xref: /openbmc/linux/arch/powerpc/kernel/prom_init.c (revision 9ab0cb30)
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 <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 <asm/prom.h>
31 #include <asm/rtas.h>
32 #include <asm/page.h>
33 #include <asm/processor.h>
34 #include <asm/irq.h>
35 #include <asm/io.h>
36 #include <asm/smp.h>
37 #include <asm/mmu.h>
38 #include <asm/iommu.h>
39 #include <asm/btext.h>
40 #include <asm/sections.h>
41 #include <asm/machdep.h>
42 #include <asm/asm-prototypes.h>
43 #include <asm/ultravisor-api.h>
44 
45 #include <linux/linux_logo.h>
46 
47 /* All of prom_init bss lives here */
48 #define __prombss __section(.bss.prominit)
49 
50 /*
51  * Eventually bump that one up
52  */
53 #define DEVTREE_CHUNK_SIZE	0x100000
54 
55 /*
56  * This is the size of the local memory reserve map that gets copied
57  * into the boot params passed to the kernel. That size is totally
58  * flexible as the kernel just reads the list until it encounters an
59  * entry with size 0, so it can be changed without breaking binary
60  * compatibility
61  */
62 #define MEM_RESERVE_MAP_SIZE	8
63 
64 /*
65  * prom_init() is called very early on, before the kernel text
66  * and data have been mapped to KERNELBASE.  At this point the code
67  * is running at whatever address it has been loaded at.
68  * On ppc32 we compile with -mrelocatable, which means that references
69  * to extern and static variables get relocated automatically.
70  * ppc64 objects are always relocatable, we just need to relocate the
71  * TOC.
72  *
73  * Because OF may have mapped I/O devices into the area starting at
74  * KERNELBASE, particularly on CHRP machines, we can't safely call
75  * OF once the kernel has been mapped to KERNELBASE.  Therefore all
76  * OF calls must be done within prom_init().
77  *
78  * ADDR is used in calls to call_prom.  The 4th and following
79  * arguments to call_prom should be 32-bit values.
80  * On ppc64, 64 bit values are truncated to 32 bits (and
81  * fortunately don't get interpreted as two arguments).
82  */
83 #define ADDR(x)		(u32)(unsigned long)(x)
84 
85 #ifdef CONFIG_PPC64
86 #define OF_WORKAROUNDS	0
87 #else
88 #define OF_WORKAROUNDS	of_workarounds
89 static int of_workarounds __prombss;
90 #endif
91 
92 #define OF_WA_CLAIM	1	/* do phys/virt claim separately, then map */
93 #define OF_WA_LONGTRAIL	2	/* work around longtrail bugs */
94 
95 #define PROM_BUG() do {						\
96         prom_printf("kernel BUG at %s line 0x%x!\n",		\
97 		    __FILE__, __LINE__);			\
98 	__builtin_trap();					\
99 } while (0)
100 
101 #ifdef DEBUG_PROM
102 #define prom_debug(x...)	prom_printf(x)
103 #else
104 #define prom_debug(x...)	do { } while (0)
105 #endif
106 
107 
108 typedef u32 prom_arg_t;
109 
110 struct prom_args {
111         __be32 service;
112         __be32 nargs;
113         __be32 nret;
114         __be32 args[10];
115 };
116 
117 struct prom_t {
118 	ihandle root;
119 	phandle chosen;
120 	int cpu;
121 	ihandle stdout;
122 	ihandle mmumap;
123 	ihandle memory;
124 };
125 
126 struct mem_map_entry {
127 	__be64	base;
128 	__be64	size;
129 };
130 
131 typedef __be32 cell_t;
132 
133 extern void __start(unsigned long r3, unsigned long r4, unsigned long r5,
134 		    unsigned long r6, unsigned long r7, unsigned long r8,
135 		    unsigned long r9);
136 
137 #ifdef CONFIG_PPC64
138 extern int enter_prom(struct prom_args *args, unsigned long entry);
139 #else
140 static inline int enter_prom(struct prom_args *args, unsigned long entry)
141 {
142 	return ((int (*)(struct prom_args *))entry)(args);
143 }
144 #endif
145 
146 extern void copy_and_flush(unsigned long dest, unsigned long src,
147 			   unsigned long size, unsigned long offset);
148 
149 /* prom structure */
150 static struct prom_t __prombss prom;
151 
152 static unsigned long __prombss prom_entry;
153 
154 static char __prombss of_stdout_device[256];
155 static char __prombss prom_scratch[256];
156 
157 static unsigned long __prombss dt_header_start;
158 static unsigned long __prombss dt_struct_start, dt_struct_end;
159 static unsigned long __prombss dt_string_start, dt_string_end;
160 
161 static unsigned long __prombss prom_initrd_start, prom_initrd_end;
162 
163 #ifdef CONFIG_PPC64
164 static int __prombss prom_iommu_force_on;
165 static int __prombss prom_iommu_off;
166 static unsigned long __prombss prom_tce_alloc_start;
167 static unsigned long __prombss prom_tce_alloc_end;
168 #endif
169 
170 #ifdef CONFIG_PPC_PSERIES
171 static bool __prombss prom_radix_disable;
172 static bool __prombss prom_xive_disable;
173 #endif
174 
175 #ifdef CONFIG_PPC_SVM
176 static bool __prombss prom_svm_enable;
177 #endif
178 
179 struct platform_support {
180 	bool hash_mmu;
181 	bool radix_mmu;
182 	bool radix_gtse;
183 	bool xive;
184 };
185 
186 /* Platforms codes are now obsolete in the kernel. Now only used within this
187  * file and ultimately gone too. Feel free to change them if you need, they
188  * are not shared with anything outside of this file anymore
189  */
190 #define PLATFORM_PSERIES	0x0100
191 #define PLATFORM_PSERIES_LPAR	0x0101
192 #define PLATFORM_LPAR		0x0001
193 #define PLATFORM_POWERMAC	0x0400
194 #define PLATFORM_GENERIC	0x0500
195 
196 static int __prombss of_platform;
197 
198 static char __prombss prom_cmd_line[COMMAND_LINE_SIZE];
199 
200 static unsigned long __prombss prom_memory_limit;
201 
202 static unsigned long __prombss alloc_top;
203 static unsigned long __prombss alloc_top_high;
204 static unsigned long __prombss alloc_bottom;
205 static unsigned long __prombss rmo_top;
206 static unsigned long __prombss ram_top;
207 
208 static struct mem_map_entry __prombss mem_reserve_map[MEM_RESERVE_MAP_SIZE];
209 static int __prombss mem_reserve_cnt;
210 
211 static cell_t __prombss regbuf[1024];
212 
213 static bool  __prombss rtas_has_query_cpu_stopped;
214 
215 
216 /*
217  * Error results ... some OF calls will return "-1" on error, some
218  * will return 0, some will return either. To simplify, here are
219  * macros to use with any ihandle or phandle return value to check if
220  * it is valid
221  */
222 
223 #define PROM_ERROR		(-1u)
224 #define PHANDLE_VALID(p)	((p) != 0 && (p) != PROM_ERROR)
225 #define IHANDLE_VALID(i)	((i) != 0 && (i) != PROM_ERROR)
226 
227 /* Copied from lib/string.c and lib/kstrtox.c */
228 
229 static int __init prom_strcmp(const char *cs, const char *ct)
230 {
231 	unsigned char c1, c2;
232 
233 	while (1) {
234 		c1 = *cs++;
235 		c2 = *ct++;
236 		if (c1 != c2)
237 			return c1 < c2 ? -1 : 1;
238 		if (!c1)
239 			break;
240 	}
241 	return 0;
242 }
243 
244 static char __init *prom_strcpy(char *dest, const char *src)
245 {
246 	char *tmp = dest;
247 
248 	while ((*dest++ = *src++) != '\0')
249 		/* nothing */;
250 	return tmp;
251 }
252 
253 static int __init prom_strncmp(const char *cs, const char *ct, size_t count)
254 {
255 	unsigned char c1, c2;
256 
257 	while (count) {
258 		c1 = *cs++;
259 		c2 = *ct++;
260 		if (c1 != c2)
261 			return c1 < c2 ? -1 : 1;
262 		if (!c1)
263 			break;
264 		count--;
265 	}
266 	return 0;
267 }
268 
269 static size_t __init prom_strlen(const char *s)
270 {
271 	const char *sc;
272 
273 	for (sc = s; *sc != '\0'; ++sc)
274 		/* nothing */;
275 	return sc - s;
276 }
277 
278 static int __init prom_memcmp(const void *cs, const void *ct, size_t count)
279 {
280 	const unsigned char *su1, *su2;
281 	int res = 0;
282 
283 	for (su1 = cs, su2 = ct; 0 < count; ++su1, ++su2, count--)
284 		if ((res = *su1 - *su2) != 0)
285 			break;
286 	return res;
287 }
288 
289 static char __init *prom_strstr(const char *s1, const char *s2)
290 {
291 	size_t l1, l2;
292 
293 	l2 = prom_strlen(s2);
294 	if (!l2)
295 		return (char *)s1;
296 	l1 = prom_strlen(s1);
297 	while (l1 >= l2) {
298 		l1--;
299 		if (!prom_memcmp(s1, s2, l2))
300 			return (char *)s1;
301 		s1++;
302 	}
303 	return NULL;
304 }
305 
306 static size_t __init prom_strlcat(char *dest, const char *src, size_t count)
307 {
308 	size_t dsize = prom_strlen(dest);
309 	size_t len = prom_strlen(src);
310 	size_t res = dsize + len;
311 
312 	/* This would be a bug */
313 	if (dsize >= count)
314 		return count;
315 
316 	dest += dsize;
317 	count -= dsize;
318 	if (len >= count)
319 		len = count-1;
320 	memcpy(dest, src, len);
321 	dest[len] = 0;
322 	return res;
323 
324 }
325 
326 #ifdef CONFIG_PPC_PSERIES
327 static int __init prom_strtobool(const char *s, bool *res)
328 {
329 	if (!s)
330 		return -EINVAL;
331 
332 	switch (s[0]) {
333 	case 'y':
334 	case 'Y':
335 	case '1':
336 		*res = true;
337 		return 0;
338 	case 'n':
339 	case 'N':
340 	case '0':
341 		*res = false;
342 		return 0;
343 	case 'o':
344 	case 'O':
345 		switch (s[1]) {
346 		case 'n':
347 		case 'N':
348 			*res = true;
349 			return 0;
350 		case 'f':
351 		case 'F':
352 			*res = false;
353 			return 0;
354 		default:
355 			break;
356 		}
357 	default:
358 		break;
359 	}
360 
361 	return -EINVAL;
362 }
363 #endif
364 
365 /* This is the one and *ONLY* place where we actually call open
366  * firmware.
367  */
368 
369 static int __init call_prom(const char *service, int nargs, int nret, ...)
370 {
371 	int i;
372 	struct prom_args args;
373 	va_list list;
374 
375 	args.service = cpu_to_be32(ADDR(service));
376 	args.nargs = cpu_to_be32(nargs);
377 	args.nret = cpu_to_be32(nret);
378 
379 	va_start(list, nret);
380 	for (i = 0; i < nargs; i++)
381 		args.args[i] = cpu_to_be32(va_arg(list, prom_arg_t));
382 	va_end(list);
383 
384 	for (i = 0; i < nret; i++)
385 		args.args[nargs+i] = 0;
386 
387 	if (enter_prom(&args, prom_entry) < 0)
388 		return PROM_ERROR;
389 
390 	return (nret > 0) ? be32_to_cpu(args.args[nargs]) : 0;
391 }
392 
393 static int __init call_prom_ret(const char *service, int nargs, int nret,
394 				prom_arg_t *rets, ...)
395 {
396 	int i;
397 	struct prom_args args;
398 	va_list list;
399 
400 	args.service = cpu_to_be32(ADDR(service));
401 	args.nargs = cpu_to_be32(nargs);
402 	args.nret = cpu_to_be32(nret);
403 
404 	va_start(list, rets);
405 	for (i = 0; i < nargs; i++)
406 		args.args[i] = cpu_to_be32(va_arg(list, prom_arg_t));
407 	va_end(list);
408 
409 	for (i = 0; i < nret; i++)
410 		args.args[nargs+i] = 0;
411 
412 	if (enter_prom(&args, prom_entry) < 0)
413 		return PROM_ERROR;
414 
415 	if (rets != NULL)
416 		for (i = 1; i < nret; ++i)
417 			rets[i-1] = be32_to_cpu(args.args[nargs+i]);
418 
419 	return (nret > 0) ? be32_to_cpu(args.args[nargs]) : 0;
420 }
421 
422 
423 static void __init prom_print(const char *msg)
424 {
425 	const char *p, *q;
426 
427 	if (prom.stdout == 0)
428 		return;
429 
430 	for (p = msg; *p != 0; p = q) {
431 		for (q = p; *q != 0 && *q != '\n'; ++q)
432 			;
433 		if (q > p)
434 			call_prom("write", 3, 1, prom.stdout, p, q - p);
435 		if (*q == 0)
436 			break;
437 		++q;
438 		call_prom("write", 3, 1, prom.stdout, ADDR("\r\n"), 2);
439 	}
440 }
441 
442 
443 /*
444  * Both prom_print_hex & prom_print_dec takes an unsigned long as input so that
445  * we do not need __udivdi3 or __umoddi3 on 32bits.
446  */
447 static void __init prom_print_hex(unsigned long val)
448 {
449 	int i, nibbles = sizeof(val)*2;
450 	char buf[sizeof(val)*2+1];
451 
452 	for (i = nibbles-1;  i >= 0;  i--) {
453 		buf[i] = (val & 0xf) + '0';
454 		if (buf[i] > '9')
455 			buf[i] += ('a'-'0'-10);
456 		val >>= 4;
457 	}
458 	buf[nibbles] = '\0';
459 	call_prom("write", 3, 1, prom.stdout, buf, nibbles);
460 }
461 
462 /* max number of decimal digits in an unsigned long */
463 #define UL_DIGITS 21
464 static void __init prom_print_dec(unsigned long val)
465 {
466 	int i, size;
467 	char buf[UL_DIGITS+1];
468 
469 	for (i = UL_DIGITS-1; i >= 0;  i--) {
470 		buf[i] = (val % 10) + '0';
471 		val = val/10;
472 		if (val == 0)
473 			break;
474 	}
475 	/* shift stuff down */
476 	size = UL_DIGITS - i;
477 	call_prom("write", 3, 1, prom.stdout, buf+i, size);
478 }
479 
480 __printf(1, 2)
481 static void __init prom_printf(const char *format, ...)
