prom_printf("WARNING: ibm,arch-vec-5-platform-support longer than expected (len: %d)\n",
1369				    prop_len);
1370		prom_getprop(prom.chosen, "ibm,arch-vec-5-platform-support", &vec, sizeof(vec));
1371		for (i = 0; i < prop_len; i += 2) {
1372			prom_debug("%d: index = 0x%x val = 0x%x\n", i / 2, vec[i], vec[i + 1]);
1373			prom_parse_platform_support(vec[i], vec[i + 1], &supported);
1374		}
1375	}
1376
1377	if (supported.radix_mmu && IS_ENABLED(CONFIG_PPC_RADIX_MMU)) {
1378		/* Radix preferred - Check if GTSE is also supported */
1379		prom_debug("Asking for radix\n");
1380		ibm_architecture_vec.vec5.mmu = OV5_FEAT(OV5_MMU_RADIX);
1381		if (supported.radix_gtse)
1382			ibm_architecture_vec.vec5.radix_ext =
1383					OV5_FEAT(OV5_RADIX_GTSE);
1384		else
1385			prom_debug("Radix GTSE isn't supported\n");
1386	} else if (supported.hash_mmu) {
1387		/* Default to hash mmu (if we can) */
1388		prom_debug("Asking for hash\n");
1389		ibm_architecture_vec.vec5.mmu = OV5_FEAT(OV5_MMU_HASH);
1390	} else {
1391		/* We're probably on a legacy hypervisor */
1392		prom_debug("Assuming legacy hash support\n");
1393	}
1394
1395	if (supported.xive) {
1396		prom_debug("Asking for XIVE\n");
1397		ibm_architecture_vec.vec5.intarch = OV5_FEAT(OV5_XIVE_EXPLOIT);
1398	}
1399}
1400
1401static void __init prom_send_capabilities(void)
1402{
1403	ihandle root;
1404	prom_arg_t ret;
1405	u32 cores;
1406
1407	/* Check ibm,arch-vec-5-platform-support and fixup vec5 if required */
1408	prom_check_platform_support();
1409
1410	root = call_prom("open", 1, 1, ADDR("/"));
1411	if (root != 0) {
1412		/* We need to tell the FW about the number of cores we support.
1413		 *
1414		 * To do that, we count the number of threads on the first core
1415		 * (we assume this is the same for all cores) and use it to
1416		 * divide NR_CPUS.
1417		 */
1418
1419		cores = DIV_ROUND_UP(NR_CPUS, prom_count_smt_threads());
1420		prom_printf("Max number of cores passed to firmware: %u (NR_CPUS = %d)\n",
1421			    cores, NR_CPUS);
1422
1423		ibm_architecture_vec.vec5.max_cpus = cpu_to_be32(cores);
1424
1425		/* try calling the ibm,client-architecture-support method */
1426		prom_printf("Calling ibm,client-architecture-support...");
1427		if (call_prom_ret("call-method", 3, 2, &ret,
1428				  ADDR("ibm,client-architecture-support"),
1429				  root,
1430				  ADDR(&ibm_architecture_vec)) == 0) {
1431			/* the call exists... */
1432			if (ret)
1433				prom_printf("\nWARNING: ibm,client-architecture"
1434					    "-support call FAILED!\n");
1435			call_prom("close", 1, 0, root);
1436			prom_printf(" done\n");
1437			return;
1438		}
1439		call_prom("close", 1, 0, root);
1440		prom_printf(" not implemented\n");
1441	}
1442
1443#ifdef __BIG_ENDIAN__
1444	{
1445		ihandle elfloader;
1446
1447		/* no ibm,client-architecture-support call, try the old way */
1448		elfloader = call_prom("open", 1, 1,
1449				      ADDR("/packages/elf-loader"));
1450		if (elfloader == 0) {
1451			prom_printf("couldn't open /packages/elf-loader\n");
1452			return;
1453		}
1454		call_prom("call-method", 3, 1, ADDR("process-elf-header"),
1455			  elfloader, ADDR(&fake_elf));
1456		call_prom("close", 1, 0, elfloader);
1457	}
1458#endif /* __BIG_ENDIAN__ */
1459}
1460#endif /* CONFIG_PPC_PSERIES */
1461
1462/*
1463 * Memory allocation strategy... our layout is normally:
1464 *
1465 *  at 14Mb or more we have vmlinux, then a gap and initrd.  In some
1466 *  rare cases, initrd might end up being before the kernel though.
1467 *  We assume this won't override the final kernel at 0, we have no
1468 *  provision to handle that in this version, but it should hopefully
1469 *  never happen.
1470 *
1471 *  alloc_top is set to the top of RMO, eventually shrink down if the
1472 *  TCEs overlap
1473 *
1474 *  alloc_bottom is set to the top of kernel/initrd
1475 *
1476 *  from there, allocations are done this way : rtas is allocated
1477 *  topmost, and the device-tree is allocated from the bottom. We try
1478 *  to grow the device-tree allocation as we progress. If we can't,
1479 *  then we fail, we don't currently have a facility to restart
1480 *  elsewhere, but that shouldn't be necessary.
1481 *
1482 *  Note that calls to reserve_mem have to be done explicitly, memory
1483 *  allocated with either alloc_up or alloc_down isn't automatically
1484 *  reserved.
1485 */
1486
1487
1488/*
1489 * Allocates memory in the RMO upward from the kernel/initrd
1490 *
1491 * When align is 0, this is a special case, it means to allocate in place
1492 * at the current location of alloc_bottom or fail (that is basically
1493 * extending the previous allocation). Used for the device-tree flattening
1494 */
1495static unsigned long __init alloc_up(unsigned long size, unsigned long align)
1496{
1497	unsigned long base = alloc_bottom;
1498	unsigned long addr = 0;
1499
1500	if (align)
1501		base = ALIGN(base, align);
1502	prom_debug("%s(%lx, %lx)\n", __func__, size, align);
1503	if (ram_top == 0)
1504		prom_panic("alloc_up() called with mem not initialized\n");
1505
1506	if (align)
1507		base = ALIGN(alloc_bottom, align);
1508	else
1509		base = alloc_bottom;
1510
1511	for(; (base + size) <= alloc_top;
1512	    base = ALIGN(base + 0x100000, align)) {
1513		prom_debug("    trying: 0x%lx\n\r", base);
1514		addr = (unsigned long)prom_claim(base, size, 0);
1515		if (addr != PROM_ERROR && addr != 0)
1516			break;
1517		addr = 0;
1518		if (align == 0)
1519			break;
1520	}
1521	if (addr == 0)
1522		return 0;
1523	alloc_bottom = addr + size;
1524
1525	prom_debug(" -> %lx\n", addr);
1526	prom_debug("  alloc_bottom : %lx\n", alloc_bottom);
1527	prom_debug("  alloc_top    : %lx\n", alloc_top);
1528	prom_debug("  alloc_top_hi : %lx\n", alloc_top_high);
1529	prom_debug("  rmo_top      : %lx\n", rmo_top);
1530	prom_debug("  ram_top      : %lx\n", ram_top);
1531
1532	return addr;
1533}
1534
1535/*
1536 * Allocates memory downward, either from top of RMO, or if highmem
1537 * is set, from the top of RAM.  Note that this one doesn't handle
1538 * failures.  It does claim memory if highmem is not set.
1539 */
1540static unsigned long __init alloc_down(unsigned long size, unsigned long align,
1541				       int highmem)
1542{
1543	unsigned long base, addr = 0;
1544
1545	prom_debug("%s(%lx, %lx, %s)\n", __func__, size, align,
1546		   highmem ? "(high)" : "(low)");
1547	if (ram_top == 0)
1548		prom_panic("alloc_down() called with mem not initialized\n");
1549
1550	if (highmem) {
1551		/* Carve out storage for the TCE table. */
1552		addr = ALIGN_DOWN(alloc_top_high - size, align);
1553		if (addr <= alloc_bottom)
1554			return 0;
1555		/* Will we bump into the RMO ? If yes, check out that we
1556		 * didn't overlap existing allocations there, if we did,
1557		 * we are dead, we must be the first in town !
1558		 */
1559		if (addr < rmo_top) {
1560			/* Good, we are first */
1561			if (alloc_top == rmo_top)
1562				alloc_top = rmo_top = addr;
1563			else
1564				return 0;
1565		}
1566		alloc_top_high = addr;
1567		goto bail;
1568	}
1569
1570	base = ALIGN_DOWN(alloc_top - size, align);
1571	for (; base > alloc_bottom;
1572	     base = ALIGN_DOWN(base - 0x100000, align))  {
1573		prom_debug("    trying: 0x%lx\n\r", base);
1574		addr = (unsigned long)prom_claim(base, size, 0);
1575		if (addr != PROM_ERROR && addr != 0)
1576			break;
1577		addr = 0;
1578	}
1579	if (addr == 0)
1580		return 0;
1581	alloc_top = addr;
1582
1583 bail:
1584	prom_debug(" -> %lx\n", addr);
1585	prom_debug("  alloc_bottom : %lx\n", alloc_bottom);
1586	prom_debug("  alloc_top    : %lx\n", alloc_top);
1587	prom_debug("  alloc_top_hi : %lx\n", alloc_top_high);
1588	prom_debug("  rmo_top      : %lx\n", rmo_top);
1589	prom_debug("  ram_top      : %lx\n", ram_top);
1590
1591	return addr;
1592}
1593
1594/*
1595 * Parse a "reg" cell
1596 */
1597static unsigned long __init prom_next_cell(int s, cell_t **cellp)
1598{
1599	cell_t *p = *cellp;
1600	unsigned long r = 0;
1601
1602	/* Ignore more than 2 cells */
1603	while (s > sizeof(unsigned long) / 4) {
1604		p++;
1605		s--;
1606	}
1607	r = be32_to_cpu(*p++);
1608#ifdef CONFIG_PPC64
1609	if (s > 1) {
1610		r <<= 32;
1611		r |= be32_to_cpu(*(p++));
1612	}
1613#endif
1614	*cellp = p;
1615	return r;
1616}
1617
1618/*
1619 * Very dumb function for adding to the memory reserve list, but
1620 * we don't need anything smarter at this point
1621 *
1622 * XXX Eventually check for collisions.  They should NEVER happen.
1623 * If problems seem to show up, it would be a good start to track
1624 * them down.
