xref: /openbmc/linux/arch/powerpc/kernel/prom_init.c (revision ccd51b9f)
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 <asm/prom.h>
30 #include <asm/rtas.h>
31 #include <asm/page.h>
32 #include <asm/processor.h>
33 #include <asm/irq.h>
34 #include <asm/io.h>
35 #include <asm/smp.h>
36 #include <asm/mmu.h>
37 #include <asm/pgtable.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 
44 #include <linux/linux_logo.h>
45 
46 /* All of prom_init bss lives here */
47 #define __prombss __section(.bss.prominit)
48 
49 /*
50  * Eventually bump that one up
51  */
52 #define DEVTREE_CHUNK_SIZE	0x100000
53 
54 /*
55  * This is the size of the local memory reserve map that gets copied
56  * into the boot params passed to the kernel. That size is totally
57  * flexible as the kernel just reads the list until it encounters an
58  * entry with size 0, so it can be changed without breaking binary
59  * compatibility
60  */
61 #define MEM_RESERVE_MAP_SIZE	8
62 
63 /*
64  * prom_init() is called very early on, before the kernel text
65  * and data have been mapped to KERNELBASE.  At this point the code
66  * is running at whatever address it has been loaded at.
67  * On ppc32 we compile with -mrelocatable, which means that references
68  * to extern and static variables get relocated automatically.
69  * ppc64 objects are always relocatable, we just need to relocate the
70  * TOC.
71  *
72  * Because OF may have mapped I/O devices into the area starting at
73  * KERNELBASE, particularly on CHRP machines, we can't safely call
74  * OF once the kernel has been mapped to KERNELBASE.  Therefore all
75  * OF calls must be done within prom_init().
76  *
77  * ADDR is used in calls to call_prom.  The 4th and following
78  * arguments to call_prom should be 32-bit values.
79  * On ppc64, 64 bit values are truncated to 32 bits (and
80  * fortunately don't get interpreted as two arguments).
81  */
82 #define ADDR(x)		(u32)(unsigned long)(x)
83 
84 #ifdef CONFIG_PPC64
85 #define OF_WORKAROUNDS	0
86 #else
87 #define OF_WORKAROUNDS	of_workarounds
88 static int of_workarounds __prombss;
89 #endif
90 
91 #define OF_WA_CLAIM	1	/* do phys/virt claim separately, then map */
92 #define OF_WA_LONGTRAIL	2	/* work around longtrail bugs */
93 
94 #define PROM_BUG() do {						\
95         prom_printf("kernel BUG at %s line 0x%x!\n",		\
96 		    __FILE__, __LINE__);			\
97         __asm__ __volatile__(".long " BUG_ILLEGAL_INSTR);	\
98 } while (0)
99 
100 #ifdef DEBUG_PROM
101 #define prom_debug(x...)	prom_printf(x)
102 #else
103 #define prom_debug(x...)	do { } while (0)
104 #endif
105 
106 
107 typedef u32 prom_arg_t;
108 
109 struct prom_args {
110         __be32 service;
111         __be32 nargs;
112         __be32 nret;
113         __be32 args[10];
114 };
115 
116 struct prom_t {
117 	ihandle root;
118 	phandle chosen;
119 	int cpu;
120 	ihandle stdout;
121 	ihandle mmumap;
122 	ihandle memory;
123 };
124 
125 struct mem_map_entry {
126 	__be64	base;
127 	__be64	size;
128 };
129 
130 typedef __be32 cell_t;
131 
132 extern void __start(unsigned long r3, unsigned long r4, unsigned long r5,
133 		    unsigned long r6, unsigned long r7, unsigned long r8,
134 		    unsigned long r9);
135 
136 #ifdef CONFIG_PPC64
137 extern int enter_prom(struct prom_args *args, unsigned long entry);
138 #else
139 static inline int enter_prom(struct prom_args *args, unsigned long entry)
140 {
141 	return ((int (*)(struct prom_args *))entry)(args);
142 }
143 #endif
144 
145 extern void copy_and_flush(unsigned long dest, unsigned long src,
146 			   unsigned long size, unsigned long offset);
147 
148 /* prom structure */
149 static struct prom_t __prombss prom;
150 
151 static unsigned long __prombss prom_entry;
152 
153 static char __prombss of_stdout_device[256];
154 static char __prombss prom_scratch[256];
155 
156 static unsigned long __prombss dt_header_start;
157 static unsigned long __prombss dt_struct_start, dt_struct_end;
158 static unsigned long __prombss dt_string_start, dt_string_end;
159 
160 static unsigned long __prombss prom_initrd_start, prom_initrd_end;
161 
162 #ifdef CONFIG_PPC64
163 static int __prombss prom_iommu_force_on;
164 static int __prombss prom_iommu_off;
165 static unsigned long __prombss prom_tce_alloc_start;
166 static unsigned long __prombss prom_tce_alloc_end;
167 #endif
168 
169 #ifdef CONFIG_PPC_PSERIES
170 static bool __prombss prom_radix_disable;
171 static bool __prombss prom_xive_disable;
172 #endif
173 
174 struct platform_support {
175 	bool hash_mmu;
176 	bool radix_mmu;
177 	bool radix_gtse;
178 	bool xive;
179 };
180 
181 /* Platforms codes are now obsolete in the kernel. Now only used within this
182  * file and ultimately gone too. Feel free to change them if you need, they
183  * are not shared with anything outside of this file anymore
184  */
185 #define PLATFORM_PSERIES	0x0100
186 #define PLATFORM_PSERIES_LPAR	0x0101
187 #define PLATFORM_LPAR		0x0001
188 #define PLATFORM_POWERMAC	0x0400
189 #define PLATFORM_GENERIC	0x0500
190 
191 static int __prombss of_platform;
192 
193 static char __prombss prom_cmd_line[COMMAND_LINE_SIZE];
194 
195 static unsigned long __prombss prom_memory_limit;
196 
197 static unsigned long __prombss alloc_top;
198 static unsigned long __prombss alloc_top_high;
199 static unsigned long __prombss alloc_bottom;
200 static unsigned long __prombss rmo_top;
201 static unsigned long __prombss ram_top;
202 
203 static struct mem_map_entry __prombss mem_reserve_map[MEM_RESERVE_MAP_SIZE];
204 static int __prombss mem_reserve_cnt;
205 
206 static cell_t __prombss regbuf[1024];
207 
208 static bool  __prombss rtas_has_query_cpu_stopped;
209 
210 
211 /*
212  * Error results ... some OF calls will return "-1" on error, some
213  * will return 0, some will return either. To simplify, here are
214  * macros to use with any ihandle or phandle return value to check if
215  * it is valid
216  */
217 
218 #define PROM_ERROR		(-1u)
219 #define PHANDLE_VALID(p)	((p) != 0 && (p) != PROM_ERROR)
220 #define IHANDLE_VALID(i)	((i) != 0 && (i) != PROM_ERROR)
221 
222 /* Copied from lib/string.c and lib/kstrtox.c */
223 
224 static int __init prom_strcmp(const char *cs, const char *ct)
225 {
226 	unsigned char c1, c2;
227 
228 	while (1) {
229 		c1 = *cs++;
230 		c2 = *ct++;
231 		if (c1 != c2)
232 			return c1 < c2 ? -1 : 1;
233 		if (!c1)
234 			break;
235 	}
236 	return 0;
237 }
238 
239 static char __init *prom_strcpy(char *dest, const char *src)
240 {
241 	char *tmp = dest;
242 
243 	while ((*dest++ = *src++) != '\0')
244 		/* nothing */;
245 	return tmp;
246 }
247 
248 static int __init prom_strncmp(const char *cs, const char *ct, size_t count)
249 {
250 	unsigned char c1, c2;
251 
252 	while (count) {
253 		c1 = *cs++;
254 		c2 = *ct++;
255 		if (c1 != c2)
256 			return c1 < c2 ? -1 : 1;
257 		if (!c1)
258 			break;
259 		count--;
260 	}
261 	return 0;
262 }
263 
264 static size_t __init prom_strlen(const char *s)
265 {
266 	const char *sc;
267 
268 	for (sc = s; *sc != '\0'; ++sc)
269 		/* nothing */;
270 	return sc - s;
271 }
272 
273 static int __init prom_memcmp(const void *cs, const void *ct, size_t count)
274 {
275 	const unsigned char *su1, *su2;
276 	int res = 0;
277 
278 	for (su1 = cs, su2 = ct; 0 < count; ++su1, ++su2, count--)
279 		if ((res = *su1 - *su2) != 0)
280 			break;
281 	return res;
282 }
283 
284 static char __init *prom_strstr(const char *s1, const char *s2)
285 {
286 	size_t l1, l2;
287 
288 	l2 = prom_strlen(s2);
289 	if (!l2)
290 		return (char *)s1;
291 	l1 = prom_strlen(s1);
292 	while (l1 >= l2) {
293 		l1--;
294 		if (!prom_memcmp(s1, s2, l2))
295 			return (char *)s1;
296 		s1++;
297 	}
298 	return NULL;
299 }
300 
301 static size_t __init prom_strlcpy(char *dest, const char *src, size_t size)
302 {
303 	size_t ret = prom_strlen(src);
304 
305 	if (size) {
306 		size_t len = (ret >= size) ? size - 1 : ret;
307 		memcpy(dest, src, len);
308 		dest[len] = '\0';
309 	}
310 	return ret;
311 }
312 
313 #ifdef CONFIG_PPC_PSERIES
314 static int __init prom_strtobool(const char *s, bool *res)
315 {
316 	if (!s)
317 		return -EINVAL;
318 
319 	switch (s[0]) {
320 	case 'y':
321 	case 'Y':
322 	case '1':
323 		*res = true;
324 		return 0;
325 	case 'n':
326 	case 'N':
327 	case '0':
328 		*res = false;
329 		return 0;
330 	case 'o':
331 	case 'O':
332 		switch (s[1]) {
333 		case 'n':
334 		case 'N':
335 			*res = true;
336 			return 0;
337 		case 'f':
338 		case 'F':
339 			*res = false;
340 			return 0;
341 		default:
342 			break;
343 		}
344 	default:
345 		break;
346 	}
347 
348 	return -EINVAL;
349 }
350 #endif
351 
352 /* This is the one and *ONLY* place where we actually call open
353  * firmware.
354  */
355 
356 static int __init call_prom(const char *service, int nargs, int nret, ...)
357 {
358 	int i;
359 	struct prom_args args;
360 	va_list list;
361 
362 	args.service = cpu_to_be32(ADDR(service));
363 	args.nargs = cpu_to_be32(nargs);
364 	args.nret = cpu_to_be32(nret);
365 
366 	va_start(list, nret);
367 	for (i = 0; i < nargs; i++)
368 		args.args[i] = cpu_to_be32(va_arg(list, prom_arg_t));
369 	va_end(list);
370 
371 	for (i = 0; i < nret; i++)
372 		args.args[nargs+i] = 0;
373 
374 	if (enter_prom(&args, prom_entry) < 0)
375 		return PROM_ERROR;
376 
377 	return (nret > 0) ? be32_to_cpu(args.args[nargs]) : 0;
378 }
379 
380 static int __init call_prom_ret(const char *service, int nargs, int nret,
381 				prom_arg_t *rets, ...)
382 {
383 	int i;
384 	struct prom_args args;
385 	va_list list;
386 
387 	args.service = cpu_to_be32(ADDR(service));
388 	args.nargs = cpu_to_be32(nargs);
389 	args.nret = cpu_to_be32(nret);
390 
391 	va_start(list, rets);
392 	for (i = 0; i < nargs; i++)
393 		args.args[i] = cpu_to_be32(va_arg(list, prom_arg_t));
394 	va_end(list);
395 
396 	for (i = 0; i < nret; i++)
397 		args.args[nargs+i] = 0;
398 
399 	if (enter_prom(&args, prom_entry) < 0)
400 		return PROM_ERROR;
401 
402 	if (rets != NULL)
403 		for (i = 1; i < nret; ++i)
404 			rets[i-1] = be32_to_cpu(args.args[nargs+i]);
405 
406 	return (nret > 0) ? be32_to_cpu(args.args[nargs]) : 0;
407 }
408 
409 
410 static void __init prom_print(const char *msg)
411 {
412 	const char *p, *q;
413 
414 	if (prom.stdout == 0)
415 		return;
416 
417 	for (p = msg; *p != 0; p = q) {
418 		for (q = p; *q != 0 && *q != '\n'; ++q)
419 			;
420 		if (q > p)
421 			call_prom("write", 3, 1, prom.stdout, p, q - p);
422 		if (*q == 0)
423 			break;
424 		++q;
425 		call_prom("write", 3, 1, prom.stdout, ADDR("\r\n"), 2);
426 	}
427 }
428 
429 
430 /*
431  * Both prom_print_hex & prom_print_dec takes an unsigned long as input so that
432  * we do not need __udivdi3 or __umoddi3 on 32bits.
433  */
434 static void __init prom_print_hex(unsigned long val)
435 {
436 	int i, nibbles = sizeof(val)*2;
437 	char buf[sizeof(val)*2+1];
438 
439 	for (i = nibbles-1;  i >= 0;  i--) {
440 		buf[i] = (val & 0xf) + '0';
441 		if (buf[i] > '9')
442 			buf[i] += ('a'-'0'-10);
443 		val >>= 4;
444 	}
445 	buf[nibbles] = '\0';
446 	call_prom("write", 3, 1, prom.stdout, buf, nibbles);
447 }
448 
449 /* max number of decimal digits in an unsigned long */
450 #define UL_DIGITS 21
451 static void __init prom_print_dec(unsigned long val)
452 {
453 	int i, size;
454 	char buf[UL_DIGITS+1];
455 
456 	for (i = UL_DIGITS-1; i >= 0;  i--) {
457 		buf[i] = (val % 10) + '0';
458 		val = val/10;
459 		if (val == 0)
460 			break;
461 	}
462 	/* shift stuff down */
463 	size = UL_DIGITS - i;
464 	call_prom("write", 3, 1, prom.stdout, buf+i, size);
465 }
466 
467 __printf(1, 2)
468 static void __init prom_printf(const char *format, ...)
