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 <linux/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 <linux/of.h>
32#include <linux/of_fdt.h>
33#include <asm/prom.h>
34#include <asm/rtas.h>
35#include <asm/page.h>
36#include <asm/processor.h>
37#include <asm/interrupt.h>
38#include <asm/irq.h>
39#include <asm/io.h>
40#include <asm/smp.h>
41#include <asm/mmu.h>
42#include <asm/iommu.h>
43#include <asm/btext.h>
44#include <asm/sections.h>
45#include <asm/setup.h>
46#include <asm/asm-prototypes.h>
47#include <asm/ultravisor-api.h>
48
49#include <linux/linux_logo.h>
50
51/* All of prom_init bss lives here */
52#define __prombss __section(".bss.prominit")
53
54/*
55 * Eventually bump that one up
56 */
57#define DEVTREE_CHUNK_SIZE 0x100000
58
59/*
60 * This is the size of the local memory reserve map that gets copied
61 * into the boot params passed to the kernel. That size is totally
62 * flexible as the kernel just reads the list until it encounters an
63 * entry with size 0, so it can be changed without breaking binary
64 * compatibility
65 */
66#define MEM_RESERVE_MAP_SIZE 8
67
68/*
69 * prom_init() is called very early on, before the kernel text
70 * and data have been mapped to KERNELBASE. At this point the code
71 * is running at whatever address it has been loaded at.
72 * On ppc32 we compile with -mrelocatable, which means that references
73 * to extern and static variables get relocated automatically.
74 * ppc64 objects are always relocatable, we just need to relocate the
75 * TOC.
76 *
77 * Because OF may have mapped I/O devices into the area starting at
78 * KERNELBASE, particularly on CHRP machines, we can't safely call
79 * OF once the kernel has been mapped to KERNELBASE. Therefore all
80 * OF calls must be done within prom_init().
81 *
82 * ADDR is used in calls to call_prom. The 4th and following
83 * arguments to call_prom should be 32-bit values.
84 * On ppc64, 64 bit values are truncated to 32 bits (and
85 * fortunately don't get interpreted as two arguments).
86 */
87#define ADDR(x) (u32)(unsigned long)(x)
88
89#ifdef CONFIG_PPC64
90#define OF_WORKAROUNDS 0
91#else
92#define OF_WORKAROUNDS of_workarounds
93static int of_workarounds __prombss;
94#endif
95
96#define OF_WA_CLAIM 1 /* do phys/virt claim separately, then map */
97#define OF_WA_LONGTRAIL 2 /* work around longtrail bugs */
98
99#ifdef DEBUG_PROM
100#define prom_debug(x...) prom_printf(x)
101#else
102#define prom_debug(x...) do { } while (0)
103#endif
104
105
106typedef u32 prom_arg_t;
107
108struct prom_args {
109 __be32 service;
110 __be32 nargs;
111 __be32 nret;
112 __be32 args[10];
113};
114
115struct prom_t {
116 ihandle root;
117 phandle chosen;
118 int cpu;
119 ihandle stdout;
120 ihandle mmumap;
121 ihandle memory;
122};
123
124struct mem_map_entry {
125 __be64 base;
126 __be64 size;
127};
128
129typedef __be32 cell_t;
130
131extern void __start(unsigned long r3, unsigned long r4, unsigned long r5,
132 unsigned long r6, unsigned long r7, unsigned long r8,
133 unsigned long r9);
134
135#ifdef CONFIG_PPC64
136extern int enter_prom(struct prom_args *args, unsigned long entry);
137#else
138static inline int enter_prom(struct prom_args *args, unsigned long entry)
139{
140 return ((int (*)(struct prom_args *))entry)(args);
141}
142#endif
143
144extern void copy_and_flush(unsigned long dest, unsigned long src,
145 unsigned long size, unsigned long offset);
146
147/* prom structure */
148static struct prom_t __prombss prom;
149
150static unsigned long __prombss prom_entry;
151
152static char __prombss of_stdout_device[256];
153static char __prombss prom_scratch[256];
154
155static unsigned long __prombss dt_header_start;
156static unsigned long __prombss dt_struct_start, dt_struct_end;
157static unsigned long __prombss dt_string_start, dt_string_end;
158
159static unsigned long __prombss prom_initrd_start, prom_initrd_end;
160
161#ifdef CONFIG_PPC64
162static int __prombss prom_iommu_force_on;
163static int __prombss prom_iommu_off;
164static unsigned long __prombss prom_tce_alloc_start;
165static unsigned long __prombss prom_tce_alloc_end;
166#endif
167
168#ifdef CONFIG_PPC_PSERIES
169static bool __prombss prom_radix_disable;
170static bool __prombss prom_radix_gtse_disable;
171static bool __prombss prom_xive_disable;
172#endif
173
174#ifdef CONFIG_PPC_SVM
175static bool __prombss prom_svm_enable;
176#endif
177
178struct platform_support {
179 bool hash_mmu;
180 bool radix_mmu;
181 bool radix_gtse;
182 bool xive;
183};
184
185/* Platforms codes are now obsolete in the kernel. Now only used within this
186 * file and ultimately gone too. Feel free to change them if you need, they
187 * are not shared with anything outside of this file anymore
188 */
189#define PLATFORM_PSERIES 0x0100
190#define PLATFORM_PSERIES_LPAR 0x0101
191#define PLATFORM_LPAR 0x0001
192#define PLATFORM_POWERMAC 0x0400
193#define PLATFORM_GENERIC 0x0500
194
195static int __prombss of_platform;
196
197static char __prombss prom_cmd_line[COMMAND_LINE_SIZE];
198
199static unsigned long __prombss prom_memory_limit;
200
201static unsigned long __prombss alloc_top;
202static unsigned long __prombss alloc_top_high;
203static unsigned long __prombss alloc_bottom;
204static unsigned long __prombss rmo_top;
205static unsigned long __prombss ram_top;
206
207static struct mem_map_entry __prombss mem_reserve_map[MEM_RESERVE_MAP_SIZE];