482 {
483 	const char *p, *q, *s;
484 	va_list args;
485 	unsigned long v;
486 	long vs;
487 	int n = 0;
488 
489 	va_start(args, format);
490 	for (p = format; *p != 0; p = q) {
491 		for (q = p; *q != 0 && *q != '\n' && *q != '%'; ++q)
492 			;
493 		if (q > p)
494 			call_prom("write", 3, 1, prom.stdout, p, q - p);
495 		if (*q == 0)
496 			break;
497 		if (*q == '\n') {
498 			++q;
499 			call_prom("write", 3, 1, prom.stdout,
500 				  ADDR("\r\n"), 2);
501 			continue;
502 		}
503 		++q;
504 		if (*q == 0)
505 			break;
506 		while (*q == 'l') {
507 			++q;
508 			++n;
509 		}
510 		switch (*q) {
511 		case 's':
512 			++q;
513 			s = va_arg(args, const char *);
514 			prom_print(s);
515 			break;
516 		case 'x':
517 			++q;
518 			switch (n) {
519 			case 0:
520 				v = va_arg(args, unsigned int);
521 				break;
522 			case 1:
523 				v = va_arg(args, unsigned long);
524 				break;
525 			case 2:
526 			default:
527 				v = va_arg(args, unsigned long long);
528 				break;
529 			}
530 			prom_print_hex(v);
531 			break;
532 		case 'u':
533 			++q;
534 			switch (n) {
535 			case 0:
536 				v = va_arg(args, unsigned int);
537 				break;
538 			case 1:
539 				v = va_arg(args, unsigned long);
540 				break;
541 			case 2:
542 			default:
543 				v = va_arg(args, unsigned long long);
544 				break;
545 			}
546 			prom_print_dec(v);
547 			break;
548 		case 'd':
549 			++q;
550 			switch (n) {
551 			case 0:
552 				vs = va_arg(args, int);
553 				break;
554 			case 1:
555 				vs = va_arg(args, long);
556 				break;
557 			case 2:
558 			default:
559 				vs = va_arg(args, long long);
560 				break;
561 			}
562 			if (vs < 0) {
563 				prom_print("-");
564 				vs = -vs;
565 			}
566 			prom_print_dec(vs);
567 			break;
568 		}
569 	}
570 	va_end(args);
571 }
572 
573 
574 static unsigned int __init prom_claim(unsigned long virt, unsigned long size,
575 				unsigned long align)
576 {
577 
578 	if (align == 0 && (OF_WORKAROUNDS & OF_WA_CLAIM)) {
579 		/*
580 		 * Old OF requires we claim physical and virtual separately
581 		 * and then map explicitly (assuming virtual mode)
582 		 */
583 		int ret;
584 		prom_arg_t result;
585 
586 		ret = call_prom_ret("call-method", 5, 2, &result,
587 				    ADDR("claim"), prom.memory,
588 				    align, size, virt);
589 		if (ret != 0 || result == -1)
590 			return -1;
591 		ret = call_prom_ret("call-method", 5, 2, &result,
592 				    ADDR("claim"), prom.mmumap,
593 				    align, size, virt);
594 		if (ret != 0) {
595 			call_prom("call-method", 4, 1, ADDR("release"),
596 				  prom.memory, size, virt);
597 			return -1;
598 		}
599 		/* the 0x12 is M (coherence) + PP == read/write */
600 		call_prom("call-method", 6, 1,
601 			  ADDR("map"), prom.mmumap, 0x12, size, virt, virt);
602 		return virt;
603 	}
604 	return call_prom("claim", 3, 1, (prom_arg_t)virt, (prom_arg_t)size,
605 			 (prom_arg_t)align);
606 }
607 
608 static void __init __attribute__((noreturn)) prom_panic(const char *reason)
609 {
610 	prom_print(reason);
611 	/* Do not call exit because it clears the screen on pmac
612 	 * it also causes some sort of double-fault on early pmacs */
613 	if (of_platform == PLATFORM_POWERMAC)
614 		asm("trap\n");
615 
616 	/* ToDo: should put up an SRC here on pSeries */
617 	call_prom("exit", 0, 0);
618 
619 	for (;;)			/* should never get here */
620 		;
621 }
622 
623 
624 static int __init prom_next_node(phandle *nodep)
625 {
626 	phandle node;
627 
628 	if ((node = *nodep) != 0
629 	    && (*nodep = call_prom("child", 1, 1, node)) != 0)
630 		return 1;
631 	if ((*nodep = call_prom("peer", 1, 1, node)) != 0)
632 		return 1;
633 	for (;;) {
634 		if ((node = call_prom("parent", 1, 1, node)) == 0)
635 			return 0;
636 		if ((*nodep = call_prom("peer", 1, 1, node)) != 0)
637 			return 1;
638 	}
639 }
640 
641 static inline int __init prom_getprop(phandle node, const char *pname,
642 				      void *value, size_t valuelen)
643 {
644 	return call_prom("getprop", 4, 1, node, ADDR(pname),
645 			 (u32)(unsigned long) value, (u32) valuelen);
646 }
647 
648 static inline int __init prom_getproplen(phandle node, const char *pname)
649 {
650 	return call_prom("getproplen", 2, 1, node, ADDR(pname));
651 }
652 
653 static void add_string(char **str, const char *q)
654 {
655 	char *p = *str;
656 
657 	while (*q)
658 		*p++ = *q++;
659 	*p++ = ' ';
660 	*str = p;
661 }
662 
663 static char *tohex(unsigned int x)
664 {
665 	static const char digits[] __initconst = "0123456789abcdef";
666 	static char result[9] __prombss;
667 	int i;
668 
669 	result[8] = 0;
670 	i = 8;
671 	do {
672 		--i;
673 		result[i] = digits[x & 0xf];
674 		x >>= 4;
675 	} while (x != 0 && i > 0);
676 	return &result[i];
677 }
678 
679 static int __init prom_setprop(phandle node, const char *nodename,
680 			       const char *pname, void *value, size_t valuelen)
681 {
682 	char cmd[256], *p;
683 
684 	if (!(OF_WORKAROUNDS & OF_WA_LONGTRAIL))
685 		return call_prom("setprop", 4, 1, node, ADDR(pname),
686 				 (u32)(unsigned long) value, (u32) valuelen);
687 
688 	/* gah... setprop doesn't work on longtrail, have to use interpret */
689 	p = cmd;
690 	add_string(&p, "dev");
691 	add_string(&p, nodename);
692 	add_string(&p, tohex((u32)(unsigned long) value));
693 	add_string(&p, tohex(valuelen));
694 	add_string(&p, tohex(ADDR(pname)));
695 	add_string(&p, tohex(prom_strlen(pname)));
696 	add_string(&p, "property");
697 	*p = 0;
698 	return call_prom("interpret", 1, 1, (u32)(unsigned long) cmd);
699 }
700 
701 /* We can't use the standard versions because of relocation headaches. */
702 #define isxdigit(c)	(('0' <= (c) && (c) <= '9') \
703 			 || ('a' <= (c) && (c) <= 'f') \
704 			 || ('A' <= (c) && (c) <= 'F'))
705 
706 #define isdigit(c)	('0' <= (c) && (c) <= '9')
707 #define islower(c)	('a' <= (c) && (c) <= 'z')
708 #define toupper(c)	(islower(c) ? ((c) - 'a' + 'A') : (c))
709 
710 static unsigned long prom_strtoul(const char *cp, const char **endp)
711 {
712 	unsigned long result = 0, base = 10, value;
713 
714 	if (*cp == '0') {
715 		base = 8;
716 		cp++;
717 		if (toupper(*cp) == 'X') {
718 			cp++;
719 			base = 16;
720 		}
721 	}
722 
723 	while (isxdigit(*cp) &&
724 	       (value = isdigit(*cp) ? *cp - '0' : toupper(*cp) - 'A' + 10) < base) {
725 		result = result * base + value;
726 		cp++;
727 	}
728 
729 	if (endp)
730 		*endp = cp;
731 
732 	return result;
733 }
734 
735 static unsigned long prom_memparse(const char *ptr, const char **retptr)
736 {
737 	unsigned long ret = prom_strtoul(ptr, retptr);
738 	int shift = 0;
739 
740 	/*
741 	 * We can't use a switch here because GCC *may* generate a
742 	 * jump table which won't work, because we're not running at
743 	 * the address we're linked at.
744 	 */
745 	if ('G' == **retptr || 'g' == **retptr)
746 		shift = 30;
747 
748 	if ('M' == **retptr || 'm' == **retptr)
749 		shift = 20;
750 
751 	if ('K' == **retptr || 'k' == **retptr)
752 		shift = 10;
753 
754 	if (shift) {
755 		ret <<= shift;
756 		(*retptr)++;
757 	}
758 
759 	return ret;
760 }
761 
762 /*
763  * Early parsing of the command line passed to the kernel, used for
764  * "mem=x" and the options that affect the iommu
765  */
766 static void __init early_cmdline_parse(void)
767 {
768 	const char *opt;
769 
770 	char *p;
771 	int l = 0;
772 
773 	prom_cmd_line[0] = 0;
774 	p = prom_cmd_line;
775 
776 	if (!IS_ENABLED(CONFIG_CMDLINE_FORCE) && (long)prom.chosen > 0)
777 		l = prom_getprop(prom.chosen, "bootargs", p, COMMAND_LINE_SIZE-1);
778 
779 	if (IS_ENABLED(CONFIG_CMDLINE_EXTEND) || l <= 0 || p[0] == '\0')
780 		prom_strlcat(prom_cmd_line, " " CONFIG_CMDLINE,
781 			     sizeof(prom_cmd_line));
782 
783 	prom_printf("command line: %s\n", prom_cmd_line);
784 
785 #ifdef CONFIG_PPC64
786 	opt = prom_strstr(prom_cmd_line, "iommu=");
787 	if (opt) {
788 		prom_printf("iommu opt is: %s\n", opt);
789 		opt += 6;
790 		while (*opt && *opt == ' ')
791 			opt++;
792 		if (!prom_strncmp(opt, "off", 3))
793 			prom_iommu_off = 1;
794 		else if (!prom_strncmp(opt, "force", 5))
795 			prom_iommu_force_on = 1;
796 	}
797 #endif
798 	opt = prom_strstr(prom_cmd_line, "mem=");
799 	if (opt) {
800 		opt += 4;
801 		prom_memory_limit = prom_memparse(opt, (const char **)&opt);
802 #ifdef CONFIG_PPC64
803 		/* Align to 16 MB == size of ppc64 large page */
804 		prom_memory_limit = ALIGN(prom_memory_limit, 0x1000000);
805 #endif
806 	}
807 
808 #ifdef CONFIG_PPC_PSERIES
809 	prom_radix_disable = !IS_ENABLED(CONFIG_PPC_RADIX_MMU_DEFAULT);
810 	opt = prom_strstr(prom_cmd_line, "disable_radix");
811 	if (opt) {
812 		opt += 13;
813 		if (*opt && *opt == '=') {
814 			bool val;
815 
816 			if (prom_strtobool(++opt, &val))
817 				prom_radix_disable = false;
818 			else
819 				prom_radix_disable = val;
820 		} else
821 			prom_radix_disable = true;
822 	}
823 	if (prom_radix_disable)
824 		prom_debug("Radix disabled from cmdline\n");
825 
826 	opt = prom_strstr(prom_cmd_line, "xive=off");
827 	if (opt) {
828 		prom_xive_disable = true;
829 		prom_debug("XIVE disabled from cmdline\n");
830 	}
831 #endif /* CONFIG_PPC_PSERIES */
832 
833 #ifdef CONFIG_PPC_SVM
834 	opt = prom_strstr(prom_cmd_line, "svm=");
835 	if (opt) {
836 		bool val;
837 
838 		opt += sizeof("svm=") - 1;
839 		if (!prom_strtobool(opt, &val))
840 			prom_svm_enable = val;
841 	}
842 #endif /* CONFIG_PPC_SVM */
843 }
844 
845 #ifdef CONFIG_PPC_PSERIES
846 /*
847  * The architecture vector has an array of PVR mask/value pairs,
848  * followed by # option vectors - 1, followed by the option vectors.
849  *
850  * See prom.h for the definition of the bits specified in the
851  * architecture vector.
852  */
853 
854 /* Firmware expects the value to be n - 1, where n is the # of vectors */
855 #define NUM_VECTORS(n)		((n) - 1)
856 
857 /*
858  * Firmware expects 1 + n - 2, where n is the length of the option vector in
859  * bytes. The 1 accounts for the length byte itself, the - 2 .. ?
860  */
861 #define VECTOR_LENGTH(n)	(1 + (n) - 2)
862 
863 struct option_vector1 {
864 	u8 byte1;
865 	u8 arch_versions;
866 	u8 arch_versions3;
867 } __packed;
868 
869 struct option_vector2 {
870 	u8 byte1;
871 	__be16 reserved;
872 	__be32 real_base;
873 	__be32 real_size;
874 	__be32 virt_base;
875 	__be32 virt_size;
876 	__be32 load_base;
877 	__be32 min_rma;
878 	__be32 min_load;
879 	u8 min_rma_percent;
880 	u8 max_pft_size;
881 } __packed;
882 
883 struct option_vector3 {
884 	u8 byte1;
885 	u8 byte2;
886 } __packed;
887 
888 struct option_vector4 {
889 	u8 byte1;
890 	u8 min_vp_cap;
891 } __packed;
892 
893 struct option_vector5 {
894 	u8 byte1;
895 	u8 byte2;
896 	u8 byte3;
897 	u8 cmo;
898 	u8 associativity;
899 	u8 bin_opts;
900 	u8 micro_checkpoint;
901 	u8 reserved0;
902 	__be32 max_cpus;
903 	__be16 papr_level;
904 	__be16 reserved1;
905 	u8 platform_facilities;
906 	u8 reserved2;
907 	__be16 reserved3;
908 	u8 subprocessors;
909 	u8 byte22;
910 	u8 intarch;
911 	u8 mmu;
912 	u8 hash_ext;
913 	u8 radix_ext;
914 } __packed;
915 
916 struct option_vector6 {
917 	u8 reserved;
918 	u8 secondary_pteg;
919 	u8 os_name;
920 } __packed;
921 
922 struct ibm_arch_vec {
923 	struct { u32 mask, val; } pvrs[14];
924 
925 	u8 num_vectors;
926 
927 	u8 vec1_len;
928 	struct option_vector1 vec1;
929 
930 	u8 vec2_len;
931 	struct option_vector2 vec2;
932 
933 	u8 vec3_len;
934 	struct option_vector3 vec3;
935 
936 	u8 vec4_len;
937 	struct option_vector4 vec4;
938 
939 	u8 vec5_len;
940 	struct option_vector5 vec5;
941 
942 	u8 vec6_len;
943 	struct option_vector6 vec6;
944 } __packed;
945 
946 static const struct ibm_arch_vec ibm_architecture_vec_template __initconst = {
947 	.pvrs = {
948 		{
949 			.mask = cpu_to_be32(0xfffe0000), /* POWER5/POWER5+ */
950 			.val  = cpu_to_be32(0x003a0000),
951 		},
952 		{
953 			.mask = cpu_to_be32(0xffff0000), /* POWER6 */
954 			.val  = cpu_to_be32(0x003e0000),
955 		},
956 		{
957 			.mask = cpu_to_be32(0xffff0000), /* POWER7 */
958 			.val  = cpu_to_be32(0x003f0000),
959 		},
960 		{
961 			.mask = cpu_to_be32(0xffff0000), /* POWER8E */
962 			.val  = cpu_to_be32(0x004b0000),
963 		},
964 		{
965 			.mask = cpu_to_be32(0xffff0000), /* POWER8NVL */
966 			.val  = cpu_to_be32(0x004c0000),
967 		},
968 		{
969 			.mask = cpu_to_be32(0xffff0000), /* POWER8 */
970 			.val  = cpu_to_be32(0x004d0000),
971 		},
972 		{
973 			.mask = cpu_to_be32(0xffff0000), /* POWER9 */
974 			.val  = cpu_to_be32(0x004e0000),
975 		},
976 		{
977 			.mask = cpu_to_be32(0xffff0000), /* POWER10 */
978 			.val  = cpu_to_be32(0x00800000),
979 		},
980 		{
981 			.mask = cpu_to_be32(0xffffffff), /* all 3.1-compliant */
982 			.val  = cpu_to_be32(0x0f000006),
983 		},
984 		{
