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 <linux/printk.h>
31#include <asm/prom.h>
32#include <asm/rtas.h>
33#include <asm/page.h>
34#include <asm/processor.h>
35#include <asm/interrupt.h>
36#include <asm/irq.h>
37#include <asm/io.h>
38#include <asm/smp.h>
39#include <asm/mmu.h>
40#include <asm/iommu.h>
41#include <asm/btext.h>
42#include <asm/sections.h>
43#include <asm/machdep.h>
44#include <asm/asm-prototypes.h>
45#include <asm/ultravisor-api.h>
46
47#include <linux/linux_logo.h>
48
49/* All of prom_init bss lives here */
50#define __prombss __section(".bss.prominit")
51
52/*
53 * Eventually bump that one up
54 */
55#define DEVTREE_CHUNK_SIZE 0x100000
56
57/*
58 * This is the size of the local memory reserve map that gets copied
59 * into the boot params passed to the kernel. That size is totally
60 * flexible as the kernel just reads the list until it encounters an
61 * entry with size 0, so it can be changed without breaking binary
62 * compatibility
63 */
64#define MEM_RESERVE_MAP_SIZE 8
65
66/*
67 * prom_init() is called very early on, before the kernel text
68 * and data have been mapped to KERNELBASE. At this point the code
69 * is running at whatever address it has been loaded at.
70 * On ppc32 we compile with -mrelocatable, which means that references
71 * to extern and static variables get relocated automatically.
72 * ppc64 objects are always relocatable, we just need to relocate the
73 * TOC.
74 *
75 * Because OF may have mapped I/O devices into the area starting at
76 * KERNELBASE, particularly on CHRP machines, we can't safely call
77 * OF once the kernel has been mapped to KERNELBASE. Therefore all
78 * OF calls must be done within prom_init().
79 *
80 * ADDR is used in calls to call_prom. The 4th and following
81 * arguments to call_prom should be 32-bit values.
82 * On ppc64, 64 bit values are truncated to 32 bits (and
83 * fortunately don't get interpreted as two arguments).
84 */
85#define ADDR(x) (u32)(unsigned long)(x)
86
87#ifdef CONFIG_PPC64
88#define OF_WORKAROUNDS 0
89#else
90#define OF_WORKAROUNDS of_workarounds
91static int of_workarounds __prombss;
92#endif
93
94#define OF_WA_CLAIM 1 /* do phys/virt claim separately, then map */
95#define OF_WA_LONGTRAIL 2 /* work around longtrail bugs */
96
97#define PROM_BUG() do { \
98 prom_printf("kernel BUG at %s line 0x%x!\n", \
99 __FILE__, __LINE__); \
100 __builtin_trap(); \
101} while (0)
102
103#ifdef DEBUG_PROM
104#define prom_debug(x...) prom_printf(x)
105#else
106#define prom_debug(x...) do { } while (0)
107#endif
108
109
110typedef u32 prom_arg_t;
111
112struct prom_args {
113 __be32 service;
114 __be32 nargs;
115 __be32 nret;
116 __be32 args[10];
117};
118
119struct prom_t {
120 ihandle root;
121 phandle chosen;
122 int cpu;
123 ihandle stdout;
124 ihandle mmumap;
125 ihandle memory;
126};
127
128struct mem_map_entry {
129 __be64 base;
130 __be64 size;
131};
132
133typedef __be32 cell_t;
134
135extern void __start(unsigned long r3, unsigned long r4, unsigned long r5,
136 unsigned long r6, unsigned long r7, unsigned long r8,
137 unsigned long r9);
138
139#ifdef CONFIG_PPC64
140extern int enter_prom(struct prom_args *args, unsigned long entry);
141#else
142static inline int enter_prom(struct prom_args *args, unsigned long entry)
143{
144 return ((int (*)(struct prom_args *))entry)(args);
145}
146#endif
147
148extern void copy_and_flush(unsigned long dest, unsigned long src,
149 unsigned long size, unsigned long offset);
150
151/* prom structure */
152static struct prom_t __prombss prom;
153
154static unsigned long __prombss prom_entry;
155
156static char __prombss of_stdout_device[256];
157static char __prombss prom_scratch[256];
158
159static unsigned long __prombss dt_header_start;
160static unsigned long __prombss dt_struct_start, dt_struct_end;
161static unsigned long __prombss dt_string_start, dt_string_end;
162
163static unsigned long __prombss prom_initrd_start, prom_initrd_end;
164
165#ifdef CONFIG_PPC64
166static int __prombss prom_iommu_force_on;
167static int __prombss prom_iommu_off;
168static unsigned long __prombss prom_tce_alloc_start;
169static unsigned long __prombss prom_tce_alloc_end;
170#endif
171
172#ifdef CONFIG_PPC_PSERIES
173static bool __prombss prom_radix_disable;
174static bool __prombss prom_radix_gtse_disable;
175static bool __prombss prom_xive_disable;
176#endif
177
178#ifdef CONFIG_PPC_SVM
179static bool __prombss prom_svm_enable;
180#endif
181
182struct platform_support {
183 bool hash_mmu;
184 bool radix_mmu;
185 bool radix_gtse;
186 bool xive;
187};
188
189/* Platforms codes are now obsolete in the kernel. Now only used within this
190 * file and ultimately gone too. Feel free to change them if you need, they
191 * are not shared with anything outside of this file anymore
192 */
193#define PLATFORM_PSERIES 0x0100
194#define PLATFORM_PSERIES_LPAR 0x0101
195#define PLATFORM_LPAR 0x0001
196#define PLATFORM_POWERMAC 0x0400
197#define PLATFORM_GENERIC 0x0500
198
199static int __prombss of_platform;
200
201static char __prombss prom_cmd_line[COMMAND_LINE_SIZE];
202
203static unsigned long __prombss prom_memory_limit;
204
205static unsigned long __prombss alloc_top;
206static unsigned long __prombss alloc_top_high;
207static unsigned long __prombss alloc_bottom;
208static unsigned long __prombss rmo_top;
209static unsigned long __prombss ram_top;
210
211static struct mem_map_entry __prombss mem_reserve_map[MEM_RESERVE_MAP_SIZE];
212static int __prombss mem_reserve_cnt;
213
214static cell_t __prombss regbuf[1024];
215
216static bool __prombss rtas_has_query_cpu_stopped;
217
218
219/*
220 * Error results ... some OF calls will return "-1" on error, some
221 * will return 0, some will return either. To simplify, here are
222 * macros to use with any ihandle or phandle return value to check if
223 * it is valid
224 */
225
226#define PROM_ERROR (-1u)
227#define PHANDLE_VALID(p) ((p) != 0 && (p) != PROM_ERROR)
228#define IHANDLE_VALID(i) ((i) != 0 && (i) != PROM_ERROR)
229
230/* Copied from lib/string.c and lib/kstrtox.c */
231
232static int __init prom_strcmp(const char *cs, const char *ct)
233{
234 unsigned char c1, c2;
235
236 while (1) {
237 c1 = *cs++;
238 c2 = *ct++;
239 if (c1 != c2)
240 return c1 < c2 ? -1 : 1;
241 if (!c1)
242 break;
243 }
244 return 0;
245}
246
247static ssize_t __init prom_strscpy_pad(char *dest, const char *src, size_t n)
248{
249 ssize_t rc;
250 size_t i;
251
252 if (n == 0 || n > INT_MAX)
253 return -E2BIG;
254
255 // Copy up to n bytes
256 for (i = 0; i < n && src[i] != '\0'; i++)
257 dest[i] = src[i];
258
259 rc = i;
260
261 // If we copied all n then we have run out of space for the nul
262 if (rc == n) {
263 // Rewind by one character to ensure nul termination
264 i--;
265 rc = -E2BIG;
266 }
267
268 for (; i < n; i++)
269 dest[i] = '\0';
270
271 return rc;
272}
273
274static int __init prom_strncmp(const char *cs, const char *ct, size_t count)
275{
276 unsigned char c1, c2;
277
278 while (count) {
279 c1 = *cs++;
280 c2 = *ct++;
281 if (c1 != c2)
282 return c1 < c2 ? -1 : 1;
283 if (!c1)
284 break;
285 count--;
286 }
287 return 0;
288}
289
290static size_t __init prom_strlen(const char *s)
291{
292 const char *sc;
293
294 for (sc = s; *sc != '\0'; ++sc)
295 /* nothing */;
296 return sc - s;
297}
298
299static int __init prom_memcmp(const void *cs, const void *ct, size_t count)
300{
301 const unsigned char *su1, *su2;
302 int res = 0;
303
304 for (su1 = cs, su2 = ct; 0 < count; ++su1, ++su2, count--)
305 if ((res = *su1 - *su2) != 0)
306 break;
307 return res;
308}
309
310static char __init *prom_strstr(const char *s1, const char *s2)
311{
312 size_t l1, l2;
313
314 l2 = prom_strlen(s2);
315 if (!l2)
316 return (char *)s1;
317 l1 = prom_strlen(s1);
318 while (l1 >= l2) {
319 l1--;
320 if (!prom_memcmp(s1, s2, l2))
321 return (char *)s1;
322 s1++;
323 }
324 return NULL;
325}
326
327static size_t __init prom_strlcat(char *dest, const char *src, size_t count)
328{
329 size_t dsize = prom_strlen(dest);
330 size_t len = prom_strlen(src);
331 size_t res = dsize + len;
332
333 /* This would be a bug */
334 if (dsize >= count)
335 return count;
336
337 dest += dsize;
338 count -= dsize;
339 if (len >= count)
340 len = count-1;
341 memcpy(dest, src, len);
342 dest[len] = 0;
343 return res;
344
345}
346
347#ifdef CONFIG_PPC_PSERIES
348static int __init prom_strtobool(const char *s, bool *res)
349{
350 if (!s)
351 return -EINVAL;
352
353 switch (s[0]) {
354 case 'y':
355 case 'Y':
356 case '1':
357 *res = true;
358 return 0;
359 case 'n':
360 case 'N':
361 case '0':
362 *res = false;
363 return 0;
364 case 'o':
365 case 'O':
366 switch (s[1]) {
367 case 'n':
368 case 'N':
369 *res = true;
370 return 0;
371 case 'f':
372 case 'F':
373 *res = false;
374 return 0;
375 default:
376 break;
377 }
378 break;
379 default:
380 break;
381 }
382
383 return -EINVAL;
384}
385#endif
386
387/* This is the one and *ONLY* place where we actually call open
388 * firmware.
389 */
390
391static int __init call_prom(const char *service, int nargs, int nret, ...)
392{
393 int i;
394 struct prom_args args;
395 va_list list;
396
397 args.service = cpu_to_be32(ADDR(service));
398 args.nargs = cpu_to_be32(nargs);
399 args.nret = cpu_to_be32(nret);
400
401 va_start(list, nret);
402 for (i = 0; i < nargs; i++)
403 args.args[i] = cpu_to_be32(va_arg(list, prom_arg_t));
404 va_end(list);
405
406 for (i = 0; i < nret; i++)
407 args.args[nargs+i] = 0;
408
409 if (enter_prom(&args, prom_entry) < 0)
410 return PROM_ERROR;
411
412 return (nret > 0) ? be32_to_cpu(args.args[nargs]) : 0;
413}
414
415static int __init call_prom_ret(const char *service, int nargs, int nret,
416 prom_arg_t *rets, ...)
417{
418 int i;
419 struct prom_args args;
420 va_list list;
421
422 args.service = cpu_to_be32(ADDR(service));
423 args.nargs = cpu_to_be32(nargs);
424 args.nret = cpu_to_be32(nret);
425
426 va_start(list, rets);
427 for (i = 0; i < nargs; i++)
428 args.args[i] = cpu_to_be32(va_arg(list, prom_arg_t));
429 va_end(list);
430
431 for (i = 0; i < nret; i++)
432 args.args[nargs+i] = 0;
433
434 if (enter_prom(&args, prom_entry) < 0)
435 return PROM_ERROR;
436
437 if (rets != NULL)
438 for (i = 1; i < nret; ++i)
439 rets[i-1] = be32_to_cpu(args.args[nargs+i]);
440
441 return (nret > 0) ? be32_to_cpu(args.args[nargs]) : 0;
442}
443
444
445static void __init prom_print(const char *msg)
446{
447 const char *p, *q;
448
449 if (prom.stdout == 0)
450 return;
451
452 for (p = msg; *p != 0; p = q) {
453 for (q = p; *q != 0 && *q != '\n'; ++q)
454 ;
455 if (q > p)
456 call_prom("write", 3, 1, prom.stdout, p, q - p);
457 if (*q == 0)
458 break;
459 ++q;
460 call_prom("write", 3, 1, prom.stdout, ADDR("\r\n"), 2);
461 }
462}
463
464
465/*
466 * Both prom_print_hex & prom_print_dec takes an unsigned long as input so that
467 * we do not need __udivdi3 or __umoddi3 on 32bits.
468 */
469static void __init prom_print_hex(unsigned long val)
470{
471 int i, nibbles = sizeof(val)*2;
472 char buf[sizeof(val)*2+1];
473
474 for (i = nibbles-1; i >= 0; i--) {
475 buf[i] = (val & 0xf) + '0';
476 if (buf[i] > '9')
477 buf[i] += ('a'-'0'-10);
478 val >>= 4;
479 }
480 buf[nibbles] = '\0';
481 call_prom("write", 3, 1, prom.stdout, buf, nibbles);
482}
483
484/* max number of decimal digits in an unsigned long */
485#define UL_DIGITS 21
486static void __init prom_print_dec(unsigned long val)
487{
488 int i, size;
489 char buf[UL_DIGITS+1];
490
491 for (i = UL_DIGITS-1; i >= 0; i--) {
492 buf[i] = (val % 10) + '0';
493 val = val/10;
494 if (val == 0)
495 break;
496 }
497 /* shift stuff down */
498 size = UL_DIGITS - i;
499 call_prom("write", 3, 1, prom.stdout, buf+i, size);
500}
501
502__printf(1, 2)
503static void __init prom_printf(const char *format, ...)
504{
505 const char *p, *q, *s;
506 va_list args;
507 unsigned long v;
508 long vs;
509 int n = 0;
510
511 va_start(args, format);
512 for (p = format; *p != 0; p = q) {
513 for (q = p; *q != 0 && *q != '\n' && *q != '%'; ++q)
514 ;
515 if (q > p)
516 call_prom("write", 3, 1, prom.stdout, p, q - p);
517 if (*q == 0)
518 break;
519 if (*q == '\n') {
520 ++q;
521 call_prom("write", 3, 1, prom.stdout,
522 ADDR("\r\n"), 2);
523 continue;
524 }
525 ++q;
526 if (*q == 0)
527 break;
528 while (*q == 'l') {
529 ++q;
530 ++n;
531 }
532 switch (*q) {
533 case 's':
534 ++q;
535 s = va_arg(args, const char *);
536 prom_print(s);
537 break;
538 case 'x':
539 ++q;
540 switch (n) {
541 case 0:
542 v = va_arg(args, unsigned int);
543 break;
544 case 1:
545 v = va_arg(args, unsigned long);
546 break;
547 case 2:
548 default:
549 v = va_arg(args, unsigned long long);
550 break;
551 }
552 prom_print_hex(v);
553 break;
554 case 'u':
555 ++q;
556 switch (n) {
557 case 0:
558 v = va_arg(args, unsigned int);
559 break;
560 case 1:
561 v = va_arg(args, unsigned long);
562 break;
563 case 2:
564 default:
565 v = va_arg(args, unsigned long long);
566 break;
567 }
568 prom_print_dec(v);
569 break;
570 case 'd':
571 ++q;
572 switch (n) {
573 case 0:
574 vs = va_arg(args, int);
575 break;
576 case 1:
577 vs = va_arg(args, long);
578 break;
579 case 2:
580 default:
581 vs = va_arg(args, long long);
582 break;
583 }
584 if (vs < 0) {
585 prom_print("-");
586 vs = -vs;
587 }
588 prom_print_dec(vs);
589 break;
590 }
591 }
592 va_end(args);
593}
594
595
596static unsigned int __init prom_claim(unsigned long virt, unsigned long size,
597 unsigned long align)
598{
599
600 if (align == 0 && (OF_WORKAROUNDS & OF_WA_CLAIM)) {
601 /*
602 * Old OF requires we claim physical and virtual separately
603 * and then map explicitly (assuming virtual mode)
604 */
605 int ret;
606 prom_arg_t result;
607
608 ret = call_prom_ret("call-method", 5, 2, &result,
609 ADDR("claim"), prom.memory,
610 align, size, virt);
611 if (ret != 0 || result == -1)
612 return -1;
613 ret = call_prom_ret("call-method", 5, 2, &result,
614 ADDR("claim"), prom.mmumap,
615 align, size, virt);
616 if (ret != 0) {
617 call_prom("call-method", 4, 1, ADDR("release"),
618 prom.memory, size, virt);
619 return -1;
620 }
621 /* the 0x12 is M (coherence) + PP == read/write */
622 call_prom("call-method", 6, 1,
623 ADDR("map"), prom.mmumap, 0x12, size, virt, virt);
624 return virt;
625 }
626 return call_prom("claim", 3, 1, (prom_arg_t)virt, (prom_arg_t)size,
627 (prom_arg_t)align);
628}
629
630static void __init __attribute__((noreturn)) prom_panic(const char *reason)
631{
632 prom_print(reason);
633 /* Do not call exit because it clears the screen on pmac
634 * it also causes some sort of double-fault on early pmacs */
635 if (of_platform == PLATFORM_POWERMAC)
636 asm("trap\n");
637
638 /* ToDo: should put up an SRC here on pSeries */
639 call_prom("exit", 0, 0);
640
641 for (;;) /* should never get here */
642 ;
643}
644
645
646static int __init prom_next_node(phandle *nodep)
647{
648 phandle node;
649
650 if ((node = *nodep) != 0
651 && (*nodep = call_prom("child", 1, 1, node)) != 0)
652 return 1;
653 if ((*nodep = call_prom("peer", 1, 1, node)) != 0)
654 return 1;
655 for (;;) {
656 if ((node = call_prom("parent", 1, 1, node)) == 0)
657 return 0;
658 if ((*nodep = call_prom("peer", 1, 1, node)) != 0)
659 return 1;
660 }
661}
662
663static inline int __init prom_getprop(phandle node, const char *pname,
664 void *value, size_t valuelen)
665{
666 return call_prom("getprop", 4, 1, node, ADDR(pname),
667 (u32)(unsigned long) value, (u32) valuelen);
668}
669
670static inline int __init prom_getproplen(phandle node, const char *pname)
671{
672 return call_prom("getproplen", 2, 1, node, ADDR(pname));
673}
674
675static void add_string(char **str, const char *q)
676{
677 char *p = *str;
678
679 while (*q)
680 *p++ = *q++;
681 *p++ = ' ';
682 *str = p;
683}
684
685static char *tohex(unsigned int x)
686{
687 static const char digits[] __initconst = "0123456789abcdef";
688 static char result[9] __prombss;
689 int i;
690
691 result[8] = 0;
692 i = 8;
693 do {
694 --i;
695 result[i] = digits[x & 0xf];
696 x >>= 4;
697 } while (x != 0 && i > 0);
698 return &result[i];
699}
700
701static int __init prom_setprop(phandle node, const char *nodename,
702 const char *pname, void *value, size_t valuelen)
703{
704 char cmd[256], *p;
705
706 if (!(OF_WORKAROUNDS & OF_WA_LONGTRAIL))
707 return call_prom("setprop", 4, 1, node, ADDR(pname),
708 (u32)(unsigned long) value, (u32) valuelen);
709
710 /* gah... setprop doesn't work on longtrail, have to use interpret */
711 p = cmd;
712 add_string(&p, "dev");
713 add_string(&p, nodename);
714 add_string(&p, tohex((u32)(unsigned long) value));
715 add_string(&p, tohex(valuelen));
716 add_string(&p, tohex(ADDR(pname)));
717 add_string(&p, tohex(prom_strlen(pname)));
718 add_string(&p, "property");
719 *p = 0;
720 return call_prom("interpret", 1, 1, (u32)(unsigned long) cmd);
721}
722
723/* We can't use the standard versions because of relocation headaches. */
724#define prom_isxdigit(c) \
725 (('0' <= (c) && (c) <= '9') || ('a' <= (c) && (c) <= 'f') || ('A' <= (c) && (c) <= 'F'))
726
727#define prom_isdigit(c) ('0' <= (c) && (c) <= '9')
728#define prom_islower(c) ('a' <= (c) && (c) <= 'z')
729#define prom_toupper(c) (prom_islower(c) ? ((c) - 'a' + 'A') : (c))
730
731static unsigned long prom_strtoul(const char *cp, const char **endp)
732{
733 unsigned long result = 0, base = 10, value;
734
735 if (*cp == '0') {
736 base = 8;
737 cp++;
738 if (prom_toupper(*cp) == 'X') {
739 cp++;
740 base = 16;
741 }
742 }
743
744 while (prom_isxdigit(*cp) &&
745 (value = prom_isdigit(*cp) ? *cp - '0' : prom_toupper(*cp) - 'A' + 10) < base) {
746 result = result * base + value;
747 cp++;
748 }
749
750 if (endp)
751 *endp = cp;
752
753 return result;
754}
755
756static unsigned long prom_memparse(const char *ptr, const char **retptr)
757{
758 unsigned long ret = prom_strtoul(ptr, retptr);
759 int shift = 0;
760
761 /*
762 * We can't use a switch here because GCC *may* generate a
763 * jump table which won't work, because we're not running at
764 * the address we're linked at.
765 */
766 if ('G' == **retptr || 'g' == **retptr)
767 shift = 30;
768
769 if ('M' == **retptr || 'm' == **retptr)
770 shift = 20;
771
772 if ('K' == **retptr || 'k' == **retptr)
773 shift = 10;
774
775 if (shift) {
776 ret <<= shift;
777 (*retptr)++;
778 }
779
780 return ret;
781}
782
783/*
784 * Early parsing of the command line passed to the kernel, used for
785 * "mem=x" and the options that affect the iommu
786 */
787static void __init early_cmdline_parse(void)
788{
789 const char *opt;
790
791 char *p;
792 int l = 0;
793
794 prom_cmd_line[0] = 0;
795 p = prom_cmd_line;
796
797 if (!IS_ENABLED(CONFIG_CMDLINE_FORCE) && (long)prom.chosen > 0)
798 l = prom_getprop(prom.chosen, "bootargs", p, COMMAND_LINE_SIZE-1);
799
800 if (IS_ENABLED(CONFIG_CMDLINE_EXTEND) || l <= 0 || p[0] == '\0')
801 prom_strlcat(prom_cmd_line, " " CONFIG_CMDLINE,
802 sizeof(prom_cmd_line));
803
804 prom_printf("command line: %s\n", prom_cmd_line);
805
806#ifdef CONFIG_PPC64
807 opt = prom_strstr(prom_cmd_line, "iommu=");
808 if (opt) {
809 prom_printf("iommu opt is: %s\n", opt);
810 opt += 6;
811 while (*opt && *opt == ' ')
812 opt++;
813 if (!prom_strncmp(opt, "off", 3))
814 prom_iommu_off = 1;
815 else if (!prom_strncmp(opt, "force", 5))
816 prom_iommu_force_on = 1;
817 }
818#endif
819 opt = prom_strstr(prom_cmd_line, "mem=");
820 if (opt) {
821 opt += 4;
822 prom_memory_limit = prom_memparse(opt, (const char **)&opt);
823#ifdef CONFIG_PPC64
824 /* Align to 16 MB == size of ppc64 large page */
825 prom_memory_limit = ALIGN(prom_memory_limit, 0x1000000);
826#endif
827 }
828
829#ifdef CONFIG_PPC_PSERIES
830 prom_radix_disable = !IS_ENABLED(CONFIG_PPC_RADIX_MMU_DEFAULT);
831 opt = prom_strstr(prom_cmd_line, "disable_radix");
832 if (opt) {
833 opt += 13;
834 if (*opt && *opt == '=') {
835 bool val;
836
837 if (prom_strtobool(++opt, &val))
838 prom_radix_disable = false;
839 else
840 prom_radix_disable = val;
841 } else
842 prom_radix_disable = true;
843 }
844 if (prom_radix_disable)
845 prom_debug("Radix disabled from cmdline\n");
846
847 opt = prom_strstr(prom_cmd_line, "radix_hcall_invalidate=on");
848 if (opt) {
849 prom_radix_gtse_disable = true;
850 prom_debug("Radix GTSE disabled from cmdline\n");
851 }
852
853 opt = prom_strstr(prom_cmd_line, "xive=off");
854 if (opt) {
855 prom_xive_disable = true;
856 prom_debug("XIVE disabled from cmdline\n");
857 }
858#endif /* CONFIG_PPC_PSERIES */
859
860#ifdef CONFIG_PPC_SVM
861 opt = prom_strstr(prom_cmd_line, "svm=");
862 if (opt) {
863 bool val;
864
865 opt += sizeof("svm=") - 1;
866 if (!prom_strtobool(opt, &val))
867 prom_svm_enable = val;
868 }
869#endif /* CONFIG_PPC_SVM */
870}
871
872#ifdef CONFIG_PPC_PSERIES
873/*
874 * The architecture vector has an array of PVR mask/value pairs,
875 * followed by # option vectors - 1, followed by the option vectors.
876 *
877 * See prom.h for the definition of the bits specified in the
878 * architecture vector.
879 */
880
881/* Firmware expects the value to be n - 1, where n is the # of vectors */
882#define NUM_VECTORS(n) ((n) - 1)
883
884/*
885 * Firmware expects 1 + n - 2, where n is the length of the option vector in
886 * bytes. The 1 accounts for the length byte itself, the - 2 .. ?