469 {
470 	const char *p, *q, *s;
471 	va_list args;
472 	unsigned long v;
473 	long vs;
474 	int n = 0;
475 
476 	va_start(args, format);
477 	for (p = format; *p != 0; p = q) {
478 		for (q = p; *q != 0 && *q != '\n' && *q != '%'; ++q)
479 			;
480 		if (q > p)
481 			call_prom("write", 3, 1, prom.stdout, p, q - p);
482 		if (*q == 0)
483 			break;
484 		if (*q == '\n') {
485 			++q;
486 			call_prom("write", 3, 1, prom.stdout,
487 				  ADDR("\r\n"), 2);
488 			continue;
489 		}
490 		++q;
491 		if (*q == 0)
492 			break;
493 		while (*q == 'l') {
494 			++q;
495 			++n;
496 		}
497 		switch (*q) {
498 		case 's':
499 			++q;
500 			s = va_arg(args, const char *);
501 			prom_print(s);
502 			break;
503 		case 'x':
504 			++q;
505 			switch (n) {
506 			case 0:
507 				v = va_arg(args, unsigned int);
508 				break;
509 			case 1:
510 				v = va_arg(args, unsigned long);
511 				break;
512 			case 2:
513 			default:
514 				v = va_arg(args, unsigned long long);
515 				break;
516 			}
517 			prom_print_hex(v);
518 			break;
519 		case 'u':
520 			++q;
521 			switch (n) {
522 			case 0:
523 				v = va_arg(args, unsigned int);
524 				break;
525 			case 1:
526 				v = va_arg(args, unsigned long);
527 				break;
528 			case 2:
529 			default:
530 				v = va_arg(args, unsigned long long);
531 				break;
532 			}
533 			prom_print_dec(v);
534 			break;
535 		case 'd':
536 			++q;
537 			switch (n) {
538 			case 0:
539 				vs = va_arg(args, int);
540 				break;
541 			case 1:
542 				vs = va_arg(args, long);
543 				break;
544 			case 2:
545 			default:
546 				vs = va_arg(args, long long);
547 				break;
548 			}
549 			if (vs < 0) {
550 				prom_print("-");
551 				vs = -vs;
552 			}
553 			prom_print_dec(vs);
554 			break;
555 		}
556 	}
557 	va_end(args);
558 }
559 
560 
561 static unsigned int __init prom_claim(unsigned long virt, unsigned long size,
562 				unsigned long align)
563 {
564 
565 	if (align == 0 && (OF_WORKAROUNDS & OF_WA_CLAIM)) {
566 		/*
567 		 * Old OF requires we claim physical and virtual separately
568 		 * and then map explicitly (assuming virtual mode)
569 		 */
570 		int ret;
571 		prom_arg_t result;
572 
573 		ret = call_prom_ret("call-method", 5, 2, &result,
574 				    ADDR("claim"), prom.memory,
575 				    align, size, virt);
576 		if (ret != 0 || result == -1)
577 			return -1;
578 		ret = call_prom_ret("call-method", 5, 2, &result,
579 				    ADDR("claim"), prom.mmumap,
580 				    align, size, virt);
581 		if (ret != 0) {
582 			call_prom("call-method", 4, 1, ADDR("release"),
583 				  prom.memory, size, virt);
584 			return -1;
585 		}
586 		/* the 0x12 is M (coherence) + PP == read/write */
587 		call_prom("call-method", 6, 1,
588 			  ADDR("map"), prom.mmumap, 0x12, size, virt, virt);
589 		return virt;
590 	}
591 	return call_prom("claim", 3, 1, (prom_arg_t)virt, (prom_arg_t)size,
592 			 (prom_arg_t)align);
593 }
594 
595 static void __init __attribute__((noreturn)) prom_panic(const char *reason)
596 {
597 	prom_print(reason);
598 	/* Do not call exit because it clears the screen on pmac
599 	 * it also causes some sort of double-fault on early pmacs */
600 	if (of_platform == PLATFORM_POWERMAC)
601 		asm("trap\n");
602 
603 	/* ToDo: should put up an SRC here on pSeries */
604 	call_prom("exit", 0, 0);
605 
606 	for (;;)			/* should never get here */
607 		;
608 }
609 
610 
611 static int __init prom_next_node(phandle *nodep)
612 {
613 	phandle node;
614 
615 	if ((node = *nodep) != 0
616 	    && (*nodep = call_prom("child", 1, 1, node)) != 0)
617 		return 1;
618 	if ((*nodep = call_prom("peer", 1, 1, node)) != 0)
619 		return 1;
620 	for (;;) {
621 		if ((node = call_prom("parent", 1, 1, node)) == 0)
622 			return 0;
623 		if ((*nodep = call_prom("peer", 1, 1, node)) != 0)
624 			return 1;
625 	}
626 }
627 
628 static inline int __init prom_getprop(phandle node, const char *pname,
629 				      void *value, size_t valuelen)
630 {
631 	return call_prom("getprop", 4, 1, node, ADDR(pname),
632 			 (u32)(unsigned long) value, (u32) valuelen);
633 }
634 
635 static inline int __init prom_getproplen(phandle node, const char *pname)
636 {
637 	return call_prom("getproplen", 2, 1, node, ADDR(pname));
638 }
639 
640 static void add_string(char **str, const char *q)
641 {
642 	char *p = *str;
643 
644 	while (*q)
645 		*p++ = *q++;
646 	*p++ = ' ';
647 	*str = p;
648 }
649 
650 static char *tohex(unsigned int x)
651 {
652 	static const char digits[] __initconst = "0123456789abcdef";
653 	static char result[9] __prombss;
654 	int i;
655 
656 	result[8] = 0;
657 	i = 8;
658 	do {
659 		--i;
660 		result[i] = digits[x & 0xf];
661 		x >>= 4;
662 	} while (x != 0 && i > 0);
663 	return &result[i];
664 }
665 
666 static int __init prom_setprop(phandle node, const char *nodename,
667 			       const char *pname, void *value, size_t valuelen)
668 {
669 	char cmd[256], *p;
670 
671 	if (!(OF_WORKAROUNDS & OF_WA_LONGTRAIL))
672 		return call_prom("setprop", 4, 1, node, ADDR(pname),
673 				 (u32)(unsigned long) value, (u32) valuelen);
674 
675 	/* gah... setprop doesn't work on longtrail, have to use interpret */
676 	p = cmd;
677 	add_string(&p, "dev");
678 	add_string(&p, nodename);
679 	add_string(&p, tohex((u32)(unsigned long) value));
680 	add_string(&p, tohex(valuelen));
681 	add_string(&p, tohex(ADDR(pname)));
682 	add_string(&p, tohex(prom_strlen(pname)));
683 	add_string(&p, "property");
684 	*p = 0;
685 	return call_prom("interpret", 1, 1, (u32)(unsigned long) cmd);
686 }
687 
688 /* We can't use the standard versions because of relocation headaches. */
689 #define isxdigit(c)	(('0' <= (c) && (c) <= '9') \
690 			 || ('a' <= (c) && (c) <= 'f') \
691 			 || ('A' <= (c) && (c) <= 'F'))
692 
693 #define isdigit(c)	('0' <= (c) && (c) <= '9')
694 #define islower(c)	('a' <= (c) && (c) <= 'z')
695 #define toupper(c)	(islower(c) ? ((c) - 'a' + 'A') : (c))
696 
697 static unsigned long prom_strtoul(const char *cp, const char **endp)
698 {
699 	unsigned long result = 0, base = 10, value;
700 
701 	if (*cp == '0') {
702 		base = 8;
703 		cp++;
704 		if (toupper(*cp) == 'X') {
705 			cp++;
706 			base = 16;
707 		}
708 	}
709 
710 	while (isxdigit(*cp) &&
711 	       (value = isdigit(*cp) ? *cp - '0' : toupper(*cp) - 'A' + 10) < base) {
712 		result = result * base + value;
713 		cp++;
714 	}
715 
716 	if (endp)
717 		*endp = cp;
718 
719 	return result;
720 }
721 
722 static unsigned long prom_memparse(const char *ptr, const char **retptr)
723 {
724 	unsigned long ret = prom_strtoul(ptr, retptr);
725 	int shift = 0;
726 
727 	/*
728 	 * We can't use a switch here because GCC *may* generate a
729 	 * jump table which won't work, because we're not running at
730 	 * the address we're linked at.
731 	 */
732 	if ('G' == **retptr || 'g' == **retptr)
733 		shift = 30;
734 
735 	if ('M' == **retptr || 'm' == **retptr)
736 		shift = 20;
737 
738 	if ('K' == **retptr || 'k' == **retptr)
739 		shift = 10;
740 
741 	if (shift) {
742 		ret <<= shift;
743 		(*retptr)++;
744 	}
745 
746 	return ret;
747 }
748 
749 /*
750  * Early parsing of the command line passed to the kernel, used for
751  * "mem=x" and the options that affect the iommu
752  */
753 static void __init early_cmdline_parse(void)
754 {
755 	const char *opt;
756 
757 	char *p;
758 	int l = 0;
759 
760 	prom_cmd_line[0] = 0;
761 	p = prom_cmd_line;
762 	if ((long)prom.chosen > 0)
763 		l = prom_getprop(prom.chosen, "bootargs", p, COMMAND_LINE_SIZE-1);
764 	if (IS_ENABLED(CONFIG_CMDLINE_BOOL) && (l <= 0 || p[0] == '\0')) /* dbl check */
765 		prom_strlcpy(prom_cmd_line, CONFIG_CMDLINE, sizeof(prom_cmd_line));
766 	prom_printf("command line: %s\n", prom_cmd_line);
767 
768 #ifdef CONFIG_PPC64
769 	opt = prom_strstr(prom_cmd_line, "iommu=");
770 	if (opt) {
771 		prom_printf("iommu opt is: %s\n", opt);
772 		opt += 6;
773 		while (*opt && *opt == ' ')
774 			opt++;
775 		if (!prom_strncmp(opt, "off", 3))
776 			prom_iommu_off = 1;
777 		else if (!prom_strncmp(opt, "force", 5))
778 			prom_iommu_force_on = 1;
779 	}
780 #endif
781 	opt = prom_strstr(prom_cmd_line, "mem=");
782 	if (opt) {
783 		opt += 4;
784 		prom_memory_limit = prom_memparse(opt, (const char **)&opt);
785 #ifdef CONFIG_PPC64
786 		/* Align to 16 MB == size of ppc64 large page */
787 		prom_memory_limit = ALIGN(prom_memory_limit, 0x1000000);
788 #endif
789 	}
790 
791 #ifdef CONFIG_PPC_PSERIES
792 	prom_radix_disable = !IS_ENABLED(CONFIG_PPC_RADIX_MMU_DEFAULT);
793 	opt = prom_strstr(prom_cmd_line, "disable_radix");
794 	if (opt) {
795 		opt += 13;
796 		if (*opt && *opt == '=') {
797 			bool val;
798 
799 			if (prom_strtobool(++opt, &val))
800 				prom_radix_disable = false;
801 			else
802 				prom_radix_disable = val;
803 		} else
804 			prom_radix_disable = true;
805 	}
806 	if (prom_radix_disable)
807 		prom_debug("Radix disabled from cmdline\n");
808 
809 	opt = prom_strstr(prom_cmd_line, "xive=off");
810 	if (opt) {
811 		prom_xive_disable = true;
812 		prom_debug("XIVE disabled from cmdline\n");
813 	}
814 #endif /* CONFIG_PPC_PSERIES */
815 }
816 
817 #ifdef CONFIG_PPC_PSERIES
818 /*
819  * The architecture vector has an array of PVR mask/value pairs,
820  * followed by # option vectors - 1, followed by the option vectors.
821  *
822  * See prom.h for the definition of the bits specified in the
823  * architecture vector.
824  */
825 
826 /* Firmware expects the value to be n - 1, where n is the # of vectors */
827 #define NUM_VECTORS(n)		((n) - 1)
828 
829 /*
830  * Firmware expects 1 + n - 2, where n is the length of the option vector in
831  * bytes. The 1 accounts for the length byte itself, the - 2 .. ?
832  */
833 #define VECTOR_LENGTH(n)	(1 + (n) - 2)
834 
835 struct option_vector1 {
836 	u8 byte1;
837 	u8 arch_versions;
838 	u8 arch_versions3;
839 } __packed;
840 
841 struct option_vector2 {
842 	u8 byte1;
843 	__be16 reserved;
844 	__be32 real_base;
845 	__be32 real_size;
846 	__be32 virt_base;
847 	__be32 virt_size;
848 	__be32 load_base;
849 	__be32 min_rma;
850 	__be32 min_load;
851 	u8 min_rma_percent;
852 	u8 max_pft_size;
853 } __packed;
854 
855 struct option_vector3 {
856 	u8 byte1;
857 	u8 byte2;
858 } __packed;
859 
860 struct option_vector4 {
861 	u8 byte1;
862 	u8 min_vp_cap;
863 } __packed;
864 
865 struct option_vector5 {
866 	u8 byte1;
867 	u8 byte2;
868 	u8 byte3;
869 	u8 cmo;
870 	u8 associativity;
871 	u8 bin_opts;
872 	u8 micro_checkpoint;
873 	u8 reserved0;
874 	__be32 max_cpus;
875 	__be16 papr_level;
876 	__be16 reserved1;
877 	u8 platform_facilities;
878 	u8 reserved2;
879 	__be16 reserved3;
880 	u8 subprocessors;
881 	u8 byte22;
882 	u8 intarch;
883 	u8 mmu;
884 	u8 hash_ext;
885 	u8 radix_ext;
886 } __packed;
887 
888 struct option_vector6 {
889 	u8 reserved;
890 	u8 secondary_pteg;
891 	u8 os_name;
892 } __packed;
893 
894 struct ibm_arch_vec {
895 	struct { u32 mask, val; } pvrs[12];
896 
897 	u8 num_vectors;
898 
899 	u8 vec1_len;
900 	struct option_vector1 vec1;
901 
902 	u8 vec2_len;
903 	struct option_vector2 vec2;
904 
905 	u8 vec3_len;
906 	struct option_vector3 vec3;
907 
908 	u8 vec4_len;
909 	struct option_vector4 vec4;
910 
911 	u8 vec5_len;
912 	struct option_vector5 vec5;
913 
914 	u8 vec6_len;
915 	struct option_vector6 vec6;
916 } __packed;
917 
918 static const struct ibm_arch_vec ibm_architecture_vec_template __initconst = {
919 	.pvrs = {
920 		{
921 			.mask = cpu_to_be32(0xfffe0000), /* POWER5/POWER5+ */
922 			.val  = cpu_to_be32(0x003a0000),
923 		},
924 		{
925 			.mask = cpu_to_be32(0xffff0000), /* POWER6 */
926 			.val  = cpu_to_be32(0x003e0000),
927 		},
928 		{
929 			.mask = cpu_to_be32(0xffff0000), /* POWER7 */
930 			.val  = cpu_to_be32(0x003f0000),
931 		},
932 		{
933 			.mask = cpu_to_be32(0xffff0000), /* POWER8E */
934 			.val  = cpu_to_be32(0x004b0000),
935 		},
936 		{
937 			.mask = cpu_to_be32(0xffff0000), /* POWER8NVL */
938 			.val  = cpu_to_be32(0x004c0000),
939 		},
940 		{
941 			.mask = cpu_to_be32(0xffff0000), /* POWER8 */
942 			.val  = cpu_to_be32(0x004d0000),
943 		},
944 		{
945 			.mask = cpu_to_be32(0xffff0000), /* POWER9 */
946 			.val  = cpu_to_be32(0x004e0000),
947 		},
948 		{
949 			.mask = cpu_to_be32(0xffffffff), /* all 3.00-compliant */
950 			.val  = cpu_to_be32(0x0f000005),
951 		},
952 		{
953 			.mask = cpu_to_be32(0xffffffff), /* all 2.07-compliant */
954 			.val  = cpu_to_be32(0x0f000004),
955 		},
956 		{
957 			.mask = cpu_to_be32(0xffffffff), /* all 2.06-compliant */
958 			.val  = cpu_to_be32(0x0f000003),
959 		},
960 		{
961 			.mask = cpu_to_be32(0xffffffff), /* all 2.05-compliant */
962 			.val  = cpu_to_be32(0x0f000002),
963 		},
964 		{
965 			.mask = cpu_to_be32(0xfffffffe), /* all 2.04-compliant and earlier */
966 			.val  = cpu_to_be32(0x0f000001),
967 		},
968 	},
969 
970 	.num_vectors = NUM_VECTORS(6),
971 
972 	.vec1_len = VECTOR_LENGTH(sizeof(struct option_vector1)),
973 	.vec1 = {
974 		.byte1 = 0,
975 		.arch_versions = OV1_PPC_2_00 | OV1_PPC_2_01 | OV1_PPC_2_02 | OV1_PPC_2_03 |
976 				 OV1_PPC_2_04 | OV1_PPC_2_05 | OV1_PPC_2_06 | OV1_PPC_2_07,
977 		.arch_versions3 = OV1_PPC_3_00,
978 	},
979 
980 	.vec2_len = VECTOR_LENGTH(sizeof(struct option_vector2)),
981 	/* option vector 2: Open Firmware options supported */
982 	.vec2 = {
983 		.byte1 = OV2_REAL_MODE,
984 		.reserved = 0,
985 		.real_base = cpu_to_be32(0xffffffff),