1625 */
1626static void __init reserve_mem(u64 base, u64 size)
1627{
1628	u64 top = base + size;
1629	unsigned long cnt = mem_reserve_cnt;
1630
1631	if (size == 0)
1632		return;
1633
1634	/* We need to always keep one empty entry so that we
1635	 * have our terminator with "size" set to 0 since we are
1636	 * dumb and just copy this entire array to the boot params
1637	 */
1638	base = ALIGN_DOWN(base, PAGE_SIZE);
1639	top = ALIGN(top, PAGE_SIZE);
1640	size = top - base;
1641
1642	if (cnt >= (MEM_RESERVE_MAP_SIZE - 1))
1643		prom_panic("Memory reserve map exhausted !\n");
1644	mem_reserve_map[cnt].base = cpu_to_be64(base);
1645	mem_reserve_map[cnt].size = cpu_to_be64(size);
1646	mem_reserve_cnt = cnt + 1;
1647}
1648
1649/*
1650 * Initialize memory allocation mechanism, parse "memory" nodes and
1651 * obtain that way the top of memory and RMO to setup out local allocator
1652 */
1653static void __init prom_init_mem(void)
1654{
1655	phandle node;
1656	char type[64];
1657	unsigned int plen;
1658	cell_t *p, *endp;
1659	__be32 val;
1660	u32 rac, rsc;
1661
1662	/*
1663	 * We iterate the memory nodes to find
1664	 * 1) top of RMO (first node)
1665	 * 2) top of memory
1666	 */
1667	val = cpu_to_be32(2);
1668	prom_getprop(prom.root, "#address-cells", &val, sizeof(val));
1669	rac = be32_to_cpu(val);
1670	val = cpu_to_be32(1);
1671	prom_getprop(prom.root, "#size-cells", &val, sizeof(rsc));
1672	rsc = be32_to_cpu(val);
1673	prom_debug("root_addr_cells: %x\n", rac);
1674	prom_debug("root_size_cells: %x\n", rsc);
1675
1676	prom_debug("scanning memory:\n");
1677
1678	for (node = 0; prom_next_node(&node); ) {
1679		type[0] = 0;
1680		prom_getprop(node, "device_type", type, sizeof(type));
1681
1682		if (type[0] == 0) {
1683			/*
1684			 * CHRP Longtrail machines have no device_type
1685			 * on the memory node, so check the name instead...
1686			 */
1687			prom_getprop(node, "name", type, sizeof(type));
1688		}
1689		if (prom_strcmp(type, "memory"))
1690			continue;
1691
1692		plen = prom_getprop(node, "reg", regbuf, sizeof(regbuf));
1693		if (plen > sizeof(regbuf)) {
1694			prom_printf("memory node too large for buffer !\n");
1695			plen = sizeof(regbuf);
1696		}
1697		p = regbuf;
1698		endp = p + (plen / sizeof(cell_t));
1699
1700#ifdef DEBUG_PROM
1701		memset(prom_scratch, 0, sizeof(prom_scratch));
1702		call_prom("package-to-path", 3, 1, node, prom_scratch,
1703			  sizeof(prom_scratch) - 1);
1704		prom_debug("  node %s :\n", prom_scratch);
1705#endif /* DEBUG_PROM */
1706
1707		while ((endp - p) >= (rac + rsc)) {
1708			unsigned long base, size;
1709
1710			base = prom_next_cell(rac, &p);
1711			size = prom_next_cell(rsc, &p);
1712
1713			if (size == 0)
1714				continue;
1715			prom_debug("    %lx %lx\n", base, size);
1716			if (base == 0 && (of_platform & PLATFORM_LPAR))
1717				rmo_top = size;
1718			if ((base + size) > ram_top)
1719				ram_top = base + size;
1720		}
1721	}
1722
1723	alloc_bottom = PAGE_ALIGN((unsigned long)&_end + 0x4000);
1724
1725	/*
1726	 * If prom_memory_limit is set we reduce the upper limits *except* for
1727	 * alloc_top_high. This must be the real top of RAM so we can put
1728	 * TCE's up there.
1729	 */
1730
1731	alloc_top_high = ram_top;
1732
1733	if (prom_memory_limit) {
1734		if (prom_memory_limit <= alloc_bottom) {
1735			prom_printf("Ignoring mem=%lx <= alloc_bottom.\n",
1736				    prom_memory_limit);
1737			prom_memory_limit = 0;
1738		} else if (prom_memory_limit >= ram_top) {
1739			prom_printf("Ignoring mem=%lx >= ram_top.\n",
1740				    prom_memory_limit);
1741			prom_memory_limit = 0;
1742		} else {
1743			ram_top = prom_memory_limit;
1744			rmo_top = min(rmo_top, prom_memory_limit);
1745		}
1746	}
1747
1748	/*
1749	 * Setup our top alloc point, that is top of RMO or top of
1750	 * segment 0 when running non-LPAR.
1751	 * Some RS64 machines have buggy firmware where claims up at
1752	 * 1GB fail.  Cap at 768MB as a workaround.
1753	 * Since 768MB is plenty of room, and we need to cap to something
1754	 * reasonable on 32-bit, cap at 768MB on all machines.
1755	 */
1756	if (!rmo_top)
1757		rmo_top = ram_top;
1758	rmo_top = min(0x30000000ul, rmo_top);
1759	alloc_top = rmo_top;
1760	alloc_top_high = ram_top;
1761
1762	/*
1763	 * Check if we have an initrd after the kernel but still inside
1764	 * the RMO.  If we do move our bottom point to after it.
1765	 */
1766	if (prom_initrd_start &&
1767	    prom_initrd_start < rmo_top &&
1768	    prom_initrd_end > alloc_bottom)
1769		alloc_bottom = PAGE_ALIGN(prom_initrd_end);
1770
1771	prom_printf("memory layout at init:\n");
1772	prom_printf("  memory_limit : %lx (16 MB aligned)\n",
1773		    prom_memory_limit);
1774	prom_printf("  alloc_bottom : %lx\n", alloc_bottom);
1775	prom_printf("  alloc_top    : %lx\n", alloc_top);
1776	prom_printf("  alloc_top_hi : %lx\n", alloc_top_high);
1777	prom_printf("  rmo_top      : %lx\n", rmo_top);
1778	prom_printf("  ram_top      : %lx\n", ram_top);
1779}
1780
1781static void __init prom_close_stdin(void)
1782{
1783	__be32 val;
1784	ihandle stdin;
1785
1786	if (prom_getprop(prom.chosen, "stdin", &val, sizeof(val)) > 0) {
1787		stdin = be32_to_cpu(val);
1788		call_prom("close", 1, 0, stdin);
1789	}
1790}
1791
1792#ifdef CONFIG_PPC_SVM
1793static int __init prom_rtas_hcall(uint64_t args)
1794{
1795	register uint64_t arg1 asm("r3") = H_RTAS;
1796	register uint64_t arg2 asm("r4") = args;
1797
1798	asm volatile("sc 1\n" : "=r" (arg1) :
1799			"r" (arg1),
1800			"r" (arg2) :);
1801	srr_regs_clobbered();
1802
1803	return arg1;
1804}
1805
1806static struct rtas_args __prombss os_term_args;
1807
1808static void __init prom_rtas_os_term(char *str)
1809{
1810	phandle rtas_node;
1811	__be32 val;
1812	u32 token;
1813
1814	prom_debug("%s: start...\n", __func__);
1815	rtas_node = call_prom("finddevice", 1, 1, ADDR("/rtas"));
1816	prom_debug("rtas_node: %x\n", rtas_node);
1817	if (!PHANDLE_VALID(rtas_node))
1818		return;
1819
1820	val = 0;
1821	prom_getprop(rtas_node, "ibm,os-term", &val, sizeof(val));
1822	token = be32_to_cpu(val);
1823	prom_debug("ibm,os-term: %x\n", token);
1824	if (token == 0)
1825		prom_panic("Could not get token for ibm,os-term\n");
1826	os_term_args.token = cpu_to_be32(token);
1827	os_term_args.nargs = cpu_to_be32(1);
1828	os_term_args.nret = cpu_to_be32(1);
1829	os_term_args.args[0] = cpu_to_be32(__pa(str));
1830	prom_rtas_hcall((uint64_t)&os_term_args);
1831}
1832#endif /* CONFIG_PPC_SVM */
1833
1834/*
1835 * Allocate room for and instantiate RTAS
1836 */
1837static void __init prom_instantiate_rtas(void)
1838{
1839	phandle rtas_node;
1840	ihandle rtas_inst;
1841	u32 base, entry = 0;
1842	__be32 val;
1843	u32 size = 0;
1844
1845	prom_debug("prom_instantiate_rtas: start...\n");
1846
1847	rtas_node = call_prom("finddevice", 1, 1, ADDR("/rtas"));
1848	prom_debug("rtas_node: %x\n", rtas_node);
1849	if (!PHANDLE_VALID(rtas_node))
1850		return;
1851
1852	val = 0;
1853	prom_getprop(rtas_node, "rtas-size", &val, sizeof(size));
1854	size = be32_to_cpu(val);
1855	if (size == 0)
1856		return;
1857
1858	base = alloc_down(size, PAGE_SIZE, 0);
1859	if (base == 0)
1860		prom_panic("Could not allocate memory for RTAS\n");
1861
1862	rtas_inst = call_prom("open", 1, 1, ADDR("/rtas"));
1863	if (!IHANDLE_VALID(rtas_inst)) {
1864		prom_printf("opening rtas package failed (%x)\n", rtas_inst);
1865		return;
1866	}
1867
1868	prom_printf("instantiating rtas at 0x%x...", base);
1869
1870	if (call_prom_ret("call-method", 3, 2, &entry,
1871			  ADDR("instantiate-rtas"),
1872			  rtas_inst, base) != 0
1873	    || entry == 0) {
1874		prom_printf(" failed\n");
1875		return;
1876	}
1877	prom_printf(" done\n");
1878
1879	reserve_mem(base, size);
1880
1881	val = cpu_to_be32(base);
1882	prom_setprop(rtas_node, "/rtas", "linux,rtas-base",
1883		     &val, sizeof(val));
1884	val = cpu_to_be32(entry);
1885	prom_setprop(rtas_node, "/rtas", "linux,rtas-entry",
1886		     &val, sizeof(val));
1887
1888	/* Check if it supports "query-cpu-stopped-state" */
1889	if (prom_getprop(rtas_node, "query-cpu-stopped-state",
1890			 &val, sizeof(val)) != PROM_ERROR)
1891		rtas_has_query_cpu_stopped = true;
1892
1893	prom_debug("rtas base     = 0x%x\n", base);
1894	prom_debug("rtas entry    = 0x%x\n", entry);
1895	prom_debug("rtas size     = 0x%x\n", size);
1896
1897	prom_debug("prom_instantiate_rtas: end...\n");
1898}
1899
1900#ifdef CONFIG_PPC64
1901/*
1902 * Allocate room for and instantiate Stored Measurement Log (SML)
1903 */
1904static void __init prom_instantiate_sml(void)
1905{
1906	phandle ibmvtpm_node;
1907	ihandle ibmvtpm_inst;
1908	u32 entry = 0, size = 0, succ = 0;
1909	u64 base;
1910	__be32 val;
1911
1912	prom_debug("prom_instantiate_sml: start...\n");
1913
