Linux Audio

Check our new training course

Loading...
v5.14.15
   1// SPDX-License-Identifier: GPL-2.0-or-later
   2/*
   3 * Procedures for interfacing to Open Firmware.
   4 *
   5 * Paul Mackerras	August 1996.
   6 * Copyright (C) 1996-2005 Paul Mackerras.
   7 * 
   8 *  Adapted for 64bit PowerPC by Dave Engebretsen and Peter Bergner.
   9 *    {engebret|bergner}@us.ibm.com 
 
 
 
 
 
  10 */
  11
  12#undef DEBUG_PROM
  13
  14/* we cannot use FORTIFY as it brings in new symbols */
  15#define __NO_FORTIFY
  16
  17#include <stdarg.h>
  18#include <linux/kernel.h>
  19#include <linux/string.h>
  20#include <linux/init.h>
  21#include <linux/threads.h>
  22#include <linux/spinlock.h>
  23#include <linux/types.h>
  24#include <linux/pci.h>
  25#include <linux/proc_fs.h>
 
  26#include <linux/delay.h>
  27#include <linux/initrd.h>
  28#include <linux/bitops.h>
  29#include <linux/pgtable.h>
  30#include <linux/printk.h>
  31#include <asm/prom.h>
  32#include <asm/rtas.h>
  33#include <asm/page.h>
  34#include <asm/processor.h>
  35#include <asm/interrupt.h>
  36#include <asm/irq.h>
  37#include <asm/io.h>
  38#include <asm/smp.h>
  39#include <asm/mmu.h>
 
  40#include <asm/iommu.h>
  41#include <asm/btext.h>
  42#include <asm/sections.h>
  43#include <asm/machdep.h>
 
  44#include <asm/asm-prototypes.h>
  45#include <asm/ultravisor-api.h>
  46
  47#include <linux/linux_logo.h>
  48
  49/* All of prom_init bss lives here */
  50#define __prombss __section(".bss.prominit")
  51
  52/*
  53 * Eventually bump that one up
  54 */
  55#define DEVTREE_CHUNK_SIZE	0x100000
  56
  57/*
  58 * This is the size of the local memory reserve map that gets copied
  59 * into the boot params passed to the kernel. That size is totally
  60 * flexible as the kernel just reads the list until it encounters an
  61 * entry with size 0, so it can be changed without breaking binary
  62 * compatibility
  63 */
  64#define MEM_RESERVE_MAP_SIZE	8
  65
  66/*
  67 * prom_init() is called very early on, before the kernel text
  68 * and data have been mapped to KERNELBASE.  At this point the code
  69 * is running at whatever address it has been loaded at.
  70 * On ppc32 we compile with -mrelocatable, which means that references
  71 * to extern and static variables get relocated automatically.
  72 * ppc64 objects are always relocatable, we just need to relocate the
  73 * TOC.
  74 *
  75 * Because OF may have mapped I/O devices into the area starting at
  76 * KERNELBASE, particularly on CHRP machines, we can't safely call
  77 * OF once the kernel has been mapped to KERNELBASE.  Therefore all
  78 * OF calls must be done within prom_init().
  79 *
  80 * ADDR is used in calls to call_prom.  The 4th and following
  81 * arguments to call_prom should be 32-bit values.
  82 * On ppc64, 64 bit values are truncated to 32 bits (and
  83 * fortunately don't get interpreted as two arguments).
  84 */
  85#define ADDR(x)		(u32)(unsigned long)(x)
  86
  87#ifdef CONFIG_PPC64
  88#define OF_WORKAROUNDS	0
  89#else
  90#define OF_WORKAROUNDS	of_workarounds
  91static int of_workarounds __prombss;
  92#endif
  93
  94#define OF_WA_CLAIM	1	/* do phys/virt claim separately, then map */
  95#define OF_WA_LONGTRAIL	2	/* work around longtrail bugs */
  96
  97#define PROM_BUG() do {						\
  98        prom_printf("kernel BUG at %s line 0x%x!\n",		\
  99		    __FILE__, __LINE__);			\
 100	__builtin_trap();					\
 101} while (0)
 102
 103#ifdef DEBUG_PROM
 104#define prom_debug(x...)	prom_printf(x)
 105#else
 106#define prom_debug(x...)	do { } while (0)
 107#endif
 108
 109
 110typedef u32 prom_arg_t;
 111
 112struct prom_args {
 113        __be32 service;
 114        __be32 nargs;
 115        __be32 nret;
 116        __be32 args[10];
 117};
 118
 119struct prom_t {
 120	ihandle root;
 121	phandle chosen;
 122	int cpu;
 123	ihandle stdout;
 124	ihandle mmumap;
 125	ihandle memory;
 126};
 127
 128struct mem_map_entry {
 129	__be64	base;
 130	__be64	size;
 131};
 132
 133typedef __be32 cell_t;
 134
 135extern void __start(unsigned long r3, unsigned long r4, unsigned long r5,
 136		    unsigned long r6, unsigned long r7, unsigned long r8,
 137		    unsigned long r9);
 138
 139#ifdef CONFIG_PPC64
 140extern int enter_prom(struct prom_args *args, unsigned long entry);
 141#else
 142static inline int enter_prom(struct prom_args *args, unsigned long entry)
 143{
 144	return ((int (*)(struct prom_args *))entry)(args);
 145}
 146#endif
 147
 148extern void copy_and_flush(unsigned long dest, unsigned long src,
 149			   unsigned long size, unsigned long offset);
 150
 151/* prom structure */
 152static struct prom_t __prombss prom;
 153
 154static unsigned long __prombss prom_entry;
 155
 156static char __prombss of_stdout_device[256];
 157static char __prombss prom_scratch[256];
 158
 159static unsigned long __prombss dt_header_start;
 160static unsigned long __prombss dt_struct_start, dt_struct_end;
 161static unsigned long __prombss dt_string_start, dt_string_end;
 162
 163static unsigned long __prombss prom_initrd_start, prom_initrd_end;
 
 
 164
 165#ifdef CONFIG_PPC64
 166static int __prombss prom_iommu_force_on;
 167static int __prombss prom_iommu_off;
 168static unsigned long __prombss prom_tce_alloc_start;
 169static unsigned long __prombss prom_tce_alloc_end;
 170#endif
 171
 172#ifdef CONFIG_PPC_PSERIES
 173static bool __prombss prom_radix_disable;
 174static bool __prombss prom_radix_gtse_disable;
 175static bool __prombss prom_xive_disable;
 176#endif
 177
 178#ifdef CONFIG_PPC_SVM
 179static bool __prombss prom_svm_enable;
 180#endif
 181
 182struct platform_support {
 183	bool hash_mmu;
 184	bool radix_mmu;
 185	bool radix_gtse;
 186	bool xive;
 187};
 188
 189/* Platforms codes are now obsolete in the kernel. Now only used within this
 190 * file and ultimately gone too. Feel free to change them if you need, they
 191 * are not shared with anything outside of this file anymore
 192 */
 193#define PLATFORM_PSERIES	0x0100
 194#define PLATFORM_PSERIES_LPAR	0x0101
 195#define PLATFORM_LPAR		0x0001
 196#define PLATFORM_POWERMAC	0x0400
 197#define PLATFORM_GENERIC	0x0500
 
 198
 199static int __prombss of_platform;
 200
 201static char __prombss prom_cmd_line[COMMAND_LINE_SIZE];
 202
 203static unsigned long __prombss prom_memory_limit;
 204
 205static unsigned long __prombss alloc_top;
 206static unsigned long __prombss alloc_top_high;
 207static unsigned long __prombss alloc_bottom;
 208static unsigned long __prombss rmo_top;
 209static unsigned long __prombss ram_top;
 210
 211static struct mem_map_entry __prombss mem_reserve_map[MEM_RESERVE_MAP_SIZE];
 212static int __prombss mem_reserve_cnt;
 213
 214static cell_t __prombss regbuf[1024];
 215
 216static bool  __prombss rtas_has_query_cpu_stopped;
 217
 218
 219/*
 220 * Error results ... some OF calls will return "-1" on error, some
 221 * will return 0, some will return either. To simplify, here are
 222 * macros to use with any ihandle or phandle return value to check if
 223 * it is valid
 224 */
 225
 226#define PROM_ERROR		(-1u)
 227#define PHANDLE_VALID(p)	((p) != 0 && (p) != PROM_ERROR)
 228#define IHANDLE_VALID(i)	((i) != 0 && (i) != PROM_ERROR)
 229
 230/* Copied from lib/string.c and lib/kstrtox.c */
 231
 232static int __init prom_strcmp(const char *cs, const char *ct)
 233{
 234	unsigned char c1, c2;
 235
 236	while (1) {
 237		c1 = *cs++;
 238		c2 = *ct++;
 239		if (c1 != c2)
 240			return c1 < c2 ? -1 : 1;
 241		if (!c1)
 242			break;
 243	}
 244	return 0;
 245}
 246
 247static ssize_t __init prom_strscpy_pad(char *dest, const char *src, size_t n)
 248{
 249	ssize_t rc;
 250	size_t i;
 251
 252	if (n == 0 || n > INT_MAX)
 253		return -E2BIG;
 254
 255	// Copy up to n bytes
 256	for (i = 0; i < n && src[i] != '\0'; i++)
 257		dest[i] = src[i];
 258
 259	rc = i;
 260
 261	// If we copied all n then we have run out of space for the nul
 262	if (rc == n) {
 263		// Rewind by one character to ensure nul termination
 264		i--;
 265		rc = -E2BIG;
 266	}
 267
 268	for (; i < n; i++)
 269		dest[i] = '\0';
 270
 271	return rc;
 272}
 273
 274static int __init prom_strncmp(const char *cs, const char *ct, size_t count)
 275{
 276	unsigned char c1, c2;
 277
 278	while (count) {
 279		c1 = *cs++;
 280		c2 = *ct++;
 281		if (c1 != c2)
 282			return c1 < c2 ? -1 : 1;
 283		if (!c1)
 284			break;
 285		count--;
 286	}
 287	return 0;
 288}
 289
 290static size_t __init prom_strlen(const char *s)
 291{
 292	const char *sc;
 293
 294	for (sc = s; *sc != '\0'; ++sc)
 295		/* nothing */;
 296	return sc - s;
 297}
 298
 299static int __init prom_memcmp(const void *cs, const void *ct, size_t count)
 300{
 301	const unsigned char *su1, *su2;
 302	int res = 0;
 303
 304	for (su1 = cs, su2 = ct; 0 < count; ++su1, ++su2, count--)
 305		if ((res = *su1 - *su2) != 0)
 306			break;
 307	return res;
 308}
 309
 310static char __init *prom_strstr(const char *s1, const char *s2)
 311{
 312	size_t l1, l2;
 313
 314	l2 = prom_strlen(s2);
 315	if (!l2)
 316		return (char *)s1;
 317	l1 = prom_strlen(s1);
 318	while (l1 >= l2) {
 319		l1--;
 320		if (!prom_memcmp(s1, s2, l2))
 321			return (char *)s1;
 322		s1++;
 323	}
 324	return NULL;
 325}
 326
 327static size_t __init prom_strlcat(char *dest, const char *src, size_t count)
 328{
 329	size_t dsize = prom_strlen(dest);
 330	size_t len = prom_strlen(src);
 331	size_t res = dsize + len;
 332
 333	/* This would be a bug */
 334	if (dsize >= count)
 335		return count;
 336
 337	dest += dsize;
 338	count -= dsize;
 339	if (len >= count)
 340		len = count-1;
 341	memcpy(dest, src, len);
 342	dest[len] = 0;
 343	return res;
 344
 345}
 346
 347#ifdef CONFIG_PPC_PSERIES
 348static int __init prom_strtobool(const char *s, bool *res)
 349{
 350	if (!s)
 351		return -EINVAL;
 352
 353	switch (s[0]) {
 354	case 'y':
 355	case 'Y':
 356	case '1':
 357		*res = true;
 358		return 0;
 359	case 'n':
 360	case 'N':
 361	case '0':
 362		*res = false;
 363		return 0;
 364	case 'o':
 365	case 'O':
 366		switch (s[1]) {
 367		case 'n':
 368		case 'N':
 369			*res = true;
 370			return 0;
 371		case 'f':
 372		case 'F':
 373			*res = false;
 374			return 0;
 375		default:
 376			break;
 377		}
 378		break;
 379	default:
 380		break;
 381	}
 382
 383	return -EINVAL;
 384}
 385#endif
 386
 387/* This is the one and *ONLY* place where we actually call open
 388 * firmware.
 389 */
 390
 391static int __init call_prom(const char *service, int nargs, int nret, ...)
 392{
 393	int i;
 394	struct prom_args args;
 395	va_list list;
 396
 397	args.service = cpu_to_be32(ADDR(service));
 398	args.nargs = cpu_to_be32(nargs);
 399	args.nret = cpu_to_be32(nret);
 400
 401	va_start(list, nret);
 402	for (i = 0; i < nargs; i++)
 403		args.args[i] = cpu_to_be32(va_arg(list, prom_arg_t));
 404	va_end(list);
 405
 406	for (i = 0; i < nret; i++)
 407		args.args[nargs+i] = 0;
 408
 409	if (enter_prom(&args, prom_entry) < 0)
 410		return PROM_ERROR;
 411
 412	return (nret > 0) ? be32_to_cpu(args.args[nargs]) : 0;
 413}
 414
 415static int __init call_prom_ret(const char *service, int nargs, int nret,
 416				prom_arg_t *rets, ...)
 417{
 418	int i;
 419	struct prom_args args;
 420	va_list list;
 421
 422	args.service = cpu_to_be32(ADDR(service));
 423	args.nargs = cpu_to_be32(nargs);
 424	args.nret = cpu_to_be32(nret);
 425
 426	va_start(list, rets);
 427	for (i = 0; i < nargs; i++)
 428		args.args[i] = cpu_to_be32(va_arg(list, prom_arg_t));
 429	va_end(list);
 430
 431	for (i = 0; i < nret; i++)
 432		args.args[nargs+i] = 0;
 433
 434	if (enter_prom(&args, prom_entry) < 0)
 435		return PROM_ERROR;
 436
 437	if (rets != NULL)
 438		for (i = 1; i < nret; ++i)
 439			rets[i-1] = be32_to_cpu(args.args[nargs+i]);
 440
 441	return (nret > 0) ? be32_to_cpu(args.args[nargs]) : 0;
 442}
 443
 444
 445static void __init prom_print(const char *msg)
 446{
 447	const char *p, *q;
 448
 449	if (prom.stdout == 0)
 450		return;
 451
 452	for (p = msg; *p != 0; p = q) {
 453		for (q = p; *q != 0 && *q != '\n'; ++q)
 454			;
 455		if (q > p)
 456			call_prom("write", 3, 1, prom.stdout, p, q - p);
 457		if (*q == 0)
 458			break;
 459		++q;
 460		call_prom("write", 3, 1, prom.stdout, ADDR("\r\n"), 2);
 461	}
 462}
 463
 464
 465/*
 466 * Both prom_print_hex & prom_print_dec takes an unsigned long as input so that
 467 * we do not need __udivdi3 or __umoddi3 on 32bits.
 468 */
 469static void __init prom_print_hex(unsigned long val)
 470{
 471	int i, nibbles = sizeof(val)*2;
 472	char buf[sizeof(val)*2+1];
 473
 474	for (i = nibbles-1;  i >= 0;  i--) {
 475		buf[i] = (val & 0xf) + '0';
 476		if (buf[i] > '9')
 477			buf[i] += ('a'-'0'-10);
 478		val >>= 4;
 479	}
 480	buf[nibbles] = '\0';
 481	call_prom("write", 3, 1, prom.stdout, buf, nibbles);
 482}
 483
 484/* max number of decimal digits in an unsigned long */
 485#define UL_DIGITS 21
 486static void __init prom_print_dec(unsigned long val)
 487{
 488	int i, size;
 489	char buf[UL_DIGITS+1];
 490
 491	for (i = UL_DIGITS-1; i >= 0;  i--) {
 492		buf[i] = (val % 10) + '0';
 493		val = val/10;
 494		if (val == 0)
 495			break;
 496	}
 497	/* shift stuff down */
 498	size = UL_DIGITS - i;
 499	call_prom("write", 3, 1, prom.stdout, buf+i, size);
 500}
 501
 502__printf(1, 2)
 503static void __init prom_printf(const char *format, ...)
 504{
 505	const char *p, *q, *s;
 506	va_list args;
 507	unsigned long v;
 508	long vs;
 509	int n = 0;
 510
 511	va_start(args, format);
 512	for (p = format; *p != 0; p = q) {
 513		for (q = p; *q != 0 && *q != '\n' && *q != '%'; ++q)
 514			;
 515		if (q > p)
 516			call_prom("write", 3, 1, prom.stdout, p, q - p);
 517		if (*q == 0)
 518			break;
 519		if (*q == '\n') {
 520			++q;
 521			call_prom("write", 3, 1, prom.stdout,
 522				  ADDR("\r\n"), 2);
 523			continue;
 524		}
 525		++q;
 526		if (*q == 0)
 527			break;
 528		while (*q == 'l') {
 529			++q;
 530			++n;
 531		}
 532		switch (*q) {
 533		case 's':
 534			++q;
 535			s = va_arg(args, const char *);
 536			prom_print(s);
 537			break;
 538		case 'x':
 539			++q;
 540			switch (n) {
 541			case 0:
 542				v = va_arg(args, unsigned int);
 543				break;
 544			case 1:
 545				v = va_arg(args, unsigned long);
 546				break;
 547			case 2:
 548			default:
 549				v = va_arg(args, unsigned long long);
 550				break;
 551			}
 552			prom_print_hex(v);
 553			break;
 554		case 'u':
 555			++q;
 556			switch (n) {
 557			case 0:
 558				v = va_arg(args, unsigned int);
 559				break;
 560			case 1:
 561				v = va_arg(args, unsigned long);
 562				break;
 563			case 2:
 564			default:
 565				v = va_arg(args, unsigned long long);
 566				break;
 567			}
 568			prom_print_dec(v);
 569			break;
 570		case 'd':
 571			++q;
 572			switch (n) {
 573			case 0:
 574				vs = va_arg(args, int);
 575				break;
 576			case 1:
 577				vs = va_arg(args, long);
 578				break;
 579			case 2:
 580			default:
 581				vs = va_arg(args, long long);
 582				break;
 583			}
 584			if (vs < 0) {
 585				prom_print("-");
 586				vs = -vs;
 587			}
 588			prom_print_dec(vs);
 589			break;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 590		}
 591	}
 592	va_end(args);
 593}
 594
 595
 596static unsigned int __init prom_claim(unsigned long virt, unsigned long size,
 597				unsigned long align)
 598{
 599
 600	if (align == 0 && (OF_WORKAROUNDS & OF_WA_CLAIM)) {
 601		/*
 602		 * Old OF requires we claim physical and virtual separately
 603		 * and then map explicitly (assuming virtual mode)
 604		 */
 605		int ret;
 606		prom_arg_t result;
 607
 608		ret = call_prom_ret("call-method", 5, 2, &result,
 609				    ADDR("claim"), prom.memory,
 610				    align, size, virt);
 611		if (ret != 0 || result == -1)
 612			return -1;
 613		ret = call_prom_ret("call-method", 5, 2, &result,
 614				    ADDR("claim"), prom.mmumap,
 615				    align, size, virt);
 616		if (ret != 0) {
 617			call_prom("call-method", 4, 1, ADDR("release"),
 618				  prom.memory, size, virt);
 619			return -1;
 620		}
 621		/* the 0x12 is M (coherence) + PP == read/write */
 622		call_prom("call-method", 6, 1,
 623			  ADDR("map"), prom.mmumap, 0x12, size, virt, virt);
 624		return virt;
 625	}
 626	return call_prom("claim", 3, 1, (prom_arg_t)virt, (prom_arg_t)size,
 627			 (prom_arg_t)align);
 628}
 629
 630static void __init __attribute__((noreturn)) prom_panic(const char *reason)
 631{
 632	prom_print(reason);
 633	/* Do not call exit because it clears the screen on pmac
 634	 * it also causes some sort of double-fault on early pmacs */
 635	if (of_platform == PLATFORM_POWERMAC)
 636		asm("trap\n");
 637
 638	/* ToDo: should put up an SRC here on pSeries */
 639	call_prom("exit", 0, 0);
 640
 641	for (;;)			/* should never get here */
 642		;
 643}
 644
 645
 646static int __init prom_next_node(phandle *nodep)
 647{
 648	phandle node;
 649
 650	if ((node = *nodep) != 0
 651	    && (*nodep = call_prom("child", 1, 1, node)) != 0)
 652		return 1;
 653	if ((*nodep = call_prom("peer", 1, 1, node)) != 0)
 654		return 1;
 655	for (;;) {
 656		if ((node = call_prom("parent", 1, 1, node)) == 0)
 657			return 0;
 658		if ((*nodep = call_prom("peer", 1, 1, node)) != 0)
 659			return 1;
 660	}
 661}
 662
 663static inline int __init prom_getprop(phandle node, const char *pname,
 664				      void *value, size_t valuelen)
 665{
 666	return call_prom("getprop", 4, 1, node, ADDR(pname),
 667			 (u32)(unsigned long) value, (u32) valuelen);
 668}
 669
 670static inline int __init prom_getproplen(phandle node, const char *pname)
 671{
 672	return call_prom("getproplen", 2, 1, node, ADDR(pname));
 673}
 674
 675static void add_string(char **str, const char *q)
 676{
 677	char *p = *str;
 678
 679	while (*q)
 680		*p++ = *q++;
 681	*p++ = ' ';
 682	*str = p;
 683}
 684
 685static char *tohex(unsigned int x)
 686{
 687	static const char digits[] __initconst = "0123456789abcdef";
 688	static char result[9] __prombss;
 689	int i;
 690
 691	result[8] = 0;
 692	i = 8;
 693	do {
 694		--i;
 695		result[i] = digits[x & 0xf];
 696		x >>= 4;
 697	} while (x != 0 && i > 0);
 698	return &result[i];
 699}
 700
 701static int __init prom_setprop(phandle node, const char *nodename,
 702			       const char *pname, void *value, size_t valuelen)
 703{
 704	char cmd[256], *p;
 705
 706	if (!(OF_WORKAROUNDS & OF_WA_LONGTRAIL))
 707		return call_prom("setprop", 4, 1, node, ADDR(pname),
 708				 (u32)(unsigned long) value, (u32) valuelen);
 709
 710	/* gah... setprop doesn't work on longtrail, have to use interpret */
 711	p = cmd;
 712	add_string(&p, "dev");
 713	add_string(&p, nodename);
 714	add_string(&p, tohex((u32)(unsigned long) value));
 715	add_string(&p, tohex(valuelen));
 716	add_string(&p, tohex(ADDR(pname)));
 717	add_string(&p, tohex(prom_strlen(pname)));
 718	add_string(&p, "property");
 719	*p = 0;
 720	return call_prom("interpret", 1, 1, (u32)(unsigned long) cmd);
 721}
 722
 723/* We can't use the standard versions because of relocation headaches. */
 724#define prom_isxdigit(c) \
 725	(('0' <= (c) && (c) <= '9') || ('a' <= (c) && (c) <= 'f') || ('A' <= (c) && (c) <= 'F'))
 726
 727#define prom_isdigit(c)	('0' <= (c) && (c) <= '9')
 728#define prom_islower(c)	('a' <= (c) && (c) <= 'z')
 729#define prom_toupper(c)	(prom_islower(c) ? ((c) - 'a' + 'A') : (c))
 