208static int __prombss mem_reserve_cnt;
209
210static cell_t __prombss regbuf[1024];
211
212static bool __prombss rtas_has_query_cpu_stopped;
213
214
215/*
216 * Error results ... some OF calls will return "-1" on error, some
217 * will return 0, some will return either. To simplify, here are
218 * macros to use with any ihandle or phandle return value to check if
219 * it is valid
220 */
221
222#define PROM_ERROR (-1u)
223#define PHANDLE_VALID(p) ((p) != 0 && (p) != PROM_ERROR)
224#define IHANDLE_VALID(i) ((i) != 0 && (i) != PROM_ERROR)
225
226/* Copied from lib/string.c and lib/kstrtox.c */
227
228static int __init prom_strcmp(const char *cs, const char *ct)
229{
230 unsigned char c1, c2;
231
232 while (1) {
233 c1 = *cs++;
234 c2 = *ct++;
235 if (c1 != c2)
236 return c1 < c2 ? -1 : 1;
237 if (!c1)
238 break;
239 }
240 return 0;
241}
242
243static ssize_t __init prom_strscpy_pad(char *dest, const char *src, size_t n)
244{
245 ssize_t rc;
246 size_t i;
247
248 if (n == 0 || n > INT_MAX)
249 return -E2BIG;
250
251 // Copy up to n bytes
252 for (i = 0; i < n && src[i] != '\0'; i++)
253 dest[i] = src[i];
254
255 rc = i;
256
257 // If we copied all n then we have run out of space for the nul
258 if (rc == n) {
259 // Rewind by one character to ensure nul termination
260 i--;
261 rc = -E2BIG;
262 }
263
264 for (; i < n; i++)
265 dest[i] = '\0';
266
267 return rc;
268}
269
270static int __init prom_strncmp(const char *cs, const char *ct, size_t count)
271{
272 unsigned char c1, c2;
273
274 while (count) {
275 c1 = *cs++;
276 c2 = *ct++;
277 if (c1 != c2)
278 return c1 < c2 ? -1 : 1;
279 if (!c1)
280 break;
281 count--;
282 }
283 return 0;
284}
285
286static size_t __init prom_strlen(const char *s)
287{
288 const char *sc;
289
290 for (sc = s; *sc != '\0'; ++sc)
291 /* nothing */;
292 return sc - s;
293}
294
295static int __init prom_memcmp(const void *cs, const void *ct, size_t count)
296{
297 const unsigned char *su1, *su2;
298 int res = 0;
299
300 for (su1 = cs, su2 = ct; 0 < count; ++su1, ++su2, count--)
301 if ((res = *su1 - *su2) != 0)
302 break;
303 return res;
304}
305
306static char __init *prom_strstr(const char *s1, const char *s2)
307{
308 size_t l1, l2;
309
310 l2 = prom_strlen(s2);
311 if (!l2)
312 return (char *)s1;
313 l1 = prom_strlen(s1);
314 while (l1 >= l2) {
315 l1--;
316 if (!prom_memcmp(s1, s2, l2))
317 return (char *)s1;
318 s1++;
319 }
320 return NULL;
321}
322
323static size_t __init prom_strlcat(char *dest, const char *src, size_t count)
324{
325 size_t dsize = prom_strlen(dest);
326 size_t len = prom_strlen(src);
327 size_t res = dsize + len;
328
329 /* This would be a bug */
330 if (dsize >= count)
331 return count;
332
333 dest += dsize;
334 count -= dsize;
335 if (len >= count)
336 len = count-1;
337 memcpy(dest, src, len);
338 dest[len] = 0;
339 return res;
340
341}
342
343#ifdef CONFIG_PPC_PSERIES
344static int __init prom_strtobool(const char *s, bool *res)
345{
346 if (!s)
347 return -EINVAL;
348
349 switch (s[0]) {
350 case 'y':
351 case 'Y':
352 case '1':
353 *res = true;
354 return 0;
355 case 'n':
356 case 'N':
357 case '0':
358 *res = false;
359 return 0;
360 case 'o':
361 case 'O':
362 switch (s[1]) {
363 case 'n':
364 case 'N':
365 *res = true;
366 return 0;
367 case 'f':
368 case 'F':
369 *res = false;
370 return 0;
371 default:
372 break;
373 }
374 break;
375 default:
376 break;
377 }
378
379 return -EINVAL;
380}
381#endif
382
383/* This is the one and *ONLY* place where we actually call open
384 * firmware.
385 */
386
387static int __init call_prom(const char *service, int nargs, int nret, ...)
388{
389 int i;
390 struct prom_args args;
391 va_list list;
392
393 args.service = cpu_to_be32(ADDR(service));
394 args.nargs = cpu_to_be32(nargs);
395 args.nret = cpu_to_be32(nret);
396
397 va_start(list, nret);
398 for (i = 0; i < nargs; i++)
399 args.args[i] = cpu_to_be32(va_arg(list, prom_arg_t));
400 va_end(list);
401
402 for (i = 0; i < nret; i++)
403 args.args[nargs+i] = 0;
404
405 if (enter_prom(&args, prom_entry) < 0)
406 return PROM_ERROR;
407
408 return (nret > 0) ? be32_to_cpu(args.args[nargs]) : 0;
409}
410
411static int __init call_prom_ret(const char *service, int nargs, int nret,
412 prom_arg_t *rets, ...)
413{
414 int i;
415 struct prom_args args;
416 va_list list;
417
418 args.service = cpu_to_be32(ADDR(service));
419 args.nargs = cpu_to_be32(nargs);
420 args.nret = cpu_to_be32(nret);
421
422 va_start(list, rets);
423 for (i = 0; i < nargs; i++)
424 args.args[i] = cpu_to_be32(va_arg(list, prom_arg_t));
425 va_end(list);
426
427 for (i = 0; i < nret; i++)
428 args.args[nargs+i] = 0;
429
430 if (enter_prom(&args, prom_entry) < 0)
431 return PROM_ERROR;
432
433 if (rets != NULL)
434 for (i = 1; i < nret; ++i)
435 rets[i-1] = be32_to_cpu(args.args[nargs+i]);
436
437 return (nret > 0) ? be32_to_cpu(args.args[nargs]) : 0;
438}
439
440
441static void __init prom_print(const char *msg)
442{
443 const char *p, *q;
444
445 if (prom.stdout == 0)
446 return;
447
448 for (p = msg; *p != 0; p = q) {
449 for (q = p; *q != 0 && *q != '\n'; ++q)
450 ;
451 if (q > p)
452 call_prom("write", 3, 1, prom.stdout, p, q - p);
453 if (*q == 0)
454 break;
455 ++q;
456 call_prom("write", 3, 1, prom.stdout, ADDR("\r\n"), 2);
457 }
458}
459
460
461/*
462 * Both prom_print_hex & prom_print_dec takes an unsigned long as input so that
463 * we do not need __udivdi3 or __umoddi3 on 32bits.
