xref: /linux/arch/powerpc/kernel/prom_init.c (revision b3c2ffd5343645fc9b46f67e8c0eaac1e2dde7b4)
1 /*
2  * Procedures for interfacing to Open Firmware.
3  *
4  * Paul Mackerras	August 1996.
5  * Copyright (C) 1996-2005 Paul Mackerras.
6  *
7  *  Adapted for 64bit PowerPC by Dave Engebretsen and Peter Bergner.
8  *    {engebret|bergner}@us.ibm.com
9  *
10  *      This program is free software; you can redistribute it and/or
11  *      modify it under the terms of the GNU General Public License
12  *      as published by the Free Software Foundation; either version
13  *      2 of the License, or (at your option) any later version.
14  */
15 
16 #undef DEBUG_PROM
17 
18 #include <stdarg.h>
19 #include <linux/config.h>
20 #include <linux/kernel.h>
21 #include <linux/string.h>
22 #include <linux/init.h>
23 #include <linux/threads.h>
24 #include <linux/spinlock.h>
25 #include <linux/types.h>
26 #include <linux/pci.h>
27 #include <linux/proc_fs.h>
28 #include <linux/stringify.h>
29 #include <linux/delay.h>
30 #include <linux/initrd.h>
31 #include <linux/bitops.h>
32 #include <asm/prom.h>
33 #include <asm/rtas.h>
34 #include <asm/page.h>
35 #include <asm/processor.h>
36 #include <asm/irq.h>
37 #include <asm/io.h>
38 #include <asm/smp.h>
39 #include <asm/system.h>
40 #include <asm/mmu.h>
41 #include <asm/pgtable.h>
42 #include <asm/pci.h>
43 #include <asm/iommu.h>
44 #include <asm/btext.h>
45 #include <asm/sections.h>
46 #include <asm/machdep.h>
47 
48 #ifdef CONFIG_LOGO_LINUX_CLUT224
49 #include <linux/linux_logo.h>
50 extern const struct linux_logo logo_linux_clut224;
51 #endif
52 
53 /*
54  * Properties whose value is longer than this get excluded from our
55  * copy of the device tree. This value does need to be big enough to
56  * ensure that we don't lose things like the interrupt-map property
57  * on a PCI-PCI bridge.
58  */
59 #define MAX_PROPERTY_LENGTH	(1UL * 1024 * 1024)
60 
61 /*
62  * Eventually bump that one up
63  */
64 #define DEVTREE_CHUNK_SIZE	0x100000
65 
66 /*
67  * This is the size of the local memory reserve map that gets copied
68  * into the boot params passed to the kernel. That size is totally
69  * flexible as the kernel just reads the list until it encounters an
70  * entry with size 0, so it can be changed without breaking binary
71  * compatibility
72  */
73 #define MEM_RESERVE_MAP_SIZE	8
74 
75 /*
76  * prom_init() is called very early on, before the kernel text
77  * and data have been mapped to KERNELBASE.  At this point the code
78  * is running at whatever address it has been loaded at.
79  * On ppc32 we compile with -mrelocatable, which means that references
80  * to extern and static variables get relocated automatically.
81  * On ppc64 we have to relocate the references explicitly with
82  * RELOC.  (Note that strings count as static variables.)
83  *
84  * Because OF may have mapped I/O devices into the area starting at
85  * KERNELBASE, particularly on CHRP machines, we can't safely call
86  * OF once the kernel has been mapped to KERNELBASE.  Therefore all
87  * OF calls must be done within prom_init().
88  *
89  * ADDR is used in calls to call_prom.  The 4th and following
90  * arguments to call_prom should be 32-bit values.
91  * On ppc64, 64 bit values are truncated to 32 bits (and
92  * fortunately don't get interpreted as two arguments).
93  */
94 #ifdef CONFIG_PPC64
95 #define RELOC(x)        (*PTRRELOC(&(x)))
96 #define ADDR(x)		(u32) add_reloc_offset((unsigned long)(x))
97 #define OF_WORKAROUNDS	0
98 #else
99 #define RELOC(x)	(x)
100 #define ADDR(x)		(u32) (x)
101 #define OF_WORKAROUNDS	of_workarounds
102 int of_workarounds;
103 #endif
104 
105 #define OF_WA_CLAIM	1	/* do phys/virt claim separately, then map */
106 #define OF_WA_LONGTRAIL	2	/* work around longtrail bugs */
107 
108 #define PROM_BUG() do {						\
109         prom_printf("kernel BUG at %s line 0x%x!\n",		\
110 		    RELOC(__FILE__), __LINE__);			\
111         __asm__ __volatile__(".long " BUG_ILLEGAL_INSTR);	\
112 } while (0)
113 
114 #ifdef DEBUG_PROM
115 #define prom_debug(x...)	prom_printf(x)
116 #else
117 #define prom_debug(x...)
118 #endif
119 
120 
121 typedef u32 prom_arg_t;
122 
123 struct prom_args {
124         u32 service;
125         u32 nargs;
126         u32 nret;
127         prom_arg_t args[10];
128 };
129 
130 struct prom_t {
131 	ihandle root;
132 	phandle chosen;
133 	int cpu;
134 	ihandle stdout;
135 	ihandle mmumap;
136 	ihandle memory;
137 };
138 
139 struct mem_map_entry {
140 	u64	base;
141 	u64	size;
142 };
143 
144 typedef u32 cell_t;
145 
146 extern void __start(unsigned long r3, unsigned long r4, unsigned long r5);
147 
148 #ifdef CONFIG_PPC64
149 extern int enter_prom(struct prom_args *args, unsigned long entry);
150 #else
151 static inline int enter_prom(struct prom_args *args, unsigned long entry)
152 {
153 	return ((int (*)(struct prom_args *))entry)(args);
154 }
155 #endif
156 
157 extern void copy_and_flush(unsigned long dest, unsigned long src,
158 			   unsigned long size, unsigned long offset);
159 
160 /* prom structure */
161 static struct prom_t __initdata prom;
162 
163 static unsigned long prom_entry __initdata;
164 
165 #define PROM_SCRATCH_SIZE 256
166 
167 static char __initdata of_stdout_device[256];
168 static char __initdata prom_scratch[PROM_SCRATCH_SIZE];
169 
170 static unsigned long __initdata dt_header_start;
171 static unsigned long __initdata dt_struct_start, dt_struct_end;
172 static unsigned long __initdata dt_string_start, dt_string_end;
173 
174 static unsigned long __initdata prom_initrd_start, prom_initrd_end;
175 
176 #ifdef CONFIG_PPC64
177 static int __initdata iommu_force_on;
178 static int __initdata ppc64_iommu_off;
179 static unsigned long __initdata prom_tce_alloc_start;
180 static unsigned long __initdata prom_tce_alloc_end;
181 #endif
182 
183 /* Platforms codes are now obsolete in the kernel. Now only used within this
184  * file and ultimately gone too. Feel free to change them if you need, they
185  * are not shared with anything outside of this file anymore
186  */
187 #define PLATFORM_PSERIES	0x0100
188 #define PLATFORM_PSERIES_LPAR	0x0101
189 #define PLATFORM_LPAR		0x0001
190 #define PLATFORM_POWERMAC	0x0400
191 #define PLATFORM_GENERIC	0x0500
192 
193 static int __initdata of_platform;
194 
195 static char __initdata prom_cmd_line[COMMAND_LINE_SIZE];
196 
197 static unsigned long __initdata alloc_top;
198 static unsigned long __initdata alloc_top_high;
199 static unsigned long __initdata alloc_bottom;
200 static unsigned long __initdata rmo_top;
201 static unsigned long __initdata ram_top;
202 
203 static struct mem_map_entry __initdata mem_reserve_map[MEM_RESERVE_MAP_SIZE];
204 static int __initdata mem_reserve_cnt;
205 
206 static cell_t __initdata regbuf[1024];
207 
208 
209 #define MAX_CPU_THREADS 2
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 
223 /* This is the one and *ONLY* place where we actually call open
224  * firmware.
225  */
226 
227 static int __init call_prom(const char *service, int nargs, int nret, ...)
228 {
229 	int i;
230 	struct prom_args args;
231 	va_list list;
232 
233 	args.service = ADDR(service);
234 	args.nargs = nargs;
235 	args.nret = nret;
236 
237 	va_start(list, nret);
238 	for (i = 0; i < nargs; i++)
239 		args.args[i] = va_arg(list, prom_arg_t);
240 	va_end(list);
241 
242 	for (i = 0; i < nret; i++)
243 		args.args[nargs+i] = 0;
244 
245 	if (enter_prom(&args, RELOC(prom_entry)) < 0)
246 		return PROM_ERROR;
247 
248 	return (nret > 0) ? args.args[nargs] : 0;
249 }
250 
251 static int __init call_prom_ret(const char *service, int nargs, int nret,
252 				prom_arg_t *rets, ...)
253 {
254 	int i;
255 	struct prom_args args;
256 	va_list list;
257 
258 	args.service = ADDR(service);
259 	args.nargs = nargs;
260 	args.nret = nret;
261 
262 	va_start(list, rets);
263 	for (i = 0; i < nargs; i++)
264 		args.args[i] = va_arg(list, prom_arg_t);
265 	va_end(list);
266 
267 	for (i = 0; i < nret; i++)
268 		args.args[nargs+i] = 0;
269 
270 	if (enter_prom(&args, RELOC(prom_entry)) < 0)
271 		return PROM_ERROR;
272 
273 	if (rets != NULL)
274 		for (i = 1; i < nret; ++i)
275 			rets[i-1] = args.args[nargs+i];
276 
277 	return (nret > 0) ? args.args[nargs] : 0;
278 }
279 
280 
281 static void __init prom_print(const char *msg)
282 {
283 	const char *p, *q;
284 	struct prom_t *_prom = &RELOC(prom);
285 
286 	if (_prom->stdout == 0)
287 		return;
288 
289 	for (p = msg; *p != 0; p = q) {
290 		for (q = p; *q != 0 && *q != '\n'; ++q)
291 			;
292 		if (q > p)
293 			call_prom("write", 3, 1, _prom->stdout, p, q - p);
294 		if (*q == 0)
295 			break;
296 		++q;
297 		call_prom("write", 3, 1, _prom->stdout, ADDR("\r\n"), 2);
298 	}
299 }
300 
301 
302 static void __init prom_print_hex(unsigned long val)
303 {
304 	int i, nibbles = sizeof(val)*2;
305 	char buf[sizeof(val)*2+1];
306 	struct prom_t *_prom = &RELOC(prom);
307 
308 	for (i = nibbles-1;  i >= 0;  i--) {
309 		buf[i] = (val & 0xf) + '0';
310 		if (buf[i] > '9')
311 			buf[i] += ('a'-'0'-10);
312 		val >>= 4;
313 	}
314 	buf[nibbles] = '\0';
315 	call_prom("write", 3, 1, _prom->stdout, buf, nibbles);
316 }
317 
318 
319 static void __init prom_printf(const char *format, ...)