985 			.mask = cpu_to_be32(0xffffffff), /* all 3.00-compliant */
986 			.val  = cpu_to_be32(0x0f000005),
987 		},
988 		{
989 			.mask = cpu_to_be32(0xffffffff), /* all 2.07-compliant */
990 			.val  = cpu_to_be32(0x0f000004),
991 		},
992 		{
993 			.mask = cpu_to_be32(0xffffffff), /* all 2.06-compliant */
994 			.val  = cpu_to_be32(0x0f000003),
995 		},
996 		{
997 			.mask = cpu_to_be32(0xffffffff), /* all 2.05-compliant */
998 			.val  = cpu_to_be32(0x0f000002),
999 		},
1000 		{
1001 			.mask = cpu_to_be32(0xfffffffe), /* all 2.04-compliant and earlier */
1002 			.val  = cpu_to_be32(0x0f000001),
1003 		},
1004 	},
1005 
1006 	.num_vectors = NUM_VECTORS(6),
1007 
1008 	.vec1_len = VECTOR_LENGTH(sizeof(struct option_vector1)),
1009 	.vec1 = {
1010 		.byte1 = 0,
1011 		.arch_versions = OV1_PPC_2_00 | OV1_PPC_2_01 | OV1_PPC_2_02 | OV1_PPC_2_03 |
1012 				 OV1_PPC_2_04 | OV1_PPC_2_05 | OV1_PPC_2_06 | OV1_PPC_2_07,
1013 		.arch_versions3 = OV1_PPC_3_00 | OV1_PPC_3_1,
1014 	},
1015 
1016 	.vec2_len = VECTOR_LENGTH(sizeof(struct option_vector2)),
1017 	/* option vector 2: Open Firmware options supported */
1018 	.vec2 = {
1019 		.byte1 = OV2_REAL_MODE,
1020 		.reserved = 0,
1021 		.real_base = cpu_to_be32(0xffffffff),
1022 		.real_size = cpu_to_be32(0xffffffff),
1023 		.virt_base = cpu_to_be32(0xffffffff),
1024 		.virt_size = cpu_to_be32(0xffffffff),
1025 		.load_base = cpu_to_be32(0xffffffff),
1026 		.min_rma = cpu_to_be32(512),		/* 512MB min RMA */
1027 		.min_load = cpu_to_be32(0xffffffff),	/* full client load */
1028 		.min_rma_percent = 0,	/* min RMA percentage of total RAM */
1029 		.max_pft_size = 48,	/* max log_2(hash table size) */
1030 	},
1031 
1032 	.vec3_len = VECTOR_LENGTH(sizeof(struct option_vector3)),
1033 	/* option vector 3: processor options supported */
1034 	.vec3 = {
1035 		.byte1 = 0,			/* don't ignore, don't halt */
1036 		.byte2 = OV3_FP | OV3_VMX | OV3_DFP,
1037 	},
1038 
1039 	.vec4_len = VECTOR_LENGTH(sizeof(struct option_vector4)),
1040 	/* option vector 4: IBM PAPR implementation */
1041 	.vec4 = {
1042 		.byte1 = 0,			/* don't halt */
1043 		.min_vp_cap = OV4_MIN_ENT_CAP,	/* minimum VP entitled capacity */
1044 	},
1045 
1046 	.vec5_len = VECTOR_LENGTH(sizeof(struct option_vector5)),
1047 	/* option vector 5: PAPR/OF options */
1048 	.vec5 = {
1049 		.byte1 = 0,				/* don't ignore, don't halt */
1050 		.byte2 = OV5_FEAT(OV5_LPAR) | OV5_FEAT(OV5_SPLPAR) | OV5_FEAT(OV5_LARGE_PAGES) |
1051 		OV5_FEAT(OV5_DRCONF_MEMORY) | OV5_FEAT(OV5_DONATE_DEDICATE_CPU) |
1052 #ifdef CONFIG_PCI_MSI
1053 		/* PCIe/MSI support.  Without MSI full PCIe is not supported */
1054 		OV5_FEAT(OV5_MSI),
1055 #else
1056 		0,
1057 #endif
1058 		.byte3 = 0,
1059 		.cmo =
1060 #ifdef CONFIG_PPC_SMLPAR
1061 		OV5_FEAT(OV5_CMO) | OV5_FEAT(OV5_XCMO),
1062 #else
1063 		0,
1064 #endif
1065 		.associativity = OV5_FEAT(OV5_TYPE1_AFFINITY) | OV5_FEAT(OV5_PRRN),
1066 		.bin_opts = OV5_FEAT(OV5_RESIZE_HPT) | OV5_FEAT(OV5_HP_EVT),
1067 		.micro_checkpoint = 0,
1068 		.reserved0 = 0,
1069 		.max_cpus = cpu_to_be32(NR_CPUS),	/* number of cores supported */
1070 		.papr_level = 0,
1071 		.reserved1 = 0,
1072 		.platform_facilities = OV5_FEAT(OV5_PFO_HW_RNG) | OV5_FEAT(OV5_PFO_HW_ENCR) | OV5_FEAT(OV5_PFO_HW_842),
1073 		.reserved2 = 0,
1074 		.reserved3 = 0,
1075 		.subprocessors = 1,
1076 		.byte22 = OV5_FEAT(OV5_DRMEM_V2) | OV5_FEAT(OV5_DRC_INFO),
1077 		.intarch = 0,
1078 		.mmu = 0,
1079 		.hash_ext = 0,
1080 		.radix_ext = 0,
1081 	},
1082 
1083 	/* option vector 6: IBM PAPR hints */
1084 	.vec6_len = VECTOR_LENGTH(sizeof(struct option_vector6)),
1085 	.vec6 = {
1086 		.reserved = 0,
1087 		.secondary_pteg = 0,
1088 		.os_name = OV6_LINUX,
1089 	},
1090 };
1091 
1092 static struct ibm_arch_vec __prombss ibm_architecture_vec  ____cacheline_aligned;
1093 
1094 /* Old method - ELF header with PT_NOTE sections only works on BE */
1095 #ifdef __BIG_ENDIAN__
1096 static const struct fake_elf {
1097 	Elf32_Ehdr	elfhdr;
1098 	Elf32_Phdr	phdr[2];
1099 	struct chrpnote {
1100 		u32	namesz;
1101 		u32	descsz;
1102 		u32	type;
1103 		char	name[8];	/* "PowerPC" */
1104 		struct chrpdesc {
1105 			u32	real_mode;
1106 			u32	real_base;
1107 			u32	real_size;
1108 			u32	virt_base;
1109 			u32	virt_size;
1110 			u32	load_base;
1111 		} chrpdesc;
1112 	} chrpnote;
1113 	struct rpanote {
1114 		u32	namesz;
1115 		u32	descsz;
1116 		u32	type;
1117 		char	name[24];	/* "IBM,RPA-Client-Config" */
1118 		struct rpadesc {
1119 			u32	lpar_affinity;
1120 			u32	min_rmo_size;
1121 			u32	min_rmo_percent;
1122 			u32	max_pft_size;
1123 			u32	splpar;
1124 			u32	min_load;
1125 			u32	new_mem_def;
1126 			u32	ignore_me;
1127 		} rpadesc;
1128 	} rpanote;
1129 } fake_elf __initconst = {
1130 	.elfhdr = {
1131 		.e_ident = { 0x7f, 'E', 'L', 'F',
1132 			     ELFCLASS32, ELFDATA2MSB, EV_CURRENT },
1133 		.e_type = ET_EXEC,	/* yeah right */
1134 		.e_machine = EM_PPC,
1135 		.e_version = EV_CURRENT,
1136 		.e_phoff = offsetof(struct fake_elf, phdr),
1137 		.e_phentsize = sizeof(Elf32_Phdr),
1138 		.e_phnum = 2
1139 	},
1140 	.phdr = {
1141 		[0] = {
1142 			.p_type = PT_NOTE,
1143 			.p_offset = offsetof(struct fake_elf, chrpnote),
1144 			.p_filesz = sizeof(struct chrpnote)
1145 		}, [1] = {
1146 			.p_type = PT_NOTE,
1147 			.p_offset = offsetof(struct fake_elf, rpanote),
1148 			.p_filesz = sizeof(struct rpanote)
1149 		}
1150 	},
1151 	.chrpnote = {
1152 		.namesz = sizeof("PowerPC"),
1153 		.descsz = sizeof(struct chrpdesc),
1154 		.type = 0x1275,
1155 		.name = "PowerPC",
1156 		.chrpdesc = {
1157 			.real_mode = ~0U,	/* ~0 means "don't care" */
1158 			.real_base = ~0U,
1159 			.real_size = ~0U,
1160 			.virt_base = ~0U,
1161 			.virt_size = ~0U,
1162 			.load_base = ~0U
1163 		},
1164 	},
1165 	.rpanote = {
1166 		.namesz = sizeof("IBM,RPA-Client-Config"),
1167 		.descsz = sizeof(struct rpadesc),
1168 		.type = 0x12759999,
1169 		.name = "IBM,RPA-Client-Config",
1170 		.rpadesc = {
1171 			.lpar_affinity = 0,
1172 			.min_rmo_size = 64,	/* in megabytes */
1173 			.min_rmo_percent = 0,
1174 			.max_pft_size = 48,	/* 2^48 bytes max PFT size */
1175 			.splpar = 1,
1176 			.min_load = ~0U,
1177 			.new_mem_def = 0
1178 		}
1179 	}
1180 };
1181 #endif /* __BIG_ENDIAN__ */
1182 
1183 static int __init prom_count_smt_threads(void)
1184 {
1185 	phandle node;
1186 	char type[64];
1187 	unsigned int plen;
1188 
1189 	/* Pick up th first CPU node we can find */
1190 	for (node = 0; prom_next_node(&node); ) {
1191 		type[0] = 0;
1192 		prom_getprop(node, "device_type", type, sizeof(type));
1193 
1194 		if (prom_strcmp(type, "cpu"))
1195 			continue;
1196 		/*
1197 		 * There is an entry for each smt thread, each entry being
1198 		 * 4 bytes long.  All cpus should have the same number of
1199 		 * smt threads, so return after finding the first.
1200 		 */
1201 		plen = prom_getproplen(node, "ibm,ppc-interrupt-server#s");
1202 		if (plen == PROM_ERROR)
1203 			break;
1204 		plen >>= 2;
1205 		prom_debug("Found %lu smt threads per core\n", (unsigned long)plen);
1206 
1207 		/* Sanity check */
1208 		if (plen < 1 || plen > 64) {
1209 			prom_printf("Threads per core %lu out of bounds, assuming 1\n",
1210 				    (unsigned long)plen);
1211 			return 1;
1212 		}
1213 		return plen;
1214 	}
1215 	prom_debug("No threads found, assuming 1 per core\n");
1216 
1217 	return 1;
1218 
1219 }
1220 
1221 static void __init prom_parse_mmu_model(u8 val,
1222 					struct platform_support *support)
1223 {
1224 	switch (val) {
1225 	case OV5_FEAT(OV5_MMU_DYNAMIC):
1226 	case OV5_FEAT(OV5_MMU_EITHER): /* Either Available */
1227 		prom_debug("MMU - either supported\n");
1228 		support->radix_mmu = !prom_radix_disable;
1229 		support->hash_mmu = true;
1230 		break;
1231 	case OV5_FEAT(OV5_MMU_RADIX): /* Only Radix */
1232 		prom_debug("MMU - radix only\n");
1233 		if (prom_radix_disable) {
1234 			/*
1235 			 * If we __have__ to do radix, we're better off ignoring
1236 			 * the command line rather than not booting.
1237 			 */
1238 			prom_printf("WARNING: Ignoring cmdline option disable_radix\n");
1239 		}
1240 		support->radix_mmu = true;
1241 		break;
1242 	case OV5_FEAT(OV5_MMU_HASH):
1243 		prom_debug("MMU - hash only\n");
1244 		support->hash_mmu = true;
1245 		break;
1246 	default:
1247 		prom_debug("Unknown mmu support option: 0x%x\n", val);
1248 		break;
1249 	}
1250 }
1251 
1252 static void __init prom_parse_xive_model(u8 val,
1253 					 struct platform_support *support)
1254 {
1255 	switch (val) {
1256 	case OV5_FEAT(OV5_XIVE_EITHER): /* Either Available */
1257 		prom_debug("XIVE - either mode supported\n");
1258 		support->xive = !prom_xive_disable;
1259 		break;
1260 	case OV5_FEAT(OV5_XIVE_EXPLOIT): /* Only Exploitation mode */
1261 		prom_debug("XIVE - exploitation mode supported\n");
1262 		if (prom_xive_disable) {
1263 			/*
1264 			 * If we __have__ to do XIVE, we're better off ignoring
1265 			 * the command line rather than not booting.
1266 			 */
1267 			prom_printf("WARNING: Ignoring cmdline option xive=off\n");
1268 		}
1269 		support->xive = true;
1270 		break;
1271 	case OV5_FEAT(OV5_XIVE_LEGACY): /* Only Legacy mode */
1272 		prom_debug("XIVE - legacy mode supported\n");
1273 		break;
1274 	default:
1275 		prom_debug("Unknown xive support option: 0x%x\n", val);
1276 		break;
1277 	}
1278 }
1279 
1280 static void __init prom_parse_platform_support(u8 index, u8 val,
1281 					       struct platform_support *support)
1282 {
1283 	switch (index) {
1284 	case OV5_INDX(OV5_MMU_SUPPORT): /* MMU Model */
1285 		prom_parse_mmu_model(val & OV5_FEAT(OV5_MMU_SUPPORT), support);
1286 		break;
1287 	case OV5_INDX(OV5_RADIX_GTSE): /* Radix Extensions */
1288 		if (val & OV5_FEAT(OV5_RADIX_GTSE)) {
1289 			prom_debug("Radix - GTSE supported\n");
1290 			support->radix_gtse = true;
1291 		}
1292 		break;
1293 	case OV5_INDX(OV5_XIVE_SUPPORT): /* Interrupt mode */
1294 		prom_parse_xive_model(val & OV5_FEAT(OV5_XIVE_SUPPORT),
1295 				      support);
1296 		break;
1297 	}
1298 }
1299 
1300 static void __init prom_check_platform_support(void)
1301 {
1302 	struct platform_support supported = {
1303 		.hash_mmu = false,
1304 		.radix_mmu = false,
1305 		.radix_gtse = false,
1306 		.xive = false
1307 	};
1308 	int prop_len = prom_getproplen(prom.chosen,
1309 				       "ibm,arch-vec-5-platform-support");
1310 
1311 	/*
1312 	 * First copy the architecture vec template
1313 	 *
1314 	 * use memcpy() instead of *vec = *vec_template so that GCC replaces it
1315 	 * by __memcpy() when KASAN is active
1316 	 */
1317 	memcpy(&ibm_architecture_vec, &ibm_architecture_vec_template,
1318 	       sizeof(ibm_architecture_vec));
1319 
1320 	if (prop_len > 1) {
1321 		int i;
1322 		u8 vec[8];
1323 		prom_debug("Found ibm,arch-vec-5-platform-support, len: %d\n",
1324 			   prop_len);
1325 		if (prop_len > sizeof(vec))
1326 			prom_printf("WARNING: ibm,arch-vec-5-platform-support longer than expected (len: %d)\n",
1327 				    prop_len);
1328 		prom_getprop(prom.chosen, "ibm,arch-vec-5-platform-support",
1329 			     &vec, sizeof(vec));
1330 		for (i = 0; i < sizeof(vec); i += 2) {
1331 			prom_debug("%d: index = 0x%x val = 0x%x\n", i / 2
1332 								  , vec[i]
1333 								  , vec[i + 1]);
1334 			prom_parse_platform_support(vec[i], vec[i + 1],
1335 						    &supported);
1336 		}
1337 	}
1338 
1339 	if (supported.radix_mmu && supported.radix_gtse &&
1340 	    IS_ENABLED(CONFIG_PPC_RADIX_MMU)) {
1341 		/* Radix preferred - but we require GTSE for now */
1342 		prom_debug("Asking for radix with GTSE\n");
1343 		ibm_architecture_vec.vec5.mmu = OV5_FEAT(OV5_MMU_RADIX);
1344 		ibm_architecture_vec.vec5.radix_ext = OV5_FEAT(OV5_RADIX_GTSE);
1345 	} else if (supported.hash_mmu) {
1346 		/* Default to hash mmu (if we can) */
1347 		prom_debug("Asking for hash\n");
1348 		ibm_architecture_vec.vec5.mmu = OV5_FEAT(OV5_MMU_HASH);
1349 	} else {
1350 		/* We're probably on a legacy hypervisor */
1351 		prom_debug("Assuming legacy hash support\n");
1352 	}
1353 
1354 	if (supported.xive) {
1355 		prom_debug("Asking for XIVE\n");
1356 		ibm_architecture_vec.vec5.intarch = OV5_FEAT(OV5_XIVE_EXPLOIT);
1357 	}
1358 }
1359 
1360 static void __init prom_send_capabilities(void)
1361 {
1362 	ihandle root;
1363 	prom_arg_t ret;
1364 	u32 cores;
1365 
1366 	/* Check ibm,arch-vec-5-platform-support and fixup vec5 if required */
1367 	prom_check_platform_support();
1368 
1369 	root = call_prom("open", 1, 1, ADDR("/"));
1370 	if (root != 0) {
1371 		/* We need to tell the FW about the number of cores we support.
1372 		 *
1373 		 * To do that, we count the number of threads on the first core
1374 		 * (we assume this is the same for all cores) and use it to
1375 		 * divide NR_CPUS.