986 		.real_size = cpu_to_be32(0xffffffff),
987 		.virt_base = cpu_to_be32(0xffffffff),
988 		.virt_size = cpu_to_be32(0xffffffff),
989 		.load_base = cpu_to_be32(0xffffffff),
990 		.min_rma = cpu_to_be32(512),		/* 512MB min RMA */
991 		.min_load = cpu_to_be32(0xffffffff),	/* full client load */
992 		.min_rma_percent = 0,	/* min RMA percentage of total RAM */
993 		.max_pft_size = 48,	/* max log_2(hash table size) */
994 	},
995 
996 	.vec3_len = VECTOR_LENGTH(sizeof(struct option_vector3)),
997 	/* option vector 3: processor options supported */
998 	.vec3 = {
999 		.byte1 = 0,			/* don't ignore, don't halt */
1000 		.byte2 = OV3_FP | OV3_VMX | OV3_DFP,
1001 	},
1002 
1003 	.vec4_len = VECTOR_LENGTH(sizeof(struct option_vector4)),
1004 	/* option vector 4: IBM PAPR implementation */
1005 	.vec4 = {
1006 		.byte1 = 0,			/* don't halt */
1007 		.min_vp_cap = OV4_MIN_ENT_CAP,	/* minimum VP entitled capacity */
1008 	},
1009 
1010 	.vec5_len = VECTOR_LENGTH(sizeof(struct option_vector5)),
1011 	/* option vector 5: PAPR/OF options */
1012 	.vec5 = {
1013 		.byte1 = 0,				/* don't ignore, don't halt */
1014 		.byte2 = OV5_FEAT(OV5_LPAR) | OV5_FEAT(OV5_SPLPAR) | OV5_FEAT(OV5_LARGE_PAGES) |
1015 		OV5_FEAT(OV5_DRCONF_MEMORY) | OV5_FEAT(OV5_DONATE_DEDICATE_CPU) |
1016 #ifdef CONFIG_PCI_MSI
1017 		/* PCIe/MSI support.  Without MSI full PCIe is not supported */
1018 		OV5_FEAT(OV5_MSI),
1019 #else
1020 		0,
1021 #endif
1022 		.byte3 = 0,
1023 		.cmo =
1024 #ifdef CONFIG_PPC_SMLPAR
1025 		OV5_FEAT(OV5_CMO) | OV5_FEAT(OV5_XCMO),
1026 #else
1027 		0,
1028 #endif
1029 		.associativity = OV5_FEAT(OV5_TYPE1_AFFINITY) | OV5_FEAT(OV5_PRRN),
1030 		.bin_opts = OV5_FEAT(OV5_RESIZE_HPT) | OV5_FEAT(OV5_HP_EVT),
1031 		.micro_checkpoint = 0,
1032 		.reserved0 = 0,
1033 		.max_cpus = cpu_to_be32(NR_CPUS),	/* number of cores supported */
1034 		.papr_level = 0,
1035 		.reserved1 = 0,
1036 		.platform_facilities = OV5_FEAT(OV5_PFO_HW_RNG) | OV5_FEAT(OV5_PFO_HW_ENCR) | OV5_FEAT(OV5_PFO_HW_842),
1037 		.reserved2 = 0,
1038 		.reserved3 = 0,
1039 		.subprocessors = 1,
1040 		.byte22 = OV5_FEAT(OV5_DRMEM_V2),
1041 		.intarch = 0,
1042 		.mmu = 0,
1043 		.hash_ext = 0,
1044 		.radix_ext = 0,
1045 	},
1046 
1047 	/* option vector 6: IBM PAPR hints */
1048 	.vec6_len = VECTOR_LENGTH(sizeof(struct option_vector6)),
1049 	.vec6 = {
1050 		.reserved = 0,
1051 		.secondary_pteg = 0,
1052 		.os_name = OV6_LINUX,
1053 	},
1054 };
1055 
1056 static struct ibm_arch_vec __prombss ibm_architecture_vec  ____cacheline_aligned;
1057 
1058 /* Old method - ELF header with PT_NOTE sections only works on BE */
1059 #ifdef __BIG_ENDIAN__
1060 static const struct fake_elf {
1061 	Elf32_Ehdr	elfhdr;
1062 	Elf32_Phdr	phdr[2];
1063 	struct chrpnote {
1064 		u32	namesz;
1065 		u32	descsz;
1066 		u32	type;
1067 		char	name[8];	/* "PowerPC" */
1068 		struct chrpdesc {
1069 			u32	real_mode;
1070 			u32	real_base;
1071 			u32	real_size;
1072 			u32	virt_base;
1073 			u32	virt_size;
1074 			u32	load_base;
1075 		} chrpdesc;
1076 	} chrpnote;
1077 	struct rpanote {
1078 		u32	namesz;
1079 		u32	descsz;
1080 		u32	type;
1081 		char	name[24];	/* "IBM,RPA-Client-Config" */
1082 		struct rpadesc {
1083 			u32	lpar_affinity;
1084 			u32	min_rmo_size;
1085 			u32	min_rmo_percent;
1086 			u32	max_pft_size;
1087 			u32	splpar;
1088 			u32	min_load;
1089 			u32	new_mem_def;
1090 			u32	ignore_me;
1091 		} rpadesc;
1092 	} rpanote;
1093 } fake_elf __initconst = {
1094 	.elfhdr = {
1095 		.e_ident = { 0x7f, 'E', 'L', 'F',
1096 			     ELFCLASS32, ELFDATA2MSB, EV_CURRENT },
1097 		.e_type = ET_EXEC,	/* yeah right */
1098 		.e_machine = EM_PPC,
1099 		.e_version = EV_CURRENT,
1100 		.e_phoff = offsetof(struct fake_elf, phdr),
1101 		.e_phentsize = sizeof(Elf32_Phdr),
1102 		.e_phnum = 2
1103 	},
1104 	.phdr = {
1105 		[0] = {
1106 			.p_type = PT_NOTE,
1107 			.p_offset = offsetof(struct fake_elf, chrpnote),
1108 			.p_filesz = sizeof(struct chrpnote)
1109 		}, [1] = {
1110 			.p_type = PT_NOTE,
1111 			.p_offset = offsetof(struct fake_elf, rpanote),
1112 			.p_filesz = sizeof(struct rpanote)
1113 		}
1114 	},
1115 	.chrpnote = {
1116 		.namesz = sizeof("PowerPC"),
1117 		.descsz = sizeof(struct chrpdesc),
1118 		.type = 0x1275,
1119 		.name = "PowerPC",
1120 		.chrpdesc = {
1121 			.real_mode = ~0U,	/* ~0 means "don't care" */
1122 			.real_base = ~0U,
1123 			.real_size = ~0U,
1124 			.virt_base = ~0U,
1125 			.virt_size = ~0U,
1126 			.load_base = ~0U
1127 		},
1128 	},
1129 	.rpanote = {
1130 		.namesz = sizeof("IBM,RPA-Client-Config"),
1131 		.descsz = sizeof(struct rpadesc),
1132 		.type = 0x12759999,
1133 		.name = "IBM,RPA-Client-Config",
1134 		.rpadesc = {
1135 			.lpar_affinity = 0,
1136 			.min_rmo_size = 64,	/* in megabytes */
1137 			.min_rmo_percent = 0,
1138 			.max_pft_size = 48,	/* 2^48 bytes max PFT size */
1139 			.splpar = 1,
1140 			.min_load = ~0U,
1141 			.new_mem_def = 0
1142 		}
1143 	}
1144 };
1145 #endif /* __BIG_ENDIAN__ */
1146 
1147 static int __init prom_count_smt_threads(void)
1148 {
1149 	phandle node;
1150 	char type[64];
1151 	unsigned int plen;
1152 
1153 	/* Pick up th first CPU node we can find */
1154 	for (node = 0; prom_next_node(&node); ) {
1155 		type[0] = 0;
1156 		prom_getprop(node, "device_type", type, sizeof(type));
1157 
1158 		if (prom_strcmp(type, "cpu"))
1159 			continue;
1160 		/*
1161 		 * There is an entry for each smt thread, each entry being
1162 		 * 4 bytes long.  All cpus should have the same number of
1163 		 * smt threads, so return after finding the first.
1164 		 */
1165 		plen = prom_getproplen(node, "ibm,ppc-interrupt-server#s");
1166 		if (plen == PROM_ERROR)
1167 			break;
1168 		plen >>= 2;
1169 		prom_debug("Found %lu smt threads per core\n", (unsigned long)plen);
1170 
1171 		/* Sanity check */
1172 		if (plen < 1 || plen > 64) {
1173 			prom_printf("Threads per core %lu out of bounds, assuming 1\n",
1174 				    (unsigned long)plen);
1175 			return 1;
1176 		}
1177 		return plen;
1178 	}
1179 	prom_debug("No threads found, assuming 1 per core\n");
1180 
1181 	return 1;
1182 
1183 }
1184 
1185 static void __init prom_parse_mmu_model(u8 val,
1186 					struct platform_support *support)
1187 {
1188 	switch (val) {
1189 	case OV5_FEAT(OV5_MMU_DYNAMIC):
1190 	case OV5_FEAT(OV5_MMU_EITHER): /* Either Available */
1191 		prom_debug("MMU - either supported\n");
1192 		support->radix_mmu = !prom_radix_disable;
1193 		support->hash_mmu = true;
1194 		break;
1195 	case OV5_FEAT(OV5_MMU_RADIX): /* Only Radix */
1196 		prom_debug("MMU - radix only\n");
1197 		if (prom_radix_disable) {
1198 			/*
1199 			 * If we __have__ to do radix, we're better off ignoring
1200 			 * the command line rather than not booting.
1201 			 */
1202 			prom_printf("WARNING: Ignoring cmdline option disable_radix\n");
1203 		}
1204 		support->radix_mmu = true;
1205 		break;
1206 	case OV5_FEAT(OV5_MMU_HASH):
1207 		prom_debug("MMU - hash only\n");
1208 		support->hash_mmu = true;
1209 		break;
1210 	default:
1211 		prom_debug("Unknown mmu support option: 0x%x\n", val);
1212 		break;
1213 	}
1214 }
1215 
1216 static void __init prom_parse_xive_model(u8 val,
1217 					 struct platform_support *support)
1218 {
1219 	switch (val) {
1220 	case OV5_FEAT(OV5_XIVE_EITHER): /* Either Available */
1221 		prom_debug("XIVE - either mode supported\n");
1222 		support->xive = !prom_xive_disable;
1223 		break;
1224 	case OV5_FEAT(OV5_XIVE_EXPLOIT): /* Only Exploitation mode */
1225 		prom_debug("XIVE - exploitation mode supported\n");
1226 		if (prom_xive_disable) {
1227 			/*
1228 			 * If we __have__ to do XIVE, we're better off ignoring
1229 			 * the command line rather than not booting.
1230 			 */
1231 			prom_printf("WARNING: Ignoring cmdline option xive=off\n");
1232 		}
1233 		support->xive = true;
1234 		break;
1235 	case OV5_FEAT(OV5_XIVE_LEGACY): /* Only Legacy mode */
1236 		prom_debug("XIVE - legacy mode supported\n");
1237 		break;
1238 	default:
1239 		prom_debug("Unknown xive support option: 0x%x\n", val);
1240 		break;
1241 	}
1242 }
1243 
1244 static void __init prom_parse_platform_support(u8 index, u8 val,
1245 					       struct platform_support *support)
1246 {
1247 	switch (index) {
1248 	case OV5_INDX(OV5_MMU_SUPPORT): /* MMU Model */
1249 		prom_parse_mmu_model(val & OV5_FEAT(OV5_MMU_SUPPORT), support);
1250 		break;
1251 	case OV5_INDX(OV5_RADIX_GTSE): /* Radix Extensions */
1252 		if (val & OV5_FEAT(OV5_RADIX_GTSE)) {
1253 			prom_debug("Radix - GTSE supported\n");
1254 			support->radix_gtse = true;
1255 		}
1256 		break;
1257 	case OV5_INDX(OV5_XIVE_SUPPORT): /* Interrupt mode */
1258 		prom_parse_xive_model(val & OV5_FEAT(OV5_XIVE_SUPPORT),
1259 				      support);
1260 		break;
1261 	}
1262 }
1263 
1264 static void __init prom_check_platform_support(void)
1265 {
1266 	struct platform_support supported = {
1267 		.hash_mmu = false,
1268 		.radix_mmu = false,
1269 		.radix_gtse = false,
1270 		.xive = false
1271 	};
1272 	int prop_len = prom_getproplen(prom.chosen,
1273 				       "ibm,arch-vec-5-platform-support");
1274 
1275 	/*
1276 	 * First copy the architecture vec template
1277 	 *
1278 	 * use memcpy() instead of *vec = *vec_template so that GCC replaces it
1279 	 * by __memcpy() when KASAN is active
1280 	 */
1281 	memcpy(&ibm_architecture_vec, &ibm_architecture_vec_template,
1282 	       sizeof(ibm_architecture_vec));
1283 
1284 	if (prop_len > 1) {
1285 		int i;
1286 		u8 vec[8];
1287 		prom_debug("Found ibm,arch-vec-5-platform-support, len: %d\n",
1288 			   prop_len);
1289 		if (prop_len > sizeof(vec))
1290 			prom_printf("WARNING: ibm,arch-vec-5-platform-support longer than expected (len: %d)\n",
1291 				    prop_len);
1292 		prom_getprop(prom.chosen, "ibm,arch-vec-5-platform-support",
1293 			     &vec, sizeof(vec));
1294 		for (i = 0; i < sizeof(vec); i += 2) {
1295 			prom_debug("%d: index = 0x%x val = 0x%x\n", i / 2
1296 								  , vec[i]
1297 								  , vec[i + 1]);
1298 			prom_parse_platform_support(vec[i], vec[i + 1],
1299 						    &supported);
1300 		}
1301 	}
1302 
1303 	if (supported.radix_mmu && supported.radix_gtse &&
1304 	    IS_ENABLED(CONFIG_PPC_RADIX_MMU)) {
1305 		/* Radix preferred - but we require GTSE for now */
1306 		prom_debug("Asking for radix with GTSE\n");
1307 		ibm_architecture_vec.vec5.mmu = OV5_FEAT(OV5_MMU_RADIX);
1308 		ibm_architecture_vec.vec5.radix_ext = OV5_FEAT(OV5_RADIX_GTSE);
1309 	} else if (supported.hash_mmu) {
1310 		/* Default to hash mmu (if we can) */
1311 		prom_debug("Asking for hash\n");
1312 		ibm_architecture_vec.vec5.mmu = OV5_FEAT(OV5_MMU_HASH);
1313 	} else {
1314 		/* We're probably on a legacy hypervisor */
1315 		prom_debug("Assuming legacy hash support\n");
1316 	}
1317 
1318 	if (supported.xive) {
1319 		prom_debug("Asking for XIVE\n");
1320 		ibm_architecture_vec.vec5.intarch = OV5_FEAT(OV5_XIVE_EXPLOIT);
1321 	}
1322 }
1323 
1324 static void __init prom_send_capabilities(void)
1325 {
1326 	ihandle root;
1327 	prom_arg_t ret;
1328 	u32 cores;
1329 
1330 	/* Check ibm,arch-vec-5-platform-support and fixup vec5 if required */
1331 	prom_check_platform_support();
1332 
1333 	root = call_prom("open", 1, 1, ADDR("/"));
1334 	if (root != 0) {
1335 		/* We need to tell the FW about the number of cores we support.
1336 		 *
1337 		 * To do that, we count the number of threads on the first core
1338 		 * (we assume this is the same for all cores) and use it to
1339 		 * divide NR_CPUS.
1340 		 */
1341 
1342 		cores = DIV_ROUND_UP(NR_CPUS, prom_count_smt_threads());
1343 		prom_printf("Max number of cores passed to firmware: %u (NR_CPUS = %d)\n",
1344 			    cores, NR_CPUS);
1345 
1346 		ibm_architecture_vec.vec5.max_cpus = cpu_to_be32(cores);
1347 
1348 		/* try calling the ibm,client-architecture-support method */
1349 		prom_printf("Calling ibm,client-architecture-support...");
1350 		if (call_prom_ret("call-method", 3, 2, &ret,
1351 				  ADDR("ibm,client-architecture-support"),
1352 				  root,
1353 				  ADDR(&ibm_architecture_vec)) == 0) {
1354 			/* the call exists... */
1355 			if (ret)
1356 				prom_printf("\nWARNING: ibm,client-architecture"
1357 					    "-support call FAILED!\n");
1358 			call_prom("close", 1, 0, root);
1359 			prom_printf(" done\n");
1360 			return;
1361 		}
1362 		call_prom("close", 1, 0, root);
1363 		prom_printf(" not implemented\n");
1364 	}
1365 
1366 #ifdef __BIG_ENDIAN__
1367 	{
1368 		ihandle elfloader;
1369 
1370 		/* no ibm,client-architecture-support call, try the old way */
1371 		elfloader = call_prom("open", 1, 1,
1372 				      ADDR("/packages/elf-loader"));
1373 		if (elfloader == 0) {
1374 			prom_printf("couldn't open /packages/elf-loader\n");
1375 			return;
1376 		}
1377 		call_prom("call-method", 3, 1, ADDR("process-elf-header"),
1378 			  elfloader, ADDR(&fake_elf));
1379 		call_prom("close", 1, 0, elfloader);
1380 	}
1381 #endif /* __BIG_ENDIAN__ */
1382 }
1383 #endif /* CONFIG_PPC_PSERIES */
1384 
1385 /*
1386  * Memory allocation strategy... our layout is normally:
1387  *
1388  *  at 14Mb or more we have vmlinux, then a gap and initrd.  In some
1389  *  rare cases, initrd might end up being before the kernel though.
1390  *  We assume this won't override the final kernel at 0, we have no
1391  *  provision to handle that in this version, but it should hopefully
1392  *  never happen.