464 */
465static void __init prom_print_hex(unsigned long val)
466{
467 int i, nibbles = sizeof(val)*2;
468 char buf[sizeof(val)*2+1];
469
470 for (i = nibbles-1; i >= 0; i--) {
471 buf[i] = (val & 0xf) + '0';
472 if (buf[i] > '9')
473 buf[i] += ('a'-'0'-10);
474 val >>= 4;
475 }
476 buf[nibbles] = '\0';
477 call_prom("write", 3, 1, prom.stdout, buf, nibbles);
478}
479
480/* max number of decimal digits in an unsigned long */
481#define UL_DIGITS 21
482static void __init prom_print_dec(unsigned long val)
483{
484 int i, size;
485 char buf[UL_DIGITS+1];
486
487 for (i = UL_DIGITS-1; i >= 0; i--) {
488 buf[i] = (val % 10) + '0';
489 val = val/10;
490 if (val == 0)
491 break;
492 }
493 /* shift stuff down */
494 size = UL_DIGITS - i;
495 call_prom("write", 3, 1, prom.stdout, buf+i, size);
496}
497
498__printf(1, 2)
499static void __init prom_printf(const char *format, ...)
500{
501 const char *p, *q, *s;
502 va_list args;
503 unsigned long v;
504 long vs;
505 int n = 0;
506
507 va_start(args, format);
508 for (p = format; *p != 0; p = q) {
509 for (q = p; *q != 0 && *q != '\n' && *q != '%'; ++q)
510 ;
511 if (q > p)
512 call_prom("write", 3, 1, prom.stdout, p, q - p);
513 if (*q == 0)
514 break;
515 if (*q == '\n') {
516 ++q;
517 call_prom("write", 3, 1, prom.stdout,
518 ADDR("\r\n"), 2);
519 continue;
520 }
521 ++q;
522 if (*q == 0)
523 break;
524 while (*q == 'l') {
525 ++q;
526 ++n;
527 }
528 switch (*q) {
529 case 's':
530 ++q;
531 s = va_arg(args, const char *);
532 prom_print(s);
533 break;
534 case 'x':
535 ++q;
536 switch (n) {
537 case 0:
538 v = va_arg(args, unsigned int);
539 break;
540 case 1:
541 v = va_arg(args, unsigned long);
542 break;
543 case 2:
544 default:
545 v = va_arg(args, unsigned long long);
546 break;
547 }
548 prom_print_hex(v);
549 break;
550 case 'u':
551 ++q;
552 switch (n) {
553 case 0:
554 v = va_arg(args, unsigned int);
555 break;
556 case 1:
557 v = va_arg(args, unsigned long);
558 break;
559 case 2:
560 default:
561 v = va_arg(args, unsigned long long);
562 break;
563 }
564 prom_print_dec(v);
565 break;
566 case 'd':
567 ++q;
568 switch (n) {
569 case 0:
570 vs = va_arg(args, int);
571 break;
572 case 1:
573 vs = va_arg(args, long);
574 break;
575 case 2:
576 default:
577 vs = va_arg(args, long long);
578 break;
579 }
580 if (vs < 0) {
581 prom_print("-");
582 vs = -vs;
583 }
584 prom_print_dec(vs);
585 break;
586 }
587 }
588 va_end(args);
589}
590
591
592static unsigned int __init prom_claim(unsigned long virt, unsigned long size,
593 unsigned long align)
594{
595
596 if (align == 0 && (OF_WORKAROUNDS & OF_WA_CLAIM)) {
597 /*
598 * Old OF requires we claim physical and virtual separately
599 * and then map explicitly (assuming virtual mode)
600 */
601 int ret;
602 prom_arg_t result;
603
604 ret = call_prom_ret("call-method", 5, 2, &result,
605 ADDR("claim"), prom.memory,
606 align, size, virt);
607 if (ret != 0 || result == -1)
608 return -1;
609 ret = call_prom_ret("call-method", 5, 2, &result,
610 ADDR("claim"), prom.mmumap,
611 align, size, virt);
612 if (ret != 0) {
613 call_prom("call-method", 4, 1, ADDR("release"),
614 prom.memory, size, virt);
615 return -1;
616 }
617 /* the 0x12 is M (coherence) + PP == read/write */
618 call_prom("call-method", 6, 1,
619 ADDR("map"), prom.mmumap, 0x12, size, virt, virt);
620 return virt;
621 }
622 return call_prom("claim", 3, 1, (prom_arg_t)virt, (prom_arg_t)size,
623 (prom_arg_t)align);
624}
625
626static void __init __attribute__((noreturn)) prom_panic(const char *reason)
627{
628 prom_print(reason);
629 /* Do not call exit because it clears the screen on pmac
630 * it also causes some sort of double-fault on early pmacs */
631 if (of_platform == PLATFORM_POWERMAC)
632 asm("trap\n");
633
634 /* ToDo: should put up an SRC here on pSeries */
635 call_prom("exit", 0, 0);
636
637 for (;;) /* should never get here */
638 ;
639}
640
641
642static int __init prom_next_node(phandle *nodep)
643{
644 phandle node;
645
646 if ((node = *nodep) != 0
647 && (*nodep = call_prom("child", 1, 1, node)) != 0)
648 return 1;
649 if ((*nodep = call_prom("peer", 1, 1, node)) != 0)
650 return 1;
651 for (;;) {
652 if ((node = call_prom("parent", 1, 1, node)) == 0)
653 return 0;
654 if ((*nodep = call_prom("peer", 1, 1, node)) != 0)
655 return 1;
656 }
657}
658
659static inline int __init prom_getprop(phandle node, const char *pname,
660 void *value, size_t valuelen)
661{
662 return call_prom("getprop", 4, 1, node, ADDR(pname),
663 (u32)(unsigned long) value, (u32) valuelen);
664}
665
666static inline int __init prom_getproplen(phandle node, const char *pname)
667{
668 return call_prom("getproplen", 2, 1, node, ADDR(pname));
669}
670
671static void __init add_string(char **str, const char *q)
672{
673 char *p = *str;
674
675 while (*q)
676 *p++ = *q++;
677 *p++ = ' ';
678 *str = p;
679}
680
681static char *__init tohex(unsigned int x)
682{
683 static const char digits[] __initconst = "0123456789abcdef";
684 static char result[9] __prombss;
685 int i;