 730
 731static unsigned long prom_strtoul(const char *cp, const char **endp)
 732{
 733	unsigned long result = 0, base = 10, value;
 734
 735	if (*cp == '0') {
 736		base = 8;
 737		cp++;
 738		if (prom_toupper(*cp) == 'X') {
 739			cp++;
 740			base = 16;
 741		}
 742	}
 743
 744	while (prom_isxdigit(*cp) &&
 745	       (value = prom_isdigit(*cp) ? *cp - '0' : prom_toupper(*cp) - 'A' + 10) < base) {
 746		result = result * base + value;
 747		cp++;
 748	}
 749
 750	if (endp)
 751		*endp = cp;
 752
 753	return result;
 754}
 755
 756static unsigned long prom_memparse(const char *ptr, const char **retptr)
 757{
 758	unsigned long ret = prom_strtoul(ptr, retptr);
 759	int shift = 0;
 760
 761	/*
 762	 * We can't use a switch here because GCC *may* generate a
 763	 * jump table which won't work, because we're not running at
 764	 * the address we're linked at.
 765	 */
 766	if ('G' == **retptr || 'g' == **retptr)
 767		shift = 30;
 768
 769	if ('M' == **retptr || 'm' == **retptr)
 770		shift = 20;
 771
 772	if ('K' == **retptr || 'k' == **retptr)
 773		shift = 10;
 774
 775	if (shift) {
 776		ret <<= shift;
 777		(*retptr)++;
 778	}
 779
 780	return ret;
 781}
 782
 783/*
 784 * Early parsing of the command line passed to the kernel, used for
 785 * "mem=x" and the options that affect the iommu
 786 */
 787static void __init early_cmdline_parse(void)
 788{
 789	const char *opt;
 790
 791	char *p;
 792	int l = 0;
 793
 794	prom_cmd_line[0] = 0;
 795	p = prom_cmd_line;
 796
 797	if (!IS_ENABLED(CONFIG_CMDLINE_FORCE) && (long)prom.chosen > 0)
 798		l = prom_getprop(prom.chosen, "bootargs", p, COMMAND_LINE_SIZE-1);
 799
 800	if (IS_ENABLED(CONFIG_CMDLINE_EXTEND) || l <= 0 || p[0] == '\0')
 801		prom_strlcat(prom_cmd_line, " " CONFIG_CMDLINE,
 802			     sizeof(prom_cmd_line));
 803
 804	prom_printf("command line: %s\n", prom_cmd_line);
 805
 806#ifdef CONFIG_PPC64
 807	opt = prom_strstr(prom_cmd_line, "iommu=");
 808	if (opt) {
 809		prom_printf("iommu opt is: %s\n", opt);
 810		opt += 6;
 811		while (*opt && *opt == ' ')
 812			opt++;
 813		if (!prom_strncmp(opt, "off", 3))
 814			prom_iommu_off = 1;
 815		else if (!prom_strncmp(opt, "force", 5))
 816			prom_iommu_force_on = 1;
 817	}
 818#endif
 819	opt = prom_strstr(prom_cmd_line, "mem=");
 820	if (opt) {
 821		opt += 4;
 822		prom_memory_limit = prom_memparse(opt, (const char **)&opt);
 823#ifdef CONFIG_PPC64
 824		/* Align to 16 MB == size of ppc64 large page */
 825		prom_memory_limit = ALIGN(prom_memory_limit, 0x1000000);
 826#endif
 827	}
 828
 829#ifdef CONFIG_PPC_PSERIES
 830	prom_radix_disable = !IS_ENABLED(CONFIG_PPC_RADIX_MMU_DEFAULT);
 831	opt = prom_strstr(prom_cmd_line, "disable_radix");
 832	if (opt) {
 833		opt += 13;
 834		if (*opt && *opt == '=') {
 835			bool val;
 836
 837			if (prom_strtobool(++opt, &val))
 838				prom_radix_disable = false;
 839			else
 840				prom_radix_disable = val;
 841		} else
 842			prom_radix_disable = true;
 843	}
 844	if (prom_radix_disable)
 845		prom_debug("Radix disabled from cmdline\n");
 846
 847	opt = prom_strstr(prom_cmd_line, "radix_hcall_invalidate=on");
 848	if (opt) {
 849		prom_radix_gtse_disable = true;
 850		prom_debug("Radix GTSE disabled from cmdline\n");
 851	}
 852
 853	opt = prom_strstr(prom_cmd_line, "xive=off");
 854	if (opt) {
 855		prom_xive_disable = true;
 856		prom_debug("XIVE disabled from cmdline\n");
 857	}
 858#endif /* CONFIG_PPC_PSERIES */
 859
 860#ifdef CONFIG_PPC_SVM
 861	opt = prom_strstr(prom_cmd_line, "svm=");
 862	if (opt) {
 863		bool val;
 864
 865		opt += sizeof("svm=") - 1;
 866		if (!prom_strtobool(opt, &val))
 867			prom_svm_enable = val;
 868	}
 869#endif /* CONFIG_PPC_SVM */
 870}
 871
 872#ifdef CONFIG_PPC_PSERIES
 873/*
 874 * The architecture vector has an array of PVR mask/value pairs,
 875 * followed by # option vectors - 1, followed by the option vectors.
 876 *
 877 * See prom.h for the definition of the bits specified in the
 878 * architecture vector.
 879 */
 880
 881/* Firmware expects the value to be n - 1, where n is the # of vectors */
 882#define NUM_VECTORS(n)		((n) - 1)
 883
 884/*
 885 * Firmware expects 1 + n - 2, where n is the length of the option vector in
 886 * bytes. The 1 accounts for the length byte itself, the - 2 .. ?
 887 */
 888#define VECTOR_LENGTH(n)	(1 + (n) - 2)
 889
 890struct option_vector1 {
 891	u8 byte1;
 892	u8 arch_versions;
 893	u8 arch_versions3;
 894} __packed;
 895
 896struct option_vector2 {
 897	u8 byte1;
 898	__be16 reserved;
 899	__be32 real_base;
 900	__be32 real_size;
 901	__be32 virt_base;
 902	__be32 virt_size;
 903	__be32 load_base;
 904	__be32 min_rma;
 905	__be32 min_load;
 906	u8 min_rma_percent;
 907	u8 max_pft_size;
 908} __packed;
 909
 910struct option_vector3 {
 911	u8 byte1;
 912	u8 byte2;
 913} __packed;
 914
 915struct option_vector4 {
 916	u8 byte1;
 917	u8 min_vp_cap;
 918} __packed;
 919
 920struct option_vector5 {
 921	u8 byte1;
 922	u8 byte2;
 923	u8 byte3;
 924	u8 cmo;
 925	u8 associativity;
 926	u8 bin_opts;
 927	u8 micro_checkpoint;
 928	u8 reserved0;
 929	__be32 max_cpus;
 930	__be16 papr_level;
 931	__be16 reserved1;
 932	u8 platform_facilities;
 933	u8 reserved2;
 934	__be16 reserved3;
 935	u8 subprocessors;
 936	u8 byte22;
 937	u8 intarch;
 938	u8 mmu;
 939	u8 hash_ext;
 940	u8 radix_ext;
 941} __packed;
 942
 943struct option_vector6 {
 944	u8 reserved;
 945	u8 secondary_pteg;
 946	u8 os_name;
 947} __packed;
 948
 949struct option_vector7 {
 950	u8 os_id[256];
 951} __packed;
 952
 953struct ibm_arch_vec {
 954	struct { u32 mask, val; } pvrs[14];
 955
 956	u8 num_vectors;
 957
 958	u8 vec1_len;
 959	struct option_vector1 vec1;
 960
 961	u8 vec2_len;
 962	struct option_vector2 vec2;
 963
 964	u8 vec3_len;
 965	struct option_vector3 vec3;
 966
 967	u8 vec4_len;
 968	struct option_vector4 vec4;
 969
 970	u8 vec5_len;
 971	struct option_vector5 vec5;
 972
 973	u8 vec6_len;
 974	struct option_vector6 vec6;
 975
 976	u8 vec7_len;
 977	struct option_vector7 vec7;
 978} __packed;
 979
 980static const struct ibm_arch_vec ibm_architecture_vec_template __initconst = {
 981	.pvrs = {
 982		{
 983			.mask = cpu_to_be32(0xfffe0000), /* POWER5/POWER5+ */
 984			.val  = cpu_to_be32(0x003a0000),
 985		},
 986		{
 987			.mask = cpu_to_be32(0xffff0000), /* POWER6 */
 988			.val  = cpu_to_be32(0x003e0000),
 989		},
 990		{
 991			.mask = cpu_to_be32(0xffff0000), /* POWER7 */
 992			.val  = cpu_to_be32(0x003f0000),
 993		},
 994		{
 995			.mask = cpu_to_be32(0xffff0000), /* POWER8E */
 996			.val  = cpu_to_be32(0x004b0000),
 997		},
 998		{
 999			.mask = cpu_to_be32(0xffff0000), /* POWER8NVL */
1000			.val  = cpu_to_be32(0x004c0000),
1001		},
1002		{
1003			.mask = cpu_to_be32(0xffff0000), /* POWER8 */
1004			.val  = cpu_to_be32(0x004d0000),
1005		},
1006		{
1007			.mask = cpu_to_be32(0xffff0000), /* POWER9 */
1008			.val  = cpu_to_be32(0x004e0000),
1009		},
1010		{
1011			.mask = cpu_to_be32(0xffff0000), /* POWER10 */
1012			.val  = cpu_to_be32(0x00800000),
1013		},
1014		{
1015			.mask = cpu_to_be32(0xffffffff), /* all 3.1-compliant */
1016			.val  = cpu_to_be32(0x0f000006),
1017		},
1018		{
1019			.mask = cpu_to_be32(0xffffffff), /* all 3.00-compliant */
1020			.val  = cpu_to_be32(0x0f000005),
1021		},
1022		{
1023			.mask = cpu_to_be32(0xffffffff), /* all 2.07-compliant */
1024			.val  = cpu_to_be32(0x0f000004),
1025		},
1026		{
1027			.mask = cpu_to_be32(0xffffffff), /* all 2.06-compliant */
1028			.val  = cpu_to_be32(0x0f000003),
1029		},
1030		{
1031			.mask = cpu_to_be32(0xffffffff), /* all 2.05-compliant */
1032			.val  = cpu_to_be32(0x0f000002),
1033		},
1034		{
1035			.mask = cpu_to_be32(0xfffffffe), /* all 2.04-compliant and earlier */
1036			.val  = cpu_to_be32(0x0f000001),
1037		},
1038	},
1039
1040	.num_vectors = NUM_VECTORS(6),
1041
1042	.vec1_len = VECTOR_LENGTH(sizeof(struct option_vector1)),
1043	.vec1 = {
1044		.byte1 = 0,
1045		.arch_versions = OV1_PPC_2_00 | OV1_PPC_2_01 | OV1_PPC_2_02 | OV1_PPC_2_03 |
1046				 OV1_PPC_2_04 | OV1_PPC_2_05 | OV1_PPC_2_06 | OV1_PPC_2_07,
1047		.arch_versions3 = OV1_PPC_3_00 | OV1_PPC_3_1,
1048	},
1049
1050	.vec2_len = VECTOR_LENGTH(sizeof(struct option_vector2)),
1051	/* option vector 2: Open Firmware options supported */
1052	.vec2 = {
1053		.byte1 = OV2_REAL_MODE,
1054		.reserved = 0,
1055		.real_base = cpu_to_be32(0xffffffff),
1056		.real_size = cpu_to_be32(0xffffffff),
1057		.virt_base = cpu_to_be32(0xffffffff),
1058		.virt_size = cpu_to_be32(0xffffffff),
1059		.load_base = cpu_to_be32(0xffffffff),
1060		.min_rma = cpu_to_be32(512),		/* 512MB min RMA */
1061		.min_load = cpu_to_be32(0xffffffff),	/* full client load */
1062		.min_rma_percent = 0,	/* min RMA percentage of total RAM */
1063		.max_pft_size = 48,	/* max log_2(hash table size) */
1064	},
1065
1066	.vec3_len = VECTOR_LENGTH(sizeof(struct option_vector3)),
1067	/* option vector 3: processor options supported */
1068	.vec3 = {
1069		.byte1 = 0,			/* don't ignore, don't halt */
1070		.byte2 = OV3_FP | OV3_VMX | OV3_DFP,
1071	},
1072
1073	.vec4_len = VECTOR_LENGTH(sizeof(struct option_vector4)),
1074	/* option vector 4: IBM PAPR implementation */
1075	.vec4 = {
1076		.byte1 = 0,			/* don't halt */
1077		.min_vp_cap = OV4_MIN_ENT_CAP,	/* minimum VP entitled capacity */
1078	},
1079
1080	.vec5_len = VECTOR_LENGTH(sizeof(struct option_vector5)),
1081	/* option vector 5: PAPR/OF options */
1082	.vec5 = {
1083		.byte1 = 0,				/* don't ignore, don't halt */
1084		.byte2 = OV5_FEAT(OV5_LPAR) | OV5_FEAT(OV5_SPLPAR) | OV5_FEAT(OV5_LARGE_PAGES) |
1085		OV5_FEAT(OV5_DRCONF_MEMORY) | OV5_FEAT(OV5_DONATE_DEDICATE_CPU) |
1086#ifdef CONFIG_PCI_MSI
1087		/* PCIe/MSI support.  Without MSI full PCIe is not supported */
1088		OV5_FEAT(OV5_MSI),
1089#else
1090		0,
1091#endif
1092		.byte3 = 0,
1093		.cmo =
1094#ifdef CONFIG_PPC_SMLPAR
1095		OV5_FEAT(OV5_CMO) | OV5_FEAT(OV5_XCMO),
1096#else
1097		0,
1098#endif
1099		.associativity = OV5_FEAT(OV5_TYPE1_AFFINITY) | OV5_FEAT(OV5_PRRN),
1100		.bin_opts = OV5_FEAT(OV5_RESIZE_HPT) | OV5_FEAT(OV5_HP_EVT),
1101		.micro_checkpoint = 0,
1102		.reserved0 = 0,
1103		.max_cpus = cpu_to_be32(NR_CPUS),	/* number of cores supported */
1104		.papr_level = 0,
1105		.reserved1 = 0,
1106		.platform_facilities = OV5_FEAT(OV5_PFO_HW_RNG) | OV5_FEAT(OV5_PFO_HW_ENCR) | OV5_FEAT(OV5_PFO_HW_842),
1107		.reserved2 = 0,
1108		.reserved3 = 0,
1109		.subprocessors = 1,
1110		.byte22 = OV5_FEAT(OV5_DRMEM_V2) | OV5_FEAT(OV5_DRC_INFO),
1111		.intarch = 0,
1112		.mmu = 0,
1113		.hash_ext = 0,
1114		.radix_ext = 0,
1115	},
1116
1117	/* option vector 6: IBM PAPR hints */
1118	.vec6_len = VECTOR_LENGTH(sizeof(struct option_vector6)),
1119	.vec6 = {
1120		.reserved = 0,
1121		.secondary_pteg = 0,
1122		.os_name = OV6_LINUX,
1123	},
1124
1125	/* option vector 7: OS Identification */
1126	.vec7_len = VECTOR_LENGTH(sizeof(struct option_vector7)),
1127};
1128
1129static struct ibm_arch_vec __prombss ibm_architecture_vec  ____cacheline_aligned;
1130
1131/* Old method - ELF header with PT_NOTE sections only works on BE */
1132#ifdef __BIG_ENDIAN__
1133static const struct fake_elf {
1134	Elf32_Ehdr	elfhdr;
1135	Elf32_Phdr	phdr[2];
1136	struct chrpnote {
1137		u32	namesz;
1138		u32	descsz;
1139		u32	type;
1140		char	name[8];	/* "PowerPC" */
1141		struct chrpdesc {
1142			u32	real_mode;
1143			u32	real_base;
1144			u32	real_size;
1145			u32	virt_base;
1146			u32	virt_size;
1147			u32	load_base;
1148		} chrpdesc;
1149	} chrpnote;
1150	struct rpanote {
1151		u32	namesz;
1152		u32	descsz;
1153		u32	type;
1154		char	name[24];	/* "IBM,RPA-Client-Config" */
1155		struct rpadesc {
1156			u32	lpar_affinity;
1157			u32	min_rmo_size;
1158			u32	min_rmo_percent;
1159			u32	max_pft_size;
1160			u32	splpar;
1161			u32	min_load;
1162			u32	new_mem_def;
1163			u32	ignore_me;
1164		} rpadesc;
1165	} rpanote;
1166} fake_elf __initconst = {
1167	.elfhdr = {
1168		.e_ident = { 0x7f, 'E', 'L', 'F',
1169			     ELFCLASS32, ELFDATA2MSB, EV_CURRENT },
1170		.e_type = ET_EXEC,	/* yeah right */
1171		.e_machine = EM_PPC,
1172		.e_version = EV_CURRENT,
1173		.e_phoff = offsetof(struct fake_elf, phdr),
1174		.e_phentsize = sizeof(Elf32_Phdr),
1175		.e_phnum = 2
1176	},
1177	.phdr = {
1178		[0] = {
1179			.p_type = PT_NOTE,
1180			.p_offset = offsetof(struct fake_elf, chrpnote),
1181			.p_filesz = sizeof(struct chrpnote)
1182		}, [1] = {
1183			.p_type = PT_NOTE,
1184			.p_offset = offsetof(struct fake_elf, rpanote),
1185			.p_filesz = sizeof(struct rpanote)
1186		}
1187	},
1188	.chrpnote = {
1189		.namesz = sizeof("PowerPC"),
1190		.descsz = sizeof(struct chrpdesc),
1191		.type = 0x1275,
1192		.name = "PowerPC",
1193		.chrpdesc = {
1194			.real_mode = ~0U,	/* ~0 means "don't care" */
1195			.real_base = ~0U,
1196			.real_size = ~0U,
1197			.virt_base = ~0U,
1198			.virt_size = ~0U,
1199			.load_base = ~0U
1200		},
1201	},
1202	.rpanote = {
1203		.namesz = sizeof("IBM,RPA-Client-Config"),
1204		.descsz = sizeof(struct rpadesc),
1205		.type = 0x12759999,
1206		.name = "IBM,RPA-Client-Config",
1207		.rpadesc = {
1208			.lpar_affinity = 0,
1209			.min_rmo_size = 64,	/* in megabytes */
1210			.min_rmo_percent = 0,
1211			.max_pft_size = 48,	/* 2^48 bytes max PFT size */
1212			.splpar = 1,
1213			.min_load = ~0U,
1214			.new_mem_def = 0
1215		}
1216	}
1217};
1218#endif /* __BIG_ENDIAN__ */
1219
1220static int __init prom_count_smt_threads(void)
1221{
1222	phandle node;
1223	char type[64];
1224	unsigned int plen;
1225
1226	/* Pick up th first CPU node we can find */
1227	for (node = 0; prom_next_node(&node); ) {
1228		type[0] = 0;
1229		prom_getprop(node, "device_type", type, sizeof(type));
1230
1231		if (prom_strcmp(type, "cpu"))
1232			continue;
1233		/*
1234		 * There is an entry for each smt thread, each entry being
1235		 * 4 bytes long.  All cpus should have the same number of
1236		 * smt threads, so return after finding the first.
1237		 */
1238		plen = prom_getproplen(node, "ibm,ppc-interrupt-server#s");
1239		if (plen == PROM_ERROR)
1240			break;
1241		plen >>= 2;
1242		prom_debug("Found %lu smt threads per core\n", (unsigned long)plen);
1243
1244		/* Sanity check */
1245		if (plen < 1 || plen > 64) {
1246			prom_printf("Threads per core %lu out of bounds, assuming 1\n",
1247				    (unsigned long)plen);
1248			return 1;
1249		}
1250		return plen;
1251	}
1252	prom_debug("No threads found, assuming 1 per core\n");
1253
1254	return 1;
1255
1256}
1257
1258static void __init prom_parse_mmu_model(u8 val,
1259					struct platform_support *support)
1260{
1261	switch (val) {
1262	case OV5_FEAT(OV5_MMU_DYNAMIC):
1263	case OV5_FEAT(OV5_MMU_EITHER): /* Either Available */
1264		prom_debug("MMU - either supported\n");
1265		support->radix_mmu = !prom_radix_disable;
1266		support->hash_mmu = true;
1267		break;
1268	case OV5_FEAT(OV5_MMU_RADIX): /* Only Radix */
1269		prom_debug("MMU - radix only\n");
1270		if (prom_radix_disable) {
1271			/*
1272			 * If we __have__ to do radix, we're better off ignoring
1273			 * the command line rather than not booting.
1274			 */
1275			prom_printf("WARNING: Ignoring cmdline option disable_radix\n");
1276		}
1277		support->radix_mmu = true;
1278		break;
1279	case OV5_FEAT(OV5_MMU_HASH):
1280		prom_debug("MMU - hash only\n");
1281		support->hash_mmu = true;
1282		break;
1283	default:
1284		prom_debug("Unknown mmu support option: 0x%x\n", val);
1285		break;
1286	}
1287}
1288
1289static void __init prom_parse_xive_model(u8 val,
1290					 struct platform_support *support)
1291{
1292	switch (val) {
1293	case OV5_FEAT(OV5_XIVE_EITHER): /* Either Available */
1294		prom_debug("XIVE - either mode supported\n");
1295		support->xive = !prom_xive_disable;
1296		break;
1297	case OV5_FEAT(OV5_XIVE_EXPLOIT): /* Only Exploitation mode */
1298		prom_debug("XIVE - exploitation mode supported\n");
1299		if (prom_xive_disable) {
1300			/*
1301			 * If we __have__ to do XIVE, we're better off ignoring
1302			 * the command line rather than not booting.
1303			 */
1304			prom_printf("WARNING: Ignoring cmdline option xive=off\n");
1305		}
1306		support->xive = true;
1307		break;
1308	case OV5_FEAT(OV5_XIVE_LEGACY): /* Only Legacy mode */
1309		prom_debug("XIVE - legacy mode supported\n");
1310		break;
1311	default:
1312		prom_debug("Unknown xive support option: 0x%x\n", val);
1313		break;
1314	}
1315}
1316
1317static void __init prom_parse_platform_support(u8 index, u8 val,
1318					       struct platform_support *support)
1319{
1320	switch (index) {
1321	case OV5_INDX(OV5_MMU_SUPPORT): /* MMU Model */
1322		prom_parse_mmu_model(val & OV5_FEAT(OV5_MMU_SUPPORT), support);
1323		break;
1324	case OV5_INDX(OV5_RADIX_GTSE): /* Radix Extensions */
1325		if (val & OV5_FEAT(OV5_RADIX_GTSE))
1326			support->radix_gtse = !prom_radix_gtse_disable;
1327		break;
1328	case OV5_INDX(OV5_XIVE_SUPPORT): /* Interrupt mode */
1329		prom_parse_xive_model(val & OV5_FEAT(OV5_XIVE_SUPPORT),
1330				      support);
1331		break;
1332	}
1333}
1334
1335static void __init prom_check_platform_support(void)
1336{
1337	struct platform_support supported = {
1338		.hash_mmu = false,
1339		.radix_mmu = false,
1340		.radix_gtse = false,
1341		.xive = false
1342	};
1343	int prop_len = prom_getproplen(prom.chosen,
1344				       "ibm,arch-vec-5-platform-support");
1345
1346	/*
1347	 * First copy the architecture vec template
1348	 *
1349	 * use memcpy() instead of *vec = *vec_template so that GCC replaces it
1350	 * by __memcpy() when KASAN is active
1351	 */
1352	memcpy(&ibm_architecture_vec, &ibm_architecture_vec_template,
1353	       sizeof(ibm_architecture_vec));
1354
1355	prom_strscpy_pad(ibm_architecture_vec.vec7.os_id, linux_banner, 256);
1356
1357	if (prop_len > 1) {
1358		int i;
1359		u8 vec[8];
1360		prom_debug("Found ibm,arch-vec-5-platform-support, len: %d\n",
1361			   prop_len);
1362		if (prop_len > sizeof(vec))
1363			prom_printf("WARNING: ibm,arch-vec-5-platform-support longer than expected (len: %d)\n",
1364				    prop_len);
1365		prom_getprop(prom.chosen, "ibm,arch-vec-5-platform-support", &vec, sizeof(vec));
1366		for (i = 0; i < prop_len; i += 2) {
1367			prom_debug("%d: index = 0x%x val = 0x%x\n", i / 2, vec[i], vec[i + 1]);
1368			prom_parse_platform_support(vec[i], vec[i + 1], &supported);
1369		}
1370	}
1371
1372	if (supported.radix_mmu && IS_ENABLED(CONFIG_PPC_RADIX_MMU)) {
1373		/* Radix preferred - Check if GTSE is also supported */
1374		prom_debug("Asking for radix\n");
1375		ibm_architecture_vec.vec5.mmu = OV5_FEAT(OV5_MMU_RADIX);
1376		if (supported.radix_gtse)
1377			ibm_architecture_vec.vec5.radix_ext =
1378					OV5_FEAT(OV5_RADIX_GTSE);
1379		else
1380			prom_debug("Radix GTSE isn't supported\n");
1381	} else if (supported.hash_mmu) {
1382		/* Default to hash mmu (if we can) */
1383		prom_debug("Asking for hash\n");
1384		ibm_architecture_vec.vec5.mmu = OV5_FEAT(OV5_MMU_HASH);
1385	} else {
1386		/* We're probably on a legacy hypervisor */
1387		prom_debug("Assuming legacy hash support\n");
1388	}
1389
1390	if (supported.xive) {
1391		prom_debug("Asking for XIVE\n");
1392		ibm_architecture_vec.vec5.intarch = OV5_FEAT(OV5_XIVE_EXPLOIT);
1393	}
1394}
1395
1396static void __init prom_send_capabilities(void)
1397{
1398	ihandle root;
1399	prom_arg_t ret;
1400	u32 cores;
1401
1402	/* Check ibm,arch-vec-5-platform-support and fixup vec5 if required */
1403	prom_check_platform_support();
1404
1405	root = call_prom("open", 1, 1, ADDR("/"));
1406	if (root != 0) {
1407		/* We need to tell the FW about the number of cores we support.
1408		 *
1409		 * To do that, we count the number of threads on the first core
1410		 * (we assume this is the same for all cores) and use it to
1411		 * divide NR_CPUS.
1412		 */
1413
1414		cores = DIV_ROUND_UP(NR_CPUS, prom_count_smt_threads());
1415		prom_printf("Max number of cores passed to firmware: %u (NR_CPUS = %d)\n",
1416			    cores, NR_CPUS);
1417
1418		ibm_architecture_vec.vec5.max_cpus = cpu_to_be32(cores);
1419
1420		/* try calling the ibm,client-architecture-support method */
1421		prom_printf("Calling ibm,client-architecture-support...");
1422		if (call_prom_ret("call-method", 3, 2, &ret,
1423				  ADDR("ibm,client-architecture-support"),
1424				  root,
1425				  ADDR(&ibm_architecture_vec)) == 0) {
1426			/* the call exists... */
1427			if (ret)
1428				prom_printf("\nWARNING: ibm,client-architecture"
1429					    "-support call FAILED!\n");
1430			call_prom("close", 1, 0, root);
1431			prom_printf(" done\n");
1432			return;
1433		}
1434		call_prom("close", 1, 0, root);
1435		prom_printf(" not implemented\n");
1436	}
1437
1438#ifdef __BIG_ENDIAN__
1439	{
1440		ihandle elfloader;
1441
1442		/* no ibm,client-architecture-support call, try the old way */
1443		elfloader = call_prom("open", 1, 1,
1444				      ADDR("/packages/elf-loader"));
1445		if (elfloader == 0) {
1446			prom_printf("couldn't open /packages/elf-loader\n");
1447			return;
1448		}
1449		call_prom("call-method", 3, 1, ADDR("process-elf-header"),
1450			  elfloader, ADDR(&fake_elf));
1451		call_prom("close", 1, 0, elfloader);
1452	}
1453#endif /* __BIG_ENDIAN__ */
1454}
1455#endif /* CONFIG_PPC_PSERIES */
1456
1457/*
1458 * Memory allocation strategy... our layout is normally:
1459 *
1460 *  at 14Mb or more we have vmlinux, then a gap and initrd.  In some
1461 *  rare cases, initrd might end up being before the kernel though.
1462 *  We assume this won't override the final kernel at 0, we have no
1463 *  provision to handle that in this version, but it should hopefully
1464 *  never happen.
1465 *
1466 *  alloc_top is set to the top of RMO, eventually shrink down if the
1467 *  TCEs overlap
1468 *
1469 *  alloc_bottom is set to the top of kernel/initrd
1470 *
1471 *  from there, allocations are done this way : rtas is allocated
1472 *  topmost, and the device-tree is allocated from the bottom. We try
1473 *  to grow the device-tree allocation as we progress. If we can't,
1474 *  then we fail, we don't currently have a facility to restart
1475 *  elsewhere, but that shouldn't be necessary.
1476 *
1477 *  Note that calls to reserve_mem have to be done explicitly, memory
1478 *  allocated with either alloc_up or alloc_down isn't automatically
1479 *  reserved.
1480 */
1481
1482
1483/*
1484 * Allocates memory in the RMO upward from the kernel/initrd
1485 *
1486 * When align is 0, this is a special case, it means to allocate in place
1487 * at the current location of alloc_bottom or fail (that is basically
1488 * extending the previous allocation). Used for the device-tree flattening
1489 */
1490static unsigned long __init alloc_up(unsigned long size, unsigned long align)
1491{
1492	unsigned long base = alloc_bottom;
1493	unsigned long addr = 0;
1494
1495	if (align)
1496		base = ALIGN(base, align);
1497	prom_debug("%s(%lx, %lx)\n", __func__, size, align);
1498	if (ram_top == 0)
1499		prom_panic("alloc_up() called with mem not initialized\n");
1500
1501	if (align)
1502		base = ALIGN(alloc_bottom, align);
1503	else
1504		base = alloc_bottom;
1505
1506	for(; (base + size) <= alloc_top; 
1507	    base = ALIGN(base + 0x100000, align)) {
1508		prom_debug("    trying: 0x%lx\n\r", base);
1509		addr = (unsigned long)prom_claim(base, size, 0);
1510		if (addr != PROM_ERROR && addr != 0)
1511			break;
1512		addr = 0;
1513		if (align == 0)
1514			break;
1515	}
1516	if (addr == 0)
1517		return 0;
1518	alloc_bottom = addr + size;
1519
1520	prom_debug(" -> %lx\n", addr);
1521	prom_debug("  alloc_bottom : %lx\n", alloc_bottom);
1522	prom_debug("  alloc_top    : %lx\n", alloc_top);
1523	prom_debug("  alloc_top_hi : %lx\n", alloc_top_high);
1524	prom_debug("  rmo_top      : %lx\n", rmo_top);
1525	prom_debug("  ram_top      : %lx\n", ram_top);
1526
1527	return addr;
1528}
1529
1530/*
1531 * Allocates memory downward, either from top of RMO, or if highmem
1532 * is set, from the top of RAM.  Note that this one doesn't handle
1533 * failures.  It does claim memory if highmem is not set.
1534 */
1535static unsigned long __init alloc_down(unsigned long size, unsigned long align,
1536				       int highmem)
1537{
1538	unsigned long base, addr = 0;
1539
1540	prom_debug("%s(%lx, %lx, %s)\n", __func__, size, align,
1541		   highmem ? "(high)" : "(low)");
1542	if (ram_top == 0)
1543		prom_panic("alloc_down() called with mem not initialized\n");
1544
1545	if (highmem) {
1546		/* Carve out storage for the TCE table. */
1547		addr = ALIGN_DOWN(alloc_top_high - size, align);
1548		if (addr <= alloc_bottom)
1549			return 0;
1550		/* Will we bump into the RMO ? If yes, check out that we
1551		 * didn't overlap existing allocations there, if we did,
1552		 * we are dead, we must be the first in town !
1553		 */
1554		if (addr < rmo_top) {
1555			/* Good, we are first */
1556			if (alloc_top == rmo_top)
1557				alloc_top = rmo_top = addr;
1558			else
1559				return 0;
1560		}
1561		alloc_top_high = addr;
1562		goto bail;
1563	}
1564
1565	base = ALIGN_DOWN(alloc_top - size, align);
1566	for (; base > alloc_bottom;
1567	     base = ALIGN_DOWN(base - 0x100000, align))  {
1568		prom_debug("    trying: 0x%lx\n\r", base);
1569		addr = (unsigned long)prom_claim(base, size, 0);
1570		if (addr != PROM_ERROR && addr != 0)
1571			break;
1572		addr = 0;
1573	}
1574	if (addr == 0)
1575		return 0;
1576	alloc_top = addr;
1577
1578 bail:
1579	prom_debug(" -> %lx\n", addr);
1580	prom_debug("  alloc_bottom : %lx\n", alloc_bottom);
1581	prom_debug("  alloc_top    : %lx\n", alloc_top);
1582	prom_debug("  alloc_top_hi : %lx\n", alloc_top_high);
1583	prom_debug("  rmo_top      : %lx\n", rmo_top);
1584	prom_debug("  ram_top      : %lx\n", ram_top);
1585
1586	return addr;
1587}
1588
1589/*
1590 * Parse a "reg" cell
1591 */
1592static unsigned long __init prom_next_cell(int s, cell_t **cellp)
1593{
1594	cell_t *p = *cellp;
1595	unsigned long r = 0;
1596
1597	/* Ignore more than 2 cells */
1598	while (s > sizeof(unsigned long) / 4) {
1599		p++;
1600		s--;
1601	}
1602	r = be32_to_cpu(*p++);
1603#ifdef CONFIG_PPC64
1604	if (s > 1) {
1605		r <<= 32;
1606		r |= be32_to_cpu(*(p++));
1607	}
1608#endif
1609	*cellp = p;
1610	return r;
1611}
1612
1613/*
1614 * Very dumb function for adding to the memory reserve list, but
1615 * we don't need anything smarter at this point
1616 *
1617 * XXX Eventually check for collisions.  They should NEVER happen.
1618 * If problems seem to show up, it would be a good start to track
1619 * them down.
1620 */
1621static void __init reserve_mem(u64 base, u64 size)
1622{
1623	u64 top = base + size;
1624	unsigned long cnt = mem_reserve_cnt;
1625
1626	if (size == 0)
1627		return;
1628
1629	/* We need to always keep one empty entry so that we
1630	 * have our terminator with "size" set to 0 since we are
1631	 * dumb and just copy this entire array to the boot params
1632	 */
1633	base = ALIGN_DOWN(base, PAGE_SIZE);
1634	top = ALIGN(top, PAGE_SIZE);
1635	size = top - base;
1636
1637	if (cnt >= (MEM_RESERVE_MAP_SIZE - 1))
1638		prom_panic("Memory reserve map exhausted !\n");
1639	mem_reserve_map[cnt].base = cpu_to_be64(base);
1640	mem_reserve_map[cnt].size = cpu_to_be64(size);
1641	mem_reserve_cnt = cnt + 1;
1642}
1643
1644/*
1645 * Initialize memory allocation mechanism, parse "memory" nodes and
1646 * obtain that way the top of memory and RMO to setup out local allocator
1647 */
1648static void __init prom_init_mem(void)
1649{
1650	phandle node;
1651	char type[64];
1652	unsigned int plen;
1653	cell_t *p, *endp;
1654	__be32 val;
1655	u32 rac, rsc;
1656
1657	/*
1658	 * We iterate the memory nodes to find
1659	 * 1) top of RMO (first node)
1660	 * 2) top of memory
1661	 */
1662	val = cpu_to_be32(2);
1663	prom_getprop(prom.root, "#address-cells", &val, sizeof(val));
1664	rac = be32_to_cpu(val);
1665	val = cpu_to_be32(1);
1666	prom_getprop(prom.root, "#size-cells", &val, sizeof(rsc));
1667	rsc = be32_to_cpu(val);
1668	prom_debug("root_addr_cells: %x\n", rac);
1669	prom_debug("root_size_cells: %x\n", rsc);
1670
1671	prom_debug("scanning memory:\n");
 
1672
1673	for (node = 0; prom_next_node(&node); ) {
1674		type[0] = 0;
1675		prom_getprop(node, "device_type", type, sizeof(type));
1676
1677		if (type[0] == 0) {
1678			/*
1679			 * CHRP Longtrail machines have no device_type
1680			 * on the memory node, so check the name instead...
1681			 */
1682			prom_getprop(node, "name", type, sizeof(type));
1683		}
1684		if (prom_strcmp(type, "memory"))
1685			continue;
1686
1687		plen = prom_getprop(node, "reg", regbuf, sizeof(regbuf));
1688		if (plen > sizeof(regbuf)) {
1689			prom_printf("memory node too large for buffer !\n");
1690			plen = sizeof(regbuf);
1691		}
1692		p = regbuf;
1693		endp = p + (plen / sizeof(cell_t));
1694
1695#ifdef DEBUG_PROM
1696		memset(prom_scratch, 0, sizeof(prom_scratch));
1697		call_prom("package-to-path", 3, 1, node, prom_scratch,
1698			  sizeof(prom_scratch) - 1);
1699		prom_debug("  node %s :\n", prom_scratch);
1700#endif /* DEBUG_PROM */
1701
1702		while ((endp - p) >= (rac + rsc)) {
1703			unsigned long base, size;
1704
1705			base = prom_next_cell(rac, &p);
1706			size = prom_next_cell(rsc, &p);
1707
1708			if (size == 0)
1709				continue;
1710			prom_debug("    %lx %lx\n", base, size);
1711			if (base == 0 && (of_platform & PLATFORM_LPAR))
1712				rmo_top = size;
1713			if ((base + size) > ram_top)
1714				ram_top = base + size;
1715		}
1716	}
1717
1718	alloc_bottom = PAGE_ALIGN((unsigned long)&_end + 0x4000);
1719
1720	/*
1721	 * If prom_memory_limit is set we reduce the upper limits *except* for
1722	 * alloc_top_high. This must be the real top of RAM so we can put
1723	 * TCE's up there.
1724	 */
1725
1726	alloc_top_high = ram_top;
1727
1728	if (prom_memory_limit) {
1729		if (prom_memory_limit <= alloc_bottom) {
1730			prom_printf("Ignoring mem=%lx <= alloc_bottom.\n",
1731				    prom_memory_limit);
1732			prom_memory_limit = 0;
1733		} else if (prom_memory_limit >= ram_top) {
1734			prom_printf("Ignoring mem=%lx >= ram_top.\n",
1735				    prom_memory_limit);
1736			prom_memory_limit = 0;
1737		} else {
1738			ram_top = prom_memory_limit;
1739			rmo_top = min(rmo_top, prom_memory_limit);
1740		}
1741	}
1742
1743	/*
1744	 * Setup our top alloc point, that is top of RMO or top of
1745	 * segment 0 when running non-LPAR.
1746	 * Some RS64 machines have buggy firmware where claims up at
1747	 * 1GB fail.  Cap at 768MB as a workaround.
1748	 * Since 768MB is plenty of room, and we need to cap to something
1749	 * reasonable on 32-bit, cap at 768MB on all machines.
1750	 */
1751	if (!rmo_top)
1752		rmo_top = ram_top;
1753	rmo_top = min(0x30000000ul, rmo_top);
1754	alloc_top = rmo_top;
1755	alloc_top_high = ram_top;
1756
1757	/*
1758	 * Check if we have an initrd after the kernel but still inside
1759	 * the RMO.  If we do move our bottom point to after it.
1760	 */
1761	if (prom_initrd_start &&
1762	    prom_initrd_start < rmo_top &&
1763	    prom_initrd_end > alloc_bottom)
1764		alloc_bottom = PAGE_ALIGN(prom_initrd_end);
1765
1766	prom_printf("memory layout at init:\n");
1767	prom_printf("  memory_limit : %lx (16 MB aligned)\n",
1768		    prom_memory_limit);
1769	prom_printf("  alloc_bottom : %lx\n", alloc_bottom);
1770	prom_printf("  alloc_top    : %lx\n", alloc_top);
1771	prom_printf("  alloc_top_hi : %lx\n", alloc_top_high);
1772	prom_printf("  rmo_top      : %lx\n", rmo_top);
1773	prom_printf("  ram_top      : %lx\n", ram_top);
1774}
1775
1776static void __init prom_close_stdin(void)
1777{
1778	__be32 val;
1779	ihandle stdin;
1780
1781	if (prom_getprop(prom.chosen, "stdin", &val, sizeof(val)) > 0) {
1782		stdin = be32_to_cpu(val);
1783		call_prom("close", 1, 0, stdin);
1784	}
1785}
1786
1787#ifdef CONFIG_PPC_SVM
1788static int prom_rtas_hcall(uint64_t args)
 