1914	ibmvtpm_node = call_prom("finddevice", 1, 1, ADDR("/vdevice/vtpm"));
1915	prom_debug("ibmvtpm_node: %x\n", ibmvtpm_node);
1916	if (!PHANDLE_VALID(ibmvtpm_node))
1917		return;
1918
1919	ibmvtpm_inst = call_prom("open", 1, 1, ADDR("/vdevice/vtpm"));
1920	if (!IHANDLE_VALID(ibmvtpm_inst)) {
1921		prom_printf("opening vtpm package failed (%x)\n", ibmvtpm_inst);
1922		return;
1923	}
1924
1925	if (prom_getprop(ibmvtpm_node, "ibm,sml-efi-reformat-supported",
1926			 &val, sizeof(val)) != PROM_ERROR) {
1927		if (call_prom_ret("call-method", 2, 2, &succ,
1928				  ADDR("reformat-sml-to-efi-alignment"),
1929				  ibmvtpm_inst) != 0 || succ == 0) {
1930			prom_printf("Reformat SML to EFI alignment failed\n");
1931			return;
1932		}
1933
1934		if (call_prom_ret("call-method", 2, 2, &size,
1935				  ADDR("sml-get-allocated-size"),
1936				  ibmvtpm_inst) != 0 || size == 0) {
1937			prom_printf("SML get allocated size failed\n");
1938			return;
1939		}
1940	} else {
1941		if (call_prom_ret("call-method", 2, 2, &size,
1942				  ADDR("sml-get-handover-size"),
1943				  ibmvtpm_inst) != 0 || size == 0) {
1944			prom_printf("SML get handover size failed\n");
1945			return;
1946		}
1947	}
1948
1949	base = alloc_down(size, PAGE_SIZE, 0);
1950	if (base == 0)
1951		prom_panic("Could not allocate memory for sml\n");
1952
1953	prom_printf("instantiating sml at 0x%llx...", base);
1954
1955	memset((void *)base, 0, size);
1956
1957	if (call_prom_ret("call-method", 4, 2, &entry,
1958			  ADDR("sml-handover"),
1959			  ibmvtpm_inst, size, base) != 0 || entry == 0) {
1960		prom_printf("SML handover failed\n");
1961		return;
1962	}
1963	prom_printf(" done\n");
1964
1965	reserve_mem(base, size);
1966
1967	prom_setprop(ibmvtpm_node, "/vdevice/vtpm", "linux,sml-base",
1968		     &base, sizeof(base));
1969	prom_setprop(ibmvtpm_node, "/vdevice/vtpm", "linux,sml-size",
1970		     &size, sizeof(size));
1971
1972	prom_debug("sml base     = 0x%llx\n", base);
1973	prom_debug("sml size     = 0x%x\n", size);
1974
1975	prom_debug("prom_instantiate_sml: end...\n");
1976}
1977
1978/*
1979 * Allocate room for and initialize TCE tables
1980 */
1981#ifdef __BIG_ENDIAN__
1982static void __init prom_initialize_tce_table(void)
1983{
1984	phandle node;
1985	ihandle phb_node;
1986	char compatible[64], type[64], model[64];
1987	char *path = prom_scratch;
1988	u64 base, align;
1989	u32 minalign, minsize;
1990	u64 tce_entry, *tce_entryp;
1991	u64 local_alloc_top, local_alloc_bottom;
1992	u64 i;
1993
1994	if (prom_iommu_off)
1995		return;
1996
1997	prom_debug("starting prom_initialize_tce_table\n");
1998
1999	/* Cache current top of allocs so we reserve a single block */
2000	local_alloc_top = alloc_top_high;
2001	local_alloc_bottom = local_alloc_top;
2002
2003	/* Search all nodes looking for PHBs. */
2004	for (node = 0; prom_next_node(&node); ) {
2005		compatible[0] = 0;
2006		type[0] = 0;
2007		model[0] = 0;
2008		prom_getprop(node, "compatible",
2009			     compatible, sizeof(compatible));
2010		prom_getprop(node, "device_type", type, sizeof(type));
2011		prom_getprop(node, "model", model, sizeof(model));
2012
2013		if ((type[0] == 0) || (prom_strstr(type, "pci") == NULL))
2014			continue;
2015
2016		/* Keep the old logic intact to avoid regression. */
2017		if (compatible[0] != 0) {
2018			if ((prom_strstr(compatible, "python") == NULL) &&
2019			    (prom_strstr(compatible, "Speedwagon") == NULL) &&
2020			    (prom_strstr(compatible, "Winnipeg") == NULL))
2021				continue;
2022		} else if (model[0] != 0) {
2023			if ((prom_strstr(model, "ython") == NULL) &&
2024			    (prom_strstr(model, "peedwagon") == NULL) &&
2025			    (prom_strstr(model, "innipeg") == NULL))
2026				continue;
2027		}
2028
2029		if (prom_getprop(node, "tce-table-minalign", &minalign,
2030				 sizeof(minalign)) == PROM_ERROR)
2031			minalign = 0;
2032		if (prom_getprop(node, "tce-table-minsize", &minsize,
2033				 sizeof(minsize)) == PROM_ERROR)
2034			minsize = 4UL << 20;
2035
2036		/*
2037		 * Even though we read what OF wants, we just set the table
2038		 * size to 4 MB.  This is enough to map 2GB of PCI DMA space.
2039		 * By doing this, we avoid the pitfalls of trying to DMA to
2040		 * MMIO space and the DMA alias hole.
2041		 */
2042		minsize = 4UL << 20;
2043
2044		/* Align to the greater of the align or size */
2045		align = max(minalign, minsize);
2046		base = alloc_down(minsize, align, 1);
2047		if (base == 0)
2048			prom_panic("ERROR, cannot find space for TCE table.\n");
2049		if (base < local_alloc_bottom)
2050			local_alloc_bottom = base;
2051
2052		/* It seems OF doesn't null-terminate the path :-( */
2053		memset(path, 0, sizeof(prom_scratch));
2054		/* Call OF to setup the TCE hardware */
2055		if (call_prom("package-to-path", 3, 1, node,
2056			      path, sizeof(prom_scratch) - 1) == PROM_ERROR) {
2057			prom_printf("package-to-path failed\n");
2058		}
2059
2060		/* Save away the TCE table attributes for later use. */
2061		prom_setprop(node, path, "linux,tce-base", &base, sizeof(base));
2062		prom_setprop(node, path, "linux,tce-size", &minsize, sizeof(minsize));
2063
2064		prom_debug("TCE table: %s\n", path);
2065		prom_debug("\tnode = 0x%x\n", node);
2066		prom_debug("\tbase = 0x%llx\n", base);
2067		prom_debug("\tsize = 0x%x\n", minsize);
2068
2069		/* Initialize the table to have a one-to-one mapping
2070		 * over the allocated size.
2071		 */
2072		tce_entryp = (u64 *)base;
2073		for (i = 0; i < (minsize >> 3) ;tce_entryp++, i++) {
2074			tce_entry = (i << PAGE_SHIFT);
2075			tce_entry |= 0x3;
2076			*tce_entryp = tce_entry;
2077		}
2078
2079		prom_printf("opening PHB %s", path);
2080		phb_node = call_prom("open", 1, 1, path);
2081		if (phb_node == 0)
2082			prom_printf("... failed\n");
2083		else
2084			prom_printf("... done\n");
2085
2086		call_prom("call-method", 6, 0, ADDR("set-64-bit-addressing"),
2087			  phb_node, -1, minsize,
2088			  (u32) base, (u32) (base >> 32));
2089		call_prom("close", 1, 0, phb_node);
2090	}
2091
2092	reserve_mem(local_alloc_bottom, local_alloc_top - local_alloc_bottom);
2093
2094	/* These are only really needed if there is a memory limit in
2095	 * effect, but we don't know so export them always. */
2096	prom_tce_alloc_start = local_alloc_bottom;
2097	prom_tce_alloc_end = local_alloc_top;
2098
2099	/* Flag the first invalid entry */
2100	prom_debug("ending prom_initialize_tce_table\n");
2101}
2102#endif /* __BIG_ENDIAN__ */
2103#endif /* CONFIG_PPC64 */
2104
2105/*
2106 * With CHRP SMP we need to use the OF to start the other processors.
2107 * We can't wait until smp_boot_cpus (the OF is trashed by then)
2108 * so we have to put the processors into a holding pattern controlled
2109 * by the kernel (not OF) before we destroy the OF.
2110 *
2111 * This uses a chunk of low memory, puts some holding pattern
2112 * code there and sends the other processors off to there until
2113 * smp_boot_cpus tells them to do something.  The holding pattern
2114 * checks that address until its cpu # is there, when it is that
2115 * cpu jumps to __secondary_start().  smp_boot_cpus() takes care
2116 * of setting those values.
2117 *
2118 * We also use physical address 0x4 here to tell when a cpu
2119 * is in its holding pattern code.
2120 *
2121 * -- Cort
2122 */
2123/*
2124 * We want to reference the copy of __secondary_hold_* in the
2125 * 0 - 0x100 address range
2126 */
2127#define LOW_ADDR(x)	(((unsigned long) &(x)) & 0xff)
2128
2129static void __init prom_hold_cpus(void)
2130{
2131	unsigned long i;
2132	phandle node;
2133	char type[64];
2134	unsigned long *spinloop
2135		= (void *) LOW_ADDR(__secondary_hold_spinloop);
2136	unsigned long *acknowledge
2137		= (void *) LOW_ADDR(__secondary_hold_acknowledge);
2138	unsigned long secondary_hold = LOW_ADDR(__secondary_hold);
2139
2140	/*
2141	 * On pseries, if RTAS supports "query-cpu-stopped-state",
2142	 * we skip this stage, the CPUs will be started by the
2143	 * kernel using RTAS.
2144	 */
2145	if ((of_platform == PLATFORM_PSERIES ||
2146	     of_platform == PLATFORM_PSERIES_LPAR) &&
2147	    rtas_has_query_cpu_stopped) {
2148		prom_printf("prom_hold_cpus: skipped\n");
2149		return;
2150	}
2151
2152	prom_debug("prom_hold_cpus: start...\n");
2153	prom_debug("    1) spinloop       = 0x%lx\n", (unsigned long)spinloop);
2154	prom_debug("    1) *spinloop      = 0x%lx\n", *spinloop);
2155	prom_debug("    1) acknowledge    = 0x%lx\n",
2156		   (unsigned long)acknowledge);
2157	prom_debug("    1) *acknowledge   = 0x%lx\n", *acknowledge);
2158	prom_debug("    1) secondary_hold = 0x%lx\n", secondary_hold);
2159
2160	/* Set the common spinloop variable, so all of the secondary cpus
2161	 * will block when they are awakened from their OF spinloop.