686
687 result[8] = 0;
688 i = 8;
689 do {
690 --i;
691 result[i] = digits[x & 0xf];
692 x >>= 4;
693 } while (x != 0 && i > 0);
694 return &result[i];
695}
696
697static int __init prom_setprop(phandle node, const char *nodename,
698 const char *pname, void *value, size_t valuelen)
699{
700 char cmd[256], *p;
701
702 if (!(OF_WORKAROUNDS & OF_WA_LONGTRAIL))
703 return call_prom("setprop", 4, 1, node, ADDR(pname),
704 (u32)(unsigned long) value, (u32) valuelen);
705
706 /* gah... setprop doesn't work on longtrail, have to use interpret */
707 p = cmd;
708 add_string(&p, "dev");
709 add_string(&p, nodename);
710 add_string(&p, tohex((u32)(unsigned long) value));
711 add_string(&p, tohex(valuelen));
712 add_string(&p, tohex(ADDR(pname)));
713 add_string(&p, tohex(prom_strlen(pname)));
714 add_string(&p, "property");
715 *p = 0;
716 return call_prom("interpret", 1, 1, (u32)(unsigned long) cmd);
717}
718
719/* We can't use the standard versions because of relocation headaches. */
720#define prom_isxdigit(c) \
721 (('0' <= (c) && (c) <= '9') || ('a' <= (c) && (c) <= 'f') || ('A' <= (c) && (c) <= 'F'))
722
723#define prom_isdigit(c) ('0' <= (c) && (c) <= '9')
724#define prom_islower(c) ('a' <= (c) && (c) <= 'z')
725#define prom_toupper(c) (prom_islower(c) ? ((c) - 'a' + 'A') : (c))
726
727static unsigned long __init prom_strtoul(const char *cp, const char **endp)
728{
729 unsigned long result = 0, base = 10, value;
730
731 if (*cp == '0') {
732 base = 8;
733 cp++;
734 if (prom_toupper(*cp) == 'X') {
735 cp++;
736 base = 16;
737 }
738 }
739
740 while (prom_isxdigit(*cp) &&
741 (value = prom_isdigit(*cp) ? *cp - '0' : prom_toupper(*cp) - 'A' + 10) < base) {
742 result = result * base + value;
743 cp++;
744 }
745
746 if (endp)
747 *endp = cp;
748
749 return result;
750}
751
752static unsigned long __init prom_memparse(const char *ptr, const char **retptr)
753{
754 unsigned long ret = prom_strtoul(ptr, retptr);
755 int shift = 0;
756
757 /*
758 * We can't use a switch here because GCC *may* generate a
759 * jump table which won't work, because we're not running at
760 * the address we're linked at.
761 */
762 if ('G' == **retptr || 'g' == **retptr)
763 shift = 30;
764
765 if ('M' == **retptr || 'm' == **retptr)
766 shift = 20;
767
768 if ('K' == **retptr || 'k' == **retptr)
769 shift = 10;
770
771 if (shift) {
772 ret <<= shift;
773 (*retptr)++;
774 }
775
776 return ret;
777}
778
779/*
780 * Early parsing of the command line passed to the kernel, used for
781 * "mem=x" and the options that affect the iommu
782 */
783static void __init early_cmdline_parse(void)
784{
785 const char *opt;
786
787 char *p;
788 int l = 0;
789
790 prom_cmd_line[0] = 0;
791 p = prom_cmd_line;
792
793 if (!IS_ENABLED(CONFIG_CMDLINE_FORCE) && (long)prom.chosen > 0)
794 l = prom_getprop(prom.chosen, "bootargs", p, COMMAND_LINE_SIZE-1);
795
796 if (IS_ENABLED(CONFIG_CMDLINE_EXTEND) || l <= 0 || p[0] == '\0')
797 prom_strlcat(prom_cmd_line, " " CONFIG_CMDLINE,
798 sizeof(prom_cmd_line));
799
800 prom_printf("command line: %s\n", prom_cmd_line);
801
802#ifdef CONFIG_PPC64
803 opt = prom_strstr(prom_cmd_line, "iommu=");
804 if (opt) {
805 prom_printf("iommu opt is: %s\n", opt);
806 opt += 6;
807 while (*opt && *opt == ' ')
808 opt++;
809 if (!prom_strncmp(opt, "off", 3))
810 prom_iommu_off = 1;
811 else if (!prom_strncmp(opt, "force", 5))
812 prom_iommu_force_on = 1;
813 }
814#endif
815 opt = prom_strstr(prom_cmd_line, "mem=");
816 if (opt) {
817 opt += 4;
818 prom_memory_limit = prom_memparse(opt, (const char **)&opt);
819#ifdef CONFIG_PPC64
820 /* Align down to 16 MB which is large page size with hash page translation */
821 prom_memory_limit = ALIGN_DOWN(prom_memory_limit, SZ_16M);
822#endif
823 }
824
825#ifdef CONFIG_PPC_PSERIES
826 prom_radix_disable = !IS_ENABLED(CONFIG_PPC_RADIX_MMU_DEFAULT);
827 opt = prom_strstr(prom_cmd_line, "disable_radix");
828 if (opt) {
829 opt += 13;
830 if (*opt && *opt == '=') {
831 bool val;
832
833 if (prom_strtobool(++opt, &val))
834 prom_radix_disable = false;
835 else
836 prom_radix_disable = val;
837 } else
838 prom_radix_disable = true;
839 }
840 if (prom_radix_disable)
841 prom_debug("Radix disabled from cmdline\n");
842
843 opt = prom_strstr(prom_cmd_line, "radix_hcall_invalidate=on");
844 if (opt) {
845 prom_radix_gtse_disable = true;
846 prom_debug("Radix GTSE disabled from cmdline\n");
847 }
848
849 opt = prom_strstr(prom_cmd_line, "xive=off");
850 if (opt) {
851 prom_xive_disable = true;
852 prom_debug("XIVE disabled from cmdline\n");
853 }
854#endif /* CONFIG_PPC_PSERIES */
855
856#ifdef CONFIG_PPC_SVM
857 opt = prom_strstr(prom_cmd_line, "svm=");
858 if (opt) {
859 bool val;
860
861 opt += sizeof("svm=") - 1;
862 if (!prom_strtobool(opt, &val))
863 prom_svm_enable = val;
864 }
865#endif /* CONFIG_PPC_SVM */
866}
867
868#ifdef CONFIG_PPC_PSERIES
869/*
870 * The architecture vector has an array of PVR mask/value pairs,
871 * followed by # option vectors - 1, followed by the option vectors.