320 {
321 	const char *p, *q, *s;
322 	va_list args;
323 	unsigned long v;
324 	struct prom_t *_prom = &RELOC(prom);
325 
326 	va_start(args, format);
327 #ifdef CONFIG_PPC64
328 	format = PTRRELOC(format);
329 #endif
330 	for (p = format; *p != 0; p = q) {
331 		for (q = p; *q != 0 && *q != '\n' && *q != '%'; ++q)
332 			;
333 		if (q > p)
334 			call_prom("write", 3, 1, _prom->stdout, p, q - p);
335 		if (*q == 0)
336 			break;
337 		if (*q == '\n') {
338 			++q;
339 			call_prom("write", 3, 1, _prom->stdout,
340 				  ADDR("\r\n"), 2);
341 			continue;
342 		}
343 		++q;
344 		if (*q == 0)
345 			break;
346 		switch (*q) {
347 		case 's':
348 			++q;
349 			s = va_arg(args, const char *);
350 			prom_print(s);
351 			break;
352 		case 'x':
353 			++q;
354 			v = va_arg(args, unsigned long);
355 			prom_print_hex(v);
356 			break;
357 		}
358 	}
359 }
360 
361 
362 static unsigned int __init prom_claim(unsigned long virt, unsigned long size,
363 				unsigned long align)
364 {
365 	struct prom_t *_prom = &RELOC(prom);
366 
367 	if (align == 0 && (OF_WORKAROUNDS & OF_WA_CLAIM)) {
368 		/*
369 		 * Old OF requires we claim physical and virtual separately
370 		 * and then map explicitly (assuming virtual mode)
371 		 */
372 		int ret;
373 		prom_arg_t result;
374 
375 		ret = call_prom_ret("call-method", 5, 2, &result,
376 				    ADDR("claim"), _prom->memory,
377 				    align, size, virt);
378 		if (ret != 0 || result == -1)
379 			return -1;
380 		ret = call_prom_ret("call-method", 5, 2, &result,
381 				    ADDR("claim"), _prom->mmumap,
382 				    align, size, virt);
383 		if (ret != 0) {
384 			call_prom("call-method", 4, 1, ADDR("release"),
385 				  _prom->memory, size, virt);
386 			return -1;
387 		}
388 		/* the 0x12 is M (coherence) + PP == read/write */
389 		call_prom("call-method", 6, 1,
390 			  ADDR("map"), _prom->mmumap, 0x12, size, virt, virt);
391 		return virt;
392 	}
393 	return call_prom("claim", 3, 1, (prom_arg_t)virt, (prom_arg_t)size,
394 			 (prom_arg_t)align);
395 }
396 
397 static void __init __attribute__((noreturn)) prom_panic(const char *reason)
398 {
399 #ifdef CONFIG_PPC64
400 	reason = PTRRELOC(reason);
401 #endif
402 	prom_print(reason);
403 	/* Do not call exit because it clears the screen on pmac
404 	 * it also causes some sort of double-fault on early pmacs */
405 	if (RELOC(of_platform) == PLATFORM_POWERMAC)
406 		asm("trap\n");
407 
408 	/* ToDo: should put up an SRC here on p/iSeries */
409 	call_prom("exit", 0, 0);
410 
411 	for (;;)			/* should never get here */
412 		;
413 }
414 
415 
416 static int __init prom_next_node(phandle *nodep)
417 {
418 	phandle node;
419 
420 	if ((node = *nodep) != 0
421 	    && (*nodep = call_prom("child", 1, 1, node)) != 0)
422 		return 1;
423 	if ((*nodep = call_prom("peer", 1, 1, node)) != 0)
424 		return 1;
425 	for (;;) {
426 		if ((node = call_prom("parent", 1, 1, node)) == 0)
427 			return 0;
428 		if ((*nodep = call_prom("peer", 1, 1, node)) != 0)
429 			return 1;
430 	}
431 }
432 
433 static int inline prom_getprop(phandle node, const char *pname,
434 			       void *value, size_t valuelen)
435 {
436 	return call_prom("getprop", 4, 1, node, ADDR(pname),
437 			 (u32)(unsigned long) value, (u32) valuelen);
438 }
439 
440 static int inline prom_getproplen(phandle node, const char *pname)
441 {
442 	return call_prom("getproplen", 2, 1, node, ADDR(pname));
443 }
444 
445 static void add_string(char **str, const char *q)
446 {
447 	char *p = *str;
448 
449 	while (*q)
450 		*p++ = *q++;
451 	*p++ = ' ';
452 	*str = p;
453 }
454 
455 static char *tohex(unsigned int x)
456 {
457 	static char digits[] = "0123456789abcdef";
458 	static char result[9];
459 	int i;
460 
461 	result[8] = 0;
462 	i = 8;
463 	do {
464 		--i;
465 		result[i] = digits[x & 0xf];
466 		x >>= 4;
467 	} while (x != 0 && i > 0);
468 	return &result[i];
469 }
470 
471 static int __init prom_setprop(phandle node, const char *nodename,
472 			       const char *pname, void *value, size_t valuelen)
473 {
474 	char cmd[256], *p;
475 
476 	if (!(OF_WORKAROUNDS & OF_WA_LONGTRAIL))
477 		return call_prom("setprop", 4, 1, node, ADDR(pname),
478 				 (u32)(unsigned long) value, (u32) valuelen);
479 
480 	/* gah... setprop doesn't work on longtrail, have to use interpret */
481 	p = cmd;
482 	add_string(&p, "dev");
483 	add_string(&p, nodename);
484 	add_string(&p, tohex((u32)(unsigned long) value));
485 	add_string(&p, tohex(valuelen));
486 	add_string(&p, tohex(ADDR(pname)));
487 	add_string(&p, tohex(strlen(RELOC(pname))));
488 	add_string(&p, "property");
489 	*p = 0;
490 	return call_prom("interpret", 1, 1, (u32)(unsigned long) cmd);
491 }
492 
493 /* We can't use the standard versions because of RELOC headaches. */
494 #define isxdigit(c)	(('0' <= (c) && (c) <= '9') \
495 			 || ('a' <= (c) && (c) <= 'f') \
496 			 || ('A' <= (c) && (c) <= 'F'))
497 
498 #define isdigit(c)	('0' <= (c) && (c) <= '9')
499 #define islower(c)	('a' <= (c) && (c) <= 'z')
500 #define toupper(c)	(islower(c) ? ((c) - 'a' + 'A') : (c))
501 
502 unsigned long prom_strtoul(const char *cp, const char **endp)
503 {
504 	unsigned long result = 0, base = 10, value;
505 
506 	if (*cp == '0') {
507 		base = 8;
508 		cp++;
509 		if (toupper(*cp) == 'X') {
510 			cp++;
511 			base = 16;
512 		}
513 	}
514 
515 	while (isxdigit(*cp) &&
516 	       (value = isdigit(*cp) ? *cp - '0' : toupper(*cp) - 'A' + 10) < base) {
517 		result = result * base + value;
518 		cp++;
519 	}
520 
521 	if (endp)
522 		*endp = cp;
523 
524 	return result;
525 }
526 
527 unsigned long prom_memparse(const char *ptr, const char **retptr)
528 {
529 	unsigned long ret = prom_strtoul(ptr, retptr);
530 	int shift = 0;
531 
532 	/*
533 	 * We can't use a switch here because GCC *may* generate a
534 	 * jump table which won't work, because we're not running at
535 	 * the address we're linked at.
536 	 */
537 	if ('G' == **retptr || 'g' == **retptr)
538 		shift = 30;
539 
540 	if ('M' == **retptr || 'm' == **retptr)
541 		shift = 20;
542 
543 	if ('K' == **retptr || 'k' == **retptr)
544 		shift = 10;
545 
546 	if (shift) {
547 		ret <<= shift;
548 		(*retptr)++;
549 	}
550 
551 	return ret;
552 }
553 
554 /*
555  * Early parsing of the command line passed to the kernel, used for
556  * "mem=x" and the options that affect the iommu
557  */
558 static void __init early_cmdline_parse(void)
559 {
560 	struct prom_t *_prom = &RELOC(prom);
561 	const char *opt;
562 	char *p;
563 	int l = 0;
564 
565 	RELOC(prom_cmd_line[0]) = 0;
566 	p = RELOC(prom_cmd_line);
567 	if ((long)_prom->chosen > 0)
568 		l = prom_getprop(_prom->chosen, "bootargs", p, COMMAND_LINE_SIZE-1);
569 #ifdef CONFIG_CMDLINE
570 	if (l <= 0 || p[0] == '\0') /* dbl check */
571 		strlcpy(RELOC(prom_cmd_line),
572 			RELOC(CONFIG_CMDLINE), sizeof(prom_cmd_line));
573 #endif /* CONFIG_CMDLINE */
574 	prom_printf("command line: %s\n", RELOC(prom_cmd_line));
575 
576 #ifdef CONFIG_PPC64
577 	opt = strstr(RELOC(prom_cmd_line), RELOC("iommu="));
578 	if (opt) {
579 		prom_printf("iommu opt is: %s\n", opt);
580 		opt += 6;
581 		while (*opt && *opt == ' ')
582 			opt++;
583 		if (!strncmp(opt, RELOC("off"), 3))
584 			RELOC(ppc64_iommu_off) = 1;
585 		else if (!strncmp(opt, RELOC("force"), 5))
586 			RELOC(iommu_force_on) = 1;
587 	}
588 #endif
589 }
590 
591 #ifdef CONFIG_PPC_PSERIES
592 /*
593  * There are two methods for telling firmware what our capabilities are.
594  * Newer machines have an "ibm,client-architecture-support" method on the
595  * root node.  For older machines, we have to call the "process-elf-header"
596  * method in the /packages/elf-loader node, passing it a fake 32-bit
597  * ELF header containing a couple of PT_NOTE sections that contain
598  * structures that contain various information.
599  */
600 
601 /*
602  * New method - extensible architecture description vector.
603  *
604  * Because the description vector contains a mix of byte and word
605  * values, we declare it as an unsigned char array, and use this
606  * macro to put word values in.
607  */
608 #define W(x)	((x) >> 24) & 0xff, ((x) >> 16) & 0xff, \
609 		((x) >> 8) & 0xff, (x) & 0xff
610 
611 /* Option vector bits - generic bits in byte 1 */
612 #define OV_IGNORE		0x80	/* ignore this vector */
613 #define OV_CESSATION_POLICY	0x40	/* halt if unsupported option present*/
614 
615 /* Option vector 1: processor architectures supported */
616 #define OV1_PPC_2_00		0x80	/* set if we support PowerPC 2.00 */
617 #define OV1_PPC_2_01		0x40	/* set if we support PowerPC 2.01 */
618 #define OV1_PPC_2_02		0x20	/* set if we support PowerPC 2.02 */
619 #define OV1_PPC_2_03		0x10	/* set if we support PowerPC 2.03 */
620 #define OV1_PPC_2_04		0x08	/* set if we support PowerPC 2.04 */
621 #define OV1_PPC_2_05		0x04	/* set if we support PowerPC 2.05 */
622 
623 /* Option vector 2: Open Firmware options supported */
624 #define OV2_REAL_MODE		0x20	/* set if we want OF in real mode */
625 
626 /* Option vector 3: processor options supported */
627 #define OV3_FP			0x80	/* floating point */
628 #define OV3_VMX			0x40	/* VMX/Altivec */
629 
630 /* Option vector 5: PAPR/OF options supported */
631 #define OV5_LPAR		0x80	/* logical partitioning supported */
632 #define OV5_SPLPAR		0x40	/* shared-processor LPAR supported */
633 /* ibm,dynamic-reconfiguration-memory property supported */
634 #define OV5_DRCONF_MEMORY	0x20
635 #define OV5_LARGE_PAGES		0x10	/* large pages supported */
636 
637 /*
638  * The architecture vector has an array of PVR mask/value pairs,
639  * followed by # option vectors - 1, followed by the option vectors.