2162	 * This must occur for both SMP and non SMP kernels, since OF will
2163	 * be trashed when we move the kernel.
2164	 */
2165	*spinloop = 0;
2166
2167	/* look for cpus */
2168	for (node = 0; prom_next_node(&node); ) {
2169		unsigned int cpu_no;
2170		__be32 reg;
2171
2172		type[0] = 0;
2173		prom_getprop(node, "device_type", type, sizeof(type));
2174		if (prom_strcmp(type, "cpu") != 0)
2175			continue;
2176
2177		/* Skip non-configured cpus. */
2178		if (prom_getprop(node, "status", type, sizeof(type)) > 0)
2179			if (prom_strcmp(type, "okay") != 0)
2180				continue;
2181
2182		reg = cpu_to_be32(-1); /* make sparse happy */
2183		prom_getprop(node, "reg", &reg, sizeof(reg));
2184		cpu_no = be32_to_cpu(reg);
2185
2186		prom_debug("cpu hw idx   = %u\n", cpu_no);
2187
2188		/* Init the acknowledge var which will be reset by
2189		 * the secondary cpu when it awakens from its OF
2190		 * spinloop.
2191		 */
2192		*acknowledge = (unsigned long)-1;
2193
2194		if (cpu_no != prom.cpu) {
2195			/* Primary Thread of non-boot cpu or any thread */
2196			prom_printf("starting cpu hw idx %u... ", cpu_no);
2197			call_prom("start-cpu", 3, 0, node,
2198				  secondary_hold, cpu_no);
2199
2200			for (i = 0; (i < 100000000) &&
2201			     (*acknowledge == ((unsigned long)-1)); i++ )
2202				mb();
2203
2204			if (*acknowledge == cpu_no)
2205				prom_printf("done\n");
2206			else
2207				prom_printf("failed: %lx\n", *acknowledge);
2208		}
2209#ifdef CONFIG_SMP
2210		else
2211			prom_printf("boot cpu hw idx %u\n", cpu_no);
2212#endif /* CONFIG_SMP */
2213	}
2214
2215	prom_debug("prom_hold_cpus: end...\n");
2216}
2217
2218
2219static void __init prom_init_client_services(unsigned long pp)
2220{
2221	/* Get a handle to the prom entry point before anything else */
2222	prom_entry = pp;
2223
2224	/* get a handle for the stdout device */
2225	prom.chosen = call_prom("finddevice", 1, 1, ADDR("/chosen"));
2226	if (!PHANDLE_VALID(prom.chosen))
2227		prom_panic("cannot find chosen"); /* msg won't be printed :( */
2228
2229	/* get device tree root */
2230	prom.root = call_prom("finddevice", 1, 1, ADDR("/"));
2231	if (!PHANDLE_VALID(prom.root))
2232		prom_panic("cannot find device tree root"); /* msg won't be printed :( */
2233
2234	prom.mmumap = 0;
2235}
2236
2237#ifdef CONFIG_PPC32
2238/*
2239 * For really old powermacs, we need to map things we claim.
2240 * For that, we need the ihandle of the mmu.
2241 * Also, on the longtrail, we need to work around other bugs.
2242 */
2243static void __init prom_find_mmu(void)
2244{
2245	phandle oprom;
2246	char version[64];
2247
2248	oprom = call_prom("finddevice", 1, 1, ADDR("/openprom"));
2249	if (!PHANDLE_VALID(oprom))
2250		return;
2251	if (prom_getprop(oprom, "model", version, sizeof(version)) <= 0)
2252		return;
2253	version[sizeof(version) - 1] = 0;
2254	/* XXX might need to add other versions here */
2255	if (prom_strcmp(version, "Open Firmware, 1.0.5") == 0)
2256		of_workarounds = OF_WA_CLAIM;
2257	else if (prom_strncmp(version, "FirmWorks,3.", 12) == 0) {
2258		of_workarounds = OF_WA_CLAIM | OF_WA_LONGTRAIL;
2259		call_prom("interpret", 1, 1, "dev /memory 0 to allow-reclaim");
2260	} else
2261		return;
2262	prom.memory = call_prom("open", 1, 1, ADDR("/memory"));
2263	prom_getprop(prom.chosen, "mmu", &prom.mmumap,
2264		     sizeof(prom.mmumap));
2265	prom.mmumap = be32_to_cpu(prom.mmumap);
2266	if (!IHANDLE_VALID(prom.memory) || !IHANDLE_VALID(prom.mmumap))
2267		of_workarounds &= ~OF_WA_CLAIM;		/* hmmm */
2268}
2269#else
2270#define prom_find_mmu()
2271#endif
2272
2273static void __init prom_init_stdout(void)
2274{
2275	char *path = of_stdout_device;
2276	char type[16];
2277	phandle stdout_node;
2278	__be32 val;
2279
2280	if (prom_getprop(prom.chosen, "stdout", &val, sizeof(val)) <= 0)
2281		prom_panic("cannot find stdout");
2282
2283	prom.stdout = be32_to_cpu(val);
2284
2285	/* Get the full OF pathname of the stdout device */
2286	memset(path, 0, 256);
2287	call_prom("instance-to-path", 3, 1, prom.stdout, path, 255);
2288	prom_printf("OF stdout device is: %s\n", of_stdout_device);
2289	prom_setprop(prom.chosen, "/chosen", "linux,stdout-path",
2290		     path, prom_strlen(path) + 1);
2291
2292	/* instance-to-package fails on PA-Semi */
2293	stdout_node = call_prom("instance-to-package", 1, 1, prom.stdout);
2294	if (stdout_node != PROM_ERROR) {
2295		val = cpu_to_be32(stdout_node);
2296
2297		/* If it's a display, note it */
2298		memset(type, 0, sizeof(type));
2299		prom_getprop(stdout_node, "device_type", type, sizeof(type));
2300		if (prom_strcmp(type, "display") == 0)
2301			prom_setprop(stdout_node, path, "linux,boot-display", NULL, 0);
2302	}
2303}
2304
2305static int __init prom_find_machine_type(void)
2306{
2307	static char compat[256] __prombss;
2308	int len, i = 0;
2309#ifdef CONFIG_PPC64
2310	phandle rtas;
2311	int x;
2312#endif
2313
2314	/* Look for a PowerMac or a Cell */
2315	len = prom_getprop(prom.root, "compatible",
2316			   compat, sizeof(compat)-1);
2317	if (len > 0) {
2318		compat[len] = 0;
2319		while (i < len) {
2320			char *p = &compat[i];
2321			int sl = prom_strlen(p);
2322			if (sl == 0)
2323				break;
2324			if (prom_strstr(p, "Power Macintosh") ||
2325			    prom_strstr(p, "MacRISC"))
2326				return PLATFORM_POWERMAC;
2327#ifdef CONFIG_PPC64
2328			/* We must make sure we don't detect the IBM Cell
2329			 * blades as pSeries due to some firmware issues,
2330			 * so we do it here.
2331			 */
2332			if (prom_strstr(p, "IBM,CBEA") ||
2333			    prom_strstr(p, "IBM,CPBW-1.0"))
2334				return PLATFORM_GENERIC;
2335#endif /* CONFIG_PPC64 */
2336			i += sl + 1;
2337		}
2338	}
2339#ifdef CONFIG_PPC64
2340	/* Try to figure out if it's an IBM pSeries or any other
2341	 * PAPR compliant platform. We assume it is if :
2342	 *  - /device_type is "chrp" (please, do NOT use that for future
2343	 *    non-IBM designs !
2344	 *  - it has /rtas
2345	 */
2346	len = prom_getprop(prom.root, "device_type",
2347			   compat, sizeof(compat)-1);
2348	if (len <= 0)
2349		return PLATFORM_GENERIC;
2350	if (prom_strcmp(compat, "chrp"))
2351		return PLATFORM_GENERIC;
2352
2353	/* Default to pSeries. We need to know if we are running LPAR */
2354	rtas = call_prom("finddevice", 1, 1, ADDR("/rtas"));
2355	if (!PHANDLE_VALID(rtas))
2356		return PLATFORM_GENERIC;
2357	x = prom_getproplen(rtas, "ibm,hypertas-functions");
2358	if (x != PROM_ERROR) {
2359		prom_debug("Hypertas detected, assuming LPAR !\n");
2360		return PLATFORM_PSERIES_LPAR;
2361	}
2362	return PLATFORM_PSERIES;
2363#else
2364	return PLATFORM_GENERIC;
2365#endif
2366}
2367
2368static int __init prom_set_color(ihandle ih, int i, int r, int g, int b)
2369{
2370	return call_prom("call-method", 6, 1, ADDR("color!"), ih, i, b, g, r);
2371}
2372
2373/*
2374 * If we have a display that we don't know how to drive,
2375 * we will want to try to execute OF's open method for it
2376 * later.  However, OF will probably fall over if we do that
2377 * we've taken over the MMU.
2378 * So we check whether we will need to open the display,
2379 * and if so, open it now.