1393  *
1394  *  alloc_top is set to the top of RMO, eventually shrink down if the
1395  *  TCEs overlap
1396  *
1397  *  alloc_bottom is set to the top of kernel/initrd
1398  *
1399  *  from there, allocations are done this way : rtas is allocated
1400  *  topmost, and the device-tree is allocated from the bottom. We try
1401  *  to grow the device-tree allocation as we progress. If we can't,
1402  *  then we fail, we don't currently have a facility to restart
1403  *  elsewhere, but that shouldn't be necessary.
1404  *
1405  *  Note that calls to reserve_mem have to be done explicitly, memory
1406  *  allocated with either alloc_up or alloc_down isn't automatically
1407  *  reserved.
1408  */
1409 
1410 
1411 /*
1412  * Allocates memory in the RMO upward from the kernel/initrd
1413  *
1414  * When align is 0, this is a special case, it means to allocate in place
1415  * at the current location of alloc_bottom or fail (that is basically
1416  * extending the previous allocation). Used for the device-tree flattening
1417  */
1418 static unsigned long __init alloc_up(unsigned long size, unsigned long align)
1419 {
1420 	unsigned long base = alloc_bottom;
1421 	unsigned long addr = 0;
1422 
1423 	if (align)
1424 		base = _ALIGN_UP(base, align);
1425 	prom_debug("%s(%lx, %lx)\n", __func__, size, align);
1426 	if (ram_top == 0)
1427 		prom_panic("alloc_up() called with mem not initialized\n");
1428 
1429 	if (align)
1430 		base = _ALIGN_UP(alloc_bottom, align);
1431 	else
1432 		base = alloc_bottom;
1433 
1434 	for(; (base + size) <= alloc_top;
1435 	    base = _ALIGN_UP(base + 0x100000, align)) {
1436 		prom_debug("    trying: 0x%lx\n\r", base);
1437 		addr = (unsigned long)prom_claim(base, size, 0);
1438 		if (addr != PROM_ERROR && addr != 0)
1439 			break;
1440 		addr = 0;
1441 		if (align == 0)
1442 			break;
1443 	}
1444 	if (addr == 0)
1445 		return 0;
1446 	alloc_bottom = addr + size;
1447 
1448 	prom_debug(" -> %lx\n", addr);
1449 	prom_debug("  alloc_bottom : %lx\n", alloc_bottom);
1450 	prom_debug("  alloc_top    : %lx\n", alloc_top);
1451 	prom_debug("  alloc_top_hi : %lx\n", alloc_top_high);
1452 	prom_debug("  rmo_top      : %lx\n", rmo_top);
1453 	prom_debug("  ram_top      : %lx\n", ram_top);
1454 
1455 	return addr;
1456 }
1457 
1458 /*
1459  * Allocates memory downward, either from top of RMO, or if highmem
1460  * is set, from the top of RAM.  Note that this one doesn't handle
1461  * failures.  It does claim memory if highmem is not set.
1462  */
1463 static unsigned long __init alloc_down(unsigned long size, unsigned long align,
1464 				       int highmem)
1465 {
1466 	unsigned long base, addr = 0;
1467 
1468 	prom_debug("%s(%lx, %lx, %s)\n", __func__, size, align,
1469 		   highmem ? "(high)" : "(low)");
1470 	if (ram_top == 0)
1471 		prom_panic("alloc_down() called with mem not initialized\n");
1472 
1473 	if (highmem) {
1474 		/* Carve out storage for the TCE table. */
1475 		addr = _ALIGN_DOWN(alloc_top_high - size, align);
1476 		if (addr <= alloc_bottom)
1477 			return 0;
1478 		/* Will we bump into the RMO ? If yes, check out that we
1479 		 * didn't overlap existing allocations there, if we did,
1480 		 * we are dead, we must be the first in town !
1481 		 */
1482 		if (addr < rmo_top) {
1483 			/* Good, we are first */
1484 			if (alloc_top == rmo_top)
1485 				alloc_top = rmo_top = addr;
1486 			else
1487 				return 0;
1488 		}
1489 		alloc_top_high = addr;
1490 		goto bail;
1491 	}
1492 
1493 	base = _ALIGN_DOWN(alloc_top - size, align);
1494 	for (; base > alloc_bottom;
1495 	     base = _ALIGN_DOWN(base - 0x100000, align))  {
1496 		prom_debug("    trying: 0x%lx\n\r", base);
1497 		addr = (unsigned long)prom_claim(base, size, 0);
1498 		if (addr != PROM_ERROR && addr != 0)
1499 			break;
1500 		addr = 0;
1501 	}
1502 	if (addr == 0)
1503 		return 0;
1504 	alloc_top = addr;
1505 
1506  bail:
1507 	prom_debug(" -> %lx\n", addr);
1508 	prom_debug("  alloc_bottom : %lx\n", alloc_bottom);
1509 	prom_debug("  alloc_top    : %lx\n", alloc_top);
1510 	prom_debug("  alloc_top_hi : %lx\n", alloc_top_high);
1511 	prom_debug("  rmo_top      : %lx\n", rmo_top);
1512 	prom_debug("  ram_top      : %lx\n", ram_top);
1513 
1514 	return addr;
1515 }
1516 
1517 /*
1518  * Parse a "reg" cell
1519  */
1520 static unsigned long __init prom_next_cell(int s, cell_t **cellp)
1521 {
1522 	cell_t *p = *cellp;
1523 	unsigned long r = 0;
1524 
1525 	/* Ignore more than 2 cells */
1526 	while (s > sizeof(unsigned long) / 4) {
1527 		p++;
1528 		s--;
1529 	}
1530 	r = be32_to_cpu(*p++);
1531 #ifdef CONFIG_PPC64
1532 	if (s > 1) {
1533 		r <<= 32;
1534 		r |= be32_to_cpu(*(p++));
1535 	}
1536 #endif
1537 	*cellp = p;
1538 	return r;
1539 }
1540 
1541 /*
1542  * Very dumb function for adding to the memory reserve list, but
1543  * we don't need anything smarter at this point
1544  *
1545  * XXX Eventually check for collisions.  They should NEVER happen.
1546  * If problems seem to show up, it would be a good start to track
1547  * them down.
1548  */
1549 static void __init reserve_mem(u64 base, u64 size)
1550 {
1551 	u64 top = base + size;
1552 	unsigned long cnt = mem_reserve_cnt;
1553 
1554 	if (size == 0)
1555 		return;
1556 
1557 	/* We need to always keep one empty entry so that we
1558 	 * have our terminator with "size" set to 0 since we are
1559 	 * dumb and just copy this entire array to the boot params
1560 	 */
1561 	base = _ALIGN_DOWN(base, PAGE_SIZE);
1562 	top = _ALIGN_UP(top, PAGE_SIZE);
1563 	size = top - base;
1564 
1565 	if (cnt >= (MEM_RESERVE_MAP_SIZE - 1))
1566 		prom_panic("Memory reserve map exhausted !\n");
1567 	mem_reserve_map[cnt].base = cpu_to_be64(base);
1568 	mem_reserve_map[cnt].size = cpu_to_be64(size);
1569 	mem_reserve_cnt = cnt + 1;
1570 }
1571 
1572 /*
1573  * Initialize memory allocation mechanism, parse "memory" nodes and
1574  * obtain that way the top of memory and RMO to setup out local allocator
1575  */
1576 static void __init prom_init_mem(void)
1577 {
1578 	phandle node;
1579 	char type[64];
1580 	unsigned int plen;
1581 	cell_t *p, *endp;
1582 	__be32 val;
1583 	u32 rac, rsc;
1584 
1585 	/*
1586 	 * We iterate the memory nodes to find
1587 	 * 1) top of RMO (first node)
1588 	 * 2) top of memory
1589 	 */
1590 	val = cpu_to_be32(2);
1591 	prom_getprop(prom.root, "#address-cells", &val, sizeof(val));
1592 	rac = be32_to_cpu(val);
1593 	val = cpu_to_be32(1);
1594 	prom_getprop(prom.root, "#size-cells", &val, sizeof(rsc));
1595 	rsc = be32_to_cpu(val);
1596 	prom_debug("root_addr_cells: %x\n", rac);
1597 	prom_debug("root_size_cells: %x\n", rsc);
1598 
1599 	prom_debug("scanning memory:\n");
1600 
1601 	for (node = 0; prom_next_node(&node); ) {
1602 		type[0] = 0;
1603 		prom_getprop(node, "device_type", type, sizeof(type));
1604 
1605 		if (type[0] == 0) {
1606 			/*
1607 			 * CHRP Longtrail machines have no device_type
1608 			 * on the memory node, so check the name instead...
1609 			 */
1610 			prom_getprop(node, "name", type, sizeof(type));
1611 		}
1612 		if (prom_strcmp(type, "memory"))
1613 			continue;
1614 
1615 		plen = prom_getprop(node, "reg", regbuf, sizeof(regbuf));
1616 		if (plen > sizeof(regbuf)) {
1617 			prom_printf("memory node too large for buffer !\n");
1618 			plen = sizeof(regbuf);
1619 		}
1620 		p = regbuf;
1621 		endp = p + (plen / sizeof(cell_t));
1622 
1623 #ifdef DEBUG_PROM
1624 		memset(prom_scratch, 0, sizeof(prom_scratch));
1625 		call_prom("package-to-path", 3, 1, node, prom_scratch,
1626 			  sizeof(prom_scratch) - 1);
1627 		prom_debug("  node %s :\n", prom_scratch);
1628 #endif /* DEBUG_PROM */
1629 
1630 		while ((endp - p) >= (rac + rsc)) {
1631 			unsigned long base, size;
1632 
1633 			base = prom_next_cell(rac, &p);
1634 			size = prom_next_cell(rsc, &p);
1635 
1636 			if (size == 0)
1637 				continue;
1638 			prom_debug("    %lx %lx\n", base, size);
1639 			if (base == 0 && (of_platform & PLATFORM_LPAR))
1640 				rmo_top = size;
1641 			if ((base + size) > ram_top)
1642 				ram_top = base + size;
1643 		}
1644 	}
1645 
1646 	alloc_bottom = PAGE_ALIGN((unsigned long)&_end + 0x4000);
1647 
1648 	/*
1649 	 * If prom_memory_limit is set we reduce the upper limits *except* for
1650 	 * alloc_top_high. This must be the real top of RAM so we can put
1651 	 * TCE's up there.
1652 	 */
1653 
1654 	alloc_top_high = ram_top;
1655 
1656 	if (prom_memory_limit) {
1657 		if (prom_memory_limit <= alloc_bottom) {
1658 			prom_printf("Ignoring mem=%lx <= alloc_bottom.\n",
1659 				    prom_memory_limit);
1660 			prom_memory_limit = 0;
1661 		} else if (prom_memory_limit >= ram_top) {
1662 			prom_printf("Ignoring mem=%lx >= ram_top.\n",
1663 				    prom_memory_limit);
1664 			prom_memory_limit = 0;
1665 		} else {
1666 			ram_top = prom_memory_limit;
1667 			rmo_top = min(rmo_top, prom_memory_limit);
1668 		}
1669 	}
1670 
1671 	/*
1672 	 * Setup our top alloc point, that is top of RMO or top of
1673 	 * segment 0 when running non-LPAR.
1674 	 * Some RS64 machines have buggy firmware where claims up at
1675 	 * 1GB fail.  Cap at 768MB as a workaround.
1676 	 * Since 768MB is plenty of room, and we need to cap to something
1677 	 * reasonable on 32-bit, cap at 768MB on all machines.
1678 	 */
1679 	if (!rmo_top)
1680 		rmo_top = ram_top;
1681 	rmo_top = min(0x30000000ul, rmo_top);
1682 	alloc_top = rmo_top;
1683 	alloc_top_high = ram_top;
1684 
1685 	/*
1686 	 * Check if we have an initrd after the kernel but still inside
1687 	 * the RMO.  If we do move our bottom point to after it.