 
 
 
 
 
 
 
 
1789{
1790	register uint64_t arg1 asm("r3") = H_RTAS;
1791	register uint64_t arg2 asm("r4") = args;
 
 
 
 
1792
1793	asm volatile("sc 1\n" : "=r" (arg1) :
1794			"r" (arg1),
1795			"r" (arg2) :);
1796	srr_regs_clobbered();
1797
1798	return arg1;
1799}
 
 
1800
1801static struct rtas_args __prombss os_term_args;
 
 
 
 
 
 
 
1802
1803static void __init prom_rtas_os_term(char *str)
1804{
1805	phandle rtas_node;
1806	__be32 val;
1807	u32 token;
1808
1809	prom_debug("%s: start...\n", __func__);
1810	rtas_node = call_prom("finddevice", 1, 1, ADDR("/rtas"));
1811	prom_debug("rtas_node: %x\n", rtas_node);
1812	if (!PHANDLE_VALID(rtas_node))
1813		return;
 
1814
1815	val = 0;
1816	prom_getprop(rtas_node, "ibm,os-term", &val, sizeof(val));
1817	token = be32_to_cpu(val);
1818	prom_debug("ibm,os-term: %x\n", token);
1819	if (token == 0)
1820		prom_panic("Could not get token for ibm,os-term\n");
1821	os_term_args.token = cpu_to_be32(token);
1822	os_term_args.nargs = cpu_to_be32(1);
1823	os_term_args.nret = cpu_to_be32(1);
1824	os_term_args.args[0] = cpu_to_be32(__pa(str));
1825	prom_rtas_hcall((uint64_t)&os_term_args);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1826}
1827#endif /* CONFIG_PPC_SVM */
 
1828
1829/*
1830 * Allocate room for and instantiate RTAS
1831 */
1832static void __init prom_instantiate_rtas(void)
1833{
1834	phandle rtas_node;
1835	ihandle rtas_inst;
1836	u32 base, entry = 0;
1837	__be32 val;
1838	u32 size = 0;
1839
1840	prom_debug("prom_instantiate_rtas: start...\n");
1841
1842	rtas_node = call_prom("finddevice", 1, 1, ADDR("/rtas"));
1843	prom_debug("rtas_node: %x\n", rtas_node);
1844	if (!PHANDLE_VALID(rtas_node))
1845		return;
1846
1847	val = 0;
1848	prom_getprop(rtas_node, "rtas-size", &val, sizeof(size));
1849	size = be32_to_cpu(val);
1850	if (size == 0)
1851		return;
1852
1853	base = alloc_down(size, PAGE_SIZE, 0);
1854	if (base == 0)
1855		prom_panic("Could not allocate memory for RTAS\n");
1856
1857	rtas_inst = call_prom("open", 1, 1, ADDR("/rtas"));
1858	if (!IHANDLE_VALID(rtas_inst)) {
1859		prom_printf("opening rtas package failed (%x)\n", rtas_inst);
1860		return;
1861	}
1862
1863	prom_printf("instantiating rtas at 0x%x...", base);
1864
1865	if (call_prom_ret("call-method", 3, 2, &entry,
1866			  ADDR("instantiate-rtas"),
1867			  rtas_inst, base) != 0
1868	    || entry == 0) {
1869		prom_printf(" failed\n");
1870		return;
1871	}
1872	prom_printf(" done\n");
1873
1874	reserve_mem(base, size);
1875
1876	val = cpu_to_be32(base);
1877	prom_setprop(rtas_node, "/rtas", "linux,rtas-base",
1878		     &val, sizeof(val));
1879	val = cpu_to_be32(entry);
1880	prom_setprop(rtas_node, "/rtas", "linux,rtas-entry",
1881		     &val, sizeof(val));
1882
1883	/* Check if it supports "query-cpu-stopped-state" */
1884	if (prom_getprop(rtas_node, "query-cpu-stopped-state",
1885			 &val, sizeof(val)) != PROM_ERROR)
1886		rtas_has_query_cpu_stopped = true;
1887
1888	prom_debug("rtas base     = 0x%x\n", base);
1889	prom_debug("rtas entry    = 0x%x\n", entry);
1890	prom_debug("rtas size     = 0x%x\n", size);
1891
1892	prom_debug("prom_instantiate_rtas: end...\n");
1893}
1894
1895#ifdef CONFIG_PPC64
1896/*
1897 * Allocate room for and instantiate Stored Measurement Log (SML)
1898 */
1899static void __init prom_instantiate_sml(void)
1900{
1901	phandle ibmvtpm_node;
1902	ihandle ibmvtpm_inst;
1903	u32 entry = 0, size = 0, succ = 0;
1904	u64 base;
1905	__be32 val;
1906
1907	prom_debug("prom_instantiate_sml: start...\n");
1908
1909	ibmvtpm_node = call_prom("finddevice", 1, 1, ADDR("/vdevice/vtpm"));
1910	prom_debug("ibmvtpm_node: %x\n", ibmvtpm_node);
1911	if (!PHANDLE_VALID(ibmvtpm_node))
1912		return;
1913
1914	ibmvtpm_inst = call_prom("open", 1, 1, ADDR("/vdevice/vtpm"));
1915	if (!IHANDLE_VALID(ibmvtpm_inst)) {
1916		prom_printf("opening vtpm package failed (%x)\n", ibmvtpm_inst);
1917		return;
1918	}
1919
1920	if (prom_getprop(ibmvtpm_node, "ibm,sml-efi-reformat-supported",
1921			 &val, sizeof(val)) != PROM_ERROR) {
1922		if (call_prom_ret("call-method", 2, 2, &succ,
1923				  ADDR("reformat-sml-to-efi-alignment"),
1924				  ibmvtpm_inst) != 0 || succ == 0) {
1925			prom_printf("Reformat SML to EFI alignment failed\n");
1926			return;
1927		}
1928
1929		if (call_prom_ret("call-method", 2, 2, &size,
1930				  ADDR("sml-get-allocated-size"),
1931				  ibmvtpm_inst) != 0 || size == 0) {
1932			prom_printf("SML get allocated size failed\n");
1933			return;
1934		}
1935	} else {
1936		if (call_prom_ret("call-method", 2, 2, &size,
1937				  ADDR("sml-get-handover-size"),
1938				  ibmvtpm_inst) != 0 || size == 0) {
1939			prom_printf("SML get handover size failed\n");
1940			return;
1941		}
1942	}
1943
1944	base = alloc_down(size, PAGE_SIZE, 0);
1945	if (base == 0)
1946		prom_panic("Could not allocate memory for sml\n");
1947
1948	prom_printf("instantiating sml at 0x%llx...", base);
1949
1950	memset((void *)base, 0, size);
1951
1952	if (call_prom_ret("call-method", 4, 2, &entry,
1953			  ADDR("sml-handover"),
1954			  ibmvtpm_inst, size, base) != 0 || entry == 0) {
1955		prom_printf("SML handover failed\n");
1956		return;
1957	}
1958	prom_printf(" done\n");
1959
1960	reserve_mem(base, size);
1961
1962	prom_setprop(ibmvtpm_node, "/vdevice/vtpm", "linux,sml-base",
1963		     &base, sizeof(base));
1964	prom_setprop(ibmvtpm_node, "/vdevice/vtpm", "linux,sml-size",
1965		     &size, sizeof(size));
1966
1967	prom_debug("sml base     = 0x%llx\n", base);
1968	prom_debug("sml size     = 0x%x\n", size);
1969
1970	prom_debug("prom_instantiate_sml: end...\n");
1971}
1972
1973/*
1974 * Allocate room for and initialize TCE tables
1975 */
1976#ifdef __BIG_ENDIAN__
1977static void __init prom_initialize_tce_table(void)
1978{
1979	phandle node;
1980	ihandle phb_node;
1981	char compatible[64], type[64], model[64];
1982	char *path = prom_scratch;
1983	u64 base, align;
1984	u32 minalign, minsize;
1985	u64 tce_entry, *tce_entryp;
1986	u64 local_alloc_top, local_alloc_bottom;
1987	u64 i;
1988
1989	if (prom_iommu_off)
1990		return;
1991
1992	prom_debug("starting prom_initialize_tce_table\n");
1993
1994	/* Cache current top of allocs so we reserve a single block */
1995	local_alloc_top = alloc_top_high;
1996	local_alloc_bottom = local_alloc_top;
1997
1998	/* Search all nodes looking for PHBs. */
1999	for (node = 0; prom_next_node(&node); ) {
2000		compatible[0] = 0;
2001		type[0] = 0;
2002		model[0] = 0;
2003		prom_getprop(node, "compatible",
2004			     compatible, sizeof(compatible));
2005		prom_getprop(node, "device_type", type, sizeof(type));
2006		prom_getprop(node, "model", model, sizeof(model));
2007
2008		if ((type[0] == 0) || (prom_strstr(type, "pci") == NULL))
2009			continue;
2010
2011		/* Keep the old logic intact to avoid regression. */
2012		if (compatible[0] != 0) {
2013			if ((prom_strstr(compatible, "python") == NULL) &&
2014			    (prom_strstr(compatible, "Speedwagon") == NULL) &&
2015			    (prom_strstr(compatible, "Winnipeg") == NULL))
2016				continue;
2017		} else if (model[0] != 0) {
2018			if ((prom_strstr(model, "ython") == NULL) &&
2019			    (prom_strstr(model, "peedwagon") == NULL) &&
2020			    (prom_strstr(model, "innipeg") == NULL))
2021				continue;
2022		}
2023
2024		if (prom_getprop(node, "tce-table-minalign", &minalign,
2025				 sizeof(minalign)) == PROM_ERROR)
2026			minalign = 0;
2027		if (prom_getprop(node, "tce-table-minsize", &minsize,
2028				 sizeof(minsize)) == PROM_ERROR)
2029			minsize = 4UL << 20;
2030
2031		/*
2032		 * Even though we read what OF wants, we just set the table
2033		 * size to 4 MB.  This is enough to map 2GB of PCI DMA space.
2034		 * By doing this, we avoid the pitfalls of trying to DMA to
2035		 * MMIO space and the DMA alias hole.
 
 
 
 
 
2036		 */
2037		minsize = 4UL << 20;
 
 
 
2038
2039		/* Align to the greater of the align or size */
2040		align = max(minalign, minsize);
2041		base = alloc_down(minsize, align, 1);
2042		if (base == 0)
2043			prom_panic("ERROR, cannot find space for TCE table.\n");
2044		if (base < local_alloc_bottom)
2045			local_alloc_bottom = base;
2046
2047		/* It seems OF doesn't null-terminate the path :-( */
2048		memset(path, 0, sizeof(prom_scratch));
2049		/* Call OF to setup the TCE hardware */
2050		if (call_prom("package-to-path", 3, 1, node,
2051			      path, sizeof(prom_scratch) - 1) == PROM_ERROR) {
2052			prom_printf("package-to-path failed\n");
2053		}
2054
2055		/* Save away the TCE table attributes for later use. */
2056		prom_setprop(node, path, "linux,tce-base", &base, sizeof(base));
2057		prom_setprop(node, path, "linux,tce-size", &minsize, sizeof(minsize));
2058
2059		prom_debug("TCE table: %s\n", path);
2060		prom_debug("\tnode = 0x%x\n", node);
2061		prom_debug("\tbase = 0x%llx\n", base);
2062		prom_debug("\tsize = 0x%x\n", minsize);
2063
2064		/* Initialize the table to have a one-to-one mapping
2065		 * over the allocated size.
2066		 */
2067		tce_entryp = (u64 *)base;
2068		for (i = 0; i < (minsize >> 3) ;tce_entryp++, i++) {
2069			tce_entry = (i << PAGE_SHIFT);
2070			tce_entry |= 0x3;
2071			*tce_entryp = tce_entry;
2072		}
2073
2074		prom_printf("opening PHB %s", path);
2075		phb_node = call_prom("open", 1, 1, path);
2076		if (phb_node == 0)
2077			prom_printf("... failed\n");
2078		else
2079			prom_printf("... done\n");
2080
2081		call_prom("call-method", 6, 0, ADDR("set-64-bit-addressing"),
2082			  phb_node, -1, minsize,
2083			  (u32) base, (u32) (base >> 32));
2084		call_prom("close", 1, 0, phb_node);
2085	}
2086
2087	reserve_mem(local_alloc_bottom, local_alloc_top - local_alloc_bottom);
2088
2089	/* These are only really needed if there is a memory limit in
2090	 * effect, but we don't know so export them always. */
2091	prom_tce_alloc_start = local_alloc_bottom;
2092	prom_tce_alloc_end = local_alloc_top;
2093
2094	/* Flag the first invalid entry */
2095	prom_debug("ending prom_initialize_tce_table\n");
2096}
2097#endif /* __BIG_ENDIAN__ */
2098#endif /* CONFIG_PPC64 */
2099
2100/*
2101 * With CHRP SMP we need to use the OF to start the other processors.
2102 * We can't wait until smp_boot_cpus (the OF is trashed by then)
2103 * so we have to put the processors into a holding pattern controlled
2104 * by the kernel (not OF) before we destroy the OF.
2105 *
2106 * This uses a chunk of low memory, puts some holding pattern
2107 * code there and sends the other processors off to there until
2108 * smp_boot_cpus tells them to do something.  The holding pattern
2109 * checks that address until its cpu # is there, when it is that
2110 * cpu jumps to __secondary_start().  smp_boot_cpus() takes care
2111 * of setting those values.
2112 *
2113 * We also use physical address 0x4 here to tell when a cpu
2114 * is in its holding pattern code.
2115 *
2116 * -- Cort
2117 */
2118/*
2119 * We want to reference the copy of __secondary_hold_* in the
2120 * 0 - 0x100 address range
2121 */
2122#define LOW_ADDR(x)	(((unsigned long) &(x)) & 0xff)
2123
2124static void __init prom_hold_cpus(void)
2125{
2126	unsigned long i;
2127	phandle node;
2128	char type[64];
2129	unsigned long *spinloop
2130		= (void *) LOW_ADDR(__secondary_hold_spinloop);
2131	unsigned long *acknowledge
2132		= (void *) LOW_ADDR(__secondary_hold_acknowledge);
2133	unsigned long secondary_hold = LOW_ADDR(__secondary_hold);
2134
2135	/*
2136	 * On pseries, if RTAS supports "query-cpu-stopped-state",
2137	 * we skip this stage, the CPUs will be started by the
2138	 * kernel using RTAS.
2139	 */
2140	if ((of_platform == PLATFORM_PSERIES ||
2141	     of_platform == PLATFORM_PSERIES_LPAR) &&
2142	    rtas_has_query_cpu_stopped) {
2143		prom_printf("prom_hold_cpus: skipped\n");
2144		return;
2145	}
2146
2147	prom_debug("prom_hold_cpus: start...\n");
2148	prom_debug("    1) spinloop       = 0x%lx\n", (unsigned long)spinloop);
2149	prom_debug("    1) *spinloop      = 0x%lx\n", *spinloop);
2150	prom_debug("    1) acknowledge    = 0x%lx\n",
2151		   (unsigned long)acknowledge);
2152	prom_debug("    1) *acknowledge   = 0x%lx\n", *acknowledge);
2153	prom_debug("    1) secondary_hold = 0x%lx\n", secondary_hold);
2154
2155	/* Set the common spinloop variable, so all of the secondary cpus
2156	 * will block when they are awakened from their OF spinloop.
2157	 * This must occur for both SMP and non SMP kernels, since OF will
2158	 * be trashed when we move the kernel.
2159	 */
2160	*spinloop = 0;
2161
2162	/* look for cpus */
2163	for (node = 0; prom_next_node(&node); ) {
2164		unsigned int cpu_no;
2165		__be32 reg;
2166
2167		type[0] = 0;
2168		prom_getprop(node, "device_type", type, sizeof(type));
2169		if (prom_strcmp(type, "cpu") != 0)
2170			continue;
2171
2172		/* Skip non-configured cpus. */
2173		if (prom_getprop(node, "status", type, sizeof(type)) > 0)
2174			if (prom_strcmp(type, "okay") != 0)
2175				continue;
2176
2177		reg = cpu_to_be32(-1); /* make sparse happy */
2178		prom_getprop(node, "reg", &reg, sizeof(reg));
2179		cpu_no = be32_to_cpu(reg);
2180
2181		prom_debug("cpu hw idx   = %u\n", cpu_no);
2182
2183		/* Init the acknowledge var which will be reset by
2184		 * the secondary cpu when it awakens from its OF
2185		 * spinloop.
2186		 */
2187		*acknowledge = (unsigned long)-1;
2188
2189		if (cpu_no != prom.cpu) {
2190			/* Primary Thread of non-boot cpu or any thread */
2191			prom_printf("starting cpu hw idx %u... ", cpu_no);
2192			call_prom("start-cpu", 3, 0, node,
2193				  secondary_hold, cpu_no);
2194
2195			for (i = 0; (i < 100000000) && 
2196			     (*acknowledge == ((unsigned long)-1)); i++ )
2197				mb();
2198
2199			if (*acknowledge == cpu_no)
2200				prom_printf("done\n");
2201			else
2202				prom_printf("failed: %lx\n", *acknowledge);
2203		}
2204#ifdef CONFIG_SMP
2205		else
2206			prom_printf("boot cpu hw idx %u\n", cpu_no);
2207#endif /* CONFIG_SMP */
2208	}
2209
2210	prom_debug("prom_hold_cpus: end...\n");
2211}
2212
2213
2214static void __init prom_init_client_services(unsigned long pp)
2215{
2216	/* Get a handle to the prom entry point before anything else */
2217	prom_entry = pp;
2218
2219	/* get a handle for the stdout device */
2220	prom.chosen = call_prom("finddevice", 1, 1, ADDR("/chosen"));
2221	if (!PHANDLE_VALID(prom.chosen))
2222		prom_panic("cannot find chosen"); /* msg won't be printed :( */
2223
2224	/* get device tree root */
2225	prom.root = call_prom("finddevice", 1, 1, ADDR("/"));
2226	if (!PHANDLE_VALID(prom.root))
2227		prom_panic("cannot find device tree root"); /* msg won't be printed :( */
2228
2229	prom.mmumap = 0;
2230}
2231
2232#ifdef CONFIG_PPC32
2233/*
2234 * For really old powermacs, we need to map things we claim.
2235 * For that, we need the ihandle of the mmu.
2236 * Also, on the longtrail, we need to work around other bugs.
2237 */
2238static void __init prom_find_mmu(void)
2239{
2240	phandle oprom;
2241	char version[64];
2242
2243	oprom = call_prom("finddevice", 1, 1, ADDR("/openprom"));
2244	if (!PHANDLE_VALID(oprom))
2245		return;
2246	if (prom_getprop(oprom, "model", version, sizeof(version)) <= 0)
2247		return;
2248	version[sizeof(version) - 1] = 0;
2249	/* XXX might need to add other versions here */
2250	if (prom_strcmp(version, "Open Firmware, 1.0.5") == 0)
2251		of_workarounds = OF_WA_CLAIM;
2252	else if (prom_strncmp(version, "FirmWorks,3.", 12) == 0) {
2253		of_workarounds = OF_WA_CLAIM | OF_WA_LONGTRAIL;
2254		call_prom("interpret", 1, 1, "dev /memory 0 to allow-reclaim");
2255	} else
2256		return;
2257	prom.memory = call_prom("open", 1, 1, ADDR("/memory"));
2258	prom_getprop(prom.chosen, "mmu", &prom.mmumap,
2259		     sizeof(prom.mmumap));
2260	prom.mmumap = be32_to_cpu(prom.mmumap);
2261	if (!IHANDLE_VALID(prom.memory) || !IHANDLE_VALID(prom.mmumap))
2262		of_workarounds &= ~OF_WA_CLAIM;		/* hmmm */
2263}
2264#else
2265#define prom_find_mmu()
2266#endif
2267
2268static void __init prom_init_stdout(void)
2269{
2270	char *path = of_stdout_device;
2271	char type[16];
2272	phandle stdout_node;
2273	__be32 val;
2274
2275	if (prom_getprop(prom.chosen, "stdout", &val, sizeof(val)) <= 0)
2276		prom_panic("cannot find stdout");
2277
2278	prom.stdout = be32_to_cpu(val);
2279
2280	/* Get the full OF pathname of the stdout device */
2281	memset(path, 0, 256);
2282	call_prom("instance-to-path", 3, 1, prom.stdout, path, 255);
2283	prom_printf("OF stdout device is: %s\n", of_stdout_device);
2284	prom_setprop(prom.chosen, "/chosen", "linux,stdout-path",
2285		     path, prom_strlen(path) + 1);
2286
2287	/* instance-to-package fails on PA-Semi */
2288	stdout_node = call_prom("instance-to-package", 1, 1, prom.stdout);
2289	if (stdout_node != PROM_ERROR) {
2290		val = cpu_to_be32(stdout_node);
 
 
2291
2292		/* If it's a display, note it */
2293		memset(type, 0, sizeof(type));
2294		prom_getprop(stdout_node, "device_type", type, sizeof(type));
2295		if (prom_strcmp(type, "display") == 0)
2296			prom_setprop(stdout_node, path, "linux,boot-display", NULL, 0);
2297	}
2298}
2299
2300static int __init prom_find_machine_type(void)
2301{
2302	char compat[256];
2303	int len, i = 0;
2304#ifdef CONFIG_PPC64
2305	phandle rtas;
2306	int x;
2307#endif
2308
2309	/* Look for a PowerMac or a Cell */
2310	len = prom_getprop(prom.root, "compatible",
2311			   compat, sizeof(compat)-1);
2312	if (len > 0) {
2313		compat[len] = 0;
2314		while (i < len) {
2315			char *p = &compat[i];
2316			int sl = prom_strlen(p);
2317			if (sl == 0)
2318				break;
2319			if (prom_strstr(p, "Power Macintosh") ||
2320			    prom_strstr(p, "MacRISC"))
2321				return PLATFORM_POWERMAC;
2322#ifdef CONFIG_PPC64
2323			/* We must make sure we don't detect the IBM Cell
2324			 * blades as pSeries due to some firmware issues,
2325			 * so we do it here.
2326			 */
2327			if (prom_strstr(p, "IBM,CBEA") ||
2328			    prom_strstr(p, "IBM,CPBW-1.0"))
2329				return PLATFORM_GENERIC;
2330#endif /* CONFIG_PPC64 */
2331			i += sl + 1;
2332		}
2333	}
2334#ifdef CONFIG_PPC64
 
 
 
 
2335	/* Try to figure out if it's an IBM pSeries or any other
2336	 * PAPR compliant platform. We assume it is if :
2337	 *  - /device_type is "chrp" (please, do NOT use that for future
2338	 *    non-IBM designs !
2339	 *  - it has /rtas
2340	 */
2341	len = prom_getprop(prom.root, "device_type",
2342			   compat, sizeof(compat)-1);
2343	if (len <= 0)
2344		return PLATFORM_GENERIC;
2345	if (prom_strcmp(compat, "chrp"))
2346		return PLATFORM_GENERIC;
2347
2348	/* Default to pSeries. We need to know if we are running LPAR */
2349	rtas = call_prom("finddevice", 1, 1, ADDR("/rtas"));
2350	if (!PHANDLE_VALID(rtas))
2351		return PLATFORM_GENERIC;
2352	x = prom_getproplen(rtas, "ibm,hypertas-functions");
2353	if (x != PROM_ERROR) {
2354		prom_debug("Hypertas detected, assuming LPAR !\n");
2355		return PLATFORM_PSERIES_LPAR;
2356	}
2357	return PLATFORM_PSERIES;
2358#else
2359	return PLATFORM_GENERIC;
2360#endif
2361}
2362
2363static int __init prom_set_color(ihandle ih, int i, int r, int g, int b)
2364{
2365	return call_prom("call-method", 6, 1, ADDR("color!"), ih, i, b, g, r);
2366}
2367
2368/*
2369 * If we have a display that we don't know how to drive,
2370 * we will want to try to execute OF's open method for it
2371 * later.  However, OF will probably fall over if we do that
2372 * we've taken over the MMU.
2373 * So we check whether we will need to open the display,
2374 * and if so, open it now.
2375 */
2376static void __init prom_check_displays(void)
2377{
2378	char type[16], *path;
2379	phandle node;
2380	ihandle ih;
2381	int i;
2382
2383	static const unsigned char default_colors[] __initconst = {
2384		0x00, 0x00, 0x00,
2385		0x00, 0x00, 0xaa,
2386		0x00, 0xaa, 0x00,
2387		0x00, 0xaa, 0xaa,
2388		0xaa, 0x00, 0x00,
2389		0xaa, 0x00, 0xaa,
2390		0xaa, 0xaa, 0x00,
2391		0xaa, 0xaa, 0xaa,
2392		0x55, 0x55, 0x55,
2393		0x55, 0x55, 0xff,
2394		0x55, 0xff, 0x55,
2395		0x55, 0xff, 0xff,
2396		0xff, 0x55, 0x55,
2397		0xff, 0x55, 0xff,
2398		0xff, 0xff, 0x55,
2399		0xff, 0xff, 0xff
2400	};
2401	const unsigned char *clut;
2402
2403	prom_debug("Looking for displays\n");
2404	for (node = 0; prom_next_node(&node); ) {
2405		memset(type, 0, sizeof(type));
2406		prom_getprop(node, "device_type", type, sizeof(type));
2407		if (prom_strcmp(type, "display") != 0)
2408			continue;
2409
2410		/* It seems OF doesn't null-terminate the path :-( */
2411		path = prom_scratch;
2412		memset(path, 0, sizeof(prom_scratch));
2413
2414		/*
2415		 * leave some room at the end of the path for appending extra
2416		 * arguments
2417		 */
2418		if (call_prom("package-to-path", 3, 1, node, path,
2419			      sizeof(prom_scratch) - 10) == PROM_ERROR)
2420			continue;
2421		prom_printf("found display   : %s, opening... ", path);
2422		
2423		ih = call_prom("open", 1, 1, path);
2424		if (ih == 0) {
2425			prom_printf("failed\n");
2426			continue;
2427		}
2428
2429		/* Success */
2430		prom_printf("done\n");
2431		prom_setprop(node, path, "linux,opened", NULL, 0);
2432
2433		/* Setup a usable color table when the appropriate
2434		 * method is available. Should update this to set-colors */
2435		clut = default_colors;
2436		for (i = 0; i < 16; i++, clut += 3)
2437			if (prom_set_color(ih, i, clut[0], clut[1],
2438					   clut[2]) != 0)
2439				break;
2440
2441#ifdef CONFIG_LOGO_LINUX_CLUT224
2442		clut = PTRRELOC(logo_linux_clut224.clut);
2443		for (i = 0; i < logo_linux_clut224.clutsize; i++, clut += 3)
2444			if (prom_set_color(ih, i + 32, clut[0], clut[1],
2445					   clut[2]) != 0)
2446				break;
2447#endif /* CONFIG_LOGO_LINUX_CLUT224 */
2448
2449#ifdef CONFIG_PPC_EARLY_DEBUG_BOOTX
2450		if (prom_getprop(node, "linux,boot-display", NULL, 0) !=
2451		    PROM_ERROR) {
2452			u32 width, height, pitch, addr;
2453
2454			prom_printf("Setting btext !\n");
2455
2456			if (prom_getprop(node, "width", &width, 4) == PROM_ERROR)
2457				return;
2458
2459			if (prom_getprop(node, "height", &height, 4) == PROM_ERROR)
2460				return;
2461
2462			if (prom_getprop(node, "linebytes", &pitch, 4) == PROM_ERROR)
2463				return;
2464
2465			if (prom_getprop(node, "address", &addr, 4) == PROM_ERROR)
2466				return;
2467
2468			prom_printf("W=%d H=%d LB=%d addr=0x%x\n",
2469				    width, height, pitch, addr);
2470			btext_setup_display(width, height, 8, pitch, addr);
2471			btext_prepare_BAT();
2472		}
2473#endif /* CONFIG_PPC_EARLY_DEBUG_BOOTX */
2474	}
2475}
2476
2477
2478/* Return (relocated) pointer to this much memory: moves initrd if reqd. */
2479static void __init *make_room(unsigned long *mem_start, unsigned long *mem_end,
2480			      unsigned long needed, unsigned long align)
2481{
2482	void *ret;
2483
2484	*mem_start = ALIGN(*mem_start, align);
2485	while ((*mem_start + needed) > *mem_end) {
2486		unsigned long room, chunk;
2487
2488		prom_debug("Chunk exhausted, claiming more at %lx...\n",
2489			   alloc_bottom);
2490		room = alloc_top - alloc_bottom;
2491		if (room > DEVTREE_CHUNK_SIZE)
2492			room = DEVTREE_CHUNK_SIZE;
2493		if (room < PAGE_SIZE)
2494			prom_panic("No memory for flatten_device_tree "
2495				   "(no room)\n");
2496		chunk = alloc_up(room, 0);
2497		if (chunk == 0)
2498			prom_panic("No memory for flatten_device_tree "
2499				   "(claim failed)\n");
2500		*mem_end = chunk + room;
2501	}
2502
2503	ret = (void *)*mem_start;
2504	*mem_start += needed;
2505
2506	return ret;
2507}
2508
2509#define dt_push_token(token, mem_start, mem_end) do { 			\
2510		void *room = make_room(mem_start, mem_end, 4, 4);	\
2511		*(__be32 *)room = cpu_to_be32(token);			\
2512	} while(0)
2513
2514static unsigned long __init dt_find_string(char *str)
2515{
2516	char *s, *os;
2517
2518	s = os = (char *)dt_string_start;
2519	s += 4;
2520	while (s <  (char *)dt_string_end) {
2521		if (prom_strcmp(s, str) == 0)
2522			return s - os;
2523		s += prom_strlen(s) + 1;
2524	}
2525	return 0;
2526}
2527
2528/*
2529 * The Open Firmware 1275 specification states properties must be 31 bytes or
2530 * less, however not all firmwares obey this. Make it 64 bytes to be safe.
2531 */
2532#define MAX_PROPERTY_NAME 64
2533
2534static void __init scan_dt_build_strings(phandle node,
2535					 unsigned long *mem_start,
2536					 unsigned long *mem_end)
2537{
2538	char *prev_name, *namep, *sstart;
2539	unsigned long soff;
2540	phandle child;
2541
2542	sstart =  (char *)dt_string_start;
2543
2544	/* get and store all property names */
2545	prev_name = "";
2546	for (;;) {
2547		/* 64 is max len of name including nul. */
2548		namep = make_room(mem_start, mem_end, MAX_PROPERTY_NAME, 1);
2549		if (call_prom("nextprop", 3, 1, node, prev_name, namep) != 1) {
2550			/* No more nodes: unwind alloc */
2551			*mem_start = (unsigned long)namep;
2552			break;
2553		}
2554
2555 		/* skip "name" */
2556		if (prom_strcmp(namep, "name") == 0) {
2557 			*mem_start = (unsigned long)namep;
2558 			prev_name = "name";
2559 			continue;
2560 		}
2561		/* get/create string entry */
2562		soff = dt_find_string(namep);
2563		if (soff != 0) {
2564			*mem_start = (unsigned long)namep;
2565			namep = sstart + soff;
2566		} else {
2567			/* Trim off some if we can */
2568			*mem_start = (unsigned long)namep + prom_strlen(namep) + 1;
2569			dt_string_end = *mem_start;
2570		}
2571		prev_name = namep;
2572	}
2573
2574	/* do all our children */
2575	child = call_prom("child", 1, 1, node);
2576	while (child != 0) {
2577		scan_dt_build_strings(child, mem_start, mem_end);
2578		child = call_prom("peer", 1, 1, child);
2579	}
2580}
2581
2582static void __init scan_dt_build_struct(phandle node, unsigned long *mem_start,
2583					unsigned long *mem_end)
2584{
2585	phandle child;
2586	char *namep, *prev_name, *sstart, *p, *ep, *lp, *path;
2587	unsigned long soff;
2588	unsigned char *valp;
2589	static char pname[MAX_PROPERTY_NAME] __prombss;
2590	int l, room, has_phandle = 0;
2591
2592	dt_push_token(OF_DT_BEGIN_NODE, mem_start, mem_end);
2593
2594	/* get the node's full name */
2595	namep = (char *)*mem_start;
2596	room = *mem_end - *mem_start;
2597	if (room > 255)
2598		room = 255;
2599	l = call_prom("package-to-path", 3, 1, node, namep, room);
2600	if (l >= 0) {
2601		/* Didn't fit?  Get more room. */
2602		if (l >= room) {
2603			if (l >= *mem_end - *mem_start)
2604				namep = make_room(mem_start, mem_end, l+1, 1);
2605			call_prom("package-to-path", 3, 1, node, namep, l);
2606		}
2607		namep[l] = '\0';
2608
2609		/* Fixup an Apple bug where they have bogus \0 chars in the
2610		 * middle of the path in some properties, and extract
2611		 * the unit name (everything after the last '/').
2612		 */
2613		for (lp = p = namep, ep = namep + l; p < ep; p++) {
2614			if (*p == '/')
2615				lp = namep;
2616			else if (*p != 0)
2617				*lp++ = *p;
2618		}
2619		*lp = 0;
2620		*mem_start = ALIGN((unsigned long)lp + 1, 4);
2621	}
2622
2623	/* get it again for debugging */
2624	path = prom_scratch;
2625	memset(path, 0, sizeof(prom_scratch));
2626	call_prom("package-to-path", 3, 1, node, path, sizeof(prom_scratch) - 1);
2627
2628	/* get and store all properties */
2629	prev_name = "";
2630	sstart = (char *)dt_string_start;
2631	for (;;) {
2632		if (call_prom("nextprop", 3, 1, node, prev_name,
2633			      pname) != 1)
2634			break;
2635
2636 		/* skip "name" */
2637		if (prom_strcmp(pname, "name") == 0) {
2638 			prev_name = "name";
2639 			continue;
2640 		}
2641
2642		/* find string offset */
2643		soff = dt_find_string(pname);
2644		if (soff == 0) {
2645			prom_printf("WARNING: Can't find string index for"
2646				    " <%s>, node %s\n", pname, path);
2647			break;
2648		}
2649		prev_name = sstart + soff;
2650
2651		/* get length */
2652		l = call_prom("getproplen", 2, 1, node, pname);
2653
2654		/* sanity checks */
2655		if (l == PROM_ERROR)
2656			continue;
2657
2658		/* push property head */
2659		dt_push_token(OF_DT_PROP, mem_start, mem_end);
2660		dt_push_token(l, mem_start, mem_end);
2661		dt_push_token(soff, mem_start, mem_end);
2662
2663		/* push property content */
2664		valp = make_room(mem_start, mem_end, l, 4);
2665		call_prom("getprop", 4, 1, node, pname, valp, l);
2666		*mem_start = ALIGN(*mem_start, 4);
2667
2668		if (!prom_strcmp(pname, "phandle"))
2669			has_phandle = 1;
2670	}
2671
2672	/* Add a "phandle" property if none already exist */
 