2380 */
2381static void __init prom_check_displays(void)
2382{
2383	char type[16], *path;
2384	phandle node;
2385	ihandle ih;
2386	int i;
2387
2388	static const unsigned char default_colors[] __initconst = {
2389		0x00, 0x00, 0x00,
2390		0x00, 0x00, 0xaa,
2391		0x00, 0xaa, 0x00,
2392		0x00, 0xaa, 0xaa,
2393		0xaa, 0x00, 0x00,
2394		0xaa, 0x00, 0xaa,
2395		0xaa, 0xaa, 0x00,
2396		0xaa, 0xaa, 0xaa,
2397		0x55, 0x55, 0x55,
2398		0x55, 0x55, 0xff,
2399		0x55, 0xff, 0x55,
2400		0x55, 0xff, 0xff,
2401		0xff, 0x55, 0x55,
2402		0xff, 0x55, 0xff,
2403		0xff, 0xff, 0x55,
2404		0xff, 0xff, 0xff
2405	};
2406	const unsigned char *clut;
2407
2408	prom_debug("Looking for displays\n");
2409	for (node = 0; prom_next_node(&node); ) {
2410		memset(type, 0, sizeof(type));
2411		prom_getprop(node, "device_type", type, sizeof(type));
2412		if (prom_strcmp(type, "display") != 0)
2413			continue;
2414
2415		/* It seems OF doesn't null-terminate the path :-( */
2416		path = prom_scratch;
2417		memset(path, 0, sizeof(prom_scratch));
2418
2419		/*
2420		 * leave some room at the end of the path for appending extra
2421		 * arguments
2422		 */
2423		if (call_prom("package-to-path", 3, 1, node, path,
2424			      sizeof(prom_scratch) - 10) == PROM_ERROR)
2425			continue;
2426		prom_printf("found display   : %s, opening... ", path);
2427
2428		ih = call_prom("open", 1, 1, path);
2429		if (ih == 0) {
2430			prom_printf("failed\n");
2431			continue;
2432		}
2433
2434		/* Success */
2435		prom_printf("done\n");
2436		prom_setprop(node, path, "linux,opened", NULL, 0);
2437
2438		/* Setup a usable color table when the appropriate
2439		 * method is available. Should update this to set-colors */
2440		clut = default_colors;
2441		for (i = 0; i < 16; i++, clut += 3)
2442			if (prom_set_color(ih, i, clut[0], clut[1],
2443					   clut[2]) != 0)
2444				break;
2445
2446#ifdef CONFIG_LOGO_LINUX_CLUT224
2447		clut = PTRRELOC(logo_linux_clut224.clut);
2448		for (i = 0; i < logo_linux_clut224.clutsize; i++, clut += 3)
2449			if (prom_set_color(ih, i + 32, clut[0], clut[1],
2450					   clut[2]) != 0)
2451				break;
2452#endif /* CONFIG_LOGO_LINUX_CLUT224 */
2453
2454#ifdef CONFIG_PPC_EARLY_DEBUG_BOOTX
2455		if (prom_getprop(node, "linux,boot-display", NULL, 0) !=
2456		    PROM_ERROR) {
2457			u32 width, height, pitch, addr;
2458
2459			prom_printf("Setting btext !\n");
2460
2461			if (prom_getprop(node, "width", &width, 4) == PROM_ERROR)
2462				return;
2463
2464			if (prom_getprop(node, "height", &height, 4) == PROM_ERROR)
2465				return;
2466
2467			if (prom_getprop(node, "linebytes", &pitch, 4) == PROM_ERROR)
2468				return;
2469
2470			if (prom_getprop(node, "address", &addr, 4) == PROM_ERROR)
2471				return;
2472
2473			prom_printf("W=%d H=%d LB=%d addr=0x%x\n",
2474				    width, height, pitch, addr);
2475			btext_setup_display(width, height, 8, pitch, addr);
2476			btext_prepare_BAT();
2477		}
2478#endif /* CONFIG_PPC_EARLY_DEBUG_BOOTX */
2479	}
2480}
2481
2482
2483/* Return (relocated) pointer to this much memory: moves initrd if reqd. */
2484static void __init *make_room(unsigned long *mem_start, unsigned long *mem_end,
2485			      unsigned long needed, unsigned long align)
2486{
2487	void *ret;
2488
2489	*mem_start = ALIGN(*mem_start, align);
2490	while ((*mem_start + needed) > *mem_end) {
2491		unsigned long room, chunk;
2492
2493		prom_debug("Chunk exhausted, claiming more at %lx...\n",
2494			   alloc_bottom);
2495		room = alloc_top - alloc_bottom;
2496		if (room > DEVTREE_CHUNK_SIZE)
2497			room = DEVTREE_CHUNK_SIZE;
2498		if (room < PAGE_SIZE)
2499			prom_panic("No memory for flatten_device_tree "
2500				   "(no room)\n");
2501		chunk = alloc_up(room, 0);
2502		if (chunk == 0)
2503			prom_panic("No memory for flatten_device_tree "
2504				   "(claim failed)\n");
2505		*mem_end = chunk + room;
2506	}
2507
2508	ret = (void *)*mem_start;
2509	*mem_start += needed;
2510
2511	return ret;
2512}
2513
2514#define dt_push_token(token, mem_start, mem_end) do { 			\
2515		void *room = make_room(mem_start, mem_end, 4, 4);	\
2516		*(__be32 *)room = cpu_to_be32(token);			\
2517	} while(0)
2518
2519static unsigned long __init dt_find_string(char *str)
2520{
2521	char *s, *os;
2522
2523	s = os = (char *)dt_string_start;
2524	s += 4;
2525	while (s <  (char *)dt_string_end) {
2526		if (prom_strcmp(s, str) == 0)
2527			return s - os;
2528		s += prom_strlen(s) + 1;
2529	}
2530	return 0;
2531}
2532
2533/*
2534 * The Open Firmware 1275 specification states properties must be 31 bytes or
2535 * less, however not all firmwares obey this. Make it 64 bytes to be safe.
2536 */
2537#define MAX_PROPERTY_NAME 64
2538
2539static void __init scan_dt_build_strings(phandle node,
2540					 unsigned long *mem_start,
2541					 unsigned long *mem_end)
2542{
2543	char *prev_name, *namep, *sstart;
2544	unsigned long soff;
2545	phandle child;
2546
2547	sstart =  (char *)dt_string_start;
2548
2549	/* get and store all property names */
2550	prev_name = "";
2551	for (;;) {
2552		/* 64 is max len of name including nul. */
2553		namep = make_room(mem_start, mem_end, MAX_PROPERTY_NAME, 1);
2554		if (call_prom("nextprop", 3, 1, node, prev_name, namep) != 1) {
2555			/* No more nodes: unwind alloc */
2556			*mem_start = (unsigned long)namep;
2557			break;
2558		}
2559
2560 		/* skip "name" */
2561		if (prom_strcmp(namep, "name") == 0) {
2562 			*mem_start = (unsigned long)namep;
2563 			prev_name = "name";
2564 			continue;
2565 		}
2566		/* get/create string entry */
2567		soff = dt_find_string(namep);
2568		if (soff != 0) {
2569			*mem_start = (unsigned long)namep;
2570			namep = sstart + soff;
2571		} else {
2572			/* Trim off some if we can */
2573			*mem_start = (unsigned long)namep + prom_strlen(namep) + 1;
2574			dt_string_end = *mem_start;
2575		}
2576		prev_name = namep;
2577	}
2578
2579	/* do all our children */
2580	child = call_prom("child", 1, 1, node);
2581	while (child != 0) {
2582		scan_dt_build_strings(child, mem_start, mem_end);
2583		child = call_prom("peer", 1, 1, child);
2584	}
2585}
2586
2587static void __init scan_dt_build_struct(phandle node, unsigned long *mem_start,
2588					unsigned long *mem_end)
2589{
2590	phandle child;
2591	char *namep, *prev_name, *sstart, *p, *ep, *lp, *path;
2592	unsigned long soff;
2593	unsigned char *valp;
2594	static char pname[MAX_PROPERTY_NAME] __prombss;
2595	int l, room, has_phandle = 0;
2596
2597	dt_push_token(OF_DT_BEGIN_NODE, mem_start, mem_end);
2598
2599	/* get the node's full name */
2600	namep = (char *)*mem_start;
2601	room = *mem_end - *mem_start;
2602	if (room > 255)
2603		room = 255;
2604	l = call_prom("package-to-path", 3, 1, node, namep, room);
2605	if (l >= 0) {
2606		/* Didn't fit?  Get more room. */
2607		if (l >= room) {
2608			if (l >= *mem_end - *mem_start)
2609				namep = make_room(mem_start, mem_end, l+1, 1);
2610			call_prom("package-to-path", 3, 1, node, namep, l);
2611		}
2612		namep[l] = '\0';
2613
2614		/* Fixup an Apple bug where they have bogus \0 chars in the
2615		 * middle of the path in some properties, and extract
2616		 * the unit name (everything after the last '/').
2617		 */
2618		for (lp = p = namep, ep = namep + l; p < ep; p++) {
2619			if (*p == '/')
2620				lp = namep;
2621			else if (*p != 0)
2622				*lp++ = *p;
2623		}
2624		*lp = 0;
2625		*mem_start = ALIGN((unsigned long)lp + 1, 4);
2626	}
2627
2628	/* get it again for debugging */
2629	path = prom_scratch;
2630	memset(path, 0, sizeof(prom_scratch));
2631	call_prom("package-to-path", 3, 1, node, path, sizeof(prom_scratch) - 1);
2632
2633	/* get and store all properties */
2634	prev_name = "";
2635	sstart = (char *)dt_string_start;
2636	for (;;) {
2637		if (call_prom("nextprop", 3, 1, node, prev_name,
2638			      pname) != 1)
2639			break;
2640
2641 		/* skip "name" */
2642		if (prom_strcmp(pname, "name") == 0) {
2643 			prev_name = "name";
2644 			continue;
2645 		}
2646
2647		/* find string offset */
2648		soff = dt_find_string(pname);
2649		if (soff == 0) {
2650			prom_printf("WARNING: Can't find string index for"
2651				    " <%s>, node %s\n", pname, path);
2652			break;
2653		}
2654		prev_name = sstart + soff;
2655
2656		/* get length */
2657		l = call_prom("getproplen", 2, 1, node, pname);
2658
2659		/* sanity checks */
2660		if (l == PROM_ERROR)
2661			continue;
2662
2663		/* push property head */
2664		dt_push_token(OF_DT_PROP, mem_start, mem_end);
2665		dt_push_token(l, mem_start, mem_end);
2666		dt_push_token(soff, mem_start, mem_end);
2667
2668		/* push property content */
2669		valp = make_room(mem_start, mem_end, l, 4);
2670		call_prom("getprop", 4, 1, node, pname, valp, l);
2671		*mem_start = ALIGN(*mem_start, 4);
2672
2673		if (!prom_strcmp(pname, "phandle"))
2674			has_phandle = 1;
2675	}
2676
2677	/* Add a "phandle" property if none already exist */
2678	if (!has_phandle) {
2679		soff = dt_find_string("phandle");
2680		if (soff == 0)
2681			prom_printf("WARNING: Can't find string index for <phandle> node %s\n", path);
2682		else {
2683			dt_push_token(OF_DT_PROP, mem_start, mem_end);
2684			dt_push_token(4, mem_start, mem_end);
2685			dt_push_token(soff, mem_start, mem_end);
2686			valp = make_room(mem_start, mem_end, 4, 4);
2687			*(__be32 *)valp = cpu_to_be32(node);
2688		}
2689	}
2690
2691	/* do all our children */
2692	child = call_prom("child", 1, 1, node);
2693	while (child != 0) {
2694		scan_dt_build_struct(child, mem_start, mem_end);
2695		child = call_prom("peer", 1, 1, child);
2696	}
2697