1376 		 */
1377 
1378 		cores = DIV_ROUND_UP(NR_CPUS, prom_count_smt_threads());
1379 		prom_printf("Max number of cores passed to firmware: %u (NR_CPUS = %d)\n",
1380 			    cores, NR_CPUS);
1381 
1382 		ibm_architecture_vec.vec5.max_cpus = cpu_to_be32(cores);
1383 
1384 		/* try calling the ibm,client-architecture-support method */
1385 		prom_printf("Calling ibm,client-architecture-support...");
1386 		if (call_prom_ret("call-method", 3, 2, &ret,
1387 				  ADDR("ibm,client-architecture-support"),
1388 				  root,
1389 				  ADDR(&ibm_architecture_vec)) == 0) {
1390 			/* the call exists... */
1391 			if (ret)
1392 				prom_printf("\nWARNING: ibm,client-architecture"
1393 					    "-support call FAILED!\n");
1394 			call_prom("close", 1, 0, root);
1395 			prom_printf(" done\n");
1396 			return;
1397 		}
1398 		call_prom("close", 1, 0, root);
1399 		prom_printf(" not implemented\n");
1400 	}
1401 
1402 #ifdef __BIG_ENDIAN__
1403 	{
1404 		ihandle elfloader;
1405 
1406 		/* no ibm,client-architecture-support call, try the old way */
1407 		elfloader = call_prom("open", 1, 1,
1408 				      ADDR("/packages/elf-loader"));
1409 		if (elfloader == 0) {
1410 			prom_printf("couldn't open /packages/elf-loader\n");
1411 			return;
1412 		}
1413 		call_prom("call-method", 3, 1, ADDR("process-elf-header"),
1414 			  elfloader, ADDR(&fake_elf));
1415 		call_prom("close", 1, 0, elfloader);
1416 	}
1417 #endif /* __BIG_ENDIAN__ */
1418 }
1419 #endif /* CONFIG_PPC_PSERIES */
1420 
1421 /*
1422  * Memory allocation strategy... our layout is normally:
1423  *
1424  *  at 14Mb or more we have vmlinux, then a gap and initrd.  In some
1425  *  rare cases, initrd might end up being before the kernel though.
1426  *  We assume this won't override the final kernel at 0, we have no
1427  *  provision to handle that in this version, but it should hopefully
1428  *  never happen.
1429  *
1430  *  alloc_top is set to the top of RMO, eventually shrink down if the
1431  *  TCEs overlap
1432  *
1433  *  alloc_bottom is set to the top of kernel/initrd
1434  *
1435  *  from there, allocations are done this way : rtas is allocated
1436  *  topmost, and the device-tree is allocated from the bottom. We try
1437  *  to grow the device-tree allocation as we progress. If we can't,
1438  *  then we fail, we don't currently have a facility to restart
1439  *  elsewhere, but that shouldn't be necessary.
1440  *
1441  *  Note that calls to reserve_mem have to be done explicitly, memory
1442  *  allocated with either alloc_up or alloc_down isn't automatically
1443  *  reserved.
1444  */
1445 
1446 
1447 /*
1448  * Allocates memory in the RMO upward from the kernel/initrd
1449  *
1450  * When align is 0, this is a special case, it means to allocate in place
1451  * at the current location of alloc_bottom or fail (that is basically
1452  * extending the previous allocation). Used for the device-tree flattening
1453  */
1454 static unsigned long __init alloc_up(unsigned long size, unsigned long align)
1455 {
1456 	unsigned long base = alloc_bottom;
1457 	unsigned long addr = 0;
1458 
1459 	if (align)
1460 		base = ALIGN(base, align);
1461 	prom_debug("%s(%lx, %lx)\n", __func__, size, align);
1462 	if (ram_top == 0)
1463 		prom_panic("alloc_up() called with mem not initialized\n");
1464 
1465 	if (align)
1466 		base = ALIGN(alloc_bottom, align);
1467 	else
1468 		base = alloc_bottom;
1469 
1470 	for(; (base + size) <= alloc_top;
1471 	    base = ALIGN(base + 0x100000, align)) {
1472 		prom_debug("    trying: 0x%lx\n\r", base);
1473 		addr = (unsigned long)prom_claim(base, size, 0);
1474 		if (addr != PROM_ERROR && addr != 0)
1475 			break;
1476 		addr = 0;
1477 		if (align == 0)
1478 			break;
1479 	}
1480 	if (addr == 0)
1481 		return 0;
1482 	alloc_bottom = addr + size;
1483 
1484 	prom_debug(" -> %lx\n", addr);
1485 	prom_debug("  alloc_bottom : %lx\n", alloc_bottom);
1486 	prom_debug("  alloc_top    : %lx\n", alloc_top);
1487 	prom_debug("  alloc_top_hi : %lx\n", alloc_top_high);
1488 	prom_debug("  rmo_top      : %lx\n", rmo_top);
1489 	prom_debug("  ram_top      : %lx\n", ram_top);
1490 
1491 	return addr;
1492 }
1493 
1494 /*
1495  * Allocates memory downward, either from top of RMO, or if highmem
1496  * is set, from the top of RAM.  Note that this one doesn't handle
1497  * failures.  It does claim memory if highmem is not set.
1498  */
1499 static unsigned long __init alloc_down(unsigned long size, unsigned long align,
1500 				       int highmem)
1501 {
1502 	unsigned long base, addr = 0;
1503 
1504 	prom_debug("%s(%lx, %lx, %s)\n", __func__, size, align,
1505 		   highmem ? "(high)" : "(low)");
1506 	if (ram_top == 0)
1507 		prom_panic("alloc_down() called with mem not initialized\n");
1508 
1509 	if (highmem) {
1510 		/* Carve out storage for the TCE table. */
1511 		addr = ALIGN_DOWN(alloc_top_high - size, align);
1512 		if (addr <= alloc_bottom)
1513 			return 0;
1514 		/* Will we bump into the RMO ? If yes, check out that we
1515 		 * didn't overlap existing allocations there, if we did,
1516 		 * we are dead, we must be the first in town !
1517 		 */
1518 		if (addr < rmo_top) {
1519 			/* Good, we are first */
1520 			if (alloc_top == rmo_top)
1521 				alloc_top = rmo_top = addr;
1522 			else
1523 				return 0;
1524 		}
1525 		alloc_top_high = addr;
1526 		goto bail;
1527 	}
1528 
1529 	base = ALIGN_DOWN(alloc_top - size, align);
1530 	for (; base > alloc_bottom;
1531 	     base = ALIGN_DOWN(base - 0x100000, align))  {
1532 		prom_debug("    trying: 0x%lx\n\r", base);
1533 		addr = (unsigned long)prom_claim(base, size, 0);
1534 		if (addr != PROM_ERROR && addr != 0)
1535 			break;
1536 		addr = 0;
1537 	}
1538 	if (addr == 0)
1539 		return 0;
1540 	alloc_top = addr;
1541 
1542  bail:
1543 	prom_debug(" -> %lx\n", addr);
1544 	prom_debug("  alloc_bottom : %lx\n", alloc_bottom);
1545 	prom_debug("  alloc_top    : %lx\n", alloc_top);
1546 	prom_debug("  alloc_top_hi : %lx\n", alloc_top_high);
1547 	prom_debug("  rmo_top      : %lx\n", rmo_top);
1548 	prom_debug("  ram_top      : %lx\n", ram_top);
1549 
1550 	return addr;
1551 }
1552 
1553 /*
1554  * Parse a "reg" cell
1555  */
1556 static unsigned long __init prom_next_cell(int s, cell_t **cellp)
1557 {
1558 	cell_t *p = *cellp;
1559 	unsigned long r = 0;
1560 
1561 	/* Ignore more than 2 cells */
1562 	while (s > sizeof(unsigned long) / 4) {
1563 		p++;
1564 		s--;
1565 	}
1566 	r = be32_to_cpu(*p++);
1567 #ifdef CONFIG_PPC64
1568 	if (s > 1) {
1569 		r <<= 32;
1570 		r |= be32_to_cpu(*(p++));
1571 	}
1572 #endif
1573 	*cellp = p;
1574 	return r;
1575 }
1576 
1577 /*
1578  * Very dumb function for adding to the memory reserve list, but
1579  * we don't need anything smarter at this point
1580  *
1581  * XXX Eventually check for collisions.  They should NEVER happen.
1582  * If problems seem to show up, it would be a good start to track
1583  * them down.
1584  */
1585 static void __init reserve_mem(u64 base, u64 size)
1586 {
1587 	u64 top = base + size;
1588 	unsigned long cnt = mem_reserve_cnt;
1589 
1590 	if (size == 0)
1591 		return;
1592 
1593 	/* We need to always keep one empty entry so that we
1594 	 * have our terminator with "size" set to 0 since we are
1595 	 * dumb and just copy this entire array to the boot params
1596 	 */
1597 	base = ALIGN_DOWN(base, PAGE_SIZE);
1598 	top = ALIGN(top, PAGE_SIZE);
1599 	size = top - base;
1600 
1601 	if (cnt >= (MEM_RESERVE_MAP_SIZE - 1))
1602 		prom_panic("Memory reserve map exhausted !\n");
1603 	mem_reserve_map[cnt].base = cpu_to_be64(base);
1604 	mem_reserve_map[cnt].size = cpu_to_be64(size);
1605 	mem_reserve_cnt = cnt + 1;
1606 }
1607 
1608 /*
1609  * Initialize memory allocation mechanism, parse "memory" nodes and
1610  * obtain that way the top of memory and RMO to setup out local allocator
1611  */
1612 static void __init prom_init_mem(void)
1613 {
1614 	phandle node;
1615 	char type[64];
1616 	unsigned int plen;
1617 	cell_t *p, *endp;
1618 	__be32 val;
1619 	u32 rac, rsc;
1620 
1621 	/*
1622 	 * We iterate the memory nodes to find
1623 	 * 1) top of RMO (first node)
1624 	 * 2) top of memory
1625 	 */
1626 	val = cpu_to_be32(2);
1627 	prom_getprop(prom.root, "#address-cells", &val, sizeof(val));
1628 	rac = be32_to_cpu(val);
1629 	val = cpu_to_be32(1);
1630 	prom_getprop(prom.root, "#size-cells", &val, sizeof(rsc));
1631 	rsc = be32_to_cpu(val);
1632 	prom_debug("root_addr_cells: %x\n", rac);
1633 	prom_debug("root_size_cells: %x\n", rsc);
1634 
1635 	prom_debug("scanning memory:\n");
1636 
1637 	for (node = 0; prom_next_node(&node); ) {
1638 		type[0] = 0;
1639 		prom_getprop(node, "device_type", type, sizeof(type));
1640 
1641 		if (type[0] == 0) {
1642 			/*
1643 			 * CHRP Longtrail machines have no device_type
1644 			 * on the memory node, so check the name instead...
1645 			 */
1646 			prom_getprop(node, "name", type, sizeof(type));
1647 		}
1648 		if (prom_strcmp(type, "memory"))
1649 			continue;
1650 
1651 		plen = prom_getprop(node, "reg", regbuf, sizeof(regbuf));
1652 		if (plen > sizeof(regbuf)) {
1653 			prom_printf("memory node too large for buffer !\n");
1654 			plen = sizeof(regbuf);
1655 		}
1656 		p = regbuf;
1657 		endp = p + (plen / sizeof(cell_t));
1658 
1659 #ifdef DEBUG_PROM
1660 		memset(prom_scratch, 0, sizeof(prom_scratch));
1661 		call_prom("package-to-path", 3, 1, node, prom_scratch,
1662 			  sizeof(prom_scratch) - 1);
1663 		prom_debug("  node %s :\n", prom_scratch);
1664 #endif /* DEBUG_PROM */
1665 
1666 		while ((endp - p) >= (rac + rsc)) {
1667 			unsigned long base, size;
1668 
1669 			base = prom_next_cell(rac, &p);
1670 			size = prom_next_cell(rsc, &p);
1671 
1672 			if (size == 0)
1673 				continue;
1674 			prom_debug("    %lx %lx\n", base, size);
1675 			if (base == 0 && (of_platform & PLATFORM_LPAR))
1676 				rmo_top = size;
1677 			if ((base + size) > ram_top)
1678 				ram_top = base + size;
1679 		}
1680 	}
1681 
1682 	alloc_bottom = PAGE_ALIGN((unsigned long)&_end + 0x4000);
1683 
1684 	/*
1685 	 * If prom_memory_limit is set we reduce the upper limits *except* for
1686 	 * alloc_top_high. This must be the real top of RAM so we can put
1687 	 * TCE's up there.
1688 	 */
1689 
1690 	alloc_top_high = ram_top;
1691 
1692 	if (prom_memory_limit) {
1693 		if (prom_memory_limit <= alloc_bottom) {
1694 			prom_printf("Ignoring mem=%lx <= alloc_bottom.\n",
1695 				    prom_memory_limit);
1696 			prom_memory_limit = 0;
1697 		} else if (prom_memory_limit >= ram_top) {
1698 			prom_printf("Ignoring mem=%lx >= ram_top.\n",
1699 				    prom_memory_limit);
1700 			prom_memory_limit = 0;
1701 		} else {
1702 			ram_top = prom_memory_limit;
1703 			rmo_top = min(rmo_top, prom_memory_limit);
1704 		}
1705 	}
1706 
1707 	/*
1708 	 * Setup our top alloc point, that is top of RMO or top of
1709 	 * segment 0 when running non-LPAR.
1710 	 * Some RS64 machines have buggy firmware where claims up at
1711 	 * 1GB fail.  Cap at 768MB as a workaround.
1712 	 * Since 768MB is plenty of room, and we need to cap to something
1713 	 * reasonable on 32-bit, cap at 768MB on all machines.
1714 	 */
1715 	if (!rmo_top)
1716 		rmo_top = ram_top;
1717 	rmo_top = min(0x30000000ul, rmo_top);
1718 	alloc_top = rmo_top;
1719 	alloc_top_high = ram_top;
1720 
1721 	/*
1722 	 * Check if we have an initrd after the kernel but still inside
1723 	 * the RMO.  If we do move our bottom point to after it.