887 */
888#define VECTOR_LENGTH(n) (1 + (n) - 2)
889
890struct option_vector1 {
891 u8 byte1;
892 u8 arch_versions;
893 u8 arch_versions3;
894} __packed;
895
896struct option_vector2 {
897 u8 byte1;
898 __be16 reserved;
899 __be32 real_base;
900 __be32 real_size;
901 __be32 virt_base;
902 __be32 virt_size;
903 __be32 load_base;
904 __be32 min_rma;
905 __be32 min_load;
906 u8 min_rma_percent;
907 u8 max_pft_size;
908} __packed;
909
910struct option_vector3 {
911 u8 byte1;
912 u8 byte2;
913} __packed;
914
915struct option_vector4 {
916 u8 byte1;
917 u8 min_vp_cap;
918} __packed;
919
920struct option_vector5 {
921 u8 byte1;
922 u8 byte2;
923 u8 byte3;
924 u8 cmo;
925 u8 associativity;
926 u8 bin_opts;
927 u8 micro_checkpoint;
928 u8 reserved0;
929 __be32 max_cpus;
930 __be16 papr_level;
931 __be16 reserved1;
932 u8 platform_facilities;
933 u8 reserved2;
934 __be16 reserved3;
935 u8 subprocessors;
936 u8 byte22;
937 u8 intarch;
938 u8 mmu;
939 u8 hash_ext;
940 u8 radix_ext;
941} __packed;
942
943struct option_vector6 {
944 u8 reserved;
945 u8 secondary_pteg;
946 u8 os_name;
947} __packed;
948
949struct option_vector7 {
950 u8 os_id[256];
951} __packed;
952
953struct ibm_arch_vec {
954 struct { u32 mask, val; } pvrs[14];
955
956 u8 num_vectors;
957
958 u8 vec1_len;
959 struct option_vector1 vec1;
960
961 u8 vec2_len;
962 struct option_vector2 vec2;
963
964 u8 vec3_len;
965 struct option_vector3 vec3;
966
967 u8 vec4_len;
968 struct option_vector4 vec4;
969
970 u8 vec5_len;
971 struct option_vector5 vec5;
972
973 u8 vec6_len;
974 struct option_vector6 vec6;
975
976 u8 vec7_len;
977 struct option_vector7 vec7;
978} __packed;
979
980static const struct ibm_arch_vec ibm_architecture_vec_template __initconst = {
981 .pvrs = {
982 {
983 .mask = cpu_to_be32(0xfffe0000), /* POWER5/POWER5+ */
984 .val = cpu_to_be32(0x003a0000),
985 },
986 {
987 .mask = cpu_to_be32(0xffff0000), /* POWER6 */
988 .val = cpu_to_be32(0x003e0000),
989 },
990 {
991 .mask = cpu_to_be32(0xffff0000), /* POWER7 */
992 .val = cpu_to_be32(0x003f0000),
993 },
994 {
995 .mask = cpu_to_be32(0xffff0000), /* POWER8E */
996 .val = cpu_to_be32(0x004b0000),
997 },
998 {
999 .mask = cpu_to_be32(0xffff0000), /* POWER8NVL */
1000 .val = cpu_to_be32(0x004c0000),
1001 },
1002 {
1003 .mask = cpu_to_be32(0xffff0000), /* POWER8 */
1004 .val = cpu_to_be32(0x004d0000),
1005 },
1006 {
1007 .mask = cpu_to_be32(0xffff0000), /* POWER9 */
1008 .val = cpu_to_be32(0x004e0000),
1009 },
1010 {
1011 .mask = cpu_to_be32(0xffff0000), /* POWER10 */
1012 .val = cpu_to_be32(0x00800000),
1013 },
1014 {
1015 .mask = cpu_to_be32(0xffffffff), /* all 3.1-compliant */
1016 .val = cpu_to_be32(0x0f000006),
1017 },
1018 {
1019 .mask = cpu_to_be32(0xffffffff), /* all 3.00-compliant */
1020 .val = cpu_to_be32(0x0f000005),
1021 },
1022 {
1023 .mask = cpu_to_be32(0xffffffff), /* all 2.07-compliant */
1024 .val = cpu_to_be32(0x0f000004),
1025 },
1026 {
1027 .mask = cpu_to_be32(0xffffffff), /* all 2.06-compliant */
1028 .val = cpu_to_be32(0x0f000003),
1029 },
1030 {
1031 .mask = cpu_to_be32(0xffffffff), /* all 2.05-compliant */
1032 .val = cpu_to_be32(0x0f000002),
1033 },
1034 {
1035 .mask = cpu_to_be32(0xfffffffe), /* all 2.04-compliant and earlier */
1036 .val = cpu_to_be32(0x0f000001),
1037 },
1038 },
1039
1040 .num_vectors = NUM_VECTORS(6),
1041
1042 .vec1_len = VECTOR_LENGTH(sizeof(struct option_vector1)),
1043 .vec1 = {
1044 .byte1 = 0,
1045 .arch_versions = OV1_PPC_2_00 | OV1_PPC_2_01 | OV1_PPC_2_02 | OV1_PPC_2_03 |
1046 OV1_PPC_2_04 | OV1_PPC_2_05 | OV1_PPC_2_06 | OV1_PPC_2_07,
1047 .arch_versions3 = OV1_PPC_3_00 | OV1_PPC_3_1,
1048 },
1049
1050 .vec2_len = VECTOR_LENGTH(sizeof(struct option_vector2)),
1051 /* option vector 2: Open Firmware options supported */
1052 .vec2 = {
1053 .byte1 = OV2_REAL_MODE,
1054 .reserved = 0,
1055 .real_base = cpu_to_be32(0xffffffff),
1056 .real_size = cpu_to_be32(0xffffffff),
1057 .virt_base = cpu_to_be32(0xffffffff),
1058 .virt_size = cpu_to_be32(0xffffffff),
1059 .load_base = cpu_to_be32(0xffffffff),
1060 .min_rma = cpu_to_be32(512), /* 512MB min RMA */
1061 .min_load = cpu_to_be32(0xffffffff), /* full client load */
1062 .min_rma_percent = 0, /* min RMA percentage of total RAM */
1063 .max_pft_size = 48, /* max log_2(hash table size) */
1064 },
1065
1066 .vec3_len = VECTOR_LENGTH(sizeof(struct option_vector3)),
1067 /* option vector 3: processor options supported */
1068 .vec3 = {
1069 .byte1 = 0, /* don't ignore, don't halt */
1070 .byte2 = OV3_FP | OV3_VMX | OV3_DFP,
1071 },
1072
1073 .vec4_len = VECTOR_LENGTH(sizeof(struct option_vector4)),
1074 /* option vector 4: IBM PAPR implementation */
1075 .vec4 = {
1076 .byte1 = 0, /* don't halt */
1077 .min_vp_cap = OV4_MIN_ENT_CAP, /* minimum VP entitled capacity */
1078 },
1079
1080 .vec5_len = VECTOR_LENGTH(sizeof(struct option_vector5)),
1081 /* option vector 5: PAPR/OF options */
1082 .vec5 = {
1083 .byte1 = 0, /* don't ignore, don't halt */
1084 .byte2 = OV5_FEAT(OV5_LPAR) | OV5_FEAT(OV5_SPLPAR) | OV5_FEAT(OV5_LARGE_PAGES) |
1085 OV5_FEAT(OV5_DRCONF_MEMORY) | OV5_FEAT(OV5_DONATE_DEDICATE_CPU) |
1086#ifdef CONFIG_PCI_MSI
1087 /* PCIe/MSI support. Without MSI full PCIe is not supported */
1088 OV5_FEAT(OV5_MSI),
1089#else
1090 0,
1091#endif
1092 .byte3 = 0,
1093 .cmo =
1094#ifdef CONFIG_PPC_SMLPAR
1095 OV5_FEAT(OV5_CMO) | OV5_FEAT(OV5_XCMO),
1096#else
1097 0,
1098#endif
1099 .associativity = OV5_FEAT(OV5_TYPE1_AFFINITY) | OV5_FEAT(OV5_PRRN),
1100 .bin_opts = OV5_FEAT(OV5_RESIZE_HPT) | OV5_FEAT(OV5_HP_EVT),
1101 .micro_checkpoint = 0,
1102 .reserved0 = 0,
1103 .max_cpus = cpu_to_be32(NR_CPUS), /* number of cores supported */
1104 .papr_level = 0,
1105 .reserved1 = 0,
1106 .platform_facilities = OV5_FEAT(OV5_PFO_HW_RNG) | OV5_FEAT(OV5_PFO_HW_ENCR) | OV5_FEAT(OV5_PFO_HW_842),
1107 .reserved2 = 0,
1108 .reserved3 = 0,
1109 .subprocessors = 1,
1110 .byte22 = OV5_FEAT(OV5_DRMEM_V2) | OV5_FEAT(OV5_DRC_INFO),
1111 .intarch = 0,
1112 .mmu = 0,
1113 .hash_ext = 0,
1114 .radix_ext = 0,
1115 },
1116
1117 /* option vector 6: IBM PAPR hints */
1118 .vec6_len = VECTOR_LENGTH(sizeof(struct option_vector6)),
1119 .vec6 = {
1120 .reserved = 0,
1121 .secondary_pteg = 0,
1122 .os_name = OV6_LINUX,
1123 },
1124
1125 /* option vector 7: OS Identification */
1126 .vec7_len = VECTOR_LENGTH(sizeof(struct option_vector7)),
1127};
1128
1129static struct ibm_arch_vec __prombss ibm_architecture_vec ____cacheline_aligned;
1130
1131/* Old method - ELF header with PT_NOTE sections only works on BE */
1132#ifdef __BIG_ENDIAN__
1133static const struct fake_elf {
1134 Elf32_Ehdr elfhdr;
1135 Elf32_Phdr phdr[2];
1136 struct chrpnote {
1137 u32 namesz;
1138 u32 descsz;
1139 u32 type;
1140 char name[8]; /* "PowerPC" */
1141 struct chrpdesc {
1142 u32 real_mode;
1143 u32 real_base;
1144 u32 real_size;
1145 u32 virt_base;
1146 u32 virt_size;
1147 u32 load_base;
1148 } chrpdesc;
1149 } chrpnote;
1150 struct rpanote {
1151 u32 namesz;
1152 u32 descsz;
1153 u32 type;
1154 char name[24]; /* "IBM,RPA-Client-Config" */
1155 struct rpadesc {
1156 u32 lpar_affinity;
1157 u32 min_rmo_size;
1158 u32 min_rmo_percent;
1159 u32 max_pft_size;
1160 u32 splpar;
1161 u32 min_load;
1162 u32 new_mem_def;
1163 u32 ignore_me;
1164 } rpadesc;
1165 } rpanote;
1166} fake_elf __initconst = {
1167 .elfhdr = {
1168 .e_ident = { 0x7f, 'E', 'L', 'F',
1169 ELFCLASS32, ELFDATA2MSB, EV_CURRENT },
1170 .e_type = ET_EXEC, /* yeah right */
1171 .e_machine = EM_PPC,
1172 .e_version = EV_CURRENT,
1173 .e_phoff = offsetof(struct fake_elf, phdr),
1174 .e_phentsize = sizeof(Elf32_Phdr),
1175 .e_phnum = 2
1176 },
1177 .phdr = {
1178 [0] = {
1179 .p_type = PT_NOTE,
1180 .p_offset = offsetof(struct fake_elf, chrpnote),
1181 .p_filesz = sizeof(struct chrpnote)
1182 }, [1] = {
1183 .p_type = PT_NOTE,
1184 .p_offset = offsetof(struct fake_elf, rpanote),
1185 .p_filesz = sizeof(struct rpanote)
1186 }
1187 },
1188 .chrpnote = {
1189 .namesz = sizeof("PowerPC"),
1190 .descsz = sizeof(struct chrpdesc),
1191 .type = 0x1275,
1192 .name = "PowerPC",
1193 .chrpdesc = {
1194 .real_mode = ~0U, /* ~0 means "don't care" */
1195 .real_base = ~0U,
1196 .real_size = ~0U,
1197 .virt_base = ~0U,
1198 .virt_size = ~0U,
1199 .load_base = ~0U
1200 },
1201 },
1202 .rpanote = {
1203 .namesz = sizeof("IBM,RPA-Client-Config"),
1204 .descsz = sizeof(struct rpadesc),
1205 .type = 0x12759999,
1206 .name = "IBM,RPA-Client-Config",
1207 .rpadesc = {
1208 .lpar_affinity = 0,
1209 .min_rmo_size = 64, /* in megabytes */
1210 .min_rmo_percent = 0,
1211 .max_pft_size = 48, /* 2^48 bytes max PFT size */
1212 .splpar = 1,
1213 .min_load = ~0U,
1214 .new_mem_def = 0
1215 }
1216 }
1217};
1218#endif /* __BIG_ENDIAN__ */
1219
1220static int __init prom_count_smt_threads(void)
1221{
1222 phandle node;
1223 char type[64];
1224 unsigned int plen;
1225
1226 /* Pick up th first CPU node we can find */
1227 for (node = 0; prom_next_node(&node); ) {
1228 type[0] = 0;
1229 prom_getprop(node, "device_type", type, sizeof(type));
1230
1231 if (prom_strcmp(type, "cpu"))
1232 continue;
1233 /*
1234 * There is an entry for each smt thread, each entry being
1235 * 4 bytes long. All cpus should have the same number of
1236 * smt threads, so return after finding the first.
1237 */
1238 plen = prom_getproplen(node, "ibm,ppc-interrupt-server#s");
1239 if (plen == PROM_ERROR)
1240 break;
1241 plen >>= 2;
1242 prom_debug("Found %lu smt threads per core\n", (unsigned long)plen);
1243
1244 /* Sanity check */
1245 if (plen < 1 || plen > 64) {
1246 prom_printf("Threads per core %lu out of bounds, assuming 1\n",
1247 (unsigned long)plen);
1248 return 1;
1249 }
1250 return plen;
1251 }
1252 prom_debug("No threads found, assuming 1 per core\n");
1253
1254 return 1;
1255
1256}
1257
1258static void __init prom_parse_mmu_model(u8 val,
1259 struct platform_support *support)
1260{
1261 switch (val) {
1262 case OV5_FEAT(OV5_MMU_DYNAMIC):
1263 case OV5_FEAT(OV5_MMU_EITHER): /* Either Available */
1264 prom_debug("MMU - either supported\n");
1265 support->radix_mmu = !prom_radix_disable;
1266 support->hash_mmu = true;
1267 break;
1268 case OV5_FEAT(OV5_MMU_RADIX): /* Only Radix */
1269 prom_debug("MMU - radix only\n");
1270 if (prom_radix_disable) {
1271 /*
1272 * If we __have__ to do radix, we're better off ignoring
1273 * the command line rather than not booting.
1274 */
1275 prom_printf("WARNING: Ignoring cmdline option disable_radix\n");
1276 }
1277 support->radix_mmu = true;
1278 break;
1279 case OV5_FEAT(OV5_MMU_HASH):
1280 prom_debug("MMU - hash only\n");
1281 support->hash_mmu = true;
1282 break;
1283 default:
1284 prom_debug("Unknown mmu support option: 0x%x\n", val);
1285 break;
1286 }
1287}
1288
1289static void __init prom_parse_xive_model(u8 val,
1290 struct platform_support *support)
1291{
1292 switch (val) {
1293 case OV5_FEAT(OV5_XIVE_EITHER): /* Either Available */
1294 prom_debug("XIVE - either mode supported\n");
1295 support->xive = !prom_xive_disable;
1296 break;
1297 case OV5_FEAT(OV5_XIVE_EXPLOIT): /* Only Exploitation mode */
1298 prom_debug("XIVE - exploitation mode supported\n");
1299 if (prom_xive_disable) {
1300 /*
1301 * If we __have__ to do XIVE, we're better off ignoring
1302 * the command line rather than not booting.
1303 */
1304 prom_printf("WARNING: Ignoring cmdline option xive=off\n");
1305 }
1306 support->xive = true;
1307 break;
1308 case OV5_FEAT(OV5_XIVE_LEGACY): /* Only Legacy mode */
1309 prom_debug("XIVE - legacy mode supported\n");
1310 break;
1311 default:
1312 prom_debug("Unknown xive support option: 0x%x\n", val);
1313 break;
1314 }
1315}
1316
1317static void __init prom_parse_platform_support(u8 index, u8 val,
1318 struct platform_support *support)
1319{
1320 switch (index) {
1321 case OV5_INDX(OV5_MMU_SUPPORT): /* MMU Model */
1322 prom_parse_mmu_model(val & OV5_FEAT(OV5_MMU_SUPPORT), support);
1323 break;
1324 case OV5_INDX(OV5_RADIX_GTSE): /* Radix Extensions */
1325 if (val & OV5_FEAT(OV5_RADIX_GTSE))
1326 support->radix_gtse = !prom_radix_gtse_disable;
1327 break;
1328 case OV5_INDX(OV5_XIVE_SUPPORT): /* Interrupt mode */
1329 prom_parse_xive_model(val & OV5_FEAT(OV5_XIVE_SUPPORT),
1330 support);
1331 break;
1332 }
1333}
1334
1335static void __init prom_check_platform_support(void)
1336{
1337 struct platform_support supported = {
1338 .hash_mmu = false,
1339 .radix_mmu = false,
1340 .radix_gtse = false,
1341 .xive = false
1342 };
1343 int prop_len = prom_getproplen(prom.chosen,
1344 "ibm,arch-vec-5-platform-support");
1345
1346 /*
1347 * First copy the architecture vec template
1348 *
1349 * use memcpy() instead of *vec = *vec_template so that GCC replaces it
1350 * by __memcpy() when KASAN is active
1351 */
1352 memcpy(&ibm_architecture_vec, &ibm_architecture_vec_template,
1353 sizeof(ibm_architecture_vec));
1354
1355 prom_strscpy_pad(ibm_architecture_vec.vec7.os_id, linux_banner, 256);
1356
1357 if (prop_len > 1) {
1358 int i;
1359 u8 vec[8];
1360 prom_debug("Found ibm,arch-vec-5-platform-support, len: %d\n",
1361 prop_len);
1362 if (prop_len > sizeof(vec))
1363 prom_printf("WARNING: ibm,arch-vec-5-platform-support longer than expected (len: %d)\n",
1364 prop_len);
1365 prom_getprop(prom.chosen, "ibm,arch-vec-5-platform-support", &vec, sizeof(vec));
1366 for (i = 0; i < prop_len; i += 2) {
1367 prom_debug("%d: index = 0x%x val = 0x%x\n", i / 2, vec[i], vec[i + 1]);
1368 prom_parse_platform_support(vec[i], vec[i + 1], &supported);
1369 }
1370 }
1371
1372 if (supported.radix_mmu && IS_ENABLED(CONFIG_PPC_RADIX_MMU)) {
1373 /* Radix preferred - Check if GTSE is also supported */
1374 prom_debug("Asking for radix\n");
1375 ibm_architecture_vec.vec5.mmu = OV5_FEAT(OV5_MMU_RADIX);
1376 if (supported.radix_gtse)
1377 ibm_architecture_vec.vec5.radix_ext =
1378 OV5_FEAT(OV5_RADIX_GTSE);
1379 else
1380 prom_debug("Radix GTSE isn't supported\n");
1381 } else if (supported.hash_mmu) {
1382 /* Default to hash mmu (if we can) */
1383 prom_debug("Asking for hash\n");
1384 ibm_architecture_vec.vec5.mmu = OV5_FEAT(OV5_MMU_HASH);
1385 } else {
1386 /* We're probably on a legacy hypervisor */
1387 prom_debug("Assuming legacy hash support\n");
1388 }
1389
1390 if (supported.xive) {
1391 prom_debug("Asking for XIVE\n");
1392 ibm_architecture_vec.vec5.intarch = OV5_FEAT(OV5_XIVE_EXPLOIT);
1393 }
1394}
1395
1396static void __init prom_send_capabilities(void)
1397{
1398 ihandle root;
1399 prom_arg_t ret;
1400 u32 cores;
1401
1402 /* Check ibm,arch-vec-5-platform-support and fixup vec5 if required */
1403 prom_check_platform_support();
1404
1405 root = call_prom("open", 1, 1, ADDR("/"));
1406 if (root != 0) {
1407 /* We need to tell the FW about the number of cores we support.
1408 *
1409 * To do that, we count the number of threads on the first core
1410 * (we assume this is the same for all cores) and use it to
1411 * divide NR_CPUS.
1412 */
1413
1414 cores = DIV_ROUND_UP(NR_CPUS, prom_count_smt_threads());
1415 prom_printf("Max number of cores passed to firmware: %u (NR_CPUS = %d)\n",
1416 cores, NR_CPUS);
1417
1418 ibm_architecture_vec.vec5.max_cpus = cpu_to_be32(cores);
1419
1420 /* try calling the ibm,client-architecture-support method */
1421 prom_printf("Calling ibm,client-architecture-support...");
1422 if (call_prom_ret("call-method", 3, 2, &ret,
1423 ADDR("ibm,client-architecture-support"),
1424 root,
1425 ADDR(&ibm_architecture_vec)) == 0) {
1426 /* the call exists... */
1427 if (ret)
1428 prom_printf("\nWARNING: ibm,client-architecture"
1429 "-support call FAILED!\n");
1430 call_prom("close", 1, 0, root);
1431 prom_printf(" done\n");
1432 return;
1433 }
1434 call_prom("close", 1, 0, root);
1435 prom_printf(" not implemented\n");
1436 }
1437
1438#ifdef __BIG_ENDIAN__
1439 {
1440 ihandle elfloader;
1441
1442 /* no ibm,client-architecture-support call, try the old way */
1443 elfloader = call_prom("open", 1, 1,
1444 ADDR("/packages/elf-loader"));
1445 if (elfloader == 0) {
1446 prom_printf("couldn't open /packages/elf-loader\n");
1447 return;
1448 }
1449 call_prom("call-method", 3, 1, ADDR("process-elf-header"),
1450 elfloader, ADDR(&fake_elf));
1451 call_prom("close", 1, 0, elfloader);
1452 }
1453#endif /* __BIG_ENDIAN__ */
1454}
1455#endif /* CONFIG_PPC_PSERIES */
1456
1457/*
1458 * Memory allocation strategy... our layout is normally:
1459 *
1460 * at 14Mb or more we have vmlinux, then a gap and initrd. In some
1461 * rare cases, initrd might end up being before the kernel though.
1462 * We assume this won't override the final kernel at 0, we have no
1463 * provision to handle that in this version, but it should hopefully
1464 * never happen.
1465 *
1466 * alloc_top is set to the top of RMO, eventually shrink down if the
1467 * TCEs overlap
1468 *
1469 * alloc_bottom is set to the top of kernel/initrd
1470 *
1471 * from there, allocations are done this way : rtas is allocated
1472 * topmost, and the device-tree is allocated from the bottom. We try
1473 * to grow the device-tree allocation as we progress. If we can't,
1474 * then we fail, we don't currently have a facility to restart
1475 * elsewhere, but that shouldn't be necessary.
1476 *
1477 * Note that calls to reserve_mem have to be done explicitly, memory
1478 * allocated with either alloc_up or alloc_down isn't automatically
1479 * reserved.
1480 */
1481
1482
1483/*
1484 * Allocates memory in the RMO upward from the kernel/initrd
1485 *
1486 * When align is 0, this is a special case, it means to allocate in place
1487 * at the current location of alloc_bottom or fail (that is basically
1488 * extending the previous allocation). Used for the device-tree flattening
1489 */
1490static unsigned long __init alloc_up(unsigned long size, unsigned long align)
1491{
1492 unsigned long base = alloc_bottom;
1493 unsigned long addr = 0;
1494
1495 if (align)
1496 base = ALIGN(base, align);
1497 prom_debug("%s(%lx, %lx)\n", __func__, size, align);
1498 if (ram_top == 0)
1499 prom_panic("alloc_up() called with mem not initialized\n");
1500
1501 if (align)
1502 base = ALIGN(alloc_bottom, align);
1503 else
1504 base = alloc_bottom;
1505
1506 for(; (base + size) <= alloc_top;
1507 base = ALIGN(base + 0x100000, align)) {
1508 prom_debug(" trying: 0x%lx\n\r", base);
1509 addr = (unsigned long)prom_claim(base, size, 0);
1510 if (addr != PROM_ERROR && addr != 0)
1511 break;
1512 addr = 0;
1513 if (align == 0)
1514 break;
1515 }
1516 if (addr == 0)
1517 return 0;
1518 alloc_bottom = addr + size;
1519
1520 prom_debug(" -> %lx\n", addr);
1521 prom_debug(" alloc_bottom : %lx\n", alloc_bottom);
1522 prom_debug(" alloc_top : %lx\n", alloc_top);
1523 prom_debug(" alloc_top_hi : %lx\n", alloc_top_high);
1524 prom_debug(" rmo_top : %lx\n", rmo_top);
1525 prom_debug(" ram_top : %lx\n", ram_top);
1526
1527 return addr;
1528}
1529
1530/*
1531 * Allocates memory downward, either from top of RMO, or if highmem
1532 * is set, from the top of RAM. Note that this one doesn't handle
1533 * failures. It does claim memory if highmem is not set.
1534 */
1535static unsigned long __init alloc_down(unsigned long size, unsigned long align,
1536 int highmem)
1537{
1538 unsigned long base, addr = 0;
1539
1540 prom_debug("%s(%lx, %lx, %s)\n", __func__, size, align,
1541 highmem ? "(high)" : "(low)");
1542 if (ram_top == 0)
1543 prom_panic("alloc_down() called with mem not initialized\n");
1544
1545 if (highmem) {
1546 /* Carve out storage for the TCE table. */
1547 addr = ALIGN_DOWN(alloc_top_high - size, align);
1548 if (addr <= alloc_bottom)
1549 return 0;
1550 /* Will we bump into the RMO ? If yes, check out that we
1551 * didn't overlap existing allocations there, if we did,
1552 * we are dead, we must be the first in town !
1553 */
1554 if (addr < rmo_top) {
1555 /* Good, we are first */
1556 if (alloc_top == rmo_top)
1557 alloc_top = rmo_top = addr;
1558 else
1559 return 0;
1560 }
1561 alloc_top_high = addr;
1562 goto bail;
1563 }
1564
1565 base = ALIGN_DOWN(alloc_top - size, align);
1566 for (; base > alloc_bottom;
1567 base = ALIGN_DOWN(base - 0x100000, align)) {
1568 prom_debug(" trying: 0x%lx\n\r", base);
1569 addr = (unsigned long)prom_claim(base, size, 0);
1570 if (addr != PROM_ERROR && addr != 0)
1571 break;
1572 addr = 0;
1573 }
1574 if (addr == 0)
1575 return 0;
1576 alloc_top = addr;
1577
1578 bail:
1579 prom_debug(" -> %lx\n", addr);
1580 prom_debug(" alloc_bottom : %lx\n", alloc_bottom);
1581 prom_debug(" alloc_top : %lx\n", alloc_top);
1582 prom_debug(" alloc_top_hi : %lx\n", alloc_top_high);
1583 prom_debug(" rmo_top : %lx\n", rmo_top);
1584 prom_debug(" ram_top : %lx\n", ram_top);
1585
1586 return addr;
1587}
1588
1589/*
1590 * Parse a "reg" cell
1591 */
1592static unsigned long __init prom_next_cell(int s, cell_t **cellp)
1593{
1594 cell_t *p = *cellp;
1595 unsigned long r = 0;
1596
1597 /* Ignore more than 2 cells */
1598 while (s > sizeof(unsigned long) / 4) {
1599 p++;
1600 s--;
1601 }
1602 r = be32_to_cpu(*p++);
1603#ifdef CONFIG_PPC64
1604 if (s > 1) {
1605 r <<= 32;
1606 r |= be32_to_cpu(*(p++));
1607 }
1608#endif
1609 *cellp = p;
1610 return r;
1611}
1612
1613/*
1614 * Very dumb function for adding to the memory reserve list, but
1615 * we don't need anything smarter at this point
1616 *
1617 * XXX Eventually check for collisions. They should NEVER happen.
1618 * If problems seem to show up, it would be a good start to track
1619 * them down.
1620 */
1621static void __init reserve_mem(u64 base, u64 size)
1622{
1623 u64 top = base + size;
1624 unsigned long cnt = mem_reserve_cnt;
1625
1626 if (size == 0)
1627 return;
1628
1629 /* We need to always keep one empty entry so that we
1630 * have our terminator with "size" set to 0 since we are
1631 * dumb and just copy this entire array to the boot params
1632 */
1633 base = ALIGN_DOWN(base, PAGE_SIZE);
1634 top = ALIGN(top, PAGE_SIZE);
1635 size = top - base;
1636
1637 if (cnt >= (MEM_RESERVE_MAP_SIZE - 1))
1638 prom_panic("Memory reserve map exhausted !\n");
1639 mem_reserve_map[cnt].base = cpu_to_be64(base);
1640 mem_reserve_map[cnt].size = cpu_to_be64(size);
1641 mem_reserve_cnt = cnt + 1;
1642}
1643
1644/*
1645 * Initialize memory allocation mechanism, parse "memory" nodes and
1646 * obtain that way the top of memory and RMO to setup out local allocator
1647 */
1648static void __init prom_init_mem(void)
1649{
1650 phandle node;
1651 char type[64];
1652 unsigned int plen;
1653 cell_t *p, *endp;
1654 __be32 val;
1655 u32 rac, rsc;
1656
1657 /*
1658 * We iterate the memory nodes to find
1659 * 1) top of RMO (first node)
1660 * 2) top of memory
1661 */
1662 val = cpu_to_be32(2);
1663 prom_getprop(prom.root, "#address-cells", &val, sizeof(val));
1664 rac = be32_to_cpu(val);
1665 val = cpu_to_be32(1);
1666 prom_getprop(prom.root, "#size-cells", &val, sizeof(rsc));
1667 rsc = be32_to_cpu(val);
1668 prom_debug("root_addr_cells: %x\n", rac);
1669 prom_debug("root_size_cells: %x\n", rsc);
1670
1671 prom_debug("scanning memory:\n");
1672
1673 for (node = 0; prom_next_node(&node); ) {
1674 type[0] = 0;
1675 prom_getprop(node, "device_type", type, sizeof(type));
1676
1677 if (type[0] == 0) {
1678 /*
1679 * CHRP Longtrail machines have no device_type
1680 * on the memory node, so check the name instead...
1681 */
1682 prom_getprop(node, "name", type, sizeof(type));
1683 }
1684 if (prom_strcmp(type, "memory"))
1685 continue;
1686
1687 plen = prom_getprop(node, "reg", regbuf, sizeof(regbuf));
1688 if (plen > sizeof(regbuf)) {
1689 prom_printf("memory node too large for buffer !\n");
1690 plen = sizeof(regbuf);
1691 }
1692 p = regbuf;
1693 endp = p + (plen / sizeof(cell_t));
1694
1695#ifdef DEBUG_PROM
1696 memset(prom_scratch, 0, sizeof(prom_scratch));
1697 call_prom("package-to-path", 3, 1, node, prom_scratch,
1698 sizeof(prom_scratch) - 1);
1699 prom_debug(" node %s :\n", prom_scratch);
1700#endif /* DEBUG_PROM */
1701
1702 while ((endp - p) >= (rac + rsc)) {
1703 unsigned long base, size;
1704
1705 base = prom_next_cell(rac, &p);
1706 size = prom_next_cell(rsc, &p);
1707
1708 if (size == 0)
1709 continue;
1710 prom_debug(" %lx %lx\n", base, size);
1711 if (base == 0 && (of_platform & PLATFORM_LPAR))
1712 rmo_top = size;
1713 if ((base + size) > ram_top)
1714 ram_top = base + size;
1715 }
1716 }
1717
1718 alloc_bottom = PAGE_ALIGN((unsigned long)&_end + 0x4000);
1719
1720 /*
1721 * If prom_memory_limit is set we reduce the upper limits *except* for
1722 * alloc_top_high. This must be the real top of RAM so we can put
1723 * TCE's up there.
1724 */
1725
1726 alloc_top_high = ram_top;
1727
1728 if (prom_memory_limit) {
1729 if (prom_memory_limit <= alloc_bottom) {
1730 prom_printf("Ignoring mem=%lx <= alloc_bottom.\n",
1731 prom_memory_limit);
1732 prom_memory_limit = 0;
1733 } else if (prom_memory_limit >= ram_top) {
1734 prom_printf("Ignoring mem=%lx >= ram_top.\n",
1735 prom_memory_limit);
1736 prom_memory_limit = 0;
1737 } else {
1738 ram_top = prom_memory_limit;
1739 rmo_top = min(rmo_top, prom_memory_limit);
1740 }
1741 }
1742
1743 /*
1744 * Setup our top alloc point, that is top of RMO or top of
1745 * segment 0 when running non-LPAR.
1746 * Some RS64 machines have buggy firmware where claims up at
1747 * 1GB fail. Cap at 768MB as a workaround.
1748 * Since 768MB is plenty of room, and we need to cap to something
1749 * reasonable on 32-bit, cap at 768MB on all machines.
1750 */
1751 if (!rmo_top)
1752 rmo_top = ram_top;
1753 rmo_top = min(0x30000000ul, rmo_top);
1754 alloc_top = rmo_top;
1755 alloc_top_high = ram_top;
1756
1757 /*
1758 * Check if we have an initrd after the kernel but still inside
1759 * the RMO. If we do move our bottom point to after it.