1688 	 */
1689 	if (prom_initrd_start &&
1690 	    prom_initrd_start < rmo_top &&
1691 	    prom_initrd_end > alloc_bottom)
1692 		alloc_bottom = PAGE_ALIGN(prom_initrd_end);
1693 
1694 	prom_printf("memory layout at init:\n");
1695 	prom_printf("  memory_limit : %lx (16 MB aligned)\n",
1696 		    prom_memory_limit);
1697 	prom_printf("  alloc_bottom : %lx\n", alloc_bottom);
1698 	prom_printf("  alloc_top    : %lx\n", alloc_top);
1699 	prom_printf("  alloc_top_hi : %lx\n", alloc_top_high);
1700 	prom_printf("  rmo_top      : %lx\n", rmo_top);
1701 	prom_printf("  ram_top      : %lx\n", ram_top);
1702 }
1703 
1704 static void __init prom_close_stdin(void)
1705 {
1706 	__be32 val;
1707 	ihandle stdin;
1708 
1709 	if (prom_getprop(prom.chosen, "stdin", &val, sizeof(val)) > 0) {
1710 		stdin = be32_to_cpu(val);
1711 		call_prom("close", 1, 0, stdin);
1712 	}
1713 }
1714 
1715 /*
1716  * Allocate room for and instantiate RTAS
1717  */
1718 static void __init prom_instantiate_rtas(void)
1719 {
1720 	phandle rtas_node;
1721 	ihandle rtas_inst;
1722 	u32 base, entry = 0;
1723 	__be32 val;
1724 	u32 size = 0;
1725 
1726 	prom_debug("prom_instantiate_rtas: start...\n");
1727 
1728 	rtas_node = call_prom("finddevice", 1, 1, ADDR("/rtas"));
1729 	prom_debug("rtas_node: %x\n", rtas_node);
1730 	if (!PHANDLE_VALID(rtas_node))
1731 		return;
1732 
1733 	val = 0;
1734 	prom_getprop(rtas_node, "rtas-size", &val, sizeof(size));
1735 	size = be32_to_cpu(val);
1736 	if (size == 0)
1737 		return;
1738 
1739 	base = alloc_down(size, PAGE_SIZE, 0);
1740 	if (base == 0)
1741 		prom_panic("Could not allocate memory for RTAS\n");
1742 
1743 	rtas_inst = call_prom("open", 1, 1, ADDR("/rtas"));
1744 	if (!IHANDLE_VALID(rtas_inst)) {
1745 		prom_printf("opening rtas package failed (%x)\n", rtas_inst);
1746 		return;
1747 	}
1748 
1749 	prom_printf("instantiating rtas at 0x%x...", base);
1750 
1751 	if (call_prom_ret("call-method", 3, 2, &entry,
1752 			  ADDR("instantiate-rtas"),
1753 			  rtas_inst, base) != 0
1754 	    || entry == 0) {
1755 		prom_printf(" failed\n");
1756 		return;
1757 	}
1758 	prom_printf(" done\n");
1759 
1760 	reserve_mem(base, size);
1761 
1762 	val = cpu_to_be32(base);
1763 	prom_setprop(rtas_node, "/rtas", "linux,rtas-base",
1764 		     &val, sizeof(val));
1765 	val = cpu_to_be32(entry);
1766 	prom_setprop(rtas_node, "/rtas", "linux,rtas-entry",
1767 		     &val, sizeof(val));
1768 
1769 	/* Check if it supports "query-cpu-stopped-state" */
1770 	if (prom_getprop(rtas_node, "query-cpu-stopped-state",
1771 			 &val, sizeof(val)) != PROM_ERROR)
1772 		rtas_has_query_cpu_stopped = true;
1773 
1774 	prom_debug("rtas base     = 0x%x\n", base);
1775 	prom_debug("rtas entry    = 0x%x\n", entry);
1776 	prom_debug("rtas size     = 0x%x\n", size);
1777 
1778 	prom_debug("prom_instantiate_rtas: end...\n");
1779 }
1780 
1781 #ifdef CONFIG_PPC64
1782 /*
1783  * Allocate room for and instantiate Stored Measurement Log (SML)
1784  */
1785 static void __init prom_instantiate_sml(void)
1786 {
1787 	phandle ibmvtpm_node;
1788 	ihandle ibmvtpm_inst;
1789 	u32 entry = 0, size = 0, succ = 0;
1790 	u64 base;
1791 	__be32 val;
1792 
1793 	prom_debug("prom_instantiate_sml: start...\n");
1794 
1795 	ibmvtpm_node = call_prom("finddevice", 1, 1, ADDR("/vdevice/vtpm"));
1796 	prom_debug("ibmvtpm_node: %x\n", ibmvtpm_node);
1797 	if (!PHANDLE_VALID(ibmvtpm_node))
1798 		return;
1799 
1800 	ibmvtpm_inst = call_prom("open", 1, 1, ADDR("/vdevice/vtpm"));
1801 	if (!IHANDLE_VALID(ibmvtpm_inst)) {
1802 		prom_printf("opening vtpm package failed (%x)\n", ibmvtpm_inst);
1803 		return;
1804 	}
1805 
1806 	if (prom_getprop(ibmvtpm_node, "ibm,sml-efi-reformat-supported",
1807 			 &val, sizeof(val)) != PROM_ERROR) {
1808 		if (call_prom_ret("call-method", 2, 2, &succ,
1809 				  ADDR("reformat-sml-to-efi-alignment"),
1810 				  ibmvtpm_inst) != 0 || succ == 0) {
1811 			prom_printf("Reformat SML to EFI alignment failed\n");
1812 			return;
1813 		}
1814 
1815 		if (call_prom_ret("call-method", 2, 2, &size,
1816 				  ADDR("sml-get-allocated-size"),
1817 				  ibmvtpm_inst) != 0 || size == 0) {
1818 			prom_printf("SML get allocated size failed\n");
1819 			return;
1820 		}
1821 	} else {
1822 		if (call_prom_ret("call-method", 2, 2, &size,
1823 				  ADDR("sml-get-handover-size"),
1824 				  ibmvtpm_inst) != 0 || size == 0) {
1825 			prom_printf("SML get handover size failed\n");
1826 			return;
1827 		}
1828 	}
1829 
1830 	base = alloc_down(size, PAGE_SIZE, 0);
1831 	if (base == 0)
1832 		prom_panic("Could not allocate memory for sml\n");
1833 
1834 	prom_printf("instantiating sml at 0x%llx...", base);
1835 
1836 	memset((void *)base, 0, size);
1837 
1838 	if (call_prom_ret("call-method", 4, 2, &entry,
1839 			  ADDR("sml-handover"),
1840 			  ibmvtpm_inst, size, base) != 0 || entry == 0) {
1841 		prom_printf("SML handover failed\n");
1842 		return;
1843 	}
1844 	prom_printf(" done\n");
1845 
1846 	reserve_mem(base, size);
1847 
1848 	prom_setprop(ibmvtpm_node, "/vdevice/vtpm", "linux,sml-base",
1849 		     &base, sizeof(base));
1850 	prom_setprop(ibmvtpm_node, "/vdevice/vtpm", "linux,sml-size",
1851 		     &size, sizeof(size));
1852 
1853 	prom_debug("sml base     = 0x%llx\n", base);
1854 	prom_debug("sml size     = 0x%x\n", size);
1855 
1856 	prom_debug("prom_instantiate_sml: end...\n");
1857 }
1858 
1859 /*
1860  * Allocate room for and initialize TCE tables
1861  */
1862 #ifdef __BIG_ENDIAN__
1863 static void __init prom_initialize_tce_table(void)
1864 {
1865 	phandle node;
1866 	ihandle phb_node;
1867 	char compatible[64], type[64], model[64];
1868 	char *path = prom_scratch;
1869 	u64 base, align;
1870 	u32 minalign, minsize;
1871 	u64 tce_entry, *tce_entryp;
1872 	u64 local_alloc_top, local_alloc_bottom;
1873 	u64 i;
1874 
1875 	if (prom_iommu_off)
1876 		return;
1877 
1878 	prom_debug("starting prom_initialize_tce_table\n");
1879 
1880 	/* Cache current top of allocs so we reserve a single block */
1881 	local_alloc_top = alloc_top_high;
1882 	local_alloc_bottom = local_alloc_top;
1883 
1884 	/* Search all nodes looking for PHBs. */
1885 	for (node = 0; prom_next_node(&node); ) {
1886 		compatible[0] = 0;
1887 		type[0] = 0;
1888 		model[0] = 0;
1889 		prom_getprop(node, "compatible",
1890 			     compatible, sizeof(compatible));
1891 		prom_getprop(node, "device_type", type, sizeof(type));
1892 		prom_getprop(node, "model", model, sizeof(model));
1893 
1894 		if ((type[0] == 0) || (prom_strstr(type, "pci") == NULL))
1895 			continue;
1896 
1897 		/* Keep the old logic intact to avoid regression. */
1898 		if (compatible[0] != 0) {
1899 			if ((prom_strstr(compatible, "python") == NULL) &&
1900 			    (prom_strstr(compatible, "Speedwagon") == NULL) &&
1901 			    (prom_strstr(compatible, "Winnipeg") == NULL))
1902 				continue;
1903 		} else if (model[0] != 0) {
1904 			if ((prom_strstr(model, "ython") == NULL) &&
1905 			    (prom_strstr(model, "peedwagon") == NULL) &&
1906 			    (prom_strstr(model, "innipeg") == NULL))
1907 				continue;
1908 		}
1909 
1910 		if (prom_getprop(node, "tce-table-minalign", &minalign,
1911 				 sizeof(minalign)) == PROM_ERROR)
1912 			minalign = 0;
1913 		if (prom_getprop(node, "tce-table-minsize", &minsize,
1914 				 sizeof(minsize)) == PROM_ERROR)
1915 			minsize = 4UL << 20;
1916 
1917 		/*
1918 		 * Even though we read what OF wants, we just set the table
1919 		 * size to 4 MB.  This is enough to map 2GB of PCI DMA space.
1920 		 * By doing this, we avoid the pitfalls of trying to DMA to
1921 		 * MMIO space and the DMA alias hole.
1922 		 */
1923 		minsize = 4UL << 20;
1924 
1925 		/* Align to the greater of the align or size */
1926 		align = max(minalign, minsize);
1927 		base = alloc_down(minsize, align, 1);
1928 		if (base == 0)
1929 			prom_panic("ERROR, cannot find space for TCE table.\n");
1930 		if (base < local_alloc_bottom)
1931 			local_alloc_bottom = base;
1932 
1933 		/* It seems OF doesn't null-terminate the path :-( */
1934 		memset(path, 0, sizeof(prom_scratch));
1935 		/* Call OF to setup the TCE hardware */
1936 		if (call_prom("package-to-path", 3, 1, node,
1937 			      path, sizeof(prom_scratch) - 1) == PROM_ERROR) {
1938 			prom_printf("package-to-path failed\n");
1939 		}
1940 
1941 		/* Save away the TCE table attributes for later use. */
1942 		prom_setprop(node, path, "linux,tce-base", &base, sizeof(base));
1943 		prom_setprop(node, path, "linux,tce-size", &minsize, sizeof(minsize));
1944 
1945 		prom_debug("TCE table: %s\n", path);
1946 		prom_debug("\tnode = 0x%x\n", node);
1947 		prom_debug("\tbase = 0x%llx\n", base);
1948 		prom_debug("\tsize = 0x%x\n", minsize);
1949 
1950 		/* Initialize the table to have a one-to-one mapping
1951 		 * over the allocated size.
1952 		 */
1953 		tce_entryp = (u64 *)base;
1954 		for (i = 0; i < (minsize >> 3) ;tce_entryp++, i++) {
1955 			tce_entry = (i << PAGE_SHIFT);
1956 			tce_entry |= 0x3;
1957 			*tce_entryp = tce_entry;
1958 		}
1959 
1960 		prom_printf("opening PHB %s", path);
1961 		phb_node = call_prom("open", 1, 1, path);
1962 		if (phb_node == 0)
1963 			prom_printf("... failed\n");
1964 		else
1965 			prom_printf("... done\n");
1966 
1967 		call_prom("call-method", 6, 0, ADDR("set-64-bit-addressing"),
1968 			  phb_node, -1, minsize,
1969 			  (u32) base, (u32) (base >> 32));
1970 		call_prom("close", 1, 0, phb_node);
1971 	}
1972 
1973 	reserve_mem(local_alloc_bottom, local_alloc_top - local_alloc_bottom);
1974 
1975 	/* These are only really needed if there is a memory limit in
1976 	 * effect, but we don't know so export them always. */
1977 	prom_tce_alloc_start = local_alloc_bottom;
1978 	prom_tce_alloc_end = local_alloc_top;
1979 
1980 	/* Flag the first invalid entry */
1981 	prom_debug("ending prom_initialize_tce_table\n");
1982 }
1983 #endif /* __BIG_ENDIAN__ */
1984 #endif /* CONFIG_PPC64 */
1985 
1986 /*
1987  * With CHRP SMP we need to use the OF to start the other processors.
1988  * We can't wait until smp_boot_cpus (the OF is trashed by then)
1989  * so we have to put the processors into a holding pattern controlled
1990  * by the kernel (not OF) before we destroy the OF.
1991  *
1992  * This uses a chunk of low memory, puts some holding pattern
1993  * code there and sends the other processors off to there until
1994  * smp_boot_cpus tells them to do something.  The holding pattern
1995  * checks that address until its cpu # is there, when it is that
1996  * cpu jumps to __secondary_start().  smp_boot_cpus() takes care
1997  * of setting those values.
1998  *
1999  * We also use physical address 0x4 here to tell when a cpu
2000  * is in its holding pattern code.
2001  *
2002  * -- Cort
2003  */
2004 /*
2005  * We want to reference the copy of __secondary_hold_* in the
2006  * 0 - 0x100 address range
2007  */
2008 #define LOW_ADDR(x)	(((unsigned long) &(x)) & 0xff)
2009 
2010 static void __init prom_hold_cpus(void)
2011 {
2012 	unsigned long i;
2013 	phandle node;
2014 	char type[64];
2015 	unsigned long *spinloop
2016 		= (void *) LOW_ADDR(__secondary_hold_spinloop);
2017 	unsigned long *acknowledge
2018 		= (void *) LOW_ADDR(__secondary_hold_acknowledge);
2019 	unsigned long secondary_hold = LOW_ADDR(__secondary_hold);
2020 
2021 	/*
2022 	 * On pseries, if RTAS supports "query-cpu-stopped-state",
2023 	 * we skip this stage, the CPUs will be started by the
2024 	 * kernel using RTAS.
2025 	 */
2026 	if ((of_platform == PLATFORM_PSERIES ||
2027 	     of_platform == PLATFORM_PSERIES_LPAR) &&
2028 	    rtas_has_query_cpu_stopped) {
2029 		prom_printf("prom_hold_cpus: skipped\n");
2030 		return;
2031 	}
2032 
2033 	prom_debug("prom_hold_cpus: start...\n");
2034 	prom_debug("    1) spinloop       = 0x%lx\n", (unsigned long)spinloop);
2035 	prom_debug("    1) *spinloop      = 0x%lx\n", *spinloop);
2036 	prom_debug("    1) acknowledge    = 0x%lx\n",
2037 		   (unsigned long)acknowledge);
2038 	prom_debug("    1) *acknowledge   = 0x%lx\n", *acknowledge);
2039 	prom_debug("    1) secondary_hold = 0x%lx\n", secondary_hold);
2040 
2041 	/* Set the common spinloop variable, so all of the secondary cpus
2042 	 * will block when they are awakened from their OF spinloop.
2043 	 * This must occur for both SMP and non SMP kernels, since OF will
2044 	 * be trashed when we move the kernel.
2045 	 */
2046 	*spinloop = 0;
2047 
2048 	/* look for cpus */
2049 	for (node = 0; prom_next_node(&node); ) {
2050 		unsigned int cpu_no;
2051 		__be32 reg;
2052 
2053 		type[0] = 0;
2054 		prom_getprop(node, "device_type", type, sizeof(type));
2055 		if (prom_strcmp(type, "cpu") != 0)
2056 			continue;
2057 
2058 		/* Skip non-configured cpus. */
2059 		if (prom_getprop(node, "status", type, sizeof(type)) > 0)
2060 			if (prom_strcmp(type, "okay") != 0)
2061 				continue;
2062 
2063 		reg = cpu_to_be32(-1); /* make sparse happy */
2064 		prom_getprop(node, "reg", &reg, sizeof(reg));
2065 		cpu_no = be32_to_cpu(reg);
2066 
2067 		prom_debug("cpu hw idx   = %u\n", cpu_no);
2068 
2069 		/* Init the acknowledge var which will be reset by
2070 		 * the secondary cpu when it awakens from its OF
2071 		 * spinloop.
2072 		 */
2073 		*acknowledge = (unsigned long)-1;
2074 
2075 		if (cpu_no != prom.cpu) {
2076 			/* Primary Thread of non-boot cpu or any thread */
2077 			prom_printf("starting cpu hw idx %u... ", cpu_no);
2078 			call_prom("start-cpu", 3, 0, node,
2079 				  secondary_hold, cpu_no);
2080 
2081 			for (i = 0; (i < 100000000) &&
2082 			     (*acknowledge == ((unsigned long)-1)); i++ )
2083 				mb();
2084 
2085 			if (*acknowledge == cpu_no)
2086 				prom_printf("done\n");
2087 			else
2088 				prom_printf("failed: %lx\n", *acknowledge);
2089 		}
2090 #ifdef CONFIG_SMP
2091 		else
2092 			prom_printf("boot cpu hw idx %u\n", cpu_no);
2093 #endif /* CONFIG_SMP */
2094 	}
2095 
2096 	prom_debug("prom_hold_cpus: end...\n");
2097 }
2098 
2099 
2100 static void __init prom_init_client_services(unsigned long pp)
2101 {
2102 	/* Get a handle to the prom entry point before anything else */
2103 	prom_entry = pp;
2104 
2105 	/* get a handle for the stdout device */
2106 	prom.chosen = call_prom("finddevice", 1, 1, ADDR("/chosen"));
2107 	if (!PHANDLE_VALID(prom.chosen))
2108 		prom_panic("cannot find chosen"); /* msg won't be printed :( */
2109 
2110 	/* get device tree root */
2111 	prom.root = call_prom("finddevice", 1, 1, ADDR("/"));
2112 	if (!PHANDLE_VALID(prom.root))
2113 		prom_panic("cannot find device tree root"); /* msg won't be printed :( */
2114 
2115 	prom.mmumap = 0;
2116 }
2117 
2118 #ifdef CONFIG_PPC32
2119 /*
2120  * For really old powermacs, we need to map things we claim.
2121  * For that, we need the ihandle of the mmu.
2122  * Also, on the longtrail, we need to work around other bugs.
2123  */
2124 static void __init prom_find_mmu(void)
2125 {
2126 	phandle oprom;
2127 	char version[64];
2128 
2129 	oprom = call_prom("finddevice", 1, 1, ADDR("/openprom"));
2130 	if (!PHANDLE_VALID(oprom))
2131 		return;
2132 	if (prom_getprop(oprom, "model", version, sizeof(version)) <= 0)
2133 		return;
2134 	version[sizeof(version) - 1] = 0;
2135 	/* XXX might need to add other versions here */
2136 	if (prom_strcmp(version, "Open Firmware, 1.0.5") == 0)
2137 		of_workarounds = OF_WA_CLAIM;
2138 	else if (prom_strncmp(version, "FirmWorks,3.", 12) == 0) {
2139 		of_workarounds = OF_WA_CLAIM | OF_WA_LONGTRAIL;
2140 		call_prom("interpret", 1, 1, "dev /memory 0 to allow-reclaim");
2141 	} else
2142 		return;
2143 	prom.memory = call_prom("open", 1, 1, ADDR("/memory"));
2144 	prom_getprop(prom.chosen, "mmu", &prom.mmumap,
2145 		     sizeof(prom.mmumap));
2146 	prom.mmumap = be32_to_cpu(prom.mmumap);
2147 	if (!IHANDLE_VALID(prom.memory) || !IHANDLE_VALID(prom.mmumap))
2148 		of_workarounds &= ~OF_WA_CLAIM;		/* hmmm */
2149 }
2150 #else
2151 #define prom_find_mmu()
2152 #endif
2153 
2154 static void __init prom_init_stdout(void)
2155 {
2156 	char *path = of_stdout_device;
2157 	char type[16];
2158 	phandle stdout_node;
2159 	__be32 val;
2160 
2161 	if (prom_getprop(prom.chosen, "stdout", &val, sizeof(val)) <= 0)
2162 		prom_panic("cannot find stdout");
2163 
2164 	prom.stdout = be32_to_cpu(val);
2165 
2166 	/* Get the full OF pathname of the stdout device */
2167 	memset(path, 0, 256);
2168 	call_prom("instance-to-path", 3, 1, prom.stdout, path, 255);
2169 	prom_printf("OF stdout device is: %s\n", of_stdout_device);
2170 	prom_setprop(prom.chosen, "/chosen", "linux,stdout-path",
2171 		     path, prom_strlen(path) + 1);
2172 
2173 	/* instance-to-package fails on PA-Semi */
2174 	stdout_node = call_prom("instance-to-package", 1, 1, prom.stdout);
2175 	if (stdout_node != PROM_ERROR) {
2176 		val = cpu_to_be32(stdout_node);
2177 
2178 		/* If it's a display, note it */
2179 		memset(type, 0, sizeof(type));
2180 		prom_getprop(stdout_node, "device_type", type, sizeof(type));
2181 		if (prom_strcmp(type, "display") == 0)
2182 			prom_setprop(stdout_node, path, "linux,boot-display", NULL, 0);
2183 	}
2184 }
2185 
2186 static int __init prom_find_machine_type(void)
2187 {
2188 	char compat[256];
2189 	int len, i = 0;
2190 #ifdef CONFIG_PPC64
2191 	phandle rtas;
2192 	int x;
2193 #endif
2194 
2195 	/* Look for a PowerMac or a Cell */
2196 	len = prom_getprop(prom.root, "compatible",
2197 			   compat, sizeof(compat)-1);
2198 	if (len > 0) {
2199 		compat[len] = 0;
2200 		while (i < len) {
2201 			char *p = &compat[i];
2202 			int sl = prom_strlen(p);
2203 			if (sl == 0)
2204 				break;
2205 			if (prom_strstr(p, "Power Macintosh") ||
2206 			    prom_strstr(p, "MacRISC"))
2207 				return PLATFORM_POWERMAC;
2208 #ifdef CONFIG_PPC64
2209 			/* We must make sure we don't detect the IBM Cell
2210 			 * blades as pSeries due to some firmware issues,
2211 			 * so we do it here.