1724 	 */
1725 	if (prom_initrd_start &&
1726 	    prom_initrd_start < rmo_top &&
1727 	    prom_initrd_end > alloc_bottom)
1728 		alloc_bottom = PAGE_ALIGN(prom_initrd_end);
1729 
1730 	prom_printf("memory layout at init:\n");
1731 	prom_printf("  memory_limit : %lx (16 MB aligned)\n",
1732 		    prom_memory_limit);
1733 	prom_printf("  alloc_bottom : %lx\n", alloc_bottom);
1734 	prom_printf("  alloc_top    : %lx\n", alloc_top);
1735 	prom_printf("  alloc_top_hi : %lx\n", alloc_top_high);
1736 	prom_printf("  rmo_top      : %lx\n", rmo_top);
1737 	prom_printf("  ram_top      : %lx\n", ram_top);
1738 }
1739 
1740 static void __init prom_close_stdin(void)
1741 {
1742 	__be32 val;
1743 	ihandle stdin;
1744 
1745 	if (prom_getprop(prom.chosen, "stdin", &val, sizeof(val)) > 0) {
1746 		stdin = be32_to_cpu(val);
1747 		call_prom("close", 1, 0, stdin);
1748 	}
1749 }
1750 
1751 #ifdef CONFIG_PPC_SVM
1752 static int prom_rtas_hcall(uint64_t args)
1753 {
1754 	register uint64_t arg1 asm("r3") = H_RTAS;
1755 	register uint64_t arg2 asm("r4") = args;
1756 
1757 	asm volatile("sc 1\n" : "=r" (arg1) :
1758 			"r" (arg1),
1759 			"r" (arg2) :);
1760 	return arg1;
1761 }
1762 
1763 static struct rtas_args __prombss os_term_args;
1764 
1765 static void __init prom_rtas_os_term(char *str)
1766 {
1767 	phandle rtas_node;
1768 	__be32 val;
1769 	u32 token;
1770 
1771 	prom_debug("%s: start...\n", __func__);
1772 	rtas_node = call_prom("finddevice", 1, 1, ADDR("/rtas"));
1773 	prom_debug("rtas_node: %x\n", rtas_node);
1774 	if (!PHANDLE_VALID(rtas_node))
1775 		return;
1776 
1777 	val = 0;
1778 	prom_getprop(rtas_node, "ibm,os-term", &val, sizeof(val));
1779 	token = be32_to_cpu(val);
1780 	prom_debug("ibm,os-term: %x\n", token);
1781 	if (token == 0)
1782 		prom_panic("Could not get token for ibm,os-term\n");
1783 	os_term_args.token = cpu_to_be32(token);
1784 	os_term_args.nargs = cpu_to_be32(1);
1785 	os_term_args.nret = cpu_to_be32(1);
1786 	os_term_args.args[0] = cpu_to_be32(__pa(str));
1787 	prom_rtas_hcall((uint64_t)&os_term_args);
1788 }
1789 #endif /* CONFIG_PPC_SVM */
1790 
1791 /*
1792  * Allocate room for and instantiate RTAS
1793  */
1794 static void __init prom_instantiate_rtas(void)
1795 {
1796 	phandle rtas_node;
1797 	ihandle rtas_inst;
1798 	u32 base, entry = 0;
1799 	__be32 val;
1800 	u32 size = 0;
1801 
1802 	prom_debug("prom_instantiate_rtas: start...\n");
1803 
1804 	rtas_node = call_prom("finddevice", 1, 1, ADDR("/rtas"));
1805 	prom_debug("rtas_node: %x\n", rtas_node);
1806 	if (!PHANDLE_VALID(rtas_node))
1807 		return;
1808 
1809 	val = 0;
1810 	prom_getprop(rtas_node, "rtas-size", &val, sizeof(size));
1811 	size = be32_to_cpu(val);
1812 	if (size == 0)
1813 		return;
1814 
1815 	base = alloc_down(size, PAGE_SIZE, 0);
1816 	if (base == 0)
1817 		prom_panic("Could not allocate memory for RTAS\n");
1818 
1819 	rtas_inst = call_prom("open", 1, 1, ADDR("/rtas"));
1820 	if (!IHANDLE_VALID(rtas_inst)) {
1821 		prom_printf("opening rtas package failed (%x)\n", rtas_inst);
1822 		return;
1823 	}
1824 
1825 	prom_printf("instantiating rtas at 0x%x...", base);
1826 
1827 	if (call_prom_ret("call-method", 3, 2, &entry,
1828 			  ADDR("instantiate-rtas"),
1829 			  rtas_inst, base) != 0
1830 	    || entry == 0) {
1831 		prom_printf(" failed\n");
1832 		return;
1833 	}
1834 	prom_printf(" done\n");
1835 
1836 	reserve_mem(base, size);
1837 
1838 	val = cpu_to_be32(base);
1839 	prom_setprop(rtas_node, "/rtas", "linux,rtas-base",
1840 		     &val, sizeof(val));
1841 	val = cpu_to_be32(entry);
1842 	prom_setprop(rtas_node, "/rtas", "linux,rtas-entry",
1843 		     &val, sizeof(val));
1844 
1845 	/* Check if it supports "query-cpu-stopped-state" */
1846 	if (prom_getprop(rtas_node, "query-cpu-stopped-state",
1847 			 &val, sizeof(val)) != PROM_ERROR)
1848 		rtas_has_query_cpu_stopped = true;
1849 
1850 	prom_debug("rtas base     = 0x%x\n", base);
1851 	prom_debug("rtas entry    = 0x%x\n", entry);
1852 	prom_debug("rtas size     = 0x%x\n", size);
1853 
1854 	prom_debug("prom_instantiate_rtas: end...\n");
1855 }
1856 
1857 #ifdef CONFIG_PPC64
1858 /*
1859  * Allocate room for and instantiate Stored Measurement Log (SML)
1860  */
1861 static void __init prom_instantiate_sml(void)
1862 {
1863 	phandle ibmvtpm_node;
1864 	ihandle ibmvtpm_inst;
1865 	u32 entry = 0, size = 0, succ = 0;
1866 	u64 base;
1867 	__be32 val;
1868 
1869 	prom_debug("prom_instantiate_sml: start...\n");
1870 
1871 	ibmvtpm_node = call_prom("finddevice", 1, 1, ADDR("/vdevice/vtpm"));
1872 	prom_debug("ibmvtpm_node: %x\n", ibmvtpm_node);
1873 	if (!PHANDLE_VALID(ibmvtpm_node))
1874 		return;
1875 
1876 	ibmvtpm_inst = call_prom("open", 1, 1, ADDR("/vdevice/vtpm"));
1877 	if (!IHANDLE_VALID(ibmvtpm_inst)) {
1878 		prom_printf("opening vtpm package failed (%x)\n", ibmvtpm_inst);
1879 		return;
1880 	}
1881 
1882 	if (prom_getprop(ibmvtpm_node, "ibm,sml-efi-reformat-supported",
1883 			 &val, sizeof(val)) != PROM_ERROR) {
1884 		if (call_prom_ret("call-method", 2, 2, &succ,
1885 				  ADDR("reformat-sml-to-efi-alignment"),
1886 				  ibmvtpm_inst) != 0 || succ == 0) {
1887 			prom_printf("Reformat SML to EFI alignment failed\n");
1888 			return;
1889 		}
1890 
1891 		if (call_prom_ret("call-method", 2, 2, &size,
1892 				  ADDR("sml-get-allocated-size"),
1893 				  ibmvtpm_inst) != 0 || size == 0) {
1894 			prom_printf("SML get allocated size failed\n");
1895 			return;
1896 		}
1897 	} else {
1898 		if (call_prom_ret("call-method", 2, 2, &size,
1899 				  ADDR("sml-get-handover-size"),
1900 				  ibmvtpm_inst) != 0 || size == 0) {
1901 			prom_printf("SML get handover size failed\n");
1902 			return;
1903 		}
1904 	}
1905 
1906 	base = alloc_down(size, PAGE_SIZE, 0);
1907 	if (base == 0)
1908 		prom_panic("Could not allocate memory for sml\n");
1909 
1910 	prom_printf("instantiating sml at 0x%llx...", base);
1911 
1912 	memset((void *)base, 0, size);
1913 
1914 	if (call_prom_ret("call-method", 4, 2, &entry,
1915 			  ADDR("sml-handover"),
1916 			  ibmvtpm_inst, size, base) != 0 || entry == 0) {
1917 		prom_printf("SML handover failed\n");
1918 		return;
1919 	}
1920 	prom_printf(" done\n");
1921 
1922 	reserve_mem(base, size);
1923 
1924 	prom_setprop(ibmvtpm_node, "/vdevice/vtpm", "linux,sml-base",
1925 		     &base, sizeof(base));
1926 	prom_setprop(ibmvtpm_node, "/vdevice/vtpm", "linux,sml-size",
1927 		     &size, sizeof(size));
1928 
1929 	prom_debug("sml base     = 0x%llx\n", base);
1930 	prom_debug("sml size     = 0x%x\n", size);
1931 
1932 	prom_debug("prom_instantiate_sml: end...\n");
1933 }
1934 
1935 /*
1936  * Allocate room for and initialize TCE tables
1937  */
1938 #ifdef __BIG_ENDIAN__
1939 static void __init prom_initialize_tce_table(void)
1940 {
1941 	phandle node;
1942 	ihandle phb_node;
1943 	char compatible[64], type[64], model[64];
1944 	char *path = prom_scratch;
1945 	u64 base, align;
1946 	u32 minalign, minsize;
1947 	u64 tce_entry, *tce_entryp;
1948 	u64 local_alloc_top, local_alloc_bottom;
1949 	u64 i;
1950 
1951 	if (prom_iommu_off)
1952 		return;
1953 
1954 	prom_debug("starting prom_initialize_tce_table\n");
1955 
1956 	/* Cache current top of allocs so we reserve a single block */
1957 	local_alloc_top = alloc_top_high;
1958 	local_alloc_bottom = local_alloc_top;
1959 
1960 	/* Search all nodes looking for PHBs. */
1961 	for (node = 0; prom_next_node(&node); ) {
1962 		compatible[0] = 0;
1963 		type[0] = 0;
1964 		model[0] = 0;
1965 		prom_getprop(node, "compatible",
1966 			     compatible, sizeof(compatible));
1967 		prom_getprop(node, "device_type", type, sizeof(type));
1968 		prom_getprop(node, "model", model, sizeof(model));
1969 
1970 		if ((type[0] == 0) || (prom_strstr(type, "pci") == NULL))
1971 			continue;
1972 
1973 		/* Keep the old logic intact to avoid regression. */
1974 		if (compatible[0] != 0) {
1975 			if ((prom_strstr(compatible, "python") == NULL) &&
1976 			    (prom_strstr(compatible, "Speedwagon") == NULL) &&
1977 			    (prom_strstr(compatible, "Winnipeg") == NULL))
1978 				continue;
1979 		} else if (model[0] != 0) {
1980 			if ((prom_strstr(model, "ython") == NULL) &&
1981 			    (prom_strstr(model, "peedwagon") == NULL) &&
1982 			    (prom_strstr(model, "innipeg") == NULL))
1983 				continue;
1984 		}
1985 
1986 		if (prom_getprop(node, "tce-table-minalign", &minalign,
1987 				 sizeof(minalign)) == PROM_ERROR)
1988 			minalign = 0;
1989 		if (prom_getprop(node, "tce-table-minsize", &minsize,
1990 				 sizeof(minsize)) == PROM_ERROR)
1991 			minsize = 4UL << 20;
1992 
1993 		/*
1994 		 * Even though we read what OF wants, we just set the table
1995 		 * size to 4 MB.  This is enough to map 2GB of PCI DMA space.
1996 		 * By doing this, we avoid the pitfalls of trying to DMA to
1997 		 * MMIO space and the DMA alias hole.
1998 		 */
1999 		minsize = 4UL << 20;
2000 
2001 		/* Align to the greater of the align or size */
2002 		align = max(minalign, minsize);
2003 		base = alloc_down(minsize, align, 1);
2004 		if (base == 0)
2005 			prom_panic("ERROR, cannot find space for TCE table.\n");
2006 		if (base < local_alloc_bottom)
2007 			local_alloc_bottom = base;
2008 
2009 		/* It seems OF doesn't null-terminate the path :-( */
2010 		memset(path, 0, sizeof(prom_scratch));
2011 		/* Call OF to setup the TCE hardware */
2012 		if (call_prom("package-to-path", 3, 1, node,
2013 			      path, sizeof(prom_scratch) - 1) == PROM_ERROR) {
2014 			prom_printf("package-to-path failed\n");
2015 		}
2016 
2017 		/* Save away the TCE table attributes for later use. */
2018 		prom_setprop(node, path, "linux,tce-base", &base, sizeof(base));
2019 		prom_setprop(node, path, "linux,tce-size", &minsize, sizeof(minsize));
2020 
2021 		prom_debug("TCE table: %s\n", path);
2022 		prom_debug("\tnode = 0x%x\n", node);
2023 		prom_debug("\tbase = 0x%llx\n", base);
2024 		prom_debug("\tsize = 0x%x\n", minsize);
2025 
2026 		/* Initialize the table to have a one-to-one mapping
2027 		 * over the allocated size.
2028 		 */
2029 		tce_entryp = (u64 *)base;
2030 		for (i = 0; i < (minsize >> 3) ;tce_entryp++, i++) {
2031 			tce_entry = (i << PAGE_SHIFT);
2032 			tce_entry |= 0x3;
2033 			*tce_entryp = tce_entry;
2034 		}
2035 
2036 		prom_printf("opening PHB %s", path);
2037 		phb_node = call_prom("open", 1, 1, path);
2038 		if (phb_node == 0)
2039 			prom_printf("... failed\n");
2040 		else
2041 			prom_printf("... done\n");
2042 
2043 		call_prom("call-method", 6, 0, ADDR("set-64-bit-addressing"),
2044 			  phb_node, -1, minsize,
2045 			  (u32) base, (u32) (base >> 32));
2046 		call_prom("close", 1, 0, phb_node);
2047 	}
2048 
2049 	reserve_mem(local_alloc_bottom, local_alloc_top - local_alloc_bottom);
2050 
2051 	/* These are only really needed if there is a memory limit in
2052 	 * effect, but we don't know so export them always. */
2053 	prom_tce_alloc_start = local_alloc_bottom;
2054 	prom_tce_alloc_end = local_alloc_top;
2055 
2056 	/* Flag the first invalid entry */
2057 	prom_debug("ending prom_initialize_tce_table\n");
2058 }
2059 #endif /* __BIG_ENDIAN__ */
2060 #endif /* CONFIG_PPC64 */
2061 
2062 /*
2063  * With CHRP SMP we need to use the OF to start the other processors.
2064  * We can't wait until smp_boot_cpus (the OF is trashed by then)
2065  * so we have to put the processors into a holding pattern controlled
2066  * by the kernel (not OF) before we destroy the OF.
2067  *
2068  * This uses a chunk of low memory, puts some holding pattern
2069  * code there and sends the other processors off to there until
2070  * smp_boot_cpus tells them to do something.  The holding pattern
2071  * checks that address until its cpu # is there, when it is that
2072  * cpu jumps to __secondary_start().  smp_boot_cpus() takes care
2073  * of setting those values.
2074  *
2075  * We also use physical address 0x4 here to tell when a cpu
2076  * is in its holding pattern code.
2077  *
2078  * -- Cort
2079  */
2080 /*
2081  * We want to reference the copy of __secondary_hold_* in the
2082  * 0 - 0x100 address range
2083  */
2084 #define LOW_ADDR(x)	(((unsigned long) &(x)) & 0xff)
2085 
2086 static void __init prom_hold_cpus(void)
2087 {
2088 	unsigned long i;
2089 	phandle node;
2090 	char type[64];
2091 	unsigned long *spinloop
2092 		= (void *) LOW_ADDR(__secondary_hold_spinloop);
2093 	unsigned long *acknowledge
2094 		= (void *) LOW_ADDR(__secondary_hold_acknowledge);
2095 	unsigned long secondary_hold = LOW_ADDR(__secondary_hold);
2096 
2097 	/*
2098 	 * On pseries, if RTAS supports "query-cpu-stopped-state",
2099 	 * we skip this stage, the CPUs will be started by the
2100 	 * kernel using RTAS.
2101 	 */
2102 	if ((of_platform == PLATFORM_PSERIES ||
2103 	     of_platform == PLATFORM_PSERIES_LPAR) &&
2104 	    rtas_has_query_cpu_stopped) {
2105 		prom_printf("prom_hold_cpus: skipped\n");
2106 		return;
2107 	}
2108 
2109 	prom_debug("prom_hold_cpus: start...\n");
2110 	prom_debug("    1) spinloop       = 0x%lx\n", (unsigned long)spinloop);
2111 	prom_debug("    1) *spinloop      = 0x%lx\n", *spinloop);
2112 	prom_debug("    1) acknowledge    = 0x%lx\n",
2113 		   (unsigned long)acknowledge);
2114 	prom_debug("    1) *acknowledge   = 0x%lx\n", *acknowledge);
2115 	prom_debug("    1) secondary_hold = 0x%lx\n", secondary_hold);
2116 
2117 	/* Set the common spinloop variable, so all of the secondary cpus
2118 	 * will block when they are awakened from their OF spinloop.
2119 	 * This must occur for both SMP and non SMP kernels, since OF will
2120 	 * be trashed when we move the kernel.
2121 	 */
2122 	*spinloop = 0;
2123 
2124 	/* look for cpus */
2125 	for (node = 0; prom_next_node(&node); ) {
2126 		unsigned int cpu_no;
2127 		__be32 reg;
2128 
2129 		type[0] = 0;
2130 		prom_getprop(node, "device_type", type, sizeof(type));
2131 		if (prom_strcmp(type, "cpu") != 0)
2132 			continue;
2133 
2134 		/* Skip non-configured cpus. */
2135 		if (prom_getprop(node, "status", type, sizeof(type)) > 0)
2136 			if (prom_strcmp(type, "okay") != 0)
2137 				continue;
2138 
2139 		reg = cpu_to_be32(-1); /* make sparse happy */
2140 		prom_getprop(node, "reg", &reg, sizeof(reg));
2141 		cpu_no = be32_to_cpu(reg);
2142 
2143 		prom_debug("cpu hw idx   = %u\n", cpu_no);
2144 
2145 		/* Init the acknowledge var which will be reset by
2146 		 * the secondary cpu when it awakens from its OF
2147 		 * spinloop.