1760 */
1761 if (prom_initrd_start &&
1762 prom_initrd_start < rmo_top &&
1763 prom_initrd_end > alloc_bottom)
1764 alloc_bottom = PAGE_ALIGN(prom_initrd_end);
1765
1766 prom_printf("memory layout at init:\n");
1767 prom_printf(" memory_limit : %lx (16 MB aligned)\n",
1768 prom_memory_limit);
1769 prom_printf(" alloc_bottom : %lx\n", alloc_bottom);
1770 prom_printf(" alloc_top : %lx\n", alloc_top);
1771 prom_printf(" alloc_top_hi : %lx\n", alloc_top_high);
1772 prom_printf(" rmo_top : %lx\n", rmo_top);
1773 prom_printf(" ram_top : %lx\n", ram_top);
1774}
1775
1776static void __init prom_close_stdin(void)
1777{
1778 __be32 val;
1779 ihandle stdin;
1780
1781 if (prom_getprop(prom.chosen, "stdin", &val, sizeof(val)) > 0) {
1782 stdin = be32_to_cpu(val);
1783 call_prom("close", 1, 0, stdin);
1784 }
1785}
1786
1787#ifdef CONFIG_PPC_SVM
1788static int prom_rtas_hcall(uint64_t args)
1789{
1790 register uint64_t arg1 asm("r3") = H_RTAS;
1791 register uint64_t arg2 asm("r4") = args;
1792
1793 asm volatile("sc 1\n" : "=r" (arg1) :
1794 "r" (arg1),
1795 "r" (arg2) :);
1796 srr_regs_clobbered();
1797
1798 return arg1;
1799}
1800
1801static struct rtas_args __prombss os_term_args;
1802
1803static void __init prom_rtas_os_term(char *str)
1804{
1805 phandle rtas_node;
1806 __be32 val;
1807 u32 token;
1808
1809 prom_debug("%s: start...\n", __func__);
1810 rtas_node = call_prom("finddevice", 1, 1, ADDR("/rtas"));
1811 prom_debug("rtas_node: %x\n", rtas_node);
1812 if (!PHANDLE_VALID(rtas_node))
1813 return;
1814
1815 val = 0;
1816 prom_getprop(rtas_node, "ibm,os-term", &val, sizeof(val));
1817 token = be32_to_cpu(val);
1818 prom_debug("ibm,os-term: %x\n", token);
1819 if (token == 0)
1820 prom_panic("Could not get token for ibm,os-term\n");
1821 os_term_args.token = cpu_to_be32(token);
1822 os_term_args.nargs = cpu_to_be32(1);
1823 os_term_args.nret = cpu_to_be32(1);
1824 os_term_args.args[0] = cpu_to_be32(__pa(str));
1825 prom_rtas_hcall((uint64_t)&os_term_args);
1826}
1827#endif /* CONFIG_PPC_SVM */
1828
1829/*
1830 * Allocate room for and instantiate RTAS
1831 */
1832static void __init prom_instantiate_rtas(void)
1833{
1834 phandle rtas_node;
1835 ihandle rtas_inst;
1836 u32 base, entry = 0;
1837 __be32 val;
1838 u32 size = 0;
1839
1840 prom_debug("prom_instantiate_rtas: start...\n");
1841
1842 rtas_node = call_prom("finddevice", 1, 1, ADDR("/rtas"));
1843 prom_debug("rtas_node: %x\n", rtas_node);
1844 if (!PHANDLE_VALID(rtas_node))
1845 return;
1846
1847 val = 0;
1848 prom_getprop(rtas_node, "rtas-size", &val, sizeof(size));
1849 size = be32_to_cpu(val);
1850 if (size == 0)
1851 return;
1852
1853 base = alloc_down(size, PAGE_SIZE, 0);
1854 if (base == 0)
1855 prom_panic("Could not allocate memory for RTAS\n");
1856
1857 rtas_inst = call_prom("open", 1, 1, ADDR("/rtas"));
1858 if (!IHANDLE_VALID(rtas_inst)) {
1859 prom_printf("opening rtas package failed (%x)\n", rtas_inst);
1860 return;
1861 }
1862
1863 prom_printf("instantiating rtas at 0x%x...", base);
1864
1865 if (call_prom_ret("call-method", 3, 2, &entry,
1866 ADDR("instantiate-rtas"),
1867 rtas_inst, base) != 0
1868 || entry == 0) {
1869 prom_printf(" failed\n");
1870 return;
1871 }
1872 prom_printf(" done\n");
1873
1874 reserve_mem(base, size);
1875
1876 val = cpu_to_be32(base);
1877 prom_setprop(rtas_node, "/rtas", "linux,rtas-base",
1878 &val, sizeof(val));
1879 val = cpu_to_be32(entry);
1880 prom_setprop(rtas_node, "/rtas", "linux,rtas-entry",
1881 &val, sizeof(val));
1882
1883 /* Check if it supports "query-cpu-stopped-state" */
1884 if (prom_getprop(rtas_node, "query-cpu-stopped-state",
1885 &val, sizeof(val)) != PROM_ERROR)
1886 rtas_has_query_cpu_stopped = true;
1887
1888 prom_debug("rtas base = 0x%x\n", base);
1889 prom_debug("rtas entry = 0x%x\n", entry);
1890 prom_debug("rtas size = 0x%x\n", size);
1891
1892 prom_debug("prom_instantiate_rtas: end...\n");
1893}
1894
1895#ifdef CONFIG_PPC64
1896/*
1897 * Allocate room for and instantiate Stored Measurement Log (SML)
1898 */
1899static void __init prom_instantiate_sml(void)
1900{
1901 phandle ibmvtpm_node;
1902 ihandle ibmvtpm_inst;
1903 u32 entry = 0, size = 0, succ = 0;
1904 u64 base;
1905 __be32 val;
1906
1907 prom_debug("prom_instantiate_sml: start...\n");
1908
1909 ibmvtpm_node = call_prom("finddevice", 1, 1, ADDR("/vdevice/vtpm"));
1910 prom_debug("ibmvtpm_node: %x\n", ibmvtpm_node);
1911 if (!PHANDLE_VALID(ibmvtpm_node))
1912 return;
1913
1914 ibmvtpm_inst = call_prom("open", 1, 1, ADDR("/vdevice/vtpm"));
1915 if (!IHANDLE_VALID(ibmvtpm_inst)) {
1916 prom_printf("opening vtpm package failed (%x)\n", ibmvtpm_inst);
1917 return;
1918 }
1919
1920 if (prom_getprop(ibmvtpm_node, "ibm,sml-efi-reformat-supported",
1921 &val, sizeof(val)) != PROM_ERROR) {
1922 if (call_prom_ret("call-method", 2, 2, &succ,
1923 ADDR("reformat-sml-to-efi-alignment"),
1924 ibmvtpm_inst) != 0 || succ == 0) {
1925 prom_printf("Reformat SML to EFI alignment failed\n");
1926 return;
1927 }
1928
1929 if (call_prom_ret("call-method", 2, 2, &size,
1930 ADDR("sml-get-allocated-size"),
1931 ibmvtpm_inst) != 0 || size == 0) {
1932 prom_printf("SML get allocated size failed\n");
1933 return;
1934 }
1935 } else {
1936 if (call_prom_ret("call-method", 2, 2, &size,
1937 ADDR("sml-get-handover-size"),
1938 ibmvtpm_inst) != 0 || size == 0) {
1939 prom_printf("SML get handover size failed\n");
1940 return;
1941 }
1942 }
1943
1944 base = alloc_down(size, PAGE_SIZE, 0);
1945 if (base == 0)
1946 prom_panic("Could not allocate memory for sml\n");
1947
1948 prom_printf("instantiating sml at 0x%llx...", base);
1949
1950 memset((void *)base, 0, size);
1951
1952 if (call_prom_ret("call-method", 4, 2, &entry,
1953 ADDR("sml-handover"),
1954 ibmvtpm_inst, size, base) != 0 || entry == 0) {
1955 prom_printf("SML handover failed\n");
1956 return;
1957 }
1958 prom_printf(" done\n");
1959
1960 reserve_mem(base, size);
1961
1962 prom_setprop(ibmvtpm_node, "/vdevice/vtpm", "linux,sml-base",
1963 &base, sizeof(base));
1964 prom_setprop(ibmvtpm_node, "/vdevice/vtpm", "linux,sml-size",
1965 &size, sizeof(size));
1966
1967 prom_debug("sml base = 0x%llx\n", base);
1968 prom_debug("sml size = 0x%x\n", size);
1969
1970 prom_debug("prom_instantiate_sml: end...\n");
1971}
1972
1973/*
1974 * Allocate room for and initialize TCE tables
1975 */
1976#ifdef __BIG_ENDIAN__
1977static void __init prom_initialize_tce_table(void)
1978{
1979 phandle node;
1980 ihandle phb_node;
1981 char compatible[64], type[64], model[64];
1982 char *path = prom_scratch;
1983 u64 base, align;
1984 u32 minalign, minsize;
1985 u64 tce_entry, *tce_entryp;
1986 u64 local_alloc_top, local_alloc_bottom;
1987 u64 i;
1988
1989 if (prom_iommu_off)
1990 return;
1991
1992 prom_debug("starting prom_initialize_tce_table\n");
1993
1994 /* Cache current top of allocs so we reserve a single block */
1995 local_alloc_top = alloc_top_high;
1996 local_alloc_bottom = local_alloc_top;
1997
1998 /* Search all nodes looking for PHBs. */
1999 for (node = 0; prom_next_node(&node); ) {
2000 compatible[0] = 0;
2001 type[0] = 0;
2002 model[0] = 0;
2003 prom_getprop(node, "compatible",
2004 compatible, sizeof(compatible));
2005 prom_getprop(node, "device_type", type, sizeof(type));
2006 prom_getprop(node, "model", model, sizeof(model));
2007
2008 if ((type[0] == 0) || (prom_strstr(type, "pci") == NULL))
2009 continue;
2010
2011 /* Keep the old logic intact to avoid regression. */
2012 if (compatible[0] != 0) {
2013 if ((prom_strstr(compatible, "python") == NULL) &&
2014 (prom_strstr(compatible, "Speedwagon") == NULL) &&
2015 (prom_strstr(compatible, "Winnipeg") == NULL))
2016 continue;
2017 } else if (model[0] != 0) {
2018 if ((prom_strstr(model, "ython") == NULL) &&
2019 (prom_strstr(model, "peedwagon") == NULL) &&
2020 (prom_strstr(model, "innipeg") == NULL))
2021 continue;
2022 }
2023
2024 if (prom_getprop(node, "tce-table-minalign", &minalign,
2025 sizeof(minalign)) == PROM_ERROR)
2026 minalign = 0;
2027 if (prom_getprop(node, "tce-table-minsize", &minsize,
2028 sizeof(minsize)) == PROM_ERROR)
2029 minsize = 4UL << 20;
2030
2031 /*
2032 * Even though we read what OF wants, we just set the table
2033 * size to 4 MB. This is enough to map 2GB of PCI DMA space.
2034 * By doing this, we avoid the pitfalls of trying to DMA to
2035 * MMIO space and the DMA alias hole.
2036 */
2037 minsize = 4UL << 20;
2038
2039 /* Align to the greater of the align or size */
2040 align = max(minalign, minsize);
2041 base = alloc_down(minsize, align, 1);
2042 if (base == 0)
2043 prom_panic("ERROR, cannot find space for TCE table.\n");
2044 if (base < local_alloc_bottom)
2045 local_alloc_bottom = base;
2046
2047 /* It seems OF doesn't null-terminate the path :-( */
2048 memset(path, 0, sizeof(prom_scratch));
2049 /* Call OF to setup the TCE hardware */
2050 if (call_prom("package-to-path", 3, 1, node,
2051 path, sizeof(prom_scratch) - 1) == PROM_ERROR) {
2052 prom_printf("package-to-path failed\n");
2053 }
2054
2055 /* Save away the TCE table attributes for later use. */
2056 prom_setprop(node, path, "linux,tce-base", &base, sizeof(base));
2057 prom_setprop(node, path, "linux,tce-size", &minsize, sizeof(minsize));
2058
2059 prom_debug("TCE table: %s\n", path);
2060 prom_debug("\tnode = 0x%x\n", node);
2061 prom_debug("\tbase = 0x%llx\n", base);
2062 prom_debug("\tsize = 0x%x\n", minsize);
2063
2064 /* Initialize the table to have a one-to-one mapping
2065 * over the allocated size.
2066 */
2067 tce_entryp = (u64 *)base;
2068 for (i = 0; i < (minsize >> 3) ;tce_entryp++, i++) {
2069 tce_entry = (i << PAGE_SHIFT);
2070 tce_entry |= 0x3;
2071 *tce_entryp = tce_entry;
2072 }
2073
2074 prom_printf("opening PHB %s", path);
2075 phb_node = call_prom("open", 1, 1, path);
2076 if (phb_node == 0)
2077 prom_printf("... failed\n");
2078 else
2079 prom_printf("... done\n");
2080
2081 call_prom("call-method", 6, 0, ADDR("set-64-bit-addressing"),
2082 phb_node, -1, minsize,
2083 (u32) base, (u32) (base >> 32));
2084 call_prom("close", 1, 0, phb_node);
2085 }
2086
2087 reserve_mem(local_alloc_bottom, local_alloc_top - local_alloc_bottom);
2088
2089 /* These are only really needed if there is a memory limit in
2090 * effect, but we don't know so export them always. */
2091 prom_tce_alloc_start = local_alloc_bottom;
2092 prom_tce_alloc_end = local_alloc_top;
2093
2094 /* Flag the first invalid entry */
2095 prom_debug("ending prom_initialize_tce_table\n");
2096}
2097#endif /* __BIG_ENDIAN__ */
2098#endif /* CONFIG_PPC64 */
2099
2100/*
2101 * With CHRP SMP we need to use the OF to start the other processors.
2102 * We can't wait until smp_boot_cpus (the OF is trashed by then)
2103 * so we have to put the processors into a holding pattern controlled
2104 * by the kernel (not OF) before we destroy the OF.
2105 *
2106 * This uses a chunk of low memory, puts some holding pattern
2107 * code there and sends the other processors off to there until
2108 * smp_boot_cpus tells them to do something. The holding pattern
2109 * checks that address until its cpu # is there, when it is that
2110 * cpu jumps to __secondary_start(). smp_boot_cpus() takes care
2111 * of setting those values.
2112 *
2113 * We also use physical address 0x4 here to tell when a cpu
2114 * is in its holding pattern code.
2115 *
2116 * -- Cort
2117 */
2118/*
2119 * We want to reference the copy of __secondary_hold_* in the
2120 * 0 - 0x100 address range
2121 */
2122#define LOW_ADDR(x) (((unsigned long) &(x)) & 0xff)
2123
2124static void __init prom_hold_cpus(void)
2125{
2126 unsigned long i;
2127 phandle node;
2128 char type[64];
2129 unsigned long *spinloop
2130 = (void *) LOW_ADDR(__secondary_hold_spinloop);
2131 unsigned long *acknowledge
2132 = (void *) LOW_ADDR(__secondary_hold_acknowledge);
2133 unsigned long secondary_hold = LOW_ADDR(__secondary_hold);
2134
2135 /*
2136 * On pseries, if RTAS supports "query-cpu-stopped-state",
2137 * we skip this stage, the CPUs will be started by the
2138 * kernel using RTAS.
2139 */
2140 if ((of_platform == PLATFORM_PSERIES ||
2141 of_platform == PLATFORM_PSERIES_LPAR) &&
2142 rtas_has_query_cpu_stopped) {
2143 prom_printf("prom_hold_cpus: skipped\n");
2144 return;
2145 }
2146
2147 prom_debug("prom_hold_cpus: start...\n");
2148 prom_debug(" 1) spinloop = 0x%lx\n", (unsigned long)spinloop);
2149 prom_debug(" 1) *spinloop = 0x%lx\n", *spinloop);
2150 prom_debug(" 1) acknowledge = 0x%lx\n",
2151 (unsigned long)acknowledge);
2152 prom_debug(" 1) *acknowledge = 0x%lx\n", *acknowledge);
2153 prom_debug(" 1) secondary_hold = 0x%lx\n", secondary_hold);
2154
2155 /* Set the common spinloop variable, so all of the secondary cpus
2156 * will block when they are awakened from their OF spinloop.
2157 * This must occur for both SMP and non SMP kernels, since OF will
2158 * be trashed when we move the kernel.
2159 */
2160 *spinloop = 0;
2161
2162 /* look for cpus */
2163 for (node = 0; prom_next_node(&node); ) {
2164 unsigned int cpu_no;
2165 __be32 reg;
2166
2167 type[0] = 0;
2168 prom_getprop(node, "device_type", type, sizeof(type));
2169 if (prom_strcmp(type, "cpu") != 0)
2170 continue;
2171
2172 /* Skip non-configured cpus. */
2173 if (prom_getprop(node, "status", type, sizeof(type)) > 0)
2174 if (prom_strcmp(type, "okay") != 0)
2175 continue;
2176
2177 reg = cpu_to_be32(-1); /* make sparse happy */
2178 prom_getprop(node, "reg", ®, sizeof(reg));
2179 cpu_no = be32_to_cpu(reg);
2180
2181 prom_debug("cpu hw idx = %u\n", cpu_no);
2182
2183 /* Init the acknowledge var which will be reset by
2184 * the secondary cpu when it awakens from its OF
2185 * spinloop.
2186 */
2187 *acknowledge = (unsigned long)-1;
2188
2189 if (cpu_no != prom.cpu) {
2190 /* Primary Thread of non-boot cpu or any thread */
2191 prom_printf("starting cpu hw idx %u... ", cpu_no);
2192 call_prom("start-cpu", 3, 0, node,
2193 secondary_hold, cpu_no);
2194
2195 for (i = 0; (i < 100000000) &&
2196 (*acknowledge == ((unsigned long)-1)); i++ )
2197 mb();
2198
2199 if (*acknowledge == cpu_no)
2200 prom_printf("done\n");
2201 else
2202 prom_printf("failed: %lx\n", *acknowledge);
2203 }
2204#ifdef CONFIG_SMP
2205 else
2206 prom_printf("boot cpu hw idx %u\n", cpu_no);
2207#endif /* CONFIG_SMP */
2208 }
2209
2210 prom_debug("prom_hold_cpus: end...\n");
2211}
2212
2213
2214static void __init prom_init_client_services(unsigned long pp)
2215{
2216 /* Get a handle to the prom entry point before anything else */
2217 prom_entry = pp;
2218
2219 /* get a handle for the stdout device */
2220 prom.chosen = call_prom("finddevice", 1, 1, ADDR("/chosen"));
2221 if (!PHANDLE_VALID(prom.chosen))
2222 prom_panic("cannot find chosen"); /* msg won't be printed :( */
2223
2224 /* get device tree root */
2225 prom.root = call_prom("finddevice", 1, 1, ADDR("/"));
2226 if (!PHANDLE_VALID(prom.root))
2227 prom_panic("cannot find device tree root"); /* msg won't be printed :( */
2228
2229 prom.mmumap = 0;
2230}
2231
2232#ifdef CONFIG_PPC32
2233/*
2234 * For really old powermacs, we need to map things we claim.
2235 * For that, we need the ihandle of the mmu.
2236 * Also, on the longtrail, we need to work around other bugs.
2237 */
2238static void __init prom_find_mmu(void)
2239{
2240 phandle oprom;
2241 char version[64];
2242
2243 oprom = call_prom("finddevice", 1, 1, ADDR("/openprom"));
2244 if (!PHANDLE_VALID(oprom))
2245 return;
2246 if (prom_getprop(oprom, "model", version, sizeof(version)) <= 0)
2247 return;
2248 version[sizeof(version) - 1] = 0;
2249 /* XXX might need to add other versions here */
2250 if (prom_strcmp(version, "Open Firmware, 1.0.5") == 0)
2251 of_workarounds = OF_WA_CLAIM;
2252 else if (prom_strncmp(version, "FirmWorks,3.", 12) == 0) {
2253 of_workarounds = OF_WA_CLAIM | OF_WA_LONGTRAIL;
2254 call_prom("interpret", 1, 1, "dev /memory 0 to allow-reclaim");
2255 } else
2256 return;
2257 prom.memory = call_prom("open", 1, 1, ADDR("/memory"));
2258 prom_getprop(prom.chosen, "mmu", &prom.mmumap,
2259 sizeof(prom.mmumap));
2260 prom.mmumap = be32_to_cpu(prom.mmumap);
2261 if (!IHANDLE_VALID(prom.memory) || !IHANDLE_VALID(prom.mmumap))
2262 of_workarounds &= ~OF_WA_CLAIM; /* hmmm */
2263}
2264#else
2265#define prom_find_mmu()
2266#endif
2267
2268static void __init prom_init_stdout(void)
2269{
2270 char *path = of_stdout_device;
2271 char type[16];
2272 phandle stdout_node;
2273 __be32 val;
2274
2275 if (prom_getprop(prom.chosen, "stdout", &val, sizeof(val)) <= 0)
2276 prom_panic("cannot find stdout");
2277
2278 prom.stdout = be32_to_cpu(val);
2279
2280 /* Get the full OF pathname of the stdout device */
2281 memset(path, 0, 256);
2282 call_prom("instance-to-path", 3, 1, prom.stdout, path, 255);
2283 prom_printf("OF stdout device is: %s\n", of_stdout_device);
2284 prom_setprop(prom.chosen, "/chosen", "linux,stdout-path",
2285 path, prom_strlen(path) + 1);
2286
2287 /* instance-to-package fails on PA-Semi */
2288 stdout_node = call_prom("instance-to-package", 1, 1, prom.stdout);
2289 if (stdout_node != PROM_ERROR) {
2290 val = cpu_to_be32(stdout_node);
2291
2292 /* If it's a display, note it */
2293 memset(type, 0, sizeof(type));
2294 prom_getprop(stdout_node, "device_type", type, sizeof(type));
2295 if (prom_strcmp(type, "display") == 0)
2296 prom_setprop(stdout_node, path, "linux,boot-display", NULL, 0);
2297 }
2298}
2299
2300static int __init prom_find_machine_type(void)
2301{
2302 char compat[256];
2303 int len, i = 0;
2304#ifdef CONFIG_PPC64
2305 phandle rtas;
2306 int x;
2307#endif
2308
2309 /* Look for a PowerMac or a Cell */
2310 len = prom_getprop(prom.root, "compatible",
2311 compat, sizeof(compat)-1);
2312 if (len > 0) {
2313 compat[len] = 0;
2314 while (i < len) {
2315 char *p = &compat[i];
2316 int sl = prom_strlen(p);
2317 if (sl == 0)
2318 break;
2319 if (prom_strstr(p, "Power Macintosh") ||
2320 prom_strstr(p, "MacRISC"))
2321 return PLATFORM_POWERMAC;
2322#ifdef CONFIG_PPC64
2323 /* We must make sure we don't detect the IBM Cell
2324 * blades as pSeries due to some firmware issues,
2325 * so we do it here.
2326 */
2327 if (prom_strstr(p, "IBM,CBEA") ||
2328 prom_strstr(p, "IBM,CPBW-1.0"))
2329 return PLATFORM_GENERIC;
2330#endif /* CONFIG_PPC64 */
2331 i += sl + 1;
2332 }
2333 }
2334#ifdef CONFIG_PPC64
2335 /* Try to figure out if it's an IBM pSeries or any other
2336 * PAPR compliant platform. We assume it is if :
2337 * - /device_type is "chrp" (please, do NOT use that for future
2338 * non-IBM designs !
2339 * - it has /rtas
2340 */
2341 len = prom_getprop(prom.root, "device_type",
2342 compat, sizeof(compat)-1);
2343 if (len <= 0)
2344 return PLATFORM_GENERIC;
2345 if (prom_strcmp(compat, "chrp"))
2346 return PLATFORM_GENERIC;
2347
2348 /* Default to pSeries. We need to know if we are running LPAR */
2349 rtas = call_prom("finddevice", 1, 1, ADDR("/rtas"));
2350 if (!PHANDLE_VALID(rtas))
2351 return PLATFORM_GENERIC;
2352 x = prom_getproplen(rtas, "ibm,hypertas-functions");
2353 if (x != PROM_ERROR) {
2354 prom_debug("Hypertas detected, assuming LPAR !\n");
2355 return PLATFORM_PSERIES_LPAR;
2356 }
2357 return PLATFORM_PSERIES;
2358#else
2359 return PLATFORM_GENERIC;
2360#endif
2361}
2362
2363static int __init prom_set_color(ihandle ih, int i, int r, int g, int b)
2364{
2365 return call_prom("call-method", 6, 1, ADDR("color!"), ih, i, b, g, r);
2366}
2367
2368/*
2369 * If we have a display that we don't know how to drive,
2370 * we will want to try to execute OF's open method for it
2371 * later. However, OF will probably fall over if we do that
2372 * we've taken over the MMU.
2373 * So we check whether we will need to open the display,
2374 * and if so, open it now.
2375 */
2376static void __init prom_check_displays(void)
2377{
2378 char type[16], *path;
2379 phandle node;
2380 ihandle ih;
2381 int i;
2382
2383 static const unsigned char default_colors[] __initconst = {
2384 0x00, 0x00, 0x00,
2385 0x00, 0x00, 0xaa,
2386 0x00, 0xaa, 0x00,
2387 0x00, 0xaa, 0xaa,
2388 0xaa, 0x00, 0x00,
2389 0xaa, 0x00, 0xaa,
2390 0xaa, 0xaa, 0x00,
2391 0xaa, 0xaa, 0xaa,
2392 0x55, 0x55, 0x55,
2393 0x55, 0x55, 0xff,
2394 0x55, 0xff, 0x55,
2395 0x55, 0xff, 0xff,
2396 0xff, 0x55, 0x55,
2397 0xff, 0x55, 0xff,
2398 0xff, 0xff, 0x55,
2399 0xff, 0xff, 0xff
2400 };
2401 const unsigned char *clut;
2402
2403 prom_debug("Looking for displays\n");
2404 for (node = 0; prom_next_node(&node); ) {
2405 memset(type, 0, sizeof(type));
2406 prom_getprop(node, "device_type", type, sizeof(type));
2407 if (prom_strcmp(type, "display") != 0)
2408 continue;
2409
2410 /* It seems OF doesn't null-terminate the path :-( */
2411 path = prom_scratch;
2412 memset(path, 0, sizeof(prom_scratch));
2413
2414 /*
2415 * leave some room at the end of the path for appending extra
2416 * arguments
2417 */
2418 if (call_prom("package-to-path", 3, 1, node, path,
2419 sizeof(prom_scratch) - 10) == PROM_ERROR)
2420 continue;
2421 prom_printf("found display : %s, opening... ", path);
2422
2423 ih = call_prom("open", 1, 1, path);
2424 if (ih == 0) {
2425 prom_printf("failed\n");
2426 continue;
2427 }
2428
2429 /* Success */
2430 prom_printf("done\n");
2431 prom_setprop(node, path, "linux,opened", NULL, 0);
2432
2433 /* Setup a usable color table when the appropriate
2434 * method is available. Should update this to set-colors */
2435 clut = default_colors;
2436 for (i = 0; i < 16; i++, clut += 3)
2437 if (prom_set_color(ih, i, clut[0], clut[1],
2438 clut[2]) != 0)
2439 break;
2440
2441#ifdef CONFIG_LOGO_LINUX_CLUT224
2442 clut = PTRRELOC(logo_linux_clut224.clut);
2443 for (i = 0; i < logo_linux_clut224.clutsize; i++, clut += 3)
2444 if (prom_set_color(ih, i + 32, clut[0], clut[1],
2445 clut[2]) != 0)
2446 break;
2447#endif /* CONFIG_LOGO_LINUX_CLUT224 */
2448
2449#ifdef CONFIG_PPC_EARLY_DEBUG_BOOTX
2450 if (prom_getprop(node, "linux,boot-display", NULL, 0) !=
2451 PROM_ERROR) {
2452 u32 width, height, pitch, addr;
2453
2454 prom_printf("Setting btext !\n");
2455
2456 if (prom_getprop(node, "width", &width, 4) == PROM_ERROR)
2457 return;
2458
2459 if (prom_getprop(node, "height", &height, 4) == PROM_ERROR)
2460 return;
2461
2462 if (prom_getprop(node, "linebytes", &pitch, 4) == PROM_ERROR)
2463 return;
2464
2465 if (prom_getprop(node, "address", &addr, 4) == PROM_ERROR)
2466 return;
2467
2468 prom_printf("W=%d H=%d LB=%d addr=0x%x\n",
2469 width, height, pitch, addr);
2470 btext_setup_display(width, height, 8, pitch, addr);
2471 btext_prepare_BAT();
2472 }
2473#endif /* CONFIG_PPC_EARLY_DEBUG_BOOTX */
2474 }
2475}
2476
2477
2478/* Return (relocated) pointer to this much memory: moves initrd if reqd. */
2479static void __init *make_room(unsigned long *mem_start, unsigned long *mem_end,
2480 unsigned long needed, unsigned long align)
2481{
2482 void *ret;
2483
2484 *mem_start = ALIGN(*mem_start, align);
2485 while ((*mem_start + needed) > *mem_end) {
2486 unsigned long room, chunk;
2487
2488 prom_debug("Chunk exhausted, claiming more at %lx...\n",
2489 alloc_bottom);
2490 room = alloc_top - alloc_bottom;
2491 if (room > DEVTREE_CHUNK_SIZE)
2492 room = DEVTREE_CHUNK_SIZE;
2493 if (room < PAGE_SIZE)
2494 prom_panic("No memory for flatten_device_tree "
2495 "(no room)\n");
2496 chunk = alloc_up(room, 0);
2497 if (chunk == 0)
2498 prom_panic("No memory for flatten_device_tree "
2499 "(claim failed)\n");
2500 *mem_end = chunk + room;
2501 }
2502
2503 ret = (void *)*mem_start;
2504 *mem_start += needed;
2505
2506 return ret;
2507}
2508
2509#define dt_push_token(token, mem_start, mem_end) do { \
2510 void *room = make_room(mem_start, mem_end, 4, 4); \
2511 *(__be32 *)room = cpu_to_be32(token); \
2512 } while(0)
2513
2514static unsigned long __init dt_find_string(char *str)
2515{
2516 char *s, *os;
2517
2518 s = os = (char *)dt_string_start;
2519 s += 4;
2520 while (s < (char *)dt_string_end) {
2521 if (prom_strcmp(s, str) == 0)
2522 return s - os;
2523 s += prom_strlen(s) + 1;
2524 }
2525 return 0;
2526}
2527
2528/*
2529 * The Open Firmware 1275 specification states properties must be 31 bytes or
2530 * less, however not all firmwares obey this. Make it 64 bytes to be safe.