 
2673	if (!has_phandle) {
2674		soff = dt_find_string("phandle");
2675		if (soff == 0)
2676			prom_printf("WARNING: Can't find string index for <phandle> node %s\n", path);
 
2677		else {
2678			dt_push_token(OF_DT_PROP, mem_start, mem_end);
2679			dt_push_token(4, mem_start, mem_end);
2680			dt_push_token(soff, mem_start, mem_end);
2681			valp = make_room(mem_start, mem_end, 4, 4);
2682			*(__be32 *)valp = cpu_to_be32(node);
2683		}
2684	}
2685
2686	/* do all our children */
2687	child = call_prom("child", 1, 1, node);
2688	while (child != 0) {
2689		scan_dt_build_struct(child, mem_start, mem_end);
2690		child = call_prom("peer", 1, 1, child);
2691	}
2692
2693	dt_push_token(OF_DT_END_NODE, mem_start, mem_end);
2694}
2695
2696static void __init flatten_device_tree(void)
2697{
2698	phandle root;
2699	unsigned long mem_start, mem_end, room;
2700	struct boot_param_header *hdr;
2701	char *namep;
2702	u64 *rsvmap;
2703
2704	/*
2705	 * Check how much room we have between alloc top & bottom (+/- a
2706	 * few pages), crop to 1MB, as this is our "chunk" size
2707	 */
2708	room = alloc_top - alloc_bottom - 0x4000;
2709	if (room > DEVTREE_CHUNK_SIZE)
2710		room = DEVTREE_CHUNK_SIZE;
2711	prom_debug("starting device tree allocs at %lx\n", alloc_bottom);
2712
2713	/* Now try to claim that */
2714	mem_start = (unsigned long)alloc_up(room, PAGE_SIZE);
2715	if (mem_start == 0)
2716		prom_panic("Can't allocate initial device-tree chunk\n");
2717	mem_end = mem_start + room;
2718
2719	/* Get root of tree */
2720	root = call_prom("peer", 1, 1, (phandle)0);
2721	if (root == (phandle)0)
2722		prom_panic ("couldn't get device tree root\n");
2723
2724	/* Build header and make room for mem rsv map */ 
2725	mem_start = ALIGN(mem_start, 4);
2726	hdr = make_room(&mem_start, &mem_end,
2727			sizeof(struct boot_param_header), 4);
2728	dt_header_start = (unsigned long)hdr;
2729	rsvmap = make_room(&mem_start, &mem_end, sizeof(mem_reserve_map), 8);
2730
2731	/* Start of strings */
2732	mem_start = PAGE_ALIGN(mem_start);
2733	dt_string_start = mem_start;
2734	mem_start += 4; /* hole */
2735
2736	/* Add "phandle" in there, we'll need it */
2737	namep = make_room(&mem_start, &mem_end, 16, 1);
2738	prom_strscpy_pad(namep, "phandle", sizeof("phandle"));
2739	mem_start = (unsigned long)namep + prom_strlen(namep) + 1;
2740
2741	/* Build string array */
2742	prom_printf("Building dt strings...\n"); 
2743	scan_dt_build_strings(root, &mem_start, &mem_end);
2744	dt_string_end = mem_start;
2745
2746	/* Build structure */
2747	mem_start = PAGE_ALIGN(mem_start);
2748	dt_struct_start = mem_start;
2749	prom_printf("Building dt structure...\n"); 
2750	scan_dt_build_struct(root, &mem_start, &mem_end);
2751	dt_push_token(OF_DT_END, &mem_start, &mem_end);
2752	dt_struct_end = PAGE_ALIGN(mem_start);
2753
2754	/* Finish header */
2755	hdr->boot_cpuid_phys = cpu_to_be32(prom.cpu);
2756	hdr->magic = cpu_to_be32(OF_DT_HEADER);
2757	hdr->totalsize = cpu_to_be32(dt_struct_end - dt_header_start);
2758	hdr->off_dt_struct = cpu_to_be32(dt_struct_start - dt_header_start);
2759	hdr->off_dt_strings = cpu_to_be32(dt_string_start - dt_header_start);
2760	hdr->dt_strings_size = cpu_to_be32(dt_string_end - dt_string_start);
2761	hdr->off_mem_rsvmap = cpu_to_be32(((unsigned long)rsvmap) - dt_header_start);
2762	hdr->version = cpu_to_be32(OF_DT_VERSION);
2763	/* Version 16 is not backward compatible */
2764	hdr->last_comp_version = cpu_to_be32(0x10);
2765
2766	/* Copy the reserve map in */
2767	memcpy(rsvmap, mem_reserve_map, sizeof(mem_reserve_map));
2768
2769#ifdef DEBUG_PROM
2770	{
2771		int i;
2772		prom_printf("reserved memory map:\n");
2773		for (i = 0; i < mem_reserve_cnt; i++)
2774			prom_printf("  %llx - %llx\n",
2775				    be64_to_cpu(mem_reserve_map[i].base),
2776				    be64_to_cpu(mem_reserve_map[i].size));
2777	}
2778#endif
2779	/* Bump mem_reserve_cnt to cause further reservations to fail
2780	 * since it's too late.
2781	 */
2782	mem_reserve_cnt = MEM_RESERVE_MAP_SIZE;
2783
2784	prom_printf("Device tree strings 0x%lx -> 0x%lx\n",
2785		    dt_string_start, dt_string_end);
2786	prom_printf("Device tree struct  0x%lx -> 0x%lx\n",
2787		    dt_struct_start, dt_struct_end);
2788}
2789
2790#ifdef CONFIG_PPC_MAPLE
2791/* PIBS Version 1.05.0000 04/26/2005 has an incorrect /ht/isa/ranges property.
2792 * The values are bad, and it doesn't even have the right number of cells. */
2793static void __init fixup_device_tree_maple(void)
2794{
2795	phandle isa;
2796	u32 rloc = 0x01002000; /* IO space; PCI device = 4 */
2797	u32 isa_ranges[6];
2798	char *name;
2799
2800	name = "/ht@0/isa@4";
2801	isa = call_prom("finddevice", 1, 1, ADDR(name));
2802	if (!PHANDLE_VALID(isa)) {
2803		name = "/ht@0/isa@6";
2804		isa = call_prom("finddevice", 1, 1, ADDR(name));
2805		rloc = 0x01003000; /* IO space; PCI device = 6 */
2806	}
2807	if (!PHANDLE_VALID(isa))
2808		return;
2809
2810	if (prom_getproplen(isa, "ranges") != 12)
2811		return;
2812	if (prom_getprop(isa, "ranges", isa_ranges, sizeof(isa_ranges))
2813		== PROM_ERROR)
2814		return;
2815
2816	if (isa_ranges[0] != 0x1 ||
2817		isa_ranges[1] != 0xf4000000 ||
2818		isa_ranges[2] != 0x00010000)
2819		return;
2820
2821	prom_printf("Fixing up bogus ISA range on Maple/Apache...\n");
2822
2823	isa_ranges[0] = 0x1;
2824	isa_ranges[1] = 0x0;
2825	isa_ranges[2] = rloc;
2826	isa_ranges[3] = 0x0;
2827	isa_ranges[4] = 0x0;
2828	isa_ranges[5] = 0x00010000;
2829	prom_setprop(isa, name, "ranges",
2830			isa_ranges, sizeof(isa_ranges));
2831}
2832
2833#define CPC925_MC_START		0xf8000000
2834#define CPC925_MC_LENGTH	0x1000000
2835/* The values for memory-controller don't have right number of cells */
2836static void __init fixup_device_tree_maple_memory_controller(void)
2837{
2838	phandle mc;
2839	u32 mc_reg[4];
2840	char *name = "/hostbridge@f8000000";
2841	u32 ac, sc;
2842
2843	mc = call_prom("finddevice", 1, 1, ADDR(name));
2844	if (!PHANDLE_VALID(mc))
2845		return;
2846
2847	if (prom_getproplen(mc, "reg") != 8)
2848		return;
2849
2850	prom_getprop(prom.root, "#address-cells", &ac, sizeof(ac));
2851	prom_getprop(prom.root, "#size-cells", &sc, sizeof(sc));
2852	if ((ac != 2) || (sc != 2))
2853		return;
2854
2855	if (prom_getprop(mc, "reg", mc_reg, sizeof(mc_reg)) == PROM_ERROR)
2856		return;
2857
2858	if (mc_reg[0] != CPC925_MC_START || mc_reg[1] != CPC925_MC_LENGTH)
2859		return;
2860
2861	prom_printf("Fixing up bogus hostbridge on Maple...\n");
2862
2863	mc_reg[0] = 0x0;
2864	mc_reg[1] = CPC925_MC_START;
2865	mc_reg[2] = 0x0;
2866	mc_reg[3] = CPC925_MC_LENGTH;
2867	prom_setprop(mc, name, "reg", mc_reg, sizeof(mc_reg));
2868}
2869#else
2870#define fixup_device_tree_maple()
2871#define fixup_device_tree_maple_memory_controller()
2872#endif
2873
2874#ifdef CONFIG_PPC_CHRP
2875/*
2876 * Pegasos and BriQ lacks the "ranges" property in the isa node
2877 * Pegasos needs decimal IRQ 14/15, not hexadecimal
2878 * Pegasos has the IDE configured in legacy mode, but advertised as native
2879 */
2880static void __init fixup_device_tree_chrp(void)
2881{
2882	phandle ph;
2883	u32 prop[6];
2884	u32 rloc = 0x01006000; /* IO space; PCI device = 12 */
2885	char *name;
2886	int rc;
2887
2888	name = "/pci@80000000/isa@c";
2889	ph = call_prom("finddevice", 1, 1, ADDR(name));
2890	if (!PHANDLE_VALID(ph)) {
2891		name = "/pci@ff500000/isa@6";
2892		ph = call_prom("finddevice", 1, 1, ADDR(name));
2893		rloc = 0x01003000; /* IO space; PCI device = 6 */
2894	}
2895	if (PHANDLE_VALID(ph)) {
2896		rc = prom_getproplen(ph, "ranges");
2897		if (rc == 0 || rc == PROM_ERROR) {
2898			prom_printf("Fixing up missing ISA range on Pegasos...\n");
2899
2900			prop[0] = 0x1;
2901			prop[1] = 0x0;
2902			prop[2] = rloc;
2903			prop[3] = 0x0;
2904			prop[4] = 0x0;
2905			prop[5] = 0x00010000;
2906			prom_setprop(ph, name, "ranges", prop, sizeof(prop));
2907		}
2908	}
2909
2910	name = "/pci@80000000/ide@C,1";
2911	ph = call_prom("finddevice", 1, 1, ADDR(name));
2912	if (PHANDLE_VALID(ph)) {
2913		prom_printf("Fixing up IDE interrupt on Pegasos...\n");
2914		prop[0] = 14;
2915		prop[1] = 0x0;
2916		prom_setprop(ph, name, "interrupts", prop, 2*sizeof(u32));
2917		prom_printf("Fixing up IDE class-code on Pegasos...\n");
2918		rc = prom_getprop(ph, "class-code", prop, sizeof(u32));
2919		if (rc == sizeof(u32)) {
2920			prop[0] &= ~0x5;
2921			prom_setprop(ph, name, "class-code", prop, sizeof(u32));
2922		}
2923	}
2924}
2925#else
2926#define fixup_device_tree_chrp()
2927#endif
2928
2929#if defined(CONFIG_PPC64) && defined(CONFIG_PPC_PMAC)
2930static void __init fixup_device_tree_pmac(void)
2931{
2932	phandle u3, i2c, mpic;
2933	u32 u3_rev;
2934	u32 interrupts[2];
2935	u32 parent;
2936
2937	/* Some G5s have a missing interrupt definition, fix it up here */
2938	u3 = call_prom("finddevice", 1, 1, ADDR("/u3@0,f8000000"));
2939	if (!PHANDLE_VALID(u3))
2940		return;
2941	i2c = call_prom("finddevice", 1, 1, ADDR("/u3@0,f8000000/i2c@f8001000"));
2942	if (!PHANDLE_VALID(i2c))
2943		return;
2944	mpic = call_prom("finddevice", 1, 1, ADDR("/u3@0,f8000000/mpic@f8040000"));
2945	if (!PHANDLE_VALID(mpic))
2946		return;
2947
2948	/* check if proper rev of u3 */
2949	if (prom_getprop(u3, "device-rev", &u3_rev, sizeof(u3_rev))
2950	    == PROM_ERROR)
2951		return;
2952	if (u3_rev < 0x35 || u3_rev > 0x39)
2953		return;
2954	/* does it need fixup ? */
2955	if (prom_getproplen(i2c, "interrupts") > 0)
2956		return;
2957
2958	prom_printf("fixing up bogus interrupts for u3 i2c...\n");
2959
2960	/* interrupt on this revision of u3 is number 0 and level */
2961	interrupts[0] = 0;
2962	interrupts[1] = 1;
2963	prom_setprop(i2c, "/u3@0,f8000000/i2c@f8001000", "interrupts",
2964		     &interrupts, sizeof(interrupts));
2965	parent = (u32)mpic;
2966	prom_setprop(i2c, "/u3@0,f8000000/i2c@f8001000", "interrupt-parent",
2967		     &parent, sizeof(parent));
2968}
2969#else
2970#define fixup_device_tree_pmac()
2971#endif
2972
2973#ifdef CONFIG_PPC_EFIKA
2974/*
2975 * The MPC5200 FEC driver requires an phy-handle property to tell it how
2976 * to talk to the phy.  If the phy-handle property is missing, then this
2977 * function is called to add the appropriate nodes and link it to the
2978 * ethernet node.
2979 */
2980static void __init fixup_device_tree_efika_add_phy(void)
2981{
2982	u32 node;
2983	char prop[64];
2984	int rv;
2985
2986	/* Check if /builtin/ethernet exists - bail if it doesn't */
2987	node = call_prom("finddevice", 1, 1, ADDR("/builtin/ethernet"));
2988	if (!PHANDLE_VALID(node))
2989		return;
2990
2991	/* Check if the phy-handle property exists - bail if it does */
2992	rv = prom_getprop(node, "phy-handle", prop, sizeof(prop));
2993	if (!rv)
2994		return;
2995
2996	/*
2997	 * At this point the ethernet device doesn't have a phy described.
2998	 * Now we need to add the missing phy node and linkage
2999	 */
3000
3001	/* Check for an MDIO bus node - if missing then create one */
3002	node = call_prom("finddevice", 1, 1, ADDR("/builtin/mdio"));
3003	if (!PHANDLE_VALID(node)) {
3004		prom_printf("Adding Ethernet MDIO node\n");
3005		call_prom("interpret", 1, 1,
3006			" s\" /builtin\" find-device"
3007			" new-device"
3008				" 1 encode-int s\" #address-cells\" property"
3009				" 0 encode-int s\" #size-cells\" property"
3010				" s\" mdio\" device-name"
3011				" s\" fsl,mpc5200b-mdio\" encode-string"
3012				" s\" compatible\" property"
3013				" 0xf0003000 0x400 reg"
3014				" 0x2 encode-int"
3015				" 0x5 encode-int encode+"
3016				" 0x3 encode-int encode+"
3017				" s\" interrupts\" property"
3018			" finish-device");
3019	}
3020
3021	/* Check for a PHY device node - if missing then create one and
3022	 * give it's phandle to the ethernet node */
3023	node = call_prom("finddevice", 1, 1,
3024			 ADDR("/builtin/mdio/ethernet-phy"));
3025	if (!PHANDLE_VALID(node)) {
3026		prom_printf("Adding Ethernet PHY node\n");
3027		call_prom("interpret", 1, 1,
3028			" s\" /builtin/mdio\" find-device"
3029			" new-device"
3030				" s\" ethernet-phy\" device-name"
3031				" 0x10 encode-int s\" reg\" property"
3032				" my-self"
3033				" ihandle>phandle"
3034			" finish-device"
3035			" s\" /builtin/ethernet\" find-device"
3036				" encode-int"
3037				" s\" phy-handle\" property"
3038			" device-end");
3039	}
3040}
3041
3042static void __init fixup_device_tree_efika(void)
3043{
3044	int sound_irq[3] = { 2, 2, 0 };
3045	int bcomm_irq[3*16] = { 3,0,0, 3,1,0, 3,2,0, 3,3,0,
3046				3,4,0, 3,5,0, 3,6,0, 3,7,0,
3047				3,8,0, 3,9,0, 3,10,0, 3,11,0,
3048				3,12,0, 3,13,0, 3,14,0, 3,15,0 };
3049	u32 node;
3050	char prop[64];
3051	int rv, len;
3052
3053	/* Check if we're really running on a EFIKA */
3054	node = call_prom("finddevice", 1, 1, ADDR("/"));
3055	if (!PHANDLE_VALID(node))
3056		return;
3057
3058	rv = prom_getprop(node, "model", prop, sizeof(prop));
3059	if (rv == PROM_ERROR)
3060		return;
3061	if (prom_strcmp(prop, "EFIKA5K2"))
3062		return;
3063
3064	prom_printf("Applying EFIKA device tree fixups\n");
3065
3066	/* Claiming to be 'chrp' is death */
3067	node = call_prom("finddevice", 1, 1, ADDR("/"));
3068	rv = prom_getprop(node, "device_type", prop, sizeof(prop));
3069	if (rv != PROM_ERROR && (prom_strcmp(prop, "chrp") == 0))
3070		prom_setprop(node, "/", "device_type", "efika", sizeof("efika"));
3071
3072	/* CODEGEN,description is exposed in /proc/cpuinfo so
3073	   fix that too */
3074	rv = prom_getprop(node, "CODEGEN,description", prop, sizeof(prop));
3075	if (rv != PROM_ERROR && (prom_strstr(prop, "CHRP")))
3076		prom_setprop(node, "/", "CODEGEN,description",
3077			     "Efika 5200B PowerPC System",
3078			     sizeof("Efika 5200B PowerPC System"));
3079
3080	/* Fixup bestcomm interrupts property */
3081	node = call_prom("finddevice", 1, 1, ADDR("/builtin/bestcomm"));
3082	if (PHANDLE_VALID(node)) {
3083		len = prom_getproplen(node, "interrupts");
3084		if (len == 12) {
3085			prom_printf("Fixing bestcomm interrupts property\n");
3086			prom_setprop(node, "/builtin/bestcom", "interrupts",
3087				     bcomm_irq, sizeof(bcomm_irq));
3088		}
3089	}
3090
3091	/* Fixup sound interrupts property */
3092	node = call_prom("finddevice", 1, 1, ADDR("/builtin/sound"));
3093	if (PHANDLE_VALID(node)) {
3094		rv = prom_getprop(node, "interrupts", prop, sizeof(prop));
3095		if (rv == PROM_ERROR) {
3096			prom_printf("Adding sound interrupts property\n");
3097			prom_setprop(node, "/builtin/sound", "interrupts",
3098				     sound_irq, sizeof(sound_irq));
3099		}
3100	}
3101
3102	/* Make sure ethernet phy-handle property exists */
3103	fixup_device_tree_efika_add_phy();
3104}
3105#else
3106#define fixup_device_tree_efika()
3107#endif
3108
3109#ifdef CONFIG_PPC_PASEMI_NEMO
3110/*
3111 * CFE supplied on Nemo is broken in several ways, biggest
3112 * problem is that it reassigns ISA interrupts to unused mpic ints.
3113 * Add an interrupt-controller property for the io-bridge to use
3114 * and correct the ints so we can attach them to an irq_domain
3115 */
3116static void __init fixup_device_tree_pasemi(void)
3117{
3118	u32 interrupts[2], parent, rval, val = 0;
3119	char *name, *pci_name;
3120	phandle iob, node;
3121
3122	/* Find the root pci node */
3123	name = "/pxp@0,e0000000";
3124	iob = call_prom("finddevice", 1, 1, ADDR(name));
3125	if (!PHANDLE_VALID(iob))
3126		return;
3127
3128	/* check if interrupt-controller node set yet */
3129	if (prom_getproplen(iob, "interrupt-controller") !=PROM_ERROR)
3130		return;
3131
3132	prom_printf("adding interrupt-controller property for SB600...\n");
3133
3134	prom_setprop(iob, name, "interrupt-controller", &val, 0);
3135
3136	pci_name = "/pxp@0,e0000000/pci@11";
3137	node = call_prom("finddevice", 1, 1, ADDR(pci_name));
3138	parent = ADDR(iob);
3139
3140	for( ; prom_next_node(&node); ) {
3141		/* scan each node for one with an interrupt */
3142		if (!PHANDLE_VALID(node))
3143			continue;
3144
3145		rval = prom_getproplen(node, "interrupts");
3146		if (rval == 0 || rval == PROM_ERROR)
3147			continue;
3148
3149		prom_getprop(node, "interrupts", &interrupts, sizeof(interrupts));
3150		if ((interrupts[0] < 212) || (interrupts[0] > 222))
3151			continue;
3152
3153		/* found a node, update both interrupts and interrupt-parent */
3154		if ((interrupts[0] >= 212) && (interrupts[0] <= 215))
3155			interrupts[0] -= 203;
3156		if ((interrupts[0] >= 216) && (interrupts[0] <= 220))
3157			interrupts[0] -= 213;
3158		if (interrupts[0] == 221)
3159			interrupts[0] = 14;
3160		if (interrupts[0] == 222)
3161			interrupts[0] = 8;
3162
3163		prom_setprop(node, pci_name, "interrupts", interrupts,
3164					sizeof(interrupts));
3165		prom_setprop(node, pci_name, "interrupt-parent", &parent,
3166					sizeof(parent));
3167	}
3168
3169	/*
3170	 * The io-bridge has device_type set to 'io-bridge' change it to 'isa'
3171	 * so that generic isa-bridge code can add the SB600 and its on-board
3172	 * peripherals.
3173	 */
3174	name = "/pxp@0,e0000000/io-bridge@0";
3175	iob = call_prom("finddevice", 1, 1, ADDR(name));
3176	if (!PHANDLE_VALID(iob))
3177		return;
3178
3179	/* device_type is already set, just change it. */
3180
3181	prom_printf("Changing device_type of SB600 node...\n");
3182
3183	prom_setprop(iob, name, "device_type", "isa", sizeof("isa"));
3184}
3185#else	/* !CONFIG_PPC_PASEMI_NEMO */
3186static inline void fixup_device_tree_pasemi(void) { }
3187#endif
3188
3189static void __init fixup_device_tree(void)
3190{
3191	fixup_device_tree_maple();
3192	fixup_device_tree_maple_memory_controller();
3193	fixup_device_tree_chrp();
3194	fixup_device_tree_pmac();
3195	fixup_device_tree_efika();
3196	fixup_device_tree_pasemi();
3197}
3198
3199static void __init prom_find_boot_cpu(void)
3200{
3201	__be32 rval;
3202	ihandle prom_cpu;
3203	phandle cpu_pkg;
3204
3205	rval = 0;
3206	if (prom_getprop(prom.chosen, "cpu", &rval, sizeof(rval)) <= 0)
3207		return;
3208	prom_cpu = be32_to_cpu(rval);
3209
3210	cpu_pkg = call_prom("instance-to-package", 1, 1, prom_cpu);
3211
3212	if (!PHANDLE_VALID(cpu_pkg))
3213		return;
3214
3215	prom_getprop(cpu_pkg, "reg", &rval, sizeof(rval));
3216	prom.cpu = be32_to_cpu(rval);
3217
3218	prom_debug("Booting CPU hw index = %d\n", prom.cpu);
3219}
3220
3221static void __init prom_check_initrd(unsigned long r3, unsigned long r4)
3222{
3223#ifdef CONFIG_BLK_DEV_INITRD
3224	if (r3 && r4 && r4 != 0xdeadbeef) {
3225		__be64 val;
3226
3227		prom_initrd_start = is_kernel_addr(r3) ? __pa(r3) : r3;
3228		prom_initrd_end = prom_initrd_start + r4;
3229
3230		val = cpu_to_be64(prom_initrd_start);
3231		prom_setprop(prom.chosen, "/chosen", "linux,initrd-start",
3232			     &val, sizeof(val));
3233		val = cpu_to_be64(prom_initrd_end);
3234		prom_setprop(prom.chosen, "/chosen", "linux,initrd-end",
3235			     &val, sizeof(val));
3236
3237		reserve_mem(prom_initrd_start,
3238			    prom_initrd_end - prom_initrd_start);
3239
3240		prom_debug("initrd_start=0x%lx\n", prom_initrd_start);
3241		prom_debug("initrd_end=0x%lx\n", prom_initrd_end);
3242	}
3243#endif /* CONFIG_BLK_DEV_INITRD */
3244}
3245
3246#ifdef CONFIG_PPC_SVM
3247/*
3248 * Perform the Enter Secure Mode ultracall.
3249 */
3250static int enter_secure_mode(unsigned long kbase, unsigned long fdt)
3251{
3252	register unsigned long r3 asm("r3") = UV_ESM;
3253	register unsigned long r4 asm("r4") = kbase;
3254	register unsigned long r5 asm("r5") = fdt;
3255
3256	asm volatile("sc 2" : "+r"(r3) : "r"(r4), "r"(r5));
3257
3258	return r3;
3259}
3260
3261/*
3262 * Call the Ultravisor to transfer us to secure memory if we have an ESM blob.
3263 */
3264static void __init setup_secure_guest(unsigned long kbase, unsigned long fdt)
3265{
3266	int ret;
3267
3268	if (!prom_svm_enable)
3269		return;
 
 
3270
3271	/* Switch to secure mode. */
3272	prom_printf("Switching to secure mode.\n");
3273
3274	/*
3275	 * The ultravisor will do an integrity check of the kernel image but we
3276	 * relocated it so the check will fail. Restore the original image by
3277	 * relocating it back to the kernel virtual base address.
3278	 */
3279	relocate(KERNELBASE);
3280
3281	ret = enter_secure_mode(kbase, fdt);
 
 
 