2698	dt_push_token(OF_DT_END_NODE, mem_start, mem_end);
2699}
2700
2701static void __init flatten_device_tree(void)
2702{
2703	phandle root;
2704	unsigned long mem_start, mem_end, room;
2705	struct boot_param_header *hdr;
2706	char *namep;
2707	u64 *rsvmap;
2708
2709	/*
2710	 * Check how much room we have between alloc top & bottom (+/- a
2711	 * few pages), crop to 1MB, as this is our "chunk" size
2712	 */
2713	room = alloc_top - alloc_bottom - 0x4000;
2714	if (room > DEVTREE_CHUNK_SIZE)
2715		room = DEVTREE_CHUNK_SIZE;
2716	prom_debug("starting device tree allocs at %lx\n", alloc_bottom);
2717
2718	/* Now try to claim that */
2719	mem_start = (unsigned long)alloc_up(room, PAGE_SIZE);
2720	if (mem_start == 0)
2721		prom_panic("Can't allocate initial device-tree chunk\n");
2722	mem_end = mem_start + room;
2723
2724	/* Get root of tree */
2725	root = call_prom("peer", 1, 1, (phandle)0);
2726	if (root == (phandle)0)
2727		prom_panic ("couldn't get device tree root\n");
2728
2729	/* Build header and make room for mem rsv map */
2730	mem_start = ALIGN(mem_start, 4);
2731	hdr = make_room(&mem_start, &mem_end,
2732			sizeof(struct boot_param_header), 4);
2733	dt_header_start = (unsigned long)hdr;
2734	rsvmap = make_room(&mem_start, &mem_end, sizeof(mem_reserve_map), 8);
2735
2736	/* Start of strings */
2737	mem_start = PAGE_ALIGN(mem_start);
2738	dt_string_start = mem_start;
2739	mem_start += 4; /* hole */
2740
2741	/* Add "phandle" in there, we'll need it */
2742	namep = make_room(&mem_start, &mem_end, 16, 1);
2743	prom_strscpy_pad(namep, "phandle", sizeof("phandle"));
2744	mem_start = (unsigned long)namep + prom_strlen(namep) + 1;
2745
2746	/* Build string array */
2747	prom_printf("Building dt strings...\n");
2748	scan_dt_build_strings(root, &mem_start, &mem_end);
2749	dt_string_end = mem_start;
2750
2751	/* Build structure */
2752	mem_start = PAGE_ALIGN(mem_start);
2753	dt_struct_start = mem_start;
2754	prom_printf("Building dt structure...\n");
2755	scan_dt_build_struct(root, &mem_start, &mem_end);
2756	dt_push_token(OF_DT_END, &mem_start, &mem_end);
2757	dt_struct_end = PAGE_ALIGN(mem_start);
2758
2759	/* Finish header */
2760	hdr->boot_cpuid_phys = cpu_to_be32(prom.cpu);
2761	hdr->magic = cpu_to_be32(OF_DT_HEADER);
2762	hdr->totalsize = cpu_to_be32(dt_struct_end - dt_header_start);
2763	hdr->off_dt_struct = cpu_to_be32(dt_struct_start - dt_header_start);
2764	hdr->off_dt_strings = cpu_to_be32(dt_string_start - dt_header_start);
2765	hdr->dt_strings_size = cpu_to_be32(dt_string_end - dt_string_start);
2766	hdr->off_mem_rsvmap = cpu_to_be32(((unsigned long)rsvmap) - dt_header_start);
2767	hdr->version = cpu_to_be32(OF_DT_VERSION);
2768	/* Version 16 is not backward compatible */
2769	hdr->last_comp_version = cpu_to_be32(0x10);
2770
2771	/* Copy the reserve map in */
2772	memcpy(rsvmap, mem_reserve_map, sizeof(mem_reserve_map));
2773
2774#ifdef DEBUG_PROM
2775	{
2776		int i;
2777		prom_printf("reserved memory map:\n");
2778		for (i = 0; i < mem_reserve_cnt; i++)
2779			prom_printf("  %llx - %llx\n",
2780				    be64_to_cpu(mem_reserve_map[i].base),
2781				    be64_to_cpu(mem_reserve_map[i].size));
2782	}
2783#endif
2784	/* Bump mem_reserve_cnt to cause further reservations to fail
2785	 * since it's too late.
2786	 */
2787	mem_reserve_cnt = MEM_RESERVE_MAP_SIZE;
2788
2789	prom_printf("Device tree strings 0x%lx -> 0x%lx\n",
2790		    dt_string_start, dt_string_end);
2791	prom_printf("Device tree struct  0x%lx -> 0x%lx\n",
2792		    dt_struct_start, dt_struct_end);
2793}
2794
2795#ifdef CONFIG_PPC_MAPLE
2796/* PIBS Version 1.05.0000 04/26/2005 has an incorrect /ht/isa/ranges property.
2797 * The values are bad, and it doesn't even have the right number of cells. */
2798static void __init fixup_device_tree_maple(void)
2799{
2800	phandle isa;
2801	u32 rloc = 0x01002000; /* IO space; PCI device = 4 */
2802	u32 isa_ranges[6];
2803	char *name;
2804
2805	name = "/ht@0/isa@4";
2806	isa = call_prom("finddevice", 1, 1, ADDR(name));
2807	if (!PHANDLE_VALID(isa)) {
2808		name = "/ht@0/isa@6";
2809		isa = call_prom("finddevice", 1, 1, ADDR(name));
2810		rloc = 0x01003000; /* IO space; PCI device = 6 */
2811	}
2812	if (!PHANDLE_VALID(isa))
2813		return;
2814
2815	if (prom_getproplen(isa, "ranges") != 12)
2816		return;
2817	if (prom_getprop(isa, "ranges", isa_ranges, sizeof(isa_ranges))
2818		== PROM_ERROR)
2819		return;
2820
2821	if (isa_ranges[0] != 0x1 ||
2822		isa_ranges[1] != 0xf4000000 ||
2823		isa_ranges[2] != 0x00010000)
2824		return;
2825
2826	prom_printf("Fixing up bogus ISA range on Maple/Apache...\n");
2827
2828	isa_ranges[0] = 0x1;
2829	isa_ranges[1] = 0x0;
2830	isa_ranges[2] = rloc;
2831	isa_ranges[3] = 0x0;
2832	isa_ranges[4] = 0x0;
2833	isa_ranges[5] = 0x00010000;
2834	prom_setprop(isa, name, "ranges",
2835			isa_ranges, sizeof(isa_ranges));
2836}
2837
2838#define CPC925_MC_START		0xf8000000
2839#define CPC925_MC_LENGTH	0x1000000
2840/* The values for memory-controller don't have right number of cells */
2841static void __init fixup_device_tree_maple_memory_controller(void)
2842{
2843	phandle mc;
2844	u32 mc_reg[4];
2845	char *name = "/hostbridge@f8000000";
2846	u32 ac, sc;
2847
2848	mc = call_prom("finddevice", 1, 1, ADDR(name));
2849	if (!PHANDLE_VALID(mc))
2850		return;
2851
2852	if (prom_getproplen(mc, "reg") != 8)
2853		return;
2854
2855	prom_getprop(prom.root, "#address-cells", &ac, sizeof(ac));
2856	prom_getprop(prom.root, "#size-cells", &sc, sizeof(sc));
2857	if ((ac != 2) || (sc != 2))
2858		return;
2859
2860	if (prom_getprop(mc, "reg", mc_reg, sizeof(mc_reg)) == PROM_ERROR)
2861		return;
2862
2863	if (mc_reg[0] != CPC925_MC_START || mc_reg[1] != CPC925_MC_LENGTH)
2864		return;
2865
2866	prom_printf("Fixing up bogus hostbridge on Maple...\n");
2867
2868	mc_reg[0] = 0x0;
2869	mc_reg[1] = CPC925_MC_START;
2870	mc_reg[2] = 0x0;
2871	mc_reg[3] = CPC925_MC_LENGTH;
2872	prom_setprop(mc, name, "reg", mc_reg, sizeof(mc_reg));
2873}
2874#else
2875#define fixup_device_tree_maple()
2876#define fixup_device_tree_maple_memory_controller()
2877#endif
2878
2879#ifdef CONFIG_PPC_CHRP
2880/*
2881 * Pegasos and BriQ lacks the "ranges" property in the isa node
2882 * Pegasos needs decimal IRQ 14/15, not hexadecimal
2883 * Pegasos has the IDE configured in legacy mode, but advertised as native
2884 */
2885static void __init fixup_device_tree_chrp(void)
2886{
2887	phandle ph;
2888	u32 prop[6];
2889	u32 rloc = 0x01006000; /* IO space; PCI device = 12 */
2890	char *name;
2891	int rc;
2892
2893	name = "/pci@80000000/isa@c";
2894	ph = call_prom("finddevice", 1, 1, ADDR(name));
2895	if (!PHANDLE_VALID(ph)) {
2896		name = "/pci@ff500000/isa@6";
2897		ph = call_prom("finddevice", 1, 1, ADDR(name));
2898		rloc = 0x01003000; /* IO space; PCI device = 6 */
2899	}
2900	if (PHANDLE_VALID(ph)) {
2901		rc = prom_getproplen(ph, "ranges");
2902		if (rc == 0 || rc == PROM_ERROR) {
2903			prom_printf("Fixing up missing ISA range on Pegasos...\n");
2904
2905			prop[0] = 0x1;
2906			prop[1] = 0x0;
2907			prop[2] = rloc;
2908			prop[3] = 0x0;
2909			prop[4] = 0x0;
2910			prop[5] = 0x00010000;
2911			prom_setprop(ph, name, "ranges", prop, sizeof(prop));
2912		}
2913	}
2914
2915	name = "/pci@80000000/ide@C,1";
2916	ph = call_prom("finddevice", 1, 1, ADDR(name));
2917	if (PHANDLE_VALID(ph)) {
2918		prom_printf("Fixing up IDE interrupt on Pegasos...\n");
2919		prop[0] = 14;
2920		prop[1] = 0x0;
2921		prom_setprop(ph, name, "interrupts", prop, 2*sizeof(u32));
2922		prom_printf("Fixing up IDE class-code on Pegasos...\n");
2923		rc = prom_getprop(ph, "class-code", prop, sizeof(u32));
2924		if (rc == sizeof(u32)) {
2925			prop[0] &= ~0x5;
2926			prom_setprop(ph, name, "class-code", prop, sizeof(u32));
2927		}
2928	}
2929}
2930#else
2931#define fixup_device_tree_chrp()
2932#endif
2933
2934#if defined(CONFIG_PPC64) && defined(CONFIG_PPC_PMAC)
2935static void __init fixup_device_tree_pmac(void)
2936{
2937	phandle u3, i2c, mpic;
2938	u32 u3_rev;
2939	u32 interrupts[2];
2940	u32 parent;
2941
2942	/* Some G5s have a missing interrupt definition, fix it up here */
2943	u3 = call_prom("finddevice", 1, 1, ADDR("/u3@0,f8000000"));
2944	if (!PHANDLE_VALID(u3))
2945		return;
2946	i2c = call_prom("finddevice", 1, 1, ADDR("/u3@0,f8000000/i2c@f8001000"));
2947	if (!PHANDLE_VALID(i2c))
2948		return;
2949	mpic = call_prom("finddevice", 1, 1, ADDR("/u3@0,f8000000/mpic@f8040000"));
2950	if (!PHANDLE_VALID(mpic))
2951		return;
2952
2953	/* check if proper rev of u3 */
2954	if (prom_getprop(u3, "device-rev", &u3_rev, sizeof(u3_rev))
2955	    == PROM_ERROR)
2956		return;
2957	if (u3_rev < 0x35 || u3_rev > 0x39)
2958		return;
2959	/* does it need fixup ? */
2960	if (prom_getproplen(i2c, "interrupts") > 0)
2961		return;
2962
2963	prom_printf("fixing up bogus interrupts for u3 i2c...\n");
2964
2965	/* interrupt on this revision of u3 is number 0 and level */
2966	interrupts[0] = 0;
2967	interrupts[1] = 1;
2968	prom_setprop(i2c, "/u3@0,f8000000/i2c@f8001000", "interrupts",
2969		     &interrupts, sizeof(interrupts));
2970	parent = (u32)mpic;
2971	prom_setprop(i2c, "/u3@0,f8000000/i2c@f8001000", "interrupt-parent",
2972		     &parent, sizeof(parent));
2973}
2974#else
2975#define fixup_device_tree_pmac()
2976#endif
2977
2978#ifdef CONFIG_PPC_EFIKA
2979/*
2980 * The MPC5200 FEC driver requires an phy-handle property to tell it how
2981 * to talk to the phy.  If the phy-handle property is missing, then this
2982 * function is called to add the appropriate nodes and link it to the
2983 * ethernet node.
