Loading...
1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * Procedures for interfacing to Open Firmware.
4 *
5 * Paul Mackerras August 1996.
6 * Copyright (C) 1996-2005 Paul Mackerras.
7 *
8 * Adapted for 64bit PowerPC by Dave Engebretsen and Peter Bergner.
9 * {engebret|bergner}@us.ibm.com
10 */
11
12#undef DEBUG_PROM
13
14/* we cannot use FORTIFY as it brings in new symbols */
15#define __NO_FORTIFY
16
17#include <stdarg.h>
18#include <linux/kernel.h>
19#include <linux/string.h>
20#include <linux/init.h>
21#include <linux/threads.h>
22#include <linux/spinlock.h>
23#include <linux/types.h>
24#include <linux/pci.h>
25#include <linux/proc_fs.h>
26#include <linux/delay.h>
27#include <linux/initrd.h>
28#include <linux/bitops.h>
29#include <linux/pgtable.h>
30#include <asm/prom.h>
31#include <asm/rtas.h>
32#include <asm/page.h>
33#include <asm/processor.h>
34#include <asm/irq.h>
35#include <asm/io.h>
36#include <asm/smp.h>
37#include <asm/mmu.h>
38#include <asm/iommu.h>
39#include <asm/btext.h>
40#include <asm/sections.h>
41#include <asm/machdep.h>
42#include <asm/asm-prototypes.h>
43#include <asm/ultravisor-api.h>
44
45#include <linux/linux_logo.h>
46
47/* All of prom_init bss lives here */
48#define __prombss __section(.bss.prominit)
49
50/*
51 * Eventually bump that one up
52 */
53#define DEVTREE_CHUNK_SIZE 0x100000
54
55/*
56 * This is the size of the local memory reserve map that gets copied
57 * into the boot params passed to the kernel. That size is totally
58 * flexible as the kernel just reads the list until it encounters an
59 * entry with size 0, so it can be changed without breaking binary
60 * compatibility
61 */
62#define MEM_RESERVE_MAP_SIZE 8
63
64/*
65 * prom_init() is called very early on, before the kernel text
66 * and data have been mapped to KERNELBASE. At this point the code
67 * is running at whatever address it has been loaded at.
68 * On ppc32 we compile with -mrelocatable, which means that references
69 * to extern and static variables get relocated automatically.
70 * ppc64 objects are always relocatable, we just need to relocate the
71 * TOC.
72 *
73 * Because OF may have mapped I/O devices into the area starting at
74 * KERNELBASE, particularly on CHRP machines, we can't safely call
75 * OF once the kernel has been mapped to KERNELBASE. Therefore all
76 * OF calls must be done within prom_init().
77 *
78 * ADDR is used in calls to call_prom. The 4th and following
79 * arguments to call_prom should be 32-bit values.
80 * On ppc64, 64 bit values are truncated to 32 bits (and
81 * fortunately don't get interpreted as two arguments).
82 */
83#define ADDR(x) (u32)(unsigned long)(x)
84
85#ifdef CONFIG_PPC64
86#define OF_WORKAROUNDS 0
87#else
88#define OF_WORKAROUNDS of_workarounds
89static int of_workarounds __prombss;
90#endif
91
92#define OF_WA_CLAIM 1 /* do phys/virt claim separately, then map */
93#define OF_WA_LONGTRAIL 2 /* work around longtrail bugs */
94
95#define PROM_BUG() do { \
96 prom_printf("kernel BUG at %s line 0x%x!\n", \
97 __FILE__, __LINE__); \
98 __builtin_trap(); \
99} while (0)
100
101#ifdef DEBUG_PROM
102#define prom_debug(x...) prom_printf(x)
103#else
104#define prom_debug(x...) do { } while (0)
105#endif
106
107
108typedef u32 prom_arg_t;
109
110struct prom_args {
111 __be32 service;
112 __be32 nargs;
113 __be32 nret;
114 __be32 args[10];
115};
116
117struct prom_t {
118 ihandle root;
119 phandle chosen;
120 int cpu;
121 ihandle stdout;
122 ihandle mmumap;
123 ihandle memory;
124};
125
126struct mem_map_entry {
127 __be64 base;
128 __be64 size;
129};
130
131typedef __be32 cell_t;
132
133extern void __start(unsigned long r3, unsigned long r4, unsigned long r5,
134 unsigned long r6, unsigned long r7, unsigned long r8,
135 unsigned long r9);
136
137#ifdef CONFIG_PPC64
138extern int enter_prom(struct prom_args *args, unsigned long entry);
139#else
140static inline int enter_prom(struct prom_args *args, unsigned long entry)
141{
142 return ((int (*)(struct prom_args *))entry)(args);
143}
144#endif
145
146extern void copy_and_flush(unsigned long dest, unsigned long src,
147 unsigned long size, unsigned long offset);
148
149/* prom structure */
150static struct prom_t __prombss prom;
151
152static unsigned long __prombss prom_entry;
153
154static char __prombss of_stdout_device[256];
155static char __prombss prom_scratch[256];
156
157static unsigned long __prombss dt_header_start;
158static unsigned long __prombss dt_struct_start, dt_struct_end;
159static unsigned long __prombss dt_string_start, dt_string_end;
160
161static unsigned long __prombss prom_initrd_start, prom_initrd_end;
162
163#ifdef CONFIG_PPC64
164static int __prombss prom_iommu_force_on;
165static int __prombss prom_iommu_off;
166static unsigned long __prombss prom_tce_alloc_start;
167static unsigned long __prombss prom_tce_alloc_end;
168#endif
169
170#ifdef CONFIG_PPC_PSERIES
171static bool __prombss prom_radix_disable;
172static bool __prombss prom_radix_gtse_disable;
173static bool __prombss prom_xive_disable;
174#endif
175
176#ifdef CONFIG_PPC_SVM
177static bool __prombss prom_svm_enable;
178#endif
179
180struct platform_support {
181 bool hash_mmu;
182 bool radix_mmu;
183 bool radix_gtse;
184 bool xive;
185};
186
187/* Platforms codes are now obsolete in the kernel. Now only used within this
188 * file and ultimately gone too. Feel free to change them if you need, they
189 * are not shared with anything outside of this file anymore
190 */
191#define PLATFORM_PSERIES 0x0100
192#define PLATFORM_PSERIES_LPAR 0x0101
193#define PLATFORM_LPAR 0x0001
194#define PLATFORM_POWERMAC 0x0400
195#define PLATFORM_GENERIC 0x0500
196
197static int __prombss of_platform;
198
199static char __prombss prom_cmd_line[COMMAND_LINE_SIZE];
200
201static unsigned long __prombss prom_memory_limit;
202
203static unsigned long __prombss alloc_top;
204static unsigned long __prombss alloc_top_high;
205static unsigned long __prombss alloc_bottom;
206static unsigned long __prombss rmo_top;
207static unsigned long __prombss ram_top;
208
209static struct mem_map_entry __prombss mem_reserve_map[MEM_RESERVE_MAP_SIZE];
210static int __prombss mem_reserve_cnt;
211
212static cell_t __prombss regbuf[1024];
213
214static bool __prombss rtas_has_query_cpu_stopped;
215
216
217/*
218 * Error results ... some OF calls will return "-1" on error, some
219 * will return 0, some will return either. To simplify, here are
220 * macros to use with any ihandle or phandle return value to check if
221 * it is valid
222 */
223
224#define PROM_ERROR (-1u)
225#define PHANDLE_VALID(p) ((p) != 0 && (p) != PROM_ERROR)
226#define IHANDLE_VALID(i) ((i) != 0 && (i) != PROM_ERROR)
227
228/* Copied from lib/string.c and lib/kstrtox.c */
229
230static int __init prom_strcmp(const char *cs, const char *ct)
231{
232 unsigned char c1, c2;
233
234 while (1) {
235 c1 = *cs++;
236 c2 = *ct++;
237 if (c1 != c2)
238 return c1 < c2 ? -1 : 1;
239 if (!c1)
240 break;
241 }
242 return 0;
243}
244
245static char __init *prom_strcpy(char *dest, const char *src)
246{
247 char *tmp = dest;
248
249 while ((*dest++ = *src++) != '\0')
250 /* nothing */;
251 return tmp;
252}
253
254static int __init prom_strncmp(const char *cs, const char *ct, size_t count)
255{
256 unsigned char c1, c2;
257
258 while (count) {
259 c1 = *cs++;
260 c2 = *ct++;
261 if (c1 != c2)
262 return c1 < c2 ? -1 : 1;
263 if (!c1)
264 break;
265 count--;
266 }
267 return 0;
268}
269
270static size_t __init prom_strlen(const char *s)
271{
272 const char *sc;
273
274 for (sc = s; *sc != '\0'; ++sc)
275 /* nothing */;
276 return sc - s;
277}
278
279static int __init prom_memcmp(const void *cs, const void *ct, size_t count)
280{
281 const unsigned char *su1, *su2;
282 int res = 0;
283
284 for (su1 = cs, su2 = ct; 0 < count; ++su1, ++su2, count--)
285 if ((res = *su1 - *su2) != 0)
286 break;
287 return res;
288}
289
290static char __init *prom_strstr(const char *s1, const char *s2)
291{
292 size_t l1, l2;
293
294 l2 = prom_strlen(s2);
295 if (!l2)
296 return (char *)s1;
297 l1 = prom_strlen(s1);
298 while (l1 >= l2) {
299 l1--;
300 if (!prom_memcmp(s1, s2, l2))
301 return (char *)s1;
302 s1++;
303 }
304 return NULL;
305}
306
307static size_t __init prom_strlcat(char *dest, const char *src, size_t count)
308{
309 size_t dsize = prom_strlen(dest);
310 size_t len = prom_strlen(src);
311 size_t res = dsize + len;
312
313 /* This would be a bug */
314 if (dsize >= count)
315 return count;
316
317 dest += dsize;
318 count -= dsize;
319 if (len >= count)
320 len = count-1;
321 memcpy(dest, src, len);
322 dest[len] = 0;
323 return res;
324
325}
326
327#ifdef CONFIG_PPC_PSERIES
328static int __init prom_strtobool(const char *s, bool *res)
329{
330 if (!s)
331 return -EINVAL;
332
333 switch (s[0]) {
334 case 'y':
335 case 'Y':
336 case '1':
337 *res = true;
338 return 0;
339 case 'n':
340 case 'N':
341 case '0':
342 *res = false;
343 return 0;
344 case 'o':
345 case 'O':
346 switch (s[1]) {
347 case 'n':
348 case 'N':
349 *res = true;
350 return 0;
351 case 'f':
352 case 'F':
353 *res = false;
354 return 0;
355 default:
356 break;
357 }
358 default:
359 break;
360 }
361
362 return -EINVAL;
363}
364#endif
365
366/* This is the one and *ONLY* place where we actually call open
367 * firmware.
368 */
369
370static int __init call_prom(const char *service, int nargs, int nret, ...)
371{
372 int i;
373 struct prom_args args;
374 va_list list;
375
376 args.service = cpu_to_be32(ADDR(service));
377 args.nargs = cpu_to_be32(nargs);
378 args.nret = cpu_to_be32(nret);
379
380 va_start(list, nret);
381 for (i = 0; i < nargs; i++)
382 args.args[i] = cpu_to_be32(va_arg(list, prom_arg_t));
383 va_end(list);
384
385 for (i = 0; i < nret; i++)
386 args.args[nargs+i] = 0;
387
388 if (enter_prom(&args, prom_entry) < 0)
389 return PROM_ERROR;
390
391 return (nret > 0) ? be32_to_cpu(args.args[nargs]) : 0;
392}
393
394static int __init call_prom_ret(const char *service, int nargs, int nret,
395 prom_arg_t *rets, ...)
396{
397 int i;
398 struct prom_args args;
399 va_list list;
400
401 args.service = cpu_to_be32(ADDR(service));
402 args.nargs = cpu_to_be32(nargs);
403 args.nret = cpu_to_be32(nret);
404
405 va_start(list, rets);
406 for (i = 0; i < nargs; i++)
407 args.args[i] = cpu_to_be32(va_arg(list, prom_arg_t));
408 va_end(list);
409
410 for (i = 0; i < nret; i++)
411 args.args[nargs+i] = 0;
412
413 if (enter_prom(&args, prom_entry) < 0)
414 return PROM_ERROR;
415
416 if (rets != NULL)
417 for (i = 1; i < nret; ++i)
418 rets[i-1] = be32_to_cpu(args.args[nargs+i]);
419
420 return (nret > 0) ? be32_to_cpu(args.args[nargs]) : 0;
421}
422
423
424static void __init prom_print(const char *msg)
425{
426 const char *p, *q;
427
428 if (prom.stdout == 0)
429 return;
430
431 for (p = msg; *p != 0; p = q) {
432 for (q = p; *q != 0 && *q != '\n'; ++q)
433 ;
434 if (q > p)
435 call_prom("write", 3, 1, prom.stdout, p, q - p);
436 if (*q == 0)
437 break;
438 ++q;
439 call_prom("write", 3, 1, prom.stdout, ADDR("\r\n"), 2);
440 }
441}
442
443
444/*
445 * Both prom_print_hex & prom_print_dec takes an unsigned long as input so that
446 * we do not need __udivdi3 or __umoddi3 on 32bits.
447 */
448static void __init prom_print_hex(unsigned long val)
449{
450 int i, nibbles = sizeof(val)*2;
451 char buf[sizeof(val)*2+1];
452
453 for (i = nibbles-1; i >= 0; i--) {
454 buf[i] = (val & 0xf) + '0';
455 if (buf[i] > '9')
456 buf[i] += ('a'-'0'-10);
457 val >>= 4;
458 }
459 buf[nibbles] = '\0';
460 call_prom("write", 3, 1, prom.stdout, buf, nibbles);
461}
462
463/* max number of decimal digits in an unsigned long */
464#define UL_DIGITS 21
465static void __init prom_print_dec(unsigned long val)
466{
467 int i, size;
468 char buf[UL_DIGITS+1];
469
470 for (i = UL_DIGITS-1; i >= 0; i--) {
471 buf[i] = (val % 10) + '0';
472 val = val/10;
473 if (val == 0)
474 break;
475 }
476 /* shift stuff down */
477 size = UL_DIGITS - i;
478 call_prom("write", 3, 1, prom.stdout, buf+i, size);
479}
480
481__printf(1, 2)
482static void __init prom_printf(const char *format, ...)
483{
484 const char *p, *q, *s;
485 va_list args;
486 unsigned long v;
487 long vs;
488 int n = 0;
489
490 va_start(args, format);
491 for (p = format; *p != 0; p = q) {
492 for (q = p; *q != 0 && *q != '\n' && *q != '%'; ++q)
493 ;
494 if (q > p)
495 call_prom("write", 3, 1, prom.stdout, p, q - p);
496 if (*q == 0)
497 break;
498 if (*q == '\n') {
499 ++q;
500 call_prom("write", 3, 1, prom.stdout,
501 ADDR("\r\n"), 2);
502 continue;
503 }
504 ++q;
505 if (*q == 0)
506 break;
507 while (*q == 'l') {
508 ++q;
509 ++n;
510 }
511 switch (*q) {
512 case 's':
513 ++q;
514 s = va_arg(args, const char *);
515 prom_print(s);
516 break;
517 case 'x':
518 ++q;
519 switch (n) {
520 case 0:
521 v = va_arg(args, unsigned int);
522 break;
523 case 1:
524 v = va_arg(args, unsigned long);
525 break;
526 case 2:
527 default:
528 v = va_arg(args, unsigned long long);
529 break;
530 }
531 prom_print_hex(v);
532 break;
533 case 'u':
534 ++q;
535 switch (n) {
536 case 0:
537 v = va_arg(args, unsigned int);
538 break;
539 case 1:
540 v = va_arg(args, unsigned long);
541 break;
542 case 2:
543 default:
544 v = va_arg(args, unsigned long long);
545 break;
546 }
547 prom_print_dec(v);
548 break;
549 case 'd':
550 ++q;
551 switch (n) {
552 case 0:
553 vs = va_arg(args, int);
554 break;
555 case 1:
556 vs = va_arg(args, long);
557 break;
558 case 2:
559 default:
560 vs = va_arg(args, long long);
561 break;
562 }
563 if (vs < 0) {
564 prom_print("-");
565 vs = -vs;
566 }
567 prom_print_dec(vs);
568 break;
569 }
570 }
571 va_end(args);
572}
573
574
575static unsigned int __init prom_claim(unsigned long virt, unsigned long size,
576 unsigned long align)
577{
578
579 if (align == 0 && (OF_WORKAROUNDS & OF_WA_CLAIM)) {
580 /*
581 * Old OF requires we claim physical and virtual separately
582 * and then map explicitly (assuming virtual mode)
583 */
584 int ret;
585 prom_arg_t result;
586
587 ret = call_prom_ret("call-method", 5, 2, &result,
588 ADDR("claim"), prom.memory,
589 align, size, virt);
590 if (ret != 0 || result == -1)
591 return -1;
592 ret = call_prom_ret("call-method", 5, 2, &result,
593 ADDR("claim"), prom.mmumap,
594 align, size, virt);
595 if (ret != 0) {
596 call_prom("call-method", 4, 1, ADDR("release"),
597 prom.memory, size, virt);
598 return -1;
599 }
600 /* the 0x12 is M (coherence) + PP == read/write */
601 call_prom("call-method", 6, 1,
602 ADDR("map"), prom.mmumap, 0x12, size, virt, virt);
603 return virt;
604 }
605 return call_prom("claim", 3, 1, (prom_arg_t)virt, (prom_arg_t)size,
606 (prom_arg_t)align);
607}
608
609static void __init __attribute__((noreturn)) prom_panic(const char *reason)
610{
611 prom_print(reason);
612 /* Do not call exit because it clears the screen on pmac
613 * it also causes some sort of double-fault on early pmacs */
614 if (of_platform == PLATFORM_POWERMAC)
615 asm("trap\n");
616
617 /* ToDo: should put up an SRC here on pSeries */
618 call_prom("exit", 0, 0);
619
620 for (;;) /* should never get here */
621 ;
622}
623
624
625static int __init prom_next_node(phandle *nodep)
626{
627 phandle node;
628
629 if ((node = *nodep) != 0
630 && (*nodep = call_prom("child", 1, 1, node)) != 0)
631 return 1;
632 if ((*nodep = call_prom("peer", 1, 1, node)) != 0)
633 return 1;
634 for (;;) {
635 if ((node = call_prom("parent", 1, 1, node)) == 0)
636 return 0;
637 if ((*nodep = call_prom("peer", 1, 1, node)) != 0)
638 return 1;
639 }
640}
641
642static inline int __init prom_getprop(phandle node, const char *pname,
643 void *value, size_t valuelen)
644{
645 return call_prom("getprop", 4, 1, node, ADDR(pname),
646 (u32)(unsigned long) value, (u32) valuelen);
647}
648
649static inline int __init prom_getproplen(phandle node, const char *pname)
650{
651 return call_prom("getproplen", 2, 1, node, ADDR(pname));
652}
653
654static void add_string(char **str, const char *q)
655{
656 char *p = *str;
657
658 while (*q)
659 *p++ = *q++;
660 *p++ = ' ';
661 *str = p;
662}
663
664static char *tohex(unsigned int x)
665{
666 static const char digits[] __initconst = "0123456789abcdef";
667 static char result[9] __prombss;
668 int i;
669
670 result[8] = 0;
671 i = 8;
672 do {
673 --i;
674 result[i] = digits[x & 0xf];
675 x >>= 4;
676 } while (x != 0 && i > 0);
677 return &result[i];
678}
679
680static int __init prom_setprop(phandle node, const char *nodename,
681 const char *pname, void *value, size_t valuelen)
682{
683 char cmd[256], *p;
684
685 if (!(OF_WORKAROUNDS & OF_WA_LONGTRAIL))
686 return call_prom("setprop", 4, 1, node, ADDR(pname),
687 (u32)(unsigned long) value, (u32) valuelen);
688
689 /* gah... setprop doesn't work on longtrail, have to use interpret */
690 p = cmd;
691 add_string(&p, "dev");
692 add_string(&p, nodename);
693 add_string(&p, tohex((u32)(unsigned long) value));
694 add_string(&p, tohex(valuelen));
695 add_string(&p, tohex(ADDR(pname)));
696 add_string(&p, tohex(prom_strlen(pname)));
697 add_string(&p, "property");
698 *p = 0;
699 return call_prom("interpret", 1, 1, (u32)(unsigned long) cmd);
700}
701
702/* We can't use the standard versions because of relocation headaches. */
703#define isxdigit(c) (('0' <= (c) && (c) <= '9') \
704 || ('a' <= (c) && (c) <= 'f') \
705 || ('A' <= (c) && (c) <= 'F'))
706
707#define isdigit(c) ('0' <= (c) && (c) <= '9')
708#define islower(c) ('a' <= (c) && (c) <= 'z')
709#define toupper(c) (islower(c) ? ((c) - 'a' + 'A') : (c))
710
711static unsigned long prom_strtoul(const char *cp, const char **endp)
712{
713 unsigned long result = 0, base = 10, value;
714
715 if (*cp == '0') {
716 base = 8;
717 cp++;
718 if (toupper(*cp) == 'X') {
719 cp++;
720 base = 16;
721 }
722 }
723
724 while (isxdigit(*cp) &&
725 (value = isdigit(*cp) ? *cp - '0' : toupper(*cp) - 'A' + 10) < base) {
726 result = result * base + value;
727 cp++;
728 }
729
730 if (endp)
731 *endp = cp;
732
733 return result;
734}
735
736static unsigned long prom_memparse(const char *ptr, const char **retptr)
737{
738 unsigned long ret = prom_strtoul(ptr, retptr);
739 int shift = 0;
740
741 /*
742 * We can't use a switch here because GCC *may* generate a
743 * jump table which won't work, because we're not running at
744 * the address we're linked at.
745 */
746 if ('G' == **retptr || 'g' == **retptr)
747 shift = 30;
748
749 if ('M' == **retptr || 'm' == **retptr)
750 shift = 20;
751
752 if ('K' == **retptr || 'k' == **retptr)
753 shift = 10;
754
755 if (shift) {
756 ret <<= shift;
757 (*retptr)++;
758 }
759
760 return ret;
761}
762
763/*
764 * Early parsing of the command line passed to the kernel, used for
765 * "mem=x" and the options that affect the iommu
766 */
767static void __init early_cmdline_parse(void)
768{
769 const char *opt;
770
771 char *p;
772 int l = 0;
773
774 prom_cmd_line[0] = 0;
775 p = prom_cmd_line;
776
777 if (!IS_ENABLED(CONFIG_CMDLINE_FORCE) && (long)prom.chosen > 0)
778 l = prom_getprop(prom.chosen, "bootargs", p, COMMAND_LINE_SIZE-1);
779
780 if (IS_ENABLED(CONFIG_CMDLINE_EXTEND) || l <= 0 || p[0] == '\0')
781 prom_strlcat(prom_cmd_line, " " CONFIG_CMDLINE,
782 sizeof(prom_cmd_line));
783
784 prom_printf("command line: %s\n", prom_cmd_line);
785
786#ifdef CONFIG_PPC64
787 opt = prom_strstr(prom_cmd_line, "iommu=");
788 if (opt) {
789 prom_printf("iommu opt is: %s\n", opt);
790 opt += 6;
791 while (*opt && *opt == ' ')
792 opt++;
793 if (!prom_strncmp(opt, "off", 3))
794 prom_iommu_off = 1;
795 else if (!prom_strncmp(opt, "force", 5))
796 prom_iommu_force_on = 1;
797 }
798#endif
799 opt = prom_strstr(prom_cmd_line, "mem=");
800 if (opt) {
801 opt += 4;
802 prom_memory_limit = prom_memparse(opt, (const char **)&opt);
803#ifdef CONFIG_PPC64
804 /* Align to 16 MB == size of ppc64 large page */
805 prom_memory_limit = ALIGN(prom_memory_limit, 0x1000000);
806#endif
807 }
808
809#ifdef CONFIG_PPC_PSERIES
810 prom_radix_disable = !IS_ENABLED(CONFIG_PPC_RADIX_MMU_DEFAULT);
811 opt = prom_strstr(prom_cmd_line, "disable_radix");
812 if (opt) {
813 opt += 13;
814 if (*opt && *opt == '=') {
815 bool val;
816
817 if (prom_strtobool(++opt, &val))
818 prom_radix_disable = false;
819 else
820 prom_radix_disable = val;
821 } else
822 prom_radix_disable = true;
823 }
824 if (prom_radix_disable)
825 prom_debug("Radix disabled from cmdline\n");
826
827 opt = prom_strstr(prom_cmd_line, "radix_hcall_invalidate=on");
828 if (opt) {
829 prom_radix_gtse_disable = true;
830 prom_debug("Radix GTSE disabled from cmdline\n");
831 }
832
833 opt = prom_strstr(prom_cmd_line, "xive=off");
834 if (opt) {
835 prom_xive_disable = true;
836 prom_debug("XIVE disabled from cmdline\n");
837 }
838#endif /* CONFIG_PPC_PSERIES */
839
840#ifdef CONFIG_PPC_SVM
841 opt = prom_strstr(prom_cmd_line, "svm=");
842 if (opt) {
843 bool val;
844
845 opt += sizeof("svm=") - 1;
846 if (!prom_strtobool(opt, &val))
847 prom_svm_enable = val;
848 }
849#endif /* CONFIG_PPC_SVM */
850}
851
852#ifdef CONFIG_PPC_PSERIES
853/*
854 * The architecture vector has an array of PVR mask/value pairs,
855 * followed by # option vectors - 1, followed by the option vectors.
856 *
857 * See prom.h for the definition of the bits specified in the
858 * architecture vector.
859 */
860
861/* Firmware expects the value to be n - 1, where n is the # of vectors */
862#define NUM_VECTORS(n) ((n) - 1)
863
864/*
865 * Firmware expects 1 + n - 2, where n is the length of the option vector in
866 * bytes. The 1 accounts for the length byte itself, the - 2 .. ?