872 *
873 * See prom.h for the definition of the bits specified in the
874 * architecture vector.
875 */
876
877/* Firmware expects the value to be n - 1, where n is the # of vectors */
878#define NUM_VECTORS(n) ((n) - 1)
879
880/*
881 * Firmware expects 1 + n - 2, where n is the length of the option vector in
882 * bytes. The 1 accounts for the length byte itself, the - 2 .. ?
883 */
884#define VECTOR_LENGTH(n) (1 + (n) - 2)
885
886struct option_vector1 {
887 u8 byte1;
888 u8 arch_versions;
889 u8 arch_versions3;
890} __packed;
891
892struct option_vector2 {
893 u8 byte1;
894 __be16 reserved;
895 __be32 real_base;
896 __be32 real_size;
897 __be32 virt_base;
898 __be32 virt_size;
899 __be32 load_base;
900 __be32 min_rma;
901 __be32 min_load;
902 u8 min_rma_percent;
903 u8 max_pft_size;
904} __packed;
905
906struct option_vector3 {
907 u8 byte1;
908 u8 byte2;
909} __packed;
910
911struct option_vector4 {
912 u8 byte1;
913 u8 min_vp_cap;
914} __packed;
915
916struct option_vector5 {
917 u8 byte1;
918 u8 byte2;
919 u8 byte3;
920 u8 cmo;
921 u8 associativity;
922 u8 bin_opts;
923 u8 micro_checkpoint;
924 u8 reserved0;
925 __be32 max_cpus;
926 __be16 papr_level;
927 __be16 reserved1;
928 u8 platform_facilities;
929 u8 reserved2;
930 __be16 reserved3;
931 u8 subprocessors;
932 u8 byte22;
933 u8 intarch;
934 u8 mmu;
935 u8 hash_ext;
936 u8 radix_ext;
937} __packed;
938
939struct option_vector6 {
940 u8 reserved;
941 u8 secondary_pteg;
942 u8 os_name;
943} __packed;
944
945struct option_vector7 {
946 u8 os_id[256];
947} __packed;
948
949struct ibm_arch_vec {
950 struct { __be32 mask, val; } pvrs[16];
951
952 u8 num_vectors;
953
954 u8 vec1_len;
955 struct option_vector1 vec1;
956
957 u8 vec2_len;
958 struct option_vector2 vec2;
959
960 u8 vec3_len;
961 struct option_vector3 vec3;
962
963 u8 vec4_len;
964 struct option_vector4 vec4;
965
966 u8 vec5_len;
967 struct option_vector5 vec5;
968
969 u8 vec6_len;
970 struct option_vector6 vec6;
971
972 u8 vec7_len;
973 struct option_vector7 vec7;
974} __packed;
975
976static const struct ibm_arch_vec ibm_architecture_vec_template __initconst = {
977 .pvrs = {
978 {
979 .mask = cpu_to_be32(0xfffe0000), /* POWER5/POWER5+ */
980 .val = cpu_to_be32(0x003a0000),
981 },
982 {
983 .mask = cpu_to_be32(0xffff0000), /* POWER6 */
984 .val = cpu_to_be32(0x003e0000),
985 },
986 {
987 .mask = cpu_to_be32(0xffff0000), /* POWER7 */
988 .val = cpu_to_be32(0x003f0000),
989 },
990 {
991 .mask = cpu_to_be32(0xffff0000), /* POWER8E */
992 .val = cpu_to_be32(0x004b0000),
993 },
994 {
995 .mask = cpu_to_be32(0xffff0000), /* POWER8NVL */
996 .val = cpu_to_be32(0x004c0000),
997 },
998 {
999 .mask = cpu_to_be32(0xffff0000), /* POWER8 */
1000 .val = cpu_to_be32(0x004d0000),
1001 },
1002 {
1003 .mask = cpu_to_be32(0xffff0000), /* POWER9 */
1004 .val = cpu_to_be32(0x004e0000),
1005 },
1006 {
1007 .mask = cpu_to_be32(0xffff0000), /* POWER10 */
1008 .val = cpu_to_be32(0x00800000),
1009 },
1010 {
1011 .mask = cpu_to_be32(0xffff0000), /* POWER11 */
1012 .val = cpu_to_be32(0x00820000),
1013 },
1014 {
1015 .mask = cpu_to_be32(0xffffffff), /* P11 compliant */
1016 .val = cpu_to_be32(0x0f000007),
1017 },
1018 {
1019 .mask = cpu_to_be32(0xffffffff), /* all 3.1-compliant */
1020 .val = cpu_to_be32(0x0f000006),
1021 },
1022 {
1023 .mask = cpu_to_be32(0xffffffff), /* all 3.00-compliant */
1024 .val = cpu_to_be32(0x0f000005),
1025 },
1026 {
1027 .mask = cpu_to_be32(0xffffffff), /* all 2.07-compliant */
1028 .val = cpu_to_be32(0x0f000004),
1029 },
1030 {
1031 .mask = cpu_to_be32(0xffffffff), /* all 2.06-compliant */
1032 .val = cpu_to_be32(0x0f000003),
1033 },
1034 {
1035 .mask = cpu_to_be32(0xffffffff), /* all 2.05-compliant */
1036 .val = cpu_to_be32(0x0f000002),
1037 },
1038 {
1039 .mask = cpu_to_be32(0xfffffffe), /* all 2.04-compliant and earlier */
1040 .val = cpu_to_be32(0x0f000001),
1041 },
1042 },
1043
1044 .num_vectors = NUM_VECTORS(6),
1045
1046 .vec1_len = VECTOR_LENGTH(sizeof(struct option_vector1)),