2984 */
2985static void __init fixup_device_tree_efika_add_phy(void)
2986{
2987	u32 node;
2988	char prop[64];
2989	int rv;
2990
2991	/* Check if /builtin/ethernet exists - bail if it doesn't */
2992	node = call_prom("finddevice", 1, 1, ADDR("/builtin/ethernet"));
2993	if (!PHANDLE_VALID(node))
2994		return;
2995
2996	/* Check if the phy-handle property exists - bail if it does */
2997	rv = prom_getprop(node, "phy-handle", prop, sizeof(prop));
2998	if (rv <= 0)
2999		return;
3000
3001	/*
3002	 * At this point the ethernet device doesn't have a phy described.
3003	 * Now we need to add the missing phy node and linkage
3004	 */
3005
3006	/* Check for an MDIO bus node - if missing then create one */
3007	node = call_prom("finddevice", 1, 1, ADDR("/builtin/mdio"));
3008	if (!PHANDLE_VALID(node)) {
3009		prom_printf("Adding Ethernet MDIO node\n");
3010		call_prom("interpret", 1, 1,
3011			" s\" /builtin\" find-device"
3012			" new-device"
3013				" 1 encode-int s\" #address-cells\" property"
3014				" 0 encode-int s\" #size-cells\" property"
3015				" s\" mdio\" device-name"
3016				" s\" fsl,mpc5200b-mdio\" encode-string"
3017				" s\" compatible\" property"
3018				" 0xf0003000 0x400 reg"
3019				" 0x2 encode-int"
3020				" 0x5 encode-int encode+"
3021				" 0x3 encode-int encode+"
3022				" s\" interrupts\" property"
3023			" finish-device");
3024	}
3025
3026	/* Check for a PHY device node - if missing then create one and
3027	 * give it's phandle to the ethernet node */
3028	node = call_prom("finddevice", 1, 1,
3029			 ADDR("/builtin/mdio/ethernet-phy"));
3030	if (!PHANDLE_VALID(node)) {
3031		prom_printf("Adding Ethernet PHY node\n");
3032		call_prom("interpret", 1, 1,
3033			" s\" /builtin/mdio\" find-device"
3034			" new-device"
3035				" s\" ethernet-phy\" device-name"
3036				" 0x10 encode-int s\" reg\" property"
3037				" my-self"
3038				" ihandle>phandle"
3039			" finish-device"
3040			" s\" /builtin/ethernet\" find-device"
3041				" encode-int"
3042				" s\" phy-handle\" property"
3043			" device-end");
3044	}
3045}
3046
3047static void __init fixup_device_tree_efika(void)
3048{
3049	int sound_irq[3] = { 2, 2, 0 };
3050	int bcomm_irq[3*16] = { 3,0,0, 3,1,0, 3,2,0, 3,3,0,
3051				3,4,0, 3,5,0, 3,6,0, 3,7,0,
3052				3,8,0, 3,9,0, 3,10,0, 3,11,0,
3053				3,12,0, 3,13,0, 3,14,0, 3,15,0 };
3054	u32 node;
3055	char prop[64];
3056	int rv, len;
3057
3058	/* Check if we're really running on a EFIKA */
3059	node = call_prom("finddevice", 1, 1, ADDR("/"));
3060	if (!PHANDLE_VALID(node))
3061		return;
3062
3063	rv = prom_getprop(node, "model", prop, sizeof(prop));
3064	if (rv == PROM_ERROR)
3065		return;
3066	if (prom_strcmp(prop, "EFIKA5K2"))
3067		return;
3068
3069	prom_printf("Applying EFIKA device tree fixups\n");
3070
3071	/* Claiming to be 'chrp' is death */
3072	node = call_prom("finddevice", 1, 1, ADDR("/"));
3073	rv = prom_getprop(node, "device_type", prop, sizeof(prop));
3074	if (rv != PROM_ERROR && (prom_strcmp(prop, "chrp") == 0))
3075		prom_setprop(node, "/", "device_type", "efika", sizeof("efika"));
3076
3077	/* CODEGEN,description is exposed in /proc/cpuinfo so
3078	   fix that too */
3079	rv = prom_getprop(node, "CODEGEN,description", prop, sizeof(prop));
3080	if (rv != PROM_ERROR && (prom_strstr(prop, "CHRP")))
3081		prom_setprop(node, "/", "CODEGEN,description",
3082			     "Efika 5200B PowerPC System",
3083			     sizeof("Efika 5200B PowerPC System"));
3084
3085	/* Fixup bestcomm interrupts property */
3086	node = call_prom("finddevice", 1, 1, ADDR("/builtin/bestcomm"));
3087	if (PHANDLE_VALID(node)) {
3088		len = prom_getproplen(node, "interrupts");
3089		if (len == 12) {
3090			prom_printf("Fixing bestcomm interrupts property\n");
3091			prom_setprop(node, "/builtin/bestcom", "interrupts",
3092				     bcomm_irq, sizeof(bcomm_irq));
3093		}
3094	}
3095
3096	/* Fixup sound interrupts property */
3097	node = call_prom("finddevice", 1, 1, ADDR("/builtin/sound"));
3098	if (PHANDLE_VALID(node)) {
3099		rv = prom_getprop(node, "interrupts", prop, sizeof(prop));
3100		if (rv == PROM_ERROR) {
3101			prom_printf("Adding sound interrupts property\n");
3102			prom_setprop(node, "/builtin/sound", "interrupts",
3103				     sound_irq, sizeof(sound_irq));
3104		}
3105	}
3106
3107	/* Make sure ethernet phy-handle property exists */
3108	fixup_device_tree_efika_add_phy();
3109}
3110#else
3111#define fixup_device_tree_efika()
3112#endif
3113
3114#ifdef CONFIG_PPC_PASEMI_NEMO
3115/*
3116 * CFE supplied on Nemo is broken in several ways, biggest
3117 * problem is that it reassigns ISA interrupts to unused mpic ints.
3118 * Add an interrupt-controller property for the io-bridge to use
3119 * and correct the ints so we can attach them to an irq_domain
3120 */
3121static void __init fixup_device_tree_pasemi(void)
3122{
3123	u32 interrupts[2], parent, rval, val = 0;
3124	char *name, *pci_name;
3125	phandle iob, node;
3126
3127	/* Find the root pci node */
3128	name = "/pxp@0,e0000000";
3129	iob = call_prom("finddevice", 1, 1, ADDR(name));
3130	if (!PHANDLE_VALID(iob))
3131		return;
3132
3133	/* check if interrupt-controller node set yet */
3134	if (prom_getproplen(iob, "interrupt-controller") !=PROM_ERROR)
3135		return;
3136
3137	prom_printf("adding interrupt-controller property for SB600...\n");
3138
3139	prom_setprop(iob, name, "interrupt-controller", &val, 0);
3140
3141	pci_name = "/pxp@0,e0000000/pci@11";
3142	node = call_prom("finddevice", 1, 1, ADDR(pci_name));
3143	parent = ADDR(iob);
3144
3145	for( ; prom_next_node(&node); ) {
3146		/* scan each node for one with an interrupt */
3147		if (!PHANDLE_VALID(node))
3148			continue;
3149
3150		rval = prom_getproplen(node, "interrupts");
3151		if (rval == 0 || rval == PROM_ERROR)
3152			continue;
3153
3154		prom_getprop(node, "interrupts", &interrupts, sizeof(interrupts));
3155		if ((interrupts[0] < 212) || (interrupts[0] > 222))
3156			continue;
3157
3158		/* found a node, update both interrupts and interrupt-parent */
3159		if ((interrupts[0] >= 212) && (interrupts[0] <= 215))
3160			interrupts[0] -= 203;
3161		if ((interrupts[0] >= 216) && (interrupts[0] <= 220))
3162			interrupts[0] -= 213;
3163		if (interrupts[0] == 221)
3164			interrupts[0] = 14;
3165		if (interrupts[0] == 222)
3166			interrupts[0] = 8;
3167
3168		prom_setprop(node, pci_name, "interrupts", interrupts,
3169					sizeof(interrupts));
3170		prom_setprop(node, pci_name, "interrupt-parent", &parent,
3171					sizeof(parent));
3172	}
3173
3174	/*
3175	 * The io-bridge has device_type set to 'io-bridge' change it to 'isa'
3176	 * so that generic isa-bridge code can add the SB600 and its on-board
3177	 * peripherals.