867 */
868#define VECTOR_LENGTH(n) (1 + (n) - 2)
869
870struct option_vector1 {
871 u8 byte1;
872 u8 arch_versions;
873 u8 arch_versions3;
874} __packed;
875
876struct option_vector2 {
877 u8 byte1;
878 __be16 reserved;
879 __be32 real_base;
880 __be32 real_size;
881 __be32 virt_base;
882 __be32 virt_size;
883 __be32 load_base;
884 __be32 min_rma;
885 __be32 min_load;
886 u8 min_rma_percent;
887 u8 max_pft_size;
888} __packed;
889
890struct option_vector3 {
891 u8 byte1;
892 u8 byte2;
893} __packed;
894
895struct option_vector4 {
896 u8 byte1;
897 u8 min_vp_cap;
898} __packed;
899
900struct option_vector5 {
901 u8 byte1;
902 u8 byte2;
903 u8 byte3;
904 u8 cmo;
905 u8 associativity;
906 u8 bin_opts;
907 u8 micro_checkpoint;
908 u8 reserved0;
909 __be32 max_cpus;
910 __be16 papr_level;
911 __be16 reserved1;
912 u8 platform_facilities;
913 u8 reserved2;
914 __be16 reserved3;
915 u8 subprocessors;
916 u8 byte22;
917 u8 intarch;
918 u8 mmu;
919 u8 hash_ext;
920 u8 radix_ext;
921} __packed;
922
923struct option_vector6 {
924 u8 reserved;
925 u8 secondary_pteg;
926 u8 os_name;
927} __packed;
928
929struct ibm_arch_vec {
930 struct { u32 mask, val; } pvrs[14];
931
932 u8 num_vectors;
933
934 u8 vec1_len;
935 struct option_vector1 vec1;
936
937 u8 vec2_len;
938 struct option_vector2 vec2;
939
940 u8 vec3_len;
941 struct option_vector3 vec3;
942
943 u8 vec4_len;
944 struct option_vector4 vec4;
945
946 u8 vec5_len;
947 struct option_vector5 vec5;
948
949 u8 vec6_len;
950 struct option_vector6 vec6;
951} __packed;
952
953static const struct ibm_arch_vec ibm_architecture_vec_template __initconst = {
954 .pvrs = {
955 {
956 .mask = cpu_to_be32(0xfffe0000), /* POWER5/POWER5+ */
957 .val = cpu_to_be32(0x003a0000),
958 },
959 {
960 .mask = cpu_to_be32(0xffff0000), /* POWER6 */
961 .val = cpu_to_be32(0x003e0000),
962 },
963 {
964 .mask = cpu_to_be32(0xffff0000), /* POWER7 */
965 .val = cpu_to_be32(0x003f0000),
966 },
967 {
968 .mask = cpu_to_be32(0xffff0000), /* POWER8E */
969 .val = cpu_to_be32(0x004b0000),
970 },
971 {
972 .mask = cpu_to_be32(0xffff0000), /* POWER8NVL */
973 .val = cpu_to_be32(0x004c0000),
974 },
975 {
976 .mask = cpu_to_be32(0xffff0000), /* POWER8 */
977 .val = cpu_to_be32(0x004d0000),
978 },
979 {
980 .mask = cpu_to_be32(0xffff0000), /* POWER9 */
981 .val = cpu_to_be32(0x004e0000),
982 },
983 {
984 .mask = cpu_to_be32(0xffff0000), /* POWER10 */
985 .val = cpu_to_be32(0x00800000),
986 },
987 {
988 .mask = cpu_to_be32(0xffffffff), /* all 3.1-compliant */
989 .val = cpu_to_be32(0x0f000006),
990 },
991 {
992 .mask = cpu_to_be32(0xffffffff), /* all 3.00-compliant */
993 .val = cpu_to_be32(0x0f000005),
994 },
995 {
996 .mask = cpu_to_be32(0xffffffff), /* all 2.07-compliant */
997 .val = cpu_to_be32(0x0f000004),
998 },
999 {
1000 .mask = cpu_to_be32(0xffffffff), /* all 2.06-compliant */
1001 .val = cpu_to_be32(0x0f000003),
1002 },
1003 {
1004 .mask = cpu_to_be32(0xffffffff), /* all 2.05-compliant */
1005 .val = cpu_to_be32(0x0f000002),
1006 },
1007 {
1008 .mask = cpu_to_be32(0xfffffffe), /* all 2.04-compliant and earlier */
1009 .val = cpu_to_be32(0x0f000001),
1010 },
1011 },
1012
1013 .num_vectors = NUM_VECTORS(6),
1014
1015 .vec1_len = VECTOR_LENGTH(sizeof(struct option_vector1)),
1016 .vec1 = {
1017 .byte1 = 0,
1018 .arch_versions = OV1_PPC_2_00 | OV1_PPC_2_01 | OV1_PPC_2_02 | OV1_PPC_2_03 |
1019 OV1_PPC_2_04 | OV1_PPC_2_05 | OV1_PPC_2_06 | OV1_PPC_2_07,
1020 .arch_versions3 = OV1_PPC_3_00 | OV1_PPC_3_1,
1021 },
1022
1023 .vec2_len = VECTOR_LENGTH(sizeof(struct option_vector2)),
1024 /* option vector 2: Open Firmware options supported */
1025 .vec2 = {
1026 .byte1 = OV2_REAL_MODE,
1027 .reserved = 0,
1028 .real_base = cpu_to_be32(0xffffffff),
1029 .real_size = cpu_to_be32(0xffffffff),
1030 .virt_base = cpu_to_be32(0xffffffff),
1031 .virt_size = cpu_to_be32(0xffffffff),
1032 .load_base = cpu_to_be32(0xffffffff),
1033 .min_rma = cpu_to_be32(512), /* 512MB min RMA */
1034 .min_load = cpu_to_be32(0xffffffff), /* full client load */
1035 .min_rma_percent = 0, /* min RMA percentage of total RAM */
1036 .max_pft_size = 48, /* max log_2(hash table size) */
1037 },
1038
1039 .vec3_len = VECTOR_LENGTH(sizeof(struct option_vector3)),
1040 /* option vector 3: processor options supported */
1041 .vec3 = {
1042 .byte1 = 0, /* don't ignore, don't halt */
1043 .byte2 = OV3_FP | OV3_VMX | OV3_DFP,
1044 },
1045
1046 .vec4_len = VECTOR_LENGTH(sizeof(struct option_vector4)),
1047 /* option vector 4: IBM PAPR implementation */
1048 .vec4 = {
1049 .byte1 = 0, /* don't halt */
1050 .min_vp_cap = OV4_MIN_ENT_CAP, /* minimum VP entitled capacity */
1051 },
1052
1053 .vec5_len = VECTOR_LENGTH(sizeof(struct option_vector5)),
1054 /* option vector 5: PAPR/OF options */
1055 .vec5 = {
1056 .byte1 = 0, /* don't ignore, don't halt */
1057 .byte2 = OV5_FEAT(OV5_LPAR) | OV5_FEAT(OV5_SPLPAR) | OV5_FEAT(OV5_LARGE_PAGES) |
1058 OV5_FEAT(OV5_DRCONF_MEMORY) | OV5_FEAT(OV5_DONATE_DEDICATE_CPU) |
1059#ifdef CONFIG_PCI_MSI
1060 /* PCIe/MSI support. Without MSI full PCIe is not supported */
1061 OV5_FEAT(OV5_MSI),
1062#else
1063 0,
1064#endif
1065 .byte3 = 0,
1066 .cmo =
1067#ifdef CONFIG_PPC_SMLPAR
1068 OV5_FEAT(OV5_CMO) | OV5_FEAT(OV5_XCMO),
1069#else
1070 0,
1071#endif
1072 .associativity = OV5_FEAT(OV5_TYPE1_AFFINITY) | OV5_FEAT(OV5_PRRN),
1073 .bin_opts = OV5_FEAT(OV5_RESIZE_HPT) | OV5_FEAT(OV5_HP_EVT),
1074 .micro_checkpoint = 0,
1075 .reserved0 = 0,
1076 .max_cpus = cpu_to_be32(NR_CPUS), /* number of cores supported */
1077 .papr_level = 0,
1078 .reserved1 = 0,
1079 .platform_facilities = OV5_FEAT(OV5_PFO_HW_RNG) | OV5_FEAT(OV5_PFO_HW_ENCR) | OV5_FEAT(OV5_PFO_HW_842),
1080 .reserved2 = 0,
1081 .reserved3 = 0,
1082 .subprocessors = 1,
1083 .byte22 = OV5_FEAT(OV5_DRMEM_V2) | OV5_FEAT(OV5_DRC_INFO),
1084 .intarch = 0,
1085 .mmu = 0,
1086 .hash_ext = 0,
1087 .radix_ext = 0,
1088 },
1089
1090 /* option vector 6: IBM PAPR hints */
1091 .vec6_len = VECTOR_LENGTH(sizeof(struct option_vector6)),
1092 .vec6 = {
1093 .reserved = 0,
1094 .secondary_pteg = 0,
1095 .os_name = OV6_LINUX,
1096 },
1097};
1098
1099static struct ibm_arch_vec __prombss ibm_architecture_vec ____cacheline_aligned;
1100
1101/* Old method - ELF header with PT_NOTE sections only works on BE */
1102#ifdef __BIG_ENDIAN__
1103static const struct fake_elf {
1104 Elf32_Ehdr elfhdr;
1105 Elf32_Phdr phdr[2];
1106 struct chrpnote {
1107 u32 namesz;
1108 u32 descsz;
1109 u32 type;
1110 char name[8]; /* "PowerPC" */
1111 struct chrpdesc {
1112 u32 real_mode;
1113 u32 real_base;
1114 u32 real_size;
1115 u32 virt_base;
1116 u32 virt_size;
1117 u32 load_base;
1118 } chrpdesc;
1119 } chrpnote;
1120 struct rpanote {
1121 u32 namesz;
1122 u32 descsz;
1123 u32 type;
1124 char name[24]; /* "IBM,RPA-Client-Config" */
1125 struct rpadesc {
1126 u32 lpar_affinity;
1127 u32 min_rmo_size;
1128 u32 min_rmo_percent;
1129 u32 max_pft_size;
1130 u32 splpar;
1131 u32 min_load;
1132 u32 new_mem_def;
1133 u32 ignore_me;
1134 } rpadesc;
1135 } rpanote;
1136} fake_elf __initconst = {
1137 .elfhdr = {
1138 .e_ident = { 0x7f, 'E', 'L', 'F',
1139 ELFCLASS32, ELFDATA2MSB, EV_CURRENT },
1140 .e_type = ET_EXEC, /* yeah right */
1141 .e_machine = EM_PPC,
1142 .e_version = EV_CURRENT,
1143 .e_phoff = offsetof(struct fake_elf, phdr),
1144 .e_phentsize = sizeof(Elf32_Phdr),
1145 .e_phnum = 2
1146 },
1147 .phdr = {
1148 [0] = {
1149 .p_type = PT_NOTE,
1150 .p_offset = offsetof(struct fake_elf, chrpnote),
1151 .p_filesz = sizeof(struct chrpnote)
1152 }, [1] = {
1153 .p_type = PT_NOTE,
1154 .p_offset = offsetof(struct fake_elf, rpanote),
1155 .p_filesz = sizeof(struct rpanote)
1156 }
1157 },
1158 .chrpnote = {
1159 .namesz = sizeof("PowerPC"),
1160 .descsz = sizeof(struct chrpdesc),
1161 .type = 0x1275,
1162 .name = "PowerPC",
1163 .chrpdesc = {
1164 .real_mode = ~0U, /* ~0 means "don't care" */
1165 .real_base = ~0U,
1166 .real_size = ~0U,
1167 .virt_base = ~0U,
1168 .virt_size = ~0U,
1169 .load_base = ~0U
1170 },
1171 },
1172 .rpanote = {
1173 .namesz = sizeof("IBM,RPA-Client-Config"),
1174 .descsz = sizeof(struct rpadesc),
1175 .type = 0x12759999,
1176 .name = "IBM,RPA-Client-Config",
1177 .rpadesc = {
1178 .lpar_affinity = 0,
1179 .min_rmo_size = 64, /* in megabytes */
1180 .min_rmo_percent = 0,
1181 .max_pft_size = 48, /* 2^48 bytes max PFT size */
1182 .splpar = 1,
1183 .min_load = ~0U,
1184 .new_mem_def = 0
1185 }
1186 }
1187};
1188#endif /* __BIG_ENDIAN__ */
1189
1190static int __init prom_count_smt_threads(void)
1191{
1192 phandle node;
1193 char type[64];
1194 unsigned int plen;
1195
1196 /* Pick up th first CPU node we can find */
1197 for (node = 0; prom_next_node(&node); ) {
1198 type[0] = 0;
1199 prom_getprop(node, "device_type", type, sizeof(type));
1200
1201 if (prom_strcmp(type, "cpu"))
1202 continue;
1203 /*
1204 * There is an entry for each smt thread, each entry being
1205 * 4 bytes long. All cpus should have the same number of
1206 * smt threads, so return after finding the first.
1207 */
1208 plen = prom_getproplen(node, "ibm,ppc-interrupt-server#s");
1209 if (plen == PROM_ERROR)
1210 break;
1211 plen >>= 2;
1212 prom_debug("Found %lu smt threads per core\n", (unsigned long)plen);
1213
1214 /* Sanity check */
1215 if (plen < 1 || plen > 64) {
1216 prom_printf("Threads per core %lu out of bounds, assuming 1\n",
1217 (unsigned long)plen);
1218 return 1;
1219 }
1220 return plen;
1221 }
1222 prom_debug("No threads found, assuming 1 per core\n");
1223
1224 return 1;
1225
1226}
1227
1228static void __init prom_parse_mmu_model(u8 val,
1229 struct platform_support *support)
1230{
1231 switch (val) {
1232 case OV5_FEAT(OV5_MMU_DYNAMIC):
1233 case OV5_FEAT(OV5_MMU_EITHER): /* Either Available */
1234 prom_debug("MMU - either supported\n");
1235 support->radix_mmu = !prom_radix_disable;
1236 support->hash_mmu = true;
1237 break;
1238 case OV5_FEAT(OV5_MMU_RADIX): /* Only Radix */
1239 prom_debug("MMU - radix only\n");
1240 if (prom_radix_disable) {
1241 /*
1242 * If we __have__ to do radix, we're better off ignoring
1243 * the command line rather than not booting.
1244 */
1245 prom_printf("WARNING: Ignoring cmdline option disable_radix\n");
1246 }
1247 support->radix_mmu = true;
1248 break;
1249 case OV5_FEAT(OV5_MMU_HASH):
1250 prom_debug("MMU - hash only\n");
1251 support->hash_mmu = true;
1252 break;
1253 default:
1254 prom_debug("Unknown mmu support option: 0x%x\n", val);
1255 break;
1256 }
1257}
1258
1259static void __init prom_parse_xive_model(u8 val,
1260 struct platform_support *support)
1261{
1262 switch (val) {
1263 case OV5_FEAT(OV5_XIVE_EITHER): /* Either Available */
1264 prom_debug("XIVE - either mode supported\n");
1265 support->xive = !prom_xive_disable;
1266 break;
1267 case OV5_FEAT(OV5_XIVE_EXPLOIT): /* Only Exploitation mode */
1268 prom_debug("XIVE - exploitation mode supported\n");
1269 if (prom_xive_disable) {
1270 /*
1271 * If we __have__ to do XIVE, we're better off ignoring
1272 * the command line rather than not booting.
1273 */
1274 prom_printf("WARNING: Ignoring cmdline option xive=off\n");
1275 }
1276 support->xive = true;
1277 break;
1278 case OV5_FEAT(OV5_XIVE_LEGACY): /* Only Legacy mode */
1279 prom_debug("XIVE - legacy mode supported\n");
1280 break;
1281 default:
1282 prom_debug("Unknown xive support option: 0x%x\n", val);
1283 break;
1284 }
1285}
1286
1287static void __init prom_parse_platform_support(u8 index, u8 val,
1288 struct platform_support *support)
1289{
1290 switch (index) {
1291 case OV5_INDX(OV5_MMU_SUPPORT): /* MMU Model */
1292 prom_parse_mmu_model(val & OV5_FEAT(OV5_MMU_SUPPORT), support);
1293 break;
1294 case OV5_INDX(OV5_RADIX_GTSE): /* Radix Extensions */
1295 if (val & OV5_FEAT(OV5_RADIX_GTSE))
1296 support->radix_gtse = !prom_radix_gtse_disable;
1297 break;
1298 case OV5_INDX(OV5_XIVE_SUPPORT): /* Interrupt mode */
1299 prom_parse_xive_model(val & OV5_FEAT(OV5_XIVE_SUPPORT),
1300 support);
1301 break;
1302 }
1303}
1304
1305static void __init prom_check_platform_support(void)
1306{
1307 struct platform_support supported = {
1308 .hash_mmu = false,
1309 .radix_mmu = false,
1310 .radix_gtse = false,
1311 .xive = false
1312 };
1313 int prop_len = prom_getproplen(prom.chosen,
1314 "ibm,arch-vec-5-platform-support");
1315
1316 /*
1317 * First copy the architecture vec template
1318 *
1319 * use memcpy() instead of *vec = *vec_template so that GCC replaces it
1320 * by __memcpy() when KASAN is active
1321 */
1322 memcpy(&ibm_architecture_vec, &ibm_architecture_vec_template,
1323 sizeof(ibm_architecture_vec));
1324
1325 if (prop_len > 1) {
1326 int i;
1327 u8 vec[8];
1328 prom_debug("Found ibm,arch-vec-5-platform-support, len: %d\n",
1329 prop_len);
1330 if (prop_len > sizeof(vec))
1331 prom_printf("WARNING: ibm,arch-vec-5-platform-support longer than expected (len: %d)\n",
1332 prop_len);
1333 prom_getprop(prom.chosen, "ibm,arch-vec-5-platform-support",
1334 &vec, sizeof(vec));
1335 for (i = 0; i < sizeof(vec); i += 2) {
1336 prom_debug("%d: index = 0x%x val = 0x%x\n", i / 2
1337 , vec[i]
1338 , vec[i + 1]);
1339 prom_parse_platform_support(vec[i], vec[i + 1],
1340 &supported);
1341 }
1342 }
1343
1344 if (supported.radix_mmu && IS_ENABLED(CONFIG_PPC_RADIX_MMU)) {
1345 /* Radix preferred - Check if GTSE is also supported */
1346 prom_debug("Asking for radix\n");
1347 ibm_architecture_vec.vec5.mmu = OV5_FEAT(OV5_MMU_RADIX);
1348 if (supported.radix_gtse)
1349 ibm_architecture_vec.vec5.radix_ext =
1350 OV5_FEAT(OV5_RADIX_GTSE);
1351 else
1352 prom_debug("Radix GTSE isn't supported\n");
1353 } else if (supported.hash_mmu) {
1354 /* Default to hash mmu (if we can) */
1355 prom_debug("Asking for hash\n");
1356 ibm_architecture_vec.vec5.mmu = OV5_FEAT(OV5_MMU_HASH);
1357 } else {
1358 /* We're probably on a legacy hypervisor */
1359 prom_debug("Assuming legacy hash support\n");
1360 }
1361
1362 if (supported.xive) {
1363 prom_debug("Asking for XIVE\n");
1364 ibm_architecture_vec.vec5.intarch = OV5_FEAT(OV5_XIVE_EXPLOIT);
1365 }
1366}
1367
1368static void __init prom_send_capabilities(void)
1369{
1370 ihandle root;
1371 prom_arg_t ret;
1372 u32 cores;
1373
1374 /* Check ibm,arch-vec-5-platform-support and fixup vec5 if required */
1375 prom_check_platform_support();
1376
1377 root = call_prom("open", 1, 1, ADDR("/"));
1378 if (root != 0) {
1379 /* We need to tell the FW about the number of cores we support.
1380 *
1381 * To do that, we count the number of threads on the first core
1382 * (we assume this is the same for all cores) and use it to
1383 * divide NR_CPUS.
1384 */
1385
1386 cores = DIV_ROUND_UP(NR_CPUS, prom_count_smt_threads());
1387 prom_printf("Max number of cores passed to firmware: %u (NR_CPUS = %d)\n",
1388 cores, NR_CPUS);
1389
1390 ibm_architecture_vec.vec5.max_cpus = cpu_to_be32(cores);
1391
1392 /* try calling the ibm,client-architecture-support method */
1393 prom_printf("Calling ibm,client-architecture-support...");
1394 if (call_prom_ret("call-method", 3, 2, &ret,
1395 ADDR("ibm,client-architecture-support"),
1396 root,
1397 ADDR(&ibm_architecture_vec)) == 0) {
1398 /* the call exists... */
1399 if (ret)
1400 prom_printf("\nWARNING: ibm,client-architecture"
1401 "-support call FAILED!\n");
1402 call_prom("close", 1, 0, root);
1403 prom_printf(" done\n");
1404 return;
1405 }
1406 call_prom("close", 1, 0, root);
1407 prom_printf(" not implemented\n");
1408 }
1409
1410#ifdef __BIG_ENDIAN__
1411 {
1412 ihandle elfloader;
1413
1414 /* no ibm,client-architecture-support call, try the old way */
1415 elfloader = call_prom("open", 1, 1,
1416 ADDR("/packages/elf-loader"));
1417 if (elfloader == 0) {
1418 prom_printf("couldn't open /packages/elf-loader\n");
1419 return;
1420 }
1421 call_prom("call-method", 3, 1, ADDR("process-elf-header"),
1422 elfloader, ADDR(&fake_elf));
1423 call_prom("close", 1, 0, elfloader);
1424 }
1425#endif /* __BIG_ENDIAN__ */
1426}
1427#endif /* CONFIG_PPC_PSERIES */
1428
1429/*
1430 * Memory allocation strategy... our layout is normally:
1431 *
1432 * at 14Mb or more we have vmlinux, then a gap and initrd. In some
1433 * rare cases, initrd might end up being before the kernel though.
1434 * We assume this won't override the final kernel at 0, we have no
1435 * provision to handle that in this version, but it should hopefully
1436 * never happen.
1437 *
1438 * alloc_top is set to the top of RMO, eventually shrink down if the
1439 * TCEs overlap
1440 *
1441 * alloc_bottom is set to the top of kernel/initrd
1442 *
1443 * from there, allocations are done this way : rtas is allocated
1444 * topmost, and the device-tree is allocated from the bottom. We try
1445 * to grow the device-tree allocation as we progress. If we can't,
1446 * then we fail, we don't currently have a facility to restart
1447 * elsewhere, but that shouldn't be necessary.
1448 *
1449 * Note that calls to reserve_mem have to be done explicitly, memory
1450 * allocated with either alloc_up or alloc_down isn't automatically
1451 * reserved.
1452 */
1453
1454
1455/*
1456 * Allocates memory in the RMO upward from the kernel/initrd
1457 *
1458 * When align is 0, this is a special case, it means to allocate in place
1459 * at the current location of alloc_bottom or fail (that is basically
1460 * extending the previous allocation). Used for the device-tree flattening
1461 */
1462static unsigned long __init alloc_up(unsigned long size, unsigned long align)
1463{
1464 unsigned long base = alloc_bottom;
1465 unsigned long addr = 0;
1466
1467 if (align)
1468 base = ALIGN(base, align);
1469 prom_debug("%s(%lx, %lx)\n", __func__, size, align);
1470 if (ram_top == 0)
1471 prom_panic("alloc_up() called with mem not initialized\n");
1472
1473 if (align)
1474 base = ALIGN(alloc_bottom, align);
1475 else
1476 base = alloc_bottom;
1477
1478 for(; (base + size) <= alloc_top;
1479 base = ALIGN(base + 0x100000, align)) {
1480 prom_debug(" trying: 0x%lx\n\r", base);
1481 addr = (unsigned long)prom_claim(base, size, 0);
1482 if (addr != PROM_ERROR && addr != 0)
1483 break;
1484 addr = 0;
1485 if (align == 0)
1486 break;
1487 }
1488 if (addr == 0)
1489 return 0;
1490 alloc_bottom = addr + size;
1491
1492 prom_debug(" -> %lx\n", addr);
1493 prom_debug(" alloc_bottom : %lx\n", alloc_bottom);
1494 prom_debug(" alloc_top : %lx\n", alloc_top);
1495 prom_debug(" alloc_top_hi : %lx\n", alloc_top_high);
1496 prom_debug(" rmo_top : %lx\n", rmo_top);
1497 prom_debug(" ram_top : %lx\n", ram_top);
1498
1499 return addr;
1500}
1501
1502/*
1503 * Allocates memory downward, either from top of RMO, or if highmem
1504 * is set, from the top of RAM. Note that this one doesn't handle
1505 * failures. It does claim memory if highmem is not set.
1506 */
1507static unsigned long __init alloc_down(unsigned long size, unsigned long align,
1508 int highmem)
1509{
1510 unsigned long base, addr = 0;
1511
1512 prom_debug("%s(%lx, %lx, %s)\n", __func__, size, align,
1513 highmem ? "(high)" : "(low)");
1514 if (ram_top == 0)
1515 prom_panic("alloc_down() called with mem not initialized\n");
1516
1517 if (highmem) {
1518 /* Carve out storage for the TCE table. */
1519 addr = ALIGN_DOWN(alloc_top_high - size, align);
1520 if (addr <= alloc_bottom)
1521 return 0;
1522 /* Will we bump into the RMO ? If yes, check out that we
1523 * didn't overlap existing allocations there, if we did,
1524 * we are dead, we must be the first in town !
1525 */
1526 if (addr < rmo_top) {
1527 /* Good, we are first */
1528 if (alloc_top == rmo_top)
1529 alloc_top = rmo_top = addr;
1530 else
1531 return 0;
1532 }
1533 alloc_top_high = addr;
1534 goto bail;
1535 }
1536
1537 base = ALIGN_DOWN(alloc_top - size, align);
1538 for (; base > alloc_bottom;
1539 base = ALIGN_DOWN(base - 0x100000, align)) {
1540 prom_debug(" trying: 0x%lx\n\r", base);
1541 addr = (unsigned long)prom_claim(base, size, 0);
1542 if (addr != PROM_ERROR && addr != 0)
1543 break;
1544 addr = 0;
1545 }
1546 if (addr == 0)
1547 return 0;
1548 alloc_top = addr;
1549
1550 bail:
1551 prom_debug(" -> %lx\n", addr);
1552 prom_debug(" alloc_bottom : %lx\n", alloc_bottom);
1553 prom_debug(" alloc_top : %lx\n", alloc_top);
1554 prom_debug(" alloc_top_hi : %lx\n", alloc_top_high);
1555 prom_debug(" rmo_top : %lx\n", rmo_top);
1556 prom_debug(" ram_top : %lx\n", ram_top);
1557
1558 return addr;
1559}
1560
1561/*
1562 * Parse a "reg" cell
1563 */
1564static unsigned long __init prom_next_cell(int s, cell_t **cellp)
1565{
1566 cell_t *p = *cellp;
1567 unsigned long r = 0;
1568
1569 /* Ignore more than 2 cells */
1570 while (s > sizeof(unsigned long) / 4) {
1571 p++;
1572 s--;
1573 }
1574 r = be32_to_cpu(*p++);
1575#ifdef CONFIG_PPC64
1576 if (s > 1) {
1577 r <<= 32;
1578 r |= be32_to_cpu(*(p++));
1579 }
1580#endif
1581 *cellp = p;
1582 return r;
1583}
1584
1585/*
1586 * Very dumb function for adding to the memory reserve list, but
1587 * we don't need anything smarter at this point
1588 *
1589 * XXX Eventually check for collisions. They should NEVER happen.
1590 * If problems seem to show up, it would be a good start to track
1591 * them down.
1592 */
1593static void __init reserve_mem(u64 base, u64 size)
1594{
1595 u64 top = base + size;
1596 unsigned long cnt = mem_reserve_cnt;
1597
1598 if (size == 0)
1599 return;
1600
1601 /* We need to always keep one empty entry so that we
1602 * have our terminator with "size" set to 0 since we are
1603 * dumb and just copy this entire array to the boot params
1604 */
1605 base = ALIGN_DOWN(base, PAGE_SIZE);
1606 top = ALIGN(top, PAGE_SIZE);
1607 size = top - base;
1608
1609 if (cnt >= (MEM_RESERVE_MAP_SIZE - 1))
1610 prom_panic("Memory reserve map exhausted !\n");
1611 mem_reserve_map[cnt].base = cpu_to_be64(base);
1612 mem_reserve_map[cnt].size = cpu_to_be64(size);
1613 mem_reserve_cnt = cnt + 1;
1614}
1615
1616/*
1617 * Initialize memory allocation mechanism, parse "memory" nodes and
1618 * obtain that way the top of memory and RMO to setup out local allocator
1619 */
1620static void __init prom_init_mem(void)
1621{
1622 phandle node;
1623 char type[64];
1624 unsigned int plen;
1625 cell_t *p, *endp;
1626 __be32 val;
1627 u32 rac, rsc;
1628
1629 /*
1630 * We iterate the memory nodes to find
1631 * 1) top of RMO (first node)
1632 * 2) top of memory
1633 */
1634 val = cpu_to_be32(2);
1635 prom_getprop(prom.root, "#address-cells", &val, sizeof(val));
1636 rac = be32_to_cpu(val);
1637 val = cpu_to_be32(1);
1638 prom_getprop(prom.root, "#size-cells", &val, sizeof(rsc));
1639 rsc = be32_to_cpu(val);
1640 prom_debug("root_addr_cells: %x\n", rac);
1641 prom_debug("root_size_cells: %x\n", rsc);
1642
1643 prom_debug("scanning memory:\n");
1644
1645 for (node = 0; prom_next_node(&node); ) {
1646 type[0] = 0;
1647 prom_getprop(node, "device_type", type, sizeof(type));
1648
1649 if (type[0] == 0) {
1650 /*
1651 * CHRP Longtrail machines have no device_type
1652 * on the memory node, so check the name instead...
1653 */
1654 prom_getprop(node, "name", type, sizeof(type));
1655 }
1656 if (prom_strcmp(type, "memory"))
1657 continue;
1658
1659 plen = prom_getprop(node, "reg", regbuf, sizeof(regbuf));
1660 if (plen > sizeof(regbuf)) {
1661 prom_printf("memory node too large for buffer !\n");
1662 plen = sizeof(regbuf);
1663 }
1664 p = regbuf;
1665 endp = p + (plen / sizeof(cell_t));
1666
1667#ifdef DEBUG_PROM
1668 memset(prom_scratch, 0, sizeof(prom_scratch));
1669 call_prom("package-to-path", 3, 1, node, prom_scratch,
1670 sizeof(prom_scratch) - 1);
1671 prom_debug(" node %s :\n", prom_scratch);
1672#endif /* DEBUG_PROM */
1673
1674 while ((endp - p) >= (rac + rsc)) {
1675 unsigned long base, size;
1676
1677 base = prom_next_cell(rac, &p);
1678 size = prom_next_cell(rsc, &p);
1679
1680 if (size == 0)
1681 continue;
1682 prom_debug(" %lx %lx\n", base, size);
1683 if (base == 0 && (of_platform & PLATFORM_LPAR))
1684 rmo_top = size;
1685 if ((base + size) > ram_top)
1686 ram_top = base + size;
1687 }
1688 }
1689
1690 alloc_bottom = PAGE_ALIGN((unsigned long)&_end + 0x4000);
1691
1692 /*
1693 * If prom_memory_limit is set we reduce the upper limits *except* for
1694 * alloc_top_high. This must be the real top of RAM so we can put
1695 * TCE's up there.
1696 */
1697
1698 alloc_top_high = ram_top;
1699
1700 if (prom_memory_limit) {
1701 if (prom_memory_limit <= alloc_bottom) {
1702 prom_printf("Ignoring mem=%lx <= alloc_bottom.\n",
1703 prom_memory_limit);
1704 prom_memory_limit = 0;
1705 } else if (prom_memory_limit >= ram_top) {
1706 prom_printf("Ignoring mem=%lx >= ram_top.\n",
1707 prom_memory_limit);
1708 prom_memory_limit = 0;
1709 } else {
1710 ram_top = prom_memory_limit;
1711 rmo_top = min(rmo_top, prom_memory_limit);
1712 }
1713 }
1714
1715 /*
1716 * Setup our top alloc point, that is top of RMO or top of
1717 * segment 0 when running non-LPAR.