640  */
641 static unsigned char ibm_architecture_vec[] = {
642 	W(0xfffe0000), W(0x003a0000),	/* POWER5/POWER5+ */
643 	W(0xffff0000), W(0x003e0000),	/* POWER6 */
644 	W(0xfffffffe), W(0x0f000001),	/* all 2.04-compliant and earlier */
645 	5 - 1,				/* 5 option vectors */
646 
647 	/* option vector 1: processor architectures supported */
648 	3 - 1,				/* length */
649 	0,				/* don't ignore, don't halt */
650 	OV1_PPC_2_00 | OV1_PPC_2_01 | OV1_PPC_2_02 | OV1_PPC_2_03 |
651 	OV1_PPC_2_04 | OV1_PPC_2_05,
652 
653 	/* option vector 2: Open Firmware options supported */
654 	34 - 1,				/* length */
655 	OV2_REAL_MODE,
656 	0, 0,
657 	W(0xffffffff),			/* real_base */
658 	W(0xffffffff),			/* real_size */
659 	W(0xffffffff),			/* virt_base */
660 	W(0xffffffff),			/* virt_size */
661 	W(0xffffffff),			/* load_base */
662 	W(64),				/* 128MB min RMA */
663 	W(0xffffffff),			/* full client load */
664 	0,				/* min RMA percentage of total RAM */
665 	48,				/* max log_2(hash table size) */
666 
667 	/* option vector 3: processor options supported */
668 	3 - 1,				/* length */
669 	0,				/* don't ignore, don't halt */
670 	OV3_FP | OV3_VMX,
671 
672 	/* option vector 4: IBM PAPR implementation */
673 	2 - 1,				/* length */
674 	0,				/* don't halt */
675 
676 	/* option vector 5: PAPR/OF options */
677 	3 - 1,				/* length */
678 	0,				/* don't ignore, don't halt */
679 	OV5_LPAR | OV5_SPLPAR | OV5_LARGE_PAGES,
680 };
681 
682 /* Old method - ELF header with PT_NOTE sections */
683 static struct fake_elf {
684 	Elf32_Ehdr	elfhdr;
685 	Elf32_Phdr	phdr[2];
686 	struct chrpnote {
687 		u32	namesz;
688 		u32	descsz;
689 		u32	type;
690 		char	name[8];	/* "PowerPC" */
691 		struct chrpdesc {
692 			u32	real_mode;
693 			u32	real_base;
694 			u32	real_size;
695 			u32	virt_base;
696 			u32	virt_size;
697 			u32	load_base;
698 		} chrpdesc;
699 	} chrpnote;
700 	struct rpanote {
701 		u32	namesz;
702 		u32	descsz;
703 		u32	type;
704 		char	name[24];	/* "IBM,RPA-Client-Config" */
705 		struct rpadesc {
706 			u32	lpar_affinity;
707 			u32	min_rmo_size;
708 			u32	min_rmo_percent;
709 			u32	max_pft_size;
710 			u32	splpar;
711 			u32	min_load;
712 			u32	new_mem_def;
713 			u32	ignore_me;
714 		} rpadesc;
715 	} rpanote;
716 } fake_elf = {
717 	.elfhdr = {
718 		.e_ident = { 0x7f, 'E', 'L', 'F',
719 			     ELFCLASS32, ELFDATA2MSB, EV_CURRENT },
720 		.e_type = ET_EXEC,	/* yeah right */
721 		.e_machine = EM_PPC,
722 		.e_version = EV_CURRENT,
723 		.e_phoff = offsetof(struct fake_elf, phdr),
724 		.e_phentsize = sizeof(Elf32_Phdr),
725 		.e_phnum = 2
726 	},
727 	.phdr = {
728 		[0] = {
729 			.p_type = PT_NOTE,
730 			.p_offset = offsetof(struct fake_elf, chrpnote),
731 			.p_filesz = sizeof(struct chrpnote)
732 		}, [1] = {
733 			.p_type = PT_NOTE,
734 			.p_offset = offsetof(struct fake_elf, rpanote),
735 			.p_filesz = sizeof(struct rpanote)
736 		}
737 	},
738 	.chrpnote = {
739 		.namesz = sizeof("PowerPC"),
740 		.descsz = sizeof(struct chrpdesc),
741 		.type = 0x1275,
742 		.name = "PowerPC",
743 		.chrpdesc = {
744 			.real_mode = ~0U,	/* ~0 means "don't care" */
745 			.real_base = ~0U,
746 			.real_size = ~0U,
747 			.virt_base = ~0U,
748 			.virt_size = ~0U,
749 			.load_base = ~0U
750 		},
751 	},
752 	.rpanote = {
753 		.namesz = sizeof("IBM,RPA-Client-Config"),
754 		.descsz = sizeof(struct rpadesc),
755 		.type = 0x12759999,
756 		.name = "IBM,RPA-Client-Config",
757 		.rpadesc = {
758 			.lpar_affinity = 0,
759 			.min_rmo_size = 64,	/* in megabytes */
760 			.min_rmo_percent = 0,
761 			.max_pft_size = 48,	/* 2^48 bytes max PFT size */
762 			.splpar = 1,
763 			.min_load = ~0U,
764 			.new_mem_def = 0
765 		}
766 	}
767 };
768 
769 static void __init prom_send_capabilities(void)
770 {
771 	ihandle elfloader, root;
772 	prom_arg_t ret;
773 
774 	root = call_prom("open", 1, 1, ADDR("/"));
775 	if (root != 0) {
776 		/* try calling the ibm,client-architecture-support method */
777 		if (call_prom_ret("call-method", 3, 2, &ret,
778 				  ADDR("ibm,client-architecture-support"),
779 				  root,
780 				  ADDR(ibm_architecture_vec)) == 0) {
781 			/* the call exists... */
782 			if (ret)
783 				prom_printf("WARNING: ibm,client-architecture"
784 					    "-support call FAILED!\n");
785 			call_prom("close", 1, 0, root);
786 			return;
787 		}
788 		call_prom("close", 1, 0, root);
789 	}
790 
791 	/* no ibm,client-architecture-support call, try the old way */
792 	elfloader = call_prom("open", 1, 1, ADDR("/packages/elf-loader"));
793 	if (elfloader == 0) {
794 		prom_printf("couldn't open /packages/elf-loader\n");
795 		return;
796 	}
797 	call_prom("call-method", 3, 1, ADDR("process-elf-header"),
798 			elfloader, ADDR(&fake_elf));
799 	call_prom("close", 1, 0, elfloader);
800 }
801 #endif
802 
803 /*
804  * Memory allocation strategy... our layout is normally:
805  *
806  *  at 14Mb or more we have vmlinux, then a gap and initrd.  In some
807  *  rare cases, initrd might end up being before the kernel though.
808  *  We assume this won't override the final kernel at 0, we have no
809  *  provision to handle that in this version, but it should hopefully
810  *  never happen.
811  *
812  *  alloc_top is set to the top of RMO, eventually shrink down if the
813  *  TCEs overlap
814  *
815  *  alloc_bottom is set to the top of kernel/initrd
816  *
817  *  from there, allocations are done this way : rtas is allocated
818  *  topmost, and the device-tree is allocated from the bottom. We try
819  *  to grow the device-tree allocation as we progress. If we can't,
820  *  then we fail, we don't currently have a facility to restart
821  *  elsewhere, but that shouldn't be necessary.
822  *
823  *  Note that calls to reserve_mem have to be done explicitly, memory
824  *  allocated with either alloc_up or alloc_down isn't automatically
825  *  reserved.
826  */
827 
828 
829 /*
830  * Allocates memory in the RMO upward from the kernel/initrd
831  *
832  * When align is 0, this is a special case, it means to allocate in place
833  * at the current location of alloc_bottom or fail (that is basically
834  * extending the previous allocation). Used for the device-tree flattening
835  */
836 static unsigned long __init alloc_up(unsigned long size, unsigned long align)
837 {
838 	unsigned long base = RELOC(alloc_bottom);
839 	unsigned long addr = 0;
840 
841 	if (align)
842 		base = _ALIGN_UP(base, align);
843 	prom_debug("alloc_up(%x, %x)\n", size, align);
844 	if (RELOC(ram_top) == 0)
845 		prom_panic("alloc_up() called with mem not initialized\n");
846 
847 	if (align)
848 		base = _ALIGN_UP(RELOC(alloc_bottom), align);
849 	else
850 		base = RELOC(alloc_bottom);
851 
852 	for(; (base + size) <= RELOC(alloc_top);
853 	    base = _ALIGN_UP(base + 0x100000, align)) {
854 		prom_debug("    trying: 0x%x\n\r", base);
855 		addr = (unsigned long)prom_claim(base, size, 0);
856 		if (addr != PROM_ERROR && addr != 0)
857 			break;
858 		addr = 0;
859 		if (align == 0)
860 			break;
861 	}
862 	if (addr == 0)
863 		return 0;
864 	RELOC(alloc_bottom) = addr;
865 
866 	prom_debug(" -> %x\n", addr);
867 	prom_debug("  alloc_bottom : %x\n", RELOC(alloc_bottom));
868 	prom_debug("  alloc_top    : %x\n", RELOC(alloc_top));
869 	prom_debug("  alloc_top_hi : %x\n", RELOC(alloc_top_high));
870 	prom_debug("  rmo_top      : %x\n", RELOC(rmo_top));
871 	prom_debug("  ram_top      : %x\n", RELOC(ram_top));
872 
873 	return addr;
874 }
875 
876 /*
877  * Allocates memory downward, either from top of RMO, or if highmem
878  * is set, from the top of RAM.  Note that this one doesn't handle
879  * failures.  It does claim memory if highmem is not set.
880  */
881 static unsigned long __init alloc_down(unsigned long size, unsigned long align,
882 				       int highmem)
883 {
884 	unsigned long base, addr = 0;
885 
886 	prom_debug("alloc_down(%x, %x, %s)\n", size, align,
887 		   highmem ? RELOC("(high)") : RELOC("(low)"));
888 	if (RELOC(ram_top) == 0)
889 		prom_panic("alloc_down() called with mem not initialized\n");
890 
891 	if (highmem) {
892 		/* Carve out storage for the TCE table. */
893 		addr = _ALIGN_DOWN(RELOC(alloc_top_high) - size, align);
894 		if (addr <= RELOC(alloc_bottom))
895 			return 0;
896 		/* Will we bump into the RMO ? If yes, check out that we
897 		 * didn't overlap existing allocations there, if we did,
898 		 * we are dead, we must be the first in town !
899 		 */
900 		if (addr < RELOC(rmo_top)) {
901 			/* Good, we are first */
902 			if (RELOC(alloc_top) == RELOC(rmo_top))
903 				RELOC(alloc_top) = RELOC(rmo_top) = addr;
904 			else
905 				return 0;
906 		}
907 		RELOC(alloc_top_high) = addr;
908 		goto bail;
909 	}
910 
911 	base = _ALIGN_DOWN(RELOC(alloc_top) - size, align);
912 	for (; base > RELOC(alloc_bottom);
913 	     base = _ALIGN_DOWN(base - 0x100000, align))  {
914 		prom_debug("    trying: 0x%x\n\r", base);
915 		addr = (unsigned long)prom_claim(base, size, 0);
916 		if (addr != PROM_ERROR && addr != 0)
917 			break;
918 		addr = 0;
919 	}
920 	if (addr == 0)
921 		return 0;
922 	RELOC(alloc_top) = addr;
923 
924  bail:
925 	prom_debug(" -> %x\n", addr);
926 	prom_debug("  alloc_bottom : %x\n", RELOC(alloc_bottom));
927 	prom_debug("  alloc_top    : %x\n", RELOC(alloc_top));
928 	prom_debug("  alloc_top_hi : %x\n", RELOC(alloc_top_high));
929 	prom_debug("  rmo_top      : %x\n", RELOC(rmo_top));
930 	prom_debug("  ram_top      : %x\n", RELOC(ram_top));
931 
932 	return addr;
933 }
934 
935 /*
936  * Parse a "reg" cell
937  */
938 static unsigned long __init prom_next_cell(int s, cell_t **cellp)
939 {
940 	cell_t *p = *cellp;
941 	unsigned long r = 0;
942 
943 	/* Ignore more than 2 cells */
944 	while (s > sizeof(unsigned long) / 4) {
945 		p++;
946 		s--;
947 	}
948 	r = *p++;
949 #ifdef CONFIG_PPC64
950 	if (s > 1) {
951 		r <<= 32;
952 		r |= *(p++);
953 	}
954 #endif
955 	*cellp = p;
956 	return r;
957 }
958 
959 /*
960  * Very dumb function for adding to the memory reserve list, but
961  * we don't need anything smarter at this point
962  *
963  * XXX Eventually check for collisions.  They should NEVER happen.