2212 			 */
2213 			if (prom_strstr(p, "IBM,CBEA") ||
2214 			    prom_strstr(p, "IBM,CPBW-1.0"))
2215 				return PLATFORM_GENERIC;
2216 #endif /* CONFIG_PPC64 */
2217 			i += sl + 1;
2218 		}
2219 	}
2220 #ifdef CONFIG_PPC64
2221 	/* Try to figure out if it's an IBM pSeries or any other
2222 	 * PAPR compliant platform. We assume it is if :
2223 	 *  - /device_type is "chrp" (please, do NOT use that for future
2224 	 *    non-IBM designs !
2225 	 *  - it has /rtas
2226 	 */
2227 	len = prom_getprop(prom.root, "device_type",
2228 			   compat, sizeof(compat)-1);
2229 	if (len <= 0)
2230 		return PLATFORM_GENERIC;
2231 	if (prom_strcmp(compat, "chrp"))
2232 		return PLATFORM_GENERIC;
2233 
2234 	/* Default to pSeries. We need to know if we are running LPAR */
2235 	rtas = call_prom("finddevice", 1, 1, ADDR("/rtas"));
2236 	if (!PHANDLE_VALID(rtas))
2237 		return PLATFORM_GENERIC;
2238 	x = prom_getproplen(rtas, "ibm,hypertas-functions");
2239 	if (x != PROM_ERROR) {
2240 		prom_debug("Hypertas detected, assuming LPAR !\n");
2241 		return PLATFORM_PSERIES_LPAR;
2242 	}
2243 	return PLATFORM_PSERIES;
2244 #else
2245 	return PLATFORM_GENERIC;
2246 #endif
2247 }
2248 
2249 static int __init prom_set_color(ihandle ih, int i, int r, int g, int b)
2250 {
2251 	return call_prom("call-method", 6, 1, ADDR("color!"), ih, i, b, g, r);
2252 }
2253 
2254 /*
2255  * If we have a display that we don't know how to drive,
2256  * we will want to try to execute OF's open method for it
2257  * later.  However, OF will probably fall over if we do that
2258  * we've taken over the MMU.
2259  * So we check whether we will need to open the display,
2260  * and if so, open it now.
2261  */
2262 static void __init prom_check_displays(void)
2263 {
2264 	char type[16], *path;
2265 	phandle node;
2266 	ihandle ih;
2267 	int i;
2268 
2269 	static const unsigned char default_colors[] __initconst = {
2270 		0x00, 0x00, 0x00,
2271 		0x00, 0x00, 0xaa,
2272 		0x00, 0xaa, 0x00,
2273 		0x00, 0xaa, 0xaa,
2274 		0xaa, 0x00, 0x00,
2275 		0xaa, 0x00, 0xaa,
2276 		0xaa, 0xaa, 0x00,
2277 		0xaa, 0xaa, 0xaa,
2278 		0x55, 0x55, 0x55,
2279 		0x55, 0x55, 0xff,
2280 		0x55, 0xff, 0x55,
2281 		0x55, 0xff, 0xff,
2282 		0xff, 0x55, 0x55,
2283 		0xff, 0x55, 0xff,
2284 		0xff, 0xff, 0x55,
2285 		0xff, 0xff, 0xff
2286 	};
2287 	const unsigned char *clut;
2288 
2289 	prom_debug("Looking for displays\n");
2290 	for (node = 0; prom_next_node(&node); ) {
2291 		memset(type, 0, sizeof(type));
2292 		prom_getprop(node, "device_type", type, sizeof(type));
2293 		if (prom_strcmp(type, "display") != 0)
2294 			continue;
2295 
2296 		/* It seems OF doesn't null-terminate the path :-( */
2297 		path = prom_scratch;
2298 		memset(path, 0, sizeof(prom_scratch));
2299 
2300 		/*
2301 		 * leave some room at the end of the path for appending extra
2302 		 * arguments
2303 		 */
2304 		if (call_prom("package-to-path", 3, 1, node, path,
2305 			      sizeof(prom_scratch) - 10) == PROM_ERROR)
2306 			continue;
2307 		prom_printf("found display   : %s, opening... ", path);
2308 
2309 		ih = call_prom("open", 1, 1, path);
2310 		if (ih == 0) {
2311 			prom_printf("failed\n");
2312 			continue;
2313 		}
2314 
2315 		/* Success */
2316 		prom_printf("done\n");
2317 		prom_setprop(node, path, "linux,opened", NULL, 0);
2318 
2319 		/* Setup a usable color table when the appropriate
2320 		 * method is available. Should update this to set-colors */
2321 		clut = default_colors;
2322 		for (i = 0; i < 16; i++, clut += 3)
2323 			if (prom_set_color(ih, i, clut[0], clut[1],
2324 					   clut[2]) != 0)
2325 				break;
2326 
2327 #ifdef CONFIG_LOGO_LINUX_CLUT224
2328 		clut = PTRRELOC(logo_linux_clut224.clut);
2329 		for (i = 0; i < logo_linux_clut224.clutsize; i++, clut += 3)
2330 			if (prom_set_color(ih, i + 32, clut[0], clut[1],
2331 					   clut[2]) != 0)
2332 				break;
2333 #endif /* CONFIG_LOGO_LINUX_CLUT224 */
2334 
2335 #ifdef CONFIG_PPC_EARLY_DEBUG_BOOTX
2336 		if (prom_getprop(node, "linux,boot-display", NULL, 0) !=
2337 		    PROM_ERROR) {
2338 			u32 width, height, pitch, addr;
2339 
2340 			prom_printf("Setting btext !\n");
2341 			prom_getprop(node, "width", &width, 4);
2342 			prom_getprop(node, "height", &height, 4);
2343 			prom_getprop(node, "linebytes", &pitch, 4);
2344 			prom_getprop(node, "address", &addr, 4);
2345 			prom_printf("W=%d H=%d LB=%d addr=0x%x\n",
2346 				    width, height, pitch, addr);
2347 			btext_setup_display(width, height, 8, pitch, addr);
2348 			btext_prepare_BAT();
2349 		}
2350 #endif /* CONFIG_PPC_EARLY_DEBUG_BOOTX */
2351 	}
2352 }
2353 
2354 
2355 /* Return (relocated) pointer to this much memory: moves initrd if reqd. */
2356 static void __init *make_room(unsigned long *mem_start, unsigned long *mem_end,
2357 			      unsigned long needed, unsigned long align)
2358 {
2359 	void *ret;
2360 
2361 	*mem_start = _ALIGN(*mem_start, align);
2362 	while ((*mem_start + needed) > *mem_end) {
2363 		unsigned long room, chunk;
2364 
2365 		prom_debug("Chunk exhausted, claiming more at %lx...\n",
2366 			   alloc_bottom);
2367 		room = alloc_top - alloc_bottom;
2368 		if (room > DEVTREE_CHUNK_SIZE)
2369 			room = DEVTREE_CHUNK_SIZE;
2370 		if (room < PAGE_SIZE)
2371 			prom_panic("No memory for flatten_device_tree "
2372 				   "(no room)\n");
2373 		chunk = alloc_up(room, 0);
2374 		if (chunk == 0)
2375 			prom_panic("No memory for flatten_device_tree "
2376 				   "(claim failed)\n");
2377 		*mem_end = chunk + room;
2378 	}
2379 
2380 	ret = (void *)*mem_start;
2381 	*mem_start += needed;
2382 
2383 	return ret;
2384 }
2385 
2386 #define dt_push_token(token, mem_start, mem_end) do { 			\
2387 		void *room = make_room(mem_start, mem_end, 4, 4);	\
2388 		*(__be32 *)room = cpu_to_be32(token);			\
2389 	} while(0)
2390 
2391 static unsigned long __init dt_find_string(char *str)
2392 {
2393 	char *s, *os;
2394 
2395 	s = os = (char *)dt_string_start;
2396 	s += 4;
2397 	while (s <  (char *)dt_string_end) {
2398 		if (prom_strcmp(s, str) == 0)
2399 			return s - os;
2400 		s += prom_strlen(s) + 1;
2401 	}
2402 	return 0;
2403 }
2404 
2405 /*
2406  * The Open Firmware 1275 specification states properties must be 31 bytes or
2407  * less, however not all firmwares obey this. Make it 64 bytes to be safe.
2408  */
2409 #define MAX_PROPERTY_NAME 64
2410 
2411 static void __init scan_dt_build_strings(phandle node,
2412 					 unsigned long *mem_start,
2413 					 unsigned long *mem_end)
2414 {
2415 	char *prev_name, *namep, *sstart;
2416 	unsigned long soff;
2417 	phandle child;
2418 
2419 	sstart =  (char *)dt_string_start;
2420 
2421 	/* get and store all property names */
2422 	prev_name = "";
2423 	for (;;) {
2424 		/* 64 is max len of name including nul. */
2425 		namep = make_room(mem_start, mem_end, MAX_PROPERTY_NAME, 1);
2426 		if (call_prom("nextprop", 3, 1, node, prev_name, namep) != 1) {
2427 			/* No more nodes: unwind alloc */
2428 			*mem_start = (unsigned long)namep;
2429 			break;
2430 		}
2431 
2432  		/* skip "name" */
2433 		if (prom_strcmp(namep, "name") == 0) {
2434  			*mem_start = (unsigned long)namep;
2435  			prev_name = "name";
2436  			continue;
2437  		}
2438 		/* get/create string entry */
2439 		soff = dt_find_string(namep);
2440 		if (soff != 0) {
2441 			*mem_start = (unsigned long)namep;
2442 			namep = sstart + soff;
2443 		} else {
2444 			/* Trim off some if we can */
2445 			*mem_start = (unsigned long)namep + prom_strlen(namep) + 1;
2446 			dt_string_end = *mem_start;
2447 		}
2448 		prev_name = namep;
2449 	}
2450 
2451 	/* do all our children */
2452 	child = call_prom("child", 1, 1, node);
2453 	while (child != 0) {
2454 		scan_dt_build_strings(child, mem_start, mem_end);
2455 		child = call_prom("peer", 1, 1, child);
2456 	}
2457 }
2458 
2459 static void __init scan_dt_build_struct(phandle node, unsigned long *mem_start,
2460 					unsigned long *mem_end)
2461 {
2462 	phandle child;
2463 	char *namep, *prev_name, *sstart, *p, *ep, *lp, *path;
2464 	unsigned long soff;
2465 	unsigned char *valp;
2466 	static char pname[MAX_PROPERTY_NAME] __prombss;
2467 	int l, room, has_phandle = 0;
2468 
2469 	dt_push_token(OF_DT_BEGIN_NODE, mem_start, mem_end);
2470 
2471 	/* get the node's full name */
2472 	namep = (char *)*mem_start;
2473 	room = *mem_end - *mem_start;
2474 	if (room > 255)
2475 		room = 255;
2476 	l = call_prom("package-to-path", 3, 1, node, namep, room);
2477 	if (l >= 0) {
2478 		/* Didn't fit?  Get more room. */
2479 		if (l >= room) {
2480 			if (l >= *mem_end - *mem_start)
2481 				namep = make_room(mem_start, mem_end, l+1, 1);
2482 			call_prom("package-to-path", 3, 1, node, namep, l);
2483 		}
2484 		namep[l] = '\0';
2485 
2486 		/* Fixup an Apple bug where they have bogus \0 chars in the
2487 		 * middle of the path in some properties, and extract
2488 		 * the unit name (everything after the last '/').
2489 		 */
2490 		for (lp = p = namep, ep = namep + l; p < ep; p++) {
2491 			if (*p == '/')
2492 				lp = namep;
2493 			else if (*p != 0)
2494 				*lp++ = *p;
2495 		}
2496 		*lp = 0;
2497 		*mem_start = _ALIGN((unsigned long)lp + 1, 4);
2498 	}
2499 
2500 	/* get it again for debugging */
2501 	path = prom_scratch;
2502 	memset(path, 0, sizeof(prom_scratch));
2503 	call_prom("package-to-path", 3, 1, node, path, sizeof(prom_scratch) - 1);
2504 
2505 	/* get and store all properties */
2506 	prev_name = "";
2507 	sstart = (char *)dt_string_start;
2508 	for (;;) {
2509 		if (call_prom("nextprop", 3, 1, node, prev_name,
2510 			      pname) != 1)
2511 			break;
2512 
2513  		/* skip "name" */
2514 		if (prom_strcmp(pname, "name") == 0) {
2515  			prev_name = "name";
2516  			continue;
2517  		}
2518 
2519 		/* find string offset */
2520 		soff = dt_find_string(pname);
2521 		if (soff == 0) {
2522 			prom_printf("WARNING: Can't find string index for"
2523 				    " <%s>, node %s\n", pname, path);
2524 			break;
2525 		}
2526 		prev_name = sstart + soff;
2527 
2528 		/* get length */
2529 		l = call_prom("getproplen", 2, 1, node, pname);
2530 
2531 		/* sanity checks */
2532 		if (l == PROM_ERROR)
2533 			continue;
2534 
2535 		/* push property head */
2536 		dt_push_token(OF_DT_PROP, mem_start, mem_end);
2537 		dt_push_token(l, mem_start, mem_end);
2538 		dt_push_token(soff, mem_start, mem_end);
2539 
2540 		/* push property content */
2541 		valp = make_room(mem_start, mem_end, l, 4);
2542 		call_prom("getprop", 4, 1, node, pname, valp, l);
2543 		*mem_start = _ALIGN(*mem_start, 4);
2544 
2545 		if (!prom_strcmp(pname, "phandle"))
2546 			has_phandle = 1;
2547 	}
2548 
2549 	/* Add a "phandle" property if none already exist */
2550 	if (!has_phandle) {
2551 		soff = dt_find_string("phandle");
2552 		if (soff == 0)
2553 			prom_printf("WARNING: Can't find string index for <phandle> node %s\n", path);
2554 		else {
2555 			dt_push_token(OF_DT_PROP, mem_start, mem_end);
2556 			dt_push_token(4, mem_start, mem_end);
2557 			dt_push_token(soff, mem_start, mem_end);
2558 			valp = make_room(mem_start, mem_end, 4, 4);
2559 			*(__be32 *)valp = cpu_to_be32(node);
2560 		}
2561 	}
2562 
2563 	/* do all our children */
2564 	child = call_prom("child", 1, 1, node);
2565 	while (child != 0) {
2566 		scan_dt_build_struct(child, mem_start, mem_end);
2567 		child = call_prom("peer", 1, 1, child);
2568 	}
2569 
2570 	dt_push_token(OF_DT_END_NODE, mem_start, mem_end);
2571 }
2572 
2573 static void __init flatten_device_tree(void)
2574 {
2575 	phandle root;
2576 	unsigned long mem_start, mem_end, room;
2577 	struct boot_param_header *hdr;
2578 	char *namep;
2579 	u64 *rsvmap;
2580 
2581 	/*
2582 	 * Check how much room we have between alloc top & bottom (+/- a
2583 	 * few pages), crop to 1MB, as this is our "chunk" size
2584 	 */
2585 	room = alloc_top - alloc_bottom - 0x4000;
2586 	if (room > DEVTREE_CHUNK_SIZE)
2587 		room = DEVTREE_CHUNK_SIZE;
2588 	prom_debug("starting device tree allocs at %lx\n", alloc_bottom);
2589 
2590 	/* Now try to claim that */
2591 	mem_start = (unsigned long)alloc_up(room, PAGE_SIZE);
2592 	if (mem_start == 0)
2593 		prom_panic("Can't allocate initial device-tree chunk\n");
2594 	mem_end = mem_start + room;
2595 
2596 	/* Get root of tree */
2597 	root = call_prom("peer", 1, 1, (phandle)0);
2598 	if (root == (phandle)0)
2599 		prom_panic ("couldn't get device tree root\n");
2600 
2601 	/* Build header and make room for mem rsv map */
2602 	mem_start = _ALIGN(mem_start, 4);
2603 	hdr = make_room(&mem_start, &mem_end,
2604 			sizeof(struct boot_param_header), 4);
2605 	dt_header_start = (unsigned long)hdr;
2606 	rsvmap = make_room(&mem_start, &mem_end, sizeof(mem_reserve_map), 8);
2607 
2608 	/* Start of strings */
2609 	mem_start = PAGE_ALIGN(mem_start);
2610 	dt_string_start = mem_start;
2611 	mem_start += 4; /* hole */
2612 
2613 	/* Add "phandle" in there, we'll need it */
2614 	namep = make_room(&mem_start, &mem_end, 16, 1);
2615 	prom_strcpy(namep, "phandle");
2616 	mem_start = (unsigned long)namep + prom_strlen(namep) + 1;
2617 
2618 	/* Build string array */
2619 	prom_printf("Building dt strings...\n");
2620 	scan_dt_build_strings(root, &mem_start, &mem_end);
2621 	dt_string_end = mem_start;
2622 
2623 	/* Build structure */
2624 	mem_start = PAGE_ALIGN(mem_start);
2625 	dt_struct_start = mem_start;
2626 	prom_printf("Building dt structure...\n");
2627 	scan_dt_build_struct(root, &mem_start, &mem_end);
2628 	dt_push_token(OF_DT_END, &mem_start, &mem_end);
2629 	dt_struct_end = PAGE_ALIGN(mem_start);
2630 
2631 	/* Finish header */
2632 	hdr->boot_cpuid_phys = cpu_to_be32(prom.cpu);
2633 	hdr->magic = cpu_to_be32(OF_DT_HEADER);
2634 	hdr->totalsize = cpu_to_be32(dt_struct_end - dt_header_start);
2635 	hdr->off_dt_struct = cpu_to_be32(dt_struct_start - dt_header_start);
2636 	hdr->off_dt_strings = cpu_to_be32(dt_string_start - dt_header_start);
2637 	hdr->dt_strings_size = cpu_to_be32(dt_string_end - dt_string_start);
2638 	hdr->off_mem_rsvmap = cpu_to_be32(((unsigned long)rsvmap) - dt_header_start);
2639 	hdr->version = cpu_to_be32(OF_DT_VERSION);
2640 	/* Version 16 is not backward compatible */
2641 	hdr->last_comp_version = cpu_to_be32(0x10);
2642 
2643 	/* Copy the reserve map in */
2644 	memcpy(rsvmap, mem_reserve_map, sizeof(mem_reserve_map));
2645 
2646 #ifdef DEBUG_PROM
2647 	{
2648 		int i;
2649 		prom_printf("reserved memory map:\n");
2650 		for (i = 0; i < mem_reserve_cnt; i++)
2651 			prom_printf("  %llx - %llx\n",
2652 				    be64_to_cpu(mem_reserve_map[i].base),
2653 				    be64_to_cpu(mem_reserve_map[i].size));
2654 	}
2655 #endif
2656 	/* Bump mem_reserve_cnt to cause further reservations to fail
2657 	 * since it's too late.