2148 		 */
2149 		*acknowledge = (unsigned long)-1;
2150 
2151 		if (cpu_no != prom.cpu) {
2152 			/* Primary Thread of non-boot cpu or any thread */
2153 			prom_printf("starting cpu hw idx %u... ", cpu_no);
2154 			call_prom("start-cpu", 3, 0, node,
2155 				  secondary_hold, cpu_no);
2156 
2157 			for (i = 0; (i < 100000000) &&
2158 			     (*acknowledge == ((unsigned long)-1)); i++ )
2159 				mb();
2160 
2161 			if (*acknowledge == cpu_no)
2162 				prom_printf("done\n");
2163 			else
2164 				prom_printf("failed: %lx\n", *acknowledge);
2165 		}
2166 #ifdef CONFIG_SMP
2167 		else
2168 			prom_printf("boot cpu hw idx %u\n", cpu_no);
2169 #endif /* CONFIG_SMP */
2170 	}
2171 
2172 	prom_debug("prom_hold_cpus: end...\n");
2173 }
2174 
2175 
2176 static void __init prom_init_client_services(unsigned long pp)
2177 {
2178 	/* Get a handle to the prom entry point before anything else */
2179 	prom_entry = pp;
2180 
2181 	/* get a handle for the stdout device */
2182 	prom.chosen = call_prom("finddevice", 1, 1, ADDR("/chosen"));
2183 	if (!PHANDLE_VALID(prom.chosen))
2184 		prom_panic("cannot find chosen"); /* msg won't be printed :( */
2185 
2186 	/* get device tree root */
2187 	prom.root = call_prom("finddevice", 1, 1, ADDR("/"));
2188 	if (!PHANDLE_VALID(prom.root))
2189 		prom_panic("cannot find device tree root"); /* msg won't be printed :( */
2190 
2191 	prom.mmumap = 0;
2192 }
2193 
2194 #ifdef CONFIG_PPC32
2195 /*
2196  * For really old powermacs, we need to map things we claim.
2197  * For that, we need the ihandle of the mmu.
2198  * Also, on the longtrail, we need to work around other bugs.
2199  */
2200 static void __init prom_find_mmu(void)
2201 {
2202 	phandle oprom;
2203 	char version[64];
2204 
2205 	oprom = call_prom("finddevice", 1, 1, ADDR("/openprom"));
2206 	if (!PHANDLE_VALID(oprom))
2207 		return;
2208 	if (prom_getprop(oprom, "model", version, sizeof(version)) <= 0)
2209 		return;
2210 	version[sizeof(version) - 1] = 0;
2211 	/* XXX might need to add other versions here */
2212 	if (prom_strcmp(version, "Open Firmware, 1.0.5") == 0)
2213 		of_workarounds = OF_WA_CLAIM;
2214 	else if (prom_strncmp(version, "FirmWorks,3.", 12) == 0) {
2215 		of_workarounds = OF_WA_CLAIM | OF_WA_LONGTRAIL;
2216 		call_prom("interpret", 1, 1, "dev /memory 0 to allow-reclaim");
2217 	} else
2218 		return;
2219 	prom.memory = call_prom("open", 1, 1, ADDR("/memory"));
2220 	prom_getprop(prom.chosen, "mmu", &prom.mmumap,
2221 		     sizeof(prom.mmumap));
2222 	prom.mmumap = be32_to_cpu(prom.mmumap);
2223 	if (!IHANDLE_VALID(prom.memory) || !IHANDLE_VALID(prom.mmumap))
2224 		of_workarounds &= ~OF_WA_CLAIM;		/* hmmm */
2225 }
2226 #else
2227 #define prom_find_mmu()
2228 #endif
2229 
2230 static void __init prom_init_stdout(void)
2231 {
2232 	char *path = of_stdout_device;
2233 	char type[16];
2234 	phandle stdout_node;
2235 	__be32 val;
2236 
2237 	if (prom_getprop(prom.chosen, "stdout", &val, sizeof(val)) <= 0)
2238 		prom_panic("cannot find stdout");
2239 
2240 	prom.stdout = be32_to_cpu(val);
2241 
2242 	/* Get the full OF pathname of the stdout device */
2243 	memset(path, 0, 256);
2244 	call_prom("instance-to-path", 3, 1, prom.stdout, path, 255);
2245 	prom_printf("OF stdout device is: %s\n", of_stdout_device);
2246 	prom_setprop(prom.chosen, "/chosen", "linux,stdout-path",
2247 		     path, prom_strlen(path) + 1);
2248 
2249 	/* instance-to-package fails on PA-Semi */
2250 	stdout_node = call_prom("instance-to-package", 1, 1, prom.stdout);
2251 	if (stdout_node != PROM_ERROR) {
2252 		val = cpu_to_be32(stdout_node);
2253 
2254 		/* If it's a display, note it */
2255 		memset(type, 0, sizeof(type));
2256 		prom_getprop(stdout_node, "device_type", type, sizeof(type));
2257 		if (prom_strcmp(type, "display") == 0)
2258 			prom_setprop(stdout_node, path, "linux,boot-display", NULL, 0);
2259 	}
2260 }
2261 
2262 static int __init prom_find_machine_type(void)
2263 {
2264 	char compat[256];
2265 	int len, i = 0;
2266 #ifdef CONFIG_PPC64
2267 	phandle rtas;
2268 	int x;
2269 #endif
2270 
2271 	/* Look for a PowerMac or a Cell */
2272 	len = prom_getprop(prom.root, "compatible",
2273 			   compat, sizeof(compat)-1);
2274 	if (len > 0) {
2275 		compat[len] = 0;
2276 		while (i < len) {
2277 			char *p = &compat[i];
2278 			int sl = prom_strlen(p);
2279 			if (sl == 0)
2280 				break;
2281 			if (prom_strstr(p, "Power Macintosh") ||
2282 			    prom_strstr(p, "MacRISC"))
2283 				return PLATFORM_POWERMAC;
2284 #ifdef CONFIG_PPC64
2285 			/* We must make sure we don't detect the IBM Cell
2286 			 * blades as pSeries due to some firmware issues,
2287 			 * so we do it here.
2288 			 */
2289 			if (prom_strstr(p, "IBM,CBEA") ||
2290 			    prom_strstr(p, "IBM,CPBW-1.0"))
2291 				return PLATFORM_GENERIC;
2292 #endif /* CONFIG_PPC64 */
2293 			i += sl + 1;
2294 		}
2295 	}
2296 #ifdef CONFIG_PPC64
2297 	/* Try to figure out if it's an IBM pSeries or any other
2298 	 * PAPR compliant platform. We assume it is if :
2299 	 *  - /device_type is "chrp" (please, do NOT use that for future
2300 	 *    non-IBM designs !
2301 	 *  - it has /rtas
2302 	 */
2303 	len = prom_getprop(prom.root, "device_type",
2304 			   compat, sizeof(compat)-1);
2305 	if (len <= 0)
2306 		return PLATFORM_GENERIC;
2307 	if (prom_strcmp(compat, "chrp"))
2308 		return PLATFORM_GENERIC;
2309 
2310 	/* Default to pSeries. We need to know if we are running LPAR */
2311 	rtas = call_prom("finddevice", 1, 1, ADDR("/rtas"));
2312 	if (!PHANDLE_VALID(rtas))
2313 		return PLATFORM_GENERIC;
2314 	x = prom_getproplen(rtas, "ibm,hypertas-functions");
2315 	if (x != PROM_ERROR) {
2316 		prom_debug("Hypertas detected, assuming LPAR !\n");
2317 		return PLATFORM_PSERIES_LPAR;
2318 	}
2319 	return PLATFORM_PSERIES;
2320 #else
2321 	return PLATFORM_GENERIC;
2322 #endif
2323 }
2324 
2325 static int __init prom_set_color(ihandle ih, int i, int r, int g, int b)
2326 {
2327 	return call_prom("call-method", 6, 1, ADDR("color!"), ih, i, b, g, r);
2328 }
2329 
2330 /*
2331  * If we have a display that we don't know how to drive,
2332  * we will want to try to execute OF's open method for it
2333  * later.  However, OF will probably fall over if we do that
2334  * we've taken over the MMU.
2335  * So we check whether we will need to open the display,
2336  * and if so, open it now.
2337  */
2338 static void __init prom_check_displays(void)
2339 {
2340 	char type[16], *path;
2341 	phandle node;
2342 	ihandle ih;
2343 	int i;
2344 
2345 	static const unsigned char default_colors[] __initconst = {
2346 		0x00, 0x00, 0x00,
2347 		0x00, 0x00, 0xaa,
2348 		0x00, 0xaa, 0x00,
2349 		0x00, 0xaa, 0xaa,
2350 		0xaa, 0x00, 0x00,
2351 		0xaa, 0x00, 0xaa,
2352 		0xaa, 0xaa, 0x00,
2353 		0xaa, 0xaa, 0xaa,
2354 		0x55, 0x55, 0x55,
2355 		0x55, 0x55, 0xff,
2356 		0x55, 0xff, 0x55,
2357 		0x55, 0xff, 0xff,
2358 		0xff, 0x55, 0x55,
2359 		0xff, 0x55, 0xff,
2360 		0xff, 0xff, 0x55,
2361 		0xff, 0xff, 0xff
2362 	};
2363 	const unsigned char *clut;
2364 
2365 	prom_debug("Looking for displays\n");
2366 	for (node = 0; prom_next_node(&node); ) {
2367 		memset(type, 0, sizeof(type));
2368 		prom_getprop(node, "device_type", type, sizeof(type));
2369 		if (prom_strcmp(type, "display") != 0)
2370 			continue;
2371 
2372 		/* It seems OF doesn't null-terminate the path :-( */
2373 		path = prom_scratch;
2374 		memset(path, 0, sizeof(prom_scratch));
2375 
2376 		/*
2377 		 * leave some room at the end of the path for appending extra
2378 		 * arguments
2379 		 */
2380 		if (call_prom("package-to-path", 3, 1, node, path,
2381 			      sizeof(prom_scratch) - 10) == PROM_ERROR)
2382 			continue;
2383 		prom_printf("found display   : %s, opening... ", path);
2384 
2385 		ih = call_prom("open", 1, 1, path);
2386 		if (ih == 0) {
2387 			prom_printf("failed\n");
2388 			continue;
2389 		}
2390 
2391 		/* Success */
2392 		prom_printf("done\n");
2393 		prom_setprop(node, path, "linux,opened", NULL, 0);
2394 
2395 		/* Setup a usable color table when the appropriate
2396 		 * method is available. Should update this to set-colors */
2397 		clut = default_colors;
2398 		for (i = 0; i < 16; i++, clut += 3)
2399 			if (prom_set_color(ih, i, clut[0], clut[1],
2400 					   clut[2]) != 0)
2401 				break;
2402 
2403 #ifdef CONFIG_LOGO_LINUX_CLUT224
2404 		clut = PTRRELOC(logo_linux_clut224.clut);
2405 		for (i = 0; i < logo_linux_clut224.clutsize; i++, clut += 3)
2406 			if (prom_set_color(ih, i + 32, clut[0], clut[1],
2407 					   clut[2]) != 0)
2408 				break;
2409 #endif /* CONFIG_LOGO_LINUX_CLUT224 */
2410 
2411 #ifdef CONFIG_PPC_EARLY_DEBUG_BOOTX
2412 		if (prom_getprop(node, "linux,boot-display", NULL, 0) !=
2413 		    PROM_ERROR) {
2414 			u32 width, height, pitch, addr;
2415 
2416 			prom_printf("Setting btext !\n");
2417 			prom_getprop(node, "width", &width, 4);
2418 			prom_getprop(node, "height", &height, 4);
2419 			prom_getprop(node, "linebytes", &pitch, 4);
2420 			prom_getprop(node, "address", &addr, 4);
2421 			prom_printf("W=%d H=%d LB=%d addr=0x%x\n",
2422 				    width, height, pitch, addr);
2423 			btext_setup_display(width, height, 8, pitch, addr);
2424 			btext_prepare_BAT();
2425 		}
2426 #endif /* CONFIG_PPC_EARLY_DEBUG_BOOTX */
2427 	}
2428 }
2429 
2430 
2431 /* Return (relocated) pointer to this much memory: moves initrd if reqd. */
2432 static void __init *make_room(unsigned long *mem_start, unsigned long *mem_end,
2433 			      unsigned long needed, unsigned long align)
2434 {
2435 	void *ret;
2436 
2437 	*mem_start = ALIGN(*mem_start, align);
2438 	while ((*mem_start + needed) > *mem_end) {
2439 		unsigned long room, chunk;
2440 
2441 		prom_debug("Chunk exhausted, claiming more at %lx...\n",
2442 			   alloc_bottom);
2443 		room = alloc_top - alloc_bottom;
2444 		if (room > DEVTREE_CHUNK_SIZE)
2445 			room = DEVTREE_CHUNK_SIZE;
2446 		if (room < PAGE_SIZE)
2447 			prom_panic("No memory for flatten_device_tree "
2448 				   "(no room)\n");
2449 		chunk = alloc_up(room, 0);
2450 		if (chunk == 0)
2451 			prom_panic("No memory for flatten_device_tree "
2452 				   "(claim failed)\n");
2453 		*mem_end = chunk + room;
2454 	}
2455 
2456 	ret = (void *)*mem_start;
2457 	*mem_start += needed;
2458 
2459 	return ret;
2460 }
2461 
2462 #define dt_push_token(token, mem_start, mem_end) do { 			\
2463 		void *room = make_room(mem_start, mem_end, 4, 4);	\
2464 		*(__be32 *)room = cpu_to_be32(token);			\
2465 	} while(0)
2466 
2467 static unsigned long __init dt_find_string(char *str)
2468 {
2469 	char *s, *os;
2470 
2471 	s = os = (char *)dt_string_start;
2472 	s += 4;
2473 	while (s <  (char *)dt_string_end) {
2474 		if (prom_strcmp(s, str) == 0)
2475 			return s - os;
2476 		s += prom_strlen(s) + 1;
2477 	}
2478 	return 0;
2479 }
2480 
2481 /*
2482  * The Open Firmware 1275 specification states properties must be 31 bytes or
2483  * less, however not all firmwares obey this. Make it 64 bytes to be safe.
2484  */
2485 #define MAX_PROPERTY_NAME 64
2486 
2487 static void __init scan_dt_build_strings(phandle node,
2488 					 unsigned long *mem_start,
2489 					 unsigned long *mem_end)
2490 {
2491 	char *prev_name, *namep, *sstart;
2492 	unsigned long soff;
2493 	phandle child;
2494 
2495 	sstart =  (char *)dt_string_start;
2496 
2497 	/* get and store all property names */
2498 	prev_name = "";
2499 	for (;;) {
2500 		/* 64 is max len of name including nul. */
2501 		namep = make_room(mem_start, mem_end, MAX_PROPERTY_NAME, 1);
2502 		if (call_prom("nextprop", 3, 1, node, prev_name, namep) != 1) {
2503 			/* No more nodes: unwind alloc */
2504 			*mem_start = (unsigned long)namep;
2505 			break;
2506 		}
2507 
2508  		/* skip "name" */
2509 		if (prom_strcmp(namep, "name") == 0) {
2510  			*mem_start = (unsigned long)namep;
2511  			prev_name = "name";
2512  			continue;
2513  		}
2514 		/* get/create string entry */
2515 		soff = dt_find_string(namep);
2516 		if (soff != 0) {
2517 			*mem_start = (unsigned long)namep;
2518 			namep = sstart + soff;
2519 		} else {
2520 			/* Trim off some if we can */
2521 			*mem_start = (unsigned long)namep + prom_strlen(namep) + 1;
2522 			dt_string_end = *mem_start;
2523 		}
2524 		prev_name = namep;
2525 	}
2526 
2527 	/* do all our children */
2528 	child = call_prom("child", 1, 1, node);
2529 	while (child != 0) {
2530 		scan_dt_build_strings(child, mem_start, mem_end);
2531 		child = call_prom("peer", 1, 1, child);
2532 	}
2533 }
2534 
2535 static void __init scan_dt_build_struct(phandle node, unsigned long *mem_start,
2536 					unsigned long *mem_end)
2537 {
2538 	phandle child;
2539 	char *namep, *prev_name, *sstart, *p, *ep, *lp, *path;
2540 	unsigned long soff;
2541 	unsigned char *valp;
2542 	static char pname[MAX_PROPERTY_NAME] __prombss;
2543 	int l, room, has_phandle = 0;
2544 
2545 	dt_push_token(OF_DT_BEGIN_NODE, mem_start, mem_end);
2546 
2547 	/* get the node's full name */
2548 	namep = (char *)*mem_start;
2549 	room = *mem_end - *mem_start;
2550 	if (room > 255)
2551 		room = 255;
2552 	l = call_prom("package-to-path", 3, 1, node, namep, room);
2553 	if (l >= 0) {
2554 		/* Didn't fit?  Get more room. */
2555 		if (l >= room) {
2556 			if (l >= *mem_end - *mem_start)
2557 				namep = make_room(mem_start, mem_end, l+1, 1);
2558 			call_prom("package-to-path", 3, 1, node, namep, l);
2559 		}
2560 		namep[l] = '\0';
2561 
2562 		/* Fixup an Apple bug where they have bogus \0 chars in the
2563 		 * middle of the path in some properties, and extract
2564 		 * the unit name (everything after the last '/').