1718 * Some RS64 machines have buggy firmware where claims up at
1719 * 1GB fail. Cap at 768MB as a workaround.
1720 * Since 768MB is plenty of room, and we need to cap to something
1721 * reasonable on 32-bit, cap at 768MB on all machines.
1722 */
1723 if (!rmo_top)
1724 rmo_top = ram_top;
1725 rmo_top = min(0x30000000ul, rmo_top);
1726 alloc_top = rmo_top;
1727 alloc_top_high = ram_top;
1728
1729 /*
1730 * Check if we have an initrd after the kernel but still inside
1731 * the RMO. If we do move our bottom point to after it.
1732 */
1733 if (prom_initrd_start &&
1734 prom_initrd_start < rmo_top &&
1735 prom_initrd_end > alloc_bottom)
1736 alloc_bottom = PAGE_ALIGN(prom_initrd_end);
1737
1738 prom_printf("memory layout at init:\n");
1739 prom_printf(" memory_limit : %lx (16 MB aligned)\n",
1740 prom_memory_limit);
1741 prom_printf(" alloc_bottom : %lx\n", alloc_bottom);
1742 prom_printf(" alloc_top : %lx\n", alloc_top);
1743 prom_printf(" alloc_top_hi : %lx\n", alloc_top_high);
1744 prom_printf(" rmo_top : %lx\n", rmo_top);
1745 prom_printf(" ram_top : %lx\n", ram_top);
1746}
1747
1748static void __init prom_close_stdin(void)
1749{
1750 __be32 val;
1751 ihandle stdin;
1752
1753 if (prom_getprop(prom.chosen, "stdin", &val, sizeof(val)) > 0) {
1754 stdin = be32_to_cpu(val);
1755 call_prom("close", 1, 0, stdin);
1756 }
1757}
1758
1759#ifdef CONFIG_PPC_SVM
1760static int prom_rtas_hcall(uint64_t args)
1761{
1762 register uint64_t arg1 asm("r3") = H_RTAS;
1763 register uint64_t arg2 asm("r4") = args;
1764
1765 asm volatile("sc 1\n" : "=r" (arg1) :
1766 "r" (arg1),
1767 "r" (arg2) :);
1768 return arg1;
1769}
1770
1771static struct rtas_args __prombss os_term_args;
1772
1773static void __init prom_rtas_os_term(char *str)
1774{
1775 phandle rtas_node;
1776 __be32 val;
1777 u32 token;
1778
1779 prom_debug("%s: start...\n", __func__);
1780 rtas_node = call_prom("finddevice", 1, 1, ADDR("/rtas"));
1781 prom_debug("rtas_node: %x\n", rtas_node);
1782 if (!PHANDLE_VALID(rtas_node))
1783 return;
1784
1785 val = 0;
1786 prom_getprop(rtas_node, "ibm,os-term", &val, sizeof(val));
1787 token = be32_to_cpu(val);
1788 prom_debug("ibm,os-term: %x\n", token);
1789 if (token == 0)
1790 prom_panic("Could not get token for ibm,os-term\n");
1791 os_term_args.token = cpu_to_be32(token);
1792 os_term_args.nargs = cpu_to_be32(1);
1793 os_term_args.nret = cpu_to_be32(1);
1794 os_term_args.args[0] = cpu_to_be32(__pa(str));
1795 prom_rtas_hcall((uint64_t)&os_term_args);
1796}
1797#endif /* CONFIG_PPC_SVM */
1798
1799/*
1800 * Allocate room for and instantiate RTAS
1801 */
1802static void __init prom_instantiate_rtas(void)
1803{
1804 phandle rtas_node;
1805 ihandle rtas_inst;
1806 u32 base, entry = 0;
1807 __be32 val;
1808 u32 size = 0;
1809
1810 prom_debug("prom_instantiate_rtas: start...\n");
1811
1812 rtas_node = call_prom("finddevice", 1, 1, ADDR("/rtas"));
1813 prom_debug("rtas_node: %x\n", rtas_node);
1814 if (!PHANDLE_VALID(rtas_node))
1815 return;
1816
1817 val = 0;
1818 prom_getprop(rtas_node, "rtas-size", &val, sizeof(size));
1819 size = be32_to_cpu(val);
1820 if (size == 0)
1821 return;
1822
1823 base = alloc_down(size, PAGE_SIZE, 0);
1824 if (base == 0)
1825 prom_panic("Could not allocate memory for RTAS\n");
1826
1827 rtas_inst = call_prom("open", 1, 1, ADDR("/rtas"));
1828 if (!IHANDLE_VALID(rtas_inst)) {
1829 prom_printf("opening rtas package failed (%x)\n", rtas_inst);
1830 return;
1831 }
1832
1833 prom_printf("instantiating rtas at 0x%x...", base);
1834
1835 if (call_prom_ret("call-method", 3, 2, &entry,
1836 ADDR("instantiate-rtas"),
1837 rtas_inst, base) != 0
1838 || entry == 0) {
1839 prom_printf(" failed\n");
1840 return;
1841 }
1842 prom_printf(" done\n");
1843
1844 reserve_mem(base, size);
1845
1846 val = cpu_to_be32(base);
1847 prom_setprop(rtas_node, "/rtas", "linux,rtas-base",
1848 &val, sizeof(val));
1849 val = cpu_to_be32(entry);
1850 prom_setprop(rtas_node, "/rtas", "linux,rtas-entry",
1851 &val, sizeof(val));
1852
1853 /* Check if it supports "query-cpu-stopped-state" */
1854 if (prom_getprop(rtas_node, "query-cpu-stopped-state",
1855 &val, sizeof(val)) != PROM_ERROR)
1856 rtas_has_query_cpu_stopped = true;
1857
1858 prom_debug("rtas base = 0x%x\n", base);
1859 prom_debug("rtas entry = 0x%x\n", entry);
1860 prom_debug("rtas size = 0x%x\n", size);
1861
1862 prom_debug("prom_instantiate_rtas: end...\n");
1863}
1864
1865#ifdef CONFIG_PPC64
1866/*
1867 * Allocate room for and instantiate Stored Measurement Log (SML)
1868 */
1869static void __init prom_instantiate_sml(void)
1870{
1871 phandle ibmvtpm_node;
1872 ihandle ibmvtpm_inst;
1873 u32 entry = 0, size = 0, succ = 0;
1874 u64 base;
1875 __be32 val;
1876
1877 prom_debug("prom_instantiate_sml: start...\n");
1878
1879 ibmvtpm_node = call_prom("finddevice", 1, 1, ADDR("/vdevice/vtpm"));
1880 prom_debug("ibmvtpm_node: %x\n", ibmvtpm_node);
1881 if (!PHANDLE_VALID(ibmvtpm_node))
1882 return;
1883
1884 ibmvtpm_inst = call_prom("open", 1, 1, ADDR("/vdevice/vtpm"));
1885 if (!IHANDLE_VALID(ibmvtpm_inst)) {
1886 prom_printf("opening vtpm package failed (%x)\n", ibmvtpm_inst);
1887 return;
1888 }
1889
1890 if (prom_getprop(ibmvtpm_node, "ibm,sml-efi-reformat-supported",
1891 &val, sizeof(val)) != PROM_ERROR) {
1892 if (call_prom_ret("call-method", 2, 2, &succ,
1893 ADDR("reformat-sml-to-efi-alignment"),
1894 ibmvtpm_inst) != 0 || succ == 0) {
1895 prom_printf("Reformat SML to EFI alignment failed\n");
1896 return;
1897 }
1898
1899 if (call_prom_ret("call-method", 2, 2, &size,
1900 ADDR("sml-get-allocated-size"),
1901 ibmvtpm_inst) != 0 || size == 0) {
1902 prom_printf("SML get allocated size failed\n");
1903 return;
1904 }
1905 } else {
1906 if (call_prom_ret("call-method", 2, 2, &size,
1907 ADDR("sml-get-handover-size"),
1908 ibmvtpm_inst) != 0 || size == 0) {
1909 prom_printf("SML get handover size failed\n");
1910 return;
1911 }
1912 }
1913
1914 base = alloc_down(size, PAGE_SIZE, 0);
1915 if (base == 0)
1916 prom_panic("Could not allocate memory for sml\n");
1917
1918 prom_printf("instantiating sml at 0x%llx...", base);
1919
1920 memset((void *)base, 0, size);
1921
1922 if (call_prom_ret("call-method", 4, 2, &entry,
1923 ADDR("sml-handover"),
1924 ibmvtpm_inst, size, base) != 0 || entry == 0) {
1925 prom_printf("SML handover failed\n");
1926 return;
1927 }
1928 prom_printf(" done\n");
1929
1930 reserve_mem(base, size);
1931
1932 prom_setprop(ibmvtpm_node, "/vdevice/vtpm", "linux,sml-base",
1933 &base, sizeof(base));
1934 prom_setprop(ibmvtpm_node, "/vdevice/vtpm", "linux,sml-size",
1935 &size, sizeof(size));
1936
1937 prom_debug("sml base = 0x%llx\n", base);
1938 prom_debug("sml size = 0x%x\n", size);
1939
1940 prom_debug("prom_instantiate_sml: end...\n");
1941}
1942
1943/*
1944 * Allocate room for and initialize TCE tables
1945 */
1946#ifdef __BIG_ENDIAN__
1947static void __init prom_initialize_tce_table(void)
1948{
1949 phandle node;
1950 ihandle phb_node;
1951 char compatible[64], type[64], model[64];
1952 char *path = prom_scratch;
1953 u64 base, align;
1954 u32 minalign, minsize;
1955 u64 tce_entry, *tce_entryp;
1956 u64 local_alloc_top, local_alloc_bottom;
1957 u64 i;
1958
1959 if (prom_iommu_off)
1960 return;
1961
1962 prom_debug("starting prom_initialize_tce_table\n");
1963
1964 /* Cache current top of allocs so we reserve a single block */
1965 local_alloc_top = alloc_top_high;
1966 local_alloc_bottom = local_alloc_top;
1967
1968 /* Search all nodes looking for PHBs. */
1969 for (node = 0; prom_next_node(&node); ) {
1970 compatible[0] = 0;
1971 type[0] = 0;
1972 model[0] = 0;
1973 prom_getprop(node, "compatible",
1974 compatible, sizeof(compatible));
1975 prom_getprop(node, "device_type", type, sizeof(type));
1976 prom_getprop(node, "model", model, sizeof(model));
1977
1978 if ((type[0] == 0) || (prom_strstr(type, "pci") == NULL))
1979 continue;
1980
1981 /* Keep the old logic intact to avoid regression. */
1982 if (compatible[0] != 0) {
1983 if ((prom_strstr(compatible, "python") == NULL) &&
1984 (prom_strstr(compatible, "Speedwagon") == NULL) &&
1985 (prom_strstr(compatible, "Winnipeg") == NULL))
1986 continue;
1987 } else if (model[0] != 0) {
1988 if ((prom_strstr(model, "ython") == NULL) &&
1989 (prom_strstr(model, "peedwagon") == NULL) &&
1990 (prom_strstr(model, "innipeg") == NULL))
1991 continue;
1992 }
1993
1994 if (prom_getprop(node, "tce-table-minalign", &minalign,
1995 sizeof(minalign)) == PROM_ERROR)
1996 minalign = 0;
1997 if (prom_getprop(node, "tce-table-minsize", &minsize,
1998 sizeof(minsize)) == PROM_ERROR)
1999 minsize = 4UL << 20;
2000
2001 /*
2002 * Even though we read what OF wants, we just set the table
2003 * size to 4 MB. This is enough to map 2GB of PCI DMA space.
2004 * By doing this, we avoid the pitfalls of trying to DMA to
2005 * MMIO space and the DMA alias hole.
2006 */
2007 minsize = 4UL << 20;
2008
2009 /* Align to the greater of the align or size */
2010 align = max(minalign, minsize);
2011 base = alloc_down(minsize, align, 1);
2012 if (base == 0)
2013 prom_panic("ERROR, cannot find space for TCE table.\n");
2014 if (base < local_alloc_bottom)
2015 local_alloc_bottom = base;
2016
2017 /* It seems OF doesn't null-terminate the path :-( */
2018 memset(path, 0, sizeof(prom_scratch));
2019 /* Call OF to setup the TCE hardware */
2020 if (call_prom("package-to-path", 3, 1, node,
2021 path, sizeof(prom_scratch) - 1) == PROM_ERROR) {
2022 prom_printf("package-to-path failed\n");
2023 }
2024
2025 /* Save away the TCE table attributes for later use. */
2026 prom_setprop(node, path, "linux,tce-base", &base, sizeof(base));
2027 prom_setprop(node, path, "linux,tce-size", &minsize, sizeof(minsize));
2028
2029 prom_debug("TCE table: %s\n", path);
2030 prom_debug("\tnode = 0x%x\n", node);
2031 prom_debug("\tbase = 0x%llx\n", base);
2032 prom_debug("\tsize = 0x%x\n", minsize);
2033
2034 /* Initialize the table to have a one-to-one mapping
2035 * over the allocated size.
2036 */
2037 tce_entryp = (u64 *)base;
2038 for (i = 0; i < (minsize >> 3) ;tce_entryp++, i++) {
2039 tce_entry = (i << PAGE_SHIFT);
2040 tce_entry |= 0x3;
2041 *tce_entryp = tce_entry;
2042 }
2043
2044 prom_printf("opening PHB %s", path);
2045 phb_node = call_prom("open", 1, 1, path);
2046 if (phb_node == 0)
2047 prom_printf("... failed\n");
2048 else
2049 prom_printf("... done\n");
2050
2051 call_prom("call-method", 6, 0, ADDR("set-64-bit-addressing"),
2052 phb_node, -1, minsize,
2053 (u32) base, (u32) (base >> 32));
2054 call_prom("close", 1, 0, phb_node);
2055 }
2056
2057 reserve_mem(local_alloc_bottom, local_alloc_top - local_alloc_bottom);
2058
2059 /* These are only really needed if there is a memory limit in
2060 * effect, but we don't know so export them always. */
2061 prom_tce_alloc_start = local_alloc_bottom;
2062 prom_tce_alloc_end = local_alloc_top;
2063
2064 /* Flag the first invalid entry */
2065 prom_debug("ending prom_initialize_tce_table\n");
2066}
2067#endif /* __BIG_ENDIAN__ */
2068#endif /* CONFIG_PPC64 */
2069
2070/*
2071 * With CHRP SMP we need to use the OF to start the other processors.
2072 * We can't wait until smp_boot_cpus (the OF is trashed by then)
2073 * so we have to put the processors into a holding pattern controlled
2074 * by the kernel (not OF) before we destroy the OF.
2075 *
2076 * This uses a chunk of low memory, puts some holding pattern
2077 * code there and sends the other processors off to there until
2078 * smp_boot_cpus tells them to do something. The holding pattern
2079 * checks that address until its cpu # is there, when it is that
2080 * cpu jumps to __secondary_start(). smp_boot_cpus() takes care
2081 * of setting those values.
2082 *
2083 * We also use physical address 0x4 here to tell when a cpu
2084 * is in its holding pattern code.
2085 *
2086 * -- Cort
2087 */
2088/*
2089 * We want to reference the copy of __secondary_hold_* in the
2090 * 0 - 0x100 address range
2091 */
2092#define LOW_ADDR(x) (((unsigned long) &(x)) & 0xff)
2093
2094static void __init prom_hold_cpus(void)
2095{
2096 unsigned long i;
2097 phandle node;
2098 char type[64];
2099 unsigned long *spinloop
2100 = (void *) LOW_ADDR(__secondary_hold_spinloop);
2101 unsigned long *acknowledge
2102 = (void *) LOW_ADDR(__secondary_hold_acknowledge);
2103 unsigned long secondary_hold = LOW_ADDR(__secondary_hold);
2104
2105 /*
2106 * On pseries, if RTAS supports "query-cpu-stopped-state",
2107 * we skip this stage, the CPUs will be started by the
2108 * kernel using RTAS.
2109 */
2110 if ((of_platform == PLATFORM_PSERIES ||
2111 of_platform == PLATFORM_PSERIES_LPAR) &&
2112 rtas_has_query_cpu_stopped) {
2113 prom_printf("prom_hold_cpus: skipped\n");
2114 return;
2115 }
2116
2117 prom_debug("prom_hold_cpus: start...\n");
2118 prom_debug(" 1) spinloop = 0x%lx\n", (unsigned long)spinloop);
2119 prom_debug(" 1) *spinloop = 0x%lx\n", *spinloop);
2120 prom_debug(" 1) acknowledge = 0x%lx\n",
2121 (unsigned long)acknowledge);
2122 prom_debug(" 1) *acknowledge = 0x%lx\n", *acknowledge);
2123 prom_debug(" 1) secondary_hold = 0x%lx\n", secondary_hold);
2124
2125 /* Set the common spinloop variable, so all of the secondary cpus
2126 * will block when they are awakened from their OF spinloop.
2127 * This must occur for both SMP and non SMP kernels, since OF will
2128 * be trashed when we move the kernel.
2129 */
2130 *spinloop = 0;
2131
2132 /* look for cpus */
2133 for (node = 0; prom_next_node(&node); ) {
2134 unsigned int cpu_no;
2135 __be32 reg;
2136
2137 type[0] = 0;
2138 prom_getprop(node, "device_type", type, sizeof(type));
2139 if (prom_strcmp(type, "cpu") != 0)
2140 continue;
2141
2142 /* Skip non-configured cpus. */
2143 if (prom_getprop(node, "status", type, sizeof(type)) > 0)
2144 if (prom_strcmp(type, "okay") != 0)
2145 continue;
2146
2147 reg = cpu_to_be32(-1); /* make sparse happy */
2148 prom_getprop(node, "reg", ®, sizeof(reg));
2149 cpu_no = be32_to_cpu(reg);
2150
2151 prom_debug("cpu hw idx = %u\n", cpu_no);
2152
2153 /* Init the acknowledge var which will be reset by
2154 * the secondary cpu when it awakens from its OF
2155 * spinloop.
2156 */
2157 *acknowledge = (unsigned long)-1;
2158
2159 if (cpu_no != prom.cpu) {
2160 /* Primary Thread of non-boot cpu or any thread */
2161 prom_printf("starting cpu hw idx %u... ", cpu_no);
2162 call_prom("start-cpu", 3, 0, node,
2163 secondary_hold, cpu_no);
2164
2165 for (i = 0; (i < 100000000) &&
2166 (*acknowledge == ((unsigned long)-1)); i++ )
2167 mb();
2168
2169 if (*acknowledge == cpu_no)
2170 prom_printf("done\n");
2171 else
2172 prom_printf("failed: %lx\n", *acknowledge);
2173 }
2174#ifdef CONFIG_SMP
2175 else
2176 prom_printf("boot cpu hw idx %u\n", cpu_no);
2177#endif /* CONFIG_SMP */
2178 }
2179
2180 prom_debug("prom_hold_cpus: end...\n");
2181}
2182
2183
2184static void __init prom_init_client_services(unsigned long pp)
2185{
2186 /* Get a handle to the prom entry point before anything else */
2187 prom_entry = pp;
2188
2189 /* get a handle for the stdout device */
2190 prom.chosen = call_prom("finddevice", 1, 1, ADDR("/chosen"));
2191 if (!PHANDLE_VALID(prom.chosen))
2192 prom_panic("cannot find chosen"); /* msg won't be printed :( */
2193
2194 /* get device tree root */
2195 prom.root = call_prom("finddevice", 1, 1, ADDR("/"));
2196 if (!PHANDLE_VALID(prom.root))
2197 prom_panic("cannot find device tree root"); /* msg won't be printed :( */
2198
2199 prom.mmumap = 0;
2200}
2201
2202#ifdef CONFIG_PPC32
2203/*
2204 * For really old powermacs, we need to map things we claim.
2205 * For that, we need the ihandle of the mmu.
2206 * Also, on the longtrail, we need to work around other bugs.
2207 */
2208static void __init prom_find_mmu(void)
2209{
2210 phandle oprom;
2211 char version[64];
2212
2213 oprom = call_prom("finddevice", 1, 1, ADDR("/openprom"));
2214 if (!PHANDLE_VALID(oprom))
2215 return;
2216 if (prom_getprop(oprom, "model", version, sizeof(version)) <= 0)
2217 return;
2218 version[sizeof(version) - 1] = 0;
2219 /* XXX might need to add other versions here */
2220 if (prom_strcmp(version, "Open Firmware, 1.0.5") == 0)
2221 of_workarounds = OF_WA_CLAIM;
2222 else if (prom_strncmp(version, "FirmWorks,3.", 12) == 0) {
2223 of_workarounds = OF_WA_CLAIM | OF_WA_LONGTRAIL;
2224 call_prom("interpret", 1, 1, "dev /memory 0 to allow-reclaim");
2225 } else
2226 return;
2227 prom.memory = call_prom("open", 1, 1, ADDR("/memory"));
2228 prom_getprop(prom.chosen, "mmu", &prom.mmumap,
2229 sizeof(prom.mmumap));
2230 prom.mmumap = be32_to_cpu(prom.mmumap);
2231 if (!IHANDLE_VALID(prom.memory) || !IHANDLE_VALID(prom.mmumap))
2232 of_workarounds &= ~OF_WA_CLAIM; /* hmmm */
2233}
2234#else
2235#define prom_find_mmu()
2236#endif
2237
2238static void __init prom_init_stdout(void)
2239{
2240 char *path = of_stdout_device;
2241 char type[16];
2242 phandle stdout_node;
2243 __be32 val;
2244
2245 if (prom_getprop(prom.chosen, "stdout", &val, sizeof(val)) <= 0)
2246 prom_panic("cannot find stdout");
2247
2248 prom.stdout = be32_to_cpu(val);
2249
2250 /* Get the full OF pathname of the stdout device */
2251 memset(path, 0, 256);
2252 call_prom("instance-to-path", 3, 1, prom.stdout, path, 255);
2253 prom_printf("OF stdout device is: %s\n", of_stdout_device);
2254 prom_setprop(prom.chosen, "/chosen", "linux,stdout-path",
2255 path, prom_strlen(path) + 1);
2256
2257 /* instance-to-package fails on PA-Semi */
2258 stdout_node = call_prom("instance-to-package", 1, 1, prom.stdout);
2259 if (stdout_node != PROM_ERROR) {
2260 val = cpu_to_be32(stdout_node);
2261
2262 /* If it's a display, note it */
2263 memset(type, 0, sizeof(type));
2264 prom_getprop(stdout_node, "device_type", type, sizeof(type));
2265 if (prom_strcmp(type, "display") == 0)
2266 prom_setprop(stdout_node, path, "linux,boot-display", NULL, 0);
2267 }
2268}
2269
2270static int __init prom_find_machine_type(void)
2271{
2272 char compat[256];
2273 int len, i = 0;
2274#ifdef CONFIG_PPC64
2275 phandle rtas;
2276 int x;
2277#endif
2278
2279 /* Look for a PowerMac or a Cell */
2280 len = prom_getprop(prom.root, "compatible",
2281 compat, sizeof(compat)-1);
2282 if (len > 0) {
2283 compat[len] = 0;
2284 while (i < len) {
2285 char *p = &compat[i];
2286 int sl = prom_strlen(p);
2287 if (sl == 0)
2288 break;
2289 if (prom_strstr(p, "Power Macintosh") ||
2290 prom_strstr(p, "MacRISC"))
2291 return PLATFORM_POWERMAC;
2292#ifdef CONFIG_PPC64
2293 /* We must make sure we don't detect the IBM Cell
2294 * blades as pSeries due to some firmware issues,
2295 * so we do it here.
2296 */
2297 if (prom_strstr(p, "IBM,CBEA") ||
2298 prom_strstr(p, "IBM,CPBW-1.0"))
2299 return PLATFORM_GENERIC;
2300#endif /* CONFIG_PPC64 */
2301 i += sl + 1;
2302 }
2303 }
2304#ifdef CONFIG_PPC64
2305 /* Try to figure out if it's an IBM pSeries or any other
2306 * PAPR compliant platform. We assume it is if :
2307 * - /device_type is "chrp" (please, do NOT use that for future
2308 * non-IBM designs !
2309 * - it has /rtas
2310 */
2311 len = prom_getprop(prom.root, "device_type",
2312 compat, sizeof(compat)-1);
2313 if (len <= 0)
2314 return PLATFORM_GENERIC;
2315 if (prom_strcmp(compat, "chrp"))
2316 return PLATFORM_GENERIC;
2317
2318 /* Default to pSeries. We need to know if we are running LPAR */
2319 rtas = call_prom("finddevice", 1, 1, ADDR("/rtas"));
2320 if (!PHANDLE_VALID(rtas))
2321 return PLATFORM_GENERIC;
2322 x = prom_getproplen(rtas, "ibm,hypertas-functions");
2323 if (x != PROM_ERROR) {
2324 prom_debug("Hypertas detected, assuming LPAR !\n");
2325 return PLATFORM_PSERIES_LPAR;
2326 }
2327 return PLATFORM_PSERIES;
2328#else
2329 return PLATFORM_GENERIC;
2330#endif
2331}
2332
2333static int __init prom_set_color(ihandle ih, int i, int r, int g, int b)
2334{
2335 return call_prom("call-method", 6, 1, ADDR("color!"), ih, i, b, g, r);
2336}
2337
2338/*
2339 * If we have a display that we don't know how to drive,
2340 * we will want to try to execute OF's open method for it
2341 * later. However, OF will probably fall over if we do that
2342 * we've taken over the MMU.
2343 * So we check whether we will need to open the display,
2344 * and if so, open it now.
2345 */
2346static void __init prom_check_displays(void)
2347{
2348 char type[16], *path;
2349 phandle node;
2350 ihandle ih;
2351 int i;
2352
2353 static const unsigned char default_colors[] __initconst = {
2354 0x00, 0x00, 0x00,
2355 0x00, 0x00, 0xaa,
2356 0x00, 0xaa, 0x00,
2357 0x00, 0xaa, 0xaa,
2358 0xaa, 0x00, 0x00,
2359 0xaa, 0x00, 0xaa,
2360 0xaa, 0xaa, 0x00,
2361 0xaa, 0xaa, 0xaa,
2362 0x55, 0x55, 0x55,
2363 0x55, 0x55, 0xff,
2364 0x55, 0xff, 0x55,
2365 0x55, 0xff, 0xff,
2366 0xff, 0x55, 0x55,
2367 0xff, 0x55, 0xff,
2368 0xff, 0xff, 0x55,
2369 0xff, 0xff, 0xff
2370 };
2371 const unsigned char *clut;
2372
2373 prom_debug("Looking for displays\n");
2374 for (node = 0; prom_next_node(&node); ) {
2375 memset(type, 0, sizeof(type));
2376 prom_getprop(node, "device_type", type, sizeof(type));
2377 if (prom_strcmp(type, "display") != 0)
2378 continue;
2379
2380 /* It seems OF doesn't null-terminate the path :-( */
2381 path = prom_scratch;
2382 memset(path, 0, sizeof(prom_scratch));
2383
2384 /*
2385 * leave some room at the end of the path for appending extra
2386 * arguments
2387 */
2388 if (call_prom("package-to-path", 3, 1, node, path,
2389 sizeof(prom_scratch) - 10) == PROM_ERROR)
2390 continue;
2391 prom_printf("found display : %s, opening... ", path);
2392
2393 ih = call_prom("open", 1, 1, path);
2394 if (ih == 0) {
2395 prom_printf("failed\n");
2396 continue;
2397 }
2398
2399 /* Success */
2400 prom_printf("done\n");
2401 prom_setprop(node, path, "linux,opened", NULL, 0);
2402
2403 /* Setup a usable color table when the appropriate
2404 * method is available. Should update this to set-colors */
2405 clut = default_colors;
2406 for (i = 0; i < 16; i++, clut += 3)
2407 if (prom_set_color(ih, i, clut[0], clut[1],
2408 clut[2]) != 0)
2409 break;
2410
2411#ifdef CONFIG_LOGO_LINUX_CLUT224
2412 clut = PTRRELOC(logo_linux_clut224.clut);
2413 for (i = 0; i < logo_linux_clut224.clutsize; i++, clut += 3)
2414 if (prom_set_color(ih, i + 32, clut[0], clut[1],
2415 clut[2]) != 0)
2416 break;
2417#endif /* CONFIG_LOGO_LINUX_CLUT224 */
2418
2419#ifdef CONFIG_PPC_EARLY_DEBUG_BOOTX
2420 if (prom_getprop(node, "linux,boot-display", NULL, 0) !=
2421 PROM_ERROR) {
2422 u32 width, height, pitch, addr;
2423
2424 prom_printf("Setting btext !\n");
2425 prom_getprop(node, "width", &width, 4);
2426 prom_getprop(node, "height", &height, 4);
2427 prom_getprop(node, "linebytes", &pitch, 4);
2428 prom_getprop(node, "address", &addr, 4);
2429 prom_printf("W=%d H=%d LB=%d addr=0x%x\n",
2430 width, height, pitch, addr);
2431 btext_setup_display(width, height, 8, pitch, addr);
2432 btext_prepare_BAT();
2433 }
2434#endif /* CONFIG_PPC_EARLY_DEBUG_BOOTX */
2435 }
2436}
2437
2438
2439/* Return (relocated) pointer to this much memory: moves initrd if reqd. */
2440static void __init *make_room(unsigned long *mem_start, unsigned long *mem_end,
2441 unsigned long needed, unsigned long align)
2442{
2443 void *ret;
2444
2445 *mem_start = ALIGN(*mem_start, align);
2446 while ((*mem_start + needed) > *mem_end) {
2447 unsigned long room, chunk;
2448
2449 prom_debug("Chunk exhausted, claiming more at %lx...\n",
2450 alloc_bottom);
2451 room = alloc_top - alloc_bottom;
2452 if (room > DEVTREE_CHUNK_SIZE)
2453 room = DEVTREE_CHUNK_SIZE;
2454 if (room < PAGE_SIZE)
2455 prom_panic("No memory for flatten_device_tree "
2456 "(no room)\n");
2457 chunk = alloc_up(room, 0);
2458 if (chunk == 0)
2459 prom_panic("No memory for flatten_device_tree "
2460 "(claim failed)\n");
2461 *mem_end = chunk + room;
2462 }
2463
2464 ret = (void *)*mem_start;
2465 *mem_start += needed;
2466
2467 return ret;
2468}
2469
2470#define dt_push_token(token, mem_start, mem_end) do { \
2471 void *room = make_room(mem_start, mem_end, 4, 4); \
2472 *(__be32 *)room = cpu_to_be32(token); \
2473 } while(0)
2474
2475static unsigned long __init dt_find_string(char *str)
2476{
2477 char *s, *os;
2478
2479 s = os = (char *)dt_string_start;
2480 s += 4;
2481 while (s < (char *)dt_string_end) {
2482 if (prom_strcmp(s, str) == 0)
2483 return s - os;
2484 s += prom_strlen(s) + 1;
2485 }
2486 return 0;
2487}
2488
2489/*
2490 * The Open Firmware 1275 specification states properties must be 31 bytes or
2491 * less, however not all firmwares obey this. Make it 64 bytes to be safe.