2658 	 */
2659 	mem_reserve_cnt = MEM_RESERVE_MAP_SIZE;
2660 
2661 	prom_printf("Device tree strings 0x%lx -> 0x%lx\n",
2662 		    dt_string_start, dt_string_end);
2663 	prom_printf("Device tree struct  0x%lx -> 0x%lx\n",
2664 		    dt_struct_start, dt_struct_end);
2665 }
2666 
2667 #ifdef CONFIG_PPC_MAPLE
2668 /* PIBS Version 1.05.0000 04/26/2005 has an incorrect /ht/isa/ranges property.
2669  * The values are bad, and it doesn't even have the right number of cells. */
2670 static void __init fixup_device_tree_maple(void)
2671 {
2672 	phandle isa;
2673 	u32 rloc = 0x01002000; /* IO space; PCI device = 4 */
2674 	u32 isa_ranges[6];
2675 	char *name;
2676 
2677 	name = "/ht@0/isa@4";
2678 	isa = call_prom("finddevice", 1, 1, ADDR(name));
2679 	if (!PHANDLE_VALID(isa)) {
2680 		name = "/ht@0/isa@6";
2681 		isa = call_prom("finddevice", 1, 1, ADDR(name));
2682 		rloc = 0x01003000; /* IO space; PCI device = 6 */
2683 	}
2684 	if (!PHANDLE_VALID(isa))
2685 		return;
2686 
2687 	if (prom_getproplen(isa, "ranges") != 12)
2688 		return;
2689 	if (prom_getprop(isa, "ranges", isa_ranges, sizeof(isa_ranges))
2690 		== PROM_ERROR)
2691 		return;
2692 
2693 	if (isa_ranges[0] != 0x1 ||
2694 		isa_ranges[1] != 0xf4000000 ||
2695 		isa_ranges[2] != 0x00010000)
2696 		return;
2697 
2698 	prom_printf("Fixing up bogus ISA range on Maple/Apache...\n");
2699 
2700 	isa_ranges[0] = 0x1;
2701 	isa_ranges[1] = 0x0;
2702 	isa_ranges[2] = rloc;
2703 	isa_ranges[3] = 0x0;
2704 	isa_ranges[4] = 0x0;
2705 	isa_ranges[5] = 0x00010000;
2706 	prom_setprop(isa, name, "ranges",
2707 			isa_ranges, sizeof(isa_ranges));
2708 }
2709 
2710 #define CPC925_MC_START		0xf8000000
2711 #define CPC925_MC_LENGTH	0x1000000
2712 /* The values for memory-controller don't have right number of cells */
2713 static void __init fixup_device_tree_maple_memory_controller(void)
2714 {
2715 	phandle mc;
2716 	u32 mc_reg[4];
2717 	char *name = "/hostbridge@f8000000";
2718 	u32 ac, sc;
2719 
2720 	mc = call_prom("finddevice", 1, 1, ADDR(name));
2721 	if (!PHANDLE_VALID(mc))
2722 		return;
2723 
2724 	if (prom_getproplen(mc, "reg") != 8)
2725 		return;
2726 
2727 	prom_getprop(prom.root, "#address-cells", &ac, sizeof(ac));
2728 	prom_getprop(prom.root, "#size-cells", &sc, sizeof(sc));
2729 	if ((ac != 2) || (sc != 2))
2730 		return;
2731 
2732 	if (prom_getprop(mc, "reg", mc_reg, sizeof(mc_reg)) == PROM_ERROR)
2733 		return;
2734 
2735 	if (mc_reg[0] != CPC925_MC_START || mc_reg[1] != CPC925_MC_LENGTH)
2736 		return;
2737 
2738 	prom_printf("Fixing up bogus hostbridge on Maple...\n");
2739 
2740 	mc_reg[0] = 0x0;
2741 	mc_reg[1] = CPC925_MC_START;
2742 	mc_reg[2] = 0x0;
2743 	mc_reg[3] = CPC925_MC_LENGTH;
2744 	prom_setprop(mc, name, "reg", mc_reg, sizeof(mc_reg));
2745 }
2746 #else
2747 #define fixup_device_tree_maple()
2748 #define fixup_device_tree_maple_memory_controller()
2749 #endif
2750 
2751 #ifdef CONFIG_PPC_CHRP
2752 /*
2753  * Pegasos and BriQ lacks the "ranges" property in the isa node
2754  * Pegasos needs decimal IRQ 14/15, not hexadecimal
2755  * Pegasos has the IDE configured in legacy mode, but advertised as native
2756  */
2757 static void __init fixup_device_tree_chrp(void)
2758 {
2759 	phandle ph;
2760 	u32 prop[6];
2761 	u32 rloc = 0x01006000; /* IO space; PCI device = 12 */
2762 	char *name;
2763 	int rc;
2764 
2765 	name = "/pci@80000000/isa@c";
2766 	ph = call_prom("finddevice", 1, 1, ADDR(name));
2767 	if (!PHANDLE_VALID(ph)) {
2768 		name = "/pci@ff500000/isa@6";
2769 		ph = call_prom("finddevice", 1, 1, ADDR(name));
2770 		rloc = 0x01003000; /* IO space; PCI device = 6 */
2771 	}
2772 	if (PHANDLE_VALID(ph)) {
2773 		rc = prom_getproplen(ph, "ranges");
2774 		if (rc == 0 || rc == PROM_ERROR) {
2775 			prom_printf("Fixing up missing ISA range on Pegasos...\n");
2776 
2777 			prop[0] = 0x1;
2778 			prop[1] = 0x0;
2779 			prop[2] = rloc;
2780 			prop[3] = 0x0;
2781 			prop[4] = 0x0;
2782 			prop[5] = 0x00010000;
2783 			prom_setprop(ph, name, "ranges", prop, sizeof(prop));
2784 		}
2785 	}
2786 
2787 	name = "/pci@80000000/ide@C,1";
2788 	ph = call_prom("finddevice", 1, 1, ADDR(name));
2789 	if (PHANDLE_VALID(ph)) {
2790 		prom_printf("Fixing up IDE interrupt on Pegasos...\n");
2791 		prop[0] = 14;
2792 		prop[1] = 0x0;
2793 		prom_setprop(ph, name, "interrupts", prop, 2*sizeof(u32));
2794 		prom_printf("Fixing up IDE class-code on Pegasos...\n");
2795 		rc = prom_getprop(ph, "class-code", prop, sizeof(u32));
2796 		if (rc == sizeof(u32)) {
2797 			prop[0] &= ~0x5;
2798 			prom_setprop(ph, name, "class-code", prop, sizeof(u32));
2799 		}
2800 	}
2801 }
2802 #else
2803 #define fixup_device_tree_chrp()
2804 #endif
2805 
2806 #if defined(CONFIG_PPC64) && defined(CONFIG_PPC_PMAC)
2807 static void __init fixup_device_tree_pmac(void)
2808 {
2809 	phandle u3, i2c, mpic;
2810 	u32 u3_rev;
2811 	u32 interrupts[2];
2812 	u32 parent;
2813 
2814 	/* Some G5s have a missing interrupt definition, fix it up here */
2815 	u3 = call_prom("finddevice", 1, 1, ADDR("/u3@0,f8000000"));
2816 	if (!PHANDLE_VALID(u3))
2817 		return;
2818 	i2c = call_prom("finddevice", 1, 1, ADDR("/u3@0,f8000000/i2c@f8001000"));
2819 	if (!PHANDLE_VALID(i2c))
2820 		return;
2821 	mpic = call_prom("finddevice", 1, 1, ADDR("/u3@0,f8000000/mpic@f8040000"));
2822 	if (!PHANDLE_VALID(mpic))
2823 		return;
2824 
2825 	/* check if proper rev of u3 */
2826 	if (prom_getprop(u3, "device-rev", &u3_rev, sizeof(u3_rev))
2827 	    == PROM_ERROR)
2828 		return;
2829 	if (u3_rev < 0x35 || u3_rev > 0x39)
2830 		return;
2831 	/* does it need fixup ? */
2832 	if (prom_getproplen(i2c, "interrupts") > 0)
2833 		return;
2834 
2835 	prom_printf("fixing up bogus interrupts for u3 i2c...\n");
2836 
2837 	/* interrupt on this revision of u3 is number 0 and level */
2838 	interrupts[0] = 0;
2839 	interrupts[1] = 1;
2840 	prom_setprop(i2c, "/u3@0,f8000000/i2c@f8001000", "interrupts",
2841 		     &interrupts, sizeof(interrupts));
2842 	parent = (u32)mpic;
2843 	prom_setprop(i2c, "/u3@0,f8000000/i2c@f8001000", "interrupt-parent",
2844 		     &parent, sizeof(parent));
2845 }
2846 #else
2847 #define fixup_device_tree_pmac()
2848 #endif
2849 
2850 #ifdef CONFIG_PPC_EFIKA
2851 /*
2852  * The MPC5200 FEC driver requires an phy-handle property to tell it how
2853  * to talk to the phy.  If the phy-handle property is missing, then this
2854  * function is called to add the appropriate nodes and link it to the
2855  * ethernet node.
2856  */
2857 static void __init fixup_device_tree_efika_add_phy(void)
2858 {
2859 	u32 node;
2860 	char prop[64];
2861 	int rv;
2862 
2863 	/* Check if /builtin/ethernet exists - bail if it doesn't */
2864 	node = call_prom("finddevice", 1, 1, ADDR("/builtin/ethernet"));
2865 	if (!PHANDLE_VALID(node))
2866 		return;
2867 
2868 	/* Check if the phy-handle property exists - bail if it does */
2869 	rv = prom_getprop(node, "phy-handle", prop, sizeof(prop));
2870 	if (!rv)
2871 		return;
2872 
2873 	/*
2874 	 * At this point the ethernet device doesn't have a phy described.
2875 	 * Now we need to add the missing phy node and linkage
2876 	 */
2877 
2878 	/* Check for an MDIO bus node - if missing then create one */
2879 	node = call_prom("finddevice", 1, 1, ADDR("/builtin/mdio"));
2880 	if (!PHANDLE_VALID(node)) {
2881 		prom_printf("Adding Ethernet MDIO node\n");
2882 		call_prom("interpret", 1, 1,
2883 			" s\" /builtin\" find-device"
2884 			" new-device"
2885 				" 1 encode-int s\" #address-cells\" property"
2886 				" 0 encode-int s\" #size-cells\" property"
2887 				" s\" mdio\" device-name"
2888 				" s\" fsl,mpc5200b-mdio\" encode-string"
2889 				" s\" compatible\" property"
2890 				" 0xf0003000 0x400 reg"
2891 				" 0x2 encode-int"
2892 				" 0x5 encode-int encode+"
2893 				" 0x3 encode-int encode+"
2894 				" s\" interrupts\" property"
2895 			" finish-device");
2896 	};
2897 
2898 	/* Check for a PHY device node - if missing then create one and
2899 	 * give it's phandle to the ethernet node */
2900 	node = call_prom("finddevice", 1, 1,
2901 			 ADDR("/builtin/mdio/ethernet-phy"));
2902 	if (!PHANDLE_VALID(node)) {
2903 		prom_printf("Adding Ethernet PHY node\n");
2904 		call_prom("interpret", 1, 1,
2905 			" s\" /builtin/mdio\" find-device"
2906 			" new-device"
2907 				" s\" ethernet-phy\" device-name"
2908 				" 0x10 encode-int s\" reg\" property"
2909 				" my-self"
2910 				" ihandle>phandle"
2911 			" finish-device"
2912 			" s\" /builtin/ethernet\" find-device"
2913 				" encode-int"
2914 				" s\" phy-handle\" property"
2915 			" device-end");
2916 	}
2917 }
2918 
2919 static void __init fixup_device_tree_efika(void)
2920 {
2921 	int sound_irq[3] = { 2, 2, 0 };
2922 	int bcomm_irq[3*16] = { 3,0,0, 3,1,0, 3,2,0, 3,3,0,
2923 				3,4,0, 3,5,0, 3,6,0, 3,7,0,
2924 				3,8,0, 3,9,0, 3,10,0, 3,11,0,
2925 				3,12,0, 3,13,0, 3,14,0, 3,15,0 };
2926 	u32 node;
2927 	char prop[64];
2928 	int rv, len;
2929 
2930 	/* Check if we're really running on a EFIKA */
2931 	node = call_prom("finddevice", 1, 1, ADDR("/"));
2932 	if (!PHANDLE_VALID(node))
2933 		return;
2934 
2935 	rv = prom_getprop(node, "model", prop, sizeof(prop));
2936 	if (rv == PROM_ERROR)
2937 		return;
2938 	if (prom_strcmp(prop, "EFIKA5K2"))
2939 		return;
2940 
2941 	prom_printf("Applying EFIKA device tree fixups\n");
2942 
2943 	/* Claiming to be 'chrp' is death */
2944 	node = call_prom("finddevice", 1, 1, ADDR("/"));
2945 	rv = prom_getprop(node, "device_type", prop, sizeof(prop));
2946 	if (rv != PROM_ERROR && (prom_strcmp(prop, "chrp") == 0))
2947 		prom_setprop(node, "/", "device_type", "efika", sizeof("efika"));
2948 
2949 	/* CODEGEN,description is exposed in /proc/cpuinfo so
2950 	   fix that too */
2951 	rv = prom_getprop(node, "CODEGEN,description", prop, sizeof(prop));
2952 	if (rv != PROM_ERROR && (prom_strstr(prop, "CHRP")))
2953 		prom_setprop(node, "/", "CODEGEN,description",
2954 			     "Efika 5200B PowerPC System",
2955 			     sizeof("Efika 5200B PowerPC System"));
2956 
2957 	/* Fixup bestcomm interrupts property */
2958 	node = call_prom("finddevice", 1, 1, ADDR("/builtin/bestcomm"));
2959 	if (PHANDLE_VALID(node)) {
2960 		len = prom_getproplen(node, "interrupts");
2961 		if (len == 12) {
2962 			prom_printf("Fixing bestcomm interrupts property\n");
2963 			prom_setprop(node, "/builtin/bestcom", "interrupts",
2964 				     bcomm_irq, sizeof(bcomm_irq));
2965 		}
2966 	}
2967 
2968 	/* Fixup sound interrupts property */
2969 	node = call_prom("finddevice", 1, 1, ADDR("/builtin/sound"));
2970 	if (PHANDLE_VALID(node)) {
2971 		rv = prom_getprop(node, "interrupts", prop, sizeof(prop));
2972 		if (rv == PROM_ERROR) {
2973 			prom_printf("Adding sound interrupts property\n");
2974 			prom_setprop(node, "/builtin/sound", "interrupts",
2975 				     sound_irq, sizeof(sound_irq));
2976 		}
2977 	}
2978 
2979 	/* Make sure ethernet phy-handle property exists */
2980 	fixup_device_tree_efika_add_phy();
2981 }
2982 #else
2983 #define fixup_device_tree_efika()
2984 #endif
2985 
2986 #ifdef CONFIG_PPC_PASEMI_NEMO
2987 /*
2988  * CFE supplied on Nemo is broken in several ways, biggest
2989  * problem is that it reassigns ISA interrupts to unused mpic ints.