964  * If problems seem to show up, it would be a good start to track
965  * them down.
966  */
967 static void reserve_mem(u64 base, u64 size)
968 {
969 	u64 top = base + size;
970 	unsigned long cnt = RELOC(mem_reserve_cnt);
971 
972 	if (size == 0)
973 		return;
974 
975 	/* We need to always keep one empty entry so that we
976 	 * have our terminator with "size" set to 0 since we are
977 	 * dumb and just copy this entire array to the boot params
978 	 */
979 	base = _ALIGN_DOWN(base, PAGE_SIZE);
980 	top = _ALIGN_UP(top, PAGE_SIZE);
981 	size = top - base;
982 
983 	if (cnt >= (MEM_RESERVE_MAP_SIZE - 1))
984 		prom_panic("Memory reserve map exhausted !\n");
985 	RELOC(mem_reserve_map)[cnt].base = base;
986 	RELOC(mem_reserve_map)[cnt].size = size;
987 	RELOC(mem_reserve_cnt) = cnt + 1;
988 }
989 
990 /*
991  * Initialize memory allocation mechanism, parse "memory" nodes and
992  * obtain that way the top of memory and RMO to setup out local allocator
993  */
994 static void __init prom_init_mem(void)
995 {
996 	phandle node;
997 	char *path, type[64];
998 	unsigned int plen;
999 	cell_t *p, *endp;
1000 	struct prom_t *_prom = &RELOC(prom);
1001 	u32 rac, rsc;
1002 
1003 	/*
1004 	 * We iterate the memory nodes to find
1005 	 * 1) top of RMO (first node)
1006 	 * 2) top of memory
1007 	 */
1008 	rac = 2;
1009 	prom_getprop(_prom->root, "#address-cells", &rac, sizeof(rac));
1010 	rsc = 1;
1011 	prom_getprop(_prom->root, "#size-cells", &rsc, sizeof(rsc));
1012 	prom_debug("root_addr_cells: %x\n", (unsigned long) rac);
1013 	prom_debug("root_size_cells: %x\n", (unsigned long) rsc);
1014 
1015 	prom_debug("scanning memory:\n");
1016 	path = RELOC(prom_scratch);
1017 
1018 	for (node = 0; prom_next_node(&node); ) {
1019 		type[0] = 0;
1020 		prom_getprop(node, "device_type", type, sizeof(type));
1021 
1022 		if (type[0] == 0) {
1023 			/*
1024 			 * CHRP Longtrail machines have no device_type
1025 			 * on the memory node, so check the name instead...
1026 			 */
1027 			prom_getprop(node, "name", type, sizeof(type));
1028 		}
1029 		if (strcmp(type, RELOC("memory")))
1030 			continue;
1031 
1032 		plen = prom_getprop(node, "reg", RELOC(regbuf), sizeof(regbuf));
1033 		if (plen > sizeof(regbuf)) {
1034 			prom_printf("memory node too large for buffer !\n");
1035 			plen = sizeof(regbuf);
1036 		}
1037 		p = RELOC(regbuf);
1038 		endp = p + (plen / sizeof(cell_t));
1039 
1040 #ifdef DEBUG_PROM
1041 		memset(path, 0, PROM_SCRATCH_SIZE);
1042 		call_prom("package-to-path", 3, 1, node, path, PROM_SCRATCH_SIZE-1);
1043 		prom_debug("  node %s :\n", path);
1044 #endif /* DEBUG_PROM */
1045 
1046 		while ((endp - p) >= (rac + rsc)) {
1047 			unsigned long base, size;
1048 
1049 			base = prom_next_cell(rac, &p);
1050 			size = prom_next_cell(rsc, &p);
1051 
1052 			if (size == 0)
1053 				continue;
1054 			prom_debug("    %x %x\n", base, size);
1055 			if (base == 0 && (RELOC(of_platform) & PLATFORM_LPAR))
1056 				RELOC(rmo_top) = size;
1057 			if ((base + size) > RELOC(ram_top))
1058 				RELOC(ram_top) = base + size;
1059 		}
1060 	}
1061 
1062 	RELOC(alloc_bottom) = PAGE_ALIGN((unsigned long)&RELOC(_end) + 0x4000);
1063 
1064 	/* Check if we have an initrd after the kernel, if we do move our bottom
1065 	 * point to after it
1066 	 */
1067 	if (RELOC(prom_initrd_start)) {
1068 		if (RELOC(prom_initrd_end) > RELOC(alloc_bottom))
1069 			RELOC(alloc_bottom) = PAGE_ALIGN(RELOC(prom_initrd_end));
1070 	}
1071 
1072 	/*
1073 	 * Setup our top alloc point, that is top of RMO or top of
1074 	 * segment 0 when running non-LPAR.
1075 	 * Some RS64 machines have buggy firmware where claims up at
1076 	 * 1GB fail.  Cap at 768MB as a workaround.
1077 	 * Since 768MB is plenty of room, and we need to cap to something
1078 	 * reasonable on 32-bit, cap at 768MB on all machines.
1079 	 */
1080 	if (!RELOC(rmo_top))
1081 		RELOC(rmo_top) = RELOC(ram_top);
1082 	RELOC(rmo_top) = min(0x30000000ul, RELOC(rmo_top));
1083 	RELOC(alloc_top) = RELOC(rmo_top);
1084 	RELOC(alloc_top_high) = RELOC(ram_top);
1085 
1086 	prom_printf("memory layout at init:\n");
1087 	prom_printf("  alloc_bottom : %x\n", RELOC(alloc_bottom));
1088 	prom_printf("  alloc_top    : %x\n", RELOC(alloc_top));
1089 	prom_printf("  alloc_top_hi : %x\n", RELOC(alloc_top_high));
1090 	prom_printf("  rmo_top      : %x\n", RELOC(rmo_top));
1091 	prom_printf("  ram_top      : %x\n", RELOC(ram_top));
1092 }
1093 
1094 
1095 /*
1096  * Allocate room for and instantiate RTAS
1097  */
1098 static void __init prom_instantiate_rtas(void)
1099 {
1100 	phandle rtas_node;
1101 	ihandle rtas_inst;
1102 	u32 base, entry = 0;
1103 	u32 size = 0;
1104 
1105 	prom_debug("prom_instantiate_rtas: start...\n");
1106 
1107 	rtas_node = call_prom("finddevice", 1, 1, ADDR("/rtas"));
1108 	prom_debug("rtas_node: %x\n", rtas_node);
1109 	if (!PHANDLE_VALID(rtas_node))
1110 		return;
1111 
1112 	prom_getprop(rtas_node, "rtas-size", &size, sizeof(size));
1113 	if (size == 0)
1114 		return;
1115 
1116 	base = alloc_down(size, PAGE_SIZE, 0);
1117 	if (base == 0) {
1118 		prom_printf("RTAS allocation failed !\n");
1119 		return;
1120 	}
1121 
1122 	rtas_inst = call_prom("open", 1, 1, ADDR("/rtas"));
1123 	if (!IHANDLE_VALID(rtas_inst)) {
1124 		prom_printf("opening rtas package failed (%x)\n", rtas_inst);
1125 		return;
1126 	}
1127 
1128 	prom_printf("instantiating rtas at 0x%x ...", base);
1129 
1130 	if (call_prom_ret("call-method", 3, 2, &entry,
1131 			  ADDR("instantiate-rtas"),
1132 			  rtas_inst, base) != 0
1133 	    || entry == 0) {
1134 		prom_printf(" failed\n");
1135 		return;
1136 	}
1137 	prom_printf(" done\n");
1138 
1139 	reserve_mem(base, size);
1140 
1141 	prom_setprop(rtas_node, "/rtas", "linux,rtas-base",
1142 		     &base, sizeof(base));
1143 	prom_setprop(rtas_node, "/rtas", "linux,rtas-entry",
1144 		     &entry, sizeof(entry));
1145 
1146 	prom_debug("rtas base     = 0x%x\n", base);
1147 	prom_debug("rtas entry    = 0x%x\n", entry);
1148 	prom_debug("rtas size     = 0x%x\n", (long)size);
1149 
1150 	prom_debug("prom_instantiate_rtas: end...\n");
1151 }
1152 
1153 #ifdef CONFIG_PPC64
1154 /*
1155  * Allocate room for and initialize TCE tables
1156  */
1157 static void __init prom_initialize_tce_table(void)
1158 {
1159 	phandle node;
1160 	ihandle phb_node;
1161 	char compatible[64], type[64], model[64];
1162 	char *path = RELOC(prom_scratch);
1163 	u64 base, align;
1164 	u32 minalign, minsize;
1165 	u64 tce_entry, *tce_entryp;
1166 	u64 local_alloc_top, local_alloc_bottom;
1167 	u64 i;
1168 
1169 	if (RELOC(ppc64_iommu_off))
1170 		return;
1171 
1172 	prom_debug("starting prom_initialize_tce_table\n");
1173 
1174 	/* Cache current top of allocs so we reserve a single block */
1175 	local_alloc_top = RELOC(alloc_top_high);
1176 	local_alloc_bottom = local_alloc_top;
1177 
1178 	/* Search all nodes looking for PHBs. */
1179 	for (node = 0; prom_next_node(&node); ) {
1180 		compatible[0] = 0;
1181 		type[0] = 0;
1182 		model[0] = 0;
1183 		prom_getprop(node, "compatible",
1184 			     compatible, sizeof(compatible));
1185 		prom_getprop(node, "device_type", type, sizeof(type));
1186 		prom_getprop(node, "model", model, sizeof(model));
1187 
1188 		if ((type[0] == 0) || (strstr(type, RELOC("pci")) == NULL))
1189 			continue;
1190 
1191 		/* Keep the old logic in tack to avoid regression. */
1192 		if (compatible[0] != 0) {
1193 			if ((strstr(compatible, RELOC("python")) == NULL) &&
1194 			    (strstr(compatible, RELOC("Speedwagon")) == NULL) &&
1195 			    (strstr(compatible, RELOC("Winnipeg")) == NULL))
1196 				continue;
1197 		} else if (model[0] != 0) {
1198 			if ((strstr(model, RELOC("ython")) == NULL) &&
1199 			    (strstr(model, RELOC("peedwagon")) == NULL) &&
1200 			    (strstr(model, RELOC("innipeg")) == NULL))
1201 				continue;
1202 		}
1203 
1204 		if (prom_getprop(node, "tce-table-minalign", &minalign,
1205 				 sizeof(minalign)) == PROM_ERROR)
1206 			minalign = 0;
1207 		if (prom_getprop(node, "tce-table-minsize", &minsize,
1208 				 sizeof(minsize)) == PROM_ERROR)
1209 			minsize = 4UL << 20;
1210 
1211 		/*
1212 		 * Even though we read what OF wants, we just set the table
1213 		 * size to 4 MB.  This is enough to map 2GB of PCI DMA space.