2531 */
2532#define MAX_PROPERTY_NAME 64
2533
2534static void __init scan_dt_build_strings(phandle node,
2535 unsigned long *mem_start,
2536 unsigned long *mem_end)
2537{
2538 char *prev_name, *namep, *sstart;
2539 unsigned long soff;
2540 phandle child;
2541
2542 sstart = (char *)dt_string_start;
2543
2544 /* get and store all property names */
2545 prev_name = "";
2546 for (;;) {
2547 /* 64 is max len of name including nul. */
2548 namep = make_room(mem_start, mem_end, MAX_PROPERTY_NAME, 1);
2549 if (call_prom("nextprop", 3, 1, node, prev_name, namep) != 1) {
2550 /* No more nodes: unwind alloc */
2551 *mem_start = (unsigned long)namep;
2552 break;
2553 }
2554
2555 /* skip "name" */
2556 if (prom_strcmp(namep, "name") == 0) {
2557 *mem_start = (unsigned long)namep;
2558 prev_name = "name";
2559 continue;
2560 }
2561 /* get/create string entry */
2562 soff = dt_find_string(namep);
2563 if (soff != 0) {
2564 *mem_start = (unsigned long)namep;
2565 namep = sstart + soff;
2566 } else {
2567 /* Trim off some if we can */
2568 *mem_start = (unsigned long)namep + prom_strlen(namep) + 1;
2569 dt_string_end = *mem_start;
2570 }
2571 prev_name = namep;
2572 }
2573
2574 /* do all our children */
2575 child = call_prom("child", 1, 1, node);
2576 while (child != 0) {
2577 scan_dt_build_strings(child, mem_start, mem_end);
2578 child = call_prom("peer", 1, 1, child);
2579 }
2580}
2581
2582static void __init scan_dt_build_struct(phandle node, unsigned long *mem_start,
2583 unsigned long *mem_end)
2584{
2585 phandle child;
2586 char *namep, *prev_name, *sstart, *p, *ep, *lp, *path;
2587 unsigned long soff;
2588 unsigned char *valp;
2589 static char pname[MAX_PROPERTY_NAME] __prombss;
2590 int l, room, has_phandle = 0;
2591
2592 dt_push_token(OF_DT_BEGIN_NODE, mem_start, mem_end);
2593
2594 /* get the node's full name */
2595 namep = (char *)*mem_start;
2596 room = *mem_end - *mem_start;
2597 if (room > 255)
2598 room = 255;
2599 l = call_prom("package-to-path", 3, 1, node, namep, room);
2600 if (l >= 0) {
2601 /* Didn't fit? Get more room. */
2602 if (l >= room) {
2603 if (l >= *mem_end - *mem_start)
2604 namep = make_room(mem_start, mem_end, l+1, 1);
2605 call_prom("package-to-path", 3, 1, node, namep, l);
2606 }
2607 namep[l] = '\0';
2608
2609 /* Fixup an Apple bug where they have bogus \0 chars in the
2610 * middle of the path in some properties, and extract
2611 * the unit name (everything after the last '/').
2612 */
2613 for (lp = p = namep, ep = namep + l; p < ep; p++) {
2614 if (*p == '/')
2615 lp = namep;
2616 else if (*p != 0)
2617 *lp++ = *p;
2618 }
2619 *lp = 0;
2620 *mem_start = ALIGN((unsigned long)lp + 1, 4);
2621 }
2622
2623 /* get it again for debugging */
2624 path = prom_scratch;
2625 memset(path, 0, sizeof(prom_scratch));
2626 call_prom("package-to-path", 3, 1, node, path, sizeof(prom_scratch) - 1);
2627
2628 /* get and store all properties */
2629 prev_name = "";
2630 sstart = (char *)dt_string_start;
2631 for (;;) {
2632 if (call_prom("nextprop", 3, 1, node, prev_name,
2633 pname) != 1)
2634 break;
2635
2636 /* skip "name" */
2637 if (prom_strcmp(pname, "name") == 0) {
2638 prev_name = "name";
2639 continue;
2640 }
2641
2642 /* find string offset */
2643 soff = dt_find_string(pname);
2644 if (soff == 0) {
2645 prom_printf("WARNING: Can't find string index for"
2646 " <%s>, node %s\n", pname, path);
2647 break;
2648 }
2649 prev_name = sstart + soff;
2650
2651 /* get length */
2652 l = call_prom("getproplen", 2, 1, node, pname);
2653
2654 /* sanity checks */
2655 if (l == PROM_ERROR)
2656 continue;
2657
2658 /* push property head */
2659 dt_push_token(OF_DT_PROP, mem_start, mem_end);
2660 dt_push_token(l, mem_start, mem_end);
2661 dt_push_token(soff, mem_start, mem_end);
2662
2663 /* push property content */
2664 valp = make_room(mem_start, mem_end, l, 4);
2665 call_prom("getprop", 4, 1, node, pname, valp, l);
2666 *mem_start = ALIGN(*mem_start, 4);
2667
2668 if (!prom_strcmp(pname, "phandle"))
2669 has_phandle = 1;
2670 }
2671
2672 /* Add a "phandle" property if none already exist */
2673 if (!has_phandle) {
2674 soff = dt_find_string("phandle");
2675 if (soff == 0)
2676 prom_printf("WARNING: Can't find string index for <phandle> node %s\n", path);
2677 else {
2678 dt_push_token(OF_DT_PROP, mem_start, mem_end);
2679 dt_push_token(4, mem_start, mem_end);
2680 dt_push_token(soff, mem_start, mem_end);
2681 valp = make_room(mem_start, mem_end, 4, 4);
2682 *(__be32 *)valp = cpu_to_be32(node);
2683 }
2684 }
2685
2686 /* do all our children */
2687 child = call_prom("child", 1, 1, node);
2688 while (child != 0) {
2689 scan_dt_build_struct(child, mem_start, mem_end);
2690 child = call_prom("peer", 1, 1, child);
2691 }
2692
2693 dt_push_token(OF_DT_END_NODE, mem_start, mem_end);
2694}
2695
2696static void __init flatten_device_tree(void)
2697{
2698 phandle root;
2699 unsigned long mem_start, mem_end, room;
2700 struct boot_param_header *hdr;
2701 char *namep;
2702 u64 *rsvmap;
2703
2704 /*
2705 * Check how much room we have between alloc top & bottom (+/- a
2706 * few pages), crop to 1MB, as this is our "chunk" size
2707 */
2708 room = alloc_top - alloc_bottom - 0x4000;
2709 if (room > DEVTREE_CHUNK_SIZE)
2710 room = DEVTREE_CHUNK_SIZE;
2711 prom_debug("starting device tree allocs at %lx\n", alloc_bottom);
2712
2713 /* Now try to claim that */
2714 mem_start = (unsigned long)alloc_up(room, PAGE_SIZE);
2715 if (mem_start == 0)
2716 prom_panic("Can't allocate initial device-tree chunk\n");
2717 mem_end = mem_start + room;
2718
2719 /* Get root of tree */
2720 root = call_prom("peer", 1, 1, (phandle)0);
2721 if (root == (phandle)0)
2722 prom_panic ("couldn't get device tree root\n");
2723
2724 /* Build header and make room for mem rsv map */
2725 mem_start = ALIGN(mem_start, 4);
2726 hdr = make_room(&mem_start, &mem_end,
2727 sizeof(struct boot_param_header), 4);
2728 dt_header_start = (unsigned long)hdr;
2729 rsvmap = make_room(&mem_start, &mem_end, sizeof(mem_reserve_map), 8);
2730
2731 /* Start of strings */
2732 mem_start = PAGE_ALIGN(mem_start);
2733 dt_string_start = mem_start;
2734 mem_start += 4; /* hole */
2735
2736 /* Add "phandle" in there, we'll need it */
2737 namep = make_room(&mem_start, &mem_end, 16, 1);
2738 prom_strscpy_pad(namep, "phandle", sizeof("phandle"));
2739 mem_start = (unsigned long)namep + prom_strlen(namep) + 1;
2740
2741 /* Build string array */
2742 prom_printf("Building dt strings...\n");
2743 scan_dt_build_strings(root, &mem_start, &mem_end);
2744 dt_string_end = mem_start;
2745
2746 /* Build structure */
2747 mem_start = PAGE_ALIGN(mem_start);
2748 dt_struct_start = mem_start;
2749 prom_printf("Building dt structure...\n");
2750 scan_dt_build_struct(root, &mem_start, &mem_end);
2751 dt_push_token(OF_DT_END, &mem_start, &mem_end);
2752 dt_struct_end = PAGE_ALIGN(mem_start);
2753
2754 /* Finish header */
2755 hdr->boot_cpuid_phys = cpu_to_be32(prom.cpu);
2756 hdr->magic = cpu_to_be32(OF_DT_HEADER);
2757 hdr->totalsize = cpu_to_be32(dt_struct_end - dt_header_start);
2758 hdr->off_dt_struct = cpu_to_be32(dt_struct_start - dt_header_start);
2759 hdr->off_dt_strings = cpu_to_be32(dt_string_start - dt_header_start);
2760 hdr->dt_strings_size = cpu_to_be32(dt_string_end - dt_string_start);
2761 hdr->off_mem_rsvmap = cpu_to_be32(((unsigned long)rsvmap) - dt_header_start);
2762 hdr->version = cpu_to_be32(OF_DT_VERSION);
2763 /* Version 16 is not backward compatible */
2764 hdr->last_comp_version = cpu_to_be32(0x10);
2765
2766 /* Copy the reserve map in */
2767 memcpy(rsvmap, mem_reserve_map, sizeof(mem_reserve_map));
2768
2769#ifdef DEBUG_PROM
2770 {
2771 int i;
2772 prom_printf("reserved memory map:\n");
2773 for (i = 0; i < mem_reserve_cnt; i++)
2774 prom_printf(" %llx - %llx\n",
2775 be64_to_cpu(mem_reserve_map[i].base),
2776 be64_to_cpu(mem_reserve_map[i].size));
2777 }
2778#endif
2779 /* Bump mem_reserve_cnt to cause further reservations to fail
2780 * since it's too late.
2781 */
2782 mem_reserve_cnt = MEM_RESERVE_MAP_SIZE;
2783
2784 prom_printf("Device tree strings 0x%lx -> 0x%lx\n",
2785 dt_string_start, dt_string_end);
2786 prom_printf("Device tree struct 0x%lx -> 0x%lx\n",
2787 dt_struct_start, dt_struct_end);
2788}
2789
2790#ifdef CONFIG_PPC_MAPLE
2791/* PIBS Version 1.05.0000 04/26/2005 has an incorrect /ht/isa/ranges property.
2792 * The values are bad, and it doesn't even have the right number of cells. */
2793static void __init fixup_device_tree_maple(void)
2794{
2795 phandle isa;
2796 u32 rloc = 0x01002000; /* IO space; PCI device = 4 */
2797 u32 isa_ranges[6];
2798 char *name;
2799
2800 name = "/ht@0/isa@4";
2801 isa = call_prom("finddevice", 1, 1, ADDR(name));
2802 if (!PHANDLE_VALID(isa)) {
2803 name = "/ht@0/isa@6";
2804 isa = call_prom("finddevice", 1, 1, ADDR(name));
2805 rloc = 0x01003000; /* IO space; PCI device = 6 */
2806 }
2807 if (!PHANDLE_VALID(isa))
2808 return;
2809
2810 if (prom_getproplen(isa, "ranges") != 12)
2811 return;
2812 if (prom_getprop(isa, "ranges", isa_ranges, sizeof(isa_ranges))
2813 == PROM_ERROR)
2814 return;
2815
2816 if (isa_ranges[0] != 0x1 ||
2817 isa_ranges[1] != 0xf4000000 ||
2818 isa_ranges[2] != 0x00010000)
2819 return;
2820
2821 prom_printf("Fixing up bogus ISA range on Maple/Apache...\n");
2822
2823 isa_ranges[0] = 0x1;
2824 isa_ranges[1] = 0x0;
2825 isa_ranges[2] = rloc;
2826 isa_ranges[3] = 0x0;
2827 isa_ranges[4] = 0x0;
2828 isa_ranges[5] = 0x00010000;
2829 prom_setprop(isa, name, "ranges",
2830 isa_ranges, sizeof(isa_ranges));
2831}
2832
2833#define CPC925_MC_START 0xf8000000
2834#define CPC925_MC_LENGTH 0x1000000
2835/* The values for memory-controller don't have right number of cells */
2836static void __init fixup_device_tree_maple_memory_controller(void)
2837{
2838 phandle mc;
2839 u32 mc_reg[4];
2840 char *name = "/hostbridge@f8000000";
2841 u32 ac, sc;
2842
2843 mc = call_prom("finddevice", 1, 1, ADDR(name));
2844 if (!PHANDLE_VALID(mc))
2845 return;
2846
2847 if (prom_getproplen(mc, "reg") != 8)
2848 return;
2849
2850 prom_getprop(prom.root, "#address-cells", &ac, sizeof(ac));
2851 prom_getprop(prom.root, "#size-cells", &sc, sizeof(sc));
2852 if ((ac != 2) || (sc != 2))
2853 return;
2854
2855 if (prom_getprop(mc, "reg", mc_reg, sizeof(mc_reg)) == PROM_ERROR)
2856 return;
2857
2858 if (mc_reg[0] != CPC925_MC_START || mc_reg[1] != CPC925_MC_LENGTH)
2859 return;
2860
2861 prom_printf("Fixing up bogus hostbridge on Maple...\n");
2862
2863 mc_reg[0] = 0x0;
2864 mc_reg[1] = CPC925_MC_START;
2865 mc_reg[2] = 0x0;
2866 mc_reg[3] = CPC925_MC_LENGTH;
2867 prom_setprop(mc, name, "reg", mc_reg, sizeof(mc_reg));
2868}
2869#else
2870#define fixup_device_tree_maple()
2871#define fixup_device_tree_maple_memory_controller()
2872#endif
2873
2874#ifdef CONFIG_PPC_CHRP
2875/*
2876 * Pegasos and BriQ lacks the "ranges" property in the isa node
2877 * Pegasos needs decimal IRQ 14/15, not hexadecimal
2878 * Pegasos has the IDE configured in legacy mode, but advertised as native
2879 */
2880static void __init fixup_device_tree_chrp(void)
2881{
2882 phandle ph;
2883 u32 prop[6];
2884 u32 rloc = 0x01006000; /* IO space; PCI device = 12 */
2885 char *name;
2886 int rc;
2887
2888 name = "/pci@80000000/isa@c";
2889 ph = call_prom("finddevice", 1, 1, ADDR(name));
2890 if (!PHANDLE_VALID(ph)) {
2891 name = "/pci@ff500000/isa@6";
2892 ph = call_prom("finddevice", 1, 1, ADDR(name));
2893 rloc = 0x01003000; /* IO space; PCI device = 6 */
2894 }
2895 if (PHANDLE_VALID(ph)) {
2896 rc = prom_getproplen(ph, "ranges");
2897 if (rc == 0 || rc == PROM_ERROR) {
2898 prom_printf("Fixing up missing ISA range on Pegasos...\n");
2899
2900 prop[0] = 0x1;
2901 prop[1] = 0x0;
2902 prop[2] = rloc;
2903 prop[3] = 0x0;
2904 prop[4] = 0x0;
2905 prop[5] = 0x00010000;
2906 prom_setprop(ph, name, "ranges", prop, sizeof(prop));
2907 }
2908 }
2909
2910 name = "/pci@80000000/ide@C,1";
2911 ph = call_prom("finddevice", 1, 1, ADDR(name));
2912 if (PHANDLE_VALID(ph)) {
2913 prom_printf("Fixing up IDE interrupt on Pegasos...\n");
2914 prop[0] = 14;
2915 prop[1] = 0x0;
2916 prom_setprop(ph, name, "interrupts", prop, 2*sizeof(u32));
2917 prom_printf("Fixing up IDE class-code on Pegasos...\n");
2918 rc = prom_getprop(ph, "class-code", prop, sizeof(u32));
2919 if (rc == sizeof(u32)) {
2920 prop[0] &= ~0x5;
2921 prom_setprop(ph, name, "class-code", prop, sizeof(u32));
2922 }
2923 }
2924}
2925#else
2926#define fixup_device_tree_chrp()
2927#endif
2928
2929#if defined(CONFIG_PPC64) && defined(CONFIG_PPC_PMAC)
2930static void __init fixup_device_tree_pmac(void)
2931{
2932 phandle u3, i2c, mpic;
2933 u32 u3_rev;
2934 u32 interrupts[2];
2935 u32 parent;
2936
2937 /* Some G5s have a missing interrupt definition, fix it up here */
2938 u3 = call_prom("finddevice", 1, 1, ADDR("/u3@0,f8000000"));
2939 if (!PHANDLE_VALID(u3))
2940 return;
2941 i2c = call_prom("finddevice", 1, 1, ADDR("/u3@0,f8000000/i2c@f8001000"));
2942 if (!PHANDLE_VALID(i2c))
2943 return;
2944 mpic = call_prom("finddevice", 1, 1, ADDR("/u3@0,f8000000/mpic@f8040000"));
2945 if (!PHANDLE_VALID(mpic))
2946 return;
2947
2948 /* check if proper rev of u3 */
2949 if (prom_getprop(u3, "device-rev", &u3_rev, sizeof(u3_rev))
2950 == PROM_ERROR)
2951 return;
2952 if (u3_rev < 0x35 || u3_rev > 0x39)
2953 return;
2954 /* does it need fixup ? */
2955 if (prom_getproplen(i2c, "interrupts") > 0)
2956 return;
2957
2958 prom_printf("fixing up bogus interrupts for u3 i2c...\n");
2959
2960 /* interrupt on this revision of u3 is number 0 and level */
2961 interrupts[0] = 0;
2962 interrupts[1] = 1;
2963 prom_setprop(i2c, "/u3@0,f8000000/i2c@f8001000", "interrupts",
2964 &interrupts, sizeof(interrupts));
2965 parent = (u32)mpic;
2966 prom_setprop(i2c, "/u3@0,f8000000/i2c@f8001000", "interrupt-parent",
2967 &parent, sizeof(parent));
2968}
2969#else
2970#define fixup_device_tree_pmac()
2971#endif
2972
2973#ifdef CONFIG_PPC_EFIKA
2974/*
2975 * The MPC5200 FEC driver requires an phy-handle property to tell it how
2976 * to talk to the phy. If the phy-handle property is missing, then this
2977 * function is called to add the appropriate nodes and link it to the
2978 * ethernet node.
2979 */
2980static void __init fixup_device_tree_efika_add_phy(void)
2981{
2982 u32 node;
2983 char prop[64];
2984 int rv;
2985
2986 /* Check if /builtin/ethernet exists - bail if it doesn't */
2987 node = call_prom("finddevice", 1, 1, ADDR("/builtin/ethernet"));
2988 if (!PHANDLE_VALID(node))
2989 return;
2990
2991 /* Check if the phy-handle property exists - bail if it does */
2992 rv = prom_getprop(node, "phy-handle", prop, sizeof(prop));
2993 if (!rv)
2994 return;
2995
2996 /*
2997 * At this point the ethernet device doesn't have a phy described.
2998 * Now we need to add the missing phy node and linkage
2999 */
3000
3001 /* Check for an MDIO bus node - if missing then create one */
3002 node = call_prom("finddevice", 1, 1, ADDR("/builtin/mdio"));
3003 if (!PHANDLE_VALID(node)) {
3004 prom_printf("Adding Ethernet MDIO node\n");
3005 call_prom("interpret", 1, 1,
3006 " s\" /builtin\" find-device"
3007 " new-device"
3008 " 1 encode-int s\" #address-cells\" property"
3009 " 0 encode-int s\" #size-cells\" property"
3010 " s\" mdio\" device-name"
3011 " s\" fsl,mpc5200b-mdio\" encode-string"
3012 " s\" compatible\" property"
3013 " 0xf0003000 0x400 reg"
3014 " 0x2 encode-int"
3015 " 0x5 encode-int encode+"
3016 " 0x3 encode-int encode+"
3017 " s\" interrupts\" property"
3018 " finish-device");
3019 }
3020
3021 /* Check for a PHY device node - if missing then create one and
3022 * give it's phandle to the ethernet node */
3023 node = call_prom("finddevice", 1, 1,
3024 ADDR("/builtin/mdio/ethernet-phy"));
3025 if (!PHANDLE_VALID(node)) {
3026 prom_printf("Adding Ethernet PHY node\n");
3027 call_prom("interpret", 1, 1,
3028 " s\" /builtin/mdio\" find-device"
3029 " new-device"
3030 " s\" ethernet-phy\" device-name"
3031 " 0x10 encode-int s\" reg\" property"
3032 " my-self"
3033 " ihandle>phandle"
3034 " finish-device"
3035 " s\" /builtin/ethernet\" find-device"
3036 " encode-int"
3037 " s\" phy-handle\" property"
3038 " device-end");
3039 }
3040}
3041
3042static void __init fixup_device_tree_efika(void)
3043{
3044 int sound_irq[3] = { 2, 2, 0 };
3045 int bcomm_irq[3*16] = { 3,0,0, 3,1,0, 3,2,0, 3,3,0,
3046 3,4,0, 3,5,0, 3,6,0, 3,7,0,
3047 3,8,0, 3,9,0, 3,10,0, 3,11,0,
3048 3,12,0, 3,13,0, 3,14,0, 3,15,0 };
3049 u32 node;
3050 char prop[64];
3051 int rv, len;
3052
3053 /* Check if we're really running on a EFIKA */
3054 node = call_prom("finddevice", 1, 1, ADDR("/"));
3055 if (!PHANDLE_VALID(node))
3056 return;
3057
3058 rv = prom_getprop(node, "model", prop, sizeof(prop));
3059 if (rv == PROM_ERROR)
3060 return;
3061 if (prom_strcmp(prop, "EFIKA5K2"))
3062 return;
3063
3064 prom_printf("Applying EFIKA device tree fixups\n");
3065
3066 /* Claiming to be 'chrp' is death */
3067 node = call_prom("finddevice", 1, 1, ADDR("/"));
3068 rv = prom_getprop(node, "device_type", prop, sizeof(prop));
3069 if (rv != PROM_ERROR && (prom_strcmp(prop, "chrp") == 0))
3070 prom_setprop(node, "/", "device_type", "efika", sizeof("efika"));
3071
3072 /* CODEGEN,description is exposed in /proc/cpuinfo so
3073 fix that too */
3074 rv = prom_getprop(node, "CODEGEN,description", prop, sizeof(prop));
3075 if (rv != PROM_ERROR && (prom_strstr(prop, "CHRP")))
3076 prom_setprop(node, "/", "CODEGEN,description",
3077 "Efika 5200B PowerPC System",
3078 sizeof("Efika 5200B PowerPC System"));
3079
3080 /* Fixup bestcomm interrupts property */
3081 node = call_prom("finddevice", 1, 1, ADDR("/builtin/bestcomm"));
3082 if (PHANDLE_VALID(node)) {
3083 len = prom_getproplen(node, "interrupts");
3084 if (len == 12) {
3085 prom_printf("Fixing bestcomm interrupts property\n");
3086 prom_setprop(node, "/builtin/bestcom", "interrupts",
3087 bcomm_irq, sizeof(bcomm_irq));
3088 }
3089 }
3090
3091 /* Fixup sound interrupts property */
3092 node = call_prom("finddevice", 1, 1, ADDR("/builtin/sound"));
3093 if (PHANDLE_VALID(node)) {
3094 rv = prom_getprop(node, "interrupts", prop, sizeof(prop));
3095 if (rv == PROM_ERROR) {
3096 prom_printf("Adding sound interrupts property\n");
3097 prom_setprop(node, "/builtin/sound", "interrupts",
3098 sound_irq, sizeof(sound_irq));
3099 }
3100 }
3101
3102 /* Make sure ethernet phy-handle property exists */
3103 fixup_device_tree_efika_add_phy();
3104}
3105#else
3106#define fixup_device_tree_efika()
3107#endif
3108
3109#ifdef CONFIG_PPC_PASEMI_NEMO
3110/*
3111 * CFE supplied on Nemo is broken in several ways, biggest
3112 * problem is that it reassigns ISA interrupts to unused mpic ints.
3113 * Add an interrupt-controller property for the io-bridge to use
3114 * and correct the ints so we can attach them to an irq_domain
3115 */
3116static void __init fixup_device_tree_pasemi(void)
3117{
3118 u32 interrupts[2], parent, rval, val = 0;
3119 char *name, *pci_name;
3120 phandle iob, node;
3121
3122 /* Find the root pci node */
3123 name = "/pxp@0,e0000000";
3124 iob = call_prom("finddevice", 1, 1, ADDR(name));
3125 if (!PHANDLE_VALID(iob))
3126 return;
3127
3128 /* check if interrupt-controller node set yet */
3129 if (prom_getproplen(iob, "interrupt-controller") !=PROM_ERROR)
3130 return;
3131
3132 prom_printf("adding interrupt-controller property for SB600...\n");
3133
3134 prom_setprop(iob, name, "interrupt-controller", &val, 0);
3135
3136 pci_name = "/pxp@0,e0000000/pci@11";
3137 node = call_prom("finddevice", 1, 1, ADDR(pci_name));
3138 parent = ADDR(iob);
3139
3140 for( ; prom_next_node(&node); ) {
3141 /* scan each node for one with an interrupt */
3142 if (!PHANDLE_VALID(node))
3143 continue;
3144
3145 rval = prom_getproplen(node, "interrupts");
3146 if (rval == 0 || rval == PROM_ERROR)
3147 continue;
3148
3149 prom_getprop(node, "interrupts", &interrupts, sizeof(interrupts));
3150 if ((interrupts[0] < 212) || (interrupts[0] > 222))
3151 continue;
3152
3153 /* found a node, update both interrupts and interrupt-parent */
3154 if ((interrupts[0] >= 212) && (interrupts[0] <= 215))
3155 interrupts[0] -= 203;
3156 if ((interrupts[0] >= 216) && (interrupts[0] <= 220))
3157 interrupts[0] -= 213;
3158 if (interrupts[0] == 221)
3159 interrupts[0] = 14;
3160 if (interrupts[0] == 222)
3161 interrupts[0] = 8;
3162
3163 prom_setprop(node, pci_name, "interrupts", interrupts,
3164 sizeof(interrupts));
3165 prom_setprop(node, pci_name, "interrupt-parent", &parent,
3166 sizeof(parent));
3167 }
3168
3169 /*
3170 * The io-bridge has device_type set to 'io-bridge' change it to 'isa'
3171 * so that generic isa-bridge code can add the SB600 and its on-board
3172 * peripherals.
3173 */
3174 name = "/pxp@0,e0000000/io-bridge@0";
3175 iob = call_prom("finddevice", 1, 1, ADDR(name));
3176 if (!PHANDLE_VALID(iob))
3177 return;
3178
3179 /* device_type is already set, just change it. */
3180
3181 prom_printf("Changing device_type of SB600 node...\n");
3182
3183 prom_setprop(iob, name, "device_type", "isa", sizeof("isa"));
3184}
3185#else /* !CONFIG_PPC_PASEMI_NEMO */
3186static inline void fixup_device_tree_pasemi(void) { }
3187#endif
3188
3189static void __init fixup_device_tree(void)
3190{
3191 fixup_device_tree_maple();
3192 fixup_device_tree_maple_memory_controller();
3193 fixup_device_tree_chrp();
3194 fixup_device_tree_pmac();
3195 fixup_device_tree_efika();
3196 fixup_device_tree_pasemi();
3197}
3198
3199static void __init prom_find_boot_cpu(void)
3200{
3201 __be32 rval;
3202 ihandle prom_cpu;
3203 phandle cpu_pkg;
3204
3205 rval = 0;
3206 if (prom_getprop(prom.chosen, "cpu", &rval, sizeof(rval)) <= 0)
3207 return;
3208 prom_cpu = be32_to_cpu(rval);
3209
3210 cpu_pkg = call_prom("instance-to-package", 1, 1, prom_cpu);
3211
3212 if (!PHANDLE_VALID(cpu_pkg))
3213 return;
3214
3215 prom_getprop(cpu_pkg, "reg", &rval, sizeof(rval));
3216 prom.cpu = be32_to_cpu(rval);
3217
3218 prom_debug("Booting CPU hw index = %d\n", prom.cpu);
3219}
3220
3221static void __init prom_check_initrd(unsigned long r3, unsigned long r4)
3222{
3223#ifdef CONFIG_BLK_DEV_INITRD
3224 if (r3 && r4 && r4 != 0xdeadbeef) {
3225 __be64 val;
3226
3227 prom_initrd_start = is_kernel_addr(r3) ? __pa(r3) : r3;
3228 prom_initrd_end = prom_initrd_start + r4;
3229
3230 val = cpu_to_be64(prom_initrd_start);
3231 prom_setprop(prom.chosen, "/chosen", "linux,initrd-start",
3232 &val, sizeof(val));
3233 val = cpu_to_be64(prom_initrd_end);
3234 prom_setprop(prom.chosen, "/chosen", "linux,initrd-end",
3235 &val, sizeof(val));
3236
3237 reserve_mem(prom_initrd_start,
3238 prom_initrd_end - prom_initrd_start);
3239
3240 prom_debug("initrd_start=0x%lx\n", prom_initrd_start);
3241 prom_debug("initrd_end=0x%lx\n", prom_initrd_end);
3242 }
3243#endif /* CONFIG_BLK_DEV_INITRD */
3244}
3245
3246#ifdef CONFIG_PPC_SVM
3247/*
3248 * Perform the Enter Secure Mode ultracall.