 
3282
3283	/* Relocate the kernel again. */
3284	relocate(kbase);
3285
3286	if (ret != U_SUCCESS) {
3287		prom_printf("Returned %d from switching to secure mode.\n", ret);
3288		prom_rtas_os_term("Switch to secure mode failed.\n");
3289	}
3290}
3291#else
3292static void __init setup_secure_guest(unsigned long kbase, unsigned long fdt)
3293{
 
 
 
 
 
 
 
3294}
3295#endif /* CONFIG_PPC_SVM */
 
3296
3297/*
3298 * We enter here early on, when the Open Firmware prom is still
3299 * handling exceptions and the MMU hash table for us.
3300 */
3301
3302unsigned long __init prom_init(unsigned long r3, unsigned long r4,
3303			       unsigned long pp,
3304			       unsigned long r6, unsigned long r7,
3305			       unsigned long kbase)
3306{	
3307	unsigned long hdr;
3308
3309#ifdef CONFIG_PPC32
3310	unsigned long offset = reloc_offset();
3311	reloc_got2(offset);
 
 
3312#endif
3313
3314	/*
3315	 * First zero the BSS
3316	 */
3317	memset(&__bss_start, 0, __bss_stop - __bss_start);
3318
3319	/*
3320	 * Init interface to Open Firmware, get some node references,
3321	 * like /chosen
3322	 */
3323	prom_init_client_services(pp);
3324
3325	/*
3326	 * See if this OF is old enough that we need to do explicit maps
3327	 * and other workarounds
3328	 */
3329	prom_find_mmu();
3330
3331	/*
3332	 * Init prom stdout device
3333	 */
3334	prom_init_stdout();
3335
3336	prom_printf("Preparing to boot %s", linux_banner);
3337
3338	/*
3339	 * Get default machine type. At this point, we do not differentiate
3340	 * between pSeries SMP and pSeries LPAR
3341	 */
3342	of_platform = prom_find_machine_type();
3343	prom_printf("Detected machine type: %x\n", of_platform);
3344
3345#ifndef CONFIG_NONSTATIC_KERNEL
3346	/* Bail if this is a kdump kernel. */
3347	if (PHYSICAL_START > 0)
3348		prom_panic("Error: You can't boot a kdump kernel from OF!\n");
3349#endif
3350
3351	/*
3352	 * Check for an initrd
3353	 */
3354	prom_check_initrd(r3, r4);
3355
3356	/*
3357	 * Do early parsing of command line
3358	 */
3359	early_cmdline_parse();
3360
3361#ifdef CONFIG_PPC_PSERIES
3362	/*
3363	 * On pSeries, inform the firmware about our capabilities
3364	 */
3365	if (of_platform == PLATFORM_PSERIES ||
3366	    of_platform == PLATFORM_PSERIES_LPAR)
3367		prom_send_capabilities();
3368#endif
3369
3370	/*
3371	 * Copy the CPU hold code
3372	 */
3373	if (of_platform != PLATFORM_POWERMAC)
3374		copy_and_flush(0, kbase, 0x100, 0);
3375
3376	/*
 
 
 
 
 
3377	 * Initialize memory management within prom_init
3378	 */
3379	prom_init_mem();
3380
3381	/*
3382	 * Determine which cpu is actually running right _now_
3383	 */
3384	prom_find_boot_cpu();
3385
3386	/* 
3387	 * Initialize display devices
3388	 */
3389	prom_check_displays();
3390
3391#if defined(CONFIG_PPC64) && defined(__BIG_ENDIAN__)
3392	/*
3393	 * Initialize IOMMU (TCE tables) on pSeries. Do that before anything else
3394	 * that uses the allocator, we need to make sure we get the top of memory
3395	 * available for us here...
3396	 */
3397	if (of_platform == PLATFORM_PSERIES)
3398		prom_initialize_tce_table();
3399#endif
3400
3401	/*
3402	 * On non-powermacs, try to instantiate RTAS. PowerMacs don't
3403	 * have a usable RTAS implementation.
3404	 */
3405	if (of_platform != PLATFORM_POWERMAC)
 
3406		prom_instantiate_rtas();
3407
 
 
 
 
 
3408#ifdef CONFIG_PPC64
3409	/* instantiate sml */
3410	prom_instantiate_sml();
3411#endif
3412
3413	/*
3414	 * On non-powermacs, put all CPUs in spin-loops.
3415	 *
3416	 * PowerMacs use a different mechanism to spin CPUs
3417	 *
3418	 * (This must be done after instanciating RTAS)
3419	 */
3420	if (of_platform != PLATFORM_POWERMAC)
 
3421		prom_hold_cpus();
3422
3423	/*
3424	 * Fill in some infos for use by the kernel later on
3425	 */
3426	if (prom_memory_limit) {
3427		__be64 val = cpu_to_be64(prom_memory_limit);
3428		prom_setprop(prom.chosen, "/chosen", "linux,memory-limit",
3429			     &val, sizeof(val));
3430	}
3431#ifdef CONFIG_PPC64
3432	if (prom_iommu_off)
3433		prom_setprop(prom.chosen, "/chosen", "linux,iommu-off",
3434			     NULL, 0);
3435
3436	if (prom_iommu_force_on)
3437		prom_setprop(prom.chosen, "/chosen", "linux,iommu-force-on",
3438			     NULL, 0);
3439
3440	if (prom_tce_alloc_start) {
3441		prom_setprop(prom.chosen, "/chosen", "linux,tce-alloc-start",
3442			     &prom_tce_alloc_start,
3443			     sizeof(prom_tce_alloc_start));
3444		prom_setprop(prom.chosen, "/chosen", "linux,tce-alloc-end",
3445			     &prom_tce_alloc_end,
3446			     sizeof(prom_tce_alloc_end));
3447	}
3448#endif
3449
3450	/*
3451	 * Fixup any known bugs in the device-tree
3452	 */
3453	fixup_device_tree();
3454
3455	/*
3456	 * Now finally create the flattened device-tree
3457	 */
3458	prom_printf("copying OF device tree...\n");
3459	flatten_device_tree();
3460
3461	/*
3462	 * in case stdin is USB and still active on IBM machines...
3463	 * Unfortunately quiesce crashes on some powermacs if we have
3464	 * closed stdin already (in particular the powerbook 101).
 
3465	 */
3466	if (of_platform != PLATFORM_POWERMAC)
 
3467		prom_close_stdin();
3468
3469	/*
3470	 * Call OF "quiesce" method to shut down pending DMA's from
3471	 * devices etc...
3472	 */
3473	prom_printf("Quiescing Open Firmware ...\n");
3474	call_prom("quiesce", 0, 0);
3475
3476	/*
3477	 * And finally, call the kernel passing it the flattened device
3478	 * tree and NULL as r5, thus triggering the new entry point which
3479	 * is common to us and kexec
3480	 */
3481	hdr = dt_header_start;
3482
3483	prom_printf("Booting Linux via __start() @ 0x%lx ...\n", kbase);
3484	prom_debug("->dt_header_start=0x%lx\n", hdr);
 
 
 
3485
3486#ifdef CONFIG_PPC32
3487	reloc_got2(-offset);
 
 
3488#endif
3489
3490	/* Move to secure memory if we're supposed to be secure guests. */
3491	setup_secure_guest(kbase, hdr);
3492
 
 
3493	__start(hdr, kbase, 0, 0, 0, 0, 0);
 
3494
3495	return 0;
3496}
v4.10.11
 
   1/*
   2 * Procedures for interfacing to Open Firmware.
   3 *
   4 * Paul Mackerras	August 1996.
   5 * Copyright (C) 1996-2005 Paul Mackerras.
   6 * 
   7 *  Adapted for 64bit PowerPC by Dave Engebretsen and Peter Bergner.
   8 *    {engebret|bergner}@us.ibm.com 
   9 *
  10 *      This program is free software; you can redistribute it and/or
  11 *      modify it under the terms of the GNU General Public License
  12 *      as published by the Free Software Foundation; either version
  13 *      2 of the License, or (at your option) any later version.
  14 */
  15
  16#undef DEBUG_PROM
  17
 
 
 
  18#include <stdarg.h>
  19#include <linux/kernel.h>
  20#include <linux/string.h>
  21#include <linux/init.h>
  22#include <linux/threads.h>
  23#include <linux/spinlock.h>
  24#include <linux/types.h>
  25#include <linux/pci.h>
  26#include <linux/proc_fs.h>
  27#include <linux/stringify.h>
  28#include <linux/delay.h>
  29#include <linux/initrd.h>
  30#include <linux/bitops.h>
 
 
  31#include <asm/prom.h>
  32#include <asm/rtas.h>
  33#include <asm/page.h>
  34#include <asm/processor.h>
 
  35#include <asm/irq.h>
  36#include <asm/io.h>
  37#include <asm/smp.h>
  38#include <asm/mmu.h>
  39#include <asm/pgtable.h>
  40#include <asm/iommu.h>
  41#include <asm/btext.h>
  42#include <asm/sections.h>
  43#include <asm/machdep.h>
  44#include <asm/opal.h>
  45#include <asm/asm-prototypes.h>
 
  46
  47#include <linux/linux_logo.h>
  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
  88int of_workarounds;
  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...)
 104#endif
 105
 106
 107typedef u32 prom_arg_t;
 108
 109struct prom_args {
 110        __be32 service;
 111        __be32 nargs;
 112        __be32 nret;
 113        __be32 args[10];
 114};
 115
 116struct prom_t {
 117	ihandle root;
 118	phandle chosen;
 119	int cpu;
 120	ihandle stdout;
 121	ihandle mmumap;
 122	ihandle memory;
 123};
 124
 125struct mem_map_entry {
 126	__be64	base;
 127	__be64	size;
 128};
 129
 130typedef __be32 cell_t;
 131
 132extern 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
 137extern int enter_prom(struct prom_args *args, unsigned long entry);
 138#else
 139static inline int enter_prom(struct prom_args *args, unsigned long entry)
 140{
 141	return ((int (*)(struct prom_args *))entry)(args);
 142}
 143#endif
 144
 145extern void copy_and_flush(unsigned long dest, unsigned long src,
 146			   unsigned long size, unsigned long offset);
 147
 148/* prom structure */
 149static struct prom_t __initdata prom;
 150
 151static unsigned long prom_entry __initdata;
 152
 153#define PROM_SCRATCH_SIZE 256
 
 154
 155static char __initdata of_stdout_device[256];
 156static char __initdata prom_scratch[PROM_SCRATCH_SIZE];
 
 157
 158static unsigned long __initdata dt_header_start;
 159static unsigned long __initdata dt_struct_start, dt_struct_end;
 160static unsigned long __initdata dt_string_start, dt_string_end;
 161
 162static unsigned long __initdata prom_initrd_start, prom_initrd_end;
 
 
 
 
 
 163
 164#ifdef CONFIG_PPC64
 165static int __initdata prom_iommu_force_on;
 166static int __initdata prom_iommu_off;
 167static unsigned long __initdata prom_tce_alloc_start;
 168static unsigned long __initdata prom_tce_alloc_end;
 
 
 
 169#endif
 170
 
 
 
 
 
 
 
 171/* Platforms codes are now obsolete in the kernel. Now only used within this
 172 * file and ultimately gone too. Feel free to change them if you need, they
 173 * are not shared with anything outside of this file anymore
 174 */
 175#define PLATFORM_PSERIES	0x0100
 176#define PLATFORM_PSERIES_LPAR	0x0101
 177#define PLATFORM_LPAR		0x0001
 178#define PLATFORM_POWERMAC	0x0400
 179#define PLATFORM_GENERIC	0x0500
 180#define PLATFORM_OPAL		0x0600
 181
 182static int __initdata of_platform;
 183
 184static char __initdata prom_cmd_line[COMMAND_LINE_SIZE];
 185
 186static unsigned long __initdata prom_memory_limit;
 187
 188static unsigned long __initdata alloc_top;
 189static unsigned long __initdata alloc_top_high;
 190static unsigned long __initdata alloc_bottom;
 191static unsigned long __initdata rmo_top;
 192static unsigned long __initdata ram_top;
 193
 194static struct mem_map_entry __initdata mem_reserve_map[MEM_RESERVE_MAP_SIZE];
 195static int __initdata mem_reserve_cnt;
 196
 197static cell_t __initdata regbuf[1024];
 198
 199static bool rtas_has_query_cpu_stopped;
 200
 201
 202/*
 203 * Error results ... some OF calls will return "-1" on error, some
 204 * will return 0, some will return either. To simplify, here are
 205 * macros to use with any ihandle or phandle return value to check if
 206 * it is valid
 207 */
 208
 209#define PROM_ERROR		(-1u)
 210#define PHANDLE_VALID(p)	((p) != 0 && (p) != PROM_ERROR)
 211#define IHANDLE_VALID(i)	((i) != 0 && (i) != PROM_ERROR)
 212
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 213
 214/* This is the one and *ONLY* place where we actually call open
 215 * firmware.
 216 */
 217
 218static int __init call_prom(const char *service, int nargs, int nret, ...)
 219{
 220	int i;
 221	struct prom_args args;
 222	va_list list;
 223
 224	args.service = cpu_to_be32(ADDR(service));
 225	args.nargs = cpu_to_be32(nargs);
 226	args.nret = cpu_to_be32(nret);
 227
 228	va_start(list, nret);
 229	for (i = 0; i < nargs; i++)
 230		args.args[i] = cpu_to_be32(va_arg(list, prom_arg_t));
 231	va_end(list);
 232
 233	for (i = 0; i < nret; i++)
 234		args.args[nargs+i] = 0;
 235
 236	if (enter_prom(&args, prom_entry) < 0)
 237		return PROM_ERROR;
 238
 239	return (nret > 0) ? be32_to_cpu(args.args[nargs]) : 0;
 240}
 241
 242static int __init call_prom_ret(const char *service, int nargs, int nret,
 243				prom_arg_t *rets, ...)
 244{
 245	int i;
 246	struct prom_args args;
 247	va_list list;
 248
 249	args.service = cpu_to_be32(ADDR(service));
 250	args.nargs = cpu_to_be32(nargs);
 251	args.nret = cpu_to_be32(nret);
 252
 253	va_start(list, rets);
 254	for (i = 0; i < nargs; i++)
 255		args.args[i] = cpu_to_be32(va_arg(list, prom_arg_t));
 256	va_end(list);
 257
 258	for (i = 0; i < nret; i++)
 259		args.args[nargs+i] = 0;
 260
 261	if (enter_prom(&args, prom_entry) < 0)
 262		return PROM_ERROR;
 263
 264	if (rets != NULL)
 265		for (i = 1; i < nret; ++i)
 266			rets[i-1] = be32_to_cpu(args.args[nargs+i]);
 267
 268	return (nret > 0) ? be32_to_cpu(args.args[nargs]) : 0;
 269}
 270
 271
 272static void __init prom_print(const char *msg)
 273{
 274	const char *p, *q;
 275
 276	if (prom.stdout == 0)
 277		return;
 278
 279	for (p = msg; *p != 0; p = q) {
 280		for (q = p; *q != 0 && *q != '\n'; ++q)
 281			;
 282		if (q > p)
 283			call_prom("write", 3, 1, prom.stdout, p, q - p);
 284		if (*q == 0)
 285			break;
 286		++q;
 287		call_prom("write", 3, 1, prom.stdout, ADDR("\r\n"), 2);
 288	}
 289}
 290
 291
 
 
 
 
 292static void __init prom_print_hex(unsigned long val)
 293{
 294	int i, nibbles = sizeof(val)*2;
 295	char buf[sizeof(val)*2+1];
 296
 297	for (i = nibbles-1;  i >= 0;  i--) {
 298		buf[i] = (val & 0xf) + '0';
 299		if (buf[i] > '9')
 300			buf[i] += ('a'-'0'-10);
 301		val >>= 4;
 302	}
 303	buf[nibbles] = '\0';
 304	call_prom("write", 3, 1, prom.stdout, buf, nibbles);
 305}
 306
 307/* max number of decimal digits in an unsigned long */
 308#define UL_DIGITS 21
 309static void __init prom_print_dec(unsigned long val)
 310{
 311	int i, size;
 312	char buf[UL_DIGITS+1];
 313
 314	for (i = UL_DIGITS-1; i >= 0;  i--) {
 315		buf[i] = (val % 10) + '0';
 316		val = val/10;
 317		if (val == 0)
 318			break;
 319	}
 320	/* shift stuff down */
 321	size = UL_DIGITS - i;
 322	call_prom("write", 3, 1, prom.stdout, buf+i, size);
 323}
 324
 
 325static void __init prom_printf(const char *format, ...)
 326{
 327	const char *p, *q, *s;
 328	va_list args;
 329	unsigned long v;
 330	long vs;
 
 331
 332	va_start(args, format);
 333	for (p = format; *p != 0; p = q) {
 334		for (q = p; *q != 0 && *q != '\n' && *q != '%'; ++q)
 335			;
 336		if (q > p)
 337			call_prom("write", 3, 1, prom.stdout, p, q - p);
 338		if (*q == 0)
 339			break;
 340		if (*q == '\n') {
 341			++q;
 342			call_prom("write", 3, 1, prom.stdout,
 343				  ADDR("\r\n"), 2);
 344			continue;
 345		}
 346		++q;
 347		if (*q == 0)
 348			break;
 
 
 
 
 349		switch (*q) {
 350		case 's':
 351			++q;
 352			s = va_arg(args, const char *);
 353			prom_print(s);
 354			break;
 355		case 'x':
 356			++q;
 357			v = va_arg(args, unsigned long);
 
 
 
 
 
 
 
 
 
 
 
 358			prom_print_hex(v);
 359			break;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 360		case 'd':
 361			++q;
 362			vs = va_arg(args, int);
 
 
 
 
 
 
 
 
 
 
 
 363			if (vs < 0) {
 364				prom_print("-");
 365				vs = -vs;
 366			}
 367			prom_print_dec(vs);
 368			break;
 369		case 'l':
 370			++q;
 371			if (*q == 0)
 372				break;
 373			else if (*q == 'x') {
 374				++q;
 375				v = va_arg(args, unsigned long);
 376				prom_print_hex(v);
 377			} else if (*q == 'u') { /* '%lu' */
 378				++q;
 379				v = va_arg(args, unsigned long);
 380				prom_print_dec(v);
 381			} else if (*q == 'd') { /* %ld */
 382				++q;
 383				vs = va_arg(args, long);
 384				if (vs < 0) {
 385					prom_print("-");
 386					vs = -vs;
 387				}
 388				prom_print_dec(vs);
 389			}
 390			break;
 391		}
 392	}
 393	va_end(args);
 394}
 395
 396
 397static unsigned int __init prom_claim(unsigned long virt, unsigned long size,
 398				unsigned long align)
 399{
 400
 401	if (align == 0 && (OF_WORKAROUNDS & OF_WA_CLAIM)) {
 402		/*
 403		 * Old OF requires we claim physical and virtual separately
 404		 * and then map explicitly (assuming virtual mode)
 405		 */
 406		int ret;
 407		prom_arg_t result;
 408
 409		ret = call_prom_ret("call-method", 5, 2, &result,
 410				    ADDR("claim"), prom.memory,
 411				    align, size, virt);
 412		if (ret != 0 || result == -1)
 413			return -1;
 414		ret = call_prom_ret("call-method", 5, 2, &result,
 415				    ADDR("claim"), prom.mmumap,
 416				    align, size, virt);
 417		if (ret != 0) {
 418			call_prom("call-method", 4, 1, ADDR("release"),
 419				  prom.memory, size, virt);
 420			return -1;
 421		}
 422		/* the 0x12 is M (coherence) + PP == read/write */
 423		call_prom("call-method", 6, 1,
 424			  ADDR("map"), prom.mmumap, 0x12, size, virt, virt);
 425		return virt;
 426	}
 427	return call_prom("claim", 3, 1, (prom_arg_t)virt, (prom_arg_t)size,
 428			 (prom_arg_t)align);
 429}
 430
 431static void __init __attribute__((noreturn)) prom_panic(const char *reason)
 432{
 433	prom_print(reason);
 434	/* Do not call exit because it clears the screen on pmac
 435	 * it also causes some sort of double-fault on early pmacs */
 436	if (of_platform == PLATFORM_POWERMAC)
 437		asm("trap\n");
 438
 439	/* ToDo: should put up an SRC here on pSeries */
 440	call_prom("exit", 0, 0);
 441
 442	for (;;)			/* should never get here */
 443		;
 444}
 445
 446
 447static int __init prom_next_node(phandle *nodep)
 448{
 449	phandle node;
 450
 451	if ((node = *nodep) != 0
 452	    && (*nodep = call_prom("child", 1, 1, node)) != 0)
 453		return 1;
 454	if ((*nodep = call_prom("peer", 1, 1, node)) != 0)
 455		return 1;
 456	for (;;) {
 457		if ((node = call_prom("parent", 1, 1, node)) == 0)
 458			return 0;
 459		if ((*nodep = call_prom("peer", 1, 1, node)) != 0)
 460			return 1;
 461	}
 462}
 463
 464static inline int prom_getprop(phandle node, const char *pname,
 465			       void *value, size_t valuelen)
 466{
 467	return call_prom("getprop", 4, 1, node, ADDR(pname),
 468			 (u32)(unsigned long) value, (u32) valuelen);
 469}
 470
 471static inline int prom_getproplen(phandle node, const char *pname)
 472{
 473	return call_prom("getproplen", 2, 1, node, ADDR(pname));
 474}
 475
 476static void add_string(char **str, const char *q)
 477{
 478	char *p = *str;
 479
 480	while (*q)
 481		*p++ = *q++;
 482	*p++ = ' ';
 483	*str = p;
 484}
 485
 486static char *tohex(unsigned int x)
 487{
 488	static char digits[] = "0123456789abcdef";
 489	static char result[9];
 490	int i;
 491
 492	result[8] = 0;
 493	i = 8;
 494	do {
 495		--i;
 496		result[i] = digits[x & 0xf];
 497		x >>= 4;
 498	} while (x != 0 && i > 0);
 499	return &result[i];
 500}
 501
 502static int __init prom_setprop(phandle node, const char *nodename,
 503			       const char *pname, void *value, size_t valuelen)
 504{
 505	char cmd[256], *p;
 506
 507	if (!(OF_WORKAROUNDS & OF_WA_LONGTRAIL))
 508		return call_prom("setprop", 4, 1, node, ADDR(pname),
 509				 (u32)(unsigned long) value, (u32) valuelen);
 510
 511	/* gah... setprop doesn't work on longtrail, have to use interpret */
 512	p = cmd;
 513	add_string(&p, "dev");
 514	add_string(&p, nodename);
 515	add_string(&p, tohex((u32)(unsigned long) value));
 516	add_string(&p, tohex(valuelen));
 517	add_string(&p, tohex(ADDR(pname)));
 518	add_string(&p, tohex(strlen(pname)));
 519	add_string(&p, "property");
 520	*p = 0;
 521	return call_prom("interpret", 1, 1, (u32)(unsigned long) cmd);
 522}
 523
 524/* We can't use the standard versions because of relocation headaches. */
 525#define isxdigit(c)	(('0' <= (c) && (c) <= '9') \
 526			 || ('a' <= (c) && (c) <= 'f') \
 527			 || ('A' <= (c) && (c) <= 'F'))
 528
 529#define isdigit(c)	('0' <= (c) && (c) <= '9')
 530#define islower(c)	('a' <= (c) && (c) <= 'z')
 531#define toupper(c)	(islower(c) ? ((c) - 'a' + 'A') : (c))
 532
 533static unsigned long prom_strtoul(const char *cp, const char **endp)
 534{
 535	unsigned long result = 0, base = 10, value;
 536
 537	if (*cp == '0') {
 538		base = 8;
 539		cp++;
 540		if (toupper(*cp) == 'X') {
 541			cp++;
 542			base = 16;
 543		}
 544	}
 545
 546	while (isxdigit(*cp) &&
 547	       (value = isdigit(*cp) ? *cp - '0' : toupper(*cp) - 'A' + 10) < base) {
 548		result = result * base + value;
 549		cp++;
 550	}
 551
 552	if (endp)
 553		*endp = cp;
 554
 555	return result;
 556}
 557
 558static unsigned long prom_memparse(const char *ptr, const char **retptr)
 559{
 560	unsigned long ret = prom_strtoul(ptr, retptr);
 561	int shift = 0;
 562
 563	/*
 564	 * We can't use a switch here because GCC *may* generate a
 565	 * jump table which won't work, because we're not running at
 566	 * the address we're linked at.
 567	 */
 568	if ('G' == **retptr || 'g' == **retptr)
 569		shift = 30;
 570
 571	if ('M' == **retptr || 'm' == **retptr)
 572		shift = 20;
 573
 574	if ('K' == **retptr || 'k' == **retptr)
 575		shift = 10;
 576
 577	if (shift) {
 578		ret <<= shift;
 579		(*retptr)++;
 580	}
 581
 582	return ret;
 583}
 584
 585/*
 586 * Early parsing of the command line passed to the kernel, used for
 587 * "mem=x" and the options that affect the iommu
 588 */
 589static void __init early_cmdline_parse(void)
 590{
 591	const char *opt;
 592
 593	char *p;
 594	int l = 0;
 595
 596	prom_cmd_line[0] = 0;
 597	p = prom_cmd_line;
 598	if ((long)prom.chosen > 0)
 
 599		l = prom_getprop(prom.chosen, "bootargs", p, COMMAND_LINE_SIZE-1);
 600#ifdef CONFIG_CMDLINE
 601	if (l <= 0 || p[0] == '\0') /* dbl check */
 602		strlcpy(prom_cmd_line,
 603			CONFIG_CMDLINE, sizeof(prom_cmd_line));
 604#endif /* CONFIG_CMDLINE */
 605	prom_printf("command line: %s\n", prom_cmd_line);
 606
 607#ifdef CONFIG_PPC64
 608	opt = strstr(prom_cmd_line, "iommu=");
 609	if (opt) {
 610		prom_printf("iommu opt is: %s\n", opt);
 611		opt += 6;
 612		while (*opt && *opt == ' ')
 613			opt++;
 614		if (!strncmp(opt, "off", 3))
 615			prom_iommu_off = 1;
 616		else if (!strncmp(opt, "force", 5))
 617			prom_iommu_force_on = 1;
 618	}
 619#endif
 620	opt = strstr(prom_cmd_line, "mem=");
 621	if (opt) {
 622		opt += 4;
 623		prom_memory_limit = prom_memparse(opt, (const char **)&opt);
 624#ifdef CONFIG_PPC64
 625		/* Align to 16 MB == size of ppc64 large page */
 626		prom_memory_limit = ALIGN(prom_memory_limit, 0x1000000);
 627#endif
 628	}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 629}
 630
 631#if defined(CONFIG_PPC_PSERIES) || defined(CONFIG_PPC_POWERNV)
 632/*
 633 * The architecture vector has an array of PVR mask/value pairs,
 634 * followed by # option vectors - 1, followed by the option vectors.
 635 *
 636 * See prom.h for the definition of the bits specified in the
 637 * architecture vector.
 638 */
 639
 640/* Firmware expects the value to be n - 1, where n is the # of vectors */
 641#define NUM_VECTORS(n)		((n) - 1)
 642
 643/*
 644 * Firmware expects 1 + n - 2, where n is the length of the option vector in
 645 * bytes. The 1 accounts for the length byte itself, the - 2 .. ?
 646 */
 647#define VECTOR_LENGTH(n)	(1 + (n) - 2)
 648
 649struct option_vector1 {
 650	u8 byte1;
 651	u8 arch_versions;
 
 652} __packed;
 653
 654struct option_vector2 {
 655	u8 byte1;
 656	__be16 reserved;
 657	__be32 real_base;
 658	__be32 real_size;
 659	__be32 virt_base;
 660	__be32 virt_size;
 661	__be32 load_base;
 662	__be32 min_rma;
 663	__be32 min_load;
 664	u8 min_rma_percent;
 665	u8 max_pft_size;
 666} __packed;
 667
 668struct option_vector3 {
 669	u8 byte1;
 670	u8 byte2;
 671} __packed;
 672
 673struct option_vector4 {
 674	u8 byte1;
 675	u8 min_vp_cap;
 676} __packed;
 677
 678struct option_vector5 {
 679	u8 byte1;
 680	u8 byte2;
 681	u8 byte3;
 682	u8 cmo;
 683	u8 associativity;
 684	u8 bin_opts;
 685	u8 micro_checkpoint;
 686	u8 reserved0;
 687	__be32 max_cpus;
 688	__be16 papr_level;
 689	__be16 reserved1;
 690	u8 platform_facilities;
 691	u8 reserved2;
 692	__be16 reserved3;
 693	u8 subprocessors;
 
 
 
 
 
 694} __packed;
 695
 696struct option_vector6 {
 697	u8 reserved;
 698	u8 secondary_pteg;
 699	u8 os_name;
 700} __packed;
 701
 
 
 
 
 702struct ibm_arch_vec {
 703	struct { u32 mask, val; } pvrs[10];
 704
 705	u8 num_vectors;
 706
 707	u8 vec1_len;
 708	struct option_vector1 vec1;
 709
 710	u8 vec2_len;
 711	struct option_vector2 vec2;
 712
 713	u8 vec3_len;
 714	struct option_vector3 vec3;
 715
 716	u8 vec4_len;
 717	struct option_vector4 vec4;
 718
 719	u8 vec5_len;
 720	struct option_vector5 vec5;
 721
 722	u8 vec6_len;
 723	struct option_vector6 vec6;
 
 
 
 724} __packed;
 725
 726struct ibm_arch_vec __cacheline_aligned ibm_architecture_vec = {
 727	.pvrs = {
 728		{
 729			.mask = cpu_to_be32(0xfffe0000), /* POWER5/POWER5+ */
 730			.val  = cpu_to_be32(0x003a0000),
 731		},
 732		{
 733			.mask = cpu_to_be32(0xffff0000), /* POWER6 */
 734			.val  = cpu_to_be32(0x003e0000),
 735		},
 736		{
 737			.mask = cpu_to_be32(0xffff0000), /* POWER7 */
 738			.val  = cpu_to_be32(0x003f0000),
 739		},
 740		{
 741			.mask = cpu_to_be32(0xffff0000), /* POWER8E */
 742			.val  = cpu_to_be32(0x004b0000),
 743		},
 744		{
 745			.mask = cpu_to_be32(0xffff0000), /* POWER8NVL */
 746			.val  = cpu_to_be32(0x004c0000),
 747		},
 748		{
 749			.mask = cpu_to_be32(0xffff0000), /* POWER8 */
 750			.val  = cpu_to_be32(0x004d0000),
 751		},
 752		{
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 753			.mask = cpu_to_be32(0xffffffff), /* all 2.07-compliant */
 754			.val  = cpu_to_be32(0x0f000004),
 755		},
 756		{
 757			.mask = cpu_to_be32(0xffffffff), /* all 2.06-compliant */
 758			.val  = cpu_to_be32(0x0f000003),
 759		},
 760		{
 761			.mask = cpu_to_be32(0xffffffff), /* all 2.05-compliant */
 762			.val  = cpu_to_be32(0x0f000002),
 763		},
 764		{
 765			.mask = cpu_to_be32(0xfffffffe), /* all 2.04-compliant and earlier */
 766			.val  = cpu_to_be32(0x0f000001),
 767		},
 768	},
 769
 770	.num_vectors = NUM_VECTORS(6),
 771
 772	.vec1_len = VECTOR_LENGTH(sizeof(struct option_vector1)),
 773	.vec1 = {
 774		.byte1 = 0,
 775		.arch_versions = OV1_PPC_2_00 | OV1_PPC_2_01 | OV1_PPC_2_02 | OV1_PPC_2_03 |
 776				 OV1_PPC_2_04 | OV1_PPC_2_05 | OV1_PPC_2_06 | OV1_PPC_2_07,
 