2565 		 */
2566 		for (lp = p = namep, ep = namep + l; p < ep; p++) {
2567 			if (*p == '/')
2568 				lp = namep;
2569 			else if (*p != 0)
2570 				*lp++ = *p;
2571 		}
2572 		*lp = 0;
2573 		*mem_start = ALIGN((unsigned long)lp + 1, 4);
2574 	}
2575 
2576 	/* get it again for debugging */
2577 	path = prom_scratch;
2578 	memset(path, 0, sizeof(prom_scratch));
2579 	call_prom("package-to-path", 3, 1, node, path, sizeof(prom_scratch) - 1);
2580 
2581 	/* get and store all properties */
2582 	prev_name = "";
2583 	sstart = (char *)dt_string_start;
2584 	for (;;) {
2585 		if (call_prom("nextprop", 3, 1, node, prev_name,
2586 			      pname) != 1)
2587 			break;
2588 
2589  		/* skip "name" */
2590 		if (prom_strcmp(pname, "name") == 0) {
2591  			prev_name = "name";
2592  			continue;
2593  		}
2594 
2595 		/* find string offset */
2596 		soff = dt_find_string(pname);
2597 		if (soff == 0) {
2598 			prom_printf("WARNING: Can't find string index for"
2599 				    " <%s>, node %s\n", pname, path);
2600 			break;
2601 		}
2602 		prev_name = sstart + soff;
2603 
2604 		/* get length */
2605 		l = call_prom("getproplen", 2, 1, node, pname);
2606 
2607 		/* sanity checks */
2608 		if (l == PROM_ERROR)
2609 			continue;
2610 
2611 		/* push property head */
2612 		dt_push_token(OF_DT_PROP, mem_start, mem_end);
2613 		dt_push_token(l, mem_start, mem_end);
2614 		dt_push_token(soff, mem_start, mem_end);
2615 
2616 		/* push property content */
2617 		valp = make_room(mem_start, mem_end, l, 4);
2618 		call_prom("getprop", 4, 1, node, pname, valp, l);
2619 		*mem_start = ALIGN(*mem_start, 4);
2620 
2621 		if (!prom_strcmp(pname, "phandle"))
2622 			has_phandle = 1;
2623 	}
2624 
2625 	/* Add a "phandle" property if none already exist */
2626 	if (!has_phandle) {
2627 		soff = dt_find_string("phandle");
2628 		if (soff == 0)
2629 			prom_printf("WARNING: Can't find string index for <phandle> node %s\n", path);
2630 		else {
2631 			dt_push_token(OF_DT_PROP, mem_start, mem_end);
2632 			dt_push_token(4, mem_start, mem_end);
2633 			dt_push_token(soff, mem_start, mem_end);
2634 			valp = make_room(mem_start, mem_end, 4, 4);
2635 			*(__be32 *)valp = cpu_to_be32(node);
2636 		}
2637 	}
2638 
2639 	/* do all our children */
2640 	child = call_prom("child", 1, 1, node);
2641 	while (child != 0) {
2642 		scan_dt_build_struct(child, mem_start, mem_end);
2643 		child = call_prom("peer", 1, 1, child);
2644 	}
2645 
2646 	dt_push_token(OF_DT_END_NODE, mem_start, mem_end);
2647 }
2648 
2649 static void __init flatten_device_tree(void)
2650 {
2651 	phandle root;
2652 	unsigned long mem_start, mem_end, room;
2653 	struct boot_param_header *hdr;
2654 	char *namep;
2655 	u64 *rsvmap;
2656 
2657 	/*
2658 	 * Check how much room we have between alloc top & bottom (+/- a
2659 	 * few pages), crop to 1MB, as this is our "chunk" size
2660 	 */
2661 	room = alloc_top - alloc_bottom - 0x4000;
2662 	if (room > DEVTREE_CHUNK_SIZE)
2663 		room = DEVTREE_CHUNK_SIZE;
2664 	prom_debug("starting device tree allocs at %lx\n", alloc_bottom);
2665 
2666 	/* Now try to claim that */
2667 	mem_start = (unsigned long)alloc_up(room, PAGE_SIZE);
2668 	if (mem_start == 0)
2669 		prom_panic("Can't allocate initial device-tree chunk\n");
2670 	mem_end = mem_start + room;
2671 
2672 	/* Get root of tree */
2673 	root = call_prom("peer", 1, 1, (phandle)0);
2674 	if (root == (phandle)0)
2675 		prom_panic ("couldn't get device tree root\n");
2676 
2677 	/* Build header and make room for mem rsv map */
2678 	mem_start = ALIGN(mem_start, 4);
2679 	hdr = make_room(&mem_start, &mem_end,
2680 			sizeof(struct boot_param_header), 4);
2681 	dt_header_start = (unsigned long)hdr;
2682 	rsvmap = make_room(&mem_start, &mem_end, sizeof(mem_reserve_map), 8);
2683 
2684 	/* Start of strings */
2685 	mem_start = PAGE_ALIGN(mem_start);
2686 	dt_string_start = mem_start;
2687 	mem_start += 4; /* hole */
2688 
2689 	/* Add "phandle" in there, we'll need it */
2690 	namep = make_room(&mem_start, &mem_end, 16, 1);
2691 	prom_strcpy(namep, "phandle");
2692 	mem_start = (unsigned long)namep + prom_strlen(namep) + 1;
2693 
2694 	/* Build string array */
2695 	prom_printf("Building dt strings...\n");
2696 	scan_dt_build_strings(root, &mem_start, &mem_end);
2697 	dt_string_end = mem_start;
2698 
2699 	/* Build structure */
2700 	mem_start = PAGE_ALIGN(mem_start);
2701 	dt_struct_start = mem_start;
2702 	prom_printf("Building dt structure...\n");
2703 	scan_dt_build_struct(root, &mem_start, &mem_end);
2704 	dt_push_token(OF_DT_END, &mem_start, &mem_end);
2705 	dt_struct_end = PAGE_ALIGN(mem_start);
2706 
2707 	/* Finish header */
2708 	hdr->boot_cpuid_phys = cpu_to_be32(prom.cpu);
2709 	hdr->magic = cpu_to_be32(OF_DT_HEADER);
2710 	hdr->totalsize = cpu_to_be32(dt_struct_end - dt_header_start);
2711 	hdr->off_dt_struct = cpu_to_be32(dt_struct_start - dt_header_start);
2712 	hdr->off_dt_strings = cpu_to_be32(dt_string_start - dt_header_start);
2713 	hdr->dt_strings_size = cpu_to_be32(dt_string_end - dt_string_start);
2714 	hdr->off_mem_rsvmap = cpu_to_be32(((unsigned long)rsvmap) - dt_header_start);
2715 	hdr->version = cpu_to_be32(OF_DT_VERSION);
2716 	/* Version 16 is not backward compatible */
2717 	hdr->last_comp_version = cpu_to_be32(0x10);
2718 
2719 	/* Copy the reserve map in */
2720 	memcpy(rsvmap, mem_reserve_map, sizeof(mem_reserve_map));
2721 
2722 #ifdef DEBUG_PROM
2723 	{
2724 		int i;
2725 		prom_printf("reserved memory map:\n");
2726 		for (i = 0; i < mem_reserve_cnt; i++)
2727 			prom_printf("  %llx - %llx\n",
2728 				    be64_to_cpu(mem_reserve_map[i].base),
2729 				    be64_to_cpu(mem_reserve_map[i].size));
2730 	}
2731 #endif
2732 	/* Bump mem_reserve_cnt to cause further reservations to fail
2733 	 * since it's too late.
2734 	 */
2735 	mem_reserve_cnt = MEM_RESERVE_MAP_SIZE;
2736 
2737 	prom_printf("Device tree strings 0x%lx -> 0x%lx\n",
2738 		    dt_string_start, dt_string_end);
2739 	prom_printf("Device tree struct  0x%lx -> 0x%lx\n",
2740 		    dt_struct_start, dt_struct_end);
2741 }
2742 
2743 #ifdef CONFIG_PPC_MAPLE
2744 /* PIBS Version 1.05.0000 04/26/2005 has an incorrect /ht/isa/ranges property.
2745  * The values are bad, and it doesn't even have the right number of cells. */
2746 static void __init fixup_device_tree_maple(void)
2747 {
2748 	phandle isa;
2749 	u32 rloc = 0x01002000; /* IO space; PCI device = 4 */
2750 	u32 isa_ranges[6];
2751 	char *name;
2752 
2753 	name = "/ht@0/isa@4";
2754 	isa = call_prom("finddevice", 1, 1, ADDR(name));
2755 	if (!PHANDLE_VALID(isa)) {
2756 		name = "/ht@0/isa@6";
2757 		isa = call_prom("finddevice", 1, 1, ADDR(name));
2758 		rloc = 0x01003000; /* IO space; PCI device = 6 */
2759 	}
2760 	if (!PHANDLE_VALID(isa))
2761 		return;
2762 
2763 	if (prom_getproplen(isa, "ranges") != 12)
2764 		return;
2765 	if (prom_getprop(isa, "ranges", isa_ranges, sizeof(isa_ranges))
2766 		== PROM_ERROR)
2767 		return;
2768 
2769 	if (isa_ranges[0] != 0x1 ||
2770 		isa_ranges[1] != 0xf4000000 ||
2771 		isa_ranges[2] != 0x00010000)
2772 		return;
2773 
2774 	prom_printf("Fixing up bogus ISA range on Maple/Apache...\n");
2775 
2776 	isa_ranges[0] = 0x1;
2777 	isa_ranges[1] = 0x0;
2778 	isa_ranges[2] = rloc;
2779 	isa_ranges[3] = 0x0;
2780 	isa_ranges[4] = 0x0;
2781 	isa_ranges[5] = 0x00010000;
2782 	prom_setprop(isa, name, "ranges",
2783 			isa_ranges, sizeof(isa_ranges));
2784 }
2785 
2786 #define CPC925_MC_START		0xf8000000
2787 #define CPC925_MC_LENGTH	0x1000000
2788 /* The values for memory-controller don't have right number of cells */
2789 static void __init fixup_device_tree_maple_memory_controller(void)
2790 {
2791 	phandle mc;
2792 	u32 mc_reg[4];
2793 	char *name = "/hostbridge@f8000000";
2794 	u32 ac, sc;
2795 
2796 	mc = call_prom("finddevice", 1, 1, ADDR(name));
2797 	if (!PHANDLE_VALID(mc))
2798 		return;
2799 
2800 	if (prom_getproplen(mc, "reg") != 8)
2801 		return;
2802 
2803 	prom_getprop(prom.root, "#address-cells", &ac, sizeof(ac));
2804 	prom_getprop(prom.root, "#size-cells", &sc, sizeof(sc));
2805 	if ((ac != 2) || (sc != 2))
2806 		return;
2807 
2808 	if (prom_getprop(mc, "reg", mc_reg, sizeof(mc_reg)) == PROM_ERROR)
2809 		return;
2810 
2811 	if (mc_reg[0] != CPC925_MC_START || mc_reg[1] != CPC925_MC_LENGTH)
2812 		return;
2813 
2814 	prom_printf("Fixing up bogus hostbridge on Maple...\n");
2815 
2816 	mc_reg[0] = 0x0;
2817 	mc_reg[1] = CPC925_MC_START;
2818 	mc_reg[2] = 0x0;
2819 	mc_reg[3] = CPC925_MC_LENGTH;
2820 	prom_setprop(mc, name, "reg", mc_reg, sizeof(mc_reg));
2821 }
2822 #else
2823 #define fixup_device_tree_maple()
2824 #define fixup_device_tree_maple_memory_controller()
2825 #endif
2826 
2827 #ifdef CONFIG_PPC_CHRP
2828 /*
2829  * Pegasos and BriQ lacks the "ranges" property in the isa node
2830  * Pegasos needs decimal IRQ 14/15, not hexadecimal
2831  * Pegasos has the IDE configured in legacy mode, but advertised as native
2832  */
2833 static void __init fixup_device_tree_chrp(void)
2834 {
2835 	phandle ph;
2836 	u32 prop[6];
2837 	u32 rloc = 0x01006000; /* IO space; PCI device = 12 */
2838 	char *name;
2839 	int rc;
2840 
2841 	name = "/pci@80000000/isa@c";
2842 	ph = call_prom("finddevice", 1, 1, ADDR(name));
2843 	if (!PHANDLE_VALID(ph)) {
2844 		name = "/pci@ff500000/isa@6";
2845 		ph = call_prom("finddevice", 1, 1, ADDR(name));
2846 		rloc = 0x01003000; /* IO space; PCI device = 6 */
2847 	}
2848 	if (PHANDLE_VALID(ph)) {
2849 		rc = prom_getproplen(ph, "ranges");
2850 		if (rc == 0 || rc == PROM_ERROR) {
2851 			prom_printf("Fixing up missing ISA range on Pegasos...\n");
2852 
2853 			prop[0] = 0x1;
2854 			prop[1] = 0x0;
2855 			prop[2] = rloc;
2856 			prop[3] = 0x0;
2857 			prop[4] = 0x0;
2858 			prop[5] = 0x00010000;
2859 			prom_setprop(ph, name, "ranges", prop, sizeof(prop));
2860 		}
2861 	}
2862 
2863 	name = "/pci@80000000/ide@C,1";
2864 	ph = call_prom("finddevice", 1, 1, ADDR(name));
2865 	if (PHANDLE_VALID(ph)) {
2866 		prom_printf("Fixing up IDE interrupt on Pegasos...\n");
2867 		prop[0] = 14;
2868 		prop[1] = 0x0;
2869 		prom_setprop(ph, name, "interrupts", prop, 2*sizeof(u32));
2870 		prom_printf("Fixing up IDE class-code on Pegasos...\n");
2871 		rc = prom_getprop(ph, "class-code", prop, sizeof(u32));
2872 		if (rc == sizeof(u32)) {
2873 			prop[0] &= ~0x5;
2874 			prom_setprop(ph, name, "class-code", prop, sizeof(u32));
2875 		}
2876 	}
2877 }
2878 #else
2879 #define fixup_device_tree_chrp()
2880 #endif
2881 
2882 #if defined(CONFIG_PPC64) && defined(CONFIG_PPC_PMAC)
2883 static void __init fixup_device_tree_pmac(void)
2884 {
2885 	phandle u3, i2c, mpic;
2886 	u32 u3_rev;
2887 	u32 interrupts[2];
2888 	u32 parent;
2889 
2890 	/* Some G5s have a missing interrupt definition, fix it up here */
2891 	u3 = call_prom("finddevice", 1, 1, ADDR("/u3@0,f8000000"));
2892 	if (!PHANDLE_VALID(u3))
2893 		return;
2894 	i2c = call_prom("finddevice", 1, 1, ADDR("/u3@0,f8000000/i2c@f8001000"));
2895 	if (!PHANDLE_VALID(i2c))
2896 		return;
2897 	mpic = call_prom("finddevice", 1, 1, ADDR("/u3@0,f8000000/mpic@f8040000"));
2898 	if (!PHANDLE_VALID(mpic))
2899 		return;
2900 
2901 	/* check if proper rev of u3 */
2902 	if (prom_getprop(u3, "device-rev", &u3_rev, sizeof(u3_rev))
2903 	    == PROM_ERROR)
2904 		return;
2905 	if (u3_rev < 0x35 || u3_rev > 0x39)
2906 		return;
2907 	/* does it need fixup ? */
2908 	if (prom_getproplen(i2c, "interrupts") > 0)
2909 		return;
2910 
2911 	prom_printf("fixing up bogus interrupts for u3 i2c...\n");
2912 
2913 	/* interrupt on this revision of u3 is number 0 and level */
2914 	interrupts[0] = 0;
2915 	interrupts[1] = 1;
2916 	prom_setprop(i2c, "/u3@0,f8000000/i2c@f8001000", "interrupts",
2917 		     &interrupts, sizeof(interrupts));
2918 	parent = (u32)mpic;
2919 	prom_setprop(i2c, "/u3@0,f8000000/i2c@f8001000", "interrupt-parent",
2920 		     &parent, sizeof(parent));
2921 }
2922 #else
2923 #define fixup_device_tree_pmac()
2924 #endif
2925 
2926 #ifdef CONFIG_PPC_EFIKA
2927 /*
2928  * The MPC5200 FEC driver requires an phy-handle property to tell it how
2929  * to talk to the phy.  If the phy-handle property is missing, then this
2930  * function is called to add the appropriate nodes and link it to the
2931  * ethernet node.