3249 */
3250static int enter_secure_mode(unsigned long kbase, unsigned long fdt)
3251{
3252 register unsigned long r3 asm("r3") = UV_ESM;
3253 register unsigned long r4 asm("r4") = kbase;
3254 register unsigned long r5 asm("r5") = fdt;
3255
3256 asm volatile("sc 2" : "+r"(r3) : "r"(r4), "r"(r5));
3257
3258 return r3;
3259}
3260
3261/*
3262 * Call the Ultravisor to transfer us to secure memory if we have an ESM blob.
3263 */
3264static void __init setup_secure_guest(unsigned long kbase, unsigned long fdt)
3265{
3266 int ret;
3267
3268 if (!prom_svm_enable)
3269 return;
3270
3271 /* Switch to secure mode. */
3272 prom_printf("Switching to secure mode.\n");
3273
3274 /*
3275 * The ultravisor will do an integrity check of the kernel image but we
3276 * relocated it so the check will fail. Restore the original image by
3277 * relocating it back to the kernel virtual base address.
3278 */
3279 relocate(KERNELBASE);
3280
3281 ret = enter_secure_mode(kbase, fdt);
3282
3283 /* Relocate the kernel again. */
3284 relocate(kbase);
3285
3286 if (ret != U_SUCCESS) {
3287 prom_printf("Returned %d from switching to secure mode.\n", ret);
3288 prom_rtas_os_term("Switch to secure mode failed.\n");
3289 }
3290}
3291#else
3292static void __init setup_secure_guest(unsigned long kbase, unsigned long fdt)
3293{
3294}
3295#endif /* CONFIG_PPC_SVM */
3296
3297/*
3298 * We enter here early on, when the Open Firmware prom is still
3299 * handling exceptions and the MMU hash table for us.
3300 */
3301
3302unsigned long __init prom_init(unsigned long r3, unsigned long r4,
3303 unsigned long pp,
3304 unsigned long r6, unsigned long r7,
3305 unsigned long kbase)
3306{
3307 unsigned long hdr;
3308
3309#ifdef CONFIG_PPC32
3310 unsigned long offset = reloc_offset();
3311 reloc_got2(offset);
3312#endif
3313
3314 /*
3315 * First zero the BSS
3316 */
3317 memset(&__bss_start, 0, __bss_stop - __bss_start);
3318
3319 /*
3320 * Init interface to Open Firmware, get some node references,
3321 * like /chosen
3322 */
3323 prom_init_client_services(pp);
3324
3325 /*
3326 * See if this OF is old enough that we need to do explicit maps
3327 * and other workarounds
3328 */
3329 prom_find_mmu();
3330
3331 /*
3332 * Init prom stdout device
3333 */
3334 prom_init_stdout();
3335
3336 prom_printf("Preparing to boot %s", linux_banner);
3337
3338 /*
3339 * Get default machine type. At this point, we do not differentiate
3340 * between pSeries SMP and pSeries LPAR
3341 */
3342 of_platform = prom_find_machine_type();
3343 prom_printf("Detected machine type: %x\n", of_platform);
3344
3345#ifndef CONFIG_NONSTATIC_KERNEL
3346 /* Bail if this is a kdump kernel. */
3347 if (PHYSICAL_START > 0)
3348 prom_panic("Error: You can't boot a kdump kernel from OF!\n");
3349#endif
3350
3351 /*
3352 * Check for an initrd
3353 */
3354 prom_check_initrd(r3, r4);
3355
3356 /*
3357 * Do early parsing of command line
3358 */
3359 early_cmdline_parse();
3360
3361#ifdef CONFIG_PPC_PSERIES
3362 /*
3363 * On pSeries, inform the firmware about our capabilities
3364 */
3365 if (of_platform == PLATFORM_PSERIES ||
3366 of_platform == PLATFORM_PSERIES_LPAR)
3367 prom_send_capabilities();
3368#endif
3369
3370 /*
3371 * Copy the CPU hold code
3372 */
3373 if (of_platform != PLATFORM_POWERMAC)
3374 copy_and_flush(0, kbase, 0x100, 0);
3375
3376 /*
3377 * Initialize memory management within prom_init
3378 */
3379 prom_init_mem();
3380
3381 /*
3382 * Determine which cpu is actually running right _now_
3383 */
3384 prom_find_boot_cpu();
3385
3386 /*
3387 * Initialize display devices
3388 */
3389 prom_check_displays();
3390
3391#if defined(CONFIG_PPC64) && defined(__BIG_ENDIAN__)
3392 /*
3393 * Initialize IOMMU (TCE tables) on pSeries. Do that before anything else
3394 * that uses the allocator, we need to make sure we get the top of memory
3395 * available for us here...
3396 */
3397 if (of_platform == PLATFORM_PSERIES)
3398 prom_initialize_tce_table();
3399#endif
3400
3401 /*
3402 * On non-powermacs, try to instantiate RTAS. PowerMacs don't
3403 * have a usable RTAS implementation.
3404 */
3405 if (of_platform != PLATFORM_POWERMAC)
3406 prom_instantiate_rtas();
3407
3408#ifdef CONFIG_PPC64
3409 /* instantiate sml */
3410 prom_instantiate_sml();
3411#endif
3412
3413 /*
3414 * On non-powermacs, put all CPUs in spin-loops.
3415 *
3416 * PowerMacs use a different mechanism to spin CPUs
3417 *
3418 * (This must be done after instanciating RTAS)
3419 */
3420 if (of_platform != PLATFORM_POWERMAC)
3421 prom_hold_cpus();
3422
3423 /*
3424 * Fill in some infos for use by the kernel later on
3425 */
3426 if (prom_memory_limit) {
3427 __be64 val = cpu_to_be64(prom_memory_limit);
3428 prom_setprop(prom.chosen, "/chosen", "linux,memory-limit",
3429 &val, sizeof(val));
3430 }
3431#ifdef CONFIG_PPC64
3432 if (prom_iommu_off)
3433 prom_setprop(prom.chosen, "/chosen", "linux,iommu-off",
3434 NULL, 0);
3435
3436 if (prom_iommu_force_on)
3437 prom_setprop(prom.chosen, "/chosen", "linux,iommu-force-on",
3438 NULL, 0);
3439
3440 if (prom_tce_alloc_start) {
3441 prom_setprop(prom.chosen, "/chosen", "linux,tce-alloc-start",
3442 &prom_tce_alloc_start,
3443 sizeof(prom_tce_alloc_start));
3444 prom_setprop(prom.chosen, "/chosen", "linux,tce-alloc-end",
3445 &prom_tce_alloc_end,
3446 sizeof(prom_tce_alloc_end));
3447 }
3448#endif
3449
3450 /*
3451 * Fixup any known bugs in the device-tree
3452 */
3453 fixup_device_tree();
3454
3455 /*
3456 * Now finally create the flattened device-tree
3457 */
3458 prom_printf("copying OF device tree...\n");
3459 flatten_device_tree();
3460
3461 /*
3462 * in case stdin is USB and still active on IBM machines...
3463 * Unfortunately quiesce crashes on some powermacs if we have
3464 * closed stdin already (in particular the powerbook 101).
3465 */
3466 if (of_platform != PLATFORM_POWERMAC)
3467 prom_close_stdin();
3468
3469 /*
3470 * Call OF "quiesce" method to shut down pending DMA's from
3471 * devices etc...
3472 */
3473 prom_printf("Quiescing Open Firmware ...\n");
3474 call_prom("quiesce", 0, 0);
3475
3476 /*
3477 * And finally, call the kernel passing it the flattened device
3478 * tree and NULL as r5, thus triggering the new entry point which
3479 * is common to us and kexec
3480 */
3481 hdr = dt_header_start;
3482
3483 prom_printf("Booting Linux via __start() @ 0x%lx ...\n", kbase);
3484 prom_debug("->dt_header_start=0x%lx\n", hdr);
3485
3486#ifdef CONFIG_PPC32
3487 reloc_got2(-offset);
3488#endif
3489
3490 /* Move to secure memory if we're supposed to be secure guests. */
3491 setup_secure_guest(kbase, hdr);
3492
3493 __start(hdr, kbase, 0, 0, 0, 0, 0);
3494
3495 return 0;
3496}
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 * On ppc64 we have to relocate the references explicitly with
70 * RELOC. (Note that strings count as static variables.)
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#ifdef CONFIG_PPC64
83#define RELOC(x) (*PTRRELOC(&(x)))
84#define ADDR(x) (u32) add_reloc_offset((unsigned long)(x))
85#define OF_WORKAROUNDS 0
86#else
87#define RELOC(x) (x)
88#define ADDR(x) (u32) (x)
89#define OF_WORKAROUNDS of_workarounds
90int of_workarounds;
91#endif
92
93#define OF_WA_CLAIM 1 /* do phys/virt claim separately, then map */
94#define OF_WA_LONGTRAIL 2 /* work around longtrail bugs */
95
96#define PROM_BUG() do { \
97 prom_printf("kernel BUG at %s line 0x%x!\n", \
98 RELOC(__FILE__), __LINE__); \
99 __asm__ __volatile__(".long " BUG_ILLEGAL_INSTR); \
100} while (0)
101
102#ifdef DEBUG_PROM
103#define prom_debug(x...) prom_printf(x)
104#else
105#define prom_debug(x...)
106#endif
107
108
109typedef u32 prom_arg_t;
110
111struct prom_args {
112 u32 service;
113 u32 nargs;
114 u32 nret;
115 prom_arg_t args[10];
116};
117
118struct prom_t {
119 ihandle root;
120 phandle chosen;
121 int cpu;
122 ihandle stdout;
123 ihandle mmumap;
124 ihandle memory;
125};
126
127struct mem_map_entry {
128 u64 base;
129 u64 size;
130};
131
132typedef u32 cell_t;
133
134extern void __start(unsigned long r3, unsigned long r4, unsigned long r5,
135 unsigned long r6, unsigned long r7, unsigned long r8,
136 unsigned long r9);
137
138#ifdef CONFIG_PPC64
139extern int enter_prom(struct prom_args *args, unsigned long entry);
140#else
141static inline int enter_prom(struct prom_args *args, unsigned long entry)
142{
143 return ((int (*)(struct prom_args *))entry)(args);
144}
145#endif
146
147extern void copy_and_flush(unsigned long dest, unsigned long src,
148 unsigned long size, unsigned long offset);
149
150/* prom structure */
151static struct prom_t __initdata prom;
152
153static unsigned long prom_entry __initdata;
154
155#define PROM_SCRATCH_SIZE 256
156
157static char __initdata of_stdout_device[256];
158static char __initdata prom_scratch[PROM_SCRATCH_SIZE];
159
160static unsigned long __initdata dt_header_start;
161static unsigned long __initdata dt_struct_start, dt_struct_end;
162static unsigned long __initdata dt_string_start, dt_string_end;
163
164static unsigned long __initdata prom_initrd_start, prom_initrd_end;
165
166#ifdef CONFIG_PPC64
167static int __initdata prom_iommu_force_on;
168static int __initdata prom_iommu_off;
169static unsigned long __initdata prom_tce_alloc_start;
170static unsigned long __initdata prom_tce_alloc_end;
171#endif
172
173/* Platforms codes are now obsolete in the kernel. Now only used within this
174 * file and ultimately gone too. Feel free to change them if you need, they
175 * are not shared with anything outside of this file anymore
176 */
177#define PLATFORM_PSERIES 0x0100
178#define PLATFORM_PSERIES_LPAR 0x0101
179#define PLATFORM_LPAR 0x0001
180#define PLATFORM_POWERMAC 0x0400
181#define PLATFORM_GENERIC 0x0500
182#define PLATFORM_OPAL 0x0600
183
184static int __initdata of_platform;
185
186static char __initdata prom_cmd_line[COMMAND_LINE_SIZE];
187
188static unsigned long __initdata prom_memory_limit;
189
190static unsigned long __initdata alloc_top;
191static unsigned long __initdata alloc_top_high;
192static unsigned long __initdata alloc_bottom;
193static unsigned long __initdata rmo_top;
194static unsigned long __initdata ram_top;
195
196static struct mem_map_entry __initdata mem_reserve_map[MEM_RESERVE_MAP_SIZE];
197static int __initdata mem_reserve_cnt;
198
199static cell_t __initdata regbuf[1024];
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 = ADDR(service);
225 args.nargs = nargs;
226 args.nret = nret;
227
228 va_start(list, nret);
229 for (i = 0; i < nargs; i++)
230 args.args[i] = 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, RELOC(prom_entry)) < 0)
237 return PROM_ERROR;
238
239 return (nret > 0) ? 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 = ADDR(service);
250 args.nargs = nargs;
251 args.nret = nret;
252
253 va_start(list, rets);
254 for (i = 0; i < nargs; i++)
255 args.args[i] = 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, RELOC(prom_entry)) < 0)
262 return PROM_ERROR;
263
264 if (rets != NULL)
265 for (i = 1; i < nret; ++i)
266 rets[i-1] = args.args[nargs+i];
267
268 return (nret > 0) ? args.args[nargs] : 0;
269}
270
271
272static void __init prom_print(const char *msg)
273{
274 const char *p, *q;
275 struct prom_t *_prom = &RELOC(prom);
276
277 if (_prom->stdout == 0)
278 return;
279
280 for (p = msg; *p != 0; p = q) {
281 for (q = p; *q != 0 && *q != '\n'; ++q)
282 ;
283 if (q > p)
284 call_prom("write", 3, 1, _prom->stdout, p, q - p);
285 if (*q == 0)
286 break;
287 ++q;
288 call_prom("write", 3, 1, _prom->stdout, ADDR("\r\n"), 2);
289 }
290}
291
292
293static void __init prom_print_hex(unsigned long val)
294{
295 int i, nibbles = sizeof(val)*2;
296 char buf[sizeof(val)*2+1];
297 struct prom_t *_prom = &RELOC(prom);
298
299 for (i = nibbles-1; i >= 0; i--) {
300 buf[i] = (val & 0xf) + '0';
301 if (buf[i] > '9')
302 buf[i] += ('a'-'0'-10);
303 val >>= 4;
304 }
305 buf[nibbles] = '\0';
306 call_prom("write", 3, 1, _prom->stdout, buf, nibbles);
307}
308
309/* max number of decimal digits in an unsigned long */
310#define UL_DIGITS 21
311static void __init prom_print_dec(unsigned long val)
312{
313 int i, size;
314 char buf[UL_DIGITS+1];
315 struct prom_t *_prom = &RELOC(prom);
316
317 for (i = UL_DIGITS-1; i >= 0; i--) {
318 buf[i] = (val % 10) + '0';
319 val = val/10;
320 if (val == 0)
321 break;
322 }
323 /* shift stuff down */
324 size = UL_DIGITS - i;
325 call_prom("write", 3, 1, _prom->stdout, buf+i, size);
326}
327
328static void __init prom_printf(const char *format, ...)
329{
330 const char *p, *q, *s;
331 va_list args;
332 unsigned long v;
333 long vs;
334 struct prom_t *_prom = &RELOC(prom);
335
336 va_start(args, format);
337#ifdef CONFIG_PPC64
338 format = PTRRELOC(format);
339#endif
340 for (p = format; *p != 0; p = q) {
341 for (q = p; *q != 0 && *q != '\n' && *q != '%'; ++q)
342 ;
343 if (q > p)
344 call_prom("write", 3, 1, _prom->stdout, p, q - p);
345 if (*q == 0)
346 break;
347 if (*q == '\n') {
348 ++q;
349 call_prom("write", 3, 1, _prom->stdout,
350 ADDR("\r\n"), 2);
351 continue;
352 }
353 ++q;
354 if (*q == 0)
355 break;
356 switch (*q) {
357 case 's':
358 ++q;
359 s = va_arg(args, const char *);
360 prom_print(s);
361 break;
362 case 'x':
363 ++q;
364 v = va_arg(args, unsigned long);
365 prom_print_hex(v);
366 break;
367 case 'd':
368 ++q;
369 vs = va_arg(args, int);
370 if (vs < 0) {
371 prom_print(RELOC("-"));
372 vs = -vs;
373 }
374 prom_print_dec(vs);
375 break;
376 case 'l':
377 ++q;
378 if (*q == 0)
379 break;
380 else if (*q == 'x') {
381 ++q;
382 v = va_arg(args, unsigned long);
383 prom_print_hex(v);
384 } else if (*q == 'u') { /* '%lu' */
385 ++q;
386 v = va_arg(args, unsigned long);
387 prom_print_dec(v);
388 } else if (*q == 'd') { /* %ld */
389 ++q;
390 vs = va_arg(args, long);
391 if (vs < 0) {
392 prom_print(RELOC("-"));
393 vs = -vs;
394 }
395 prom_print_dec(vs);
396 }
397 break;
398 }
399 }
400}
401
402
403static unsigned int __init prom_claim(unsigned long virt, unsigned long size,
404 unsigned long align)
405{
406 struct prom_t *_prom = &RELOC(prom);
407
408 if (align == 0 && (OF_WORKAROUNDS & OF_WA_CLAIM)) {
409 /*
410 * Old OF requires we claim physical and virtual separately
411 * and then map explicitly (assuming virtual mode)
412 */
413 int ret;
414 prom_arg_t result;
415
416 ret = call_prom_ret("call-method", 5, 2, &result,
417 ADDR("claim"), _prom->memory,
418 align, size, virt);
419 if (ret != 0 || result == -1)
420 return -1;
421 ret = call_prom_ret("call-method", 5, 2, &result,
422 ADDR("claim"), _prom->mmumap,
423 align, size, virt);
424 if (ret != 0) {
425 call_prom("call-method", 4, 1, ADDR("release"),
426 _prom->memory, size, virt);
427 return -1;
428 }
429 /* the 0x12 is M (coherence) + PP == read/write */
430 call_prom("call-method", 6, 1,
431 ADDR("map"), _prom->mmumap, 0x12, size, virt, virt);
432 return virt;
433 }
434 return call_prom("claim", 3, 1, (prom_arg_t)virt, (prom_arg_t)size,
435 (prom_arg_t)align);
436}
437
438static void __init __attribute__((noreturn)) prom_panic(const char *reason)
439{
440#ifdef CONFIG_PPC64
441 reason = PTRRELOC(reason);
442#endif
443 prom_print(reason);
444 /* Do not call exit because it clears the screen on pmac
445 * it also causes some sort of double-fault on early pmacs */
446 if (RELOC(of_platform) == PLATFORM_POWERMAC)
447 asm("trap\n");
448
449 /* ToDo: should put up an SRC here on pSeries */
450 call_prom("exit", 0, 0);
451
452 for (;;) /* should never get here */
453 ;
454}
455
456
457static int __init prom_next_node(phandle *nodep)
458{
459 phandle node;
460
461 if ((node = *nodep) != 0
462 && (*nodep = call_prom("child", 1, 1, node)) != 0)
463 return 1;
464 if ((*nodep = call_prom("peer", 1, 1, node)) != 0)
465 return 1;
466 for (;;) {
467 if ((node = call_prom("parent", 1, 1, node)) == 0)
468 return 0;
469 if ((*nodep = call_prom("peer", 1, 1, node)) != 0)
470 return 1;
471 }
472}
473
474static int inline prom_getprop(phandle node, const char *pname,
475 void *value, size_t valuelen)
476{
477 return call_prom("getprop", 4, 1, node, ADDR(pname),
478 (u32)(unsigned long) value, (u32) valuelen);
479}
480
481static int inline prom_getproplen(phandle node, const char *pname)
482{
483 return call_prom("getproplen", 2, 1, node, ADDR(pname));
484}
485
486static void add_string(char **str, const char *q)
487{
488 char *p = *str;
489
490 while (*q)
491 *p++ = *q++;
492 *p++ = ' ';
493 *str = p;
494}
495
496static char *tohex(unsigned int x)
497{
498 static char digits[] = "0123456789abcdef";
499 static char result[9];
500 int i;
501
502 result[8] = 0;
503 i = 8;
504 do {
505 --i;
506 result[i] = digits[x & 0xf];
507 x >>= 4;
508 } while (x != 0 && i > 0);
509 return &result[i];
510}
511
512static int __init prom_setprop(phandle node, const char *nodename,
513 const char *pname, void *value, size_t valuelen)
514{
515 char cmd[256], *p;
516
517 if (!(OF_WORKAROUNDS & OF_WA_LONGTRAIL))
518 return call_prom("setprop", 4, 1, node, ADDR(pname),
519 (u32)(unsigned long) value, (u32) valuelen);
520
521 /* gah... setprop doesn't work on longtrail, have to use interpret */
522 p = cmd;
523 add_string(&p, "dev");
524 add_string(&p, nodename);
525 add_string(&p, tohex((u32)(unsigned long) value));
526 add_string(&p, tohex(valuelen));
527 add_string(&p, tohex(ADDR(pname)));
528 add_string(&p, tohex(strlen(RELOC(pname))));
529 add_string(&p, "property");
530 *p = 0;
531 return call_prom("interpret", 1, 1, (u32)(unsigned long) cmd);
532}
533
534/* We can't use the standard versions because of RELOC headaches. */
535#define isxdigit(c) (('0' <= (c) && (c) <= '9') \
536 || ('a' <= (c) && (c) <= 'f') \
537 || ('A' <= (c) && (c) <= 'F'))
538
539#define isdigit(c) ('0' <= (c) && (c) <= '9')
540#define islower(c) ('a' <= (c) && (c) <= 'z')
541#define toupper(c) (islower(c) ? ((c) - 'a' + 'A') : (c))
542
543unsigned long prom_strtoul(const char *cp, const char **endp)
544{
545 unsigned long result = 0, base = 10, value;
546
547 if (*cp == '0') {
548 base = 8;
549 cp++;
550 if (toupper(*cp) == 'X') {
551 cp++;
552 base = 16;
553 }
554 }
555
556 while (isxdigit(*cp) &&
557 (value = isdigit(*cp) ? *cp - '0' : toupper(*cp) - 'A' + 10) < base) {
558 result = result * base + value;
559 cp++;
560 }
561
562 if (endp)
563 *endp = cp;
564
565 return result;
566}
567
568unsigned long prom_memparse(const char *ptr, const char **retptr)
569{
570 unsigned long ret = prom_strtoul(ptr, retptr);
571 int shift = 0;
572
573 /*
574 * We can't use a switch here because GCC *may* generate a
575 * jump table which won't work, because we're not running at
576 * the address we're linked at.
577 */
578 if ('G' == **retptr || 'g' == **retptr)
579 shift = 30;
580
581 if ('M' == **retptr || 'm' == **retptr)
582 shift = 20;
583
584 if ('K' == **retptr || 'k' == **retptr)
585 shift = 10;
586
587 if (shift) {
588 ret <<= shift;
589 (*retptr)++;
590 }
591
592 return ret;
593}
594
595/*
596 * Early parsing of the command line passed to the kernel, used for
597 * "mem=x" and the options that affect the iommu
598 */
599static void __init early_cmdline_parse(void)
600{
601 struct prom_t *_prom = &RELOC(prom);
602 const char *opt;
603
604 char *p;
605 int l = 0;
606
607 RELOC(prom_cmd_line[0]) = 0;
608 p = RELOC(prom_cmd_line);
609 if ((long)_prom->chosen > 0)
610 l = prom_getprop(_prom->chosen, "bootargs", p, COMMAND_LINE_SIZE-1);
611#ifdef CONFIG_CMDLINE
612 if (l <= 0 || p[0] == '\0') /* dbl check */
613 strlcpy(RELOC(prom_cmd_line),
614 RELOC(CONFIG_CMDLINE), sizeof(prom_cmd_line));
615#endif /* CONFIG_CMDLINE */
616 prom_printf("command line: %s\n", RELOC(prom_cmd_line));
617
618#ifdef CONFIG_PPC64
619 opt = strstr(RELOC(prom_cmd_line), RELOC("iommu="));
620 if (opt) {
621 prom_printf("iommu opt is: %s\n", opt);
622 opt += 6;
623 while (*opt && *opt == ' ')
624 opt++;
625 if (!strncmp(opt, RELOC("off"), 3))
626 RELOC(prom_iommu_off) = 1;
627 else if (!strncmp(opt, RELOC("force"), 5))
628 RELOC(prom_iommu_force_on) = 1;
629 }
630#endif
631 opt = strstr(RELOC(prom_cmd_line), RELOC("mem="));
632 if (opt) {
633 opt += 4;
634 RELOC(prom_memory_limit) = prom_memparse(opt, (const char **)&opt);
635#ifdef CONFIG_PPC64
636 /* Align to 16 MB == size of ppc64 large page */
637 RELOC(prom_memory_limit) = ALIGN(RELOC(prom_memory_limit), 0x1000000);
638#endif
639 }
640}
641
642#if defined(CONFIG_PPC_PSERIES) || defined(CONFIG_PPC_POWERNV)
643/*
644 * There are two methods for telling firmware what our capabilities are.
645 * Newer machines have an "ibm,client-architecture-support" method on the
646 * root node. For older machines, we have to call the "process-elf-header"
647 * method in the /packages/elf-loader node, passing it a fake 32-bit
648 * ELF header containing a couple of PT_NOTE sections that contain
649 * structures that contain various information.
650 */
651
652/*
653 * New method - extensible architecture description vector.
654 *
655 * Because the description vector contains a mix of byte and word
656 * values, we declare it as an unsigned char array, and use this
657 * macro to put word values in.
658 */
659#define W(x) ((x) >> 24) & 0xff, ((x) >> 16) & 0xff, \
660 ((x) >> 8) & 0xff, (x) & 0xff
661
662/* Option vector bits - generic bits in byte 1 */
663#define OV_IGNORE 0x80 /* ignore this vector */
664#define OV_CESSATION_POLICY 0x40 /* halt if unsupported option present*/
665
666/* Option vector 1: processor architectures supported */
667#define OV1_PPC_2_00 0x80 /* set if we support PowerPC 2.00 */
668#define OV1_PPC_2_01 0x40 /* set if we support PowerPC 2.01 */
669#define OV1_PPC_2_02 0x20 /* set if we support PowerPC 2.02 */
670#define OV1_PPC_2_03 0x10 /* set if we support PowerPC 2.03 */
671#define OV1_PPC_2_04 0x08 /* set if we support PowerPC 2.04 */
672#define OV1_PPC_2_05 0x04 /* set if we support PowerPC 2.05 */
673#define OV1_PPC_2_06 0x02 /* set if we support PowerPC 2.06 */
674
675/* Option vector 2: Open Firmware options supported */
676#define OV2_REAL_MODE 0x20 /* set if we want OF in real mode */
677
678/* Option vector 3: processor options supported */
679#define OV3_FP 0x80 /* floating point */
680#define OV3_VMX 0x40 /* VMX/Altivec */
681#define OV3_DFP 0x20 /* decimal FP */
682
683/* Option vector 4: IBM PAPR implementation */
684#define OV4_MIN_ENT_CAP 0x01 /* minimum VP entitled capacity */
685
686/* Option vector 5: PAPR/OF options supported */
687#define OV5_LPAR 0x80 /* logical partitioning supported */
688#define OV5_SPLPAR 0x40 /* shared-processor LPAR supported */
689/* ibm,dynamic-reconfiguration-memory property supported */
690#define OV5_DRCONF_MEMORY 0x20
691#define OV5_LARGE_PAGES 0x10 /* large pages supported */
692#define OV5_DONATE_DEDICATE_CPU 0x02 /* donate dedicated CPU support */
693/* PCIe/MSI support. Without MSI full PCIe is not supported */
694#ifdef CONFIG_PCI_MSI
695#define OV5_MSI 0x01 /* PCIe/MSI support */
696#else
697#define OV5_MSI 0x00
698#endif /* CONFIG_PCI_MSI */
699#ifdef CONFIG_PPC_SMLPAR
700#define OV5_CMO 0x80 /* Cooperative Memory Overcommitment */
701#define OV5_XCMO 0x40 /* Page Coalescing */
702#else
703#define OV5_CMO 0x00
704#define OV5_XCMO 0x00
705#endif
706#define OV5_TYPE1_AFFINITY 0x80 /* Type 1 NUMA affinity */
707#define OV5_PFO_HW_RNG 0x80 /* PFO Random Number Generator */
708#define OV5_PFO_HW_ENCR 0x20 /* PFO Encryption Accelerator */
709
710/* Option Vector 6: IBM PAPR hints */
711#define OV6_LINUX 0x02 /* Linux is our OS */
712
713/*
714 * The architecture vector has an array of PVR mask/value pairs,
715 * followed by # option vectors - 1, followed by the option vectors.
716 */
717static unsigned char ibm_architecture_vec[] = {
718 W(0xfffe0000), W(0x003a0000), /* POWER5/POWER5+ */
719 W(0xffff0000), W(0x003e0000), /* POWER6 */
720 W(0xffff0000), W(0x003f0000), /* POWER7 */
721 W(0xffffffff), W(0x0f000003), /* all 2.06-compliant */
722 W(0xffffffff), W(0x0f000002), /* all 2.05-compliant */
723 W(0xfffffffe), W(0x0f000001), /* all 2.04-compliant and earlier */
724 6 - 1, /* 6 option vectors */
725
726 /* option vector 1: processor architectures supported */
727 3 - 2, /* length */
728 0, /* don't ignore, don't halt */
729 OV1_PPC_2_00 | OV1_PPC_2_01 | OV1_PPC_2_02 | OV1_PPC_2_03 |
730 OV1_PPC_2_04 | OV1_PPC_2_05 | OV1_PPC_2_06,
731
732 /* option vector 2: Open Firmware options supported */
733 34 - 2, /* length */
734 OV2_REAL_MODE,
735 0, 0,
736 W(0xffffffff), /* real_base */
737 W(0xffffffff), /* real_size */
738 W(0xffffffff), /* virt_base */
739 W(0xffffffff), /* virt_size */
740 W(0xffffffff), /* load_base */
741 W(256), /* 256MB min RMA */
742 W(0xffffffff), /* full client load */
743 0, /* min RMA percentage of total RAM */
744 48, /* max log_2(hash table size) */
745
746 /* option vector 3: processor options supported */
747 3 - 2, /* length */
748 0, /* don't ignore, don't halt */
749 OV3_FP | OV3_VMX | OV3_DFP,
750
751 /* option vector 4: IBM PAPR implementation */
752 3 - 2, /* length */
753 0, /* don't halt */
754 OV4_MIN_ENT_CAP, /* minimum VP entitled capacity */
755
756 /* option vector 5: PAPR/OF options */
757 18 - 2, /* length */
758 0, /* don't ignore, don't halt */
759 OV5_LPAR | OV5_SPLPAR | OV5_LARGE_PAGES | OV5_DRCONF_MEMORY |
760 OV5_DONATE_DEDICATE_CPU | OV5_MSI,
761 0,
762 OV5_CMO | OV5_XCMO,
763 OV5_TYPE1_AFFINITY,
764 0,
765 0,
766 0,
767 /* WARNING: The offset of the "number of cores" field below
768 * must match by the macro below. Update the definition if
769 * the structure layout changes.