1047 .vec1 = {
1048 .byte1 = 0,
1049 .arch_versions = OV1_PPC_2_00 | OV1_PPC_2_01 | OV1_PPC_2_02 | OV1_PPC_2_03 |
1050 OV1_PPC_2_04 | OV1_PPC_2_05 | OV1_PPC_2_06 | OV1_PPC_2_07,
1051 .arch_versions3 = OV1_PPC_3_00 | OV1_PPC_3_1,
1052 },
1053
1054 .vec2_len = VECTOR_LENGTH(sizeof(struct option_vector2)),
1055 /* option vector 2: Open Firmware options supported */
1056 .vec2 = {
1057 .byte1 = OV2_REAL_MODE,
1058 .reserved = 0,
1059 .real_base = cpu_to_be32(0xffffffff),
1060 .real_size = cpu_to_be32(0xffffffff),
1061 .virt_base = cpu_to_be32(0xffffffff),
1062 .virt_size = cpu_to_be32(0xffffffff),
1063 .load_base = cpu_to_be32(0xffffffff),
1064 .min_rma = cpu_to_be32(512), /* 512MB min RMA */
1065 .min_load = cpu_to_be32(0xffffffff), /* full client load */
1066 .min_rma_percent = 0, /* min RMA percentage of total RAM */
1067 .max_pft_size = 48, /* max log_2(hash table size) */
1068 },
1069
1070 .vec3_len = VECTOR_LENGTH(sizeof(struct option_vector3)),
1071 /* option vector 3: processor options supported */
1072 .vec3 = {
1073 .byte1 = 0, /* don't ignore, don't halt */
1074 .byte2 = OV3_FP | OV3_VMX | OV3_DFP,
1075 },
1076
1077 .vec4_len = VECTOR_LENGTH(sizeof(struct option_vector4)),
1078 /* option vector 4: IBM PAPR implementation */
1079 .vec4 = {
1080 .byte1 = 0, /* don't halt */
1081 .min_vp_cap = OV4_MIN_ENT_CAP, /* minimum VP entitled capacity */
1082 },
1083
1084 .vec5_len = VECTOR_LENGTH(sizeof(struct option_vector5)),
1085 /* option vector 5: PAPR/OF options */
1086 .vec5 = {
1087 .byte1 = 0, /* don't ignore, don't halt */
1088 .byte2 = OV5_FEAT(OV5_LPAR) | OV5_FEAT(OV5_SPLPAR) | OV5_FEAT(OV5_LARGE_PAGES) |
1089 OV5_FEAT(OV5_DRCONF_MEMORY) | OV5_FEAT(OV5_DONATE_DEDICATE_CPU) |
1090#ifdef CONFIG_PCI_MSI
1091 /* PCIe/MSI support. Without MSI full PCIe is not supported */
1092 OV5_FEAT(OV5_MSI),
1093#else
1094 0,
1095#endif
1096 .byte3 = 0,
1097 .cmo =
1098#ifdef CONFIG_PPC_SMLPAR
1099 OV5_FEAT(OV5_CMO) | OV5_FEAT(OV5_XCMO),
1100#else
1101 0,
1102#endif
1103 .associativity = OV5_FEAT(OV5_FORM1_AFFINITY) | OV5_FEAT(OV5_PRRN) |
1104 OV5_FEAT(OV5_FORM2_AFFINITY),
1105 .bin_opts = OV5_FEAT(OV5_RESIZE_HPT) | OV5_FEAT(OV5_HP_EVT),
1106 .micro_checkpoint = 0,
1107 .reserved0 = 0,
1108 .max_cpus = cpu_to_be32(NR_CPUS), /* number of cores supported */
1109 .papr_level = 0,
1110 .reserved1 = 0,
1111 .platform_facilities = OV5_FEAT(OV5_PFO_HW_RNG) | OV5_FEAT(OV5_PFO_HW_ENCR) | OV5_FEAT(OV5_PFO_HW_842),
1112 .reserved2 = 0,
1113 .reserved3 = 0,
1114 .subprocessors = 1,
1115 .byte22 = OV5_FEAT(OV5_DRMEM_V2) | OV5_FEAT(OV5_DRC_INFO),
1116 .intarch = 0,
1117 .mmu = 0,
1118 .hash_ext = 0,
1119 .radix_ext = 0,
1120 },
1121
1122 /* option vector 6: IBM PAPR hints */
1123 .vec6_len = VECTOR_LENGTH(sizeof(struct option_vector6)),
1124 .vec6 = {
1125 .reserved = 0,
1126 .secondary_pteg = 0,
1127 .os_name = OV6_LINUX,
1128 },
1129
1130 /* option vector 7: OS Identification */
1131 .vec7_len = VECTOR_LENGTH(sizeof(struct option_vector7)),
1132};
1133
1134static struct ibm_arch_vec __prombss ibm_architecture_vec ____cacheline_aligned;
1135
1136/* Old method - ELF header with PT_NOTE sections only works on BE */
1137#ifdef __BIG_ENDIAN__
1138static const struct fake_elf {
1139 Elf32_Ehdr elfhdr;
1140 Elf32_Phdr phdr[2];
1141 struct chrpnote {
1142 u32 namesz;
1143 u32 descsz;
1144 u32 type;
1145 char name[8]; /* "PowerPC" */
1146 struct chrpdesc {
1147 u32 real_mode;
1148 u32 real_base;
1149 u32 real_size;
1150 u32 virt_base;
1151 u32 virt_size;
1152 u32 load_base;
1153 } chrpdesc;
1154 } chrpnote;
1155 struct rpanote {
1156 u32 namesz;
1157 u32 descsz;
1158 u32 type;
1159 char name[24]; /* "IBM,RPA-Client-Config" */
1160 struct rpadesc {
1161 u32 lpar_affinity;
1162 u32 min_rmo_size;
1163 u32 min_rmo_percent;
1164 u32 max_pft_size;
1165 u32 splpar;
1166 u32 min_load;
1167 u32 new_mem_def;
1168 u32 ignore_me;
1169 } rpadesc;
1170 } rpanote;
1171} fake_elf __initconst = {