2492 */
2493#define MAX_PROPERTY_NAME 64
2494
2495static void __init scan_dt_build_strings(phandle node,
2496 unsigned long *mem_start,
2497 unsigned long *mem_end)
2498{
2499 char *prev_name, *namep, *sstart;
2500 unsigned long soff;
2501 phandle child;
2502
2503 sstart = (char *)dt_string_start;
2504
2505 /* get and store all property names */
2506 prev_name = "";
2507 for (;;) {
2508 /* 64 is max len of name including nul. */
2509 namep = make_room(mem_start, mem_end, MAX_PROPERTY_NAME, 1);
2510 if (call_prom("nextprop", 3, 1, node, prev_name, namep) != 1) {
2511 /* No more nodes: unwind alloc */
2512 *mem_start = (unsigned long)namep;
2513 break;
2514 }
2515
2516 /* skip "name" */
2517 if (prom_strcmp(namep, "name") == 0) {
2518 *mem_start = (unsigned long)namep;
2519 prev_name = "name";
2520 continue;
2521 }
2522 /* get/create string entry */
2523 soff = dt_find_string(namep);
2524 if (soff != 0) {
2525 *mem_start = (unsigned long)namep;
2526 namep = sstart + soff;
2527 } else {
2528 /* Trim off some if we can */
2529 *mem_start = (unsigned long)namep + prom_strlen(namep) + 1;
2530 dt_string_end = *mem_start;
2531 }
2532 prev_name = namep;
2533 }
2534
2535 /* do all our children */
2536 child = call_prom("child", 1, 1, node);
2537 while (child != 0) {
2538 scan_dt_build_strings(child, mem_start, mem_end);
2539 child = call_prom("peer", 1, 1, child);
2540 }
2541}
2542
2543static void __init scan_dt_build_struct(phandle node, unsigned long *mem_start,
2544 unsigned long *mem_end)
2545{
2546 phandle child;
2547 char *namep, *prev_name, *sstart, *p, *ep, *lp, *path;
2548 unsigned long soff;
2549 unsigned char *valp;
2550 static char pname[MAX_PROPERTY_NAME] __prombss;
2551 int l, room, has_phandle = 0;
2552
2553 dt_push_token(OF_DT_BEGIN_NODE, mem_start, mem_end);
2554
2555 /* get the node's full name */
2556 namep = (char *)*mem_start;
2557 room = *mem_end - *mem_start;
2558 if (room > 255)
2559 room = 255;
2560 l = call_prom("package-to-path", 3, 1, node, namep, room);
2561 if (l >= 0) {
2562 /* Didn't fit? Get more room. */
2563 if (l >= room) {
2564 if (l >= *mem_end - *mem_start)
2565 namep = make_room(mem_start, mem_end, l+1, 1);
2566 call_prom("package-to-path", 3, 1, node, namep, l);
2567 }
2568 namep[l] = '\0';
2569
2570 /* Fixup an Apple bug where they have bogus \0 chars in the
2571 * middle of the path in some properties, and extract
2572 * the unit name (everything after the last '/').
2573 */
2574 for (lp = p = namep, ep = namep + l; p < ep; p++) {
2575 if (*p == '/')
2576 lp = namep;
2577 else if (*p != 0)
2578 *lp++ = *p;
2579 }
2580 *lp = 0;
2581 *mem_start = ALIGN((unsigned long)lp + 1, 4);
2582 }
2583
2584 /* get it again for debugging */
2585 path = prom_scratch;
2586 memset(path, 0, sizeof(prom_scratch));
2587 call_prom("package-to-path", 3, 1, node, path, sizeof(prom_scratch) - 1);
2588
2589 /* get and store all properties */
2590 prev_name = "";
2591 sstart = (char *)dt_string_start;
2592 for (;;) {
2593 if (call_prom("nextprop", 3, 1, node, prev_name,
2594 pname) != 1)
2595 break;
2596
2597 /* skip "name" */
2598 if (prom_strcmp(pname, "name") == 0) {
2599 prev_name = "name";
2600 continue;
2601 }
2602
2603 /* find string offset */
2604 soff = dt_find_string(pname);
2605 if (soff == 0) {
2606 prom_printf("WARNING: Can't find string index for"
2607 " <%s>, node %s\n", pname, path);
2608 break;
2609 }
2610 prev_name = sstart + soff;
2611
2612 /* get length */
2613 l = call_prom("getproplen", 2, 1, node, pname);
2614
2615 /* sanity checks */
2616 if (l == PROM_ERROR)
2617 continue;
2618
2619 /* push property head */
2620 dt_push_token(OF_DT_PROP, mem_start, mem_end);
2621 dt_push_token(l, mem_start, mem_end);
2622 dt_push_token(soff, mem_start, mem_end);
2623
2624 /* push property content */
2625 valp = make_room(mem_start, mem_end, l, 4);
2626 call_prom("getprop", 4, 1, node, pname, valp, l);
2627 *mem_start = ALIGN(*mem_start, 4);
2628
2629 if (!prom_strcmp(pname, "phandle"))
2630 has_phandle = 1;
2631 }
2632
2633 /* Add a "phandle" property if none already exist */
2634 if (!has_phandle) {
2635 soff = dt_find_string("phandle");
2636 if (soff == 0)
2637 prom_printf("WARNING: Can't find string index for <phandle> node %s\n", path);
2638 else {
2639 dt_push_token(OF_DT_PROP, mem_start, mem_end);
2640 dt_push_token(4, mem_start, mem_end);
2641 dt_push_token(soff, mem_start, mem_end);
2642 valp = make_room(mem_start, mem_end, 4, 4);
2643 *(__be32 *)valp = cpu_to_be32(node);
2644 }
2645 }
2646
2647 /* do all our children */
2648 child = call_prom("child", 1, 1, node);
2649 while (child != 0) {
2650 scan_dt_build_struct(child, mem_start, mem_end);
2651 child = call_prom("peer", 1, 1, child);
2652 }
2653
2654 dt_push_token(OF_DT_END_NODE, mem_start, mem_end);
2655}
2656
2657static void __init flatten_device_tree(void)
2658{
2659 phandle root;
2660 unsigned long mem_start, mem_end, room;
2661 struct boot_param_header *hdr;
2662 char *namep;
2663 u64 *rsvmap;
2664
2665 /*
2666 * Check how much room we have between alloc top & bottom (+/- a
2667 * few pages), crop to 1MB, as this is our "chunk" size
2668 */
2669 room = alloc_top - alloc_bottom - 0x4000;
2670 if (room > DEVTREE_CHUNK_SIZE)
2671 room = DEVTREE_CHUNK_SIZE;
2672 prom_debug("starting device tree allocs at %lx\n", alloc_bottom);
2673
2674 /* Now try to claim that */
2675 mem_start = (unsigned long)alloc_up(room, PAGE_SIZE);
2676 if (mem_start == 0)
2677 prom_panic("Can't allocate initial device-tree chunk\n");
2678 mem_end = mem_start + room;
2679
2680 /* Get root of tree */
2681 root = call_prom("peer", 1, 1, (phandle)0);
2682 if (root == (phandle)0)
2683 prom_panic ("couldn't get device tree root\n");
2684
2685 /* Build header and make room for mem rsv map */
2686 mem_start = ALIGN(mem_start, 4);
2687 hdr = make_room(&mem_start, &mem_end,
2688 sizeof(struct boot_param_header), 4);
2689 dt_header_start = (unsigned long)hdr;
2690 rsvmap = make_room(&mem_start, &mem_end, sizeof(mem_reserve_map), 8);
2691
2692 /* Start of strings */
2693 mem_start = PAGE_ALIGN(mem_start);
2694 dt_string_start = mem_start;
2695 mem_start += 4; /* hole */
2696
2697 /* Add "phandle" in there, we'll need it */
2698 namep = make_room(&mem_start, &mem_end, 16, 1);
2699 prom_strcpy(namep, "phandle");
2700 mem_start = (unsigned long)namep + prom_strlen(namep) + 1;
2701
2702 /* Build string array */
2703 prom_printf("Building dt strings...\n");
2704 scan_dt_build_strings(root, &mem_start, &mem_end);
2705 dt_string_end = mem_start;
2706
2707 /* Build structure */
2708 mem_start = PAGE_ALIGN(mem_start);
2709 dt_struct_start = mem_start;
2710 prom_printf("Building dt structure...\n");
2711 scan_dt_build_struct(root, &mem_start, &mem_end);
2712 dt_push_token(OF_DT_END, &mem_start, &mem_end);
2713 dt_struct_end = PAGE_ALIGN(mem_start);
2714
2715 /* Finish header */
2716 hdr->boot_cpuid_phys = cpu_to_be32(prom.cpu);
2717 hdr->magic = cpu_to_be32(OF_DT_HEADER);
2718 hdr->totalsize = cpu_to_be32(dt_struct_end - dt_header_start);
2719 hdr->off_dt_struct = cpu_to_be32(dt_struct_start - dt_header_start);
2720 hdr->off_dt_strings = cpu_to_be32(dt_string_start - dt_header_start);
2721 hdr->dt_strings_size = cpu_to_be32(dt_string_end - dt_string_start);
2722 hdr->off_mem_rsvmap = cpu_to_be32(((unsigned long)rsvmap) - dt_header_start);
2723 hdr->version = cpu_to_be32(OF_DT_VERSION);
2724 /* Version 16 is not backward compatible */
2725 hdr->last_comp_version = cpu_to_be32(0x10);
2726
2727 /* Copy the reserve map in */
2728 memcpy(rsvmap, mem_reserve_map, sizeof(mem_reserve_map));
2729
2730#ifdef DEBUG_PROM
2731 {
2732 int i;
2733 prom_printf("reserved memory map:\n");
2734 for (i = 0; i < mem_reserve_cnt; i++)
2735 prom_printf(" %llx - %llx\n",
2736 be64_to_cpu(mem_reserve_map[i].base),
2737 be64_to_cpu(mem_reserve_map[i].size));
2738 }
2739#endif
2740 /* Bump mem_reserve_cnt to cause further reservations to fail
2741 * since it's too late.
2742 */
2743 mem_reserve_cnt = MEM_RESERVE_MAP_SIZE;
2744
2745 prom_printf("Device tree strings 0x%lx -> 0x%lx\n",
2746 dt_string_start, dt_string_end);
2747 prom_printf("Device tree struct 0x%lx -> 0x%lx\n",
2748 dt_struct_start, dt_struct_end);
2749}
2750
2751#ifdef CONFIG_PPC_MAPLE
2752/* PIBS Version 1.05.0000 04/26/2005 has an incorrect /ht/isa/ranges property.
2753 * The values are bad, and it doesn't even have the right number of cells. */
2754static void __init fixup_device_tree_maple(void)
2755{
2756 phandle isa;
2757 u32 rloc = 0x01002000; /* IO space; PCI device = 4 */
2758 u32 isa_ranges[6];
2759 char *name;
2760
2761 name = "/ht@0/isa@4";
2762 isa = call_prom("finddevice", 1, 1, ADDR(name));
2763 if (!PHANDLE_VALID(isa)) {
2764 name = "/ht@0/isa@6";
2765 isa = call_prom("finddevice", 1, 1, ADDR(name));
2766 rloc = 0x01003000; /* IO space; PCI device = 6 */
2767 }
2768 if (!PHANDLE_VALID(isa))
2769 return;
2770
2771 if (prom_getproplen(isa, "ranges") != 12)
2772 return;
2773 if (prom_getprop(isa, "ranges", isa_ranges, sizeof(isa_ranges))
2774 == PROM_ERROR)
2775 return;
2776
2777 if (isa_ranges[0] != 0x1 ||
2778 isa_ranges[1] != 0xf4000000 ||
2779 isa_ranges[2] != 0x00010000)
2780 return;
2781
2782 prom_printf("Fixing up bogus ISA range on Maple/Apache...\n");
2783
2784 isa_ranges[0] = 0x1;
2785 isa_ranges[1] = 0x0;
2786 isa_ranges[2] = rloc;
2787 isa_ranges[3] = 0x0;
2788 isa_ranges[4] = 0x0;
2789 isa_ranges[5] = 0x00010000;
2790 prom_setprop(isa, name, "ranges",
2791 isa_ranges, sizeof(isa_ranges));
2792}
2793
2794#define CPC925_MC_START 0xf8000000
2795#define CPC925_MC_LENGTH 0x1000000
2796/* The values for memory-controller don't have right number of cells */
2797static void __init fixup_device_tree_maple_memory_controller(void)
2798{
2799 phandle mc;
2800 u32 mc_reg[4];
2801 char *name = "/hostbridge@f8000000";
2802 u32 ac, sc;
2803
2804 mc = call_prom("finddevice", 1, 1, ADDR(name));
2805 if (!PHANDLE_VALID(mc))
2806 return;
2807
2808 if (prom_getproplen(mc, "reg") != 8)
2809 return;
2810
2811 prom_getprop(prom.root, "#address-cells", &ac, sizeof(ac));
2812 prom_getprop(prom.root, "#size-cells", &sc, sizeof(sc));
2813 if ((ac != 2) || (sc != 2))
2814 return;
2815
2816 if (prom_getprop(mc, "reg", mc_reg, sizeof(mc_reg)) == PROM_ERROR)
2817 return;
2818
2819 if (mc_reg[0] != CPC925_MC_START || mc_reg[1] != CPC925_MC_LENGTH)
2820 return;
2821
2822 prom_printf("Fixing up bogus hostbridge on Maple...\n");
2823
2824 mc_reg[0] = 0x0;
2825 mc_reg[1] = CPC925_MC_START;
2826 mc_reg[2] = 0x0;
2827 mc_reg[3] = CPC925_MC_LENGTH;
2828 prom_setprop(mc, name, "reg", mc_reg, sizeof(mc_reg));
2829}
2830#else
2831#define fixup_device_tree_maple()
2832#define fixup_device_tree_maple_memory_controller()
2833#endif
2834
2835#ifdef CONFIG_PPC_CHRP
2836/*
2837 * Pegasos and BriQ lacks the "ranges" property in the isa node
2838 * Pegasos needs decimal IRQ 14/15, not hexadecimal
2839 * Pegasos has the IDE configured in legacy mode, but advertised as native
2840 */
2841static void __init fixup_device_tree_chrp(void)
2842{
2843 phandle ph;
2844 u32 prop[6];
2845 u32 rloc = 0x01006000; /* IO space; PCI device = 12 */
2846 char *name;
2847 int rc;
2848
2849 name = "/pci@80000000/isa@c";
2850 ph = call_prom("finddevice", 1, 1, ADDR(name));
2851 if (!PHANDLE_VALID(ph)) {
2852 name = "/pci@ff500000/isa@6";
2853 ph = call_prom("finddevice", 1, 1, ADDR(name));
2854 rloc = 0x01003000; /* IO space; PCI device = 6 */
2855 }
2856 if (PHANDLE_VALID(ph)) {
2857 rc = prom_getproplen(ph, "ranges");
2858 if (rc == 0 || rc == PROM_ERROR) {
2859 prom_printf("Fixing up missing ISA range on Pegasos...\n");
2860
2861 prop[0] = 0x1;
2862 prop[1] = 0x0;
2863 prop[2] = rloc;
2864 prop[3] = 0x0;
2865 prop[4] = 0x0;
2866 prop[5] = 0x00010000;
2867 prom_setprop(ph, name, "ranges", prop, sizeof(prop));
2868 }
2869 }
2870
2871 name = "/pci@80000000/ide@C,1";
2872 ph = call_prom("finddevice", 1, 1, ADDR(name));
2873 if (PHANDLE_VALID(ph)) {
2874 prom_printf("Fixing up IDE interrupt on Pegasos...\n");
2875 prop[0] = 14;
2876 prop[1] = 0x0;
2877 prom_setprop(ph, name, "interrupts", prop, 2*sizeof(u32));
2878 prom_printf("Fixing up IDE class-code on Pegasos...\n");
2879 rc = prom_getprop(ph, "class-code", prop, sizeof(u32));
2880 if (rc == sizeof(u32)) {
2881 prop[0] &= ~0x5;
2882 prom_setprop(ph, name, "class-code", prop, sizeof(u32));
2883 }
2884 }
2885}
2886#else
2887#define fixup_device_tree_chrp()
2888#endif
2889
2890#if defined(CONFIG_PPC64) && defined(CONFIG_PPC_PMAC)
2891static void __init fixup_device_tree_pmac(void)
2892{
2893 phandle u3, i2c, mpic;
2894 u32 u3_rev;
2895 u32 interrupts[2];
2896 u32 parent;
2897
2898 /* Some G5s have a missing interrupt definition, fix it up here */
2899 u3 = call_prom("finddevice", 1, 1, ADDR("/u3@0,f8000000"));
2900 if (!PHANDLE_VALID(u3))
2901 return;
2902 i2c = call_prom("finddevice", 1, 1, ADDR("/u3@0,f8000000/i2c@f8001000"));
2903 if (!PHANDLE_VALID(i2c))
2904 return;
2905 mpic = call_prom("finddevice", 1, 1, ADDR("/u3@0,f8000000/mpic@f8040000"));
2906 if (!PHANDLE_VALID(mpic))
2907 return;
2908
2909 /* check if proper rev of u3 */
2910 if (prom_getprop(u3, "device-rev", &u3_rev, sizeof(u3_rev))
2911 == PROM_ERROR)
2912 return;
2913 if (u3_rev < 0x35 || u3_rev > 0x39)
2914 return;
2915 /* does it need fixup ? */
2916 if (prom_getproplen(i2c, "interrupts") > 0)
2917 return;
2918
2919 prom_printf("fixing up bogus interrupts for u3 i2c...\n");
2920
2921 /* interrupt on this revision of u3 is number 0 and level */
2922 interrupts[0] = 0;
2923 interrupts[1] = 1;
2924 prom_setprop(i2c, "/u3@0,f8000000/i2c@f8001000", "interrupts",
2925 &interrupts, sizeof(interrupts));
2926 parent = (u32)mpic;
2927 prom_setprop(i2c, "/u3@0,f8000000/i2c@f8001000", "interrupt-parent",
2928 &parent, sizeof(parent));
2929}
2930#else
2931#define fixup_device_tree_pmac()
2932#endif
2933
2934#ifdef CONFIG_PPC_EFIKA
2935/*
2936 * The MPC5200 FEC driver requires an phy-handle property to tell it how
2937 * to talk to the phy. If the phy-handle property is missing, then this
2938 * function is called to add the appropriate nodes and link it to the
2939 * ethernet node.
2940 */
2941static void __init fixup_device_tree_efika_add_phy(void)
2942{
2943 u32 node;
2944 char prop[64];
2945 int rv;
2946
2947 /* Check if /builtin/ethernet exists - bail if it doesn't */
2948 node = call_prom("finddevice", 1, 1, ADDR("/builtin/ethernet"));
2949 if (!PHANDLE_VALID(node))
2950 return;
2951
2952 /* Check if the phy-handle property exists - bail if it does */
2953 rv = prom_getprop(node, "phy-handle", prop, sizeof(prop));
2954 if (!rv)
2955 return;
2956
2957 /*
2958 * At this point the ethernet device doesn't have a phy described.
2959 * Now we need to add the missing phy node and linkage
2960 */
2961
2962 /* Check for an MDIO bus node - if missing then create one */
2963 node = call_prom("finddevice", 1, 1, ADDR("/builtin/mdio"));
2964 if (!PHANDLE_VALID(node)) {
2965 prom_printf("Adding Ethernet MDIO node\n");
2966 call_prom("interpret", 1, 1,
2967 " s\" /builtin\" find-device"
2968 " new-device"
2969 " 1 encode-int s\" #address-cells\" property"
2970 " 0 encode-int s\" #size-cells\" property"
2971 " s\" mdio\" device-name"
2972 " s\" fsl,mpc5200b-mdio\" encode-string"
2973 " s\" compatible\" property"
2974 " 0xf0003000 0x400 reg"
2975 " 0x2 encode-int"
2976 " 0x5 encode-int encode+"
2977 " 0x3 encode-int encode+"
2978 " s\" interrupts\" property"
2979 " finish-device");
2980 };
2981
2982 /* Check for a PHY device node - if missing then create one and
2983 * give it's phandle to the ethernet node */
2984 node = call_prom("finddevice", 1, 1,
2985 ADDR("/builtin/mdio/ethernet-phy"));
2986 if (!PHANDLE_VALID(node)) {
2987 prom_printf("Adding Ethernet PHY node\n");
2988 call_prom("interpret", 1, 1,
2989 " s\" /builtin/mdio\" find-device"
2990 " new-device"
2991 " s\" ethernet-phy\" device-name"
2992 " 0x10 encode-int s\" reg\" property"
2993 " my-self"
2994 " ihandle>phandle"
2995 " finish-device"
2996 " s\" /builtin/ethernet\" find-device"
2997 " encode-int"
2998 " s\" phy-handle\" property"
2999 " device-end");
3000 }
3001}
3002
3003static void __init fixup_device_tree_efika(void)
3004{
3005 int sound_irq[3] = { 2, 2, 0 };
3006 int bcomm_irq[3*16] = { 3,0,0, 3,1,0, 3,2,0, 3,3,0,
3007 3,4,0, 3,5,0, 3,6,0, 3,7,0,
3008 3,8,0, 3,9,0, 3,10,0, 3,11,0,
3009 3,12,0, 3,13,0, 3,14,0, 3,15,0 };
3010 u32 node;
3011 char prop[64];
3012 int rv, len;
3013
3014 /* Check if we're really running on a EFIKA */
3015 node = call_prom("finddevice", 1, 1, ADDR("/"));
3016 if (!PHANDLE_VALID(node))
3017 return;
3018
3019 rv = prom_getprop(node, "model", prop, sizeof(prop));
3020 if (rv == PROM_ERROR)
3021 return;
3022 if (prom_strcmp(prop, "EFIKA5K2"))
3023 return;
3024
3025 prom_printf("Applying EFIKA device tree fixups\n");
3026
3027 /* Claiming to be 'chrp' is death */
3028 node = call_prom("finddevice", 1, 1, ADDR("/"));
3029 rv = prom_getprop(node, "device_type", prop, sizeof(prop));
3030 if (rv != PROM_ERROR && (prom_strcmp(prop, "chrp") == 0))
3031 prom_setprop(node, "/", "device_type", "efika", sizeof("efika"));
3032
3033 /* CODEGEN,description is exposed in /proc/cpuinfo so
3034 fix that too */
3035 rv = prom_getprop(node, "CODEGEN,description", prop, sizeof(prop));
3036 if (rv != PROM_ERROR && (prom_strstr(prop, "CHRP")))
3037 prom_setprop(node, "/", "CODEGEN,description",
3038 "Efika 5200B PowerPC System",
3039 sizeof("Efika 5200B PowerPC System"));
3040
3041 /* Fixup bestcomm interrupts property */
3042 node = call_prom("finddevice", 1, 1, ADDR("/builtin/bestcomm"));
3043 if (PHANDLE_VALID(node)) {
3044 len = prom_getproplen(node, "interrupts");
3045 if (len == 12) {
3046 prom_printf("Fixing bestcomm interrupts property\n");
3047 prom_setprop(node, "/builtin/bestcom", "interrupts",
3048 bcomm_irq, sizeof(bcomm_irq));
3049 }
3050 }
3051
3052 /* Fixup sound interrupts property */
3053 node = call_prom("finddevice", 1, 1, ADDR("/builtin/sound"));
3054 if (PHANDLE_VALID(node)) {
3055 rv = prom_getprop(node, "interrupts", prop, sizeof(prop));
3056 if (rv == PROM_ERROR) {
3057 prom_printf("Adding sound interrupts property\n");
3058 prom_setprop(node, "/builtin/sound", "interrupts",
3059 sound_irq, sizeof(sound_irq));
3060 }
3061 }
3062
3063 /* Make sure ethernet phy-handle property exists */
3064 fixup_device_tree_efika_add_phy();
3065}
3066#else
3067#define fixup_device_tree_efika()
3068#endif
3069
3070#ifdef CONFIG_PPC_PASEMI_NEMO
3071/*
3072 * CFE supplied on Nemo is broken in several ways, biggest
3073 * problem is that it reassigns ISA interrupts to unused mpic ints.
3074 * Add an interrupt-controller property for the io-bridge to use
3075 * and correct the ints so we can attach them to an irq_domain
3076 */
3077static void __init fixup_device_tree_pasemi(void)
3078{
3079 u32 interrupts[2], parent, rval, val = 0;
3080 char *name, *pci_name;
3081 phandle iob, node;
3082
3083 /* Find the root pci node */
3084 name = "/pxp@0,e0000000";
3085 iob = call_prom("finddevice", 1, 1, ADDR(name));
3086 if (!PHANDLE_VALID(iob))
3087 return;
3088
3089 /* check if interrupt-controller node set yet */
3090 if (prom_getproplen(iob, "interrupt-controller") !=PROM_ERROR)
3091 return;
3092
3093 prom_printf("adding interrupt-controller property for SB600...\n");
3094
3095 prom_setprop(iob, name, "interrupt-controller", &val, 0);
3096
3097 pci_name = "/pxp@0,e0000000/pci@11";
3098 node = call_prom("finddevice", 1, 1, ADDR(pci_name));
3099 parent = ADDR(iob);
3100
3101 for( ; prom_next_node(&node); ) {
3102 /* scan each node for one with an interrupt */
3103 if (!PHANDLE_VALID(node))
3104 continue;
3105
3106 rval = prom_getproplen(node, "interrupts");
3107 if (rval == 0 || rval == PROM_ERROR)
3108 continue;
3109
3110 prom_getprop(node, "interrupts", &interrupts, sizeof(interrupts));
3111 if ((interrupts[0] < 212) || (interrupts[0] > 222))
3112 continue;
3113
3114 /* found a node, update both interrupts and interrupt-parent */
3115 if ((interrupts[0] >= 212) && (interrupts[0] <= 215))
3116 interrupts[0] -= 203;
3117 if ((interrupts[0] >= 216) && (interrupts[0] <= 220))
3118 interrupts[0] -= 213;
3119 if (interrupts[0] == 221)
3120 interrupts[0] = 14;
3121 if (interrupts[0] == 222)
3122 interrupts[0] = 8;
3123
3124 prom_setprop(node, pci_name, "interrupts", interrupts,
3125 sizeof(interrupts));
3126 prom_setprop(node, pci_name, "interrupt-parent", &parent,
3127 sizeof(parent));
3128 }
3129
3130 /*
3131 * The io-bridge has device_type set to 'io-bridge' change it to 'isa'
3132 * so that generic isa-bridge code can add the SB600 and its on-board
3133 * peripherals.