770 */
771#define IBM_ARCH_VEC_NRCORES_OFFSET 101
772 W(NR_CPUS), /* number of cores supported */
773 0,
774 0,
775 0,
776 0,
777 OV5_PFO_HW_RNG | OV5_PFO_HW_ENCR,
778
779 /* option vector 6: IBM PAPR hints */
780 4 - 2, /* length */
781 0,
782 0,
783 OV6_LINUX,
784
785};
786
787/* Old method - ELF header with PT_NOTE sections */
788static struct fake_elf {
789 Elf32_Ehdr elfhdr;
790 Elf32_Phdr phdr[2];
791 struct chrpnote {
792 u32 namesz;
793 u32 descsz;
794 u32 type;
795 char name[8]; /* "PowerPC" */
796 struct chrpdesc {
797 u32 real_mode;
798 u32 real_base;
799 u32 real_size;
800 u32 virt_base;
801 u32 virt_size;
802 u32 load_base;
803 } chrpdesc;
804 } chrpnote;
805 struct rpanote {
806 u32 namesz;
807 u32 descsz;
808 u32 type;
809 char name[24]; /* "IBM,RPA-Client-Config" */
810 struct rpadesc {
811 u32 lpar_affinity;
812 u32 min_rmo_size;
813 u32 min_rmo_percent;
814 u32 max_pft_size;
815 u32 splpar;
816 u32 min_load;
817 u32 new_mem_def;
818 u32 ignore_me;
819 } rpadesc;
820 } rpanote;
821} fake_elf = {
822 .elfhdr = {
823 .e_ident = { 0x7f, 'E', 'L', 'F',
824 ELFCLASS32, ELFDATA2MSB, EV_CURRENT },
825 .e_type = ET_EXEC, /* yeah right */
826 .e_machine = EM_PPC,
827 .e_version = EV_CURRENT,
828 .e_phoff = offsetof(struct fake_elf, phdr),
829 .e_phentsize = sizeof(Elf32_Phdr),
830 .e_phnum = 2
831 },
832 .phdr = {
833 [0] = {
834 .p_type = PT_NOTE,
835 .p_offset = offsetof(struct fake_elf, chrpnote),
836 .p_filesz = sizeof(struct chrpnote)
837 }, [1] = {
838 .p_type = PT_NOTE,
839 .p_offset = offsetof(struct fake_elf, rpanote),
840 .p_filesz = sizeof(struct rpanote)
841 }
842 },
843 .chrpnote = {
844 .namesz = sizeof("PowerPC"),
845 .descsz = sizeof(struct chrpdesc),
846 .type = 0x1275,
847 .name = "PowerPC",
848 .chrpdesc = {
849 .real_mode = ~0U, /* ~0 means "don't care" */
850 .real_base = ~0U,
851 .real_size = ~0U,
852 .virt_base = ~0U,
853 .virt_size = ~0U,
854 .load_base = ~0U
855 },
856 },
857 .rpanote = {
858 .namesz = sizeof("IBM,RPA-Client-Config"),
859 .descsz = sizeof(struct rpadesc),
860 .type = 0x12759999,
861 .name = "IBM,RPA-Client-Config",
862 .rpadesc = {
863 .lpar_affinity = 0,
864 .min_rmo_size = 64, /* in megabytes */
865 .min_rmo_percent = 0,
866 .max_pft_size = 48, /* 2^48 bytes max PFT size */
867 .splpar = 1,
868 .min_load = ~0U,
869 .new_mem_def = 0
870 }
871 }
872};
873
874static int __init prom_count_smt_threads(void)
875{
876 phandle node;
877 char type[64];
878 unsigned int plen;
879
880 /* Pick up th first CPU node we can find */
881 for (node = 0; prom_next_node(&node); ) {
882 type[0] = 0;
883 prom_getprop(node, "device_type", type, sizeof(type));
884
885 if (strcmp(type, RELOC("cpu")))
886 continue;
887 /*
888 * There is an entry for each smt thread, each entry being
889 * 4 bytes long. All cpus should have the same number of
890 * smt threads, so return after finding the first.
891 */
892 plen = prom_getproplen(node, "ibm,ppc-interrupt-server#s");
893 if (plen == PROM_ERROR)
894 break;
895 plen >>= 2;
896 prom_debug("Found %lu smt threads per core\n", (unsigned long)plen);
897
898 /* Sanity check */
899 if (plen < 1 || plen > 64) {
900 prom_printf("Threads per core %lu out of bounds, assuming 1\n",
901 (unsigned long)plen);
902 return 1;
903 }
904 return plen;
905 }
906 prom_debug("No threads found, assuming 1 per core\n");
907
908 return 1;
909
910}
911
912
913static void __init prom_send_capabilities(void)
914{
915 ihandle elfloader, root;
916 prom_arg_t ret;
917 u32 *cores;
918
919 root = call_prom("open", 1, 1, ADDR("/"));
920 if (root != 0) {
921 /* We need to tell the FW about the number of cores we support.
922 *
923 * To do that, we count the number of threads on the first core
924 * (we assume this is the same for all cores) and use it to
925 * divide NR_CPUS.
926 */
927 cores = (u32 *)PTRRELOC(&ibm_architecture_vec[IBM_ARCH_VEC_NRCORES_OFFSET]);
928 if (*cores != NR_CPUS) {
929 prom_printf("WARNING ! "
930 "ibm_architecture_vec structure inconsistent: %lu!\n",
931 *cores);
932 } else {
933 *cores = DIV_ROUND_UP(NR_CPUS, prom_count_smt_threads());
934 prom_printf("Max number of cores passed to firmware: %lu (NR_CPUS = %lu)\n",
935 *cores, NR_CPUS);
936 }
937
938 /* try calling the ibm,client-architecture-support method */
939 prom_printf("Calling ibm,client-architecture-support...");
940 if (call_prom_ret("call-method", 3, 2, &ret,
941 ADDR("ibm,client-architecture-support"),
942 root,
943 ADDR(ibm_architecture_vec)) == 0) {
944 /* the call exists... */
945 if (ret)
946 prom_printf("\nWARNING: ibm,client-architecture"
947 "-support call FAILED!\n");
948 call_prom("close", 1, 0, root);
949 prom_printf(" done\n");
950 return;
951 }
952 call_prom("close", 1, 0, root);
953 prom_printf(" not implemented\n");
954 }
955
956 /* no ibm,client-architecture-support call, try the old way */
957 elfloader = call_prom("open", 1, 1, ADDR("/packages/elf-loader"));
958 if (elfloader == 0) {
959 prom_printf("couldn't open /packages/elf-loader\n");
960 return;
961 }
962 call_prom("call-method", 3, 1, ADDR("process-elf-header"),
963 elfloader, ADDR(&fake_elf));
964 call_prom("close", 1, 0, elfloader);
965}
966#endif
967
968/*
969 * Memory allocation strategy... our layout is normally:
970 *
971 * at 14Mb or more we have vmlinux, then a gap and initrd. In some
972 * rare cases, initrd might end up being before the kernel though.
973 * We assume this won't override the final kernel at 0, we have no
974 * provision to handle that in this version, but it should hopefully
975 * never happen.
976 *
977 * alloc_top is set to the top of RMO, eventually shrink down if the
978 * TCEs overlap
979 *
980 * alloc_bottom is set to the top of kernel/initrd
981 *
982 * from there, allocations are done this way : rtas is allocated
983 * topmost, and the device-tree is allocated from the bottom. We try
984 * to grow the device-tree allocation as we progress. If we can't,
985 * then we fail, we don't currently have a facility to restart
986 * elsewhere, but that shouldn't be necessary.
987 *
988 * Note that calls to reserve_mem have to be done explicitly, memory
989 * allocated with either alloc_up or alloc_down isn't automatically
990 * reserved.
991 */
992
993
994/*
995 * Allocates memory in the RMO upward from the kernel/initrd
996 *
997 * When align is 0, this is a special case, it means to allocate in place
998 * at the current location of alloc_bottom or fail (that is basically
999 * extending the previous allocation). Used for the device-tree flattening
1000 */
1001static unsigned long __init alloc_up(unsigned long size, unsigned long align)
1002{
1003 unsigned long base = RELOC(alloc_bottom);
1004 unsigned long addr = 0;
1005
1006 if (align)
1007 base = _ALIGN_UP(base, align);
1008 prom_debug("alloc_up(%x, %x)\n", size, align);
1009 if (RELOC(ram_top) == 0)
1010 prom_panic("alloc_up() called with mem not initialized\n");
1011
1012 if (align)
1013 base = _ALIGN_UP(RELOC(alloc_bottom), align);
1014 else
1015 base = RELOC(alloc_bottom);
1016
1017 for(; (base + size) <= RELOC(alloc_top);
1018 base = _ALIGN_UP(base + 0x100000, align)) {
1019 prom_debug(" trying: 0x%x\n\r", base);
1020 addr = (unsigned long)prom_claim(base, size, 0);
1021 if (addr != PROM_ERROR && addr != 0)
1022 break;
1023 addr = 0;
1024 if (align == 0)
1025 break;
1026 }
1027 if (addr == 0)
1028 return 0;
1029 RELOC(alloc_bottom) = addr + size;
1030
1031 prom_debug(" -> %x\n", addr);
1032 prom_debug(" alloc_bottom : %x\n", RELOC(alloc_bottom));
1033 prom_debug(" alloc_top : %x\n", RELOC(alloc_top));
1034 prom_debug(" alloc_top_hi : %x\n", RELOC(alloc_top_high));
1035 prom_debug(" rmo_top : %x\n", RELOC(rmo_top));
1036 prom_debug(" ram_top : %x\n", RELOC(ram_top));
1037
1038 return addr;
1039}
1040
1041/*
1042 * Allocates memory downward, either from top of RMO, or if highmem
1043 * is set, from the top of RAM. Note that this one doesn't handle
1044 * failures. It does claim memory if highmem is not set.
1045 */
1046static unsigned long __init alloc_down(unsigned long size, unsigned long align,
1047 int highmem)
1048{
1049 unsigned long base, addr = 0;
1050
1051 prom_debug("alloc_down(%x, %x, %s)\n", size, align,
1052 highmem ? RELOC("(high)") : RELOC("(low)"));
1053 if (RELOC(ram_top) == 0)
1054 prom_panic("alloc_down() called with mem not initialized\n");
1055
1056 if (highmem) {
1057 /* Carve out storage for the TCE table. */
1058 addr = _ALIGN_DOWN(RELOC(alloc_top_high) - size, align);
1059 if (addr <= RELOC(alloc_bottom))
1060 return 0;
1061 /* Will we bump into the RMO ? If yes, check out that we
1062 * didn't overlap existing allocations there, if we did,
1063 * we are dead, we must be the first in town !
1064 */
1065 if (addr < RELOC(rmo_top)) {
1066 /* Good, we are first */
1067 if (RELOC(alloc_top) == RELOC(rmo_top))
1068 RELOC(alloc_top) = RELOC(rmo_top) = addr;
1069 else
1070 return 0;
1071 }
1072 RELOC(alloc_top_high) = addr;
1073 goto bail;
1074 }
1075
1076 base = _ALIGN_DOWN(RELOC(alloc_top) - size, align);
1077 for (; base > RELOC(alloc_bottom);
1078 base = _ALIGN_DOWN(base - 0x100000, align)) {
1079 prom_debug(" trying: 0x%x\n\r", base);
1080 addr = (unsigned long)prom_claim(base, size, 0);
1081 if (addr != PROM_ERROR && addr != 0)
1082 break;
1083 addr = 0;
1084 }
1085 if (addr == 0)
1086 return 0;
1087 RELOC(alloc_top) = addr;
1088
1089 bail:
1090 prom_debug(" -> %x\n", addr);
1091 prom_debug(" alloc_bottom : %x\n", RELOC(alloc_bottom));
1092 prom_debug(" alloc_top : %x\n", RELOC(alloc_top));
1093 prom_debug(" alloc_top_hi : %x\n", RELOC(alloc_top_high));
1094 prom_debug(" rmo_top : %x\n", RELOC(rmo_top));
1095 prom_debug(" ram_top : %x\n", RELOC(ram_top));
1096
1097 return addr;
1098}
1099
1100/*
1101 * Parse a "reg" cell
1102 */
1103static unsigned long __init prom_next_cell(int s, cell_t **cellp)
1104{
1105 cell_t *p = *cellp;
1106 unsigned long r = 0;
1107
1108 /* Ignore more than 2 cells */
1109 while (s > sizeof(unsigned long) / 4) {
1110 p++;
1111 s--;
1112 }
1113 r = *p++;
1114#ifdef CONFIG_PPC64
1115 if (s > 1) {
1116 r <<= 32;
1117 r |= *(p++);
1118 }
1119#endif
1120 *cellp = p;
1121 return r;
1122}
1123
1124/*
1125 * Very dumb function for adding to the memory reserve list, but
1126 * we don't need anything smarter at this point
1127 *
1128 * XXX Eventually check for collisions. They should NEVER happen.
1129 * If problems seem to show up, it would be a good start to track
1130 * them down.
1131 */
1132static void __init reserve_mem(u64 base, u64 size)
1133{
1134 u64 top = base + size;
1135 unsigned long cnt = RELOC(mem_reserve_cnt);
1136
1137 if (size == 0)
1138 return;
1139
1140 /* We need to always keep one empty entry so that we
1141 * have our terminator with "size" set to 0 since we are
1142 * dumb and just copy this entire array to the boot params
1143 */
1144 base = _ALIGN_DOWN(base, PAGE_SIZE);
1145 top = _ALIGN_UP(top, PAGE_SIZE);
1146 size = top - base;
1147
1148 if (cnt >= (MEM_RESERVE_MAP_SIZE - 1))
1149 prom_panic("Memory reserve map exhausted !\n");
1150 RELOC(mem_reserve_map)[cnt].base = base;
1151 RELOC(mem_reserve_map)[cnt].size = size;
1152 RELOC(mem_reserve_cnt) = cnt + 1;
1153}
1154
1155/*
1156 * Initialize memory allocation mechanism, parse "memory" nodes and
1157 * obtain that way the top of memory and RMO to setup out local allocator
1158 */
1159static void __init prom_init_mem(void)
1160{
1161 phandle node;
1162 char *path, type[64];
1163 unsigned int plen;
1164 cell_t *p, *endp;
1165 struct prom_t *_prom = &RELOC(prom);
1166 u32 rac, rsc;
1167
1168 /*
1169 * We iterate the memory nodes to find
1170 * 1) top of RMO (first node)
1171 * 2) top of memory
1172 */
1173 rac = 2;
1174 prom_getprop(_prom->root, "#address-cells", &rac, sizeof(rac));
1175 rsc = 1;
1176 prom_getprop(_prom->root, "#size-cells", &rsc, sizeof(rsc));
1177 prom_debug("root_addr_cells: %x\n", (unsigned long) rac);
1178 prom_debug("root_size_cells: %x\n", (unsigned long) rsc);
1179
1180 prom_debug("scanning memory:\n");
1181 path = RELOC(prom_scratch);
1182
1183 for (node = 0; prom_next_node(&node); ) {
1184 type[0] = 0;
1185 prom_getprop(node, "device_type", type, sizeof(type));
1186
1187 if (type[0] == 0) {
1188 /*
1189 * CHRP Longtrail machines have no device_type
1190 * on the memory node, so check the name instead...
1191 */
1192 prom_getprop(node, "name", type, sizeof(type));
1193 }
1194 if (strcmp(type, RELOC("memory")))
1195 continue;
1196
1197 plen = prom_getprop(node, "reg", RELOC(regbuf), sizeof(regbuf));
1198 if (plen > sizeof(regbuf)) {
1199 prom_printf("memory node too large for buffer !\n");
1200 plen = sizeof(regbuf);
1201 }
1202 p = RELOC(regbuf);
1203 endp = p + (plen / sizeof(cell_t));
1204
1205#ifdef DEBUG_PROM
1206 memset(path, 0, PROM_SCRATCH_SIZE);
1207 call_prom("package-to-path", 3, 1, node, path, PROM_SCRATCH_SIZE-1);
1208 prom_debug(" node %s :\n", path);
1209#endif /* DEBUG_PROM */
1210
1211 while ((endp - p) >= (rac + rsc)) {
1212 unsigned long base, size;
1213
1214 base = prom_next_cell(rac, &p);
1215 size = prom_next_cell(rsc, &p);
1216
1217 if (size == 0)
1218 continue;
1219 prom_debug(" %x %x\n", base, size);
1220 if (base == 0 && (RELOC(of_platform) & PLATFORM_LPAR))
1221 RELOC(rmo_top) = size;
1222 if ((base + size) > RELOC(ram_top))
1223 RELOC(ram_top) = base + size;
1224 }
1225 }
1226
1227 RELOC(alloc_bottom) = PAGE_ALIGN((unsigned long)&RELOC(_end) + 0x4000);
1228
1229 /*
1230 * If prom_memory_limit is set we reduce the upper limits *except* for
1231 * alloc_top_high. This must be the real top of RAM so we can put
1232 * TCE's up there.
1233 */
1234
1235 RELOC(alloc_top_high) = RELOC(ram_top);
1236
1237 if (RELOC(prom_memory_limit)) {
1238 if (RELOC(prom_memory_limit) <= RELOC(alloc_bottom)) {
1239 prom_printf("Ignoring mem=%x <= alloc_bottom.\n",
1240 RELOC(prom_memory_limit));
1241 RELOC(prom_memory_limit) = 0;
1242 } else if (RELOC(prom_memory_limit) >= RELOC(ram_top)) {
1243 prom_printf("Ignoring mem=%x >= ram_top.\n",
1244 RELOC(prom_memory_limit));
1245 RELOC(prom_memory_limit) = 0;
1246 } else {
1247 RELOC(ram_top) = RELOC(prom_memory_limit);
1248 RELOC(rmo_top) = min(RELOC(rmo_top), RELOC(prom_memory_limit));
1249 }
1250 }
1251
1252 /*
1253 * Setup our top alloc point, that is top of RMO or top of
1254 * segment 0 when running non-LPAR.
1255 * Some RS64 machines have buggy firmware where claims up at
1256 * 1GB fail. Cap at 768MB as a workaround.
1257 * Since 768MB is plenty of room, and we need to cap to something
1258 * reasonable on 32-bit, cap at 768MB on all machines.
1259 */
1260 if (!RELOC(rmo_top))
1261 RELOC(rmo_top) = RELOC(ram_top);
1262 RELOC(rmo_top) = min(0x30000000ul, RELOC(rmo_top));
1263 RELOC(alloc_top) = RELOC(rmo_top);
1264 RELOC(alloc_top_high) = RELOC(ram_top);
1265
1266 /*
1267 * Check if we have an initrd after the kernel but still inside
1268 * the RMO. If we do move our bottom point to after it.
1269 */
1270 if (RELOC(prom_initrd_start) &&
1271 RELOC(prom_initrd_start) < RELOC(rmo_top) &&
1272 RELOC(prom_initrd_end) > RELOC(alloc_bottom))
1273 RELOC(alloc_bottom) = PAGE_ALIGN(RELOC(prom_initrd_end));
1274
1275 prom_printf("memory layout at init:\n");
1276 prom_printf(" memory_limit : %x (16 MB aligned)\n", RELOC(prom_memory_limit));
1277 prom_printf(" alloc_bottom : %x\n", RELOC(alloc_bottom));
1278 prom_printf(" alloc_top : %x\n", RELOC(alloc_top));
1279 prom_printf(" alloc_top_hi : %x\n", RELOC(alloc_top_high));
1280 prom_printf(" rmo_top : %x\n", RELOC(rmo_top));
1281 prom_printf(" ram_top : %x\n", RELOC(ram_top));
1282}
1283
1284static void __init prom_close_stdin(void)
1285{
1286 struct prom_t *_prom = &RELOC(prom);
1287 ihandle val;
1288
1289 if (prom_getprop(_prom->chosen, "stdin", &val, sizeof(val)) > 0)
1290 call_prom("close", 1, 0, val);
1291}
1292
1293#ifdef CONFIG_PPC_POWERNV
1294
1295static u64 __initdata prom_opal_size;
1296static u64 __initdata prom_opal_align;
1297static int __initdata prom_rtas_start_cpu;
1298static u64 __initdata prom_rtas_data;
1299static u64 __initdata prom_rtas_entry;
1300
1301#ifdef CONFIG_PPC_EARLY_DEBUG_OPAL
1302static u64 __initdata prom_opal_base;
1303static u64 __initdata prom_opal_entry;
1304#endif
1305
1306/* XXX Don't change this structure without updating opal-takeover.S */
1307static struct opal_secondary_data {
1308 s64 ack; /* 0 */
1309 u64 go; /* 8 */
1310 struct opal_takeover_args args; /* 16 */
1311} opal_secondary_data;
1312
1313extern char opal_secondary_entry;
1314
1315static void __init prom_query_opal(void)
1316{
1317 long rc;
1318
1319 /* We must not query for OPAL presence on a machine that
1320 * supports TNK takeover (970 blades), as this uses the same
1321 * h-call with different arguments and will crash
1322 */
1323 if (PHANDLE_VALID(call_prom("finddevice", 1, 1,
1324 ADDR("/tnk-memory-map")))) {
1325 prom_printf("TNK takeover detected, skipping OPAL check\n");
1326 return;
1327 }
1328
1329 prom_printf("Querying for OPAL presence... ");
1330 rc = opal_query_takeover(&RELOC(prom_opal_size),
1331 &RELOC(prom_opal_align));
1332 prom_debug("(rc = %ld) ", rc);
1333 if (rc != 0) {
1334 prom_printf("not there.\n");
1335 return;
1336 }
1337 RELOC(of_platform) = PLATFORM_OPAL;
1338 prom_printf(" there !\n");
1339 prom_debug(" opal_size = 0x%lx\n", RELOC(prom_opal_size));
1340 prom_debug(" opal_align = 0x%lx\n", RELOC(prom_opal_align));
1341 if (RELOC(prom_opal_align) < 0x10000)
1342 RELOC(prom_opal_align) = 0x10000;
1343}
1344
1345static int prom_rtas_call(int token, int nargs, int nret, int *outputs, ...)
1346{
1347 struct rtas_args rtas_args;
1348 va_list list;
1349 int i;
1350
1351 rtas_args.token = token;
1352 rtas_args.nargs = nargs;
1353 rtas_args.nret = nret;
1354 rtas_args.rets = (rtas_arg_t *)&(rtas_args.args[nargs]);
1355 va_start(list, outputs);
1356 for (i = 0; i < nargs; ++i)
1357 rtas_args.args[i] = va_arg(list, rtas_arg_t);
1358 va_end(list);
1359
1360 for (i = 0; i < nret; ++i)
1361 rtas_args.rets[i] = 0;
1362
1363 opal_enter_rtas(&rtas_args, RELOC(prom_rtas_data),
1364 RELOC(prom_rtas_entry));
1365
1366 if (nret > 1 && outputs != NULL)
1367 for (i = 0; i < nret-1; ++i)
1368 outputs[i] = rtas_args.rets[i+1];
1369 return (nret > 0)? rtas_args.rets[0]: 0;
1370}
1371
1372static void __init prom_opal_hold_cpus(void)
1373{
1374 int i, cnt, cpu, rc;
1375 long j;
1376 phandle node;
1377 char type[64];
1378 u32 servers[8];
1379 struct prom_t *_prom = &RELOC(prom);
1380 void *entry = (unsigned long *)&RELOC(opal_secondary_entry);
1381 struct opal_secondary_data *data = &RELOC(opal_secondary_data);
1382
1383 prom_debug("prom_opal_hold_cpus: start...\n");
1384 prom_debug(" - entry = 0x%x\n", entry);
1385 prom_debug(" - data = 0x%x\n", data);
1386
1387 data->ack = -1;
1388 data->go = 0;
1389
1390 /* look for cpus */
1391 for (node = 0; prom_next_node(&node); ) {
1392 type[0] = 0;
1393 prom_getprop(node, "device_type", type, sizeof(type));
1394 if (strcmp(type, RELOC("cpu")) != 0)
1395 continue;
1396
1397 /* Skip non-configured cpus. */
1398 if (prom_getprop(node, "status", type, sizeof(type)) > 0)
1399 if (strcmp(type, RELOC("okay")) != 0)
1400 continue;
1401
1402 cnt = prom_getprop(node, "ibm,ppc-interrupt-server#s", servers,
1403 sizeof(servers));
1404 if (cnt == PROM_ERROR)
1405 break;
1406 cnt >>= 2;
1407 for (i = 0; i < cnt; i++) {
1408 cpu = servers[i];
1409 prom_debug("CPU %d ... ", cpu);
1410 if (cpu == _prom->cpu) {
1411 prom_debug("booted !\n");
1412 continue;
1413 }
1414 prom_debug("starting ... ");
1415
1416 /* Init the acknowledge var which will be reset by
1417 * the secondary cpu when it awakens from its OF
1418 * spinloop.