1214 		 * By doing this, we avoid the pitfalls of trying to DMA to
1215 		 * MMIO space and the DMA alias hole.
1216 		 *
1217 		 * On POWER4, firmware sets the TCE region by assuming
1218 		 * each TCE table is 8MB. Using this memory for anything
1219 		 * else will impact performance, so we always allocate 8MB.
1220 		 * Anton
1221 		 */
1222 		if (__is_processor(PV_POWER4) || __is_processor(PV_POWER4p))
1223 			minsize = 8UL << 20;
1224 		else
1225 			minsize = 4UL << 20;
1226 
1227 		/* Align to the greater of the align or size */
1228 		align = max(minalign, minsize);
1229 		base = alloc_down(minsize, align, 1);
1230 		if (base == 0)
1231 			prom_panic("ERROR, cannot find space for TCE table.\n");
1232 		if (base < local_alloc_bottom)
1233 			local_alloc_bottom = base;
1234 
1235 		/* It seems OF doesn't null-terminate the path :-( */
1236 		memset(path, 0, sizeof(path));
1237 		/* Call OF to setup the TCE hardware */
1238 		if (call_prom("package-to-path", 3, 1, node,
1239 			      path, PROM_SCRATCH_SIZE-1) == PROM_ERROR) {
1240 			prom_printf("package-to-path failed\n");
1241 		}
1242 
1243 		/* Save away the TCE table attributes for later use. */
1244 		prom_setprop(node, path, "linux,tce-base", &base, sizeof(base));
1245 		prom_setprop(node, path, "linux,tce-size", &minsize, sizeof(minsize));
1246 
1247 		prom_debug("TCE table: %s\n", path);
1248 		prom_debug("\tnode = 0x%x\n", node);
1249 		prom_debug("\tbase = 0x%x\n", base);
1250 		prom_debug("\tsize = 0x%x\n", minsize);
1251 
1252 		/* Initialize the table to have a one-to-one mapping
1253 		 * over the allocated size.
1254 		 */
1255 		tce_entryp = (unsigned long *)base;
1256 		for (i = 0; i < (minsize >> 3) ;tce_entryp++, i++) {
1257 			tce_entry = (i << PAGE_SHIFT);
1258 			tce_entry |= 0x3;
1259 			*tce_entryp = tce_entry;
1260 		}
1261 
1262 		prom_printf("opening PHB %s", path);
1263 		phb_node = call_prom("open", 1, 1, path);
1264 		if (phb_node == 0)
1265 			prom_printf("... failed\n");
1266 		else
1267 			prom_printf("... done\n");
1268 
1269 		call_prom("call-method", 6, 0, ADDR("set-64-bit-addressing"),
1270 			  phb_node, -1, minsize,
1271 			  (u32) base, (u32) (base >> 32));
1272 		call_prom("close", 1, 0, phb_node);
1273 	}
1274 
1275 	reserve_mem(local_alloc_bottom, local_alloc_top - local_alloc_bottom);
1276 
1277 	/* These are only really needed if there is a memory limit in
1278 	 * effect, but we don't know so export them always. */
1279 	RELOC(prom_tce_alloc_start) = local_alloc_bottom;
1280 	RELOC(prom_tce_alloc_end) = local_alloc_top;
1281 
1282 	/* Flag the first invalid entry */
1283 	prom_debug("ending prom_initialize_tce_table\n");
1284 }
1285 #endif
1286 
1287 /*
1288  * With CHRP SMP we need to use the OF to start the other processors.
1289  * We can't wait until smp_boot_cpus (the OF is trashed by then)
1290  * so we have to put the processors into a holding pattern controlled
1291  * by the kernel (not OF) before we destroy the OF.
1292  *
1293  * This uses a chunk of low memory, puts some holding pattern
1294  * code there and sends the other processors off to there until
1295  * smp_boot_cpus tells them to do something.  The holding pattern
1296  * checks that address until its cpu # is there, when it is that
1297  * cpu jumps to __secondary_start().  smp_boot_cpus() takes care
1298  * of setting those values.
1299  *
1300  * We also use physical address 0x4 here to tell when a cpu
1301  * is in its holding pattern code.
1302  *
1303  * -- Cort
1304  */
1305 extern void __secondary_hold(void);
1306 extern unsigned long __secondary_hold_spinloop;
1307 extern unsigned long __secondary_hold_acknowledge;
1308 
1309 /*
1310  * We want to reference the copy of __secondary_hold_* in the
1311  * 0 - 0x100 address range
1312  */
1313 #define LOW_ADDR(x)	(((unsigned long) &(x)) & 0xff)
1314 
1315 static void __init prom_hold_cpus(void)
1316 {
1317 	unsigned long i;
1318 	unsigned int reg;
1319 	phandle node;
1320 	char type[64];
1321 	int cpuid = 0;
1322 	unsigned int interrupt_server[MAX_CPU_THREADS];
1323 	unsigned int cpu_threads, hw_cpu_num;
1324 	int propsize;
1325 	struct prom_t *_prom = &RELOC(prom);
1326 	unsigned long *spinloop
1327 		= (void *) LOW_ADDR(__secondary_hold_spinloop);
1328 	unsigned long *acknowledge
1329 		= (void *) LOW_ADDR(__secondary_hold_acknowledge);
1330 #ifdef CONFIG_PPC64
1331 	/* __secondary_hold is actually a descriptor, not the text address */
1332 	unsigned long secondary_hold
1333 		= __pa(*PTRRELOC((unsigned long *)__secondary_hold));
1334 #else
1335 	unsigned long secondary_hold = LOW_ADDR(__secondary_hold);
1336 #endif
1337 
1338 	prom_debug("prom_hold_cpus: start...\n");
1339 	prom_debug("    1) spinloop       = 0x%x\n", (unsigned long)spinloop);
1340 	prom_debug("    1) *spinloop      = 0x%x\n", *spinloop);
1341 	prom_debug("    1) acknowledge    = 0x%x\n",
1342 		   (unsigned long)acknowledge);
1343 	prom_debug("    1) *acknowledge   = 0x%x\n", *acknowledge);
1344 	prom_debug("    1) secondary_hold = 0x%x\n", secondary_hold);
1345 
1346 	/* Set the common spinloop variable, so all of the secondary cpus
1347 	 * will block when they are awakened from their OF spinloop.
1348 	 * This must occur for both SMP and non SMP kernels, since OF will
1349 	 * be trashed when we move the kernel.
1350 	 */
1351 	*spinloop = 0;
1352 
1353 	/* look for cpus */
1354 	for (node = 0; prom_next_node(&node); ) {
1355 		type[0] = 0;
1356 		prom_getprop(node, "device_type", type, sizeof(type));
1357 		if (strcmp(type, RELOC("cpu")) != 0)
1358 			continue;
1359 
1360 		/* Skip non-configured cpus. */
1361 		if (prom_getprop(node, "status", type, sizeof(type)) > 0)
1362 			if (strcmp(type, RELOC("okay")) != 0)
1363 				continue;
1364 
1365 		reg = -1;
1366 		prom_getprop(node, "reg", &reg, sizeof(reg));
1367 
1368 		prom_debug("\ncpuid        = 0x%x\n", cpuid);
1369 		prom_debug("cpu hw idx   = 0x%x\n", reg);
1370 
1371 		/* Init the acknowledge var which will be reset by
1372 		 * the secondary cpu when it awakens from its OF
1373 		 * spinloop.
1374 		 */
1375 		*acknowledge = (unsigned long)-1;
1376 
1377 		propsize = prom_getprop(node, "ibm,ppc-interrupt-server#s",
1378 					&interrupt_server,
1379 					sizeof(interrupt_server));
1380 		if (propsize < 0) {
1381 			/* no property.  old hardware has no SMT */
1382 			cpu_threads = 1;
1383 			interrupt_server[0] = reg; /* fake it with phys id */
1384 		} else {
1385 			/* We have a threaded processor */
1386 			cpu_threads = propsize / sizeof(u32);
1387 			if (cpu_threads > MAX_CPU_THREADS) {
1388 				prom_printf("SMT: too many threads!\n"
1389 					    "SMT: found %x, max is %x\n",
1390 					    cpu_threads, MAX_CPU_THREADS);
1391 				cpu_threads = 1; /* ToDo: panic? */
1392 			}
1393 		}
1394 
1395 		hw_cpu_num = interrupt_server[0];
1396 		if (hw_cpu_num != _prom->cpu) {
1397 			/* Primary Thread of non-boot cpu */
1398 			prom_printf("%x : starting cpu hw idx %x... ", cpuid, reg);
1399 			call_prom("start-cpu", 3, 0, node,
1400 				  secondary_hold, reg);
1401 
1402 			for (i = 0; (i < 100000000) &&
1403 			     (*acknowledge == ((unsigned long)-1)); i++ )
1404 				mb();
1405 
1406 			if (*acknowledge == reg)
1407 				prom_printf("done\n");
1408 			else
1409 				prom_printf("failed: %x\n", *acknowledge);
1410 		}
1411 #ifdef CONFIG_SMP
1412 		else
1413 			prom_printf("%x : boot cpu     %x\n", cpuid, reg);
1414 #endif /* CONFIG_SMP */
1415 
1416 		/* Reserve cpu #s for secondary threads.   They start later. */
1417 		cpuid += cpu_threads;
1418 	}
1419 
1420 	if (cpuid > NR_CPUS)
1421 		prom_printf("WARNING: maximum CPUs (" __stringify(NR_CPUS)
1422 			    ") exceeded: ignoring extras\n");
1423 
1424 	prom_debug("prom_hold_cpus: end...\n");
1425 }
1426 
1427 
1428 static void __init prom_init_client_services(unsigned long pp)
1429 {
1430 	struct prom_t *_prom = &RELOC(prom);
1431 
1432 	/* Get a handle to the prom entry point before anything else */
1433 	RELOC(prom_entry) = pp;
1434 
1435 	/* get a handle for the stdout device */
1436 	_prom->chosen = call_prom("finddevice", 1, 1, ADDR("/chosen"));
1437 	if (!PHANDLE_VALID(_prom->chosen))
1438 		prom_panic("cannot find chosen"); /* msg won't be printed :( */
1439 
1440 	/* get device tree root */
1441 	_prom->root = call_prom("finddevice", 1, 1, ADDR("/"));
1442 	if (!PHANDLE_VALID(_prom->root))
1443 		prom_panic("cannot find device tree root"); /* msg won't be printed :( */
1444 
1445 	_prom->mmumap = 0;
1446 }
1447 
1448 #ifdef CONFIG_PPC32
1449 /*
1450  * For really old powermacs, we need to map things we claim.
1451  * For that, we need the ihandle of the mmu.
1452  * Also, on the longtrail, we need to work around other bugs.