 777	},
 778
 779	.vec2_len = VECTOR_LENGTH(sizeof(struct option_vector2)),
 780	/* option vector 2: Open Firmware options supported */
 781	.vec2 = {
 782		.byte1 = OV2_REAL_MODE,
 783		.reserved = 0,
 784		.real_base = cpu_to_be32(0xffffffff),
 785		.real_size = cpu_to_be32(0xffffffff),
 786		.virt_base = cpu_to_be32(0xffffffff),
 787		.virt_size = cpu_to_be32(0xffffffff),
 788		.load_base = cpu_to_be32(0xffffffff),
 789		.min_rma = cpu_to_be32(256),		/* 256MB min RMA */
 790		.min_load = cpu_to_be32(0xffffffff),	/* full client load */
 791		.min_rma_percent = 0,	/* min RMA percentage of total RAM */
 792		.max_pft_size = 48,	/* max log_2(hash table size) */
 793	},
 794
 795	.vec3_len = VECTOR_LENGTH(sizeof(struct option_vector3)),
 796	/* option vector 3: processor options supported */
 797	.vec3 = {
 798		.byte1 = 0,			/* don't ignore, don't halt */
 799		.byte2 = OV3_FP | OV3_VMX | OV3_DFP,
 800	},
 801
 802	.vec4_len = VECTOR_LENGTH(sizeof(struct option_vector4)),
 803	/* option vector 4: IBM PAPR implementation */
 804	.vec4 = {
 805		.byte1 = 0,			/* don't halt */
 806		.min_vp_cap = OV4_MIN_ENT_CAP,	/* minimum VP entitled capacity */
 807	},
 808
 809	.vec5_len = VECTOR_LENGTH(sizeof(struct option_vector5)),
 810	/* option vector 5: PAPR/OF options */
 811	.vec5 = {
 812		.byte1 = 0,				/* don't ignore, don't halt */
 813		.byte2 = OV5_FEAT(OV5_LPAR) | OV5_FEAT(OV5_SPLPAR) | OV5_FEAT(OV5_LARGE_PAGES) |
 814		OV5_FEAT(OV5_DRCONF_MEMORY) | OV5_FEAT(OV5_DONATE_DEDICATE_CPU) |
 815#ifdef CONFIG_PCI_MSI
 816		/* PCIe/MSI support.  Without MSI full PCIe is not supported */
 817		OV5_FEAT(OV5_MSI),
 818#else
 819		0,
 820#endif
 821		.byte3 = 0,
 822		.cmo =
 823#ifdef CONFIG_PPC_SMLPAR
 824		OV5_FEAT(OV5_CMO) | OV5_FEAT(OV5_XCMO),
 825#else
 826		0,
 827#endif
 828		.associativity = OV5_FEAT(OV5_TYPE1_AFFINITY) | OV5_FEAT(OV5_PRRN),
 829		.bin_opts = 0,
 830		.micro_checkpoint = 0,
 831		.reserved0 = 0,
 832		.max_cpus = cpu_to_be32(NR_CPUS),	/* number of cores supported */
 833		.papr_level = 0,
 834		.reserved1 = 0,
 835		.platform_facilities = OV5_FEAT(OV5_PFO_HW_RNG) | OV5_FEAT(OV5_PFO_HW_ENCR) | OV5_FEAT(OV5_PFO_HW_842),
 836		.reserved2 = 0,
 837		.reserved3 = 0,
 838		.subprocessors = 1,
 
 
 
 
 
 839	},
 840
 841	/* option vector 6: IBM PAPR hints */
 842	.vec6_len = VECTOR_LENGTH(sizeof(struct option_vector6)),
 843	.vec6 = {
 844		.reserved = 0,
 845		.secondary_pteg = 0,
 846		.os_name = OV6_LINUX,
 847	},
 
 
 
 848};
 849
 
 
 850/* Old method - ELF header with PT_NOTE sections only works on BE */
 851#ifdef __BIG_ENDIAN__
 852static struct fake_elf {
 853	Elf32_Ehdr	elfhdr;
 854	Elf32_Phdr	phdr[2];
 855	struct chrpnote {
 856		u32	namesz;
 857		u32	descsz;
 858		u32	type;
 859		char	name[8];	/* "PowerPC" */
 860		struct chrpdesc {
 861			u32	real_mode;
 862			u32	real_base;
 863			u32	real_size;
 864			u32	virt_base;
 865			u32	virt_size;
 866			u32	load_base;
 867		} chrpdesc;
 868	} chrpnote;
 869	struct rpanote {
 870		u32	namesz;
 871		u32	descsz;
 872		u32	type;
 873		char	name[24];	/* "IBM,RPA-Client-Config" */
 874		struct rpadesc {
 875			u32	lpar_affinity;
 876			u32	min_rmo_size;
 877			u32	min_rmo_percent;
 878			u32	max_pft_size;
 879			u32	splpar;
 880			u32	min_load;
 881			u32	new_mem_def;
 882			u32	ignore_me;
 883		} rpadesc;
 884	} rpanote;
 885} fake_elf = {
 886	.elfhdr = {
 887		.e_ident = { 0x7f, 'E', 'L', 'F',
 888			     ELFCLASS32, ELFDATA2MSB, EV_CURRENT },
 889		.e_type = ET_EXEC,	/* yeah right */
 890		.e_machine = EM_PPC,
 891		.e_version = EV_CURRENT,
 892		.e_phoff = offsetof(struct fake_elf, phdr),
 893		.e_phentsize = sizeof(Elf32_Phdr),
 894		.e_phnum = 2
 895	},
 896	.phdr = {
 897		[0] = {
 898			.p_type = PT_NOTE,
 899			.p_offset = offsetof(struct fake_elf, chrpnote),
 900			.p_filesz = sizeof(struct chrpnote)
 901		}, [1] = {
 902			.p_type = PT_NOTE,
 903			.p_offset = offsetof(struct fake_elf, rpanote),
 904			.p_filesz = sizeof(struct rpanote)
 905		}
 906	},
 907	.chrpnote = {
 908		.namesz = sizeof("PowerPC"),
 909		.descsz = sizeof(struct chrpdesc),
 910		.type = 0x1275,
 911		.name = "PowerPC",
 912		.chrpdesc = {
 913			.real_mode = ~0U,	/* ~0 means "don't care" */
 914			.real_base = ~0U,
 915			.real_size = ~0U,
 916			.virt_base = ~0U,
 917			.virt_size = ~0U,
 918			.load_base = ~0U
 919		},
 920	},
 921	.rpanote = {
 922		.namesz = sizeof("IBM,RPA-Client-Config"),
 923		.descsz = sizeof(struct rpadesc),
 924		.type = 0x12759999,
 925		.name = "IBM,RPA-Client-Config",
 926		.rpadesc = {
 927			.lpar_affinity = 0,
 928			.min_rmo_size = 64,	/* in megabytes */
 929			.min_rmo_percent = 0,
 930			.max_pft_size = 48,	/* 2^48 bytes max PFT size */
 931			.splpar = 1,
 932			.min_load = ~0U,
 933			.new_mem_def = 0
 934		}
 935	}
 936};
 937#endif /* __BIG_ENDIAN__ */
 938
 939static int __init prom_count_smt_threads(void)
 940{
 941	phandle node;
 942	char type[64];
 943	unsigned int plen;
 944
 945	/* Pick up th first CPU node we can find */
 946	for (node = 0; prom_next_node(&node); ) {
 947		type[0] = 0;
 948		prom_getprop(node, "device_type", type, sizeof(type));
 949
 950		if (strcmp(type, "cpu"))
 951			continue;
 952		/*
 953		 * There is an entry for each smt thread, each entry being
 954		 * 4 bytes long.  All cpus should have the same number of
 955		 * smt threads, so return after finding the first.
 956		 */
 957		plen = prom_getproplen(node, "ibm,ppc-interrupt-server#s");
 958		if (plen == PROM_ERROR)
 959			break;
 960		plen >>= 2;
 961		prom_debug("Found %lu smt threads per core\n", (unsigned long)plen);
 962
 963		/* Sanity check */
 964		if (plen < 1 || plen > 64) {
 965			prom_printf("Threads per core %lu out of bounds, assuming 1\n",
 966				    (unsigned long)plen);
 967			return 1;
 968		}
 969		return plen;
 970	}
 971	prom_debug("No threads found, assuming 1 per core\n");
 972
 973	return 1;
 974
 975}
 976
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 977
 978static void __init prom_send_capabilities(void)
 979{
 980	ihandle root;
 981	prom_arg_t ret;
 982	u32 cores;
 983
 
 
 
 984	root = call_prom("open", 1, 1, ADDR("/"));
 985	if (root != 0) {
 986		/* We need to tell the FW about the number of cores we support.
 987		 *
 988		 * To do that, we count the number of threads on the first core
 989		 * (we assume this is the same for all cores) and use it to
 990		 * divide NR_CPUS.
 991		 */
 992
 993		cores = DIV_ROUND_UP(NR_CPUS, prom_count_smt_threads());
 994		prom_printf("Max number of cores passed to firmware: %lu (NR_CPUS = %lu)\n",
 995			    cores, NR_CPUS);
 996
 997		ibm_architecture_vec.vec5.max_cpus = cpu_to_be32(cores);
 998
 999		/* try calling the ibm,client-architecture-support method */
1000		prom_printf("Calling ibm,client-architecture-support...");
1001		if (call_prom_ret("call-method", 3, 2, &ret,
1002				  ADDR("ibm,client-architecture-support"),
1003				  root,
1004				  ADDR(&ibm_architecture_vec)) == 0) {
1005			/* the call exists... */
1006			if (ret)
1007				prom_printf("\nWARNING: ibm,client-architecture"
1008					    "-support call FAILED!\n");
1009			call_prom("close", 1, 0, root);
1010			prom_printf(" done\n");
1011			return;
1012		}
1013		call_prom("close", 1, 0, root);
1014		prom_printf(" not implemented\n");
1015	}
1016
1017#ifdef __BIG_ENDIAN__
1018	{
1019		ihandle elfloader;
1020
1021		/* no ibm,client-architecture-support call, try the old way */
1022		elfloader = call_prom("open", 1, 1,
1023				      ADDR("/packages/elf-loader"));
1024		if (elfloader == 0) {
1025			prom_printf("couldn't open /packages/elf-loader\n");
1026			return;
1027		}
1028		call_prom("call-method", 3, 1, ADDR("process-elf-header"),
1029			  elfloader, ADDR(&fake_elf));
1030		call_prom("close", 1, 0, elfloader);
1031	}
1032#endif /* __BIG_ENDIAN__ */
1033}
1034#endif /* #if defined(CONFIG_PPC_PSERIES) || defined(CONFIG_PPC_POWERNV) */
1035
1036/*
1037 * Memory allocation strategy... our layout is normally:
1038 *
1039 *  at 14Mb or more we have vmlinux, then a gap and initrd.  In some
1040 *  rare cases, initrd might end up being before the kernel though.
1041 *  We assume this won't override the final kernel at 0, we have no
1042 *  provision to handle that in this version, but it should hopefully
1043 *  never happen.
1044 *
1045 *  alloc_top is set to the top of RMO, eventually shrink down if the
1046 *  TCEs overlap
1047 *
1048 *  alloc_bottom is set to the top of kernel/initrd
1049 *
1050 *  from there, allocations are done this way : rtas is allocated
1051 *  topmost, and the device-tree is allocated from the bottom. We try
1052 *  to grow the device-tree allocation as we progress. If we can't,
1053 *  then we fail, we don't currently have a facility to restart
1054 *  elsewhere, but that shouldn't be necessary.
1055 *
1056 *  Note that calls to reserve_mem have to be done explicitly, memory
1057 *  allocated with either alloc_up or alloc_down isn't automatically
1058 *  reserved.
1059 */
1060
1061
1062/*
1063 * Allocates memory in the RMO upward from the kernel/initrd
1064 *
1065 * When align is 0, this is a special case, it means to allocate in place
1066 * at the current location of alloc_bottom or fail (that is basically
1067 * extending the previous allocation). Used for the device-tree flattening
1068 */
1069static unsigned long __init alloc_up(unsigned long size, unsigned long align)
1070{
1071	unsigned long base = alloc_bottom;
1072	unsigned long addr = 0;
1073
1074	if (align)
1075		base = _ALIGN_UP(base, align);
1076	prom_debug("alloc_up(%x, %x)\n", size, align);
1077	if (ram_top == 0)
1078		prom_panic("alloc_up() called with mem not initialized\n");
1079
1080	if (align)
1081		base = _ALIGN_UP(alloc_bottom, align);
1082	else
1083		base = alloc_bottom;
1084
1085	for(; (base + size) <= alloc_top; 
1086	    base = _ALIGN_UP(base + 0x100000, align)) {
1087		prom_debug("    trying: 0x%x\n\r", base);
1088		addr = (unsigned long)prom_claim(base, size, 0);
1089		if (addr != PROM_ERROR && addr != 0)
1090			break;
1091		addr = 0;
1092		if (align == 0)
1093			break;
1094	}
1095	if (addr == 0)
1096		return 0;
1097	alloc_bottom = addr + size;
1098
1099	prom_debug(" -> %x\n", addr);
1100	prom_debug("  alloc_bottom : %x\n", alloc_bottom);
1101	prom_debug("  alloc_top    : %x\n", alloc_top);
1102	prom_debug("  alloc_top_hi : %x\n", alloc_top_high);
1103	prom_debug("  rmo_top      : %x\n", rmo_top);
1104	prom_debug("  ram_top      : %x\n", ram_top);
1105
1106	return addr;
1107}
1108
1109/*
1110 * Allocates memory downward, either from top of RMO, or if highmem
1111 * is set, from the top of RAM.  Note that this one doesn't handle
1112 * failures.  It does claim memory if highmem is not set.
1113 */
1114static unsigned long __init alloc_down(unsigned long size, unsigned long align,
1115				       int highmem)
1116{
1117	unsigned long base, addr = 0;
1118
1119	prom_debug("alloc_down(%x, %x, %s)\n", size, align,
1120		   highmem ? "(high)" : "(low)");
1121	if (ram_top == 0)
1122		prom_panic("alloc_down() called with mem not initialized\n");
1123
1124	if (highmem) {
1125		/* Carve out storage for the TCE table. */
1126		addr = _ALIGN_DOWN(alloc_top_high - size, align);
1127		if (addr <= alloc_bottom)
1128			return 0;
1129		/* Will we bump into the RMO ? If yes, check out that we
1130		 * didn't overlap existing allocations there, if we did,
1131		 * we are dead, we must be the first in town !
1132		 */
1133		if (addr < rmo_top) {
1134			/* Good, we are first */
1135			if (alloc_top == rmo_top)
1136				alloc_top = rmo_top = addr;
1137			else
1138				return 0;
1139		}
1140		alloc_top_high = addr;
1141		goto bail;
1142	}
1143
1144	base = _ALIGN_DOWN(alloc_top - size, align);
1145	for (; base > alloc_bottom;
1146	     base = _ALIGN_DOWN(base - 0x100000, align))  {
1147		prom_debug("    trying: 0x%x\n\r", base);
1148		addr = (unsigned long)prom_claim(base, size, 0);
1149		if (addr != PROM_ERROR && addr != 0)
1150			break;
1151		addr = 0;
1152	}
1153	if (addr == 0)
1154		return 0;
1155	alloc_top = addr;
1156
1157 bail:
1158	prom_debug(" -> %x\n", addr);
1159	prom_debug("  alloc_bottom : %x\n", alloc_bottom);
1160	prom_debug("  alloc_top    : %x\n", alloc_top);
1161	prom_debug("  alloc_top_hi : %x\n", alloc_top_high);
1162	prom_debug("  rmo_top      : %x\n", rmo_top);
1163	prom_debug("  ram_top      : %x\n", ram_top);
1164
1165	return addr;
1166}
1167
1168/*
1169 * Parse a "reg" cell
1170 */
1171static unsigned long __init prom_next_cell(int s, cell_t **cellp)
1172{
1173	cell_t *p = *cellp;
1174	unsigned long r = 0;
1175
1176	/* Ignore more than 2 cells */
1177	while (s > sizeof(unsigned long) / 4) {
1178		p++;
1179		s--;
1180	}
1181	r = be32_to_cpu(*p++);
1182#ifdef CONFIG_PPC64
1183	if (s > 1) {
1184		r <<= 32;
1185		r |= be32_to_cpu(*(p++));
1186	}
1187#endif
1188	*cellp = p;
1189	return r;
1190}
1191
1192/*
1193 * Very dumb function for adding to the memory reserve list, but
1194 * we don't need anything smarter at this point
1195 *
1196 * XXX Eventually check for collisions.  They should NEVER happen.
1197 * If problems seem to show up, it would be a good start to track
1198 * them down.
1199 */
1200static void __init reserve_mem(u64 base, u64 size)
1201{
1202	u64 top = base + size;
1203	unsigned long cnt = mem_reserve_cnt;
1204
1205	if (size == 0)
1206		return;
1207
1208	/* We need to always keep one empty entry so that we
1209	 * have our terminator with "size" set to 0 since we are
1210	 * dumb and just copy this entire array to the boot params
1211	 */
1212	base = _ALIGN_DOWN(base, PAGE_SIZE);
1213	top = _ALIGN_UP(top, PAGE_SIZE);
1214	size = top - base;
1215
1216	if (cnt >= (MEM_RESERVE_MAP_SIZE - 1))
1217		prom_panic("Memory reserve map exhausted !\n");
1218	mem_reserve_map[cnt].base = cpu_to_be64(base);
1219	mem_reserve_map[cnt].size = cpu_to_be64(size);
1220	mem_reserve_cnt = cnt + 1;
1221}
1222
1223/*
1224 * Initialize memory allocation mechanism, parse "memory" nodes and
1225 * obtain that way the top of memory and RMO to setup out local allocator
1226 */
1227static void __init prom_init_mem(void)
1228{
1229	phandle node;
1230	char *path, type[64];
1231	unsigned int plen;
1232	cell_t *p, *endp;
1233	__be32 val;
1234	u32 rac, rsc;
1235
1236	/*
1237	 * We iterate the memory nodes to find
1238	 * 1) top of RMO (first node)
1239	 * 2) top of memory
1240	 */
1241	val = cpu_to_be32(2);
1242	prom_getprop(prom.root, "#address-cells", &val, sizeof(val));
1243	rac = be32_to_cpu(val);
1244	val = cpu_to_be32(1);
1245	prom_getprop(prom.root, "#size-cells", &val, sizeof(rsc));
1246	rsc = be32_to_cpu(val);
1247	prom_debug("root_addr_cells: %x\n", rac);
1248	prom_debug("root_size_cells: %x\n", rsc);
1249
1250	prom_debug("scanning memory:\n");
1251	path = prom_scratch;
1252
1253	for (node = 0; prom_next_node(&node); ) {
1254		type[0] = 0;
1255		prom_getprop(node, "device_type", type, sizeof(type));
1256
1257		if (type[0] == 0) {
1258			/*
1259			 * CHRP Longtrail machines have no device_type
1260			 * on the memory node, so check the name instead...
1261			 */
1262			prom_getprop(node, "name", type, sizeof(type));
1263		}
1264		if (strcmp(type, "memory"))
1265			continue;
1266
1267		plen = prom_getprop(node, "reg", regbuf, sizeof(regbuf));
1268		if (plen > sizeof(regbuf)) {
1269			prom_printf("memory node too large for buffer !\n");
1270			plen = sizeof(regbuf);
1271		}
1272		p = regbuf;
1273		endp = p + (plen / sizeof(cell_t));
1274
1275#ifdef DEBUG_PROM
1276		memset(path, 0, PROM_SCRATCH_SIZE);
1277		call_prom("package-to-path", 3, 1, node, path, PROM_SCRATCH_SIZE-1);
1278		prom_debug("  node %s :\n", path);
 
1279#endif /* DEBUG_PROM */
1280
1281		while ((endp - p) >= (rac + rsc)) {
1282			unsigned long base, size;
1283
1284			base = prom_next_cell(rac, &p);
1285			size = prom_next_cell(rsc, &p);
1286
1287			if (size == 0)
1288				continue;
1289			prom_debug("    %x %x\n", base, size);
1290			if (base == 0 && (of_platform & PLATFORM_LPAR))
1291				rmo_top = size;
1292			if ((base + size) > ram_top)
1293				ram_top = base + size;
1294		}
1295	}
1296
1297	alloc_bottom = PAGE_ALIGN((unsigned long)&_end + 0x4000);
1298
1299	/*
1300	 * If prom_memory_limit is set we reduce the upper limits *except* for
1301	 * alloc_top_high. This must be the real top of RAM so we can put
1302	 * TCE's up there.
1303	 */
1304
1305	alloc_top_high = ram_top;
1306
1307	if (prom_memory_limit) {
1308		if (prom_memory_limit <= alloc_bottom) {
1309			prom_printf("Ignoring mem=%x <= alloc_bottom.\n",
1310				prom_memory_limit);
1311			prom_memory_limit = 0;
1312		} else if (prom_memory_limit >= ram_top) {
1313			prom_printf("Ignoring mem=%x >= ram_top.\n",
1314				prom_memory_limit);
1315			prom_memory_limit = 0;
1316		} else {
1317			ram_top = prom_memory_limit;
1318			rmo_top = min(rmo_top, prom_memory_limit);
1319		}
1320	}
1321
1322	/*
1323	 * Setup our top alloc point, that is top of RMO or top of
1324	 * segment 0 when running non-LPAR.
1325	 * Some RS64 machines have buggy firmware where claims up at
1326	 * 1GB fail.  Cap at 768MB as a workaround.
1327	 * Since 768MB is plenty of room, and we need to cap to something
1328	 * reasonable on 32-bit, cap at 768MB on all machines.
1329	 */
1330	if (!rmo_top)
1331		rmo_top = ram_top;
1332	rmo_top = min(0x30000000ul, rmo_top);
1333	alloc_top = rmo_top;
1334	alloc_top_high = ram_top;
1335
1336	/*
1337	 * Check if we have an initrd after the kernel but still inside
1338	 * the RMO.  If we do move our bottom point to after it.
1339	 */
1340	if (prom_initrd_start &&
1341	    prom_initrd_start < rmo_top &&
1342	    prom_initrd_end > alloc_bottom)
1343		alloc_bottom = PAGE_ALIGN(prom_initrd_end);
1344
1345	prom_printf("memory layout at init:\n");
1346	prom_printf("  memory_limit : %x (16 MB aligned)\n", prom_memory_limit);
1347	prom_printf("  alloc_bottom : %x\n", alloc_bottom);
1348	prom_printf("  alloc_top    : %x\n", alloc_top);
1349	prom_printf("  alloc_top_hi : %x\n", alloc_top_high);
1350	prom_printf("  rmo_top      : %x\n", rmo_top);
1351	prom_printf("  ram_top      : %x\n", ram_top);
 
1352}
1353
1354static void __init prom_close_stdin(void)
1355{
1356	__be32 val;
1357	ihandle stdin;
1358
1359	if (prom_getprop(prom.chosen, "stdin", &val, sizeof(val)) > 0) {
1360		stdin = be32_to_cpu(val);
1361		call_prom("close", 1, 0, stdin);
1362	}
1363}
1364
1365#ifdef CONFIG_PPC_POWERNV
1366
1367#ifdef CONFIG_PPC_EARLY_DEBUG_OPAL
1368static u64 __initdata prom_opal_base;
1369static u64 __initdata prom_opal_entry;
1370#endif
1371
1372/*
1373 * Allocate room for and instantiate OPAL
1374 */
1375static void __init prom_instantiate_opal(void)
1376{
1377	phandle opal_node;
1378	ihandle opal_inst;
1379	u64 base, entry;
1380	u64 size = 0, align = 0x10000;
1381	__be64 val64;
1382	u32 rets[2];
1383
1384	prom_debug("prom_instantiate_opal: start...\n");
 
 
 
1385
1386	opal_node = call_prom("finddevice", 1, 1, ADDR("/ibm,opal"));
1387	prom_debug("opal_node: %x\n", opal_node);
1388	if (!PHANDLE_VALID(opal_node))
1389		return;
1390
1391	val64 = 0;
1392	prom_getprop(opal_node, "opal-runtime-size", &val64, sizeof(val64));
1393	size = be64_to_cpu(val64);
1394	if (size == 0)
1395		return;
1396	val64 = 0;
1397	prom_getprop(opal_node, "opal-runtime-alignment", &val64,sizeof(val64));
1398	align = be64_to_cpu(val64);
1399
1400	base = alloc_down(size, align, 0);
1401	if (base == 0) {
1402		prom_printf("OPAL allocation failed !\n");
1403		return;
1404	}
1405
1406	opal_inst = call_prom("open", 1, 1, ADDR("/ibm,opal"));
1407	if (!IHANDLE_VALID(opal_inst)) {
1408		prom_printf("opening opal package failed (%x)\n", opal_inst);
 