2990  * Add an interrupt-controller property for the io-bridge to use
2991  * and correct the ints so we can attach them to an irq_domain
2992  */
2993 static void __init fixup_device_tree_pasemi(void)
2994 {
2995 	u32 interrupts[2], parent, rval, val = 0;
2996 	char *name, *pci_name;
2997 	phandle iob, node;
2998 
2999 	/* Find the root pci node */
3000 	name = "/pxp@0,e0000000";
3001 	iob = call_prom("finddevice", 1, 1, ADDR(name));
3002 	if (!PHANDLE_VALID(iob))
3003 		return;
3004 
3005 	/* check if interrupt-controller node set yet */
3006 	if (prom_getproplen(iob, "interrupt-controller") !=PROM_ERROR)
3007 		return;
3008 
3009 	prom_printf("adding interrupt-controller property for SB600...\n");
3010 
3011 	prom_setprop(iob, name, "interrupt-controller", &val, 0);
3012 
3013 	pci_name = "/pxp@0,e0000000/pci@11";
3014 	node = call_prom("finddevice", 1, 1, ADDR(pci_name));
3015 	parent = ADDR(iob);
3016 
3017 	for( ; prom_next_node(&node); ) {
3018 		/* scan each node for one with an interrupt */
3019 		if (!PHANDLE_VALID(node))
3020 			continue;
3021 
3022 		rval = prom_getproplen(node, "interrupts");
3023 		if (rval == 0 || rval == PROM_ERROR)
3024 			continue;
3025 
3026 		prom_getprop(node, "interrupts", &interrupts, sizeof(interrupts));
3027 		if ((interrupts[0] < 212) || (interrupts[0] > 222))
3028 			continue;
3029 
3030 		/* found a node, update both interrupts and interrupt-parent */
3031 		if ((interrupts[0] >= 212) && (interrupts[0] <= 215))
3032 			interrupts[0] -= 203;
3033 		if ((interrupts[0] >= 216) && (interrupts[0] <= 220))
3034 			interrupts[0] -= 213;
3035 		if (interrupts[0] == 221)
3036 			interrupts[0] = 14;
3037 		if (interrupts[0] == 222)
3038 			interrupts[0] = 8;
3039 
3040 		prom_setprop(node, pci_name, "interrupts", interrupts,
3041 					sizeof(interrupts));
3042 		prom_setprop(node, pci_name, "interrupt-parent", &parent,
3043 					sizeof(parent));
3044 	}
3045 
3046 	/*
3047 	 * The io-bridge has device_type set to 'io-bridge' change it to 'isa'
3048 	 * so that generic isa-bridge code can add the SB600 and its on-board
3049 	 * peripherals.
3050 	 */
3051 	name = "/pxp@0,e0000000/io-bridge@0";
3052 	iob = call_prom("finddevice", 1, 1, ADDR(name));
3053 	if (!PHANDLE_VALID(iob))
3054 		return;
3055 
3056 	/* device_type is already set, just change it. */
3057 
3058 	prom_printf("Changing device_type of SB600 node...\n");
3059 
3060 	prom_setprop(iob, name, "device_type", "isa", sizeof("isa"));
3061 }
3062 #else	/* !CONFIG_PPC_PASEMI_NEMO */
3063 static inline void fixup_device_tree_pasemi(void) { }
3064 #endif
3065 
3066 static void __init fixup_device_tree(void)
3067 {
3068 	fixup_device_tree_maple();
3069 	fixup_device_tree_maple_memory_controller();
3070 	fixup_device_tree_chrp();
3071 	fixup_device_tree_pmac();
3072 	fixup_device_tree_efika();
3073 	fixup_device_tree_pasemi();
3074 }
3075 
3076 static void __init prom_find_boot_cpu(void)
3077 {
3078 	__be32 rval;
3079 	ihandle prom_cpu;
3080 	phandle cpu_pkg;
3081 
3082 	rval = 0;
3083 	if (prom_getprop(prom.chosen, "cpu", &rval, sizeof(rval)) <= 0)
3084 		return;
3085 	prom_cpu = be32_to_cpu(rval);
3086 
3087 	cpu_pkg = call_prom("instance-to-package", 1, 1, prom_cpu);
3088 
3089 	if (!PHANDLE_VALID(cpu_pkg))
3090 		return;
3091 
3092 	prom_getprop(cpu_pkg, "reg", &rval, sizeof(rval));
3093 	prom.cpu = be32_to_cpu(rval);
3094 
3095 	prom_debug("Booting CPU hw index = %d\n", prom.cpu);
3096 }
3097 
3098 static void __init prom_check_initrd(unsigned long r3, unsigned long r4)
3099 {
3100 #ifdef CONFIG_BLK_DEV_INITRD
3101 	if (r3 && r4 && r4 != 0xdeadbeef) {
3102 		__be64 val;
3103 
3104 		prom_initrd_start = is_kernel_addr(r3) ? __pa(r3) : r3;
3105 		prom_initrd_end = prom_initrd_start + r4;
3106 
3107 		val = cpu_to_be64(prom_initrd_start);
3108 		prom_setprop(prom.chosen, "/chosen", "linux,initrd-start",
3109 			     &val, sizeof(val));
3110 		val = cpu_to_be64(prom_initrd_end);
3111 		prom_setprop(prom.chosen, "/chosen", "linux,initrd-end",
3112 			     &val, sizeof(val));
3113 
3114 		reserve_mem(prom_initrd_start,
3115 			    prom_initrd_end - prom_initrd_start);
3116 
3117 		prom_debug("initrd_start=0x%lx\n", prom_initrd_start);
3118 		prom_debug("initrd_end=0x%lx\n", prom_initrd_end);
3119 	}
3120 #endif /* CONFIG_BLK_DEV_INITRD */
3121 }
3122 
3123 #ifdef CONFIG_PPC64
3124 #ifdef CONFIG_RELOCATABLE
3125 static void reloc_toc(void)
3126 {
3127 }
3128 
3129 static void unreloc_toc(void)
3130 {
3131 }
3132 #else
3133 static void __reloc_toc(unsigned long offset, unsigned long nr_entries)
3134 {
3135 	unsigned long i;
3136 	unsigned long *toc_entry;
3137 
3138 	/* Get the start of the TOC by using r2 directly. */
3139 	asm volatile("addi %0,2,-0x8000" : "=b" (toc_entry));
3140 
3141 	for (i = 0; i < nr_entries; i++) {
3142 		*toc_entry = *toc_entry + offset;
3143 		toc_entry++;
3144 	}
3145 }
3146 
3147 static void reloc_toc(void)
3148 {
3149 	unsigned long offset = reloc_offset();
3150 	unsigned long nr_entries =
3151 		(__prom_init_toc_end - __prom_init_toc_start) / sizeof(long);
3152 
3153 	__reloc_toc(offset, nr_entries);
3154 
3155 	mb();
3156 }
3157 
3158 static void unreloc_toc(void)
3159 {
3160 	unsigned long offset = reloc_offset();
3161 	unsigned long nr_entries =
3162 		(__prom_init_toc_end - __prom_init_toc_start) / sizeof(long);
3163 
3164 	mb();
3165 
3166 	__reloc_toc(-offset, nr_entries);
3167 }
3168 #endif
3169 #endif
3170 
3171 /*
3172  * We enter here early on, when the Open Firmware prom is still
3173  * handling exceptions and the MMU hash table for us.
3174  */
3175 
3176 unsigned long __init prom_init(unsigned long r3, unsigned long r4,
3177 			       unsigned long pp,
3178 			       unsigned long r6, unsigned long r7,
3179 			       unsigned long kbase)
3180 {
3181 	unsigned long hdr;
3182 
3183 #ifdef CONFIG_PPC32
3184 	unsigned long offset = reloc_offset();
3185 	reloc_got2(offset);
3186 #else
3187 	reloc_toc();
3188 #endif
3189 
3190 	/*
3191 	 * First zero the BSS
3192 	 */
3193 	memset(&__bss_start, 0, __bss_stop - __bss_start);
3194 
3195 	/*
3196 	 * Init interface to Open Firmware, get some node references,
3197 	 * like /chosen
3198 	 */
3199 	prom_init_client_services(pp);
3200 
3201 	/*
3202 	 * See if this OF is old enough that we need to do explicit maps
3203 	 * and other workarounds
3204 	 */
3205 	prom_find_mmu();
3206 
3207 	/*
3208 	 * Init prom stdout device
3209 	 */
3210 	prom_init_stdout();
3211 
3212 	prom_printf("Preparing to boot %s", linux_banner);
3213 
3214 	/*
3215 	 * Get default machine type. At this point, we do not differentiate
3216 	 * between pSeries SMP and pSeries LPAR
3217 	 */
3218 	of_platform = prom_find_machine_type();
3219 	prom_printf("Detected machine type: %x\n", of_platform);
3220 
3221 #ifndef CONFIG_NONSTATIC_KERNEL
3222 	/* Bail if this is a kdump kernel. */
3223 	if (PHYSICAL_START > 0)
3224 		prom_panic("Error: You can't boot a kdump kernel from OF!\n");
3225 #endif
3226 
3227 	/*
3228 	 * Check for an initrd
3229 	 */
3230 	prom_check_initrd(r3, r4);
3231 
3232 	/*
3233 	 * Do early parsing of command line
3234 	 */
3235 	early_cmdline_parse();
3236 
3237 #ifdef CONFIG_PPC_PSERIES
3238 	/*
3239 	 * On pSeries, inform the firmware about our capabilities
3240 	 */
3241 	if (of_platform == PLATFORM_PSERIES ||
3242 	    of_platform == PLATFORM_PSERIES_LPAR)
3243 		prom_send_capabilities();
3244 #endif
3245 
3246 	/*
3247 	 * Copy the CPU hold code
3248 	 */
3249 	if (of_platform != PLATFORM_POWERMAC)
3250 		copy_and_flush(0, kbase, 0x100, 0);
3251 
3252 	/*
3253 	 * Initialize memory management within prom_init
3254 	 */
3255 	prom_init_mem();
3256 
3257 	/*
3258 	 * Determine which cpu is actually running right _now_
3259 	 */
3260 	prom_find_boot_cpu();
3261 
3262 	/*
3263 	 * Initialize display devices
3264 	 */
3265 	prom_check_displays();
3266 
3267 #if defined(CONFIG_PPC64) && defined(__BIG_ENDIAN__)
3268 	/*
3269 	 * Initialize IOMMU (TCE tables) on pSeries. Do that before anything else
3270 	 * that uses the allocator, we need to make sure we get the top of memory
3271 	 * available for us here...
3272 	 */
3273 	if (of_platform == PLATFORM_PSERIES)
3274 		prom_initialize_tce_table();
3275 #endif
3276 
3277 	/*
3278 	 * On non-powermacs, try to instantiate RTAS. PowerMacs don't
3279 	 * have a usable RTAS implementation.
3280 	 */
3281 	if (of_platform != PLATFORM_POWERMAC)
3282 		prom_instantiate_rtas();
3283 
3284 #ifdef CONFIG_PPC64
3285 	/* instantiate sml */
3286 	prom_instantiate_sml();
3287 #endif
3288 
3289 	/*
3290 	 * On non-powermacs, put all CPUs in spin-loops.
3291 	 *
3292 	 * PowerMacs use a different mechanism to spin CPUs
3293 	 *
3294 	 * (This must be done after instanciating RTAS)
3295 	 */
3296 	if (of_platform != PLATFORM_POWERMAC)
3297 		prom_hold_cpus();
3298 
3299 	/*
3300 	 * Fill in some infos for use by the kernel later on
3301 	 */
3302 	if (prom_memory_limit) {
3303 		__be64 val = cpu_to_be64(prom_memory_limit);
3304 		prom_setprop(prom.chosen, "/chosen", "linux,memory-limit",
3305 			     &val, sizeof(val));
3306 	}
3307 #ifdef CONFIG_PPC64
3308 	if (prom_iommu_off)
3309 		prom_setprop(prom.chosen, "/chosen", "linux,iommu-off",
3310 			     NULL, 0);
3311 
3312 	if (prom_iommu_force_on)
3313 		prom_setprop(prom.chosen, "/chosen", "linux,iommu-force-on",
3314 			     NULL, 0);
3315 
3316 	if (prom_tce_alloc_start) {
3317 		prom_setprop(prom.chosen, "/chosen", "linux,tce-alloc-start",
3318 			     &prom_tce_alloc_start,
3319 			     sizeof(prom_tce_alloc_start));
3320 		prom_setprop(prom.chosen, "/chosen", "linux,tce-alloc-end",
3321 			     &prom_tce_alloc_end,
3322 			     sizeof(prom_tce_alloc_end));
3323 	}
3324 #endif
3325 
3326 	/*
3327 	 * Fixup any known bugs in the device-tree
3328 	 */
3329 	fixup_device_tree();
3330 
3331 	/*
3332 	 * Now finally create the flattened device-tree
3333 	 */
3334 	prom_printf("copying OF device tree...\n");
3335 	flatten_device_tree();
3336 
3337 	/*
3338 	 * in case stdin is USB and still active on IBM machines...
3339 	 * Unfortunately quiesce crashes on some powermacs if we have
3340 	 * closed stdin already (in particular the powerbook 101).
3341 	 */
3342 	if (of_platform != PLATFORM_POWERMAC)
3343 		prom_close_stdin();
3344 
3345 	/*
3346 	 * Call OF "quiesce" method to shut down pending DMA's from
3347 	 * devices etc...
3348 	 */
3349 	prom_printf("Quiescing Open Firmware ...\n");
3350 	call_prom("quiesce", 0, 0);
3351 
3352 	/*
3353 	 * And finally, call the kernel passing it the flattened device
3354 	 * tree and NULL as r5, thus triggering the new entry point which
3355 	 * is common to us and kexec
3356 	 */
3357 	hdr = dt_header_start;
3358 
3359 	/* Don't print anything after quiesce under OPAL, it crashes OFW */
3360 	prom_printf("Booting Linux via __start() @ 0x%lx ...\n", kbase);
3361 	prom_debug("->dt_header_start=0x%lx\n", hdr);
3362 
3363 #ifdef CONFIG_PPC32
3364 	reloc_got2(-offset);
3365 #else
3366 	unreloc_toc();
3367 #endif
3368 
3369 	__start(hdr, kbase, 0, 0, 0, 0, 0);
3370 
3371 	return 0;
3372 }
3373