1172 .elfhdr = {
1173 .e_ident = { 0x7f, 'E', 'L', 'F',
1174 ELFCLASS32, ELFDATA2MSB, EV_CURRENT },
1175 .e_type = ET_EXEC, /* yeah right */
1176 .e_machine = EM_PPC,
1177 .e_version = EV_CURRENT,
1178 .e_phoff = offsetof(struct fake_elf, phdr),
1179 .e_phentsize = sizeof(Elf32_Phdr),
1180 .e_phnum = 2
1181 },
1182 .phdr = {
1183 [0] = {
1184 .p_type = PT_NOTE,
1185 .p_offset = offsetof(struct fake_elf, chrpnote),
1186 .p_filesz = sizeof(struct chrpnote)
1187 }, [1] = {
1188 .p_type = PT_NOTE,
1189 .p_offset = offsetof(struct fake_elf, rpanote),
1190 .p_filesz = sizeof(struct rpanote)
1191 }
1192 },
1193 .chrpnote = {
1194 .namesz = sizeof("PowerPC"),
1195 .descsz = sizeof(struct chrpdesc),
1196 .type = 0x1275,
1197 .name = "PowerPC",
1198 .chrpdesc = {
1199 .real_mode = ~0U, /* ~0 means "don't care" */
1200 .real_base = ~0U,
1201 .real_size = ~0U,
1202 .virt_base = ~0U,
1203 .virt_size = ~0U,
1204 .load_base = ~0U
1205 },
1206 },
1207 .rpanote = {
1208 .namesz = sizeof("IBM,RPA-Client-Config"),
1209 .descsz = sizeof(struct rpadesc),
1210 .type = 0x12759999,
1211 .name = "IBM,RPA-Client-Config",
1212 .rpadesc = {
1213 .lpar_affinity = 0,
1214 .min_rmo_size = 64, /* in megabytes */
1215 .min_rmo_percent = 0,
1216 .max_pft_size = 48, /* 2^48 bytes max PFT size */
1217 .splpar = 1,
1218 .min_load = ~0U,
1219 .new_mem_def = 0
1220 }
1221 }
1222};
1223#endif /* __BIG_ENDIAN__ */
1224
1225static int __init prom_count_smt_threads(void)
1226{
1227 phandle node;
1228 char type[64];
1229 unsigned int plen;
1230
1231 /* Pick up th first CPU node we can find */
1232 for (node = 0; prom_next_node(&node); ) {
1233 type[0] = 0;
1234 prom_getprop(node, "device_type", type, sizeof(type));
1235
1236 if (prom_strcmp(type, "cpu"))
1237 continue;
1238 /*
1239 * There is an entry for each smt thread, each entry being
1240 * 4 bytes long. All cpus should have the same number of
1241 * smt threads, so return after finding the first.
1242 */
1243 plen = prom_getproplen(node, "ibm,ppc-interrupt-server#s");
1244 if (plen == PROM_ERROR)
1245 break;
1246 plen >>= 2;
1247 prom_debug("Found %lu smt threads per core\n", (unsigned long)plen);
1248
1249 /* Sanity check */
1250 if (plen < 1 || plen > 64) {
1251 prom_printf("Threads per core %lu out of bounds, assuming 1\n",
1252 (unsigned long)plen);
1253 return 1;
1254 }
1255 return plen;
1256 }
1257 prom_debug("No threads found, assuming 1 per core\n");
1258
1259 return 1;
1260
1261}
1262
1263static void __init prom_parse_mmu_model(u8 val,
1264 struct platform_support *support)
1265{
1266 switch (val) {
1267 case OV5_FEAT(OV5_MMU_DYNAMIC):
1268 case OV5_FEAT(OV5_MMU_EITHER): /* Either Available */
1269 prom_debug("MMU - either supported\n");
1270 support->radix_mmu = !prom_radix_disable;
1271 support->hash_mmu = true;
1272 break;
1273 case OV5_FEAT(OV5_MMU_RADIX): /* Only Radix */
1274 prom_debug("MMU - radix only\n");
1275 if (prom_radix_disable) {
1276 /*
1277 * If we __have__ to do radix, we're better off ignoring
1278 * the command line rather than not booting.
1279 */
1280 prom_printf("WARNING: Ignoring cmdline option disable_radix\n");
1281 }
1282 support->radix_mmu = true;
1283 break;
1284 case OV5_FEAT(OV5_MMU_HASH):
1285 prom_debug("MMU - hash only\n");
1286 support->hash_mmu = true;
1287 break;
1288 default:
1289 prom_debug("Unknown mmu support option: 0x%x\n", val);
1290 break;
1291 }
1292}
1293
1294static void __init prom_parse_xive_model(u8 val,
1295 struct platform_support *support)
1296{
1297 switch (val) {
1298 case OV5_FEAT(OV5_XIVE_EITHER): /* Either Available */
1299 prom_debug("XIVE - either mode supported\n");
1300 support->xive = !prom_xive_disable;
1301 break;
1302 case OV5_FEAT(OV5_XIVE_EXPLOIT): /* Only Exploitation mode */
1303 prom_debug("XIVE - exploitation mode supported\n");
1304 if (prom_xive_disable) {
1305 /*
1306 * If we __have__ to do XIVE, we're better off ignoring
1307 * the command line rather than not booting.