3134 */
3135 name = "/pxp@0,e0000000/io-bridge@0";
3136 iob = call_prom("finddevice", 1, 1, ADDR(name));
3137 if (!PHANDLE_VALID(iob))
3138 return;
3139
3140 /* device_type is already set, just change it. */
3141
3142 prom_printf("Changing device_type of SB600 node...\n");
3143
3144 prom_setprop(iob, name, "device_type", "isa", sizeof("isa"));
3145}
3146#else /* !CONFIG_PPC_PASEMI_NEMO */
3147static inline void fixup_device_tree_pasemi(void) { }
3148#endif
3149
3150static void __init fixup_device_tree(void)
3151{
3152 fixup_device_tree_maple();
3153 fixup_device_tree_maple_memory_controller();
3154 fixup_device_tree_chrp();
3155 fixup_device_tree_pmac();
3156 fixup_device_tree_efika();
3157 fixup_device_tree_pasemi();
3158}
3159
3160static void __init prom_find_boot_cpu(void)
3161{
3162 __be32 rval;
3163 ihandle prom_cpu;
3164 phandle cpu_pkg;
3165
3166 rval = 0;
3167 if (prom_getprop(prom.chosen, "cpu", &rval, sizeof(rval)) <= 0)
3168 return;
3169 prom_cpu = be32_to_cpu(rval);
3170
3171 cpu_pkg = call_prom("instance-to-package", 1, 1, prom_cpu);
3172
3173 if (!PHANDLE_VALID(cpu_pkg))
3174 return;
3175
3176 prom_getprop(cpu_pkg, "reg", &rval, sizeof(rval));
3177 prom.cpu = be32_to_cpu(rval);
3178
3179 prom_debug("Booting CPU hw index = %d\n", prom.cpu);
3180}
3181
3182static void __init prom_check_initrd(unsigned long r3, unsigned long r4)
3183{
3184#ifdef CONFIG_BLK_DEV_INITRD
3185 if (r3 && r4 && r4 != 0xdeadbeef) {
3186 __be64 val;
3187
3188 prom_initrd_start = is_kernel_addr(r3) ? __pa(r3) : r3;
3189 prom_initrd_end = prom_initrd_start + r4;
3190
3191 val = cpu_to_be64(prom_initrd_start);
3192 prom_setprop(prom.chosen, "/chosen", "linux,initrd-start",
3193 &val, sizeof(val));
3194 val = cpu_to_be64(prom_initrd_end);
3195 prom_setprop(prom.chosen, "/chosen", "linux,initrd-end",
3196 &val, sizeof(val));
3197
3198 reserve_mem(prom_initrd_start,
3199 prom_initrd_end - prom_initrd_start);
3200
3201 prom_debug("initrd_start=0x%lx\n", prom_initrd_start);
3202 prom_debug("initrd_end=0x%lx\n", prom_initrd_end);
3203 }
3204#endif /* CONFIG_BLK_DEV_INITRD */
3205}
3206
3207#ifdef CONFIG_PPC64
3208#ifdef CONFIG_RELOCATABLE
3209static void reloc_toc(void)
3210{
3211}
3212
3213static void unreloc_toc(void)
3214{
3215}
3216#else
3217static void __reloc_toc(unsigned long offset, unsigned long nr_entries)
3218{
3219 unsigned long i;
3220 unsigned long *toc_entry;
3221
3222 /* Get the start of the TOC by using r2 directly. */
3223 asm volatile("addi %0,2,-0x8000" : "=b" (toc_entry));
3224
3225 for (i = 0; i < nr_entries; i++) {
3226 *toc_entry = *toc_entry + offset;
3227 toc_entry++;
3228 }
3229}
3230
3231static void reloc_toc(void)
3232{
3233 unsigned long offset = reloc_offset();
3234 unsigned long nr_entries =
3235 (__prom_init_toc_end - __prom_init_toc_start) / sizeof(long);
3236
3237 __reloc_toc(offset, nr_entries);
3238
3239 mb();
3240}
3241
3242static void unreloc_toc(void)
3243{
3244 unsigned long offset = reloc_offset();
3245 unsigned long nr_entries =
3246 (__prom_init_toc_end - __prom_init_toc_start) / sizeof(long);
3247
3248 mb();
3249
3250 __reloc_toc(-offset, nr_entries);
3251}
3252#endif
3253#endif
3254
3255#ifdef CONFIG_PPC_SVM
3256/*
3257 * Perform the Enter Secure Mode ultracall.
3258 */
3259static int enter_secure_mode(unsigned long kbase, unsigned long fdt)
3260{
3261 register unsigned long r3 asm("r3") = UV_ESM;
3262 register unsigned long r4 asm("r4") = kbase;
3263 register unsigned long r5 asm("r5") = fdt;
3264
3265 asm volatile("sc 2" : "+r"(r3) : "r"(r4), "r"(r5));
3266
3267 return r3;
3268}
3269
3270/*
3271 * Call the Ultravisor to transfer us to secure memory if we have an ESM blob.
3272 */
3273static void __init setup_secure_guest(unsigned long kbase, unsigned long fdt)
3274{
3275 int ret;
3276
3277 if (!prom_svm_enable)
3278 return;
3279
3280 /* Switch to secure mode. */
3281 prom_printf("Switching to secure mode.\n");
3282
3283 /*
3284 * The ultravisor will do an integrity check of the kernel image but we
3285 * relocated it so the check will fail. Restore the original image by
3286 * relocating it back to the kernel virtual base address.
3287 */
3288 if (IS_ENABLED(CONFIG_RELOCATABLE))
3289 relocate(KERNELBASE);
3290
3291 ret = enter_secure_mode(kbase, fdt);
3292
3293 /* Relocate the kernel again. */
3294 if (IS_ENABLED(CONFIG_RELOCATABLE))
3295 relocate(kbase);
3296
3297 if (ret != U_SUCCESS) {
3298 prom_printf("Returned %d from switching to secure mode.\n", ret);
3299 prom_rtas_os_term("Switch to secure mode failed.\n");
3300 }
3301}
3302#else
3303static void __init setup_secure_guest(unsigned long kbase, unsigned long fdt)
3304{
3305}
3306#endif /* CONFIG_PPC_SVM */
3307
3308/*
3309 * We enter here early on, when the Open Firmware prom is still
3310 * handling exceptions and the MMU hash table for us.
3311 */
3312
3313unsigned long __init prom_init(unsigned long r3, unsigned long r4,
3314 unsigned long pp,
3315 unsigned long r6, unsigned long r7,
3316 unsigned long kbase)
3317{
3318 unsigned long hdr;
3319
3320#ifdef CONFIG_PPC32
3321 unsigned long offset = reloc_offset();
3322 reloc_got2(offset);
3323#else
3324 reloc_toc();
3325#endif
3326
3327 /*
3328 * First zero the BSS
3329 */
3330 memset(&__bss_start, 0, __bss_stop - __bss_start);
3331
3332 /*
3333 * Init interface to Open Firmware, get some node references,
3334 * like /chosen
3335 */
3336 prom_init_client_services(pp);
3337
3338 /*
3339 * See if this OF is old enough that we need to do explicit maps
3340 * and other workarounds
3341 */
3342 prom_find_mmu();
3343
3344 /*
3345 * Init prom stdout device
3346 */
3347 prom_init_stdout();
3348
3349 prom_printf("Preparing to boot %s", linux_banner);
3350
3351 /*
3352 * Get default machine type. At this point, we do not differentiate
3353 * between pSeries SMP and pSeries LPAR
3354 */
3355 of_platform = prom_find_machine_type();
3356 prom_printf("Detected machine type: %x\n", of_platform);
3357
3358#ifndef CONFIG_NONSTATIC_KERNEL
3359 /* Bail if this is a kdump kernel. */
3360 if (PHYSICAL_START > 0)
3361 prom_panic("Error: You can't boot a kdump kernel from OF!\n");
3362#endif
3363
3364 /*
3365 * Check for an initrd
3366 */
3367 prom_check_initrd(r3, r4);
3368
3369 /*
3370 * Do early parsing of command line
3371 */
3372 early_cmdline_parse();
3373
3374#ifdef CONFIG_PPC_PSERIES
3375 /*
3376 * On pSeries, inform the firmware about our capabilities
3377 */
3378 if (of_platform == PLATFORM_PSERIES ||
3379 of_platform == PLATFORM_PSERIES_LPAR)
3380 prom_send_capabilities();
3381#endif
3382
3383 /*
3384 * Copy the CPU hold code
3385 */
3386 if (of_platform != PLATFORM_POWERMAC)
3387 copy_and_flush(0, kbase, 0x100, 0);
3388
3389 /*
3390 * Initialize memory management within prom_init
3391 */
3392 prom_init_mem();
3393
3394 /*
3395 * Determine which cpu is actually running right _now_
3396 */
3397 prom_find_boot_cpu();
3398
3399 /*
3400 * Initialize display devices
3401 */
3402 prom_check_displays();
3403
3404#if defined(CONFIG_PPC64) && defined(__BIG_ENDIAN__)
3405 /*
3406 * Initialize IOMMU (TCE tables) on pSeries. Do that before anything else
3407 * that uses the allocator, we need to make sure we get the top of memory
3408 * available for us here...
3409 */
3410 if (of_platform == PLATFORM_PSERIES)
3411 prom_initialize_tce_table();
3412#endif
3413
3414 /*
3415 * On non-powermacs, try to instantiate RTAS. PowerMacs don't
3416 * have a usable RTAS implementation.
3417 */
3418 if (of_platform != PLATFORM_POWERMAC)
3419 prom_instantiate_rtas();
3420
3421#ifdef CONFIG_PPC64
3422 /* instantiate sml */
3423 prom_instantiate_sml();
3424#endif
3425
3426 /*
3427 * On non-powermacs, put all CPUs in spin-loops.
3428 *
3429 * PowerMacs use a different mechanism to spin CPUs
3430 *
3431 * (This must be done after instanciating RTAS)
3432 */
3433 if (of_platform != PLATFORM_POWERMAC)
3434 prom_hold_cpus();
3435
3436 /*
3437 * Fill in some infos for use by the kernel later on
3438 */
3439 if (prom_memory_limit) {
3440 __be64 val = cpu_to_be64(prom_memory_limit);
3441 prom_setprop(prom.chosen, "/chosen", "linux,memory-limit",
3442 &val, sizeof(val));
3443 }
3444#ifdef CONFIG_PPC64
3445 if (prom_iommu_off)
3446 prom_setprop(prom.chosen, "/chosen", "linux,iommu-off",
3447 NULL, 0);
3448
3449 if (prom_iommu_force_on)
3450 prom_setprop(prom.chosen, "/chosen", "linux,iommu-force-on",
3451 NULL, 0);
3452
3453 if (prom_tce_alloc_start) {
3454 prom_setprop(prom.chosen, "/chosen", "linux,tce-alloc-start",
3455 &prom_tce_alloc_start,
3456 sizeof(prom_tce_alloc_start));
3457 prom_setprop(prom.chosen, "/chosen", "linux,tce-alloc-end",
3458 &prom_tce_alloc_end,
3459 sizeof(prom_tce_alloc_end));
3460 }
3461#endif
3462
3463 /*
3464 * Fixup any known bugs in the device-tree
3465 */
3466 fixup_device_tree();
3467
3468 /*
3469 * Now finally create the flattened device-tree
3470 */
3471 prom_printf("copying OF device tree...\n");
3472 flatten_device_tree();
3473
3474 /*
3475 * in case stdin is USB and still active on IBM machines...
3476 * Unfortunately quiesce crashes on some powermacs if we have
3477 * closed stdin already (in particular the powerbook 101).
3478 */
3479 if (of_platform != PLATFORM_POWERMAC)
3480 prom_close_stdin();
3481
3482 /*
3483 * Call OF "quiesce" method to shut down pending DMA's from
3484 * devices etc...
3485 */
3486 prom_printf("Quiescing Open Firmware ...\n");
3487 call_prom("quiesce", 0, 0);
3488
3489 /*
3490 * And finally, call the kernel passing it the flattened device
3491 * tree and NULL as r5, thus triggering the new entry point which
3492 * is common to us and kexec
3493 */
3494 hdr = dt_header_start;
3495
3496 prom_printf("Booting Linux via __start() @ 0x%lx ...\n", kbase);
3497 prom_debug("->dt_header_start=0x%lx\n", hdr);
3498
3499#ifdef CONFIG_PPC32
3500 reloc_got2(-offset);
3501#else
3502 unreloc_toc();
3503#endif
3504
3505 /* Move to secure memory if we're supposed to be secure guests. */
3506 setup_secure_guest(kbase, hdr);
3507
3508 __start(hdr, kbase, 0, 0, 0, 0, 0);
3509
3510 return 0;
3511}
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/pci.h>
41#include <asm/iommu.h>
42#include <asm/btext.h>
43#include <asm/sections.h>
44#include <asm/machdep.h>
45#include <asm/opal.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}
394
395
396static unsigned int __init prom_claim(unsigned long virt, unsigned long size,
397 unsigned long align)
398{
399
400 if (align == 0 && (OF_WORKAROUNDS & OF_WA_CLAIM)) {
401 /*
402 * Old OF requires we claim physical and virtual separately
403 * and then map explicitly (assuming virtual mode)
404 */
405 int ret;
406 prom_arg_t result;
407
408 ret = call_prom_ret("call-method", 5, 2, &result,
409 ADDR("claim"), prom.memory,
410 align, size, virt);
411 if (ret != 0 || result == -1)
412 return -1;
413 ret = call_prom_ret("call-method", 5, 2, &result,
414 ADDR("claim"), prom.mmumap,
415 align, size, virt);
416 if (ret != 0) {
417 call_prom("call-method", 4, 1, ADDR("release"),
418 prom.memory, size, virt);
419 return -1;
420 }
421 /* the 0x12 is M (coherence) + PP == read/write */
422 call_prom("call-method", 6, 1,
423 ADDR("map"), prom.mmumap, 0x12, size, virt, virt);
424 return virt;
425 }
426 return call_prom("claim", 3, 1, (prom_arg_t)virt, (prom_arg_t)size,
427 (prom_arg_t)align);
428}
429
430static void __init __attribute__((noreturn)) prom_panic(const char *reason)
431{
432 prom_print(reason);
433 /* Do not call exit because it clears the screen on pmac
434 * it also causes some sort of double-fault on early pmacs */
435 if (of_platform == PLATFORM_POWERMAC)
436 asm("trap\n");
437
438 /* ToDo: should put up an SRC here on pSeries */
439 call_prom("exit", 0, 0);
440
441 for (;;) /* should never get here */
442 ;
443}
444
445
446static int __init prom_next_node(phandle *nodep)
447{
448 phandle node;
449
450 if ((node = *nodep) != 0
451 && (*nodep = call_prom("child", 1, 1, node)) != 0)
452 return 1;
453 if ((*nodep = call_prom("peer", 1, 1, node)) != 0)
454 return 1;
455 for (;;) {
456 if ((node = call_prom("parent", 1, 1, node)) == 0)
457 return 0;
458 if ((*nodep = call_prom("peer", 1, 1, node)) != 0)
459 return 1;
460 }
461}
462
463static int inline prom_getprop(phandle node, const char *pname,
464 void *value, size_t valuelen)
465{
466 return call_prom("getprop", 4, 1, node, ADDR(pname),
467 (u32)(unsigned long) value, (u32) valuelen);
468}
469
470static int inline prom_getproplen(phandle node, const char *pname)
471{
472 return call_prom("getproplen", 2, 1, node, ADDR(pname));
473}
474
475static void add_string(char **str, const char *q)
476{
477 char *p = *str;
478
479 while (*q)
480 *p++ = *q++;
481 *p++ = ' ';
482 *str = p;
483}
484
485static char *tohex(unsigned int x)
486{
487 static char digits[] = "0123456789abcdef";
488 static char result[9];
489 int i;
490
491 result[8] = 0;
492 i = 8;
493 do {
494 --i;
495 result[i] = digits[x & 0xf];
496 x >>= 4;
497 } while (x != 0 && i > 0);
498 return &result[i];
499}
500
501static int __init prom_setprop(phandle node, const char *nodename,
502 const char *pname, void *value, size_t valuelen)
503{
504 char cmd[256], *p;
505
506 if (!(OF_WORKAROUNDS & OF_WA_LONGTRAIL))
507 return call_prom("setprop", 4, 1, node, ADDR(pname),
508 (u32)(unsigned long) value, (u32) valuelen);
509
510 /* gah... setprop doesn't work on longtrail, have to use interpret */
511 p = cmd;
512 add_string(&p, "dev");
513 add_string(&p, nodename);
514 add_string(&p, tohex((u32)(unsigned long) value));
515 add_string(&p, tohex(valuelen));
516 add_string(&p, tohex(ADDR(pname)));
517 add_string(&p, tohex(strlen(pname)));
518 add_string(&p, "property");
519 *p = 0;
520 return call_prom("interpret", 1, 1, (u32)(unsigned long) cmd);
521}
522
523/* We can't use the standard versions because of relocation headaches. */
524#define isxdigit(c) (('0' <= (c) && (c) <= '9') \
525 || ('a' <= (c) && (c) <= 'f') \
526 || ('A' <= (c) && (c) <= 'F'))
527
528#define isdigit(c) ('0' <= (c) && (c) <= '9')
529#define islower(c) ('a' <= (c) && (c) <= 'z')
530#define toupper(c) (islower(c) ? ((c) - 'a' + 'A') : (c))
531
532static unsigned long prom_strtoul(const char *cp, const char **endp)
533{
534 unsigned long result = 0, base = 10, value;
535
536 if (*cp == '0') {
537 base = 8;
538 cp++;
539 if (toupper(*cp) == 'X') {
540 cp++;
541 base = 16;
542 }
543 }
544
545 while (isxdigit(*cp) &&
546 (value = isdigit(*cp) ? *cp - '0' : toupper(*cp) - 'A' + 10) < base) {
547 result = result * base + value;
548 cp++;
549 }
550
551 if (endp)
552 *endp = cp;
553
554 return result;
555}
556
557static unsigned long prom_memparse(const char *ptr, const char **retptr)
558{
559 unsigned long ret = prom_strtoul(ptr, retptr);
560 int shift = 0;
561
562 /*
563 * We can't use a switch here because GCC *may* generate a
564 * jump table which won't work, because we're not running at
565 * the address we're linked at.
566 */
567 if ('G' == **retptr || 'g' == **retptr)
568 shift = 30;
569
570 if ('M' == **retptr || 'm' == **retptr)
571 shift = 20;
572
573 if ('K' == **retptr || 'k' == **retptr)
574 shift = 10;
575
576 if (shift) {
577 ret <<= shift;
578 (*retptr)++;
579 }
580
581 return ret;
582}
583
584/*
585 * Early parsing of the command line passed to the kernel, used for
586 * "mem=x" and the options that affect the iommu
587 */
588static void __init early_cmdline_parse(void)
589{
590 const char *opt;
591
592 char *p;
593 int l = 0;
594
595 prom_cmd_line[0] = 0;
596 p = prom_cmd_line;
597 if ((long)prom.chosen > 0)
598 l = prom_getprop(prom.chosen, "bootargs", p, COMMAND_LINE_SIZE-1);
599#ifdef CONFIG_CMDLINE
600 if (l <= 0 || p[0] == '\0') /* dbl check */
601 strlcpy(prom_cmd_line,
602 CONFIG_CMDLINE, sizeof(prom_cmd_line));
603#endif /* CONFIG_CMDLINE */
604 prom_printf("command line: %s\n", prom_cmd_line);
605
606#ifdef CONFIG_PPC64
607 opt = strstr(prom_cmd_line, "iommu=");
608 if (opt) {
609 prom_printf("iommu opt is: %s\n", opt);
610 opt += 6;
611 while (*opt && *opt == ' ')
612 opt++;
613 if (!strncmp(opt, "off", 3))
614 prom_iommu_off = 1;
615 else if (!strncmp(opt, "force", 5))
616 prom_iommu_force_on = 1;
617 }
618#endif
619 opt = strstr(prom_cmd_line, "mem=");
620 if (opt) {
621 opt += 4;
622 prom_memory_limit = prom_memparse(opt, (const char **)&opt);
623#ifdef CONFIG_PPC64
624 /* Align to 16 MB == size of ppc64 large page */
625 prom_memory_limit = ALIGN(prom_memory_limit, 0x1000000);
626#endif
627 }
628}
629
630#if defined(CONFIG_PPC_PSERIES) || defined(CONFIG_PPC_POWERNV)
631/*
632 * The architecture vector has an array of PVR mask/value pairs,
633 * followed by # option vectors - 1, followed by the option vectors.
634 *
635 * See prom.h for the definition of the bits specified in the
636 * architecture vector.
637 *
638 * Because the description vector contains a mix of byte and word
639 * values, we declare it as an unsigned char array, and use this
640 * macro to put word values in.
641 */
642#define W(x) ((x) >> 24) & 0xff, ((x) >> 16) & 0xff, \
643 ((x) >> 8) & 0xff, (x) & 0xff
644
645unsigned char ibm_architecture_vec[] = {
646 W(0xfffe0000), W(0x003a0000), /* POWER5/POWER5+ */
647 W(0xffff0000), W(0x003e0000), /* POWER6 */
648 W(0xffff0000), W(0x003f0000), /* POWER7 */
649 W(0xffff0000), W(0x004b0000), /* POWER8E */
650 W(0xffff0000), W(0x004d0000), /* POWER8 */
651 W(0xffffffff), W(0x0f000004), /* all 2.07-compliant */
652 W(0xffffffff), W(0x0f000003), /* all 2.06-compliant */
653 W(0xffffffff), W(0x0f000002), /* all 2.05-compliant */
654 W(0xfffffffe), W(0x0f000001), /* all 2.04-compliant and earlier */
655 6 - 1, /* 6 option vectors */
656
657 /* option vector 1: processor architectures supported */
658 3 - 2, /* length */
659 0, /* don't ignore, don't halt */
660 OV1_PPC_2_00 | OV1_PPC_2_01 | OV1_PPC_2_02 | OV1_PPC_2_03 |
661 OV1_PPC_2_04 | OV1_PPC_2_05 | OV1_PPC_2_06 | OV1_PPC_2_07,
662
663 /* option vector 2: Open Firmware options supported */
664 34 - 2, /* length */
665 OV2_REAL_MODE,
666 0, 0,
667 W(0xffffffff), /* real_base */
668 W(0xffffffff), /* real_size */
669 W(0xffffffff), /* virt_base */
670 W(0xffffffff), /* virt_size */
671 W(0xffffffff), /* load_base */
672 W(256), /* 256MB min RMA */
673 W(0xffffffff), /* full client load */
674 0, /* min RMA percentage of total RAM */
675 48, /* max log_2(hash table size) */
676
677 /* option vector 3: processor options supported */
678 3 - 2, /* length */
679 0, /* don't ignore, don't halt */
680 OV3_FP | OV3_VMX | OV3_DFP,
681
682 /* option vector 4: IBM PAPR implementation */
683 3 - 2, /* length */
684 0, /* don't halt */
685 OV4_MIN_ENT_CAP, /* minimum VP entitled capacity */
686
687 /* option vector 5: PAPR/OF options */
688 19 - 2, /* length */
689 0, /* don't ignore, don't halt */
690 OV5_FEAT(OV5_LPAR) | OV5_FEAT(OV5_SPLPAR) | OV5_FEAT(OV5_LARGE_PAGES) |
691 OV5_FEAT(OV5_DRCONF_MEMORY) | OV5_FEAT(OV5_DONATE_DEDICATE_CPU) |
692#ifdef CONFIG_PCI_MSI
693 /* PCIe/MSI support. Without MSI full PCIe is not supported */
694 OV5_FEAT(OV5_MSI),
695#else
696 0,
697#endif
698 0,
699#ifdef CONFIG_PPC_SMLPAR
700 OV5_FEAT(OV5_CMO) | OV5_FEAT(OV5_XCMO),
701#else
702 0,
703#endif
704 OV5_FEAT(OV5_TYPE1_AFFINITY) | OV5_FEAT(OV5_PRRN),
705 0,
706 0,
707 0,
708 /* WARNING: The offset of the "number of cores" field below
709 * must match by the macro below. Update the definition if
710 * the structure layout changes.
711 */
712#define IBM_ARCH_VEC_NRCORES_OFFSET 125
713 W(NR_CPUS), /* number of cores supported */
714 0,
715 0,
716 0,
717 0,
718 OV5_FEAT(OV5_PFO_HW_RNG) | OV5_FEAT(OV5_PFO_HW_ENCR) |
719 OV5_FEAT(OV5_PFO_HW_842),
720 OV5_FEAT(OV5_SUB_PROCESSORS),
721 /* option vector 6: IBM PAPR hints */
722 4 - 2, /* length */
723 0,
724 0,
725 OV6_LINUX,
726
727};
728
729/* Old method - ELF header with PT_NOTE sections only works on BE */
730#ifdef __BIG_ENDIAN__
731static struct fake_elf {
732 Elf32_Ehdr elfhdr;
733 Elf32_Phdr phdr[2];
734 struct chrpnote {
735 u32 namesz;
736 u32 descsz;
737 u32 type;
738 char name[8]; /* "PowerPC" */
739 struct chrpdesc {
740 u32 real_mode;
741 u32 real_base;
742 u32 real_size;
743 u32 virt_base;
744 u32 virt_size;
745 u32 load_base;
746 } chrpdesc;
747 } chrpnote;
748 struct rpanote {
749 u32 namesz;
750 u32 descsz;
751 u32 type;
752 char name[24]; /* "IBM,RPA-Client-Config" */
753 struct rpadesc {
754 u32 lpar_affinity;
755 u32 min_rmo_size;
756 u32 min_rmo_percent;
757 u32 max_pft_size;
758 u32 splpar;
759 u32 min_load;
760 u32 new_mem_def;
761 u32 ignore_me;
762 } rpadesc;
763 } rpanote;
764} fake_elf = {
765 .elfhdr = {
766 .e_ident = { 0x7f, 'E', 'L', 'F',
767 ELFCLASS32, ELFDATA2MSB, EV_CURRENT },
768 .e_type = ET_EXEC, /* yeah right */
769 .e_machine = EM_PPC,
770 .e_version = EV_CURRENT,
771 .e_phoff = offsetof(struct fake_elf, phdr),
772 .e_phentsize = sizeof(Elf32_Phdr),
773 .e_phnum = 2
774 },
775 .phdr = {
776 [0] = {
777 .p_type = PT_NOTE,
778 .p_offset = offsetof(struct fake_elf, chrpnote),
779 .p_filesz = sizeof(struct chrpnote)
780 }, [1] = {
781 .p_type = PT_NOTE,
782 .p_offset = offsetof(struct fake_elf, rpanote),
783 .p_filesz = sizeof(struct rpanote)
784 }
785 },
786 .chrpnote = {
787 .namesz = sizeof("PowerPC"),
788 .descsz = sizeof(struct chrpdesc),
789 .type = 0x1275,
790 .name = "PowerPC",
791 .chrpdesc = {
792 .real_mode = ~0U, /* ~0 means "don't care" */
793 .real_base = ~0U,
794 .real_size = ~0U,
795 .virt_base = ~0U,
796 .virt_size = ~0U,
797 .load_base = ~0U
798 },
799 },
800 .rpanote = {
801 .namesz = sizeof("IBM,RPA-Client-Config"),
802 .descsz = sizeof(struct rpadesc),
803 .type = 0x12759999,
804 .name = "IBM,RPA-Client-Config",
805 .rpadesc = {
806 .lpar_affinity = 0,
807 .min_rmo_size = 64, /* in megabytes */
808 .min_rmo_percent = 0,
809 .max_pft_size = 48, /* 2^48 bytes max PFT size */
810 .splpar = 1,
811 .min_load = ~0U,
812 .new_mem_def = 0
813 }
814 }
815};
816#endif /* __BIG_ENDIAN__ */
817
818static int __init prom_count_smt_threads(void)
819{
820 phandle node;
821 char type[64];
822 unsigned int plen;
823
824 /* Pick up th first CPU node we can find */
825 for (node = 0; prom_next_node(&node); ) {
826 type[0] = 0;
827 prom_getprop(node, "device_type", type, sizeof(type));
828
829 if (strcmp(type, "cpu"))
830 continue;
831 /*
832 * There is an entry for each smt thread, each entry being
833 * 4 bytes long. All cpus should have the same number of
834 * smt threads, so return after finding the first.
835 */
836 plen = prom_getproplen(node, "ibm,ppc-interrupt-server#s");
837 if (plen == PROM_ERROR)
838 break;
839 plen >>= 2;
840 prom_debug("Found %lu smt threads per core\n", (unsigned long)plen);
841
842 /* Sanity check */
843 if (plen < 1 || plen > 64) {
844 prom_printf("Threads per core %lu out of bounds, assuming 1\n",
845 (unsigned long)plen);
846 return 1;
847 }
848 return plen;
849 }
850 prom_debug("No threads found, assuming 1 per core\n");
851
852 return 1;
853
854}
855
856
857static void __init prom_send_capabilities(void)
858{
859 ihandle root;
860 prom_arg_t ret;
861 u32 cores;
862 unsigned char *ptcores;
863
864 root = call_prom("open", 1, 1, ADDR("/"));
865 if (root != 0) {
866 /* We need to tell the FW about the number of cores we support.
867 *
868 * To do that, we count the number of threads on the first core
869 * (we assume this is the same for all cores) and use it to
870 * divide NR_CPUS.
871 */
872
873 /* The core value may start at an odd address. If such a word
874 * access is made at a cache line boundary, this leads to an
875 * exception which may not be handled at this time.
876 * Forcing a per byte access to avoid exception.