1453  */
1454 static void __init prom_find_mmu(void)
1455 {
1456 	struct prom_t *_prom = &RELOC(prom);
1457 	phandle oprom;
1458 	char version[64];
1459 
1460 	oprom = call_prom("finddevice", 1, 1, ADDR("/openprom"));
1461 	if (!PHANDLE_VALID(oprom))
1462 		return;
1463 	if (prom_getprop(oprom, "model", version, sizeof(version)) <= 0)
1464 		return;
1465 	version[sizeof(version) - 1] = 0;
1466 	/* XXX might need to add other versions here */
1467 	if (strcmp(version, "Open Firmware, 1.0.5") == 0)
1468 		of_workarounds = OF_WA_CLAIM;
1469 	else if (strncmp(version, "FirmWorks,3.", 12) == 0) {
1470 		of_workarounds = OF_WA_CLAIM | OF_WA_LONGTRAIL;
1471 		call_prom("interpret", 1, 1, "dev /memory 0 to allow-reclaim");
1472 	} else
1473 		return;
1474 	_prom->memory = call_prom("open", 1, 1, ADDR("/memory"));
1475 	prom_getprop(_prom->chosen, "mmu", &_prom->mmumap,
1476 		     sizeof(_prom->mmumap));
1477 	if (!IHANDLE_VALID(_prom->memory) || !IHANDLE_VALID(_prom->mmumap))
1478 		of_workarounds &= ~OF_WA_CLAIM;		/* hmmm */
1479 }
1480 #else
1481 #define prom_find_mmu()
1482 #endif
1483 
1484 static void __init prom_init_stdout(void)
1485 {
1486 	struct prom_t *_prom = &RELOC(prom);
1487 	char *path = RELOC(of_stdout_device);
1488 	char type[16];
1489 	u32 val;
1490 
1491 	if (prom_getprop(_prom->chosen, "stdout", &val, sizeof(val)) <= 0)
1492 		prom_panic("cannot find stdout");
1493 
1494 	_prom->stdout = val;
1495 
1496 	/* Get the full OF pathname of the stdout device */
1497 	memset(path, 0, 256);
1498 	call_prom("instance-to-path", 3, 1, _prom->stdout, path, 255);
1499 	val = call_prom("instance-to-package", 1, 1, _prom->stdout);
1500 	prom_setprop(_prom->chosen, "/chosen", "linux,stdout-package",
1501 		     &val, sizeof(val));
1502 	prom_printf("OF stdout device is: %s\n", RELOC(of_stdout_device));
1503 	prom_setprop(_prom->chosen, "/chosen", "linux,stdout-path",
1504 		     path, strlen(path) + 1);
1505 
1506 	/* If it's a display, note it */
1507 	memset(type, 0, sizeof(type));
1508 	prom_getprop(val, "device_type", type, sizeof(type));
1509 	if (strcmp(type, RELOC("display")) == 0)
1510 		prom_setprop(val, path, "linux,boot-display", NULL, 0);
1511 }
1512 
1513 static void __init prom_close_stdin(void)
1514 {
1515 	struct prom_t *_prom = &RELOC(prom);
1516 	ihandle val;
1517 
1518 	if (prom_getprop(_prom->chosen, "stdin", &val, sizeof(val)) > 0)
1519 		call_prom("close", 1, 0, val);
1520 }
1521 
1522 static int __init prom_find_machine_type(void)
1523 {
1524 	struct prom_t *_prom = &RELOC(prom);
1525 	char compat[256];
1526 	int len, i = 0;
1527 #ifdef CONFIG_PPC64
1528 	phandle rtas;
1529 	int x;
1530 #endif
1531 
1532 	/* Look for a PowerMac */
1533 	len = prom_getprop(_prom->root, "compatible",
1534 			   compat, sizeof(compat)-1);
1535 	if (len > 0) {
1536 		compat[len] = 0;
1537 		while (i < len) {
1538 			char *p = &compat[i];
1539 			int sl = strlen(p);
1540 			if (sl == 0)
1541 				break;
1542 			if (strstr(p, RELOC("Power Macintosh")) ||
1543 			    strstr(p, RELOC("MacRISC")))
1544 				return PLATFORM_POWERMAC;
1545 #ifdef CONFIG_PPC64
1546 			/* We must make sure we don't detect the IBM Cell
1547 			 * blades as pSeries due to some firmware issues,
1548 			 * so we do it here.
1549 			 */
1550 			if (strstr(p, RELOC("IBM,CBEA")) ||
1551 			    strstr(p, RELOC("IBM,CPBW-1.0")))
1552 				return PLATFORM_GENERIC;
1553 #endif /* CONFIG_PPC64 */
1554 			i += sl + 1;
1555 		}
1556 	}
1557 #ifdef CONFIG_PPC64
1558 	/* If not a mac, try to figure out if it's an IBM pSeries or any other
1559 	 * PAPR compliant platform. We assume it is if :
1560 	 *  - /device_type is "chrp" (please, do NOT use that for future
1561 	 *    non-IBM designs !
1562 	 *  - it has /rtas
1563 	 */
1564 	len = prom_getprop(_prom->root, "device_type",
1565 			   compat, sizeof(compat)-1);
1566 	if (len <= 0)
1567 		return PLATFORM_GENERIC;
1568 	if (strcmp(compat, RELOC("chrp")))
1569 		return PLATFORM_GENERIC;
1570 
1571 	/* Default to pSeries. We need to know if we are running LPAR */
1572 	rtas = call_prom("finddevice", 1, 1, ADDR("/rtas"));
1573 	if (!PHANDLE_VALID(rtas))
1574 		return PLATFORM_GENERIC;
1575 	x = prom_getproplen(rtas, "ibm,hypertas-functions");
1576 	if (x != PROM_ERROR) {
1577 		prom_printf("Hypertas detected, assuming LPAR !\n");
1578 		return PLATFORM_PSERIES_LPAR;
1579 	}
1580 	return PLATFORM_PSERIES;
1581 #else
1582 	return PLATFORM_GENERIC;
1583 #endif
1584 }
1585 
1586 static int __init prom_set_color(ihandle ih, int i, int r, int g, int b)
1587 {
1588 	return call_prom("call-method", 6, 1, ADDR("color!"), ih, i, b, g, r);
1589 }
1590 
1591 /*
1592  * If we have a display that we don't know how to drive,
1593  * we will want to try to execute OF's open method for it
1594  * later.  However, OF will probably fall over if we do that
1595  * we've taken over the MMU.
1596  * So we check whether we will need to open the display,
1597  * and if so, open it now.
1598  */
1599 static void __init prom_check_displays(void)
1600 {
1601 	char type[16], *path;
1602 	phandle node;
1603 	ihandle ih;
1604 	int i;
1605 
1606 	static unsigned char default_colors[] = {
1607 		0x00, 0x00, 0x00,
1608 		0x00, 0x00, 0xaa,
1609 		0x00, 0xaa, 0x00,
1610 		0x00, 0xaa, 0xaa,
1611 		0xaa, 0x00, 0x00,
1612 		0xaa, 0x00, 0xaa,
1613 		0xaa, 0xaa, 0x00,
1614 		0xaa, 0xaa, 0xaa,
1615 		0x55, 0x55, 0x55,
1616 		0x55, 0x55, 0xff,
1617 		0x55, 0xff, 0x55,
1618 		0x55, 0xff, 0xff,
1619 		0xff, 0x55, 0x55,
1620 		0xff, 0x55, 0xff,
1621 		0xff, 0xff, 0x55,
1622 		0xff, 0xff, 0xff
1623 	};
1624 	const unsigned char *clut;
1625 
1626 	prom_printf("Looking for displays\n");
1627 	for (node = 0; prom_next_node(&node); ) {
1628 		memset(type, 0, sizeof(type));
1629 		prom_getprop(node, "device_type", type, sizeof(type));
1630 		if (strcmp(type, RELOC("display")) != 0)
1631 			continue;
1632 
1633 		/* It seems OF doesn't null-terminate the path :-( */
1634 		path = RELOC(prom_scratch);
1635 		memset(path, 0, PROM_SCRATCH_SIZE);
1636 
1637 		/*
1638 		 * leave some room at the end of the path for appending extra
1639 		 * arguments
1640 		 */
1641 		if (call_prom("package-to-path", 3, 1, node, path,
1642 			      PROM_SCRATCH_SIZE-10) == PROM_ERROR)
1643 			continue;
1644 		prom_printf("found display   : %s, opening ... ", path);
1645 
1646 		ih = call_prom("open", 1, 1, path);
1647 		if (ih == 0) {
1648 			prom_printf("failed\n");
1649 			continue;
1650 		}
1651 
1652 		/* Success */
1653 		prom_printf("done\n");
1654 		prom_setprop(node, path, "linux,opened", NULL, 0);
1655 
1656 		/* Setup a usable color table when the appropriate
1657 		 * method is available. Should update this to set-colors */
1658 		clut = RELOC(default_colors);
1659 		for (i = 0; i < 32; i++, clut += 3)
1660 			if (prom_set_color(ih, i, clut[0], clut[1],
1661 					   clut[2]) != 0)
1662 				break;
1663 
1664 #ifdef CONFIG_LOGO_LINUX_CLUT224
1665 		clut = PTRRELOC(RELOC(logo_linux_clut224.clut));
1666 		for (i = 0; i < RELOC(logo_linux_clut224.clutsize); i++, clut += 3)
1667 			if (prom_set_color(ih, i + 32, clut[0], clut[1],
1668 					   clut[2]) != 0)
1669 				break;
1670 #endif /* CONFIG_LOGO_LINUX_CLUT224 */
1671 	}
1672 }
1673 
1674 
1675 /* Return (relocated) pointer to this much memory: moves initrd if reqd. */
1676 static void __init *make_room(unsigned long *mem_start, unsigned long *mem_end,
1677 			      unsigned long needed, unsigned long align)
1678 {
1679 	void *ret;
1680 
1681 	*mem_start = _ALIGN(*mem_start, align);
1682 	while ((*mem_start + needed) > *mem_end) {
1683 		unsigned long room, chunk;
1684 
1685 		prom_debug("Chunk exhausted, claiming more at %x...\n",
1686 			   RELOC(alloc_bottom));
1687 		room = RELOC(alloc_top) - RELOC(alloc_bottom);
1688 		if (room > DEVTREE_CHUNK_SIZE)
1689 			room = DEVTREE_CHUNK_SIZE;
1690 		if (room < PAGE_SIZE)
1691 			prom_panic("No memory for flatten_device_tree (no room)");
1692 		chunk = alloc_up(room, 0);
1693 		if (chunk == 0)
1694 			prom_panic("No memory for flatten_device_tree (claim failed)");
1695 		*mem_end = RELOC(alloc_top);
1696 	}
1697 
1698 	ret = (void *)*mem_start;
1699 	*mem_start += needed;
1700 
1701 	return ret;
1702 }
1703 
1704 #define dt_push_token(token, mem_start, mem_end) \
1705 	do { *((u32 *)make_room(mem_start, mem_end, 4, 4)) = token; } while(0)
1706 
1707 static unsigned long __init dt_find_string(char *str)
1708 {
1709 	char *s, *os;
1710 
1711 	s = os = (char *)RELOC(dt_string_start);
1712 	s += 4;
1713 	while (s <  (char *)RELOC(dt_string_end)) {
1714 		if (strcmp(s, str) == 0)
1715 			return s - os;
1716 		s += strlen(s) + 1;
1717 	}
1718 	return 0;
1719 }
1720 
1721 /*
1722  * The Open Firmware 1275 specification states properties must be 31 bytes or
1723  * less, however not all firmwares obey this. Make it 64 bytes to be safe.