1419 */
1420 data->ack = -1;
1421 rc = prom_rtas_call(RELOC(prom_rtas_start_cpu), 3, 1,
1422 NULL, cpu, entry, data);
1423 prom_debug("rtas rc=%d ...", rc);
1424
1425 for (j = 0; j < 100000000 && data->ack == -1; j++) {
1426 HMT_low();
1427 mb();
1428 }
1429 HMT_medium();
1430 if (data->ack != -1)
1431 prom_debug("done, PIR=0x%x\n", data->ack);
1432 else
1433 prom_debug("timeout !\n");
1434 }
1435 }
1436 prom_debug("prom_opal_hold_cpus: end...\n");
1437}
1438
1439static void __init prom_opal_takeover(void)
1440{
1441 struct opal_secondary_data *data = &RELOC(opal_secondary_data);
1442 struct opal_takeover_args *args = &data->args;
1443 u64 align = RELOC(prom_opal_align);
1444 u64 top_addr, opal_addr;
1445
1446 args->k_image = (u64)RELOC(_stext);
1447 args->k_size = _end - _stext;
1448 args->k_entry = 0;
1449 args->k_entry2 = 0x60;
1450
1451 top_addr = _ALIGN_UP(args->k_size, align);
1452
1453 if (RELOC(prom_initrd_start) != 0) {
1454 args->rd_image = RELOC(prom_initrd_start);
1455 args->rd_size = RELOC(prom_initrd_end) - args->rd_image;
1456 args->rd_loc = top_addr;
1457 top_addr = _ALIGN_UP(args->rd_loc + args->rd_size, align);
1458 }
1459
1460 /* Pickup an address for the HAL. We want to go really high
1461 * up to avoid problem with future kexecs. On the other hand
1462 * we don't want to be all over the TCEs on P5IOC2 machines
1463 * which are going to be up there too. We assume the machine
1464 * has plenty of memory, and we ask for the HAL for now to
1465 * be just below the 1G point, or above the initrd
1466 */
1467 opal_addr = _ALIGN_DOWN(0x40000000 - RELOC(prom_opal_size), align);
1468 if (opal_addr < top_addr)
1469 opal_addr = top_addr;
1470 args->hal_addr = opal_addr;
1471
1472 /* Copy the command line to the kernel image */
1473 strlcpy(RELOC(boot_command_line), RELOC(prom_cmd_line),
1474 COMMAND_LINE_SIZE);
1475
1476 prom_debug(" k_image = 0x%lx\n", args->k_image);
1477 prom_debug(" k_size = 0x%lx\n", args->k_size);
1478 prom_debug(" k_entry = 0x%lx\n", args->k_entry);
1479 prom_debug(" k_entry2 = 0x%lx\n", args->k_entry2);
1480 prom_debug(" hal_addr = 0x%lx\n", args->hal_addr);
1481 prom_debug(" rd_image = 0x%lx\n", args->rd_image);
1482 prom_debug(" rd_size = 0x%lx\n", args->rd_size);
1483 prom_debug(" rd_loc = 0x%lx\n", args->rd_loc);
1484 prom_printf("Performing OPAL takeover,this can take a few minutes..\n");
1485 prom_close_stdin();
1486 mb();
1487 data->go = 1;
1488 for (;;)
1489 opal_do_takeover(args);
1490}
1491
1492/*
1493 * Allocate room for and instantiate OPAL
1494 */
1495static void __init prom_instantiate_opal(void)
1496{
1497 phandle opal_node;
1498 ihandle opal_inst;
1499 u64 base, entry;
1500 u64 size = 0, align = 0x10000;
1501 u32 rets[2];
1502
1503 prom_debug("prom_instantiate_opal: start...\n");
1504
1505 opal_node = call_prom("finddevice", 1, 1, ADDR("/ibm,opal"));
1506 prom_debug("opal_node: %x\n", opal_node);
1507 if (!PHANDLE_VALID(opal_node))
1508 return;
1509
1510 prom_getprop(opal_node, "opal-runtime-size", &size, sizeof(size));
1511 if (size == 0)
1512 return;
1513 prom_getprop(opal_node, "opal-runtime-alignment", &align,
1514 sizeof(align));
1515
1516 base = alloc_down(size, align, 0);
1517 if (base == 0) {
1518 prom_printf("OPAL allocation failed !\n");
1519 return;
1520 }
1521
1522 opal_inst = call_prom("open", 1, 1, ADDR("/ibm,opal"));
1523 if (!IHANDLE_VALID(opal_inst)) {
1524 prom_printf("opening opal package failed (%x)\n", opal_inst);
1525 return;
1526 }
1527
1528 prom_printf("instantiating opal at 0x%x...", base);
1529
1530 if (call_prom_ret("call-method", 4, 3, rets,
1531 ADDR("load-opal-runtime"),
1532 opal_inst,
1533 base >> 32, base & 0xffffffff) != 0
1534 || (rets[0] == 0 && rets[1] == 0)) {
1535 prom_printf(" failed\n");
1536 return;
1537 }
1538 entry = (((u64)rets[0]) << 32) | rets[1];
1539
1540 prom_printf(" done\n");
1541
1542 reserve_mem(base, size);
1543
1544 prom_debug("opal base = 0x%x\n", base);
1545 prom_debug("opal align = 0x%x\n", align);
1546 prom_debug("opal entry = 0x%x\n", entry);
1547 prom_debug("opal size = 0x%x\n", (long)size);
1548
1549 prom_setprop(opal_node, "/ibm,opal", "opal-base-address",
1550 &base, sizeof(base));
1551 prom_setprop(opal_node, "/ibm,opal", "opal-entry-address",
1552 &entry, sizeof(entry));
1553
1554#ifdef CONFIG_PPC_EARLY_DEBUG_OPAL
1555 RELOC(prom_opal_base) = base;
1556 RELOC(prom_opal_entry) = entry;
1557#endif
1558 prom_debug("prom_instantiate_opal: end...\n");
1559}
1560
1561#endif /* CONFIG_PPC_POWERNV */
1562
1563/*
1564 * Allocate room for and instantiate RTAS
1565 */
1566static void __init prom_instantiate_rtas(void)
1567{
1568 phandle rtas_node;
1569 ihandle rtas_inst;
1570 u32 base, entry = 0;
1571 u32 size = 0;
1572
1573 prom_debug("prom_instantiate_rtas: start...\n");
1574
1575 rtas_node = call_prom("finddevice", 1, 1, ADDR("/rtas"));
1576 prom_debug("rtas_node: %x\n", rtas_node);
1577 if (!PHANDLE_VALID(rtas_node))
1578 return;
1579
1580 prom_getprop(rtas_node, "rtas-size", &size, sizeof(size));
1581 if (size == 0)
1582 return;
1583
1584 base = alloc_down(size, PAGE_SIZE, 0);
1585 if (base == 0)
1586 prom_panic("Could not allocate memory for RTAS\n");
1587
1588 rtas_inst = call_prom("open", 1, 1, ADDR("/rtas"));
1589 if (!IHANDLE_VALID(rtas_inst)) {
1590 prom_printf("opening rtas package failed (%x)\n", rtas_inst);
1591 return;
1592 }
1593
1594 prom_printf("instantiating rtas at 0x%x...", base);
1595
1596 if (call_prom_ret("call-method", 3, 2, &entry,
1597 ADDR("instantiate-rtas"),
1598 rtas_inst, base) != 0
1599 || entry == 0) {
1600 prom_printf(" failed\n");
1601 return;
1602 }
1603 prom_printf(" done\n");
1604
1605 reserve_mem(base, size);
1606
1607 prom_setprop(rtas_node, "/rtas", "linux,rtas-base",
1608 &base, sizeof(base));
1609 prom_setprop(rtas_node, "/rtas", "linux,rtas-entry",
1610 &entry, sizeof(entry));
1611
1612#ifdef CONFIG_PPC_POWERNV
1613 /* PowerVN takeover hack */
1614 RELOC(prom_rtas_data) = base;
1615 RELOC(prom_rtas_entry) = entry;
1616 prom_getprop(rtas_node, "start-cpu", &RELOC(prom_rtas_start_cpu), 4);
1617#endif
1618 prom_debug("rtas base = 0x%x\n", base);
1619 prom_debug("rtas entry = 0x%x\n", entry);
1620 prom_debug("rtas size = 0x%x\n", (long)size);
1621
1622 prom_debug("prom_instantiate_rtas: end...\n");
1623}
1624
1625#ifdef CONFIG_PPC64
1626/*
1627 * Allocate room for and initialize TCE tables
1628 */
1629static void __init prom_initialize_tce_table(void)
1630{
1631 phandle node;
1632 ihandle phb_node;
1633 char compatible[64], type[64], model[64];
1634 char *path = RELOC(prom_scratch);
1635 u64 base, align;
1636 u32 minalign, minsize;
1637 u64 tce_entry, *tce_entryp;
1638 u64 local_alloc_top, local_alloc_bottom;
1639 u64 i;
1640
1641 if (RELOC(prom_iommu_off))
1642 return;
1643
1644 prom_debug("starting prom_initialize_tce_table\n");
1645
1646 /* Cache current top of allocs so we reserve a single block */
1647 local_alloc_top = RELOC(alloc_top_high);
1648 local_alloc_bottom = local_alloc_top;
1649
1650 /* Search all nodes looking for PHBs. */
1651 for (node = 0; prom_next_node(&node); ) {
1652 compatible[0] = 0;
1653 type[0] = 0;
1654 model[0] = 0;
1655 prom_getprop(node, "compatible",
1656 compatible, sizeof(compatible));
1657 prom_getprop(node, "device_type", type, sizeof(type));
1658 prom_getprop(node, "model", model, sizeof(model));
1659
1660 if ((type[0] == 0) || (strstr(type, RELOC("pci")) == NULL))
1661 continue;
1662
1663 /* Keep the old logic intact to avoid regression. */
1664 if (compatible[0] != 0) {
1665 if ((strstr(compatible, RELOC("python")) == NULL) &&
1666 (strstr(compatible, RELOC("Speedwagon")) == NULL) &&
1667 (strstr(compatible, RELOC("Winnipeg")) == NULL))
1668 continue;
1669 } else if (model[0] != 0) {
1670 if ((strstr(model, RELOC("ython")) == NULL) &&
1671 (strstr(model, RELOC("peedwagon")) == NULL) &&
1672 (strstr(model, RELOC("innipeg")) == NULL))
1673 continue;
1674 }
1675
1676 if (prom_getprop(node, "tce-table-minalign", &minalign,
1677 sizeof(minalign)) == PROM_ERROR)
1678 minalign = 0;
1679 if (prom_getprop(node, "tce-table-minsize", &minsize,
1680 sizeof(minsize)) == PROM_ERROR)
1681 minsize = 4UL << 20;
1682
1683 /*
1684 * Even though we read what OF wants, we just set the table
1685 * size to 4 MB. This is enough to map 2GB of PCI DMA space.
1686 * By doing this, we avoid the pitfalls of trying to DMA to
1687 * MMIO space and the DMA alias hole.
1688 *
1689 * On POWER4, firmware sets the TCE region by assuming
1690 * each TCE table is 8MB. Using this memory for anything
1691 * else will impact performance, so we always allocate 8MB.
1692 * Anton
1693 */
1694 if (__is_processor(PV_POWER4) || __is_processor(PV_POWER4p))
1695 minsize = 8UL << 20;
1696 else
1697 minsize = 4UL << 20;
1698
1699 /* Align to the greater of the align or size */
1700 align = max(minalign, minsize);
1701 base = alloc_down(minsize, align, 1);
1702 if (base == 0)
1703 prom_panic("ERROR, cannot find space for TCE table.\n");
1704 if (base < local_alloc_bottom)
1705 local_alloc_bottom = base;
1706
1707 /* It seems OF doesn't null-terminate the path :-( */
1708 memset(path, 0, PROM_SCRATCH_SIZE);
1709 /* Call OF to setup the TCE hardware */
1710 if (call_prom("package-to-path", 3, 1, node,
1711 path, PROM_SCRATCH_SIZE-1) == PROM_ERROR) {
1712 prom_printf("package-to-path failed\n");
1713 }
1714
1715 /* Save away the TCE table attributes for later use. */
1716 prom_setprop(node, path, "linux,tce-base", &base, sizeof(base));
1717 prom_setprop(node, path, "linux,tce-size", &minsize, sizeof(minsize));
1718
1719 prom_debug("TCE table: %s\n", path);
1720 prom_debug("\tnode = 0x%x\n", node);
1721 prom_debug("\tbase = 0x%x\n", base);
1722 prom_debug("\tsize = 0x%x\n", minsize);
1723
1724 /* Initialize the table to have a one-to-one mapping
1725 * over the allocated size.
1726 */
1727 tce_entryp = (u64 *)base;
1728 for (i = 0; i < (minsize >> 3) ;tce_entryp++, i++) {
1729 tce_entry = (i << PAGE_SHIFT);
1730 tce_entry |= 0x3;
1731 *tce_entryp = tce_entry;
1732 }
1733
1734 prom_printf("opening PHB %s", path);
1735 phb_node = call_prom("open", 1, 1, path);
1736 if (phb_node == 0)
1737 prom_printf("... failed\n");
1738 else
1739 prom_printf("... done\n");
1740
1741 call_prom("call-method", 6, 0, ADDR("set-64-bit-addressing"),
1742 phb_node, -1, minsize,
1743 (u32) base, (u32) (base >> 32));
1744 call_prom("close", 1, 0, phb_node);
1745 }
1746
1747 reserve_mem(local_alloc_bottom, local_alloc_top - local_alloc_bottom);
1748
1749 /* These are only really needed if there is a memory limit in
1750 * effect, but we don't know so export them always. */
1751 RELOC(prom_tce_alloc_start) = local_alloc_bottom;
1752 RELOC(prom_tce_alloc_end) = local_alloc_top;
1753
1754 /* Flag the first invalid entry */
1755 prom_debug("ending prom_initialize_tce_table\n");
1756}
1757#endif
1758
1759/*
1760 * With CHRP SMP we need to use the OF to start the other processors.
1761 * We can't wait until smp_boot_cpus (the OF is trashed by then)
1762 * so we have to put the processors into a holding pattern controlled
1763 * by the kernel (not OF) before we destroy the OF.
1764 *
1765 * This uses a chunk of low memory, puts some holding pattern
1766 * code there and sends the other processors off to there until
1767 * smp_boot_cpus tells them to do something. The holding pattern
1768 * checks that address until its cpu # is there, when it is that
1769 * cpu jumps to __secondary_start(). smp_boot_cpus() takes care
1770 * of setting those values.
1771 *
1772 * We also use physical address 0x4 here to tell when a cpu
1773 * is in its holding pattern code.
1774 *
1775 * -- Cort
1776 */
1777/*
1778 * We want to reference the copy of __secondary_hold_* in the
1779 * 0 - 0x100 address range
1780 */
1781#define LOW_ADDR(x) (((unsigned long) &(x)) & 0xff)
1782
1783static void __init prom_hold_cpus(void)
1784{
1785 unsigned long i;
1786 unsigned int reg;
1787 phandle node;
1788 char type[64];
1789 struct prom_t *_prom = &RELOC(prom);
1790 unsigned long *spinloop
1791 = (void *) LOW_ADDR(__secondary_hold_spinloop);
1792 unsigned long *acknowledge
1793 = (void *) LOW_ADDR(__secondary_hold_acknowledge);
1794 unsigned long secondary_hold = LOW_ADDR(__secondary_hold);
1795
1796 prom_debug("prom_hold_cpus: start...\n");
1797 prom_debug(" 1) spinloop = 0x%x\n", (unsigned long)spinloop);
1798 prom_debug(" 1) *spinloop = 0x%x\n", *spinloop);
1799 prom_debug(" 1) acknowledge = 0x%x\n",
1800 (unsigned long)acknowledge);
1801 prom_debug(" 1) *acknowledge = 0x%x\n", *acknowledge);
1802 prom_debug(" 1) secondary_hold = 0x%x\n", secondary_hold);
1803
1804 /* Set the common spinloop variable, so all of the secondary cpus
1805 * will block when they are awakened from their OF spinloop.
1806 * This must occur for both SMP and non SMP kernels, since OF will
1807 * be trashed when we move the kernel.
1808 */
1809 *spinloop = 0;
1810
1811 /* look for cpus */
1812 for (node = 0; prom_next_node(&node); ) {
1813 type[0] = 0;
1814 prom_getprop(node, "device_type", type, sizeof(type));
1815 if (strcmp(type, RELOC("cpu")) != 0)
1816 continue;
1817
1818 /* Skip non-configured cpus. */
1819 if (prom_getprop(node, "status", type, sizeof(type)) > 0)
1820 if (strcmp(type, RELOC("okay")) != 0)
1821 continue;
1822
1823 reg = -1;
1824 prom_getprop(node, "reg", ®, sizeof(reg));
1825
1826 prom_debug("cpu hw idx = %lu\n", reg);
1827
1828 /* Init the acknowledge var which will be reset by
1829 * the secondary cpu when it awakens from its OF
1830 * spinloop.
1831 */
1832 *acknowledge = (unsigned long)-1;
1833
1834 if (reg != _prom->cpu) {
1835 /* Primary Thread of non-boot cpu or any thread */
1836 prom_printf("starting cpu hw idx %lu... ", reg);
1837 call_prom("start-cpu", 3, 0, node,
1838 secondary_hold, reg);
1839
1840 for (i = 0; (i < 100000000) &&
1841 (*acknowledge == ((unsigned long)-1)); i++ )
1842 mb();
1843
1844 if (*acknowledge == reg)
1845 prom_printf("done\n");
1846 else
1847 prom_printf("failed: %x\n", *acknowledge);
1848 }
1849#ifdef CONFIG_SMP
1850 else
1851 prom_printf("boot cpu hw idx %lu\n", reg);
1852#endif /* CONFIG_SMP */
1853 }
1854
1855 prom_debug("prom_hold_cpus: end...\n");
1856}
1857
1858
1859static void __init prom_init_client_services(unsigned long pp)
1860{
1861 struct prom_t *_prom = &RELOC(prom);
1862
1863 /* Get a handle to the prom entry point before anything else */
1864 RELOC(prom_entry) = pp;
1865
1866 /* get a handle for the stdout device */
1867 _prom->chosen = call_prom("finddevice", 1, 1, ADDR("/chosen"));
1868 if (!PHANDLE_VALID(_prom->chosen))
1869 prom_panic("cannot find chosen"); /* msg won't be printed :( */
1870
1871 /* get device tree root */
1872 _prom->root = call_prom("finddevice", 1, 1, ADDR("/"));
1873 if (!PHANDLE_VALID(_prom->root))
1874 prom_panic("cannot find device tree root"); /* msg won't be printed :( */
1875
1876 _prom->mmumap = 0;
1877}
1878
1879#ifdef CONFIG_PPC32
1880/*
1881 * For really old powermacs, we need to map things we claim.
1882 * For that, we need the ihandle of the mmu.
1883 * Also, on the longtrail, we need to work around other bugs.
1884 */
1885static void __init prom_find_mmu(void)
1886{
1887 struct prom_t *_prom = &RELOC(prom);
1888 phandle oprom;
1889 char version[64];
1890
1891 oprom = call_prom("finddevice", 1, 1, ADDR("/openprom"));
1892 if (!PHANDLE_VALID(oprom))
1893 return;
1894 if (prom_getprop(oprom, "model", version, sizeof(version)) <= 0)
1895 return;
1896 version[sizeof(version) - 1] = 0;
1897 /* XXX might need to add other versions here */
1898 if (strcmp(version, "Open Firmware, 1.0.5") == 0)
1899 of_workarounds = OF_WA_CLAIM;
1900 else if (strncmp(version, "FirmWorks,3.", 12) == 0) {
1901 of_workarounds = OF_WA_CLAIM | OF_WA_LONGTRAIL;
1902 call_prom("interpret", 1, 1, "dev /memory 0 to allow-reclaim");
1903 } else
1904 return;
1905 _prom->memory = call_prom("open", 1, 1, ADDR("/memory"));
1906 prom_getprop(_prom->chosen, "mmu", &_prom->mmumap,
1907 sizeof(_prom->mmumap));
1908 if (!IHANDLE_VALID(_prom->memory) || !IHANDLE_VALID(_prom->mmumap))
1909 of_workarounds &= ~OF_WA_CLAIM; /* hmmm */
1910}
1911#else
1912#define prom_find_mmu()
1913#endif
1914
1915static void __init prom_init_stdout(void)
1916{
1917 struct prom_t *_prom = &RELOC(prom);
1918 char *path = RELOC(of_stdout_device);
1919 char type[16];
1920 u32 val;
1921
1922 if (prom_getprop(_prom->chosen, "stdout", &val, sizeof(val)) <= 0)
1923 prom_panic("cannot find stdout");
1924
1925 _prom->stdout = val;
1926
1927 /* Get the full OF pathname of the stdout device */
1928 memset(path, 0, 256);
1929 call_prom("instance-to-path", 3, 1, _prom->stdout, path, 255);
1930 val = call_prom("instance-to-package", 1, 1, _prom->stdout);
1931 prom_setprop(_prom->chosen, "/chosen", "linux,stdout-package",
1932 &val, sizeof(val));
1933 prom_printf("OF stdout device is: %s\n", RELOC(of_stdout_device));
1934 prom_setprop(_prom->chosen, "/chosen", "linux,stdout-path",
1935 path, strlen(path) + 1);
1936
1937 /* If it's a display, note it */
1938 memset(type, 0, sizeof(type));
1939 prom_getprop(val, "device_type", type, sizeof(type));
1940 if (strcmp(type, RELOC("display")) == 0)
1941 prom_setprop(val, path, "linux,boot-display", NULL, 0);
1942}
1943
1944static int __init prom_find_machine_type(void)
1945{
1946 struct prom_t *_prom = &RELOC(prom);
1947 char compat[256];
1948 int len, i = 0;
1949#ifdef CONFIG_PPC64
1950 phandle rtas;
1951 int x;
1952#endif
1953
1954 /* Look for a PowerMac or a Cell */
1955 len = prom_getprop(_prom->root, "compatible",
1956 compat, sizeof(compat)-1);
1957 if (len > 0) {
1958 compat[len] = 0;
1959 while (i < len) {
1960 char *p = &compat[i];
1961 int sl = strlen(p);
1962 if (sl == 0)
1963 break;
1964 if (strstr(p, RELOC("Power Macintosh")) ||
1965 strstr(p, RELOC("MacRISC")))
1966 return PLATFORM_POWERMAC;
1967#ifdef CONFIG_PPC64
1968 /* We must make sure we don't detect the IBM Cell
1969 * blades as pSeries due to some firmware issues,
1970 * so we do it here.
1971 */
1972 if (strstr(p, RELOC("IBM,CBEA")) ||
1973 strstr(p, RELOC("IBM,CPBW-1.0")))
1974 return PLATFORM_GENERIC;
1975#endif /* CONFIG_PPC64 */
1976 i += sl + 1;
1977 }
1978 }
1979#ifdef CONFIG_PPC64
1980 /* Try to detect OPAL */
1981 if (PHANDLE_VALID(call_prom("finddevice", 1, 1, ADDR("/ibm,opal"))))
1982 return PLATFORM_OPAL;
1983
1984 /* Try to figure out if it's an IBM pSeries or any other
1985 * PAPR compliant platform. We assume it is if :
1986 * - /device_type is "chrp" (please, do NOT use that for future
1987 * non-IBM designs !
1988 * - it has /rtas
1989 */
1990 len = prom_getprop(_prom->root, "device_type",
1991 compat, sizeof(compat)-1);
1992 if (len <= 0)
1993 return PLATFORM_GENERIC;
1994 if (strcmp(compat, RELOC("chrp")))
1995 return PLATFORM_GENERIC;
1996
1997 /* Default to pSeries. We need to know if we are running LPAR */
1998 rtas = call_prom("finddevice", 1, 1, ADDR("/rtas"));
1999 if (!PHANDLE_VALID(rtas))
2000 return PLATFORM_GENERIC;
2001 x = prom_getproplen(rtas, "ibm,hypertas-functions");
2002 if (x != PROM_ERROR) {
2003 prom_debug("Hypertas detected, assuming LPAR !\n");
2004 return PLATFORM_PSERIES_LPAR;
2005 }
2006 return PLATFORM_PSERIES;
2007#else
2008 return PLATFORM_GENERIC;
2009#endif
2010}
2011
2012static int __init prom_set_color(ihandle ih, int i, int r, int g, int b)
2013{
2014 return call_prom("call-method", 6, 1, ADDR("color!"), ih, i, b, g, r);
2015}
2016
2017/*
2018 * If we have a display that we don't know how to drive,
2019 * we will want to try to execute OF's open method for it
2020 * later. However, OF will probably fall over if we do that
2021 * we've taken over the MMU.
2022 * So we check whether we will need to open the display,
2023 * and if so, open it now.
2024 */
2025static void __init prom_check_displays(void)
2026{
2027 char type[16], *path;
2028 phandle node;
2029 ihandle ih;
2030 int i;
2031
2032 static unsigned char default_colors[] = {
2033 0x00, 0x00, 0x00,
2034 0x00, 0x00, 0xaa,
2035 0x00, 0xaa, 0x00,
2036 0x00, 0xaa, 0xaa,
2037 0xaa, 0x00, 0x00,
2038 0xaa, 0x00, 0xaa,
2039 0xaa, 0xaa, 0x00,
2040 0xaa, 0xaa, 0xaa,
2041 0x55, 0x55, 0x55,
2042 0x55, 0x55, 0xff,
2043 0x55, 0xff, 0x55,
2044 0x55, 0xff, 0xff,
2045 0xff, 0x55, 0x55,
2046 0xff, 0x55, 0xff,
2047 0xff, 0xff, 0x55,
2048 0xff, 0xff, 0xff
2049 };
2050 const unsigned char *clut;
2051
2052 prom_debug("Looking for displays\n");
2053 for (node = 0; prom_next_node(&node); ) {
2054 memset(type, 0, sizeof(type));
2055 prom_getprop(node, "device_type", type, sizeof(type));
2056 if (strcmp(type, RELOC("display")) != 0)
2057 continue;
2058
2059 /* It seems OF doesn't null-terminate the path :-( */
2060 path = RELOC(prom_scratch);
2061 memset(path, 0, PROM_SCRATCH_SIZE);
2062
2063 /*
2064 * leave some room at the end of the path for appending extra
2065 * arguments
2066 */
2067 if (call_prom("package-to-path", 3, 1, node, path,
2068 PROM_SCRATCH_SIZE-10) == PROM_ERROR)
2069 continue;
2070 prom_printf("found display : %s, opening... ", path);
2071
2072 ih = call_prom("open", 1, 1, path);
2073 if (ih == 0) {
2074 prom_printf("failed\n");
2075 continue;
2076 }
2077
2078 /* Success */
2079 prom_printf("done\n");
2080 prom_setprop(node, path, "linux,opened", NULL, 0);
2081
2082 /* Setup a usable color table when the appropriate
2083 * method is available. Should update this to set-colors */
2084 clut = RELOC(default_colors);
2085 for (i = 0; i < 16; i++, clut += 3)
2086 if (prom_set_color(ih, i, clut[0], clut[1],
2087 clut[2]) != 0)
2088 break;
2089
2090#ifdef CONFIG_LOGO_LINUX_CLUT224
2091 clut = PTRRELOC(RELOC(logo_linux_clut224.clut));
2092 for (i = 0; i < RELOC(logo_linux_clut224.clutsize); i++, clut += 3)
2093 if (prom_set_color(ih, i + 32, clut[0], clut[1],
2094 clut[2]) != 0)
2095 break;
2096#endif /* CONFIG_LOGO_LINUX_CLUT224 */
2097 }
2098}
2099
2100
2101/* Return (relocated) pointer to this much memory: moves initrd if reqd. */
2102static void __init *make_room(unsigned long *mem_start, unsigned long *mem_end,
2103 unsigned long needed, unsigned long align)
2104{
2105 void *ret;
2106
2107 *mem_start = _ALIGN(*mem_start, align);
2108 while ((*mem_start + needed) > *mem_end) {
2109 unsigned long room, chunk;
2110
2111 prom_debug("Chunk exhausted, claiming more at %x...\n",
2112 RELOC(alloc_bottom));
2113 room = RELOC(alloc_top) - RELOC(alloc_bottom);
2114 if (room > DEVTREE_CHUNK_SIZE)
2115 room = DEVTREE_CHUNK_SIZE;
2116 if (room < PAGE_SIZE)
2117 prom_panic("No memory for flatten_device_tree "
2118 "(no room)\n");
2119 chunk = alloc_up(room, 0);
2120 if (chunk == 0)
2121 prom_panic("No memory for flatten_device_tree "
2122 "(claim failed)\n");
2123 *mem_end = chunk + room;
2124 }
2125
2126 ret = (void *)*mem_start;
2127 *mem_start += needed;
2128
2129 return ret;
2130}
2131
2132#define dt_push_token(token, mem_start, mem_end) \
2133 do { *((u32 *)make_room(mem_start, mem_end, 4, 4)) = token; } while(0)
2134
2135static unsigned long __init dt_find_string(char *str)
2136{
2137 char *s, *os;
2138
2139 s = os = (char *)RELOC(dt_string_start);
2140 s += 4;
2141 while (s < (char *)RELOC(dt_string_end)) {
2142 if (strcmp(s, str) == 0)
2143 return s - os;
2144 s += strlen(s) + 1;
2145 }
2146 return 0;
2147}
2148
2149/*
2150 * The Open Firmware 1275 specification states properties must be 31 bytes or
2151 * less, however not all firmwares obey this. Make it 64 bytes to be safe.
2152 */
2153#define MAX_PROPERTY_NAME 64
2154
2155static void __init scan_dt_build_strings(phandle node,
2156 unsigned long *mem_start,
2157 unsigned long *mem_end)
2158{
2159 char *prev_name, *namep, *sstart;
2160 unsigned long soff;
2161 phandle child;
2162
2163 sstart = (char *)RELOC(dt_string_start);
2164
2165 /* get and store all property names */
2166 prev_name = RELOC("");
2167 for (;;) {
2168 /* 64 is max len of name including nul. */
2169 namep = make_room(mem_start, mem_end, MAX_PROPERTY_NAME, 1);
2170 if (call_prom("nextprop", 3, 1, node, prev_name, namep) != 1) {
2171 /* No more nodes: unwind alloc */
2172 *mem_start = (unsigned long)namep;
2173 break;
2174 }
2175
2176 /* skip "name" */
2177 if (strcmp(namep, RELOC("name")) == 0) {
2178 *mem_start = (unsigned long)namep;
2179 prev_name = RELOC("name");
2180 continue;
2181 }
2182 /* get/create string entry */
2183 soff = dt_find_string(namep);
2184 if (soff != 0) {
2185 *mem_start = (unsigned long)namep;
2186 namep = sstart + soff;
2187 } else {
2188 /* Trim off some if we can */
2189 *mem_start = (unsigned long)namep + strlen(namep) + 1;
2190 RELOC(dt_string_end) = *mem_start;
2191 }
2192 prev_name = namep;
2193 }
2194
2195 /* do all our children */
2196 child = call_prom("child", 1, 1, node);
2197 while (child != 0) {
2198 scan_dt_build_strings(child, mem_start, mem_end);
2199 child = call_prom("peer", 1, 1, child);
2200 }
2201}
2202
2203static void __init scan_dt_build_struct(phandle node, unsigned long *mem_start,
2204 unsigned long *mem_end)
2205{
2206 phandle child;
2207 char *namep, *prev_name, *sstart, *p, *ep, *lp, *path;
2208 unsigned long soff;
2209 unsigned char *valp;
2210 static char pname[MAX_PROPERTY_NAME];
2211 int l, room, has_phandle = 0;
2212
2213 dt_push_token(OF_DT_BEGIN_NODE, mem_start, mem_end);
2214
2215 /* get the node's full name */
2216 namep = (char *)*mem_start;
2217 room = *mem_end - *mem_start;
2218 if (room > 255)
2219 room = 255;
2220 l = call_prom("package-to-path", 3, 1, node, namep, room);
2221 if (l >= 0) {
2222 /* Didn't fit? Get more room. */
2223 if (l >= room) {
2224 if (l >= *mem_end - *mem_start)
2225 namep = make_room(mem_start, mem_end, l+1, 1);
2226 call_prom("package-to-path", 3, 1, node, namep, l);
2227 }
2228 namep[l] = '\0';
2229
2230 /* Fixup an Apple bug where they have bogus \0 chars in the
2231 * middle of the path in some properties, and extract
2232 * the unit name (everything after the last '/').