877 */
878 ptcores = &ibm_architecture_vec[IBM_ARCH_VEC_NRCORES_OFFSET];
879 cores = 0;
880 cores |= ptcores[0] << 24;
881 cores |= ptcores[1] << 16;
882 cores |= ptcores[2] << 8;
883 cores |= ptcores[3];
884 if (cores != NR_CPUS) {
885 prom_printf("WARNING ! "
886 "ibm_architecture_vec structure inconsistent: %lu!\n",
887 cores);
888 } else {
889 cores = DIV_ROUND_UP(NR_CPUS, prom_count_smt_threads());
890 prom_printf("Max number of cores passed to firmware: %lu (NR_CPUS = %lu)\n",
891 cores, NR_CPUS);
892 ptcores[0] = (cores >> 24) & 0xff;
893 ptcores[1] = (cores >> 16) & 0xff;
894 ptcores[2] = (cores >> 8) & 0xff;
895 ptcores[3] = cores & 0xff;
896 }
897
898 /* try calling the ibm,client-architecture-support method */
899 prom_printf("Calling ibm,client-architecture-support...");
900 if (call_prom_ret("call-method", 3, 2, &ret,
901 ADDR("ibm,client-architecture-support"),
902 root,
903 ADDR(ibm_architecture_vec)) == 0) {
904 /* the call exists... */
905 if (ret)
906 prom_printf("\nWARNING: ibm,client-architecture"
907 "-support call FAILED!\n");
908 call_prom("close", 1, 0, root);
909 prom_printf(" done\n");
910 return;
911 }
912 call_prom("close", 1, 0, root);
913 prom_printf(" not implemented\n");
914 }
915
916#ifdef __BIG_ENDIAN__
917 {
918 ihandle elfloader;
919
920 /* no ibm,client-architecture-support call, try the old way */
921 elfloader = call_prom("open", 1, 1,
922 ADDR("/packages/elf-loader"));
923 if (elfloader == 0) {
924 prom_printf("couldn't open /packages/elf-loader\n");
925 return;
926 }
927 call_prom("call-method", 3, 1, ADDR("process-elf-header"),
928 elfloader, ADDR(&fake_elf));
929 call_prom("close", 1, 0, elfloader);
930 }
931#endif /* __BIG_ENDIAN__ */
932}
933#endif /* #if defined(CONFIG_PPC_PSERIES) || defined(CONFIG_PPC_POWERNV) */
934
935/*
936 * Memory allocation strategy... our layout is normally:
937 *
938 * at 14Mb or more we have vmlinux, then a gap and initrd. In some
939 * rare cases, initrd might end up being before the kernel though.
940 * We assume this won't override the final kernel at 0, we have no
941 * provision to handle that in this version, but it should hopefully
942 * never happen.
943 *
944 * alloc_top is set to the top of RMO, eventually shrink down if the
945 * TCEs overlap
946 *
947 * alloc_bottom is set to the top of kernel/initrd
948 *
949 * from there, allocations are done this way : rtas is allocated
950 * topmost, and the device-tree is allocated from the bottom. We try
951 * to grow the device-tree allocation as we progress. If we can't,
952 * then we fail, we don't currently have a facility to restart
953 * elsewhere, but that shouldn't be necessary.
954 *
955 * Note that calls to reserve_mem have to be done explicitly, memory
956 * allocated with either alloc_up or alloc_down isn't automatically
957 * reserved.
958 */
959
960
961/*
962 * Allocates memory in the RMO upward from the kernel/initrd
963 *
964 * When align is 0, this is a special case, it means to allocate in place
965 * at the current location of alloc_bottom or fail (that is basically
966 * extending the previous allocation). Used for the device-tree flattening
967 */
968static unsigned long __init alloc_up(unsigned long size, unsigned long align)
969{
970 unsigned long base = alloc_bottom;
971 unsigned long addr = 0;
972
973 if (align)
974 base = _ALIGN_UP(base, align);
975 prom_debug("alloc_up(%x, %x)\n", size, align);
976 if (ram_top == 0)
977 prom_panic("alloc_up() called with mem not initialized\n");
978
979 if (align)
980 base = _ALIGN_UP(alloc_bottom, align);
981 else
982 base = alloc_bottom;
983
984 for(; (base + size) <= alloc_top;
985 base = _ALIGN_UP(base + 0x100000, align)) {
986 prom_debug(" trying: 0x%x\n\r", base);
987 addr = (unsigned long)prom_claim(base, size, 0);
988 if (addr != PROM_ERROR && addr != 0)
989 break;
990 addr = 0;
991 if (align == 0)
992 break;
993 }
994 if (addr == 0)
995 return 0;
996 alloc_bottom = addr + size;
997
998 prom_debug(" -> %x\n", addr);
999 prom_debug(" alloc_bottom : %x\n", alloc_bottom);
1000 prom_debug(" alloc_top : %x\n", alloc_top);
1001 prom_debug(" alloc_top_hi : %x\n", alloc_top_high);
1002 prom_debug(" rmo_top : %x\n", rmo_top);
1003 prom_debug(" ram_top : %x\n", ram_top);
1004
1005 return addr;
1006}
1007
1008/*
1009 * Allocates memory downward, either from top of RMO, or if highmem
1010 * is set, from the top of RAM. Note that this one doesn't handle
1011 * failures. It does claim memory if highmem is not set.
1012 */
1013static unsigned long __init alloc_down(unsigned long size, unsigned long align,
1014 int highmem)
1015{
1016 unsigned long base, addr = 0;
1017
1018 prom_debug("alloc_down(%x, %x, %s)\n", size, align,
1019 highmem ? "(high)" : "(low)");
1020 if (ram_top == 0)
1021 prom_panic("alloc_down() called with mem not initialized\n");
1022
1023 if (highmem) {
1024 /* Carve out storage for the TCE table. */
1025 addr = _ALIGN_DOWN(alloc_top_high - size, align);
1026 if (addr <= alloc_bottom)
1027 return 0;
1028 /* Will we bump into the RMO ? If yes, check out that we
1029 * didn't overlap existing allocations there, if we did,
1030 * we are dead, we must be the first in town !
1031 */
1032 if (addr < rmo_top) {
1033 /* Good, we are first */
1034 if (alloc_top == rmo_top)
1035 alloc_top = rmo_top = addr;
1036 else
1037 return 0;
1038 }
1039 alloc_top_high = addr;
1040 goto bail;
1041 }
1042
1043 base = _ALIGN_DOWN(alloc_top - size, align);
1044 for (; base > alloc_bottom;
1045 base = _ALIGN_DOWN(base - 0x100000, align)) {
1046 prom_debug(" trying: 0x%x\n\r", base);
1047 addr = (unsigned long)prom_claim(base, size, 0);
1048 if (addr != PROM_ERROR && addr != 0)
1049 break;
1050 addr = 0;
1051 }
1052 if (addr == 0)
1053 return 0;
1054 alloc_top = addr;
1055
1056 bail:
1057 prom_debug(" -> %x\n", addr);
1058 prom_debug(" alloc_bottom : %x\n", alloc_bottom);
1059 prom_debug(" alloc_top : %x\n", alloc_top);
1060 prom_debug(" alloc_top_hi : %x\n", alloc_top_high);
1061 prom_debug(" rmo_top : %x\n", rmo_top);
1062 prom_debug(" ram_top : %x\n", ram_top);
1063
1064 return addr;
1065}
1066
1067/*
1068 * Parse a "reg" cell
1069 */
1070static unsigned long __init prom_next_cell(int s, cell_t **cellp)
1071{
1072 cell_t *p = *cellp;
1073 unsigned long r = 0;
1074
1075 /* Ignore more than 2 cells */
1076 while (s > sizeof(unsigned long) / 4) {
1077 p++;
1078 s--;
1079 }
1080 r = be32_to_cpu(*p++);
1081#ifdef CONFIG_PPC64
1082 if (s > 1) {
1083 r <<= 32;
1084 r |= be32_to_cpu(*(p++));
1085 }
1086#endif
1087 *cellp = p;
1088 return r;
1089}
1090
1091/*
1092 * Very dumb function for adding to the memory reserve list, but
1093 * we don't need anything smarter at this point
1094 *
1095 * XXX Eventually check for collisions. They should NEVER happen.
1096 * If problems seem to show up, it would be a good start to track
1097 * them down.
1098 */
1099static void __init reserve_mem(u64 base, u64 size)
1100{
1101 u64 top = base + size;
1102 unsigned long cnt = mem_reserve_cnt;
1103
1104 if (size == 0)
1105 return;
1106
1107 /* We need to always keep one empty entry so that we
1108 * have our terminator with "size" set to 0 since we are
1109 * dumb and just copy this entire array to the boot params
1110 */
1111 base = _ALIGN_DOWN(base, PAGE_SIZE);
1112 top = _ALIGN_UP(top, PAGE_SIZE);
1113 size = top - base;
1114
1115 if (cnt >= (MEM_RESERVE_MAP_SIZE - 1))
1116 prom_panic("Memory reserve map exhausted !\n");
1117 mem_reserve_map[cnt].base = cpu_to_be64(base);
1118 mem_reserve_map[cnt].size = cpu_to_be64(size);
1119 mem_reserve_cnt = cnt + 1;
1120}
1121
1122/*
1123 * Initialize memory allocation mechanism, parse "memory" nodes and
1124 * obtain that way the top of memory and RMO to setup out local allocator
1125 */
1126static void __init prom_init_mem(void)
1127{
1128 phandle node;
1129 char *path, type[64];
1130 unsigned int plen;
1131 cell_t *p, *endp;
1132 __be32 val;
1133 u32 rac, rsc;
1134
1135 /*
1136 * We iterate the memory nodes to find
1137 * 1) top of RMO (first node)
1138 * 2) top of memory
1139 */
1140 val = cpu_to_be32(2);
1141 prom_getprop(prom.root, "#address-cells", &val, sizeof(val));
1142 rac = be32_to_cpu(val);
1143 val = cpu_to_be32(1);
1144 prom_getprop(prom.root, "#size-cells", &val, sizeof(rsc));
1145 rsc = be32_to_cpu(val);
1146 prom_debug("root_addr_cells: %x\n", rac);
1147 prom_debug("root_size_cells: %x\n", rsc);
1148
1149 prom_debug("scanning memory:\n");
1150 path = prom_scratch;
1151
1152 for (node = 0; prom_next_node(&node); ) {
1153 type[0] = 0;
1154 prom_getprop(node, "device_type", type, sizeof(type));
1155
1156 if (type[0] == 0) {
1157 /*
1158 * CHRP Longtrail machines have no device_type
1159 * on the memory node, so check the name instead...
1160 */
1161 prom_getprop(node, "name", type, sizeof(type));
1162 }
1163 if (strcmp(type, "memory"))
1164 continue;
1165
1166 plen = prom_getprop(node, "reg", regbuf, sizeof(regbuf));
1167 if (plen > sizeof(regbuf)) {
1168 prom_printf("memory node too large for buffer !\n");
1169 plen = sizeof(regbuf);
1170 }
1171 p = regbuf;
1172 endp = p + (plen / sizeof(cell_t));
1173
1174#ifdef DEBUG_PROM
1175 memset(path, 0, PROM_SCRATCH_SIZE);
1176 call_prom("package-to-path", 3, 1, node, path, PROM_SCRATCH_SIZE-1);
1177 prom_debug(" node %s :\n", path);
1178#endif /* DEBUG_PROM */
1179
1180 while ((endp - p) >= (rac + rsc)) {
1181 unsigned long base, size;
1182
1183 base = prom_next_cell(rac, &p);
1184 size = prom_next_cell(rsc, &p);
1185
1186 if (size == 0)
1187 continue;
1188 prom_debug(" %x %x\n", base, size);
1189 if (base == 0 && (of_platform & PLATFORM_LPAR))
1190 rmo_top = size;
1191 if ((base + size) > ram_top)
1192 ram_top = base + size;
1193 }
1194 }
1195
1196 alloc_bottom = PAGE_ALIGN((unsigned long)&_end + 0x4000);
1197
1198 /*
1199 * If prom_memory_limit is set we reduce the upper limits *except* for
1200 * alloc_top_high. This must be the real top of RAM so we can put
1201 * TCE's up there.
1202 */
1203
1204 alloc_top_high = ram_top;
1205
1206 if (prom_memory_limit) {
1207 if (prom_memory_limit <= alloc_bottom) {
1208 prom_printf("Ignoring mem=%x <= alloc_bottom.\n",
1209 prom_memory_limit);
1210 prom_memory_limit = 0;
1211 } else if (prom_memory_limit >= ram_top) {
1212 prom_printf("Ignoring mem=%x >= ram_top.\n",
1213 prom_memory_limit);
1214 prom_memory_limit = 0;
1215 } else {
1216 ram_top = prom_memory_limit;
1217 rmo_top = min(rmo_top, prom_memory_limit);
1218 }
1219 }
1220
1221 /*
1222 * Setup our top alloc point, that is top of RMO or top of
1223 * segment 0 when running non-LPAR.
1224 * Some RS64 machines have buggy firmware where claims up at
1225 * 1GB fail. Cap at 768MB as a workaround.
1226 * Since 768MB is plenty of room, and we need to cap to something
1227 * reasonable on 32-bit, cap at 768MB on all machines.
1228 */
1229 if (!rmo_top)
1230 rmo_top = ram_top;
1231 rmo_top = min(0x30000000ul, rmo_top);
1232 alloc_top = rmo_top;
1233 alloc_top_high = ram_top;
1234
1235 /*
1236 * Check if we have an initrd after the kernel but still inside
1237 * the RMO. If we do move our bottom point to after it.
1238 */
1239 if (prom_initrd_start &&
1240 prom_initrd_start < rmo_top &&
1241 prom_initrd_end > alloc_bottom)
1242 alloc_bottom = PAGE_ALIGN(prom_initrd_end);
1243
1244 prom_printf("memory layout at init:\n");
1245 prom_printf(" memory_limit : %x (16 MB aligned)\n", prom_memory_limit);
1246 prom_printf(" alloc_bottom : %x\n", alloc_bottom);
1247 prom_printf(" alloc_top : %x\n", alloc_top);
1248 prom_printf(" alloc_top_hi : %x\n", alloc_top_high);
1249 prom_printf(" rmo_top : %x\n", rmo_top);
1250 prom_printf(" ram_top : %x\n", ram_top);
1251}
1252
1253static void __init prom_close_stdin(void)
1254{
1255 __be32 val;
1256 ihandle stdin;
1257
1258 if (prom_getprop(prom.chosen, "stdin", &val, sizeof(val)) > 0) {
1259 stdin = be32_to_cpu(val);
1260 call_prom("close", 1, 0, stdin);
1261 }
1262}
1263
1264#ifdef CONFIG_PPC_POWERNV
1265
1266#ifdef CONFIG_PPC_EARLY_DEBUG_OPAL
1267static u64 __initdata prom_opal_base;
1268static u64 __initdata prom_opal_entry;
1269#endif
1270
1271#ifdef __BIG_ENDIAN__
1272/* XXX Don't change this structure without updating opal-takeover.S */
1273static struct opal_secondary_data {
1274 s64 ack; /* 0 */
1275 u64 go; /* 8 */
1276 struct opal_takeover_args args; /* 16 */
1277} opal_secondary_data;
1278
1279static u64 __initdata prom_opal_align;
1280static u64 __initdata prom_opal_size;
1281static int __initdata prom_rtas_start_cpu;
1282static u64 __initdata prom_rtas_data;
1283static u64 __initdata prom_rtas_entry;
1284
1285extern char opal_secondary_entry;
1286
1287static void __init prom_query_opal(void)
1288{
1289 long rc;
1290
1291 /* We must not query for OPAL presence on a machine that
1292 * supports TNK takeover (970 blades), as this uses the same
1293 * h-call with different arguments and will crash
1294 */
1295 if (PHANDLE_VALID(call_prom("finddevice", 1, 1,
1296 ADDR("/tnk-memory-map")))) {
1297 prom_printf("TNK takeover detected, skipping OPAL check\n");
1298 return;
1299 }
1300
1301 prom_printf("Querying for OPAL presence... ");
1302
1303 rc = opal_query_takeover(&prom_opal_size,
1304 &prom_opal_align);
1305 prom_debug("(rc = %ld) ", rc);
1306 if (rc != 0) {
1307 prom_printf("not there.\n");
1308 return;
1309 }
1310 of_platform = PLATFORM_OPAL;
1311 prom_printf(" there !\n");
1312 prom_debug(" opal_size = 0x%lx\n", prom_opal_size);
1313 prom_debug(" opal_align = 0x%lx\n", prom_opal_align);
1314 if (prom_opal_align < 0x10000)
1315 prom_opal_align = 0x10000;
1316}
1317
1318static int __init prom_rtas_call(int token, int nargs, int nret,
1319 int *outputs, ...)
1320{
1321 struct rtas_args rtas_args;
1322 va_list list;
1323 int i;
1324
1325 rtas_args.token = token;
1326 rtas_args.nargs = nargs;
1327 rtas_args.nret = nret;
1328 rtas_args.rets = (rtas_arg_t *)&(rtas_args.args[nargs]);
1329 va_start(list, outputs);
1330 for (i = 0; i < nargs; ++i)
1331 rtas_args.args[i] = va_arg(list, rtas_arg_t);
1332 va_end(list);
1333
1334 for (i = 0; i < nret; ++i)
1335 rtas_args.rets[i] = 0;
1336
1337 opal_enter_rtas(&rtas_args, prom_rtas_data,
1338 prom_rtas_entry);
1339
1340 if (nret > 1 && outputs != NULL)
1341 for (i = 0; i < nret-1; ++i)
1342 outputs[i] = rtas_args.rets[i+1];
1343 return (nret > 0)? rtas_args.rets[0]: 0;
1344}
1345
1346static void __init prom_opal_hold_cpus(void)
1347{
1348 int i, cnt, cpu, rc;
1349 long j;
1350 phandle node;
1351 char type[64];
1352 u32 servers[8];
1353 void *entry = (unsigned long *)&opal_secondary_entry;
1354 struct opal_secondary_data *data = &opal_secondary_data;
1355
1356 prom_debug("prom_opal_hold_cpus: start...\n");
1357 prom_debug(" - entry = 0x%x\n", entry);
1358 prom_debug(" - data = 0x%x\n", data);
1359
1360 data->ack = -1;
1361 data->go = 0;
1362
1363 /* look for cpus */
1364 for (node = 0; prom_next_node(&node); ) {
1365 type[0] = 0;
1366 prom_getprop(node, "device_type", type, sizeof(type));
1367 if (strcmp(type, "cpu") != 0)
1368 continue;
1369
1370 /* Skip non-configured cpus. */
1371 if (prom_getprop(node, "status", type, sizeof(type)) > 0)
1372 if (strcmp(type, "okay") != 0)
1373 continue;
1374
1375 cnt = prom_getprop(node, "ibm,ppc-interrupt-server#s", servers,
1376 sizeof(servers));
1377 if (cnt == PROM_ERROR)
1378 break;
1379 cnt >>= 2;
1380 for (i = 0; i < cnt; i++) {
1381 cpu = servers[i];
1382 prom_debug("CPU %d ... ", cpu);
1383 if (cpu == prom.cpu) {
1384 prom_debug("booted !\n");
1385 continue;
1386 }
1387 prom_debug("starting ... ");
1388
1389 /* Init the acknowledge var which will be reset by
1390 * the secondary cpu when it awakens from its OF
1391 * spinloop.
1392 */
1393 data->ack = -1;
1394 rc = prom_rtas_call(prom_rtas_start_cpu, 3, 1,
1395 NULL, cpu, entry, data);
1396 prom_debug("rtas rc=%d ...", rc);
1397
1398 for (j = 0; j < 100000000 && data->ack == -1; j++) {
1399 HMT_low();
1400 mb();
1401 }
1402 HMT_medium();
1403 if (data->ack != -1)
1404 prom_debug("done, PIR=0x%x\n", data->ack);
1405 else
1406 prom_debug("timeout !\n");
1407 }
1408 }
1409 prom_debug("prom_opal_hold_cpus: end...\n");
1410}
1411
1412static void __init prom_opal_takeover(void)
1413{
1414 struct opal_secondary_data *data = &opal_secondary_data;
1415 struct opal_takeover_args *args = &data->args;
1416 u64 align = prom_opal_align;
1417 u64 top_addr, opal_addr;
1418
1419 args->k_image = (u64)_stext;
1420 args->k_size = _end - _stext;
1421 args->k_entry = 0;
1422 args->k_entry2 = 0x60;
1423
1424 top_addr = _ALIGN_UP(args->k_size, align);
1425
1426 if (prom_initrd_start != 0) {
1427 args->rd_image = prom_initrd_start;
1428 args->rd_size = prom_initrd_end - args->rd_image;
1429 args->rd_loc = top_addr;
1430 top_addr = _ALIGN_UP(args->rd_loc + args->rd_size, align);
1431 }
1432
1433 /* Pickup an address for the HAL. We want to go really high
1434 * up to avoid problem with future kexecs. On the other hand
1435 * we don't want to be all over the TCEs on P5IOC2 machines
1436 * which are going to be up there too. We assume the machine
1437 * has plenty of memory, and we ask for the HAL for now to
1438 * be just below the 1G point, or above the initrd
1439 */
1440 opal_addr = _ALIGN_DOWN(0x40000000 - prom_opal_size, align);
1441 if (opal_addr < top_addr)
1442 opal_addr = top_addr;
1443 args->hal_addr = opal_addr;
1444
1445 /* Copy the command line to the kernel image */
1446 strlcpy(boot_command_line, prom_cmd_line,
1447 COMMAND_LINE_SIZE);
1448
1449 prom_debug(" k_image = 0x%lx\n", args->k_image);
1450 prom_debug(" k_size = 0x%lx\n", args->k_size);
1451 prom_debug(" k_entry = 0x%lx\n", args->k_entry);
1452 prom_debug(" k_entry2 = 0x%lx\n", args->k_entry2);
1453 prom_debug(" hal_addr = 0x%lx\n", args->hal_addr);
1454 prom_debug(" rd_image = 0x%lx\n", args->rd_image);
1455 prom_debug(" rd_size = 0x%lx\n", args->rd_size);
1456 prom_debug(" rd_loc = 0x%lx\n", args->rd_loc);
1457 prom_printf("Performing OPAL takeover,this can take a few minutes..\n");
1458 prom_close_stdin();
1459 mb();
1460 data->go = 1;
1461 for (;;)
1462 opal_do_takeover(args);
1463}
1464#endif /* __BIG_ENDIAN__ */
1465
1466/*
1467 * Allocate room for and instantiate OPAL
1468 */
1469static void __init prom_instantiate_opal(void)
1470{
1471 phandle opal_node;
1472 ihandle opal_inst;
1473 u64 base, entry;
1474 u64 size = 0, align = 0x10000;
1475 __be64 val64;
1476 u32 rets[2];
1477
1478 prom_debug("prom_instantiate_opal: start...\n");
1479
1480 opal_node = call_prom("finddevice", 1, 1, ADDR("/ibm,opal"));
1481 prom_debug("opal_node: %x\n", opal_node);
1482 if (!PHANDLE_VALID(opal_node))
1483 return;
1484
1485 val64 = 0;
1486 prom_getprop(opal_node, "opal-runtime-size", &val64, sizeof(val64));
1487 size = be64_to_cpu(val64);
1488 if (size == 0)
1489 return;
1490 val64 = 0;
1491 prom_getprop(opal_node, "opal-runtime-alignment", &val64,sizeof(val64));
1492 align = be64_to_cpu(val64);
1493
1494 base = alloc_down(size, align, 0);
1495 if (base == 0) {
1496 prom_printf("OPAL allocation failed !\n");
1497 return;
1498 }
1499
1500 opal_inst = call_prom("open", 1, 1, ADDR("/ibm,opal"));
1501 if (!IHANDLE_VALID(opal_inst)) {
1502 prom_printf("opening opal package failed (%x)\n", opal_inst);
1503 return;
1504 }
1505
1506 prom_printf("instantiating opal at 0x%x...", base);
1507
1508 if (call_prom_ret("call-method", 4, 3, rets,
1509 ADDR("load-opal-runtime"),
1510 opal_inst,
1511 base >> 32, base & 0xffffffff) != 0
1512 || (rets[0] == 0 && rets[1] == 0)) {
1513 prom_printf(" failed\n");
1514 return;
1515 }
1516 entry = (((u64)rets[0]) << 32) | rets[1];
1517
1518 prom_printf(" done\n");
1519
1520 reserve_mem(base, size);
1521
1522 prom_debug("opal base = 0x%x\n", base);
1523 prom_debug("opal align = 0x%x\n", align);
1524 prom_debug("opal entry = 0x%x\n", entry);
1525 prom_debug("opal size = 0x%x\n", (long)size);
1526
1527 prom_setprop(opal_node, "/ibm,opal", "opal-base-address",
1528 &base, sizeof(base));
1529 prom_setprop(opal_node, "/ibm,opal", "opal-entry-address",
1530 &entry, sizeof(entry));
1531
1532#ifdef CONFIG_PPC_EARLY_DEBUG_OPAL
1533 prom_opal_base = base;
1534 prom_opal_entry = entry;
1535#endif
1536 prom_debug("prom_instantiate_opal: end...\n");
1537}
1538
1539#endif /* CONFIG_PPC_POWERNV */
1540
1541/*
1542 * Allocate room for and instantiate RTAS
1543 */
1544static void __init prom_instantiate_rtas(void)
1545{
1546 phandle rtas_node;
1547 ihandle rtas_inst;
1548 u32 base, entry = 0;
1549 __be32 val;
1550 u32 size = 0;
1551
1552 prom_debug("prom_instantiate_rtas: start...\n");
1553
1554 rtas_node = call_prom("finddevice", 1, 1, ADDR("/rtas"));
1555 prom_debug("rtas_node: %x\n", rtas_node);
1556 if (!PHANDLE_VALID(rtas_node))
1557 return;
1558
1559 val = 0;
1560 prom_getprop(rtas_node, "rtas-size", &val, sizeof(size));
1561 size = be32_to_cpu(val);
1562 if (size == 0)
1563 return;
1564
1565 base = alloc_down(size, PAGE_SIZE, 0);
1566 if (base == 0)
1567 prom_panic("Could not allocate memory for RTAS\n");
1568
1569 rtas_inst = call_prom("open", 1, 1, ADDR("/rtas"));
1570 if (!IHANDLE_VALID(rtas_inst)) {
1571 prom_printf("opening rtas package failed (%x)\n", rtas_inst);
1572 return;
1573 }
1574
1575 prom_printf("instantiating rtas at 0x%x...", base);
1576
1577 if (call_prom_ret("call-method", 3, 2, &entry,
1578 ADDR("instantiate-rtas"),
1579 rtas_inst, base) != 0
1580 || entry == 0) {
1581 prom_printf(" failed\n");
1582 return;
1583 }
1584 prom_printf(" done\n");
1585
1586 reserve_mem(base, size);
1587
1588 val = cpu_to_be32(base);
1589 prom_setprop(rtas_node, "/rtas", "linux,rtas-base",
1590 &val, sizeof(val));
1591 val = cpu_to_be32(entry);
1592 prom_setprop(rtas_node, "/rtas", "linux,rtas-entry",
1593 &val, sizeof(val));
1594
1595 /* Check if it supports "query-cpu-stopped-state" */
1596 if (prom_getprop(rtas_node, "query-cpu-stopped-state",
1597 &val, sizeof(val)) != PROM_ERROR)
1598 rtas_has_query_cpu_stopped = true;
1599
1600#if defined(CONFIG_PPC_POWERNV) && defined(__BIG_ENDIAN__)
1601 /* PowerVN takeover hack */
1602 prom_rtas_data = base;
1603 prom_rtas_entry = entry;
1604 prom_getprop(rtas_node, "start-cpu", &prom_rtas_start_cpu, 4);
1605#endif
1606 prom_debug("rtas base = 0x%x\n", base);
1607 prom_debug("rtas entry = 0x%x\n", entry);
1608 prom_debug("rtas size = 0x%x\n", (long)size);
1609
1610 prom_debug("prom_instantiate_rtas: end...\n");
1611}
1612
1613#ifdef CONFIG_PPC64
1614/*
1615 * Allocate room for and instantiate Stored Measurement Log (SML)
1616 */
1617static void __init prom_instantiate_sml(void)
1618{
1619 phandle ibmvtpm_node;
1620 ihandle ibmvtpm_inst;
1621 u32 entry = 0, size = 0;
1622 u64 base;
1623
1624 prom_debug("prom_instantiate_sml: start...\n");
1625
1626 ibmvtpm_node = call_prom("finddevice", 1, 1, ADDR("/ibm,vtpm"));
1627 prom_debug("ibmvtpm_node: %x\n", ibmvtpm_node);
1628 if (!PHANDLE_VALID(ibmvtpm_node))
1629 return;
1630
1631 ibmvtpm_inst = call_prom("open", 1, 1, ADDR("/ibm,vtpm"));
1632 if (!IHANDLE_VALID(ibmvtpm_inst)) {
1633 prom_printf("opening vtpm package failed (%x)\n", ibmvtpm_inst);
1634 return;
1635 }
1636
1637 if (call_prom_ret("call-method", 2, 2, &size,
1638 ADDR("sml-get-handover-size"),
1639 ibmvtpm_inst) != 0 || size == 0) {
1640 prom_printf("SML get handover size failed\n");
1641 return;
1642 }
1643
1644 base = alloc_down(size, PAGE_SIZE, 0);
1645 if (base == 0)
1646 prom_panic("Could not allocate memory for sml\n");
1647
1648 prom_printf("instantiating sml at 0x%x...", base);
1649
1650 if (call_prom_ret("call-method", 4, 2, &entry,
1651 ADDR("sml-handover"),
1652 ibmvtpm_inst, size, base) != 0 || entry == 0) {
1653 prom_printf("SML handover failed\n");
1654 return;
1655 }
1656 prom_printf(" done\n");
1657
1658 reserve_mem(base, size);
1659
1660 prom_setprop(ibmvtpm_node, "/ibm,vtpm", "linux,sml-base",
1661 &base, sizeof(base));
1662 prom_setprop(ibmvtpm_node, "/ibm,vtpm", "linux,sml-size",
1663 &size, sizeof(size));
1664
1665 prom_debug("sml base = 0x%x\n", base);
1666 prom_debug("sml size = 0x%x\n", (long)size);
1667
1668 prom_debug("prom_instantiate_sml: end...\n");
1669}
1670
1671/*
1672 * Allocate room for and initialize TCE tables
1673 */
1674#ifdef __BIG_ENDIAN__
1675static void __init prom_initialize_tce_table(void)
1676{
1677 phandle node;
1678 ihandle phb_node;
1679 char compatible[64], type[64], model[64];
1680 char *path = prom_scratch;
1681 u64 base, align;
1682 u32 minalign, minsize;
1683 u64 tce_entry, *tce_entryp;
1684 u64 local_alloc_top, local_alloc_bottom;
1685 u64 i;
1686
1687 if (prom_iommu_off)
1688 return;
1689
1690 prom_debug("starting prom_initialize_tce_table\n");
1691
1692 /* Cache current top of allocs so we reserve a single block */
1693 local_alloc_top = alloc_top_high;
1694 local_alloc_bottom = local_alloc_top;
1695
1696 /* Search all nodes looking for PHBs. */
1697 for (node = 0; prom_next_node(&node); ) {
1698 compatible[0] = 0;
1699 type[0] = 0;
1700 model[0] = 0;
1701 prom_getprop(node, "compatible",
1702 compatible, sizeof(compatible));
1703 prom_getprop(node, "device_type", type, sizeof(type));
1704 prom_getprop(node, "model", model, sizeof(model));
1705
1706 if ((type[0] == 0) || (strstr(type, "pci") == NULL))
1707 continue;
1708
1709 /* Keep the old logic intact to avoid regression. */
1710 if (compatible[0] != 0) {
1711 if ((strstr(compatible, "python") == NULL) &&
1712 (strstr(compatible, "Speedwagon") == NULL) &&
1713 (strstr(compatible, "Winnipeg") == NULL))
1714 continue;
1715 } else if (model[0] != 0) {
1716 if ((strstr(model, "ython") == NULL) &&
1717 (strstr(model, "peedwagon") == NULL) &&
1718 (strstr(model, "innipeg") == NULL))
1719 continue;
1720 }
1721
1722 if (prom_getprop(node, "tce-table-minalign", &minalign,
1723 sizeof(minalign)) == PROM_ERROR)
1724 minalign = 0;
1725 if (prom_getprop(node, "tce-table-minsize", &minsize,
1726 sizeof(minsize)) == PROM_ERROR)
1727 minsize = 4UL << 20;
1728
1729 /*
1730 * Even though we read what OF wants, we just set the table
1731 * size to 4 MB. This is enough to map 2GB of PCI DMA space.