1724  */
1725 #define MAX_PROPERTY_NAME 64
1726 
1727 static void __init scan_dt_build_strings(phandle node,
1728 					 unsigned long *mem_start,
1729 					 unsigned long *mem_end)
1730 {
1731 	char *prev_name, *namep, *sstart;
1732 	unsigned long soff;
1733 	phandle child;
1734 
1735 	sstart =  (char *)RELOC(dt_string_start);
1736 
1737 	/* get and store all property names */
1738 	prev_name = RELOC("");
1739 	for (;;) {
1740 		/* 64 is max len of name including nul. */
1741 		namep = make_room(mem_start, mem_end, MAX_PROPERTY_NAME, 1);
1742 		if (call_prom("nextprop", 3, 1, node, prev_name, namep) != 1) {
1743 			/* No more nodes: unwind alloc */
1744 			*mem_start = (unsigned long)namep;
1745 			break;
1746 		}
1747 
1748  		/* skip "name" */
1749  		if (strcmp(namep, RELOC("name")) == 0) {
1750  			*mem_start = (unsigned long)namep;
1751  			prev_name = RELOC("name");
1752  			continue;
1753  		}
1754 		/* get/create string entry */
1755 		soff = dt_find_string(namep);
1756 		if (soff != 0) {
1757 			*mem_start = (unsigned long)namep;
1758 			namep = sstart + soff;
1759 		} else {
1760 			/* Trim off some if we can */
1761 			*mem_start = (unsigned long)namep + strlen(namep) + 1;
1762 			RELOC(dt_string_end) = *mem_start;
1763 		}
1764 		prev_name = namep;
1765 	}
1766 
1767 	/* do all our children */
1768 	child = call_prom("child", 1, 1, node);
1769 	while (child != 0) {
1770 		scan_dt_build_strings(child, mem_start, mem_end);
1771 		child = call_prom("peer", 1, 1, child);
1772 	}
1773 }
1774 
1775 static void __init scan_dt_build_struct(phandle node, unsigned long *mem_start,
1776 					unsigned long *mem_end)
1777 {
1778 	phandle child;
1779 	char *namep, *prev_name, *sstart, *p, *ep, *lp, *path;
1780 	unsigned long soff;
1781 	unsigned char *valp;
1782 	static char pname[MAX_PROPERTY_NAME];
1783 	int l, room;
1784 
1785 	dt_push_token(OF_DT_BEGIN_NODE, mem_start, mem_end);
1786 
1787 	/* get the node's full name */
1788 	namep = (char *)*mem_start;
1789 	room = *mem_end - *mem_start;
1790 	if (room > 255)
1791 		room = 255;
1792 	l = call_prom("package-to-path", 3, 1, node, namep, room);
1793 	if (l >= 0) {
1794 		/* Didn't fit?  Get more room. */
1795 		if (l >= room) {
1796 			if (l >= *mem_end - *mem_start)
1797 				namep = make_room(mem_start, mem_end, l+1, 1);
1798 			call_prom("package-to-path", 3, 1, node, namep, l);
1799 		}
1800 		namep[l] = '\0';
1801 
1802 		/* Fixup an Apple bug where they have bogus \0 chars in the
1803 		 * middle of the path in some properties, and extract
1804 		 * the unit name (everything after the last '/').
1805 		 */
1806 		for (lp = p = namep, ep = namep + l; p < ep; p++) {
1807 			if (*p == '/')
1808 				lp = namep;
1809 			else if (*p != 0)
1810 				*lp++ = *p;
1811 		}
1812 		*lp = 0;
1813 		*mem_start = _ALIGN((unsigned long)lp + 1, 4);
1814 	}
1815 
1816 	/* get it again for debugging */
1817 	path = RELOC(prom_scratch);
1818 	memset(path, 0, PROM_SCRATCH_SIZE);
1819 	call_prom("package-to-path", 3, 1, node, path, PROM_SCRATCH_SIZE-1);
1820 
1821 	/* get and store all properties */
1822 	prev_name = RELOC("");
1823 	sstart = (char *)RELOC(dt_string_start);
1824 	for (;;) {
1825 		if (call_prom("nextprop", 3, 1, node, prev_name,
1826 			      RELOC(pname)) != 1)
1827 			break;
1828 
1829  		/* skip "name" */
1830  		if (strcmp(RELOC(pname), RELOC("name")) == 0) {
1831  			prev_name = RELOC("name");
1832  			continue;
1833  		}
1834 
1835 		/* find string offset */
1836 		soff = dt_find_string(RELOC(pname));
1837 		if (soff == 0) {
1838 			prom_printf("WARNING: Can't find string index for"
1839 				    " <%s>, node %s\n", RELOC(pname), path);
1840 			break;
1841 		}
1842 		prev_name = sstart + soff;
1843 
1844 		/* get length */
1845 		l = call_prom("getproplen", 2, 1, node, RELOC(pname));
1846 
1847 		/* sanity checks */
1848 		if (l == PROM_ERROR)
1849 			continue;
1850 		if (l > MAX_PROPERTY_LENGTH) {
1851 			prom_printf("WARNING: ignoring large property ");
1852 			/* It seems OF doesn't null-terminate the path :-( */
1853 			prom_printf("[%s] ", path);
1854 			prom_printf("%s length 0x%x\n", RELOC(pname), l);
1855 			continue;
1856 		}
1857 
1858 		/* push property head */
1859 		dt_push_token(OF_DT_PROP, mem_start, mem_end);
1860 		dt_push_token(l, mem_start, mem_end);
1861 		dt_push_token(soff, mem_start, mem_end);
1862 
1863 		/* push property content */
1864 		valp = make_room(mem_start, mem_end, l, 4);
1865 		call_prom("getprop", 4, 1, node, RELOC(pname), valp, l);
1866 		*mem_start = _ALIGN(*mem_start, 4);
1867 	}
1868 
1869 	/* Add a "linux,phandle" property. */
1870 	soff = dt_find_string(RELOC("linux,phandle"));
1871 	if (soff == 0)
1872 		prom_printf("WARNING: Can't find string index for"
1873 			    " <linux-phandle> node %s\n", path);
1874 	else {
1875 		dt_push_token(OF_DT_PROP, mem_start, mem_end);
1876 		dt_push_token(4, mem_start, mem_end);
1877 		dt_push_token(soff, mem_start, mem_end);
1878 		valp = make_room(mem_start, mem_end, 4, 4);
1879 		*(u32 *)valp = node;
1880 	}
1881 
1882 	/* do all our children */
1883 	child = call_prom("child", 1, 1, node);
1884 	while (child != 0) {
1885 		scan_dt_build_struct(child, mem_start, mem_end);
1886 		child = call_prom("peer", 1, 1, child);
1887 	}
1888 
1889 	dt_push_token(OF_DT_END_NODE, mem_start, mem_end);
1890 }
1891 
1892 static void __init flatten_device_tree(void)
1893 {
1894 	phandle root;
1895 	unsigned long mem_start, mem_end, room;
1896 	struct boot_param_header *hdr;
1897 	struct prom_t *_prom = &RELOC(prom);
1898 	char *namep;
1899 	u64 *rsvmap;
1900 
1901 	/*
1902 	 * Check how much room we have between alloc top & bottom (+/- a
1903 	 * few pages), crop to 4Mb, as this is our "chuck" size
1904 	 */
1905 	room = RELOC(alloc_top) - RELOC(alloc_bottom) - 0x4000;
1906 	if (room > DEVTREE_CHUNK_SIZE)
1907 		room = DEVTREE_CHUNK_SIZE;
1908 	prom_debug("starting device tree allocs at %x\n", RELOC(alloc_bottom));
1909 
1910 	/* Now try to claim that */
1911 	mem_start = (unsigned long)alloc_up(room, PAGE_SIZE);
1912 	if (mem_start == 0)
1913 		prom_panic("Can't allocate initial device-tree chunk\n");
1914 	mem_end = RELOC(alloc_top);
1915 
1916 	/* Get root of tree */
1917 	root = call_prom("peer", 1, 1, (phandle)0);
1918 	if (root == (phandle)0)
1919 		prom_panic ("couldn't get device tree root\n");
1920 
1921 	/* Build header and make room for mem rsv map */
1922 	mem_start = _ALIGN(mem_start, 4);
1923 	hdr = make_room(&mem_start, &mem_end,
1924 			sizeof(struct boot_param_header), 4);
1925 	RELOC(dt_header_start) = (unsigned long)hdr;
1926 	rsvmap = make_room(&mem_start, &mem_end, sizeof(mem_reserve_map), 8);
1927 
1928 	/* Start of strings */
1929 	mem_start = PAGE_ALIGN(mem_start);
1930 	RELOC(dt_string_start) = mem_start;
1931 	mem_start += 4; /* hole */
1932 
1933 	/* Add "linux,phandle" in there, we'll need it */
1934 	namep = make_room(&mem_start, &mem_end, 16, 1);
1935 	strcpy(namep, RELOC("linux,phandle"));
1936 	mem_start = (unsigned long)namep + strlen(namep) + 1;
1937 
1938 	/* Build string array */
1939 	prom_printf("Building dt strings...\n");
1940 	scan_dt_build_strings(root, &mem_start, &mem_end);
1941 	RELOC(dt_string_end) = mem_start;
1942 
1943 	/* Build structure */
1944 	mem_start = PAGE_ALIGN(mem_start);
1945 	RELOC(dt_struct_start) = mem_start;
1946 	prom_printf("Building dt structure...\n");
1947 	scan_dt_build_struct(root, &mem_start, &mem_end);
1948 	dt_push_token(OF_DT_END, &mem_start, &mem_end);
1949 	RELOC(dt_struct_end) = PAGE_ALIGN(mem_start);
1950 
1951 	/* Finish header */
1952 	hdr->boot_cpuid_phys = _prom->cpu;
1953 	hdr->magic = OF_DT_HEADER;
1954 	hdr->totalsize = RELOC(dt_struct_end) - RELOC(dt_header_start);
1955 	hdr->off_dt_struct = RELOC(dt_struct_start) - RELOC(dt_header_start);
1956 	hdr->off_dt_strings = RELOC(dt_string_start) - RELOC(dt_header_start);
1957 	hdr->dt_strings_size = RELOC(dt_string_end) - RELOC(dt_string_start);
1958 	hdr->off_mem_rsvmap = ((unsigned long)rsvmap) - RELOC(dt_header_start);
1959 	hdr->version = OF_DT_VERSION;
1960 	/* Version 16 is not backward compatible */
1961 	hdr->last_comp_version = 0x10;
1962 
1963 	/* Copy the reserve map in */
1964 	memcpy(rsvmap, RELOC(mem_reserve_map), sizeof(mem_reserve_map));
1965 
1966 #ifdef DEBUG_PROM
1967 	{
1968 		int i;
1969 		prom_printf("reserved memory map:\n");
1970 		for (i = 0; i < RELOC(mem_reserve_cnt); i++)
1971 			prom_printf("  %x - %x\n",
1972 				    RELOC(mem_reserve_map)[i].base,
1973 				    RELOC(mem_reserve_map)[i].size);
1974 	}
1975 #endif
1976 	/* Bump mem_reserve_cnt to cause further reservations to fail
1977 	 * since it's too late.
1978 	 */
1979 	RELOC(mem_reserve_cnt) = MEM_RESERVE_MAP_SIZE;
1980 
1981 	prom_printf("Device tree strings 0x%x -> 0x%x\n",
1982 		    RELOC(dt_string_start), RELOC(dt_string_end));
1983 	prom_printf("Device tree struct  0x%x -> 0x%x\n",
1984 		    RELOC(dt_struct_start), RELOC(dt_struct_end));
1985 
1986 }
1987 
1988 #ifdef CONFIG_PPC_MAPLE
1989 /* PIBS Version 1.05.0000 04/26/2005 has an incorrect /ht/isa/ranges property.