2233 */
2234 for (lp = p = namep, ep = namep + l; p < ep; p++) {
2235 if (*p == '/')
2236 lp = namep;
2237 else if (*p != 0)
2238 *lp++ = *p;
2239 }
2240 *lp = 0;
2241 *mem_start = _ALIGN((unsigned long)lp + 1, 4);
2242 }
2243
2244 /* get it again for debugging */
2245 path = RELOC(prom_scratch);
2246 memset(path, 0, PROM_SCRATCH_SIZE);
2247 call_prom("package-to-path", 3, 1, node, path, PROM_SCRATCH_SIZE-1);
2248
2249 /* get and store all properties */
2250 prev_name = RELOC("");
2251 sstart = (char *)RELOC(dt_string_start);
2252 for (;;) {
2253 if (call_prom("nextprop", 3, 1, node, prev_name,
2254 RELOC(pname)) != 1)
2255 break;
2256
2257 /* skip "name" */
2258 if (strcmp(RELOC(pname), RELOC("name")) == 0) {
2259 prev_name = RELOC("name");
2260 continue;
2261 }
2262
2263 /* find string offset */
2264 soff = dt_find_string(RELOC(pname));
2265 if (soff == 0) {
2266 prom_printf("WARNING: Can't find string index for"
2267 " <%s>, node %s\n", RELOC(pname), path);
2268 break;
2269 }
2270 prev_name = sstart + soff;
2271
2272 /* get length */
2273 l = call_prom("getproplen", 2, 1, node, RELOC(pname));
2274
2275 /* sanity checks */
2276 if (l == PROM_ERROR)
2277 continue;
2278
2279 /* push property head */
2280 dt_push_token(OF_DT_PROP, mem_start, mem_end);
2281 dt_push_token(l, mem_start, mem_end);
2282 dt_push_token(soff, mem_start, mem_end);
2283
2284 /* push property content */
2285 valp = make_room(mem_start, mem_end, l, 4);
2286 call_prom("getprop", 4, 1, node, RELOC(pname), valp, l);
2287 *mem_start = _ALIGN(*mem_start, 4);
2288
2289 if (!strcmp(RELOC(pname), RELOC("phandle")))
2290 has_phandle = 1;
2291 }
2292
2293 /* Add a "linux,phandle" property if no "phandle" property already
2294 * existed (can happen with OPAL)
2295 */
2296 if (!has_phandle) {
2297 soff = dt_find_string(RELOC("linux,phandle"));
2298 if (soff == 0)
2299 prom_printf("WARNING: Can't find string index for"
2300 " <linux-phandle> node %s\n", path);
2301 else {
2302 dt_push_token(OF_DT_PROP, mem_start, mem_end);
2303 dt_push_token(4, mem_start, mem_end);
2304 dt_push_token(soff, mem_start, mem_end);
2305 valp = make_room(mem_start, mem_end, 4, 4);
2306 *(u32 *)valp = node;
2307 }
2308 }
2309
2310 /* do all our children */
2311 child = call_prom("child", 1, 1, node);
2312 while (child != 0) {
2313 scan_dt_build_struct(child, mem_start, mem_end);
2314 child = call_prom("peer", 1, 1, child);
2315 }
2316
2317 dt_push_token(OF_DT_END_NODE, mem_start, mem_end);
2318}
2319
2320static void __init flatten_device_tree(void)
2321{
2322 phandle root;
2323 unsigned long mem_start, mem_end, room;
2324 struct boot_param_header *hdr;
2325 struct prom_t *_prom = &RELOC(prom);
2326 char *namep;
2327 u64 *rsvmap;
2328
2329 /*
2330 * Check how much room we have between alloc top & bottom (+/- a
2331 * few pages), crop to 1MB, as this is our "chunk" size
2332 */
2333 room = RELOC(alloc_top) - RELOC(alloc_bottom) - 0x4000;
2334 if (room > DEVTREE_CHUNK_SIZE)
2335 room = DEVTREE_CHUNK_SIZE;
2336 prom_debug("starting device tree allocs at %x\n", RELOC(alloc_bottom));
2337
2338 /* Now try to claim that */
2339 mem_start = (unsigned long)alloc_up(room, PAGE_SIZE);
2340 if (mem_start == 0)
2341 prom_panic("Can't allocate initial device-tree chunk\n");
2342 mem_end = mem_start + room;
2343
2344 /* Get root of tree */
2345 root = call_prom("peer", 1, 1, (phandle)0);
2346 if (root == (phandle)0)
2347 prom_panic ("couldn't get device tree root\n");
2348
2349 /* Build header and make room for mem rsv map */
2350 mem_start = _ALIGN(mem_start, 4);
2351 hdr = make_room(&mem_start, &mem_end,
2352 sizeof(struct boot_param_header), 4);
2353 RELOC(dt_header_start) = (unsigned long)hdr;
2354 rsvmap = make_room(&mem_start, &mem_end, sizeof(mem_reserve_map), 8);
2355
2356 /* Start of strings */
2357 mem_start = PAGE_ALIGN(mem_start);
2358 RELOC(dt_string_start) = mem_start;
2359 mem_start += 4; /* hole */
2360
2361 /* Add "linux,phandle" in there, we'll need it */
2362 namep = make_room(&mem_start, &mem_end, 16, 1);
2363 strcpy(namep, RELOC("linux,phandle"));
2364 mem_start = (unsigned long)namep + strlen(namep) + 1;
2365
2366 /* Build string array */
2367 prom_printf("Building dt strings...\n");
2368 scan_dt_build_strings(root, &mem_start, &mem_end);
2369 RELOC(dt_string_end) = mem_start;
2370
2371 /* Build structure */
2372 mem_start = PAGE_ALIGN(mem_start);
2373 RELOC(dt_struct_start) = mem_start;
2374 prom_printf("Building dt structure...\n");
2375 scan_dt_build_struct(root, &mem_start, &mem_end);
2376 dt_push_token(OF_DT_END, &mem_start, &mem_end);
2377 RELOC(dt_struct_end) = PAGE_ALIGN(mem_start);
2378
2379 /* Finish header */
2380 hdr->boot_cpuid_phys = _prom->cpu;
2381 hdr->magic = OF_DT_HEADER;
2382 hdr->totalsize = RELOC(dt_struct_end) - RELOC(dt_header_start);
2383 hdr->off_dt_struct = RELOC(dt_struct_start) - RELOC(dt_header_start);
2384 hdr->off_dt_strings = RELOC(dt_string_start) - RELOC(dt_header_start);
2385 hdr->dt_strings_size = RELOC(dt_string_end) - RELOC(dt_string_start);
2386 hdr->off_mem_rsvmap = ((unsigned long)rsvmap) - RELOC(dt_header_start);
2387 hdr->version = OF_DT_VERSION;
2388 /* Version 16 is not backward compatible */
2389 hdr->last_comp_version = 0x10;
2390
2391 /* Copy the reserve map in */
2392 memcpy(rsvmap, RELOC(mem_reserve_map), sizeof(mem_reserve_map));
2393
2394#ifdef DEBUG_PROM
2395 {
2396 int i;
2397 prom_printf("reserved memory map:\n");
2398 for (i = 0; i < RELOC(mem_reserve_cnt); i++)
2399 prom_printf(" %x - %x\n",
2400 RELOC(mem_reserve_map)[i].base,
2401 RELOC(mem_reserve_map)[i].size);
2402 }
2403#endif
2404 /* Bump mem_reserve_cnt to cause further reservations to fail
2405 * since it's too late.
2406 */
2407 RELOC(mem_reserve_cnt) = MEM_RESERVE_MAP_SIZE;
2408
2409 prom_printf("Device tree strings 0x%x -> 0x%x\n",
2410 RELOC(dt_string_start), RELOC(dt_string_end));
2411 prom_printf("Device tree struct 0x%x -> 0x%x\n",
2412 RELOC(dt_struct_start), RELOC(dt_struct_end));
2413
2414}
2415
2416#ifdef CONFIG_PPC_MAPLE
2417/* PIBS Version 1.05.0000 04/26/2005 has an incorrect /ht/isa/ranges property.
2418 * The values are bad, and it doesn't even have the right number of cells. */
2419static void __init fixup_device_tree_maple(void)
2420{
2421 phandle isa;
2422 u32 rloc = 0x01002000; /* IO space; PCI device = 4 */
2423 u32 isa_ranges[6];
2424 char *name;
2425
2426 name = "/ht@0/isa@4";
2427 isa = call_prom("finddevice", 1, 1, ADDR(name));
2428 if (!PHANDLE_VALID(isa)) {
2429 name = "/ht@0/isa@6";
2430 isa = call_prom("finddevice", 1, 1, ADDR(name));
2431 rloc = 0x01003000; /* IO space; PCI device = 6 */
2432 }
2433 if (!PHANDLE_VALID(isa))
2434 return;
2435
2436 if (prom_getproplen(isa, "ranges") != 12)
2437 return;
2438 if (prom_getprop(isa, "ranges", isa_ranges, sizeof(isa_ranges))
2439 == PROM_ERROR)
2440 return;
2441
2442 if (isa_ranges[0] != 0x1 ||
2443 isa_ranges[1] != 0xf4000000 ||
2444 isa_ranges[2] != 0x00010000)
2445 return;
2446
2447 prom_printf("Fixing up bogus ISA range on Maple/Apache...\n");
2448
2449 isa_ranges[0] = 0x1;
2450 isa_ranges[1] = 0x0;
2451 isa_ranges[2] = rloc;
2452 isa_ranges[3] = 0x0;
2453 isa_ranges[4] = 0x0;
2454 isa_ranges[5] = 0x00010000;
2455 prom_setprop(isa, name, "ranges",
2456 isa_ranges, sizeof(isa_ranges));
2457}
2458
2459#define CPC925_MC_START 0xf8000000
2460#define CPC925_MC_LENGTH 0x1000000
2461/* The values for memory-controller don't have right number of cells */
2462static void __init fixup_device_tree_maple_memory_controller(void)
2463{
2464 phandle mc;
2465 u32 mc_reg[4];
2466 char *name = "/hostbridge@f8000000";
2467 struct prom_t *_prom = &RELOC(prom);
2468 u32 ac, sc;
2469
2470 mc = call_prom("finddevice", 1, 1, ADDR(name));
2471 if (!PHANDLE_VALID(mc))
2472 return;
2473
2474 if (prom_getproplen(mc, "reg") != 8)
2475 return;
2476
2477 prom_getprop(_prom->root, "#address-cells", &ac, sizeof(ac));
2478 prom_getprop(_prom->root, "#size-cells", &sc, sizeof(sc));
2479 if ((ac != 2) || (sc != 2))
2480 return;
2481
2482 if (prom_getprop(mc, "reg", mc_reg, sizeof(mc_reg)) == PROM_ERROR)
2483 return;
2484
2485 if (mc_reg[0] != CPC925_MC_START || mc_reg[1] != CPC925_MC_LENGTH)
2486 return;
2487
2488 prom_printf("Fixing up bogus hostbridge on Maple...\n");
2489
2490 mc_reg[0] = 0x0;
2491 mc_reg[1] = CPC925_MC_START;
2492 mc_reg[2] = 0x0;
2493 mc_reg[3] = CPC925_MC_LENGTH;
2494 prom_setprop(mc, name, "reg", mc_reg, sizeof(mc_reg));
2495}
2496#else
2497#define fixup_device_tree_maple()
2498#define fixup_device_tree_maple_memory_controller()
2499#endif
2500
2501#ifdef CONFIG_PPC_CHRP
2502/*
2503 * Pegasos and BriQ lacks the "ranges" property in the isa node
2504 * Pegasos needs decimal IRQ 14/15, not hexadecimal
2505 * Pegasos has the IDE configured in legacy mode, but advertised as native
2506 */
2507static void __init fixup_device_tree_chrp(void)
2508{
2509 phandle ph;
2510 u32 prop[6];
2511 u32 rloc = 0x01006000; /* IO space; PCI device = 12 */
2512 char *name;
2513 int rc;
2514
2515 name = "/pci@80000000/isa@c";
2516 ph = call_prom("finddevice", 1, 1, ADDR(name));
2517 if (!PHANDLE_VALID(ph)) {
2518 name = "/pci@ff500000/isa@6";
2519 ph = call_prom("finddevice", 1, 1, ADDR(name));
2520 rloc = 0x01003000; /* IO space; PCI device = 6 */
2521 }
2522 if (PHANDLE_VALID(ph)) {
2523 rc = prom_getproplen(ph, "ranges");
2524 if (rc == 0 || rc == PROM_ERROR) {
2525 prom_printf("Fixing up missing ISA range on Pegasos...\n");
2526
2527 prop[0] = 0x1;
2528 prop[1] = 0x0;
2529 prop[2] = rloc;
2530 prop[3] = 0x0;
2531 prop[4] = 0x0;
2532 prop[5] = 0x00010000;
2533 prom_setprop(ph, name, "ranges", prop, sizeof(prop));
2534 }
2535 }
2536
2537 name = "/pci@80000000/ide@C,1";
2538 ph = call_prom("finddevice", 1, 1, ADDR(name));
2539 if (PHANDLE_VALID(ph)) {
2540 prom_printf("Fixing up IDE interrupt on Pegasos...\n");
2541 prop[0] = 14;
2542 prop[1] = 0x0;
2543 prom_setprop(ph, name, "interrupts", prop, 2*sizeof(u32));
2544 prom_printf("Fixing up IDE class-code on Pegasos...\n");
2545 rc = prom_getprop(ph, "class-code", prop, sizeof(u32));
2546 if (rc == sizeof(u32)) {
2547 prop[0] &= ~0x5;
2548 prom_setprop(ph, name, "class-code", prop, sizeof(u32));
2549 }
2550 }
2551}
2552#else
2553#define fixup_device_tree_chrp()
2554#endif
2555
2556#if defined(CONFIG_PPC64) && defined(CONFIG_PPC_PMAC)
2557static void __init fixup_device_tree_pmac(void)
2558{
2559 phandle u3, i2c, mpic;
2560 u32 u3_rev;
2561 u32 interrupts[2];
2562 u32 parent;
2563
2564 /* Some G5s have a missing interrupt definition, fix it up here */
2565 u3 = call_prom("finddevice", 1, 1, ADDR("/u3@0,f8000000"));
2566 if (!PHANDLE_VALID(u3))
2567 return;
2568 i2c = call_prom("finddevice", 1, 1, ADDR("/u3@0,f8000000/i2c@f8001000"));
2569 if (!PHANDLE_VALID(i2c))
2570 return;
2571 mpic = call_prom("finddevice", 1, 1, ADDR("/u3@0,f8000000/mpic@f8040000"));
2572 if (!PHANDLE_VALID(mpic))
2573 return;
2574
2575 /* check if proper rev of u3 */
2576 if (prom_getprop(u3, "device-rev", &u3_rev, sizeof(u3_rev))
2577 == PROM_ERROR)
2578 return;
2579 if (u3_rev < 0x35 || u3_rev > 0x39)
2580 return;
2581 /* does it need fixup ? */
2582 if (prom_getproplen(i2c, "interrupts") > 0)
2583 return;
2584
2585 prom_printf("fixing up bogus interrupts for u3 i2c...\n");
2586
2587 /* interrupt on this revision of u3 is number 0 and level */
2588 interrupts[0] = 0;
2589 interrupts[1] = 1;
2590 prom_setprop(i2c, "/u3@0,f8000000/i2c@f8001000", "interrupts",
2591 &interrupts, sizeof(interrupts));
2592 parent = (u32)mpic;
2593 prom_setprop(i2c, "/u3@0,f8000000/i2c@f8001000", "interrupt-parent",
2594 &parent, sizeof(parent));
2595}
2596#else
2597#define fixup_device_tree_pmac()
2598#endif
2599
2600#ifdef CONFIG_PPC_EFIKA
2601/*
2602 * The MPC5200 FEC driver requires an phy-handle property to tell it how
2603 * to talk to the phy. If the phy-handle property is missing, then this
2604 * function is called to add the appropriate nodes and link it to the
2605 * ethernet node.
2606 */
2607static void __init fixup_device_tree_efika_add_phy(void)
2608{
2609 u32 node;
2610 char prop[64];
2611 int rv;
2612
2613 /* Check if /builtin/ethernet exists - bail if it doesn't */
2614 node = call_prom("finddevice", 1, 1, ADDR("/builtin/ethernet"));
2615 if (!PHANDLE_VALID(node))
2616 return;
2617
2618 /* Check if the phy-handle property exists - bail if it does */
2619 rv = prom_getprop(node, "phy-handle", prop, sizeof(prop));
2620 if (!rv)
2621 return;
2622
2623 /*
2624 * At this point the ethernet device doesn't have a phy described.
2625 * Now we need to add the missing phy node and linkage
2626 */
2627
2628 /* Check for an MDIO bus node - if missing then create one */
2629 node = call_prom("finddevice", 1, 1, ADDR("/builtin/mdio"));
2630 if (!PHANDLE_VALID(node)) {
2631 prom_printf("Adding Ethernet MDIO node\n");
2632 call_prom("interpret", 1, 1,
2633 " s\" /builtin\" find-device"
2634 " new-device"
2635 " 1 encode-int s\" #address-cells\" property"
2636 " 0 encode-int s\" #size-cells\" property"
2637 " s\" mdio\" device-name"
2638 " s\" fsl,mpc5200b-mdio\" encode-string"
2639 " s\" compatible\" property"
2640 " 0xf0003000 0x400 reg"
2641 " 0x2 encode-int"
2642 " 0x5 encode-int encode+"
2643 " 0x3 encode-int encode+"
2644 " s\" interrupts\" property"
2645 " finish-device");
2646 };
2647
2648 /* Check for a PHY device node - if missing then create one and
2649 * give it's phandle to the ethernet node */
2650 node = call_prom("finddevice", 1, 1,
2651 ADDR("/builtin/mdio/ethernet-phy"));
2652 if (!PHANDLE_VALID(node)) {
2653 prom_printf("Adding Ethernet PHY node\n");
2654 call_prom("interpret", 1, 1,
2655 " s\" /builtin/mdio\" find-device"
2656 " new-device"
2657 " s\" ethernet-phy\" device-name"
2658 " 0x10 encode-int s\" reg\" property"
2659 " my-self"
2660 " ihandle>phandle"
2661 " finish-device"
2662 " s\" /builtin/ethernet\" find-device"
2663 " encode-int"
2664 " s\" phy-handle\" property"
2665 " device-end");
2666 }
2667}
2668
2669static void __init fixup_device_tree_efika(void)
2670{
2671 int sound_irq[3] = { 2, 2, 0 };
2672 int bcomm_irq[3*16] = { 3,0,0, 3,1,0, 3,2,0, 3,3,0,
2673 3,4,0, 3,5,0, 3,6,0, 3,7,0,
2674 3,8,0, 3,9,0, 3,10,0, 3,11,0,
2675 3,12,0, 3,13,0, 3,14,0, 3,15,0 };
2676 u32 node;
2677 char prop[64];
2678 int rv, len;
2679
2680 /* Check if we're really running on a EFIKA */
2681 node = call_prom("finddevice", 1, 1, ADDR("/"));
2682 if (!PHANDLE_VALID(node))
2683 return;
2684
2685 rv = prom_getprop(node, "model", prop, sizeof(prop));
2686 if (rv == PROM_ERROR)
2687 return;
2688 if (strcmp(prop, "EFIKA5K2"))
2689 return;
2690
2691 prom_printf("Applying EFIKA device tree fixups\n");
2692
2693 /* Claiming to be 'chrp' is death */
2694 node = call_prom("finddevice", 1, 1, ADDR("/"));
2695 rv = prom_getprop(node, "device_type", prop, sizeof(prop));
2696 if (rv != PROM_ERROR && (strcmp(prop, "chrp") == 0))
2697 prom_setprop(node, "/", "device_type", "efika", sizeof("efika"));
2698
2699 /* CODEGEN,description is exposed in /proc/cpuinfo so
2700 fix that too */
2701 rv = prom_getprop(node, "CODEGEN,description", prop, sizeof(prop));
2702 if (rv != PROM_ERROR && (strstr(prop, "CHRP")))
2703 prom_setprop(node, "/", "CODEGEN,description",
2704 "Efika 5200B PowerPC System",
2705 sizeof("Efika 5200B PowerPC System"));
2706
2707 /* Fixup bestcomm interrupts property */
2708 node = call_prom("finddevice", 1, 1, ADDR("/builtin/bestcomm"));
2709 if (PHANDLE_VALID(node)) {
2710 len = prom_getproplen(node, "interrupts");
2711 if (len == 12) {
2712 prom_printf("Fixing bestcomm interrupts property\n");
2713 prom_setprop(node, "/builtin/bestcom", "interrupts",
2714 bcomm_irq, sizeof(bcomm_irq));
2715 }
2716 }
2717
2718 /* Fixup sound interrupts property */
2719 node = call_prom("finddevice", 1, 1, ADDR("/builtin/sound"));
2720 if (PHANDLE_VALID(node)) {
2721 rv = prom_getprop(node, "interrupts", prop, sizeof(prop));
2722 if (rv == PROM_ERROR) {
2723 prom_printf("Adding sound interrupts property\n");
2724 prom_setprop(node, "/builtin/sound", "interrupts",
2725 sound_irq, sizeof(sound_irq));
2726 }
2727 }
2728
2729 /* Make sure ethernet phy-handle property exists */
2730 fixup_device_tree_efika_add_phy();
2731}
2732#else
2733#define fixup_device_tree_efika()
2734#endif
2735
2736static void __init fixup_device_tree(void)
2737{
2738 fixup_device_tree_maple();
2739 fixup_device_tree_maple_memory_controller();
2740 fixup_device_tree_chrp();
2741 fixup_device_tree_pmac();
2742 fixup_device_tree_efika();
2743}
2744
2745static void __init prom_find_boot_cpu(void)
2746{
2747 struct prom_t *_prom = &RELOC(prom);
2748 u32 getprop_rval;
2749 ihandle prom_cpu;
2750 phandle cpu_pkg;
2751
2752 _prom->cpu = 0;
2753 if (prom_getprop(_prom->chosen, "cpu", &prom_cpu, sizeof(prom_cpu)) <= 0)
2754 return;
2755
2756 cpu_pkg = call_prom("instance-to-package", 1, 1, prom_cpu);
2757
2758 prom_getprop(cpu_pkg, "reg", &getprop_rval, sizeof(getprop_rval));
2759 _prom->cpu = getprop_rval;
2760
2761 prom_debug("Booting CPU hw index = %lu\n", _prom->cpu);
2762}
2763
2764static void __init prom_check_initrd(unsigned long r3, unsigned long r4)
2765{
2766#ifdef CONFIG_BLK_DEV_INITRD
2767 struct prom_t *_prom = &RELOC(prom);
2768
2769 if (r3 && r4 && r4 != 0xdeadbeef) {
2770 unsigned long val;
2771
2772 RELOC(prom_initrd_start) = is_kernel_addr(r3) ? __pa(r3) : r3;
2773 RELOC(prom_initrd_end) = RELOC(prom_initrd_start) + r4;
2774
2775 val = RELOC(prom_initrd_start);
2776 prom_setprop(_prom->chosen, "/chosen", "linux,initrd-start",
2777 &val, sizeof(val));
2778 val = RELOC(prom_initrd_end);
2779 prom_setprop(_prom->chosen, "/chosen", "linux,initrd-end",
2780 &val, sizeof(val));
2781
2782 reserve_mem(RELOC(prom_initrd_start),
2783 RELOC(prom_initrd_end) - RELOC(prom_initrd_start));
2784
2785 prom_debug("initrd_start=0x%x\n", RELOC(prom_initrd_start));
2786 prom_debug("initrd_end=0x%x\n", RELOC(prom_initrd_end));
2787 }
2788#endif /* CONFIG_BLK_DEV_INITRD */
2789}
2790
2791
2792/*
2793 * We enter here early on, when the Open Firmware prom is still
2794 * handling exceptions and the MMU hash table for us.
2795 */
2796
2797unsigned long __init prom_init(unsigned long r3, unsigned long r4,
2798 unsigned long pp,
2799 unsigned long r6, unsigned long r7,
2800 unsigned long kbase)
2801{
2802 struct prom_t *_prom;
2803 unsigned long hdr;
2804
2805#ifdef CONFIG_PPC32
2806 unsigned long offset = reloc_offset();
2807 reloc_got2(offset);
2808#endif
2809
2810 _prom = &RELOC(prom);
2811
2812 /*
2813 * First zero the BSS
2814 */
2815 memset(&RELOC(__bss_start), 0, __bss_stop - __bss_start);
2816
2817 /*
2818 * Init interface to Open Firmware, get some node references,
2819 * like /chosen
2820 */
2821 prom_init_client_services(pp);
2822
2823 /*
2824 * See if this OF is old enough that we need to do explicit maps
2825 * and other workarounds
2826 */
2827 prom_find_mmu();
2828
2829 /*
2830 * Init prom stdout device
2831 */
2832 prom_init_stdout();
2833
2834 prom_printf("Preparing to boot %s", RELOC(linux_banner));
2835
2836 /*
2837 * Get default machine type. At this point, we do not differentiate
2838 * between pSeries SMP and pSeries LPAR
2839 */
2840 RELOC(of_platform) = prom_find_machine_type();
2841 prom_printf("Detected machine type: %x\n", RELOC(of_platform));
2842
2843#ifndef CONFIG_NONSTATIC_KERNEL
2844 /* Bail if this is a kdump kernel. */
2845 if (PHYSICAL_START > 0)
2846 prom_panic("Error: You can't boot a kdump kernel from OF!\n");
2847#endif
2848
2849 /*
2850 * Check for an initrd
2851 */
2852 prom_check_initrd(r3, r4);
2853
2854#if defined(CONFIG_PPC_PSERIES) || defined(CONFIG_PPC_POWERNV)
2855 /*
2856 * On pSeries, inform the firmware about our capabilities
2857 */
2858 if (RELOC(of_platform) == PLATFORM_PSERIES ||
2859 RELOC(of_platform) == PLATFORM_PSERIES_LPAR)
2860 prom_send_capabilities();
2861#endif
2862
2863 /*
2864 * Copy the CPU hold code
2865 */
2866 if (RELOC(of_platform) != PLATFORM_POWERMAC)
2867 copy_and_flush(0, kbase, 0x100, 0);
2868
2869 /*
2870 * Do early parsing of command line
2871 */
2872 early_cmdline_parse();
2873
2874 /*
2875 * Initialize memory management within prom_init
2876 */
2877 prom_init_mem();
2878
2879 /*
2880 * Determine which cpu is actually running right _now_
2881 */
2882 prom_find_boot_cpu();
2883
2884 /*
2885 * Initialize display devices
2886 */
2887 prom_check_displays();
2888
2889#ifdef CONFIG_PPC64
2890 /*
2891 * Initialize IOMMU (TCE tables) on pSeries. Do that before anything else
2892 * that uses the allocator, we need to make sure we get the top of memory
2893 * available for us here...
2894 */
2895 if (RELOC(of_platform) == PLATFORM_PSERIES)
2896 prom_initialize_tce_table();
2897#endif
2898
2899 /*
2900 * On non-powermacs, try to instantiate RTAS. PowerMacs don't
2901 * have a usable RTAS implementation.
2902 */
2903 if (RELOC(of_platform) != PLATFORM_POWERMAC &&
2904 RELOC(of_platform) != PLATFORM_OPAL)
2905 prom_instantiate_rtas();
2906
2907#ifdef CONFIG_PPC_POWERNV
2908 /* Detect HAL and try instanciating it & doing takeover */
2909 if (RELOC(of_platform) == PLATFORM_PSERIES_LPAR) {
2910 prom_query_opal();
2911 if (RELOC(of_platform) == PLATFORM_OPAL) {
2912 prom_opal_hold_cpus();
2913 prom_opal_takeover();
2914 }
2915 } else if (RELOC(of_platform) == PLATFORM_OPAL)
2916 prom_instantiate_opal();
2917#endif
2918
2919 /*
2920 * On non-powermacs, put all CPUs in spin-loops.
2921 *
2922 * PowerMacs use a different mechanism to spin CPUs
2923 */
2924 if (RELOC(of_platform) != PLATFORM_POWERMAC &&
2925 RELOC(of_platform) != PLATFORM_OPAL)
2926 prom_hold_cpus();
2927
2928 /*
2929 * Fill in some infos for use by the kernel later on
2930 */
2931 if (RELOC(prom_memory_limit))
2932 prom_setprop(_prom->chosen, "/chosen", "linux,memory-limit",
2933 &RELOC(prom_memory_limit),
2934 sizeof(prom_memory_limit));
2935#ifdef CONFIG_PPC64
2936 if (RELOC(prom_iommu_off))
2937 prom_setprop(_prom->chosen, "/chosen", "linux,iommu-off",
2938 NULL, 0);
2939
2940 if (RELOC(prom_iommu_force_on))
2941 prom_setprop(_prom->chosen, "/chosen", "linux,iommu-force-on",
2942 NULL, 0);
2943
2944 if (RELOC(prom_tce_alloc_start)) {
2945 prom_setprop(_prom->chosen, "/chosen", "linux,tce-alloc-start",
2946 &RELOC(prom_tce_alloc_start),
2947 sizeof(prom_tce_alloc_start));
2948 prom_setprop(_prom->chosen, "/chosen", "linux,tce-alloc-end",
2949 &RELOC(prom_tce_alloc_end),
2950 sizeof(prom_tce_alloc_end));
2951 }
2952#endif
2953
2954 /*
2955 * Fixup any known bugs in the device-tree
2956 */
2957 fixup_device_tree();
2958
2959 /*
2960 * Now finally create the flattened device-tree
2961 */
2962 prom_printf("copying OF device tree...\n");
2963 flatten_device_tree();
2964
2965 /*
2966 * in case stdin is USB and still active on IBM machines...
2967 * Unfortunately quiesce crashes on some powermacs if we have
2968 * closed stdin already (in particular the powerbook 101). It
2969 * appears that the OPAL version of OFW doesn't like it either.
2970 */
2971 if (RELOC(of_platform) != PLATFORM_POWERMAC &&
2972 RELOC(of_platform) != PLATFORM_OPAL)
2973 prom_close_stdin();
2974
2975 /*
2976 * Call OF "quiesce" method to shut down pending DMA's from
2977 * devices etc...
2978 */
2979 prom_printf("Calling quiesce...\n");
2980 call_prom("quiesce", 0, 0);
2981
2982 /*
2983 * And finally, call the kernel passing it the flattened device
2984 * tree and NULL as r5, thus triggering the new entry point which
2985 * is common to us and kexec
2986 */
2987 hdr = RELOC(dt_header_start);
2988
2989 /* Don't print anything after quiesce under OPAL, it crashes OFW */
2990 if (RELOC(of_platform) != PLATFORM_OPAL) {
2991 prom_printf("returning from prom_init\n");
2992 prom_debug("->dt_header_start=0x%x\n", hdr);
2993 }
2994
2995#ifdef CONFIG_PPC32
2996 reloc_got2(-offset);
2997#endif
2998
2999#ifdef CONFIG_PPC_EARLY_DEBUG_OPAL
3000 /* OPAL early debug gets the OPAL base & entry in r8 and r9 */
3001 __start(hdr, kbase, 0, 0, 0,
3002 RELOC(prom_opal_base), RELOC(prom_opal_entry));
3003#else
3004 __start(hdr, kbase, 0, 0, 0, 0, 0);
3005#endif
3006
3007 return 0;
3008}