3178	 */
3179	name = "/pxp@0,e0000000/io-bridge@0";
3180	iob = call_prom("finddevice", 1, 1, ADDR(name));
3181	if (!PHANDLE_VALID(iob))
3182		return;
3183
3184	/* device_type is already set, just change it. */
3185
3186	prom_printf("Changing device_type of SB600 node...\n");
3187
3188	prom_setprop(iob, name, "device_type", "isa", sizeof("isa"));
3189}
3190#else	/* !CONFIG_PPC_PASEMI_NEMO */
3191static inline void fixup_device_tree_pasemi(void) { }
3192#endif
3193
3194static void __init fixup_device_tree(void)
3195{
3196	fixup_device_tree_maple();
3197	fixup_device_tree_maple_memory_controller();
3198	fixup_device_tree_chrp();
3199	fixup_device_tree_pmac();
3200	fixup_device_tree_efika();
3201	fixup_device_tree_pasemi();
3202}
3203
3204static void __init prom_find_boot_cpu(void)
3205{
3206	__be32 rval;
3207	ihandle prom_cpu;
3208	phandle cpu_pkg;
3209
3210	rval = 0;
3211	if (prom_getprop(prom.chosen, "cpu", &rval, sizeof(rval)) <= 0)
3212		return;
3213	prom_cpu = be32_to_cpu(rval);
3214
3215	cpu_pkg = call_prom("instance-to-package", 1, 1, prom_cpu);
3216
3217	if (!PHANDLE_VALID(cpu_pkg))
3218		return;
3219
3220	prom_getprop(cpu_pkg, "reg", &rval, sizeof(rval));
3221	prom.cpu = be32_to_cpu(rval);
3222
3223	prom_debug("Booting CPU hw index = %d\n", prom.cpu);
3224}
3225
3226static void __init prom_check_initrd(unsigned long r3, unsigned long r4)
3227{
3228#ifdef CONFIG_BLK_DEV_INITRD
3229	if (r3 && r4 && r4 != 0xdeadbeef) {
3230		__be64 val;
3231
3232		prom_initrd_start = is_kernel_addr(r3) ? __pa(r3) : r3;
3233		prom_initrd_end = prom_initrd_start + r4;
3234
3235		val = cpu_to_be64(prom_initrd_start);
3236		prom_setprop(prom.chosen, "/chosen", "linux,initrd-start",
3237			     &val, sizeof(val));
3238		val = cpu_to_be64(prom_initrd_end);
3239		prom_setprop(prom.chosen, "/chosen", "linux,initrd-end",
3240			     &val, sizeof(val));
3241
3242		reserve_mem(prom_initrd_start,
3243			    prom_initrd_end - prom_initrd_start);
3244
3245		prom_debug("initrd_start=0x%lx\n", prom_initrd_start);
3246		prom_debug("initrd_end=0x%lx\n", prom_initrd_end);
3247	}
3248#endif /* CONFIG_BLK_DEV_INITRD */
3249}
3250
3251#ifdef CONFIG_PPC_SVM
3252/*
3253 * Perform the Enter Secure Mode ultracall.
3254 */
3255static int __init enter_secure_mode(unsigned long kbase, unsigned long fdt)
3256{
3257	register unsigned long r3 asm("r3") = UV_ESM;
3258	register unsigned long r4 asm("r4") = kbase;
3259	register unsigned long r5 asm("r5") = fdt;
3260
3261	asm volatile("sc 2" : "+r"(r3) : "r"(r4), "r"(r5));
3262
3263	return r3;
3264}
3265
3266/*
3267 * Call the Ultravisor to transfer us to secure memory if we have an ESM blob.
3268 */
3269static void __init setup_secure_guest(unsigned long kbase, unsigned long fdt)
3270{
3271	int ret;
3272
3273	if (!prom_svm_enable)
3274		return;
3275
3276	/* Switch to secure mode. */
3277	prom_printf("Switching to secure mode.\n");
3278
3279	/*
3280	 * The ultravisor will do an integrity check of the kernel image but we
3281	 * relocated it so the check will fail. Restore the original image by
3282	 * relocating it back to the kernel virtual base address.
3283	 */
3284	relocate(KERNELBASE);
3285
3286	ret = enter_secure_mode(kbase, fdt);
3287
3288	/* Relocate the kernel again. */
3289	relocate(kbase);
3290
3291	if (ret != U_SUCCESS) {
3292		prom_printf("Returned %d from switching to secure mode.\n", ret);
3293		prom_rtas_os_term("Switch to secure mode failed.\n");
3294	}
3295}
3296#else
3297static void __init setup_secure_guest(unsigned long kbase, unsigned long fdt)
3298{
3299}
3300#endif /* CONFIG_PPC_SVM */
3301
3302/*
3303 * We enter here early on, when the Open Firmware prom is still
3304 * handling exceptions and the MMU hash table for us.
3305 */
3306
3307unsigned long __init prom_init(unsigned long r3, unsigned long r4,
3308			       unsigned long pp,
3309			       unsigned long r6, unsigned long r7,
3310			       unsigned long kbase)
3311{
3312	unsigned long hdr;
3313
3314#ifdef CONFIG_PPC32
3315	unsigned long offset = reloc_offset();
3316	reloc_got2(offset);
3317#endif
3318
3319	/*
3320	 * First zero the BSS
3321	 */
3322	memset(&__bss_start, 0, __bss_stop - __bss_start);
3323
3324	/*
3325	 * Init interface to Open Firmware, get some node references,
3326	 * like /chosen
3327	 */
3328	prom_init_client_services(pp);
3329
3330	/*
3331	 * See if this OF is old enough that we need to do explicit maps
3332	 * and other workarounds
3333	 */
3334	prom_find_mmu();
3335
3336	/*
3337	 * Init prom stdout device
3338	 */
3339	prom_init_stdout();
3340
3341	prom_printf("Preparing to boot %s", linux_banner);
3342
3343	/*
3344	 * Get default machine type. At this point, we do not differentiate
3345	 * between pSeries SMP and pSeries LPAR
3346	 */
3347	of_platform = prom_find_machine_type();
3348	prom_printf("Detected machine type: %x\n", of_platform);
3349
3350#ifndef CONFIG_NONSTATIC_KERNEL
3351	/* Bail if this is a kdump kernel. */
3352	if (PHYSICAL_START > 0)
3353		prom_panic("Error: You can't boot a kdump kernel from OF!\n");
3354#endif
3355
3356	/*
3357	 * Check for an initrd
3358	 */
3359	prom_check_initrd(r3, r4);
3360
3361	/*
3362	 * Do early parsing of command line
3363	 */
3364	early_cmdline_parse();
3365
3366#ifdef CONFIG_PPC_PSERIES
3367	/*
3368	 * On pSeries, inform the firmware about our capabilities
3369	 */
3370	if (of_platform == PLATFORM_PSERIES ||
3371	    of_platform == PLATFORM_PSERIES_LPAR)
3372		prom_send_capabilities();
3373#endif
3374
3375	/*
3376	 * Copy the CPU hold code
3377	 */
3378	if (of_platform != PLATFORM_POWERMAC)
3379		copy_and_flush(0, kbase, 0x100, 0);
3380
3381	/*
3382	 * Initialize memory management within prom_init
3383	 */
3384	prom_init_mem();
3385
3386	/*
3387	 * Determine which cpu is actually running right _now_
3388	 */
3389	prom_find_boot_cpu();
3390
3391	/*
3392	 * Initialize display devices
3393	 */
3394	prom_check_displays();
3395
3396#if defined(CONFIG_PPC64) && defined(__BIG_ENDIAN__)
3397	/*
3398	 * Initialize IOMMU (TCE tables) on pSeries. Do that before anything else
3399	 * that uses the allocator, we need to make sure we get the top of memory
3400	 * available for us here...
3401	 */
3402	if (of_platform == PLATFORM_PSERIES)
3403		prom_initialize_tce_table();
3404#endif
3405
3406	/*
3407	 * On non-powermacs, try to instantiate RTAS. PowerMacs don't
3408	 * have a usable RTAS implementation.
3409	 */
3410	if (of_platform != PLATFORM_POWERMAC)
3411		prom_instantiate_rtas();
3412
3413#ifdef CONFIG_PPC64
3414	/* instantiate sml */
3415	prom_instantiate_sml();
3416#endif
3417
3418	/*
3419	 * On non-powermacs, put all CPUs in spin-loops.
3420	 *
3421	 * PowerMacs use a different mechanism to spin CPUs
3422	 *
3423	 * (This must be done after instantiating RTAS)
3424	 */
3425	if (of_platform != PLATFORM_POWERMAC)
3426		prom_hold_cpus();
3427
3428	/*
3429	 * Fill in some infos for use by the kernel later on
3430	 */
3431	if (prom_memory_limit) {
3432		__be64 val = cpu_to_be64(prom_memory_limit);
3433		prom_setprop(prom.chosen, "/chosen", "linux,memory-limit",
3434			     &val, sizeof(val));
3435	}
3436#ifdef CONFIG_PPC64
3437	if (prom_iommu_off)
3438		prom_setprop(prom.chosen, "/chosen", "linux,iommu-off",
3439			     NULL, 0);
3440
3441	if (prom_iommu_force_on)
3442		prom_setprop(prom.chosen, "/chosen", "linux,iommu-force-on",
3443			     NULL, 0);
3444
3445	if (prom_tce_alloc_start) {
3446		prom_setprop(prom.chosen, "/chosen", "linux,tce-alloc-start",
3447			     &prom_tce_alloc_start,
3448			     sizeof(prom_tce_alloc_start));
3449		prom_setprop(prom.chosen, "/chosen", "linux,tce-alloc-end",
3450			     &prom_tce_alloc_end,
3451			     sizeof(prom_tce_alloc_end));
3452	}
3453#endif
3454
3455	/*
3456	 * Fixup any known bugs in the device-tree
3457	 */
3458	fixup_device_tree();
3459
3460	/*
3461	 * Now finally create the flattened device-tree
3462	 */
3463	prom_printf("copying OF device tree...\n");
3464	flatten_device_tree();
3465
3466	/*
3467	 * in case stdin is USB and still active on IBM machines...
3468	 * Unfortunately quiesce crashes on some powermacs if we have
3469	 * closed stdin already (in particular the powerbook 101).
3470	 */
3471	if (of_platform != PLATFORM_POWERMAC)
3472		prom_close_stdin();
3473
3474	/*
3475	 * Call OF "quiesce" method to shut down pending DMA's from
3476	 * devices etc...
3477	 */
3478	prom_printf("Quiescing Open Firmware ...\n");
3479	call_prom("quiesce", 0, 0);
3480
3481	/*
3482	 * And finally, call the kernel passing it the flattened device
3483	 * tree and NULL as r5, thus triggering the new entry point which
3484	 * is common to us and kexec
3485	 */
3486	hdr = dt_header_start;
3487
3488	prom_printf("Booting Linux via __start() @ 0x%lx ...\n", kbase);
3489	prom_debug("->dt_header_start=0x%lx\n", hdr);
3490
3491#ifdef CONFIG_PPC32
3492	reloc_got2(-offset);
3493#endif
3494
3495	/* Move to secure memory if we're supposed to be secure guests. */
3496	setup_secure_guest(kbase, hdr);
3497
3498	__start(hdr, kbase, 0, 0, 0, 0, 0);
3499
3500	return 0;
3501}
3502