1990  * The values are bad, and it doesn't even have the right number of cells. */
1991 static void __init fixup_device_tree_maple(void)
1992 {
1993 	phandle isa;
1994 	u32 isa_ranges[6];
1995 
1996 	isa = call_prom("finddevice", 1, 1, ADDR("/ht@0/isa@4"));
1997 	if (!PHANDLE_VALID(isa))
1998 		return;
1999 
2000 	if (prom_getprop(isa, "ranges", isa_ranges, sizeof(isa_ranges))
2001 		== PROM_ERROR)
2002 		return;
2003 
2004 	if (isa_ranges[0] != 0x1 ||
2005 		isa_ranges[1] != 0xf4000000 ||
2006 		isa_ranges[2] != 0x00010000)
2007 		return;
2008 
2009 	prom_printf("fixing up bogus ISA range on Maple...\n");
2010 
2011 	isa_ranges[0] = 0x1;
2012 	isa_ranges[1] = 0x0;
2013 	isa_ranges[2] = 0x01002000; /* IO space; PCI device = 4 */
2014 	isa_ranges[3] = 0x0;
2015 	isa_ranges[4] = 0x0;
2016 	isa_ranges[5] = 0x00010000;
2017 	prom_setprop(isa, "/ht@0/isa@4", "ranges",
2018 			isa_ranges, sizeof(isa_ranges));
2019 }
2020 #else
2021 #define fixup_device_tree_maple()
2022 #endif
2023 
2024 #if defined(CONFIG_PPC64) && defined(CONFIG_PPC_PMAC)
2025 static void __init fixup_device_tree_pmac(void)
2026 {
2027 	phandle u3, i2c, mpic;
2028 	u32 u3_rev;
2029 	u32 interrupts[2];
2030 	u32 parent;
2031 
2032 	/* Some G5s have a missing interrupt definition, fix it up here */
2033 	u3 = call_prom("finddevice", 1, 1, ADDR("/u3@0,f8000000"));
2034 	if (!PHANDLE_VALID(u3))
2035 		return;
2036 	i2c = call_prom("finddevice", 1, 1, ADDR("/u3@0,f8000000/i2c@f8001000"));
2037 	if (!PHANDLE_VALID(i2c))
2038 		return;
2039 	mpic = call_prom("finddevice", 1, 1, ADDR("/u3@0,f8000000/mpic@f8040000"));
2040 	if (!PHANDLE_VALID(mpic))
2041 		return;
2042 
2043 	/* check if proper rev of u3 */
2044 	if (prom_getprop(u3, "device-rev", &u3_rev, sizeof(u3_rev))
2045 	    == PROM_ERROR)
2046 		return;
2047 	if (u3_rev < 0x35 || u3_rev > 0x39)
2048 		return;
2049 	/* does it need fixup ? */
2050 	if (prom_getproplen(i2c, "interrupts") > 0)
2051 		return;
2052 
2053 	prom_printf("fixing up bogus interrupts for u3 i2c...\n");
2054 
2055 	/* interrupt on this revision of u3 is number 0 and level */
2056 	interrupts[0] = 0;
2057 	interrupts[1] = 1;
2058 	prom_setprop(i2c, "/u3@0,f8000000/i2c@f8001000", "interrupts",
2059 		     &interrupts, sizeof(interrupts));
2060 	parent = (u32)mpic;
2061 	prom_setprop(i2c, "/u3@0,f8000000/i2c@f8001000", "interrupt-parent",
2062 		     &parent, sizeof(parent));
2063 }
2064 #else
2065 #define fixup_device_tree_pmac()
2066 #endif
2067 
2068 static void __init fixup_device_tree(void)
2069 {
2070 	fixup_device_tree_maple();
2071 	fixup_device_tree_pmac();
2072 }
2073 
2074 static void __init prom_find_boot_cpu(void)
2075 {
2076        	struct prom_t *_prom = &RELOC(prom);
2077 	u32 getprop_rval;
2078 	ihandle prom_cpu;
2079 	phandle cpu_pkg;
2080 
2081 	_prom->cpu = 0;
2082 	if (prom_getprop(_prom->chosen, "cpu", &prom_cpu, sizeof(prom_cpu)) <= 0)
2083 		return;
2084 
2085 	cpu_pkg = call_prom("instance-to-package", 1, 1, prom_cpu);
2086 
2087 	prom_getprop(cpu_pkg, "reg", &getprop_rval, sizeof(getprop_rval));
2088 	_prom->cpu = getprop_rval;
2089 
2090 	prom_debug("Booting CPU hw index = 0x%x\n", _prom->cpu);
2091 }
2092 
2093 static void __init prom_check_initrd(unsigned long r3, unsigned long r4)
2094 {
2095 #ifdef CONFIG_BLK_DEV_INITRD
2096        	struct prom_t *_prom = &RELOC(prom);
2097 
2098 	if (r3 && r4 && r4 != 0xdeadbeef) {
2099 		unsigned long val;
2100 
2101 		RELOC(prom_initrd_start) = is_kernel_addr(r3) ? __pa(r3) : r3;
2102 		RELOC(prom_initrd_end) = RELOC(prom_initrd_start) + r4;
2103 
2104 		val = RELOC(prom_initrd_start);
2105 		prom_setprop(_prom->chosen, "/chosen", "linux,initrd-start",
2106 			     &val, sizeof(val));
2107 		val = RELOC(prom_initrd_end);
2108 		prom_setprop(_prom->chosen, "/chosen", "linux,initrd-end",
2109 			     &val, sizeof(val));
2110 
2111 		reserve_mem(RELOC(prom_initrd_start),
2112 			    RELOC(prom_initrd_end) - RELOC(prom_initrd_start));
2113 
2114 		prom_debug("initrd_start=0x%x\n", RELOC(prom_initrd_start));
2115 		prom_debug("initrd_end=0x%x\n", RELOC(prom_initrd_end));
2116 	}
2117 #endif /* CONFIG_BLK_DEV_INITRD */
2118 }
2119 
2120 /*
2121  * We enter here early on, when the Open Firmware prom is still
2122  * handling exceptions and the MMU hash table for us.
2123  */
2124 
2125 unsigned long __init prom_init(unsigned long r3, unsigned long r4,
2126 			       unsigned long pp,
2127 			       unsigned long r6, unsigned long r7)
2128 {
2129        	struct prom_t *_prom;
2130 	unsigned long hdr;
2131 	unsigned long offset = reloc_offset();
2132 
2133 #ifdef CONFIG_PPC32
2134 	reloc_got2(offset);
2135 #endif
2136 
2137 	_prom = &RELOC(prom);
2138 
2139 	/*
2140 	 * First zero the BSS
2141 	 */
2142 	memset(&RELOC(__bss_start), 0, __bss_stop - __bss_start);
2143 
2144 	/*
2145 	 * Init interface to Open Firmware, get some node references,
2146 	 * like /chosen
2147 	 */
2148 	prom_init_client_services(pp);
2149 
2150 	/*
2151 	 * See if this OF is old enough that we need to do explicit maps
2152 	 * and other workarounds
2153 	 */
2154 	prom_find_mmu();
2155 
2156 	/*
2157 	 * Init prom stdout device
2158 	 */
2159 	prom_init_stdout();
2160 
2161 	/*
2162 	 * Get default machine type. At this point, we do not differentiate
2163 	 * between pSeries SMP and pSeries LPAR
2164 	 */
2165 	RELOC(of_platform) = prom_find_machine_type();
2166 
2167 	/* Bail if this is a kdump kernel. */
2168 	if (PHYSICAL_START > 0)
2169 		prom_panic("Error: You can't boot a kdump kernel from OF!\n");
2170 
2171 	/*
2172 	 * Check for an initrd
2173 	 */
2174 	prom_check_initrd(r3, r4);
2175 
2176 #ifdef CONFIG_PPC_PSERIES
2177 	/*
2178 	 * On pSeries, inform the firmware about our capabilities
2179 	 */
2180 	if (RELOC(of_platform) == PLATFORM_PSERIES ||
2181 	    RELOC(of_platform) == PLATFORM_PSERIES_LPAR)
2182 		prom_send_capabilities();
2183 #endif
2184 
2185 	/*
2186 	 * Copy the CPU hold code
2187 	 */
2188        	if (RELOC(of_platform) != PLATFORM_POWERMAC)
2189        		copy_and_flush(0, KERNELBASE + offset, 0x100, 0);
2190 
2191 	/*
2192 	 * Do early parsing of command line
2193 	 */
2194 	early_cmdline_parse();
2195 
2196 	/*
2197 	 * Initialize memory management within prom_init
2198 	 */
2199 	prom_init_mem();
2200 
2201 	/*
2202 	 * Determine which cpu is actually running right _now_
2203 	 */
2204 	prom_find_boot_cpu();
2205 
2206 	/*
2207 	 * Initialize display devices
2208 	 */
2209 	prom_check_displays();
2210 
2211 #ifdef CONFIG_PPC64
2212 	/*
2213 	 * Initialize IOMMU (TCE tables) on pSeries. Do that before anything else
2214 	 * that uses the allocator, we need to make sure we get the top of memory
2215 	 * available for us here...
2216 	 */
2217 	if (RELOC(of_platform) == PLATFORM_PSERIES)
2218 		prom_initialize_tce_table();
2219 #endif
2220 
2221 	/*
2222 	 * On non-powermacs, try to instantiate RTAS and puts all CPUs
2223 	 * in spin-loops. PowerMacs don't have a working RTAS and use
2224 	 * a different way to spin CPUs
2225 	 */
2226 	if (RELOC(of_platform) != PLATFORM_POWERMAC) {
2227 		prom_instantiate_rtas();
2228 		prom_hold_cpus();
2229 	}
2230 
2231 	/*
2232 	 * Fill in some infos for use by the kernel later on
2233 	 */
2234 #ifdef CONFIG_PPC64
2235 	if (RELOC(ppc64_iommu_off))
2236 		prom_setprop(_prom->chosen, "/chosen", "linux,iommu-off",
2237 			     NULL, 0);
2238 
2239 	if (RELOC(iommu_force_on))
2240 		prom_setprop(_prom->chosen, "/chosen", "linux,iommu-force-on",
2241 			     NULL, 0);
2242 
2243 	if (RELOC(prom_tce_alloc_start)) {
2244 		prom_setprop(_prom->chosen, "/chosen", "linux,tce-alloc-start",
2245 			     &RELOC(prom_tce_alloc_start),
2246 			     sizeof(prom_tce_alloc_start));
2247 		prom_setprop(_prom->chosen, "/chosen", "linux,tce-alloc-end",
2248 			     &RELOC(prom_tce_alloc_end),
2249 			     sizeof(prom_tce_alloc_end));
2250 	}
2251 #endif
2252 
2253 	/*
2254 	 * Fixup any known bugs in the device-tree
2255 	 */
2256 	fixup_device_tree();
2257 
2258 	/*
2259 	 * Now finally create the flattened device-tree
2260 	 */
2261 	prom_printf("copying OF device tree ...\n");
2262 	flatten_device_tree();
2263 
2264 	/*
2265 	 * in case stdin is USB and still active on IBM machines...
2266 	 * Unfortunately quiesce crashes on some powermacs if we have
2267 	 * closed stdin already (in particular the powerbook 101).
2268 	 */
2269 	if (RELOC(of_platform) != PLATFORM_POWERMAC)
2270 		prom_close_stdin();
2271 
2272 	/*
2273 	 * Call OF "quiesce" method to shut down pending DMA's from
2274 	 * devices etc...
2275 	 */
2276 	prom_printf("Calling quiesce ...\n");
2277 	call_prom("quiesce", 0, 0);
2278 
2279 	/*
2280 	 * And finally, call the kernel passing it the flattened device
2281 	 * tree and NULL as r5, thus triggering the new entry point which
2282 	 * is common to us and kexec
2283 	 */
2284 	hdr = RELOC(dt_header_start);
2285 	prom_printf("returning from prom_init\n");
2286 	prom_debug("->dt_header_start=0x%x\n", hdr);
2287 
2288 #ifdef CONFIG_PPC32
2289 	reloc_got2(-offset);
2290 #endif
2291 
2292 	__start(hdr, KERNELBASE + offset, 0);
2293 
2294 	return 0;
2295 }
2296