1732 * By doing this, we avoid the pitfalls of trying to DMA to
1733 * MMIO space and the DMA alias hole.
1734 *
1735 * On POWER4, firmware sets the TCE region by assuming
1736 * each TCE table is 8MB. Using this memory for anything
1737 * else will impact performance, so we always allocate 8MB.
1738 * Anton
1739 */
1740 if (pvr_version_is(PVR_POWER4) || pvr_version_is(PVR_POWER4p))
1741 minsize = 8UL << 20;
1742 else
1743 minsize = 4UL << 20;
1744
1745 /* Align to the greater of the align or size */
1746 align = max(minalign, minsize);
1747 base = alloc_down(minsize, align, 1);
1748 if (base == 0)
1749 prom_panic("ERROR, cannot find space for TCE table.\n");
1750 if (base < local_alloc_bottom)
1751 local_alloc_bottom = base;
1752
1753 /* It seems OF doesn't null-terminate the path :-( */
1754 memset(path, 0, PROM_SCRATCH_SIZE);
1755 /* Call OF to setup the TCE hardware */
1756 if (call_prom("package-to-path", 3, 1, node,
1757 path, PROM_SCRATCH_SIZE-1) == PROM_ERROR) {
1758 prom_printf("package-to-path failed\n");
1759 }
1760
1761 /* Save away the TCE table attributes for later use. */
1762 prom_setprop(node, path, "linux,tce-base", &base, sizeof(base));
1763 prom_setprop(node, path, "linux,tce-size", &minsize, sizeof(minsize));
1764
1765 prom_debug("TCE table: %s\n", path);
1766 prom_debug("\tnode = 0x%x\n", node);
1767 prom_debug("\tbase = 0x%x\n", base);
1768 prom_debug("\tsize = 0x%x\n", minsize);
1769
1770 /* Initialize the table to have a one-to-one mapping
1771 * over the allocated size.
1772 */
1773 tce_entryp = (u64 *)base;
1774 for (i = 0; i < (minsize >> 3) ;tce_entryp++, i++) {
1775 tce_entry = (i << PAGE_SHIFT);
1776 tce_entry |= 0x3;
1777 *tce_entryp = tce_entry;
1778 }
1779
1780 prom_printf("opening PHB %s", path);
1781 phb_node = call_prom("open", 1, 1, path);
1782 if (phb_node == 0)
1783 prom_printf("... failed\n");
1784 else
1785 prom_printf("... done\n");
1786
1787 call_prom("call-method", 6, 0, ADDR("set-64-bit-addressing"),
1788 phb_node, -1, minsize,
1789 (u32) base, (u32) (base >> 32));
1790 call_prom("close", 1, 0, phb_node);
1791 }
1792
1793 reserve_mem(local_alloc_bottom, local_alloc_top - local_alloc_bottom);
1794
1795 /* These are only really needed if there is a memory limit in
1796 * effect, but we don't know so export them always. */
1797 prom_tce_alloc_start = local_alloc_bottom;
1798 prom_tce_alloc_end = local_alloc_top;
1799
1800 /* Flag the first invalid entry */
1801 prom_debug("ending prom_initialize_tce_table\n");
1802}
1803#endif /* __BIG_ENDIAN__ */
1804#endif /* CONFIG_PPC64 */
1805
1806/*
1807 * With CHRP SMP we need to use the OF to start the other processors.
1808 * We can't wait until smp_boot_cpus (the OF is trashed by then)
1809 * so we have to put the processors into a holding pattern controlled
1810 * by the kernel (not OF) before we destroy the OF.
1811 *
1812 * This uses a chunk of low memory, puts some holding pattern
1813 * code there and sends the other processors off to there until
1814 * smp_boot_cpus tells them to do something. The holding pattern
1815 * checks that address until its cpu # is there, when it is that
1816 * cpu jumps to __secondary_start(). smp_boot_cpus() takes care
1817 * of setting those values.
1818 *
1819 * We also use physical address 0x4 here to tell when a cpu
1820 * is in its holding pattern code.
1821 *
1822 * -- Cort
1823 */
1824/*
1825 * We want to reference the copy of __secondary_hold_* in the
1826 * 0 - 0x100 address range
1827 */
1828#define LOW_ADDR(x) (((unsigned long) &(x)) & 0xff)
1829
1830static void __init prom_hold_cpus(void)
1831{
1832 unsigned long i;
1833 phandle node;
1834 char type[64];
1835 unsigned long *spinloop
1836 = (void *) LOW_ADDR(__secondary_hold_spinloop);
1837 unsigned long *acknowledge
1838 = (void *) LOW_ADDR(__secondary_hold_acknowledge);
1839 unsigned long secondary_hold = LOW_ADDR(__secondary_hold);
1840
1841 /*
1842 * On pseries, if RTAS supports "query-cpu-stopped-state",
1843 * we skip this stage, the CPUs will be started by the
1844 * kernel using RTAS.
1845 */
1846 if ((of_platform == PLATFORM_PSERIES ||
1847 of_platform == PLATFORM_PSERIES_LPAR) &&
1848 rtas_has_query_cpu_stopped) {
1849 prom_printf("prom_hold_cpus: skipped\n");
1850 return;
1851 }
1852
1853 prom_debug("prom_hold_cpus: start...\n");
1854 prom_debug(" 1) spinloop = 0x%x\n", (unsigned long)spinloop);
1855 prom_debug(" 1) *spinloop = 0x%x\n", *spinloop);
1856 prom_debug(" 1) acknowledge = 0x%x\n",
1857 (unsigned long)acknowledge);
1858 prom_debug(" 1) *acknowledge = 0x%x\n", *acknowledge);
1859 prom_debug(" 1) secondary_hold = 0x%x\n", secondary_hold);
1860
1861 /* Set the common spinloop variable, so all of the secondary cpus
1862 * will block when they are awakened from their OF spinloop.
1863 * This must occur for both SMP and non SMP kernels, since OF will
1864 * be trashed when we move the kernel.
1865 */
1866 *spinloop = 0;
1867
1868 /* look for cpus */
1869 for (node = 0; prom_next_node(&node); ) {
1870 unsigned int cpu_no;
1871 __be32 reg;
1872
1873 type[0] = 0;
1874 prom_getprop(node, "device_type", type, sizeof(type));
1875 if (strcmp(type, "cpu") != 0)
1876 continue;
1877
1878 /* Skip non-configured cpus. */
1879 if (prom_getprop(node, "status", type, sizeof(type)) > 0)
1880 if (strcmp(type, "okay") != 0)
1881 continue;
1882
1883 reg = cpu_to_be32(-1); /* make sparse happy */
1884 prom_getprop(node, "reg", ®, sizeof(reg));
1885 cpu_no = be32_to_cpu(reg);
1886
1887 prom_debug("cpu hw idx = %lu\n", cpu_no);
1888
1889 /* Init the acknowledge var which will be reset by
1890 * the secondary cpu when it awakens from its OF
1891 * spinloop.
1892 */
1893 *acknowledge = (unsigned long)-1;
1894
1895 if (cpu_no != prom.cpu) {
1896 /* Primary Thread of non-boot cpu or any thread */
1897 prom_printf("starting cpu hw idx %lu... ", cpu_no);
1898 call_prom("start-cpu", 3, 0, node,
1899 secondary_hold, cpu_no);
1900
1901 for (i = 0; (i < 100000000) &&
1902 (*acknowledge == ((unsigned long)-1)); i++ )
1903 mb();
1904
1905 if (*acknowledge == cpu_no)
1906 prom_printf("done\n");
1907 else
1908 prom_printf("failed: %x\n", *acknowledge);
1909 }
1910#ifdef CONFIG_SMP
1911 else
1912 prom_printf("boot cpu hw idx %lu\n", cpu_no);
1913#endif /* CONFIG_SMP */
1914 }
1915
1916 prom_debug("prom_hold_cpus: end...\n");
1917}
1918
1919
1920static void __init prom_init_client_services(unsigned long pp)
1921{
1922 /* Get a handle to the prom entry point before anything else */
1923 prom_entry = pp;
1924
1925 /* get a handle for the stdout device */
1926 prom.chosen = call_prom("finddevice", 1, 1, ADDR("/chosen"));
1927 if (!PHANDLE_VALID(prom.chosen))
1928 prom_panic("cannot find chosen"); /* msg won't be printed :( */
1929
1930 /* get device tree root */
1931 prom.root = call_prom("finddevice", 1, 1, ADDR("/"));
1932 if (!PHANDLE_VALID(prom.root))
1933 prom_panic("cannot find device tree root"); /* msg won't be printed :( */
1934
1935 prom.mmumap = 0;
1936}
1937
1938#ifdef CONFIG_PPC32
1939/*
1940 * For really old powermacs, we need to map things we claim.
1941 * For that, we need the ihandle of the mmu.
1942 * Also, on the longtrail, we need to work around other bugs.
1943 */
1944static void __init prom_find_mmu(void)
1945{
1946 phandle oprom;
1947 char version[64];
1948
1949 oprom = call_prom("finddevice", 1, 1, ADDR("/openprom"));
1950 if (!PHANDLE_VALID(oprom))
1951 return;
1952 if (prom_getprop(oprom, "model", version, sizeof(version)) <= 0)
1953 return;
1954 version[sizeof(version) - 1] = 0;
1955 /* XXX might need to add other versions here */
1956 if (strcmp(version, "Open Firmware, 1.0.5") == 0)
1957 of_workarounds = OF_WA_CLAIM;
1958 else if (strncmp(version, "FirmWorks,3.", 12) == 0) {
1959 of_workarounds = OF_WA_CLAIM | OF_WA_LONGTRAIL;
1960 call_prom("interpret", 1, 1, "dev /memory 0 to allow-reclaim");
1961 } else
1962 return;
1963 prom.memory = call_prom("open", 1, 1, ADDR("/memory"));
1964 prom_getprop(prom.chosen, "mmu", &prom.mmumap,
1965 sizeof(prom.mmumap));
1966 prom.mmumap = be32_to_cpu(prom.mmumap);
1967 if (!IHANDLE_VALID(prom.memory) || !IHANDLE_VALID(prom.mmumap))
1968 of_workarounds &= ~OF_WA_CLAIM; /* hmmm */
1969}
1970#else
1971#define prom_find_mmu()
1972#endif
1973
1974static void __init prom_init_stdout(void)
1975{
1976 char *path = of_stdout_device;
1977 char type[16];
1978 phandle stdout_node;
1979 __be32 val;
1980
1981 if (prom_getprop(prom.chosen, "stdout", &val, sizeof(val)) <= 0)
1982 prom_panic("cannot find stdout");
1983
1984 prom.stdout = be32_to_cpu(val);
1985
1986 /* Get the full OF pathname of the stdout device */
1987 memset(path, 0, 256);
1988 call_prom("instance-to-path", 3, 1, prom.stdout, path, 255);
1989 prom_printf("OF stdout device is: %s\n", of_stdout_device);
1990 prom_setprop(prom.chosen, "/chosen", "linux,stdout-path",
1991 path, strlen(path) + 1);
1992
1993 /* instance-to-package fails on PA-Semi */
1994 stdout_node = call_prom("instance-to-package", 1, 1, prom.stdout);
1995 if (stdout_node != PROM_ERROR) {
1996 val = cpu_to_be32(stdout_node);
1997 prom_setprop(prom.chosen, "/chosen", "linux,stdout-package",
1998 &val, sizeof(val));
1999
2000 /* If it's a display, note it */
2001 memset(type, 0, sizeof(type));
2002 prom_getprop(stdout_node, "device_type", type, sizeof(type));
2003 if (strcmp(type, "display") == 0)
2004 prom_setprop(stdout_node, path, "linux,boot-display", NULL, 0);
2005 }
2006}
2007
2008static int __init prom_find_machine_type(void)
2009{
2010 char compat[256];
2011 int len, i = 0;
2012#ifdef CONFIG_PPC64
2013 phandle rtas;
2014 int x;
2015#endif
2016
2017 /* Look for a PowerMac or a Cell */
2018 len = prom_getprop(prom.root, "compatible",
2019 compat, sizeof(compat)-1);
2020 if (len > 0) {
2021 compat[len] = 0;
2022 while (i < len) {
2023 char *p = &compat[i];
2024 int sl = strlen(p);
2025 if (sl == 0)
2026 break;
2027 if (strstr(p, "Power Macintosh") ||
2028 strstr(p, "MacRISC"))
2029 return PLATFORM_POWERMAC;
2030#ifdef CONFIG_PPC64
2031 /* We must make sure we don't detect the IBM Cell
2032 * blades as pSeries due to some firmware issues,
2033 * so we do it here.
2034 */
2035 if (strstr(p, "IBM,CBEA") ||
2036 strstr(p, "IBM,CPBW-1.0"))
2037 return PLATFORM_GENERIC;
2038#endif /* CONFIG_PPC64 */
2039 i += sl + 1;
2040 }
2041 }
2042#ifdef CONFIG_PPC64
2043 /* Try to detect OPAL */
2044 if (PHANDLE_VALID(call_prom("finddevice", 1, 1, ADDR("/ibm,opal"))))
2045 return PLATFORM_OPAL;
2046
2047 /* Try to figure out if it's an IBM pSeries or any other
2048 * PAPR compliant platform. We assume it is if :
2049 * - /device_type is "chrp" (please, do NOT use that for future
2050 * non-IBM designs !
2051 * - it has /rtas
2052 */
2053 len = prom_getprop(prom.root, "device_type",
2054 compat, sizeof(compat)-1);
2055 if (len <= 0)
2056 return PLATFORM_GENERIC;
2057 if (strcmp(compat, "chrp"))
2058 return PLATFORM_GENERIC;
2059
2060 /* Default to pSeries. We need to know if we are running LPAR */
2061 rtas = call_prom("finddevice", 1, 1, ADDR("/rtas"));
2062 if (!PHANDLE_VALID(rtas))
2063 return PLATFORM_GENERIC;
2064 x = prom_getproplen(rtas, "ibm,hypertas-functions");
2065 if (x != PROM_ERROR) {
2066 prom_debug("Hypertas detected, assuming LPAR !\n");
2067 return PLATFORM_PSERIES_LPAR;
2068 }
2069 return PLATFORM_PSERIES;
2070#else
2071 return PLATFORM_GENERIC;
2072#endif
2073}
2074
2075static int __init prom_set_color(ihandle ih, int i, int r, int g, int b)
2076{
2077 return call_prom("call-method", 6, 1, ADDR("color!"), ih, i, b, g, r);
2078}
2079
2080/*
2081 * If we have a display that we don't know how to drive,
2082 * we will want to try to execute OF's open method for it
2083 * later. However, OF will probably fall over if we do that
2084 * we've taken over the MMU.
2085 * So we check whether we will need to open the display,
2086 * and if so, open it now.
2087 */
2088static void __init prom_check_displays(void)
2089{
2090 char type[16], *path;
2091 phandle node;
2092 ihandle ih;
2093 int i;
2094
2095 static unsigned char default_colors[] = {
2096 0x00, 0x00, 0x00,
2097 0x00, 0x00, 0xaa,
2098 0x00, 0xaa, 0x00,
2099 0x00, 0xaa, 0xaa,
2100 0xaa, 0x00, 0x00,
2101 0xaa, 0x00, 0xaa,
2102 0xaa, 0xaa, 0x00,
2103 0xaa, 0xaa, 0xaa,
2104 0x55, 0x55, 0x55,
2105 0x55, 0x55, 0xff,
2106 0x55, 0xff, 0x55,
2107 0x55, 0xff, 0xff,
2108 0xff, 0x55, 0x55,
2109 0xff, 0x55, 0xff,
2110 0xff, 0xff, 0x55,
2111 0xff, 0xff, 0xff
2112 };
2113 const unsigned char *clut;
2114
2115 prom_debug("Looking for displays\n");
2116 for (node = 0; prom_next_node(&node); ) {
2117 memset(type, 0, sizeof(type));
2118 prom_getprop(node, "device_type", type, sizeof(type));
2119 if (strcmp(type, "display") != 0)
2120 continue;
2121
2122 /* It seems OF doesn't null-terminate the path :-( */
2123 path = prom_scratch;
2124 memset(path, 0, PROM_SCRATCH_SIZE);
2125
2126 /*
2127 * leave some room at the end of the path for appending extra
2128 * arguments
2129 */
2130 if (call_prom("package-to-path", 3, 1, node, path,
2131 PROM_SCRATCH_SIZE-10) == PROM_ERROR)
2132 continue;
2133 prom_printf("found display : %s, opening... ", path);
2134
2135 ih = call_prom("open", 1, 1, path);
2136 if (ih == 0) {
2137 prom_printf("failed\n");
2138 continue;
2139 }
2140
2141 /* Success */
2142 prom_printf("done\n");
2143 prom_setprop(node, path, "linux,opened", NULL, 0);
2144
2145 /* Setup a usable color table when the appropriate
2146 * method is available. Should update this to set-colors */
2147 clut = default_colors;
2148 for (i = 0; i < 16; i++, clut += 3)
2149 if (prom_set_color(ih, i, clut[0], clut[1],
2150 clut[2]) != 0)
2151 break;
2152
2153#ifdef CONFIG_LOGO_LINUX_CLUT224
2154 clut = PTRRELOC(logo_linux_clut224.clut);
2155 for (i = 0; i < logo_linux_clut224.clutsize; i++, clut += 3)
2156 if (prom_set_color(ih, i + 32, clut[0], clut[1],
2157 clut[2]) != 0)
2158 break;
2159#endif /* CONFIG_LOGO_LINUX_CLUT224 */
2160
2161#ifdef CONFIG_PPC_EARLY_DEBUG_BOOTX
2162 if (prom_getprop(node, "linux,boot-display", NULL, 0) !=
2163 PROM_ERROR) {
2164 u32 width, height, pitch, addr;
2165
2166 prom_printf("Setting btext !\n");
2167 prom_getprop(node, "width", &width, 4);
2168 prom_getprop(node, "height", &height, 4);
2169 prom_getprop(node, "linebytes", &pitch, 4);
2170 prom_getprop(node, "address", &addr, 4);
2171 prom_printf("W=%d H=%d LB=%d addr=0x%x\n",
2172 width, height, pitch, addr);
2173 btext_setup_display(width, height, 8, pitch, addr);
2174 }
2175#endif /* CONFIG_PPC_EARLY_DEBUG_BOOTX */
2176 }
2177}
2178
2179
2180/* Return (relocated) pointer to this much memory: moves initrd if reqd. */
2181static void __init *make_room(unsigned long *mem_start, unsigned long *mem_end,
2182 unsigned long needed, unsigned long align)
2183{
2184 void *ret;
2185
2186 *mem_start = _ALIGN(*mem_start, align);
2187 while ((*mem_start + needed) > *mem_end) {
2188 unsigned long room, chunk;
2189
2190 prom_debug("Chunk exhausted, claiming more at %x...\n",
2191 alloc_bottom);
2192 room = alloc_top - alloc_bottom;
2193 if (room > DEVTREE_CHUNK_SIZE)
2194 room = DEVTREE_CHUNK_SIZE;
2195 if (room < PAGE_SIZE)
2196 prom_panic("No memory for flatten_device_tree "
2197 "(no room)\n");
2198 chunk = alloc_up(room, 0);
2199 if (chunk == 0)
2200 prom_panic("No memory for flatten_device_tree "
2201 "(claim failed)\n");
2202 *mem_end = chunk + room;
2203 }
2204
2205 ret = (void *)*mem_start;
2206 *mem_start += needed;
2207
2208 return ret;
2209}
2210
2211#define dt_push_token(token, mem_start, mem_end) do { \
2212 void *room = make_room(mem_start, mem_end, 4, 4); \
2213 *(__be32 *)room = cpu_to_be32(token); \
2214 } while(0)
2215
2216static unsigned long __init dt_find_string(char *str)
2217{
2218 char *s, *os;
2219
2220 s = os = (char *)dt_string_start;
2221 s += 4;
2222 while (s < (char *)dt_string_end) {
2223 if (strcmp(s, str) == 0)
2224 return s - os;
2225 s += strlen(s) + 1;
2226 }
2227 return 0;
2228}
2229
2230/*
2231 * The Open Firmware 1275 specification states properties must be 31 bytes or
2232 * less, however not all firmwares obey this. Make it 64 bytes to be safe.
2233 */
2234#define MAX_PROPERTY_NAME 64
2235
2236static void __init scan_dt_build_strings(phandle node,
2237 unsigned long *mem_start,
2238 unsigned long *mem_end)
2239{
2240 char *prev_name, *namep, *sstart;
2241 unsigned long soff;
2242 phandle child;
2243
2244 sstart = (char *)dt_string_start;
2245
2246 /* get and store all property names */
2247 prev_name = "";
2248 for (;;) {
2249 /* 64 is max len of name including nul. */
2250 namep = make_room(mem_start, mem_end, MAX_PROPERTY_NAME, 1);
2251 if (call_prom("nextprop", 3, 1, node, prev_name, namep) != 1) {
2252 /* No more nodes: unwind alloc */
2253 *mem_start = (unsigned long)namep;
2254 break;
2255 }
2256
2257 /* skip "name" */
2258 if (strcmp(namep, "name") == 0) {
2259 *mem_start = (unsigned long)namep;
2260 prev_name = "name";
2261 continue;
2262 }
2263 /* get/create string entry */
2264 soff = dt_find_string(namep);
2265 if (soff != 0) {
2266 *mem_start = (unsigned long)namep;
2267 namep = sstart + soff;
2268 } else {
2269 /* Trim off some if we can */
2270 *mem_start = (unsigned long)namep + strlen(namep) + 1;
2271 dt_string_end = *mem_start;
2272 }
2273 prev_name = namep;
2274 }
2275
2276 /* do all our children */
2277 child = call_prom("child", 1, 1, node);
2278 while (child != 0) {
2279 scan_dt_build_strings(child, mem_start, mem_end);
2280 child = call_prom("peer", 1, 1, child);
2281 }
2282}
2283
2284static void __init scan_dt_build_struct(phandle node, unsigned long *mem_start,
2285 unsigned long *mem_end)
2286{
2287 phandle child;
2288 char *namep, *prev_name, *sstart, *p, *ep, *lp, *path;
2289 unsigned long soff;
2290 unsigned char *valp;
2291 static char pname[MAX_PROPERTY_NAME];
2292 int l, room, has_phandle = 0;
2293
2294 dt_push_token(OF_DT_BEGIN_NODE, mem_start, mem_end);
2295
2296 /* get the node's full name */
2297 namep = (char *)*mem_start;
2298 room = *mem_end - *mem_start;
2299 if (room > 255)
2300 room = 255;
2301 l = call_prom("package-to-path", 3, 1, node, namep, room);
2302 if (l >= 0) {
2303 /* Didn't fit? Get more room. */
2304 if (l >= room) {
2305 if (l >= *mem_end - *mem_start)
2306 namep = make_room(mem_start, mem_end, l+1, 1);
2307 call_prom("package-to-path", 3, 1, node, namep, l);
2308 }
2309 namep[l] = '\0';
2310
2311 /* Fixup an Apple bug where they have bogus \0 chars in the
2312 * middle of the path in some properties, and extract
2313 * the unit name (everything after the last '/').