1409		return;
1410	}
1411
1412	prom_printf("instantiating opal at 0x%x...", base);
1413
1414	if (call_prom_ret("call-method", 4, 3, rets,
1415			  ADDR("load-opal-runtime"),
1416			  opal_inst,
1417			  base >> 32, base & 0xffffffff) != 0
1418	    || (rets[0] == 0 && rets[1] == 0)) {
1419		prom_printf(" failed\n");
1420		return;
1421	}
1422	entry = (((u64)rets[0]) << 32) | rets[1];
1423
1424	prom_printf(" done\n");
1425
1426	reserve_mem(base, size);
1427
1428	prom_debug("opal base     = 0x%x\n", base);
1429	prom_debug("opal align    = 0x%x\n", align);
1430	prom_debug("opal entry    = 0x%x\n", entry);
1431	prom_debug("opal size     = 0x%x\n", (long)size);
1432
1433	prom_setprop(opal_node, "/ibm,opal", "opal-base-address",
1434		     &base, sizeof(base));
1435	prom_setprop(opal_node, "/ibm,opal", "opal-entry-address",
1436		     &entry, sizeof(entry));
1437
1438#ifdef CONFIG_PPC_EARLY_DEBUG_OPAL
1439	prom_opal_base = base;
1440	prom_opal_entry = entry;
1441#endif
1442	prom_debug("prom_instantiate_opal: end...\n");
1443}
1444
1445#endif /* CONFIG_PPC_POWERNV */
1446
1447/*
1448 * Allocate room for and instantiate RTAS
1449 */
1450static void __init prom_instantiate_rtas(void)
1451{
1452	phandle rtas_node;
1453	ihandle rtas_inst;
1454	u32 base, entry = 0;
1455	__be32 val;
1456	u32 size = 0;
1457
1458	prom_debug("prom_instantiate_rtas: start...\n");
1459
1460	rtas_node = call_prom("finddevice", 1, 1, ADDR("/rtas"));
1461	prom_debug("rtas_node: %x\n", rtas_node);
1462	if (!PHANDLE_VALID(rtas_node))
1463		return;
1464
1465	val = 0;
1466	prom_getprop(rtas_node, "rtas-size", &val, sizeof(size));
1467	size = be32_to_cpu(val);
1468	if (size == 0)
1469		return;
1470
1471	base = alloc_down(size, PAGE_SIZE, 0);
1472	if (base == 0)
1473		prom_panic("Could not allocate memory for RTAS\n");
1474
1475	rtas_inst = call_prom("open", 1, 1, ADDR("/rtas"));
1476	if (!IHANDLE_VALID(rtas_inst)) {
1477		prom_printf("opening rtas package failed (%x)\n", rtas_inst);
1478		return;
1479	}
1480
1481	prom_printf("instantiating rtas at 0x%x...", base);
1482
1483	if (call_prom_ret("call-method", 3, 2, &entry,
1484			  ADDR("instantiate-rtas"),
1485			  rtas_inst, base) != 0
1486	    || entry == 0) {
1487		prom_printf(" failed\n");
1488		return;
1489	}
1490	prom_printf(" done\n");
1491
1492	reserve_mem(base, size);
1493
1494	val = cpu_to_be32(base);
1495	prom_setprop(rtas_node, "/rtas", "linux,rtas-base",
1496		     &val, sizeof(val));
1497	val = cpu_to_be32(entry);
1498	prom_setprop(rtas_node, "/rtas", "linux,rtas-entry",
1499		     &val, sizeof(val));
1500
1501	/* Check if it supports "query-cpu-stopped-state" */
1502	if (prom_getprop(rtas_node, "query-cpu-stopped-state",
1503			 &val, sizeof(val)) != PROM_ERROR)
1504		rtas_has_query_cpu_stopped = true;
1505
1506	prom_debug("rtas base     = 0x%x\n", base);
1507	prom_debug("rtas entry    = 0x%x\n", entry);
1508	prom_debug("rtas size     = 0x%x\n", (long)size);
1509
1510	prom_debug("prom_instantiate_rtas: end...\n");
1511}
1512
1513#ifdef CONFIG_PPC64
1514/*
1515 * Allocate room for and instantiate Stored Measurement Log (SML)
1516 */
1517static void __init prom_instantiate_sml(void)
1518{
1519	phandle ibmvtpm_node;
1520	ihandle ibmvtpm_inst;
1521	u32 entry = 0, size = 0, succ = 0;
1522	u64 base;
1523	__be32 val;
1524
1525	prom_debug("prom_instantiate_sml: start...\n");
1526
1527	ibmvtpm_node = call_prom("finddevice", 1, 1, ADDR("/vdevice/vtpm"));
1528	prom_debug("ibmvtpm_node: %x\n", ibmvtpm_node);
1529	if (!PHANDLE_VALID(ibmvtpm_node))
1530		return;
1531
1532	ibmvtpm_inst = call_prom("open", 1, 1, ADDR("/vdevice/vtpm"));
1533	if (!IHANDLE_VALID(ibmvtpm_inst)) {
1534		prom_printf("opening vtpm package failed (%x)\n", ibmvtpm_inst);
1535		return;
1536	}
1537
1538	if (prom_getprop(ibmvtpm_node, "ibm,sml-efi-reformat-supported",
1539			 &val, sizeof(val)) != PROM_ERROR) {
1540		if (call_prom_ret("call-method", 2, 2, &succ,
1541				  ADDR("reformat-sml-to-efi-alignment"),
1542				  ibmvtpm_inst) != 0 || succ == 0) {
1543			prom_printf("Reformat SML to EFI alignment failed\n");
1544			return;
1545		}
1546
1547		if (call_prom_ret("call-method", 2, 2, &size,
1548				  ADDR("sml-get-allocated-size"),
1549				  ibmvtpm_inst) != 0 || size == 0) {
1550			prom_printf("SML get allocated size failed\n");
1551			return;
1552		}
1553	} else {
1554		if (call_prom_ret("call-method", 2, 2, &size,
1555				  ADDR("sml-get-handover-size"),
1556				  ibmvtpm_inst) != 0 || size == 0) {
1557			prom_printf("SML get handover size failed\n");
1558			return;
1559		}
1560	}
1561
1562	base = alloc_down(size, PAGE_SIZE, 0);
1563	if (base == 0)
1564		prom_panic("Could not allocate memory for sml\n");
1565
1566	prom_printf("instantiating sml at 0x%x...", base);
1567
1568	memset((void *)base, 0, size);
1569
1570	if (call_prom_ret("call-method", 4, 2, &entry,
1571			  ADDR("sml-handover"),
1572			  ibmvtpm_inst, size, base) != 0 || entry == 0) {
1573		prom_printf("SML handover failed\n");
1574		return;
1575	}
1576	prom_printf(" done\n");
1577
1578	reserve_mem(base, size);
1579
1580	prom_setprop(ibmvtpm_node, "/vdevice/vtpm", "linux,sml-base",
1581		     &base, sizeof(base));
1582	prom_setprop(ibmvtpm_node, "/vdevice/vtpm", "linux,sml-size",
1583		     &size, sizeof(size));
1584
1585	prom_debug("sml base     = 0x%x\n", base);
1586	prom_debug("sml size     = 0x%x\n", (long)size);
1587
1588	prom_debug("prom_instantiate_sml: end...\n");
1589}
1590
1591/*
1592 * Allocate room for and initialize TCE tables
1593 */
1594#ifdef __BIG_ENDIAN__
1595static void __init prom_initialize_tce_table(void)
1596{
1597	phandle node;
1598	ihandle phb_node;
1599	char compatible[64], type[64], model[64];
1600	char *path = prom_scratch;
1601	u64 base, align;
1602	u32 minalign, minsize;
1603	u64 tce_entry, *tce_entryp;
1604	u64 local_alloc_top, local_alloc_bottom;
1605	u64 i;
1606
1607	if (prom_iommu_off)
1608		return;
1609
1610	prom_debug("starting prom_initialize_tce_table\n");
1611
1612	/* Cache current top of allocs so we reserve a single block */
1613	local_alloc_top = alloc_top_high;
1614	local_alloc_bottom = local_alloc_top;
1615
1616	/* Search all nodes looking for PHBs. */
1617	for (node = 0; prom_next_node(&node); ) {
1618		compatible[0] = 0;
1619		type[0] = 0;
1620		model[0] = 0;
1621		prom_getprop(node, "compatible",
1622			     compatible, sizeof(compatible));
1623		prom_getprop(node, "device_type", type, sizeof(type));
1624		prom_getprop(node, "model", model, sizeof(model));
1625
1626		if ((type[0] == 0) || (strstr(type, "pci") == NULL))
1627			continue;
1628
1629		/* Keep the old logic intact to avoid regression. */
1630		if (compatible[0] != 0) {
1631			if ((strstr(compatible, "python") == NULL) &&
1632			    (strstr(compatible, "Speedwagon") == NULL) &&
1633			    (strstr(compatible, "Winnipeg") == NULL))
1634				continue;
1635		} else if (model[0] != 0) {
1636			if ((strstr(model, "ython") == NULL) &&
1637			    (strstr(model, "peedwagon") == NULL) &&
1638			    (strstr(model, "innipeg") == NULL))
1639				continue;
1640		}
1641
1642		if (prom_getprop(node, "tce-table-minalign", &minalign,
1643				 sizeof(minalign)) == PROM_ERROR)
1644			minalign = 0;
1645		if (prom_getprop(node, "tce-table-minsize", &minsize,
1646				 sizeof(minsize)) == PROM_ERROR)
1647			minsize = 4UL << 20;
1648
1649		/*
1650		 * Even though we read what OF wants, we just set the table
1651		 * size to 4 MB.  This is enough to map 2GB of PCI DMA space.
1652		 * By doing this, we avoid the pitfalls of trying to DMA to
1653		 * MMIO space and the DMA alias hole.
1654		 *
1655		 * On POWER4, firmware sets the TCE region by assuming
1656		 * each TCE table is 8MB. Using this memory for anything
1657		 * else will impact performance, so we always allocate 8MB.
1658		 * Anton
1659		 */
1660		if (pvr_version_is(PVR_POWER4) || pvr_version_is(PVR_POWER4p))
1661			minsize = 8UL << 20;
1662		else
1663			minsize = 4UL << 20;
1664
1665		/* Align to the greater of the align or size */
1666		align = max(minalign, minsize);
1667		base = alloc_down(minsize, align, 1);
1668		if (base == 0)
1669			prom_panic("ERROR, cannot find space for TCE table.\n");
1670		if (base < local_alloc_bottom)
1671			local_alloc_bottom = base;
1672
1673		/* It seems OF doesn't null-terminate the path :-( */
1674		memset(path, 0, PROM_SCRATCH_SIZE);
1675		/* Call OF to setup the TCE hardware */
1676		if (call_prom("package-to-path", 3, 1, node,
1677			      path, PROM_SCRATCH_SIZE-1) == PROM_ERROR) {
1678			prom_printf("package-to-path failed\n");
1679		}
1680
1681		/* Save away the TCE table attributes for later use. */
1682		prom_setprop(node, path, "linux,tce-base", &base, sizeof(base));
1683		prom_setprop(node, path, "linux,tce-size", &minsize, sizeof(minsize));
1684
1685		prom_debug("TCE table: %s\n", path);
1686		prom_debug("\tnode = 0x%x\n", node);
1687		prom_debug("\tbase = 0x%x\n", base);
1688		prom_debug("\tsize = 0x%x\n", minsize);
1689
1690		/* Initialize the table to have a one-to-one mapping
1691		 * over the allocated size.
1692		 */
1693		tce_entryp = (u64 *)base;
1694		for (i = 0; i < (minsize >> 3) ;tce_entryp++, i++) {
1695			tce_entry = (i << PAGE_SHIFT);
1696			tce_entry |= 0x3;
1697			*tce_entryp = tce_entry;
1698		}
1699
1700		prom_printf("opening PHB %s", path);
1701		phb_node = call_prom("open", 1, 1, path);
1702		if (phb_node == 0)
1703			prom_printf("... failed\n");
1704		else
1705			prom_printf("... done\n");
1706
1707		call_prom("call-method", 6, 0, ADDR("set-64-bit-addressing"),
1708			  phb_node, -1, minsize,
1709			  (u32) base, (u32) (base >> 32));
1710		call_prom("close", 1, 0, phb_node);
1711	}
1712
1713	reserve_mem(local_alloc_bottom, local_alloc_top - local_alloc_bottom);
1714
1715	/* These are only really needed if there is a memory limit in
1716	 * effect, but we don't know so export them always. */
1717	prom_tce_alloc_start = local_alloc_bottom;
1718	prom_tce_alloc_end = local_alloc_top;
1719
1720	/* Flag the first invalid entry */
1721	prom_debug("ending prom_initialize_tce_table\n");
1722}
1723#endif /* __BIG_ENDIAN__ */
1724#endif /* CONFIG_PPC64 */
1725
1726/*
1727 * With CHRP SMP we need to use the OF to start the other processors.
1728 * We can't wait until smp_boot_cpus (the OF is trashed by then)
1729 * so we have to put the processors into a holding pattern controlled
1730 * by the kernel (not OF) before we destroy the OF.
1731 *
1732 * This uses a chunk of low memory, puts some holding pattern
1733 * code there and sends the other processors off to there until
1734 * smp_boot_cpus tells them to do something.  The holding pattern
1735 * checks that address until its cpu # is there, when it is that
1736 * cpu jumps to __secondary_start().  smp_boot_cpus() takes care
1737 * of setting those values.
1738 *
1739 * We also use physical address 0x4 here to tell when a cpu
1740 * is in its holding pattern code.
1741 *
1742 * -- Cort
1743 */
1744/*
1745 * We want to reference the copy of __secondary_hold_* in the
1746 * 0 - 0x100 address range
1747 */
1748#define LOW_ADDR(x)	(((unsigned long) &(x)) & 0xff)
1749
1750static void __init prom_hold_cpus(void)
1751{
1752	unsigned long i;
1753	phandle node;
1754	char type[64];
1755	unsigned long *spinloop
1756		= (void *) LOW_ADDR(__secondary_hold_spinloop);
1757	unsigned long *acknowledge
1758		= (void *) LOW_ADDR(__secondary_hold_acknowledge);
1759	unsigned long secondary_hold = LOW_ADDR(__secondary_hold);
1760
1761	/*
1762	 * On pseries, if RTAS supports "query-cpu-stopped-state",
1763	 * we skip this stage, the CPUs will be started by the
1764	 * kernel using RTAS.
1765	 */
1766	if ((of_platform == PLATFORM_PSERIES ||
1767	     of_platform == PLATFORM_PSERIES_LPAR) &&
1768	    rtas_has_query_cpu_stopped) {
1769		prom_printf("prom_hold_cpus: skipped\n");
1770		return;
1771	}
1772
1773	prom_debug("prom_hold_cpus: start...\n");
1774	prom_debug("    1) spinloop       = 0x%x\n", (unsigned long)spinloop);
1775	prom_debug("    1) *spinloop      = 0x%x\n", *spinloop);
1776	prom_debug("    1) acknowledge    = 0x%x\n",
1777		   (unsigned long)acknowledge);
1778	prom_debug("    1) *acknowledge   = 0x%x\n", *acknowledge);
1779	prom_debug("    1) secondary_hold = 0x%x\n", secondary_hold);
1780
1781	/* Set the common spinloop variable, so all of the secondary cpus
1782	 * will block when they are awakened from their OF spinloop.
1783	 * This must occur for both SMP and non SMP kernels, since OF will
1784	 * be trashed when we move the kernel.
1785	 */
1786	*spinloop = 0;
1787
1788	/* look for cpus */
1789	for (node = 0; prom_next_node(&node); ) {
1790		unsigned int cpu_no;
1791		__be32 reg;
1792
1793		type[0] = 0;
1794		prom_getprop(node, "device_type", type, sizeof(type));
1795		if (strcmp(type, "cpu") != 0)
1796			continue;
1797
1798		/* Skip non-configured cpus. */
1799		if (prom_getprop(node, "status", type, sizeof(type)) > 0)
1800			if (strcmp(type, "okay") != 0)
1801				continue;
1802
1803		reg = cpu_to_be32(-1); /* make sparse happy */
1804		prom_getprop(node, "reg", &reg, sizeof(reg));
1805		cpu_no = be32_to_cpu(reg);
1806
1807		prom_debug("cpu hw idx   = %lu\n", cpu_no);
1808
1809		/* Init the acknowledge var which will be reset by
1810		 * the secondary cpu when it awakens from its OF
1811		 * spinloop.
1812		 */
1813		*acknowledge = (unsigned long)-1;
1814
1815		if (cpu_no != prom.cpu) {
1816			/* Primary Thread of non-boot cpu or any thread */
1817			prom_printf("starting cpu hw idx %lu... ", cpu_no);
1818			call_prom("start-cpu", 3, 0, node,
1819				  secondary_hold, cpu_no);
1820
1821			for (i = 0; (i < 100000000) && 
1822			     (*acknowledge == ((unsigned long)-1)); i++ )
1823				mb();
1824
1825			if (*acknowledge == cpu_no)
1826				prom_printf("done\n");
1827			else
1828				prom_printf("failed: %x\n", *acknowledge);
1829		}
1830#ifdef CONFIG_SMP
1831		else
1832			prom_printf("boot cpu hw idx %lu\n", cpu_no);
1833#endif /* CONFIG_SMP */
1834	}
1835
1836	prom_debug("prom_hold_cpus: end...\n");
1837}
1838
1839
1840static void __init prom_init_client_services(unsigned long pp)
1841{
1842	/* Get a handle to the prom entry point before anything else */
1843	prom_entry = pp;
1844
1845	/* get a handle for the stdout device */
1846	prom.chosen = call_prom("finddevice", 1, 1, ADDR("/chosen"));
1847	if (!PHANDLE_VALID(prom.chosen))
1848		prom_panic("cannot find chosen"); /* msg won't be printed :( */
1849
1850	/* get device tree root */
1851	prom.root = call_prom("finddevice", 1, 1, ADDR("/"));
1852	if (!PHANDLE_VALID(prom.root))
1853		prom_panic("cannot find device tree root"); /* msg won't be printed :( */
1854
1855	prom.mmumap = 0;
1856}
1857
1858#ifdef CONFIG_PPC32
1859/*
1860 * For really old powermacs, we need to map things we claim.
1861 * For that, we need the ihandle of the mmu.
1862 * Also, on the longtrail, we need to work around other bugs.
1863 */
1864static void __init prom_find_mmu(void)
1865{
1866	phandle oprom;
1867	char version[64];
1868
1869	oprom = call_prom("finddevice", 1, 1, ADDR("/openprom"));
1870	if (!PHANDLE_VALID(oprom))
1871		return;
1872	if (prom_getprop(oprom, "model", version, sizeof(version)) <= 0)
1873		return;
1874	version[sizeof(version) - 1] = 0;
1875	/* XXX might need to add other versions here */
1876	if (strcmp(version, "Open Firmware, 1.0.5") == 0)
1877		of_workarounds = OF_WA_CLAIM;
1878	else if (strncmp(version, "FirmWorks,3.", 12) == 0) {
1879		of_workarounds = OF_WA_CLAIM | OF_WA_LONGTRAIL;
1880		call_prom("interpret", 1, 1, "dev /memory 0 to allow-reclaim");
1881	} else
1882		return;
1883	prom.memory = call_prom("open", 1, 1, ADDR("/memory"));
1884	prom_getprop(prom.chosen, "mmu", &prom.mmumap,
1885		     sizeof(prom.mmumap));
1886	prom.mmumap = be32_to_cpu(prom.mmumap);
1887	if (!IHANDLE_VALID(prom.memory) || !IHANDLE_VALID(prom.mmumap))
1888		of_workarounds &= ~OF_WA_CLAIM;		/* hmmm */
1889}
1890#else
1891#define prom_find_mmu()
1892#endif
1893
1894static void __init prom_init_stdout(void)
1895{
1896	char *path = of_stdout_device;
1897	char type[16];
1898	phandle stdout_node;
1899	__be32 val;
1900
1901	if (prom_getprop(prom.chosen, "stdout", &val, sizeof(val)) <= 0)
1902		prom_panic("cannot find stdout");
1903
1904	prom.stdout = be32_to_cpu(val);
1905
1906	/* Get the full OF pathname of the stdout device */
1907	memset(path, 0, 256);
1908	call_prom("instance-to-path", 3, 1, prom.stdout, path, 255);
1909	prom_printf("OF stdout device is: %s\n", of_stdout_device);
1910	prom_setprop(prom.chosen, "/chosen", "linux,stdout-path",
1911		     path, strlen(path) + 1);
1912
1913	/* instance-to-package fails on PA-Semi */
1914	stdout_node = call_prom("instance-to-package", 1, 1, prom.stdout);
1915	if (stdout_node != PROM_ERROR) {
1916		val = cpu_to_be32(stdout_node);
1917		prom_setprop(prom.chosen, "/chosen", "linux,stdout-package",
1918			     &val, sizeof(val));
1919
1920		/* If it's a display, note it */
1921		memset(type, 0, sizeof(type));
1922		prom_getprop(stdout_node, "device_type", type, sizeof(type));
1923		if (strcmp(type, "display") == 0)
1924			prom_setprop(stdout_node, path, "linux,boot-display", NULL, 0);
1925	}
1926}
1927
1928static int __init prom_find_machine_type(void)
1929{
1930	char compat[256];
1931	int len, i = 0;
1932#ifdef CONFIG_PPC64
1933	phandle rtas;
1934	int x;
1935#endif
1936
1937	/* Look for a PowerMac or a Cell */
1938	len = prom_getprop(prom.root, "compatible",
1939			   compat, sizeof(compat)-1);
1940	if (len > 0) {
1941		compat[len] = 0;
1942		while (i < len) {
1943			char *p = &compat[i];
1944			int sl = strlen(p);
1945			if (sl == 0)
1946				break;
1947			if (strstr(p, "Power Macintosh") ||
1948			    strstr(p, "MacRISC"))
1949				return PLATFORM_POWERMAC;
1950#ifdef CONFIG_PPC64
1951			/* We must make sure we don't detect the IBM Cell
1952			 * blades as pSeries due to some firmware issues,
1953			 * so we do it here.
1954			 */
1955			if (strstr(p, "IBM,CBEA") ||
1956			    strstr(p, "IBM,CPBW-1.0"))
1957				return PLATFORM_GENERIC;
1958#endif /* CONFIG_PPC64 */
1959			i += sl + 1;
1960		}
1961	}
1962#ifdef CONFIG_PPC64
1963	/* Try to detect OPAL */
1964	if (PHANDLE_VALID(call_prom("finddevice", 1, 1, ADDR("/ibm,opal"))))
1965		return PLATFORM_OPAL;
1966
1967	/* Try to figure out if it's an IBM pSeries or any other
1968	 * PAPR compliant platform. We assume it is if :
1969	 *  - /device_type is "chrp" (please, do NOT use that for future
1970	 *    non-IBM designs !
1971	 *  - it has /rtas
1972	 */
1973	len = prom_getprop(prom.root, "device_type",
1974			   compat, sizeof(compat)-1);
1975	if (len <= 0)
1976		return PLATFORM_GENERIC;
1977	if (strcmp(compat, "chrp"))
1978		return PLATFORM_GENERIC;
1979
1980	/* Default to pSeries. We need to know if we are running LPAR */
1981	rtas = call_prom("finddevice", 1, 1, ADDR("/rtas"));
1982	if (!PHANDLE_VALID(rtas))
1983		return PLATFORM_GENERIC;
1984	x = prom_getproplen(rtas, "ibm,hypertas-functions");
1985	if (x != PROM_ERROR) {
1986		prom_debug("Hypertas detected, assuming LPAR !\n");
1987		return PLATFORM_PSERIES_LPAR;
1988	}
1989	return PLATFORM_PSERIES;
1990#else
1991	return PLATFORM_GENERIC;
1992#endif
1993}
1994
1995static int __init prom_set_color(ihandle ih, int i, int r, int g, int b)
1996{
1997	return call_prom("call-method", 6, 1, ADDR("color!"), ih, i, b, g, r);
1998}
1999
2000/*
2001 * If we have a display that we don't know how to drive,
2002 * we will want to try to execute OF's open method for it
2003 * later.  However, OF will probably fall over if we do that
2004 * we've taken over the MMU.
2005 * So we check whether we will need to open the display,
2006 * and if so, open it now.
2007 */
2008static void __init prom_check_displays(void)
2009{
2010	char type[16], *path;
2011	phandle node;
2012	ihandle ih;
2013	int i;
2014
2015	static unsigned char default_colors[] = {
2016		0x00, 0x00, 0x00,
2017		0x00, 0x00, 0xaa,
2018		0x00, 0xaa, 0x00,
2019		0x00, 0xaa, 0xaa,
2020		0xaa, 0x00, 0x00,
2021		0xaa, 0x00, 0xaa,
2022		0xaa, 0xaa, 0x00,
2023		0xaa, 0xaa, 0xaa,
2024		0x55, 0x55, 0x55,
2025		0x55, 0x55, 0xff,
2026		0x55, 0xff, 0x55,
2027		0x55, 0xff, 0xff,
2028		0xff, 0x55, 0x55,
2029		0xff, 0x55, 0xff,
2030		0xff, 0xff, 0x55,
2031		0xff, 0xff, 0xff
2032	};
2033	const unsigned char *clut;
2034
2035	prom_debug("Looking for displays\n");
2036	for (node = 0; prom_next_node(&node); ) {
2037		memset(type, 0, sizeof(type));
2038		prom_getprop(node, "device_type", type, sizeof(type));
2039		if (strcmp(type, "display") != 0)
2040			continue;
2041
2042		/* It seems OF doesn't null-terminate the path :-( */
2043		path = prom_scratch;
2044		memset(path, 0, PROM_SCRATCH_SIZE);
2045
2046		/*
2047		 * leave some room at the end of the path for appending extra
2048		 * arguments
2049		 */
2050		if (call_prom("package-to-path", 3, 1, node, path,
2051			      PROM_SCRATCH_SIZE-10) == PROM_ERROR)
2052			continue;
2053		prom_printf("found display   : %s, opening... ", path);
2054		
2055		ih = call_prom("open", 1, 1, path);
2056		if (ih == 0) {
2057			prom_printf("failed\n");
2058			continue;
2059		}
2060
2061		/* Success */
2062		prom_printf("done\n");
2063		prom_setprop(node, path, "linux,opened", NULL, 0);
2064
2065		/* Setup a usable color table when the appropriate
2066		 * method is available. Should update this to set-colors */
2067		clut = default_colors;
2068		for (i = 0; i < 16; i++, clut += 3)
2069			if (prom_set_color(ih, i, clut[0], clut[1],
2070					   clut[2]) != 0)
2071				break;
2072
2073#ifdef CONFIG_LOGO_LINUX_CLUT224
2074		clut = PTRRELOC(logo_linux_clut224.clut);
2075		for (i = 0; i < logo_linux_clut224.clutsize; i++, clut += 3)
2076			if (prom_set_color(ih, i + 32, clut[0], clut[1],
2077					   clut[2]) != 0)
2078				break;
2079#endif /* CONFIG_LOGO_LINUX_CLUT224 */
2080
2081#ifdef CONFIG_PPC_EARLY_DEBUG_BOOTX
2082		if (prom_getprop(node, "linux,boot-display", NULL, 0) !=
2083		    PROM_ERROR) {
2084			u32 width, height, pitch, addr;
2085
2086			prom_printf("Setting btext !\n");
2087			prom_getprop(node, "width", &width, 4);
2088			prom_getprop(node, "height", &height, 4);
2089			prom_getprop(node, "linebytes", &pitch, 4);
2090			prom_getprop(node, "address", &addr, 4);
 
 
 
 
 
 
 
 
 
2091			prom_printf("W=%d H=%d LB=%d addr=0x%x\n",
2092				    width, height, pitch, addr);
2093			btext_setup_display(width, height, 8, pitch, addr);
 