2932  */
2933 static void __init fixup_device_tree_efika_add_phy(void)
2934 {
2935 	u32 node;
2936 	char prop[64];
2937 	int rv;
2938 
2939 	/* Check if /builtin/ethernet exists - bail if it doesn't */
2940 	node = call_prom("finddevice", 1, 1, ADDR("/builtin/ethernet"));
2941 	if (!PHANDLE_VALID(node))
2942 		return;
2943 
2944 	/* Check if the phy-handle property exists - bail if it does */
2945 	rv = prom_getprop(node, "phy-handle", prop, sizeof(prop));
2946 	if (!rv)
2947 		return;
2948 
2949 	/*
2950 	 * At this point the ethernet device doesn't have a phy described.
2951 	 * Now we need to add the missing phy node and linkage
2952 	 */
2953 
2954 	/* Check for an MDIO bus node - if missing then create one */
2955 	node = call_prom("finddevice", 1, 1, ADDR("/builtin/mdio"));
2956 	if (!PHANDLE_VALID(node)) {
2957 		prom_printf("Adding Ethernet MDIO node\n");
2958 		call_prom("interpret", 1, 1,
2959 			" s\" /builtin\" find-device"
2960 			" new-device"
2961 				" 1 encode-int s\" #address-cells\" property"
2962 				" 0 encode-int s\" #size-cells\" property"
2963 				" s\" mdio\" device-name"
2964 				" s\" fsl,mpc5200b-mdio\" encode-string"
2965 				" s\" compatible\" property"
2966 				" 0xf0003000 0x400 reg"
2967 				" 0x2 encode-int"
2968 				" 0x5 encode-int encode+"
2969 				" 0x3 encode-int encode+"
2970 				" s\" interrupts\" property"
2971 			" finish-device");
2972 	};
2973 
2974 	/* Check for a PHY device node - if missing then create one and
2975 	 * give it's phandle to the ethernet node */
2976 	node = call_prom("finddevice", 1, 1,
2977 			 ADDR("/builtin/mdio/ethernet-phy"));
2978 	if (!PHANDLE_VALID(node)) {
2979 		prom_printf("Adding Ethernet PHY node\n");
2980 		call_prom("interpret", 1, 1,
2981 			" s\" /builtin/mdio\" find-device"
2982 			" new-device"
2983 				" s\" ethernet-phy\" device-name"
2984 				" 0x10 encode-int s\" reg\" property"
2985 				" my-self"
2986 				" ihandle>phandle"
2987 			" finish-device"
2988 			" s\" /builtin/ethernet\" find-device"
2989 				" encode-int"
2990 				" s\" phy-handle\" property"
2991 			" device-end");
2992 	}
2993 }
2994 
2995 static void __init fixup_device_tree_efika(void)
2996 {
2997 	int sound_irq[3] = { 2, 2, 0 };
2998 	int bcomm_irq[3*16] = { 3,0,0, 3,1,0, 3,2,0, 3,3,0,
2999 				3,4,0, 3,5,0, 3,6,0, 3,7,0,
3000 				3,8,0, 3,9,0, 3,10,0, 3,11,0,
3001 				3,12,0, 3,13,0, 3,14,0, 3,15,0 };
3002 	u32 node;
3003 	char prop[64];
3004 	int rv, len;
3005 
3006 	/* Check if we're really running on a EFIKA */
3007 	node = call_prom("finddevice", 1, 1, ADDR("/"));
3008 	if (!PHANDLE_VALID(node))
3009 		return;
3010 
3011 	rv = prom_getprop(node, "model", prop, sizeof(prop));
3012 	if (rv == PROM_ERROR)
3013 		return;
3014 	if (prom_strcmp(prop, "EFIKA5K2"))
3015 		return;
3016 
3017 	prom_printf("Applying EFIKA device tree fixups\n");
3018 
3019 	/* Claiming to be 'chrp' is death */
3020 	node = call_prom("finddevice", 1, 1, ADDR("/"));
3021 	rv = prom_getprop(node, "device_type", prop, sizeof(prop));
3022 	if (rv != PROM_ERROR && (prom_strcmp(prop, "chrp") == 0))
3023 		prom_setprop(node, "/", "device_type", "efika", sizeof("efika"));
3024 
3025 	/* CODEGEN,description is exposed in /proc/cpuinfo so
3026 	   fix that too */
3027 	rv = prom_getprop(node, "CODEGEN,description", prop, sizeof(prop));
3028 	if (rv != PROM_ERROR && (prom_strstr(prop, "CHRP")))
3029 		prom_setprop(node, "/", "CODEGEN,description",
3030 			     "Efika 5200B PowerPC System",
3031 			     sizeof("Efika 5200B PowerPC System"));
3032 
3033 	/* Fixup bestcomm interrupts property */
3034 	node = call_prom("finddevice", 1, 1, ADDR("/builtin/bestcomm"));
3035 	if (PHANDLE_VALID(node)) {
3036 		len = prom_getproplen(node, "interrupts");
3037 		if (len == 12) {
3038 			prom_printf("Fixing bestcomm interrupts property\n");
3039 			prom_setprop(node, "/builtin/bestcom", "interrupts",
3040 				     bcomm_irq, sizeof(bcomm_irq));
3041 		}
3042 	}
3043 
3044 	/* Fixup sound interrupts property */
3045 	node = call_prom("finddevice", 1, 1, ADDR("/builtin/sound"));
3046 	if (PHANDLE_VALID(node)) {
3047 		rv = prom_getprop(node, "interrupts", prop, sizeof(prop));
3048 		if (rv == PROM_ERROR) {
3049 			prom_printf("Adding sound interrupts property\n");
3050 			prom_setprop(node, "/builtin/sound", "interrupts",
3051 				     sound_irq, sizeof(sound_irq));
3052 		}
3053 	}
3054 
3055 	/* Make sure ethernet phy-handle property exists */
3056 	fixup_device_tree_efika_add_phy();
3057 }
3058 #else
3059 #define fixup_device_tree_efika()
3060 #endif
3061 
3062 #ifdef CONFIG_PPC_PASEMI_NEMO
3063 /*
3064  * CFE supplied on Nemo is broken in several ways, biggest
3065  * problem is that it reassigns ISA interrupts to unused mpic ints.
3066  * Add an interrupt-controller property for the io-bridge to use
3067  * and correct the ints so we can attach them to an irq_domain
3068  */
3069 static void __init fixup_device_tree_pasemi(void)
3070 {
3071 	u32 interrupts[2], parent, rval, val = 0;
3072 	char *name, *pci_name;
3073 	phandle iob, node;
3074 
3075 	/* Find the root pci node */
3076 	name = "/pxp@0,e0000000";
3077 	iob = call_prom("finddevice", 1, 1, ADDR(name));
3078 	if (!PHANDLE_VALID(iob))
3079 		return;
3080 
3081 	/* check if interrupt-controller node set yet */
3082 	if (prom_getproplen(iob, "interrupt-controller") !=PROM_ERROR)
3083 		return;
3084 
3085 	prom_printf("adding interrupt-controller property for SB600...\n");
3086 
3087 	prom_setprop(iob, name, "interrupt-controller", &val, 0);
3088 
3089 	pci_name = "/pxp@0,e0000000/pci@11";
3090 	node = call_prom("finddevice", 1, 1, ADDR(pci_name));
3091 	parent = ADDR(iob);
3092 
3093 	for( ; prom_next_node(&node); ) {
3094 		/* scan each node for one with an interrupt */
3095 		if (!PHANDLE_VALID(node))
3096 			continue;
3097 
3098 		rval = prom_getproplen(node, "interrupts");
3099 		if (rval == 0 || rval == PROM_ERROR)
3100 			continue;
3101 
3102 		prom_getprop(node, "interrupts", &interrupts, sizeof(interrupts));
3103 		if ((interrupts[0] < 212) || (interrupts[0] > 222))
3104 			continue;
3105 
3106 		/* found a node, update both interrupts and interrupt-parent */
3107 		if ((interrupts[0] >= 212) && (interrupts[0] <= 215))
3108 			interrupts[0] -= 203;
3109 		if ((interrupts[0] >= 216) && (interrupts[0] <= 220))
3110 			interrupts[0] -= 213;
3111 		if (interrupts[0] == 221)
3112 			interrupts[0] = 14;
3113 		if (interrupts[0] == 222)
3114 			interrupts[0] = 8;
3115 
3116 		prom_setprop(node, pci_name, "interrupts", interrupts,
3117 					sizeof(interrupts));
3118 		prom_setprop(node, pci_name, "interrupt-parent", &parent,
3119 					sizeof(parent));
3120 	}
3121 
3122 	/*
3123 	 * The io-bridge has device_type set to 'io-bridge' change it to 'isa'
3124 	 * so that generic isa-bridge code can add the SB600 and its on-board
3125 	 * peripherals.
3126 	 */
3127 	name = "/pxp@0,e0000000/io-bridge@0";
3128 	iob = call_prom("finddevice", 1, 1, ADDR(name));
3129 	if (!PHANDLE_VALID(iob))
3130 		return;
3131 
3132 	/* device_type is already set, just change it. */
3133 
3134 	prom_printf("Changing device_type of SB600 node...\n");
3135 
3136 	prom_setprop(iob, name, "device_type", "isa", sizeof("isa"));
3137 }
3138 #else	/* !CONFIG_PPC_PASEMI_NEMO */
3139 static inline void fixup_device_tree_pasemi(void) { }
3140 #endif
3141 
3142 static void __init fixup_device_tree(void)
3143 {
3144 	fixup_device_tree_maple();
3145 	fixup_device_tree_maple_memory_controller();
3146 	fixup_device_tree_chrp();
3147 	fixup_device_tree_pmac();
3148 	fixup_device_tree_efika();
3149 	fixup_device_tree_pasemi();
3150 }
3151 
3152 static void __init prom_find_boot_cpu(void)
3153 {
3154 	__be32 rval;
3155 	ihandle prom_cpu;
3156 	phandle cpu_pkg;
3157 
3158 	rval = 0;
3159 	if (prom_getprop(prom.chosen, "cpu", &rval, sizeof(rval)) <= 0)
3160 		return;
3161 	prom_cpu = be32_to_cpu(rval);
3162 
3163 	cpu_pkg = call_prom("instance-to-package", 1, 1, prom_cpu);
3164 
3165 	if (!PHANDLE_VALID(cpu_pkg))
3166 		return;
3167 
3168 	prom_getprop(cpu_pkg, "reg", &rval, sizeof(rval));
3169 	prom.cpu = be32_to_cpu(rval);
3170 
3171 	prom_debug("Booting CPU hw index = %d\n", prom.cpu);
3172 }
3173 
3174 static void __init prom_check_initrd(unsigned long r3, unsigned long r4)
3175 {
3176 #ifdef CONFIG_BLK_DEV_INITRD
3177 	if (r3 && r4 && r4 != 0xdeadbeef) {
3178 		__be64 val;
3179 
3180 		prom_initrd_start = is_kernel_addr(r3) ? __pa(r3) : r3;
3181 		prom_initrd_end = prom_initrd_start + r4;
3182 
3183 		val = cpu_to_be64(prom_initrd_start);
3184 		prom_setprop(prom.chosen, "/chosen", "linux,initrd-start",
3185 			     &val, sizeof(val));
3186 		val = cpu_to_be64(prom_initrd_end);
3187 		prom_setprop(prom.chosen, "/chosen", "linux,initrd-end",
3188 			     &val, sizeof(val));
3189 
3190 		reserve_mem(prom_initrd_start,
3191 			    prom_initrd_end - prom_initrd_start);
3192 
3193 		prom_debug("initrd_start=0x%lx\n", prom_initrd_start);
3194 		prom_debug("initrd_end=0x%lx\n", prom_initrd_end);
3195 	}
3196 #endif /* CONFIG_BLK_DEV_INITRD */
3197 }
3198 
3199 #ifdef CONFIG_PPC64
3200 #ifdef CONFIG_RELOCATABLE
3201 static void reloc_toc(void)
3202 {
3203 }
3204 
3205 static void unreloc_toc(void)
3206 {
3207 }
3208 #else
3209 static void __reloc_toc(unsigned long offset, unsigned long nr_entries)
3210 {
3211 	unsigned long i;
3212 	unsigned long *toc_entry;
3213 
3214 	/* Get the start of the TOC by using r2 directly. */
3215 	asm volatile("addi %0,2,-0x8000" : "=b" (toc_entry));
3216 
3217 	for (i = 0; i < nr_entries; i++) {
3218 		*toc_entry = *toc_entry + offset;
3219 		toc_entry++;
3220 	}
3221 }
3222 
3223 static void reloc_toc(void)
3224 {
3225 	unsigned long offset = reloc_offset();
3226 	unsigned long nr_entries =
3227 		(__prom_init_toc_end - __prom_init_toc_start) / sizeof(long);
3228 
3229 	__reloc_toc(offset, nr_entries);
3230 
3231 	mb();
3232 }
3233 
3234 static void unreloc_toc(void)
3235 {
3236 	unsigned long offset = reloc_offset();
3237 	unsigned long nr_entries =
3238 		(__prom_init_toc_end - __prom_init_toc_start) / sizeof(long);
3239 
3240 	mb();
3241 
3242 	__reloc_toc(-offset, nr_entries);
3243 }
3244 #endif
3245 #endif
3246 
3247 #ifdef CONFIG_PPC_SVM
3248 /*
3249  * Perform the Enter Secure Mode ultracall.
3250  */
3251 static int enter_secure_mode(unsigned long kbase, unsigned long fdt)
3252 {
3253 	register unsigned long r3 asm("r3") = UV_ESM;
3254 	register unsigned long r4 asm("r4") = kbase;
3255 	register unsigned long r5 asm("r5") = fdt;
3256 
3257 	asm volatile("sc 2" : "+r"(r3) : "r"(r4), "r"(r5));
3258 
3259 	return r3;
3260 }
3261 
3262 /*
3263  * Call the Ultravisor to transfer us to secure memory if we have an ESM blob.
3264  */
3265 static void setup_secure_guest(unsigned long kbase, unsigned long fdt)
3266 {
3267 	int ret;
3268 
3269 	if (!prom_svm_enable)
3270 		return;
3271 
3272 	/* Switch to secure mode. */
3273 	prom_printf("Switching to secure mode.\n");
3274 
3275 	/*
3276 	 * The ultravisor will do an integrity check of the kernel image but we
3277 	 * relocated it so the check will fail. Restore the original image by
3278 	 * relocating it back to the kernel virtual base address.
3279 	 */
3280 	if (IS_ENABLED(CONFIG_RELOCATABLE))
3281 		relocate(KERNELBASE);
3282 
3283 	ret = enter_secure_mode(kbase, fdt);
3284 
3285 	/* Relocate the kernel again. */
3286 	if (IS_ENABLED(CONFIG_RELOCATABLE))
3287 		relocate(kbase);
3288 
3289 	if (ret != U_SUCCESS) {
3290 		prom_printf("Returned %d from switching to secure mode.\n", ret);
3291 		prom_rtas_os_term("Switch to secure mode failed.\n");
3292 	}
3293 }
3294 #else
3295 static void setup_secure_guest(unsigned long kbase, unsigned long fdt)
3296 {
3297 }
3298 #endif /* CONFIG_PPC_SVM */
3299 
3300 /*
3301  * We enter here early on, when the Open Firmware prom is still
3302  * handling exceptions and the MMU hash table for us.
3303  */
3304 
3305 unsigned long __init prom_init(unsigned long r3, unsigned long r4,
3306 			       unsigned long pp,
3307 			       unsigned long r6, unsigned long r7,
3308 			       unsigned long kbase)
3309 {
3310 	unsigned long hdr;
3311 
3312 #ifdef CONFIG_PPC32
3313 	unsigned long offset = reloc_offset();
3314 	reloc_got2(offset);
3315 #else
3316 	reloc_toc();
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 instanciating 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 #else
3494 	unreloc_toc();
3495 #endif
3496 
3497 	/* Move to secure memory if we're supposed to be secure guests. */
3498 	setup_secure_guest(kbase, hdr);
3499 
3500 	__start(hdr, kbase, 0, 0, 0, 0, 0);
3501 
3502 	return 0;
3503 }
3504