2314 */
2315 for (lp = p = namep, ep = namep + l; p < ep; p++) {
2316 if (*p == '/')
2317 lp = namep;
2318 else if (*p != 0)
2319 *lp++ = *p;
2320 }
2321 *lp = 0;
2322 *mem_start = _ALIGN((unsigned long)lp + 1, 4);
2323 }
2324
2325 /* get it again for debugging */
2326 path = prom_scratch;
2327 memset(path, 0, PROM_SCRATCH_SIZE);
2328 call_prom("package-to-path", 3, 1, node, path, PROM_SCRATCH_SIZE-1);
2329
2330 /* get and store all properties */
2331 prev_name = "";
2332 sstart = (char *)dt_string_start;
2333 for (;;) {
2334 if (call_prom("nextprop", 3, 1, node, prev_name,
2335 pname) != 1)
2336 break;
2337
2338 /* skip "name" */
2339 if (strcmp(pname, "name") == 0) {
2340 prev_name = "name";
2341 continue;
2342 }
2343
2344 /* find string offset */
2345 soff = dt_find_string(pname);
2346 if (soff == 0) {
2347 prom_printf("WARNING: Can't find string index for"
2348 " <%s>, node %s\n", pname, path);
2349 break;
2350 }
2351 prev_name = sstart + soff;
2352
2353 /* get length */
2354 l = call_prom("getproplen", 2, 1, node, pname);
2355
2356 /* sanity checks */
2357 if (l == PROM_ERROR)
2358 continue;
2359
2360 /* push property head */
2361 dt_push_token(OF_DT_PROP, mem_start, mem_end);
2362 dt_push_token(l, mem_start, mem_end);
2363 dt_push_token(soff, mem_start, mem_end);
2364
2365 /* push property content */
2366 valp = make_room(mem_start, mem_end, l, 4);
2367 call_prom("getprop", 4, 1, node, pname, valp, l);
2368 *mem_start = _ALIGN(*mem_start, 4);
2369
2370 if (!strcmp(pname, "phandle"))
2371 has_phandle = 1;
2372 }
2373
2374 /* Add a "linux,phandle" property if no "phandle" property already
2375 * existed (can happen with OPAL)
2376 */
2377 if (!has_phandle) {
2378 soff = dt_find_string("linux,phandle");
2379 if (soff == 0)
2380 prom_printf("WARNING: Can't find string index for"
2381 " <linux-phandle> node %s\n", path);
2382 else {
2383 dt_push_token(OF_DT_PROP, mem_start, mem_end);
2384 dt_push_token(4, mem_start, mem_end);
2385 dt_push_token(soff, mem_start, mem_end);
2386 valp = make_room(mem_start, mem_end, 4, 4);
2387 *(__be32 *)valp = cpu_to_be32(node);
2388 }
2389 }
2390
2391 /* do all our children */
2392 child = call_prom("child", 1, 1, node);
2393 while (child != 0) {
2394 scan_dt_build_struct(child, mem_start, mem_end);
2395 child = call_prom("peer", 1, 1, child);
2396 }
2397
2398 dt_push_token(OF_DT_END_NODE, mem_start, mem_end);
2399}
2400
2401static void __init flatten_device_tree(void)
2402{
2403 phandle root;
2404 unsigned long mem_start, mem_end, room;
2405 struct boot_param_header *hdr;
2406 char *namep;
2407 u64 *rsvmap;
2408
2409 /*
2410 * Check how much room we have between alloc top & bottom (+/- a
2411 * few pages), crop to 1MB, as this is our "chunk" size
2412 */
2413 room = alloc_top - alloc_bottom - 0x4000;
2414 if (room > DEVTREE_CHUNK_SIZE)
2415 room = DEVTREE_CHUNK_SIZE;
2416 prom_debug("starting device tree allocs at %x\n", alloc_bottom);
2417
2418 /* Now try to claim that */
2419 mem_start = (unsigned long)alloc_up(room, PAGE_SIZE);
2420 if (mem_start == 0)
2421 prom_panic("Can't allocate initial device-tree chunk\n");
2422 mem_end = mem_start + room;
2423
2424 /* Get root of tree */
2425 root = call_prom("peer", 1, 1, (phandle)0);
2426 if (root == (phandle)0)
2427 prom_panic ("couldn't get device tree root\n");
2428
2429 /* Build header and make room for mem rsv map */
2430 mem_start = _ALIGN(mem_start, 4);
2431 hdr = make_room(&mem_start, &mem_end,
2432 sizeof(struct boot_param_header), 4);
2433 dt_header_start = (unsigned long)hdr;
2434 rsvmap = make_room(&mem_start, &mem_end, sizeof(mem_reserve_map), 8);
2435
2436 /* Start of strings */
2437 mem_start = PAGE_ALIGN(mem_start);
2438 dt_string_start = mem_start;
2439 mem_start += 4; /* hole */
2440
2441 /* Add "linux,phandle" in there, we'll need it */
2442 namep = make_room(&mem_start, &mem_end, 16, 1);
2443 strcpy(namep, "linux,phandle");
2444 mem_start = (unsigned long)namep + strlen(namep) + 1;
2445
2446 /* Build string array */
2447 prom_printf("Building dt strings...\n");
2448 scan_dt_build_strings(root, &mem_start, &mem_end);
2449 dt_string_end = mem_start;
2450
2451 /* Build structure */
2452 mem_start = PAGE_ALIGN(mem_start);
2453 dt_struct_start = mem_start;
2454 prom_printf("Building dt structure...\n");
2455 scan_dt_build_struct(root, &mem_start, &mem_end);
2456 dt_push_token(OF_DT_END, &mem_start, &mem_end);
2457 dt_struct_end = PAGE_ALIGN(mem_start);
2458
2459 /* Finish header */
2460 hdr->boot_cpuid_phys = cpu_to_be32(prom.cpu);
2461 hdr->magic = cpu_to_be32(OF_DT_HEADER);
2462 hdr->totalsize = cpu_to_be32(dt_struct_end - dt_header_start);
2463 hdr->off_dt_struct = cpu_to_be32(dt_struct_start - dt_header_start);
2464 hdr->off_dt_strings = cpu_to_be32(dt_string_start - dt_header_start);
2465 hdr->dt_strings_size = cpu_to_be32(dt_string_end - dt_string_start);
2466 hdr->off_mem_rsvmap = cpu_to_be32(((unsigned long)rsvmap) - dt_header_start);
2467 hdr->version = cpu_to_be32(OF_DT_VERSION);
2468 /* Version 16 is not backward compatible */
2469 hdr->last_comp_version = cpu_to_be32(0x10);
2470
2471 /* Copy the reserve map in */
2472 memcpy(rsvmap, mem_reserve_map, sizeof(mem_reserve_map));
2473
2474#ifdef DEBUG_PROM
2475 {
2476 int i;
2477 prom_printf("reserved memory map:\n");
2478 for (i = 0; i < mem_reserve_cnt; i++)
2479 prom_printf(" %x - %x\n",
2480 be64_to_cpu(mem_reserve_map[i].base),
2481 be64_to_cpu(mem_reserve_map[i].size));
2482 }
2483#endif
2484 /* Bump mem_reserve_cnt to cause further reservations to fail
2485 * since it's too late.
2486 */
2487 mem_reserve_cnt = MEM_RESERVE_MAP_SIZE;
2488
2489 prom_printf("Device tree strings 0x%x -> 0x%x\n",
2490 dt_string_start, dt_string_end);
2491 prom_printf("Device tree struct 0x%x -> 0x%x\n",
2492 dt_struct_start, dt_struct_end);
2493}
2494
2495#ifdef CONFIG_PPC_MAPLE
2496/* PIBS Version 1.05.0000 04/26/2005 has an incorrect /ht/isa/ranges property.
2497 * The values are bad, and it doesn't even have the right number of cells. */
2498static void __init fixup_device_tree_maple(void)
2499{
2500 phandle isa;
2501 u32 rloc = 0x01002000; /* IO space; PCI device = 4 */
2502 u32 isa_ranges[6];
2503 char *name;
2504
2505 name = "/ht@0/isa@4";
2506 isa = call_prom("finddevice", 1, 1, ADDR(name));
2507 if (!PHANDLE_VALID(isa)) {
2508 name = "/ht@0/isa@6";
2509 isa = call_prom("finddevice", 1, 1, ADDR(name));
2510 rloc = 0x01003000; /* IO space; PCI device = 6 */
2511 }
2512 if (!PHANDLE_VALID(isa))
2513 return;
2514
2515 if (prom_getproplen(isa, "ranges") != 12)
2516 return;
2517 if (prom_getprop(isa, "ranges", isa_ranges, sizeof(isa_ranges))
2518 == PROM_ERROR)
2519 return;
2520
2521 if (isa_ranges[0] != 0x1 ||
2522 isa_ranges[1] != 0xf4000000 ||
2523 isa_ranges[2] != 0x00010000)
2524 return;
2525
2526 prom_printf("Fixing up bogus ISA range on Maple/Apache...\n");
2527
2528 isa_ranges[0] = 0x1;
2529 isa_ranges[1] = 0x0;
2530 isa_ranges[2] = rloc;
2531 isa_ranges[3] = 0x0;
2532 isa_ranges[4] = 0x0;
2533 isa_ranges[5] = 0x00010000;
2534 prom_setprop(isa, name, "ranges",
2535 isa_ranges, sizeof(isa_ranges));
2536}
2537
2538#define CPC925_MC_START 0xf8000000
2539#define CPC925_MC_LENGTH 0x1000000
2540/* The values for memory-controller don't have right number of cells */
2541static void __init fixup_device_tree_maple_memory_controller(void)
2542{
2543 phandle mc;
2544 u32 mc_reg[4];
2545 char *name = "/hostbridge@f8000000";
2546 u32 ac, sc;
2547
2548 mc = call_prom("finddevice", 1, 1, ADDR(name));
2549 if (!PHANDLE_VALID(mc))
2550 return;
2551
2552 if (prom_getproplen(mc, "reg") != 8)
2553 return;
2554
2555 prom_getprop(prom.root, "#address-cells", &ac, sizeof(ac));
2556 prom_getprop(prom.root, "#size-cells", &sc, sizeof(sc));
2557 if ((ac != 2) || (sc != 2))
2558 return;
2559
2560 if (prom_getprop(mc, "reg", mc_reg, sizeof(mc_reg)) == PROM_ERROR)
2561 return;
2562
2563 if (mc_reg[0] != CPC925_MC_START || mc_reg[1] != CPC925_MC_LENGTH)
2564 return;
2565
2566 prom_printf("Fixing up bogus hostbridge on Maple...\n");
2567
2568 mc_reg[0] = 0x0;
2569 mc_reg[1] = CPC925_MC_START;
2570 mc_reg[2] = 0x0;
2571 mc_reg[3] = CPC925_MC_LENGTH;
2572 prom_setprop(mc, name, "reg", mc_reg, sizeof(mc_reg));
2573}
2574#else
2575#define fixup_device_tree_maple()
2576#define fixup_device_tree_maple_memory_controller()
2577#endif
2578
2579#ifdef CONFIG_PPC_CHRP
2580/*
2581 * Pegasos and BriQ lacks the "ranges" property in the isa node
2582 * Pegasos needs decimal IRQ 14/15, not hexadecimal
2583 * Pegasos has the IDE configured in legacy mode, but advertised as native
2584 */
2585static void __init fixup_device_tree_chrp(void)
2586{
2587 phandle ph;
2588 u32 prop[6];
2589 u32 rloc = 0x01006000; /* IO space; PCI device = 12 */
2590 char *name;
2591 int rc;
2592
2593 name = "/pci@80000000/isa@c";
2594 ph = call_prom("finddevice", 1, 1, ADDR(name));
2595 if (!PHANDLE_VALID(ph)) {
2596 name = "/pci@ff500000/isa@6";
2597 ph = call_prom("finddevice", 1, 1, ADDR(name));
2598 rloc = 0x01003000; /* IO space; PCI device = 6 */
2599 }
2600 if (PHANDLE_VALID(ph)) {
2601 rc = prom_getproplen(ph, "ranges");
2602 if (rc == 0 || rc == PROM_ERROR) {
2603 prom_printf("Fixing up missing ISA range on Pegasos...\n");
2604
2605 prop[0] = 0x1;
2606 prop[1] = 0x0;
2607 prop[2] = rloc;
2608 prop[3] = 0x0;
2609 prop[4] = 0x0;
2610 prop[5] = 0x00010000;
2611 prom_setprop(ph, name, "ranges", prop, sizeof(prop));
2612 }
2613 }
2614
2615 name = "/pci@80000000/ide@C,1";
2616 ph = call_prom("finddevice", 1, 1, ADDR(name));
2617 if (PHANDLE_VALID(ph)) {
2618 prom_printf("Fixing up IDE interrupt on Pegasos...\n");
2619 prop[0] = 14;
2620 prop[1] = 0x0;
2621 prom_setprop(ph, name, "interrupts", prop, 2*sizeof(u32));
2622 prom_printf("Fixing up IDE class-code on Pegasos...\n");
2623 rc = prom_getprop(ph, "class-code", prop, sizeof(u32));
2624 if (rc == sizeof(u32)) {
2625 prop[0] &= ~0x5;
2626 prom_setprop(ph, name, "class-code", prop, sizeof(u32));
2627 }
2628 }
2629}
2630#else
2631#define fixup_device_tree_chrp()
2632#endif
2633
2634#if defined(CONFIG_PPC64) && defined(CONFIG_PPC_PMAC)
2635static void __init fixup_device_tree_pmac(void)
2636{
2637 phandle u3, i2c, mpic;
2638 u32 u3_rev;
2639 u32 interrupts[2];
2640 u32 parent;
2641
2642 /* Some G5s have a missing interrupt definition, fix it up here */
2643 u3 = call_prom("finddevice", 1, 1, ADDR("/u3@0,f8000000"));
2644 if (!PHANDLE_VALID(u3))
2645 return;
2646 i2c = call_prom("finddevice", 1, 1, ADDR("/u3@0,f8000000/i2c@f8001000"));
2647 if (!PHANDLE_VALID(i2c))
2648 return;
2649 mpic = call_prom("finddevice", 1, 1, ADDR("/u3@0,f8000000/mpic@f8040000"));
2650 if (!PHANDLE_VALID(mpic))
2651 return;
2652
2653 /* check if proper rev of u3 */
2654 if (prom_getprop(u3, "device-rev", &u3_rev, sizeof(u3_rev))
2655 == PROM_ERROR)
2656 return;
2657 if (u3_rev < 0x35 || u3_rev > 0x39)
2658 return;
2659 /* does it need fixup ? */
2660 if (prom_getproplen(i2c, "interrupts") > 0)
2661 return;
2662
2663 prom_printf("fixing up bogus interrupts for u3 i2c...\n");
2664
2665 /* interrupt on this revision of u3 is number 0 and level */
2666 interrupts[0] = 0;
2667 interrupts[1] = 1;
2668 prom_setprop(i2c, "/u3@0,f8000000/i2c@f8001000", "interrupts",
2669 &interrupts, sizeof(interrupts));
2670 parent = (u32)mpic;
2671 prom_setprop(i2c, "/u3@0,f8000000/i2c@f8001000", "interrupt-parent",
2672 &parent, sizeof(parent));
2673}
2674#else
2675#define fixup_device_tree_pmac()
2676#endif
2677
2678#ifdef CONFIG_PPC_EFIKA
2679/*
2680 * The MPC5200 FEC driver requires an phy-handle property to tell it how
2681 * to talk to the phy. If the phy-handle property is missing, then this
2682 * function is called to add the appropriate nodes and link it to the
2683 * ethernet node.
2684 */
2685static void __init fixup_device_tree_efika_add_phy(void)
2686{
2687 u32 node;
2688 char prop[64];
2689 int rv;
2690
2691 /* Check if /builtin/ethernet exists - bail if it doesn't */
2692 node = call_prom("finddevice", 1, 1, ADDR("/builtin/ethernet"));
2693 if (!PHANDLE_VALID(node))
2694 return;
2695
2696 /* Check if the phy-handle property exists - bail if it does */
2697 rv = prom_getprop(node, "phy-handle", prop, sizeof(prop));
2698 if (!rv)
2699 return;
2700
2701 /*
2702 * At this point the ethernet device doesn't have a phy described.
2703 * Now we need to add the missing phy node and linkage
2704 */
2705
2706 /* Check for an MDIO bus node - if missing then create one */
2707 node = call_prom("finddevice", 1, 1, ADDR("/builtin/mdio"));
2708 if (!PHANDLE_VALID(node)) {
2709 prom_printf("Adding Ethernet MDIO node\n");
2710 call_prom("interpret", 1, 1,
2711 " s\" /builtin\" find-device"
2712 " new-device"
2713 " 1 encode-int s\" #address-cells\" property"
2714 " 0 encode-int s\" #size-cells\" property"
2715 " s\" mdio\" device-name"
2716 " s\" fsl,mpc5200b-mdio\" encode-string"
2717 " s\" compatible\" property"
2718 " 0xf0003000 0x400 reg"
2719 " 0x2 encode-int"
2720 " 0x5 encode-int encode+"
2721 " 0x3 encode-int encode+"
2722 " s\" interrupts\" property"
2723 " finish-device");
2724 };
2725
2726 /* Check for a PHY device node - if missing then create one and
2727 * give it's phandle to the ethernet node */
2728 node = call_prom("finddevice", 1, 1,
2729 ADDR("/builtin/mdio/ethernet-phy"));
2730 if (!PHANDLE_VALID(node)) {
2731 prom_printf("Adding Ethernet PHY node\n");
2732 call_prom("interpret", 1, 1,
2733 " s\" /builtin/mdio\" find-device"
2734 " new-device"
2735 " s\" ethernet-phy\" device-name"
2736 " 0x10 encode-int s\" reg\" property"
2737 " my-self"
2738 " ihandle>phandle"
2739 " finish-device"
2740 " s\" /builtin/ethernet\" find-device"
2741 " encode-int"
2742 " s\" phy-handle\" property"
2743 " device-end");
2744 }
2745}
2746
2747static void __init fixup_device_tree_efika(void)
2748{
2749 int sound_irq[3] = { 2, 2, 0 };
2750 int bcomm_irq[3*16] = { 3,0,0, 3,1,0, 3,2,0, 3,3,0,
2751 3,4,0, 3,5,0, 3,6,0, 3,7,0,
2752 3,8,0, 3,9,0, 3,10,0, 3,11,0,
2753 3,12,0, 3,13,0, 3,14,0, 3,15,0 };
2754 u32 node;
2755 char prop[64];
2756 int rv, len;
2757
2758 /* Check if we're really running on a EFIKA */
2759 node = call_prom("finddevice", 1, 1, ADDR("/"));
2760 if (!PHANDLE_VALID(node))
2761 return;
2762
2763 rv = prom_getprop(node, "model", prop, sizeof(prop));
2764 if (rv == PROM_ERROR)
2765 return;
2766 if (strcmp(prop, "EFIKA5K2"))
2767 return;
2768
2769 prom_printf("Applying EFIKA device tree fixups\n");
2770
2771 /* Claiming to be 'chrp' is death */
2772 node = call_prom("finddevice", 1, 1, ADDR("/"));
2773 rv = prom_getprop(node, "device_type", prop, sizeof(prop));
2774 if (rv != PROM_ERROR && (strcmp(prop, "chrp") == 0))
2775 prom_setprop(node, "/", "device_type", "efika", sizeof("efika"));
2776
2777 /* CODEGEN,description is exposed in /proc/cpuinfo so
2778 fix that too */
2779 rv = prom_getprop(node, "CODEGEN,description", prop, sizeof(prop));
2780 if (rv != PROM_ERROR && (strstr(prop, "CHRP")))
2781 prom_setprop(node, "/", "CODEGEN,description",
2782 "Efika 5200B PowerPC System",
2783 sizeof("Efika 5200B PowerPC System"));
2784
2785 /* Fixup bestcomm interrupts property */
2786 node = call_prom("finddevice", 1, 1, ADDR("/builtin/bestcomm"));
2787 if (PHANDLE_VALID(node)) {
2788 len = prom_getproplen(node, "interrupts");
2789 if (len == 12) {
2790 prom_printf("Fixing bestcomm interrupts property\n");
2791 prom_setprop(node, "/builtin/bestcom", "interrupts",
2792 bcomm_irq, sizeof(bcomm_irq));
2793 }
2794 }
2795
2796 /* Fixup sound interrupts property */
2797 node = call_prom("finddevice", 1, 1, ADDR("/builtin/sound"));
2798 if (PHANDLE_VALID(node)) {
2799 rv = prom_getprop(node, "interrupts", prop, sizeof(prop));
2800 if (rv == PROM_ERROR) {
2801 prom_printf("Adding sound interrupts property\n");
2802 prom_setprop(node, "/builtin/sound", "interrupts",
2803 sound_irq, sizeof(sound_irq));
2804 }
2805 }
2806
2807 /* Make sure ethernet phy-handle property exists */
2808 fixup_device_tree_efika_add_phy();
2809}
2810#else
2811#define fixup_device_tree_efika()
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}
2822
2823static void __init prom_find_boot_cpu(void)
2824{
2825 __be32 rval;
2826 ihandle prom_cpu;
2827 phandle cpu_pkg;
2828
2829 rval = 0;
2830 if (prom_getprop(prom.chosen, "cpu", &rval, sizeof(rval)) <= 0)
2831 return;
2832 prom_cpu = be32_to_cpu(rval);
2833
2834 cpu_pkg = call_prom("instance-to-package", 1, 1, prom_cpu);
2835
2836 prom_getprop(cpu_pkg, "reg", &rval, sizeof(rval));
2837 prom.cpu = be32_to_cpu(rval);
2838
2839 prom_debug("Booting CPU hw index = %lu\n", prom.cpu);
2840}
2841
2842static void __init prom_check_initrd(unsigned long r3, unsigned long r4)
2843{
2844#ifdef CONFIG_BLK_DEV_INITRD
2845 if (r3 && r4 && r4 != 0xdeadbeef) {
2846 __be64 val;
2847
2848 prom_initrd_start = is_kernel_addr(r3) ? __pa(r3) : r3;
2849 prom_initrd_end = prom_initrd_start + r4;
2850
2851 val = cpu_to_be64(prom_initrd_start);
2852 prom_setprop(prom.chosen, "/chosen", "linux,initrd-start",
2853 &val, sizeof(val));
2854 val = cpu_to_be64(prom_initrd_end);
2855 prom_setprop(prom.chosen, "/chosen", "linux,initrd-end",
2856 &val, sizeof(val));
2857
2858 reserve_mem(prom_initrd_start,
2859 prom_initrd_end - prom_initrd_start);
2860
2861 prom_debug("initrd_start=0x%x\n", prom_initrd_start);
2862 prom_debug("initrd_end=0x%x\n", prom_initrd_end);
2863 }
2864#endif /* CONFIG_BLK_DEV_INITRD */
2865}
2866
2867#ifdef CONFIG_PPC64
2868#ifdef CONFIG_RELOCATABLE
2869static void reloc_toc(void)
2870{
2871}
2872
2873static void unreloc_toc(void)
2874{
2875}
2876#else
2877static void __reloc_toc(unsigned long offset, unsigned long nr_entries)
2878{
2879 unsigned long i;
2880 unsigned long *toc_entry;
2881
2882 /* Get the start of the TOC by using r2 directly. */
2883 asm volatile("addi %0,2,-0x8000" : "=b" (toc_entry));
2884
2885 for (i = 0; i < nr_entries; i++) {
2886 *toc_entry = *toc_entry + offset;
2887 toc_entry++;
2888 }
2889}
2890
2891static void reloc_toc(void)
2892{
2893 unsigned long offset = reloc_offset();
2894 unsigned long nr_entries =
2895 (__prom_init_toc_end - __prom_init_toc_start) / sizeof(long);
2896
2897 __reloc_toc(offset, nr_entries);
2898
2899 mb();
2900}
2901
2902static void unreloc_toc(void)
2903{
2904 unsigned long offset = reloc_offset();
2905 unsigned long nr_entries =
2906 (__prom_init_toc_end - __prom_init_toc_start) / sizeof(long);
2907
2908 mb();
2909
2910 __reloc_toc(-offset, nr_entries);
2911}
2912#endif
2913#endif
2914
2915/*
2916 * We enter here early on, when the Open Firmware prom is still
2917 * handling exceptions and the MMU hash table for us.
2918 */
2919
2920unsigned long __init prom_init(unsigned long r3, unsigned long r4,
2921 unsigned long pp,
2922 unsigned long r6, unsigned long r7,
2923 unsigned long kbase)
2924{
2925 unsigned long hdr;
2926
2927#ifdef CONFIG_PPC32
2928 unsigned long offset = reloc_offset();
2929 reloc_got2(offset);
2930#else
2931 reloc_toc();
2932#endif
2933
2934 /*
2935 * First zero the BSS
2936 */
2937 memset(&__bss_start, 0, __bss_stop - __bss_start);
2938
2939 /*
2940 * Init interface to Open Firmware, get some node references,
2941 * like /chosen
2942 */
2943 prom_init_client_services(pp);
2944
2945 /*
2946 * See if this OF is old enough that we need to do explicit maps
2947 * and other workarounds
2948 */
2949 prom_find_mmu();
2950
2951 /*
2952 * Init prom stdout device
2953 */
2954 prom_init_stdout();
2955
2956 prom_printf("Preparing to boot %s", linux_banner);
2957
2958 /*
2959 * Get default machine type. At this point, we do not differentiate
2960 * between pSeries SMP and pSeries LPAR
2961 */
2962 of_platform = prom_find_machine_type();
2963 prom_printf("Detected machine type: %x\n", of_platform);
2964
2965#ifndef CONFIG_NONSTATIC_KERNEL
2966 /* Bail if this is a kdump kernel. */
2967 if (PHYSICAL_START > 0)
2968 prom_panic("Error: You can't boot a kdump kernel from OF!\n");
2969#endif
2970
2971 /*
2972 * Check for an initrd
2973 */
2974 prom_check_initrd(r3, r4);
2975
2976#if defined(CONFIG_PPC_PSERIES) || defined(CONFIG_PPC_POWERNV)
2977 /*
2978 * On pSeries, inform the firmware about our capabilities
2979 */
2980 if (of_platform == PLATFORM_PSERIES ||
2981 of_platform == PLATFORM_PSERIES_LPAR)
2982 prom_send_capabilities();
2983#endif
2984
2985 /*
2986 * Copy the CPU hold code
2987 */
2988 if (of_platform != PLATFORM_POWERMAC)
2989 copy_and_flush(0, kbase, 0x100, 0);
2990
2991 /*
2992 * Do early parsing of command line
2993 */
2994 early_cmdline_parse();
2995
2996 /*
2997 * Initialize memory management within prom_init
2998 */
2999 prom_init_mem();
3000
3001 /*
3002 * Determine which cpu is actually running right _now_
3003 */
3004 prom_find_boot_cpu();
3005
3006 /*
3007 * Initialize display devices
3008 */
3009 prom_check_displays();
3010
3011#if defined(CONFIG_PPC64) && defined(__BIG_ENDIAN__)
3012 /*
3013 * Initialize IOMMU (TCE tables) on pSeries. Do that before anything else
3014 * that uses the allocator, we need to make sure we get the top of memory
3015 * available for us here...
3016 */
3017 if (of_platform == PLATFORM_PSERIES)
3018 prom_initialize_tce_table();
3019#endif
3020
3021 /*
3022 * On non-powermacs, try to instantiate RTAS. PowerMacs don't
3023 * have a usable RTAS implementation.
3024 */
3025 if (of_platform != PLATFORM_POWERMAC &&
3026 of_platform != PLATFORM_OPAL)
3027 prom_instantiate_rtas();
3028
3029#ifdef CONFIG_PPC_POWERNV
3030#ifdef __BIG_ENDIAN__
3031 /* Detect HAL and try instanciating it & doing takeover */
3032 if (of_platform == PLATFORM_PSERIES_LPAR) {
3033 prom_query_opal();
3034 if (of_platform == PLATFORM_OPAL) {
3035 prom_opal_hold_cpus();
3036 prom_opal_takeover();
3037 }
3038 } else
3039#endif /* __BIG_ENDIAN__ */
3040 if (of_platform == PLATFORM_OPAL)
3041 prom_instantiate_opal();
3042#endif /* CONFIG_PPC_POWERNV */
3043
3044#ifdef CONFIG_PPC64
3045 /* instantiate sml */
3046 prom_instantiate_sml();
3047#endif
3048
3049 /*
3050 * On non-powermacs, put all CPUs in spin-loops.
3051 *
3052 * PowerMacs use a different mechanism to spin CPUs
3053 *
3054 * (This must be done after instanciating RTAS)
3055 */
3056 if (of_platform != PLATFORM_POWERMAC &&
3057 of_platform != PLATFORM_OPAL)
3058 prom_hold_cpus();
3059
3060 /*
3061 * Fill in some infos for use by the kernel later on
3062 */
3063 if (prom_memory_limit) {
3064 __be64 val = cpu_to_be64(prom_memory_limit);
3065 prom_setprop(prom.chosen, "/chosen", "linux,memory-limit",
3066 &val, sizeof(val));
3067 }
3068#ifdef CONFIG_PPC64
3069 if (prom_iommu_off)
3070 prom_setprop(prom.chosen, "/chosen", "linux,iommu-off",
3071 NULL, 0);
3072
3073 if (prom_iommu_force_on)
3074 prom_setprop(prom.chosen, "/chosen", "linux,iommu-force-on",
3075 NULL, 0);
3076
3077 if (prom_tce_alloc_start) {
3078 prom_setprop(prom.chosen, "/chosen", "linux,tce-alloc-start",
3079 &prom_tce_alloc_start,
3080 sizeof(prom_tce_alloc_start));
3081 prom_setprop(prom.chosen, "/chosen", "linux,tce-alloc-end",
3082 &prom_tce_alloc_end,
3083 sizeof(prom_tce_alloc_end));
3084 }
3085#endif
3086
3087 /*
3088 * Fixup any known bugs in the device-tree
3089 */
3090 fixup_device_tree();
3091
3092 /*
3093 * Now finally create the flattened device-tree
3094 */
3095 prom_printf("copying OF device tree...\n");
3096 flatten_device_tree();
3097
3098 /*
3099 * in case stdin is USB and still active on IBM machines...
3100 * Unfortunately quiesce crashes on some powermacs if we have
3101 * closed stdin already (in particular the powerbook 101). It
3102 * appears that the OPAL version of OFW doesn't like it either.
3103 */
3104 if (of_platform != PLATFORM_POWERMAC &&
3105 of_platform != PLATFORM_OPAL)
3106 prom_close_stdin();
3107
3108 /*
3109 * Call OF "quiesce" method to shut down pending DMA's from
3110 * devices etc...
3111 */
3112 prom_printf("Calling quiesce...\n");
3113 call_prom("quiesce", 0, 0);
3114
3115 /*
3116 * And finally, call the kernel passing it the flattened device
3117 * tree and NULL as r5, thus triggering the new entry point which
3118 * is common to us and kexec
3119 */
3120 hdr = dt_header_start;
3121
3122 /* Don't print anything after quiesce under OPAL, it crashes OFW */
3123 if (of_platform != PLATFORM_OPAL) {
3124 prom_printf("returning from prom_init\n");
3125 prom_debug("->dt_header_start=0x%x\n", hdr);
3126 }
3127
3128#ifdef CONFIG_PPC32
3129 reloc_got2(-offset);
3130#else
3131 unreloc_toc();
3132#endif
3133
3134#ifdef CONFIG_PPC_EARLY_DEBUG_OPAL
3135 /* OPAL early debug gets the OPAL base & entry in r8 and r9 */
3136 __start(hdr, kbase, 0, 0, 0,
3137 prom_opal_base, prom_opal_entry);
3138#else
3139 __start(hdr, kbase, 0, 0, 0, 0, 0);
3140#endif
3141
3142 return 0;
3143}