2094		}
2095#endif /* CONFIG_PPC_EARLY_DEBUG_BOOTX */
2096	}
2097}
2098
2099
2100/* Return (relocated) pointer to this much memory: moves initrd if reqd. */
2101static void __init *make_room(unsigned long *mem_start, unsigned long *mem_end,
2102			      unsigned long needed, unsigned long align)
2103{
2104	void *ret;
2105
2106	*mem_start = _ALIGN(*mem_start, align);
2107	while ((*mem_start + needed) > *mem_end) {
2108		unsigned long room, chunk;
2109
2110		prom_debug("Chunk exhausted, claiming more at %x...\n",
2111			   alloc_bottom);
2112		room = alloc_top - alloc_bottom;
2113		if (room > DEVTREE_CHUNK_SIZE)
2114			room = DEVTREE_CHUNK_SIZE;
2115		if (room < PAGE_SIZE)
2116			prom_panic("No memory for flatten_device_tree "
2117				   "(no room)\n");
2118		chunk = alloc_up(room, 0);
2119		if (chunk == 0)
2120			prom_panic("No memory for flatten_device_tree "
2121				   "(claim failed)\n");
2122		*mem_end = chunk + room;
2123	}
2124
2125	ret = (void *)*mem_start;
2126	*mem_start += needed;
2127
2128	return ret;
2129}
2130
2131#define dt_push_token(token, mem_start, mem_end) do { 			\
2132		void *room = make_room(mem_start, mem_end, 4, 4);	\
2133		*(__be32 *)room = cpu_to_be32(token);			\
2134	} while(0)
2135
2136static unsigned long __init dt_find_string(char *str)
2137{
2138	char *s, *os;
2139
2140	s = os = (char *)dt_string_start;
2141	s += 4;
2142	while (s <  (char *)dt_string_end) {
2143		if (strcmp(s, str) == 0)
2144			return s - os;
2145		s += strlen(s) + 1;
2146	}
2147	return 0;
2148}
2149
2150/*
2151 * The Open Firmware 1275 specification states properties must be 31 bytes or
2152 * less, however not all firmwares obey this. Make it 64 bytes to be safe.
2153 */
2154#define MAX_PROPERTY_NAME 64
2155
2156static void __init scan_dt_build_strings(phandle node,
2157					 unsigned long *mem_start,
2158					 unsigned long *mem_end)
2159{
2160	char *prev_name, *namep, *sstart;
2161	unsigned long soff;
2162	phandle child;
2163
2164	sstart =  (char *)dt_string_start;
2165
2166	/* get and store all property names */
2167	prev_name = "";
2168	for (;;) {
2169		/* 64 is max len of name including nul. */
2170		namep = make_room(mem_start, mem_end, MAX_PROPERTY_NAME, 1);
2171		if (call_prom("nextprop", 3, 1, node, prev_name, namep) != 1) {
2172			/* No more nodes: unwind alloc */
2173			*mem_start = (unsigned long)namep;
2174			break;
2175		}
2176
2177 		/* skip "name" */
2178 		if (strcmp(namep, "name") == 0) {
2179 			*mem_start = (unsigned long)namep;
2180 			prev_name = "name";
2181 			continue;
2182 		}
2183		/* get/create string entry */
2184		soff = dt_find_string(namep);
2185		if (soff != 0) {
2186			*mem_start = (unsigned long)namep;
2187			namep = sstart + soff;
2188		} else {
2189			/* Trim off some if we can */
2190			*mem_start = (unsigned long)namep + strlen(namep) + 1;
2191			dt_string_end = *mem_start;
2192		}
2193		prev_name = namep;
2194	}
2195
2196	/* do all our children */
2197	child = call_prom("child", 1, 1, node);
2198	while (child != 0) {
2199		scan_dt_build_strings(child, mem_start, mem_end);
2200		child = call_prom("peer", 1, 1, child);
2201	}
2202}
2203
2204static void __init scan_dt_build_struct(phandle node, unsigned long *mem_start,
2205					unsigned long *mem_end)
2206{
2207	phandle child;
2208	char *namep, *prev_name, *sstart, *p, *ep, *lp, *path;
2209	unsigned long soff;
2210	unsigned char *valp;
2211	static char pname[MAX_PROPERTY_NAME];
2212	int l, room, has_phandle = 0;
2213
2214	dt_push_token(OF_DT_BEGIN_NODE, mem_start, mem_end);
2215
2216	/* get the node's full name */
2217	namep = (char *)*mem_start;
2218	room = *mem_end - *mem_start;
2219	if (room > 255)
2220		room = 255;
2221	l = call_prom("package-to-path", 3, 1, node, namep, room);
2222	if (l >= 0) {
2223		/* Didn't fit?  Get more room. */
2224		if (l >= room) {
2225			if (l >= *mem_end - *mem_start)
2226				namep = make_room(mem_start, mem_end, l+1, 1);
2227			call_prom("package-to-path", 3, 1, node, namep, l);
2228		}
2229		namep[l] = '\0';
2230
2231		/* Fixup an Apple bug where they have bogus \0 chars in the
2232		 * middle of the path in some properties, and extract
2233		 * the unit name (everything after the last '/').
2234		 */
2235		for (lp = p = namep, ep = namep + l; p < ep; p++) {
2236			if (*p == '/')
2237				lp = namep;
2238			else if (*p != 0)
2239				*lp++ = *p;
2240		}
2241		*lp = 0;
2242		*mem_start = _ALIGN((unsigned long)lp + 1, 4);
2243	}
2244
2245	/* get it again for debugging */
2246	path = prom_scratch;
2247	memset(path, 0, PROM_SCRATCH_SIZE);
2248	call_prom("package-to-path", 3, 1, node, path, PROM_SCRATCH_SIZE-1);
2249
2250	/* get and store all properties */
2251	prev_name = "";
2252	sstart = (char *)dt_string_start;
2253	for (;;) {
2254		if (call_prom("nextprop", 3, 1, node, prev_name,
2255			      pname) != 1)
2256			break;
2257
2258 		/* skip "name" */
2259 		if (strcmp(pname, "name") == 0) {
2260 			prev_name = "name";
2261 			continue;
2262 		}
2263
2264		/* find string offset */
2265		soff = dt_find_string(pname);
2266		if (soff == 0) {
2267			prom_printf("WARNING: Can't find string index for"
2268				    " <%s>, node %s\n", pname, path);
2269			break;
2270		}
2271		prev_name = sstart + soff;
2272
2273		/* get length */
2274		l = call_prom("getproplen", 2, 1, node, pname);
2275
2276		/* sanity checks */
2277		if (l == PROM_ERROR)
2278			continue;
2279
2280		/* push property head */
2281		dt_push_token(OF_DT_PROP, mem_start, mem_end);
2282		dt_push_token(l, mem_start, mem_end);
2283		dt_push_token(soff, mem_start, mem_end);
2284
2285		/* push property content */
2286		valp = make_room(mem_start, mem_end, l, 4);
2287		call_prom("getprop", 4, 1, node, pname, valp, l);
2288		*mem_start = _ALIGN(*mem_start, 4);
2289
2290		if (!strcmp(pname, "phandle"))
2291			has_phandle = 1;
2292	}
2293
2294	/* Add a "linux,phandle" property if no "phandle" property already
2295	 * existed (can happen with OPAL)
2296	 */
2297	if (!has_phandle) {
2298		soff = dt_find_string("linux,phandle");
2299		if (soff == 0)
2300			prom_printf("WARNING: Can't find string index for"
2301				    " <linux-phandle> node %s\n", path);
2302		else {
2303			dt_push_token(OF_DT_PROP, mem_start, mem_end);
2304			dt_push_token(4, mem_start, mem_end);
2305			dt_push_token(soff, mem_start, mem_end);
2306			valp = make_room(mem_start, mem_end, 4, 4);
2307			*(__be32 *)valp = cpu_to_be32(node);
2308		}
2309	}
2310
2311	/* do all our children */
2312	child = call_prom("child", 1, 1, node);
2313	while (child != 0) {
2314		scan_dt_build_struct(child, mem_start, mem_end);
2315		child = call_prom("peer", 1, 1, child);
2316	}
2317
2318	dt_push_token(OF_DT_END_NODE, mem_start, mem_end);
2319}
2320
2321static void __init flatten_device_tree(void)
2322{
2323	phandle root;
2324	unsigned long mem_start, mem_end, room;
2325	struct boot_param_header *hdr;
2326	char *namep;
2327	u64 *rsvmap;
2328
2329	/*
2330	 * Check how much room we have between alloc top & bottom (+/- a
2331	 * few pages), crop to 1MB, as this is our "chunk" size
2332	 */
2333	room = alloc_top - alloc_bottom - 0x4000;
2334	if (room > DEVTREE_CHUNK_SIZE)
2335		room = DEVTREE_CHUNK_SIZE;
2336	prom_debug("starting device tree allocs at %x\n", alloc_bottom);
2337
2338	/* Now try to claim that */
2339	mem_start = (unsigned long)alloc_up(room, PAGE_SIZE);
2340	if (mem_start == 0)
2341		prom_panic("Can't allocate initial device-tree chunk\n");
2342	mem_end = mem_start + room;
2343
2344	/* Get root of tree */
2345	root = call_prom("peer", 1, 1, (phandle)0);
2346	if (root == (phandle)0)
2347		prom_panic ("couldn't get device tree root\n");
2348
2349	/* Build header and make room for mem rsv map */ 
2350	mem_start = _ALIGN(mem_start, 4);
2351	hdr = make_room(&mem_start, &mem_end,
2352			sizeof(struct boot_param_header), 4);
2353	dt_header_start = (unsigned long)hdr;
2354	rsvmap = make_room(&mem_start, &mem_end, sizeof(mem_reserve_map), 8);
2355
2356	/* Start of strings */
2357	mem_start = PAGE_ALIGN(mem_start);
2358	dt_string_start = mem_start;
2359	mem_start += 4; /* hole */
2360
2361	/* Add "linux,phandle" in there, we'll need it */
2362	namep = make_room(&mem_start, &mem_end, 16, 1);
2363	strcpy(namep, "linux,phandle");
2364	mem_start = (unsigned long)namep + strlen(namep) + 1;
2365
2366	/* Build string array */
2367	prom_printf("Building dt strings...\n"); 
2368	scan_dt_build_strings(root, &mem_start, &mem_end);
2369	dt_string_end = mem_start;
2370
2371	/* Build structure */
2372	mem_start = PAGE_ALIGN(mem_start);
2373	dt_struct_start = mem_start;
2374	prom_printf("Building dt structure...\n"); 
2375	scan_dt_build_struct(root, &mem_start, &mem_end);
2376	dt_push_token(OF_DT_END, &mem_start, &mem_end);
2377	dt_struct_end = PAGE_ALIGN(mem_start);
2378
2379	/* Finish header */
2380	hdr->boot_cpuid_phys = cpu_to_be32(prom.cpu);
2381	hdr->magic = cpu_to_be32(OF_DT_HEADER);
2382	hdr->totalsize = cpu_to_be32(dt_struct_end - dt_header_start);
2383	hdr->off_dt_struct = cpu_to_be32(dt_struct_start - dt_header_start);
2384	hdr->off_dt_strings = cpu_to_be32(dt_string_start - dt_header_start);
2385	hdr->dt_strings_size = cpu_to_be32(dt_string_end - dt_string_start);
2386	hdr->off_mem_rsvmap = cpu_to_be32(((unsigned long)rsvmap) - dt_header_start);
2387	hdr->version = cpu_to_be32(OF_DT_VERSION);
2388	/* Version 16 is not backward compatible */
2389	hdr->last_comp_version = cpu_to_be32(0x10);
2390
2391	/* Copy the reserve map in */
2392	memcpy(rsvmap, mem_reserve_map, sizeof(mem_reserve_map));
2393
2394#ifdef DEBUG_PROM
2395	{
2396		int i;
2397		prom_printf("reserved memory map:\n");
2398		for (i = 0; i < mem_reserve_cnt; i++)
2399			prom_printf("  %x - %x\n",
2400				    be64_to_cpu(mem_reserve_map[i].base),
2401				    be64_to_cpu(mem_reserve_map[i].size));
2402	}
2403#endif
2404	/* Bump mem_reserve_cnt to cause further reservations to fail
2405	 * since it's too late.
2406	 */
2407	mem_reserve_cnt = MEM_RESERVE_MAP_SIZE;
2408
2409	prom_printf("Device tree strings 0x%x -> 0x%x\n",
2410		    dt_string_start, dt_string_end);
2411	prom_printf("Device tree struct  0x%x -> 0x%x\n",
2412		    dt_struct_start, dt_struct_end);
2413}
2414
2415#ifdef CONFIG_PPC_MAPLE
2416/* PIBS Version 1.05.0000 04/26/2005 has an incorrect /ht/isa/ranges property.
2417 * The values are bad, and it doesn't even have the right number of cells. */
2418static void __init fixup_device_tree_maple(void)
2419{
2420	phandle isa;
2421	u32 rloc = 0x01002000; /* IO space; PCI device = 4 */
2422	u32 isa_ranges[6];
2423	char *name;
2424
2425	name = "/ht@0/isa@4";
2426	isa = call_prom("finddevice", 1, 1, ADDR(name));
2427	if (!PHANDLE_VALID(isa)) {
2428		name = "/ht@0/isa@6";
2429		isa = call_prom("finddevice", 1, 1, ADDR(name));
2430		rloc = 0x01003000; /* IO space; PCI device = 6 */
2431	}
2432	if (!PHANDLE_VALID(isa))
2433		return;
2434
2435	if (prom_getproplen(isa, "ranges") != 12)
2436		return;
2437	if (prom_getprop(isa, "ranges", isa_ranges, sizeof(isa_ranges))
2438		== PROM_ERROR)
2439		return;
2440
2441	if (isa_ranges[0] != 0x1 ||
2442		isa_ranges[1] != 0xf4000000 ||
2443		isa_ranges[2] != 0x00010000)
2444		return;
2445
2446	prom_printf("Fixing up bogus ISA range on Maple/Apache...\n");
2447
2448	isa_ranges[0] = 0x1;
2449	isa_ranges[1] = 0x0;
2450	isa_ranges[2] = rloc;
2451	isa_ranges[3] = 0x0;
2452	isa_ranges[4] = 0x0;
2453	isa_ranges[5] = 0x00010000;
2454	prom_setprop(isa, name, "ranges",
2455			isa_ranges, sizeof(isa_ranges));
2456}
2457
2458#define CPC925_MC_START		0xf8000000
2459#define CPC925_MC_LENGTH	0x1000000
2460/* The values for memory-controller don't have right number of cells */
2461static void __init fixup_device_tree_maple_memory_controller(void)
2462{
2463	phandle mc;
2464	u32 mc_reg[4];
2465	char *name = "/hostbridge@f8000000";
2466	u32 ac, sc;
2467
2468	mc = call_prom("finddevice", 1, 1, ADDR(name));
2469	if (!PHANDLE_VALID(mc))
2470		return;
2471
2472	if (prom_getproplen(mc, "reg") != 8)
2473		return;
2474
2475	prom_getprop(prom.root, "#address-cells", &ac, sizeof(ac));
2476	prom_getprop(prom.root, "#size-cells", &sc, sizeof(sc));
2477	if ((ac != 2) || (sc != 2))
2478		return;
2479
2480	if (prom_getprop(mc, "reg", mc_reg, sizeof(mc_reg)) == PROM_ERROR)
2481		return;
2482
2483	if (mc_reg[0] != CPC925_MC_START || mc_reg[1] != CPC925_MC_LENGTH)
2484		return;
2485
2486	prom_printf("Fixing up bogus hostbridge on Maple...\n");
2487
2488	mc_reg[0] = 0x0;
2489	mc_reg[1] = CPC925_MC_START;
2490	mc_reg[2] = 0x0;
2491	mc_reg[3] = CPC925_MC_LENGTH;
2492	prom_setprop(mc, name, "reg", mc_reg, sizeof(mc_reg));
2493}
2494#else
2495#define fixup_device_tree_maple()
2496#define fixup_device_tree_maple_memory_controller()
2497#endif
2498
2499#ifdef CONFIG_PPC_CHRP
2500/*
2501 * Pegasos and BriQ lacks the "ranges" property in the isa node
2502 * Pegasos needs decimal IRQ 14/15, not hexadecimal
2503 * Pegasos has the IDE configured in legacy mode, but advertised as native
2504 */
2505static void __init fixup_device_tree_chrp(void)
2506{
2507	phandle ph;
2508	u32 prop[6];
2509	u32 rloc = 0x01006000; /* IO space; PCI device = 12 */
2510	char *name;
2511	int rc;
2512
2513	name = "/pci@80000000/isa@c";
2514	ph = call_prom("finddevice", 1, 1, ADDR(name));
2515	if (!PHANDLE_VALID(ph)) {
2516		name = "/pci@ff500000/isa@6";
2517		ph = call_prom("finddevice", 1, 1, ADDR(name));
2518		rloc = 0x01003000; /* IO space; PCI device = 6 */
2519	}
2520	if (PHANDLE_VALID(ph)) {
2521		rc = prom_getproplen(ph, "ranges");
2522		if (rc == 0 || rc == PROM_ERROR) {
2523			prom_printf("Fixing up missing ISA range on Pegasos...\n");
2524
2525			prop[0] = 0x1;
2526			prop[1] = 0x0;
2527			prop[2] = rloc;
2528			prop[3] = 0x0;
2529			prop[4] = 0x0;
2530			prop[5] = 0x00010000;
2531			prom_setprop(ph, name, "ranges", prop, sizeof(prop));
2532		}
2533	}
2534
2535	name = "/pci@80000000/ide@C,1";
2536	ph = call_prom("finddevice", 1, 1, ADDR(name));
2537	if (PHANDLE_VALID(ph)) {
2538		prom_printf("Fixing up IDE interrupt on Pegasos...\n");
2539		prop[0] = 14;
2540		prop[1] = 0x0;
2541		prom_setprop(ph, name, "interrupts", prop, 2*sizeof(u32));
2542		prom_printf("Fixing up IDE class-code on Pegasos...\n");
2543		rc = prom_getprop(ph, "class-code", prop, sizeof(u32));
2544		if (rc == sizeof(u32)) {
2545			prop[0] &= ~0x5;
2546			prom_setprop(ph, name, "class-code", prop, sizeof(u32));
2547		}
2548	}
2549}
2550#else
2551#define fixup_device_tree_chrp()
2552#endif
2553
2554#if defined(CONFIG_PPC64) && defined(CONFIG_PPC_PMAC)
2555static void __init fixup_device_tree_pmac(void)
2556{
2557	phandle u3, i2c, mpic;
2558	u32 u3_rev;
2559	u32 interrupts[2];
2560	u32 parent;
2561
2562	/* Some G5s have a missing interrupt definition, fix it up here */
2563	u3 = call_prom("finddevice", 1, 1, ADDR("/u3@0,f8000000"));
2564	if (!PHANDLE_VALID(u3))
2565		return;
2566	i2c = call_prom("finddevice", 1, 1, ADDR("/u3@0,f8000000/i2c@f8001000"));
2567	if (!PHANDLE_VALID(i2c))
2568		return;
2569	mpic = call_prom("finddevice", 1, 1, ADDR("/u3@0,f8000000/mpic@f8040000"));
2570	if (!PHANDLE_VALID(mpic))
2571		return;
2572
2573	/* check if proper rev of u3 */
2574	if (prom_getprop(u3, "device-rev", &u3_rev, sizeof(u3_rev))
2575	    == PROM_ERROR)
2576		return;
2577	if (u3_rev < 0x35 || u3_rev > 0x39)
2578		return;
2579	/* does it need fixup ? */
2580	if (prom_getproplen(i2c, "interrupts") > 0)
2581		return;
2582
2583	prom_printf("fixing up bogus interrupts for u3 i2c...\n");
2584
2585	/* interrupt on this revision of u3 is number 0 and level */
2586	interrupts[0] = 0;
2587	interrupts[1] = 1;
2588	prom_setprop(i2c, "/u3@0,f8000000/i2c@f8001000", "interrupts",
2589		     &interrupts, sizeof(interrupts));
2590	parent = (u32)mpic;
2591	prom_setprop(i2c, "/u3@0,f8000000/i2c@f8001000", "interrupt-parent",
2592		     &parent, sizeof(parent));
2593}
2594#else
2595#define fixup_device_tree_pmac()
2596#endif
2597
2598#ifdef CONFIG_PPC_EFIKA
2599/*
2600 * The MPC5200 FEC driver requires an phy-handle property to tell it how
2601 * to talk to the phy.  If the phy-handle property is missing, then this
2602 * function is called to add the appropriate nodes and link it to the
2603 * ethernet node.
2604 */
2605static void __init fixup_device_tree_efika_add_phy(void)
2606{
2607	u32 node;
2608	char prop[64];
2609	int rv;
2610
2611	/* Check if /builtin/ethernet exists - bail if it doesn't */
2612	node = call_prom("finddevice", 1, 1, ADDR("/builtin/ethernet"));
2613	if (!PHANDLE_VALID(node))
2614		return;
2615
2616	/* Check if the phy-handle property exists - bail if it does */
2617	rv = prom_getprop(node, "phy-handle", prop, sizeof(prop));
2618	if (!rv)
2619		return;
2620
2621	/*
2622	 * At this point the ethernet device doesn't have a phy described.
2623	 * Now we need to add the missing phy node and linkage
2624	 */
2625
2626	/* Check for an MDIO bus node - if missing then create one */
2627	node = call_prom("finddevice", 1, 1, ADDR("/builtin/mdio"));
2628	if (!PHANDLE_VALID(node)) {
2629		prom_printf("Adding Ethernet MDIO node\n");
2630		call_prom("interpret", 1, 1,
2631			" s\" /builtin\" find-device"
2632			" new-device"
2633				" 1 encode-int s\" #address-cells\" property"
2634				" 0 encode-int s\" #size-cells\" property"
2635				" s\" mdio\" device-name"
2636				" s\" fsl,mpc5200b-mdio\" encode-string"
2637				" s\" compatible\" property"
2638				" 0xf0003000 0x400 reg"
2639				" 0x2 encode-int"
2640				" 0x5 encode-int encode+"
2641				" 0x3 encode-int encode+"
2642				" s\" interrupts\" property"
2643			" finish-device");
2644	};
2645
2646	/* Check for a PHY device node - if missing then create one and
2647	 * give it's phandle to the ethernet node */
2648	node = call_prom("finddevice", 1, 1,
2649			 ADDR("/builtin/mdio/ethernet-phy"));
2650	if (!PHANDLE_VALID(node)) {
2651		prom_printf("Adding Ethernet PHY node\n");
2652		call_prom("interpret", 1, 1,
2653			" s\" /builtin/mdio\" find-device"
2654			" new-device"
2655				" s\" ethernet-phy\" device-name"
2656				" 0x10 encode-int s\" reg\" property"
2657				" my-self"
2658				" ihandle>phandle"
2659			" finish-device"
2660			" s\" /builtin/ethernet\" find-device"
2661				" encode-int"
2662				" s\" phy-handle\" property"
2663			" device-end");
2664	}
2665}
2666
2667static void __init fixup_device_tree_efika(void)
2668{
2669	int sound_irq[3] = { 2, 2, 0 };
2670	int bcomm_irq[3*16] = { 3,0,0, 3,1,0, 3,2,0, 3,3,0,
2671				3,4,0, 3,5,0, 3,6,0, 3,7,0,
2672				3,8,0, 3,9,0, 3,10,0, 3,11,0,
2673				3,12,0, 3,13,0, 3,14,0, 3,15,0 };
2674	u32 node;
2675	char prop[64];
2676	int rv, len;
2677
2678	/* Check if we're really running on a EFIKA */
2679	node = call_prom("finddevice", 1, 1, ADDR("/"));
2680	if (!PHANDLE_VALID(node))
2681		return;
2682
2683	rv = prom_getprop(node, "model", prop, sizeof(prop));
2684	if (rv == PROM_ERROR)
2685		return;
2686	if (strcmp(prop, "EFIKA5K2"))
2687		return;
2688
2689	prom_printf("Applying EFIKA device tree fixups\n");
2690
2691	/* Claiming to be 'chrp' is death */
2692	node = call_prom("finddevice", 1, 1, ADDR("/"));
2693	rv = prom_getprop(node, "device_type", prop, sizeof(prop));
2694	if (rv != PROM_ERROR && (strcmp(prop, "chrp") == 0))
2695		prom_setprop(node, "/", "device_type", "efika", sizeof("efika"));
2696
2697	/* CODEGEN,description is exposed in /proc/cpuinfo so
2698	   fix that too */
2699	rv = prom_getprop(node, "CODEGEN,description", prop, sizeof(prop));
2700	if (rv != PROM_ERROR && (strstr(prop, "CHRP")))
2701		prom_setprop(node, "/", "CODEGEN,description",
2702			     "Efika 5200B PowerPC System",
2703			     sizeof("Efika 5200B PowerPC System"));
2704
2705	/* Fixup bestcomm interrupts property */
2706	node = call_prom("finddevice", 1, 1, ADDR("/builtin/bestcomm"));
2707	if (PHANDLE_VALID(node)) {
2708		len = prom_getproplen(node, "interrupts");
2709		if (len == 12) {
2710			prom_printf("Fixing bestcomm interrupts property\n");
2711			prom_setprop(node, "/builtin/bestcom", "interrupts",
2712				     bcomm_irq, sizeof(bcomm_irq));
2713		}
2714	}
2715
2716	/* Fixup sound interrupts property */
2717	node = call_prom("finddevice", 1, 1, ADDR("/builtin/sound"));
2718	if (PHANDLE_VALID(node)) {
2719		rv = prom_getprop(node, "interrupts", prop, sizeof(prop));
2720		if (rv == PROM_ERROR) {
2721			prom_printf("Adding sound interrupts property\n");
2722			prom_setprop(node, "/builtin/sound", "interrupts",
2723				     sound_irq, sizeof(sound_irq));
2724		}
2725	}
2726
2727	/* Make sure ethernet phy-handle property exists */
2728	fixup_device_tree_efika_add_phy();
2729}
2730#else
2731#define fixup_device_tree_efika()
2732#endif
2733
2734#ifdef CONFIG_PPC_PASEMI_NEMO
2735/*
2736 * CFE supplied on Nemo is broken in several ways, biggest
2737 * problem is that it reassigns ISA interrupts to unused mpic ints.
2738 * Add an interrupt-controller property for the io-bridge to use
2739 * and correct the ints so we can attach them to an irq_domain
2740 */
2741static void __init fixup_device_tree_pasemi(void)
2742{
2743	u32 interrupts[2], parent, rval, val = 0;
2744	char *name, *pci_name;
2745	phandle iob, node;
2746
2747	/* Find the root pci node */
2748	name = "/pxp@0,e0000000";
2749	iob = call_prom("finddevice", 1, 1, ADDR(name));
2750	if (!PHANDLE_VALID(iob))
2751		return;
2752
2753	/* check if interrupt-controller node set yet */
2754	if (prom_getproplen(iob, "interrupt-controller") !=PROM_ERROR)
2755		return;
2756
2757	prom_printf("adding interrupt-controller property for SB600...\n");
2758
2759	prom_setprop(iob, name, "interrupt-controller", &val, 0);
2760
2761	pci_name = "/pxp@0,e0000000/pci@11";
2762	node = call_prom("finddevice", 1, 1, ADDR(pci_name));
2763	parent = ADDR(iob);
2764
2765	for( ; prom_next_node(&node); ) {
2766		/* scan each node for one with an interrupt */
2767		if (!PHANDLE_VALID(node))
2768			continue;
2769
2770		rval = prom_getproplen(node, "interrupts");
2771		if (rval == 0 || rval == PROM_ERROR)
2772			continue;
2773
2774		prom_getprop(node, "interrupts", &interrupts, sizeof(interrupts));
2775		if ((interrupts[0] < 212) || (interrupts[0] > 222))
2776			continue;
2777
2778		/* found a node, update both interrupts and interrupt-parent */
2779		if ((interrupts[0] >= 212) && (interrupts[0] <= 215))
2780			interrupts[0] -= 203;
2781		if ((interrupts[0] >= 216) && (interrupts[0] <= 220))
2782			interrupts[0] -= 213;
2783		if (interrupts[0] == 221)
2784			interrupts[0] = 14;
2785		if (interrupts[0] == 222)
2786			interrupts[0] = 8;
2787
2788		prom_setprop(node, pci_name, "interrupts", interrupts,
2789					sizeof(interrupts));
2790		prom_setprop(node, pci_name, "interrupt-parent", &parent,
2791					sizeof(parent));
2792	}
2793
2794	/*
2795	 * The io-bridge has device_type set to 'io-bridge' change it to 'isa'
2796	 * so that generic isa-bridge code can add the SB600 and its on-board
2797	 * peripherals.
2798	 */
2799	name = "/pxp@0,e0000000/io-bridge@0";
2800	iob = call_prom("finddevice", 1, 1, ADDR(name));
2801	if (!PHANDLE_VALID(iob))
2802		return;
2803
2804	/* device_type is already set, just change it. */
2805
2806	prom_printf("Changing device_type of SB600 node...\n");
2807
2808	prom_setprop(iob, name, "device_type", "isa", sizeof("isa"));
2809}
2810#else	/* !CONFIG_PPC_PASEMI_NEMO */
2811static inline void fixup_device_tree_pasemi(void) { }
2812#endif
2813
2814static void __init fixup_device_tree(void)
2815{
2816	fixup_device_tree_maple();
2817	fixup_device_tree_maple_memory_controller();
2818	fixup_device_tree_chrp();
2819	fixup_device_tree_pmac();
2820	fixup_device_tree_efika();
2821	fixup_device_tree_pasemi();
2822}
2823
2824static void __init prom_find_boot_cpu(void)
2825{
2826	__be32 rval;
2827	ihandle prom_cpu;
2828	phandle cpu_pkg;
2829
2830	rval = 0;
2831	if (prom_getprop(prom.chosen, "cpu", &rval, sizeof(rval)) <= 0)
2832		return;
2833	prom_cpu = be32_to_cpu(rval);
2834
2835	cpu_pkg = call_prom("instance-to-package", 1, 1, prom_cpu);
2836
2837	if (!PHANDLE_VALID(cpu_pkg))
2838		return;
2839
2840	prom_getprop(cpu_pkg, "reg", &rval, sizeof(rval));
2841	prom.cpu = be32_to_cpu(rval);
2842
2843	prom_debug("Booting CPU hw index = %lu\n", prom.cpu);
2844}
2845
2846static void __init prom_check_initrd(unsigned long r3, unsigned long r4)
2847{
2848#ifdef CONFIG_BLK_DEV_INITRD
2849	if (r3 && r4 && r4 != 0xdeadbeef) {
2850		__be64 val;
2851
2852		prom_initrd_start = is_kernel_addr(r3) ? __pa(r3) : r3;
2853		prom_initrd_end = prom_initrd_start + r4;
2854
2855		val = cpu_to_be64(prom_initrd_start);
2856		prom_setprop(prom.chosen, "/chosen", "linux,initrd-start",
2857			     &val, sizeof(val));
2858		val = cpu_to_be64(prom_initrd_end);
2859		prom_setprop(prom.chosen, "/chosen", "linux,initrd-end",
2860			     &val, sizeof(val));
2861
2862		reserve_mem(prom_initrd_start,
2863			    prom_initrd_end - prom_initrd_start);
2864
2865		prom_debug("initrd_start=0x%x\n", prom_initrd_start);
2866		prom_debug("initrd_end=0x%x\n", prom_initrd_end);
2867	}
2868#endif /* CONFIG_BLK_DEV_INITRD */
2869}
2870
2871#ifdef CONFIG_PPC64
2872#ifdef CONFIG_RELOCATABLE
2873static void reloc_toc(void)
 
 
2874{
 
 
 
 
 
 
 
2875}
2876
2877static void unreloc_toc(void)
 
 
 
2878{
2879}
2880#else
2881static void __reloc_toc(unsigned long offset, unsigned long nr_entries)
2882{
2883	unsigned long i;
2884	unsigned long *toc_entry;
2885
2886	/* Get the start of the TOC by using r2 directly. */
2887	asm volatile("addi %0,2,-0x8000" : "=b" (toc_entry));
2888
2889	for (i = 0; i < nr_entries; i++) {
2890		*toc_entry = *toc_entry + offset;
2891		toc_entry++;
2892	}
2893}
 
2894
2895static void reloc_toc(void)
2896{
2897	unsigned long offset = reloc_offset();
2898	unsigned long nr_entries =
2899		(__prom_init_toc_end - __prom_init_toc_start) / sizeof(long);
2900
2901	__reloc_toc(offset, nr_entries);
 
2902
2903	mb();
 
 
 
2904}
2905
2906static void unreloc_toc(void)
2907{
2908	unsigned long offset = reloc_offset();
2909	unsigned long nr_entries =
2910		(__prom_init_toc_end - __prom_init_toc_start) / sizeof(long);
2911
2912	mb();
2913
2914	__reloc_toc(-offset, nr_entries);
2915}
2916#endif
2917#endif
2918
2919/*
2920 * We enter here early on, when the Open Firmware prom is still
2921 * handling exceptions and the MMU hash table for us.
2922 */
2923
2924unsigned long __init prom_init(unsigned long r3, unsigned long r4,
2925			       unsigned long pp,
2926			       unsigned long r6, unsigned long r7,
2927			       unsigned long kbase)
2928{	
2929	unsigned long hdr;
2930
2931#ifdef CONFIG_PPC32
2932	unsigned long offset = reloc_offset();
2933	reloc_got2(offset);
2934#else
2935	reloc_toc();
2936#endif
2937
2938	/*
2939	 * First zero the BSS
2940	 */
2941	memset(&__bss_start, 0, __bss_stop - __bss_start);
2942
2943	/*
2944	 * Init interface to Open Firmware, get some node references,
2945	 * like /chosen
2946	 */
2947	prom_init_client_services(pp);
2948
2949	/*
2950	 * See if this OF is old enough that we need to do explicit maps
2951	 * and other workarounds
2952	 */
2953	prom_find_mmu();
2954
2955	/*
2956	 * Init prom stdout device
2957	 */
2958	prom_init_stdout();
2959
2960	prom_printf("Preparing to boot %s", linux_banner);
2961
2962	/*
2963	 * Get default machine type. At this point, we do not differentiate
2964	 * between pSeries SMP and pSeries LPAR
2965	 */
2966	of_platform = prom_find_machine_type();
2967	prom_printf("Detected machine type: %x\n", of_platform);
2968
2969#ifndef CONFIG_NONSTATIC_KERNEL
2970	/* Bail if this is a kdump kernel. */
2971	if (PHYSICAL_START > 0)
2972		prom_panic("Error: You can't boot a kdump kernel from OF!\n");
2973#endif
2974
2975	/*
2976	 * Check for an initrd
2977	 */
2978	prom_check_initrd(r3, r4);
2979
2980#if defined(CONFIG_PPC_PSERIES) || defined(CONFIG_PPC_POWERNV)
 
 
 
 
 
2981	/*
2982	 * On pSeries, inform the firmware about our capabilities
2983	 */
2984	if (of_platform == PLATFORM_PSERIES ||
2985	    of_platform == PLATFORM_PSERIES_LPAR)
2986		prom_send_capabilities();
2987#endif
2988
2989	/*
2990	 * Copy the CPU hold code
2991	 */
2992	if (of_platform != PLATFORM_POWERMAC)
2993		copy_and_flush(0, kbase, 0x100, 0);
2994
2995	/*
2996	 * Do early parsing of command line
2997	 */
2998	early_cmdline_parse();
2999
3000	/*
3001	 * Initialize memory management within prom_init
3002	 */
3003	prom_init_mem();
3004
3005	/*
3006	 * Determine which cpu is actually running right _now_
3007	 */
3008	prom_find_boot_cpu();
3009
3010	/* 
3011	 * Initialize display devices
3012	 */
3013	prom_check_displays();
3014
3015#if defined(CONFIG_PPC64) && defined(__BIG_ENDIAN__)
3016	/*
3017	 * Initialize IOMMU (TCE tables) on pSeries. Do that before anything else
3018	 * that uses the allocator, we need to make sure we get the top of memory
3019	 * available for us here...
3020	 */
3021	if (of_platform == PLATFORM_PSERIES)
3022		prom_initialize_tce_table();
3023#endif
3024
3025	/*
3026	 * On non-powermacs, try to instantiate RTAS. PowerMacs don't
3027	 * have a usable RTAS implementation.
3028	 */
3029	if (of_platform != PLATFORM_POWERMAC &&
3030	    of_platform != PLATFORM_OPAL)
3031		prom_instantiate_rtas();
3032
3033#ifdef CONFIG_PPC_POWERNV
3034	if (of_platform == PLATFORM_OPAL)
3035		prom_instantiate_opal();
3036#endif /* CONFIG_PPC_POWERNV */
3037
3038#ifdef CONFIG_PPC64
3039	/* instantiate sml */
3040	prom_instantiate_sml();
3041#endif
3042
3043	/*
3044	 * On non-powermacs, put all CPUs in spin-loops.
3045	 *
3046	 * PowerMacs use a different mechanism to spin CPUs
3047	 *
3048	 * (This must be done after instanciating RTAS)
3049	 */
3050	if (of_platform != PLATFORM_POWERMAC &&
3051	    of_platform != PLATFORM_OPAL)
3052		prom_hold_cpus();
3053
3054	/*
3055	 * Fill in some infos for use by the kernel later on
3056	 */
3057	if (prom_memory_limit) {
3058		__be64 val = cpu_to_be64(prom_memory_limit);
3059		prom_setprop(prom.chosen, "/chosen", "linux,memory-limit",
3060			     &val, sizeof(val));
3061	}
3062#ifdef CONFIG_PPC64
3063	if (prom_iommu_off)
3064		prom_setprop(prom.chosen, "/chosen", "linux,iommu-off",
3065			     NULL, 0);
3066
3067	if (prom_iommu_force_on)
3068		prom_setprop(prom.chosen, "/chosen", "linux,iommu-force-on",
3069			     NULL, 0);
3070
3071	if (prom_tce_alloc_start) {
3072		prom_setprop(prom.chosen, "/chosen", "linux,tce-alloc-start",
3073			     &prom_tce_alloc_start,
3074			     sizeof(prom_tce_alloc_start));
3075		prom_setprop(prom.chosen, "/chosen", "linux,tce-alloc-end",
3076			     &prom_tce_alloc_end,
3077			     sizeof(prom_tce_alloc_end));
3078	}
3079#endif
3080
3081	/*
3082	 * Fixup any known bugs in the device-tree
3083	 */
3084	fixup_device_tree();
3085
3086	/*
3087	 * Now finally create the flattened device-tree
3088	 */
3089	prom_printf("copying OF device tree...\n");
3090	flatten_device_tree();
3091
3092	/*
3093	 * in case stdin is USB and still active on IBM machines...
3094	 * Unfortunately quiesce crashes on some powermacs if we have
3095	 * closed stdin already (in particular the powerbook 101). It
3096	 * appears that the OPAL version of OFW doesn't like it either.
3097	 */
3098	if (of_platform != PLATFORM_POWERMAC &&
3099	    of_platform != PLATFORM_OPAL)
3100		prom_close_stdin();
3101
3102	/*
3103	 * Call OF "quiesce" method to shut down pending DMA's from
3104	 * devices etc...
3105	 */
3106	prom_printf("Quiescing Open Firmware ...\n");
3107	call_prom("quiesce", 0, 0);
3108
3109	/*
3110	 * And finally, call the kernel passing it the flattened device
3111	 * tree and NULL as r5, thus triggering the new entry point which
3112	 * is common to us and kexec
3113	 */
3114	hdr = dt_header_start;
3115
3116	/* Don't print anything after quiesce under OPAL, it crashes OFW */
3117	if (of_platform != PLATFORM_OPAL) {
3118		prom_printf("Booting Linux via __start() @ 0x%lx ...\n", kbase);
3119		prom_debug("->dt_header_start=0x%x\n", hdr);
3120	}
3121
3122#ifdef CONFIG_PPC32
3123	reloc_got2(-offset);
3124#else
3125	unreloc_toc();
3126#endif
3127
3128#ifdef CONFIG_PPC_EARLY_DEBUG_OPAL
3129	/* OPAL early debug gets the OPAL base & entry in r8 and r9 */
3130	__start(hdr, kbase, 0, 0, 0,
3131		prom_opal_base, prom_opal_entry);
3132#else
3133	__start(hdr, kbase, 0, 0, 0, 0, 0);
3134#endif
3135
3136	return 0;
3137}