1308 */
1309 prom_printf("WARNING: Ignoring cmdline option xive=off\n");
1310 }
1311 support->xive = true;
1312 break;
1313 case OV5_FEAT(OV5_XIVE_LEGACY): /* Only Legacy mode */
1314 prom_debug("XIVE - legacy mode supported\n");
1315 break;
1316 default:
1317 prom_debug("Unknown xive support option: 0x%x\n", val);
1318 break;
1319 }
1320}
1321
1322static void __init prom_parse_platform_support(u8 index, u8 val,
1323 struct platform_support *support)
1324{
1325 switch (index) {
1326 case OV5_INDX(OV5_MMU_SUPPORT): /* MMU Model */
1327 prom_parse_mmu_model(val & OV5_FEAT(OV5_MMU_SUPPORT), support);
1328 break;
1329 case OV5_INDX(OV5_RADIX_GTSE): /* Radix Extensions */
1330 if (val & OV5_FEAT(OV5_RADIX_GTSE))
1331 support->radix_gtse = !prom_radix_gtse_disable;
1332 break;
1333 case OV5_INDX(OV5_XIVE_SUPPORT): /* Interrupt mode */
1334 prom_parse_xive_model(val & OV5_FEAT(OV5_XIVE_SUPPORT),
1335 support);
1336 break;
1337 }
1338}
1339
1340static void __init prom_check_platform_support(void)
1341{
1342 struct platform_support supported = {
1343 .hash_mmu = false,
1344 .radix_mmu = false,
1345 .radix_gtse = false,
1346 .xive = false
1347 };
1348 int prop_len = prom_getproplen(prom.chosen,
1349 "ibm,arch-vec-5-platform-support");
1350
1351 /*
1352 * First copy the architecture vec template
1353 *
1354 * use memcpy() instead of *vec = *vec_template so that GCC replaces it
1355 * by __memcpy() when KASAN is active
1356 */
1357 memcpy(&ibm_architecture_vec, &ibm_architecture_vec_template,
1358 sizeof(ibm_architecture_vec));
1359
1360 prom_strscpy_pad(ibm_architecture_vec.vec7.os_id, linux_banner, 256);
1361
1362 if (prop_len > 1) {
1363 int i;
1364 u8 vec[8];
1365 prom_debug("Found ibm,arch-vec-5-platform-support, len: %d\n",
1366 prop_len);
1367 if (prop_len > sizeof(vec))
1368 prom_printf("WARNING: ibm,arch-vec-5-platform-support longer than expected (len: %d)\n",
1369 prop_len);
1370 prom_getprop(prom.chosen, "ibm,arch-vec-5-platform-support", &vec, sizeof(vec));
1371 for (i = 0; i < prop_len; i += 2) {
1372 prom_debug("%d: index = 0x%x val = 0x%x\n", i / 2, vec[i], vec[i + 1]);
1373 prom_parse_platform_support(vec[i], vec[i + 1], &supported);
1374 }
1375 }
1376
1377 if (supported.radix_mmu && IS_ENABLED(CONFIG_PPC_RADIX_MMU)) {
1378 /* Radix preferred - Check if GTSE is also supported */
1379 prom_debug("Asking for radix\n");
1380 ibm_architecture_vec.vec5.mmu = OV5_FEAT(OV5_MMU_RADIX);
1381 if (supported.radix_gtse)
1382 ibm_architecture_vec.vec5.radix_ext =
1383 OV5_FEAT(OV5_RADIX_GTSE);
1384 else
1385 prom_debug("Radix GTSE isn't supported\n");
1386 } else if (supported.hash_mmu) {
1387 /* Default to hash mmu (if we can) */
1388 prom_debug("Asking for hash\n");
1389 ibm_architecture_vec.vec5.mmu = OV5_FEAT(OV5_MMU_HASH);
1390 } else {
1391 /* We're probably on a legacy hypervisor */
1392 prom_debug("Assuming legacy hash support\n");
1393 }
1394
1395 if (supported.xive) {
1396 prom_debug("Asking for XIVE\n");
1397 ibm_architecture_vec.vec5.intarch = OV5_FEAT(OV5_XIVE_EXPLOIT);
1398 }
1399}
1400
1401static void __init prom_send_capabilities(void)
1402{
1403 ihandle root;
1404 prom_arg_t ret;
1405 u32 cores;
1406
1407 /* Check ibm,arch-vec-5-platform-support and fixup vec5 if required */
1408 prom_check_platform_support();
1409
1410 root = call_prom("open", 1, 1, ADDR("/"));
1411 if (root != 0) {
1412 /* We need to tell the FW about the number of cores we support.
1413 *
1414 * To do that, we count the number of threads on the first core
1415 * (we assume this is the same for all cores) and use it to
1416 * divide NR_CPUS.
1417 */
1418
1419 cores = DIV_ROUND_UP(NR_CPUS, prom_count_smt_threads());
1420 prom_printf("Max number of cores passed to firmware: %u (NR_CPUS = %d)\n",
1421 cores, NR_CPUS);
1422
1423 ibm_architecture_vec.vec5.max_cpus = cpu_to_be32(cores);
1424
1425 /* try calling the ibm,client-architecture-support method */
1426 prom_printf("Calling ibm,client-architecture-support...");
1427 if (call_prom_ret("call-method", 3, 2, &ret,
1428 ADDR("ibm,client-architecture-support"),
1429 root,
1430 ADDR(&ibm_architecture_vec)) == 0) {
1431 /* the call exists... */
1432 if (ret)
1433 prom_printf("\nWARNING: ibm,client-architecture"
1434 "-support call FAILED!\n");
1435 call_prom("close", 1, 0, root);
1436 prom_printf(" done\n");
1437 return;
1438 }
1439 call_prom("close", 1, 0, root);
1440 prom_printf(" not implemented\n");
1441 }
1442
1443#ifdef __BIG_ENDIAN__
1444 {
1445 ihandle elfloader;
1446
1447 /* no ibm,client-architecture-support call, try the old way */
1448 elfloader = call_prom("open", 1, 1,
1449 ADDR("/packages/elf-loader"));
1450 if (elfloader == 0) {
1451 prom_printf("couldn't open /packages/elf-loader\n");
1452 return;
1453 }
1454 call_prom("call-method", 3, 1, ADDR("process-elf-header"),
1455 elfloader, ADDR(&fake_elf));
1456 call_prom("close", 1, 0, elfloader);
1457 }
1458#endif /* __BIG_ENDIAN__ */
1459}
1460#endif /* CONFIG_PPC_PSERIES */
1461
1462/*
1463 * Memory allocation strategy... our layout is normally:
1464 *
1465 * at 14Mb or more we have vmlinux, then a gap and initrd. In some
1466 * rare cases, initrd might end up being before the kernel though.
1467 * We assume this won't override the final kernel at 0, we have no
1468 * provision to handle that in this version, but it should hopefully
1469 * never happen.
1470 *
1471 * alloc_top is set to the top of RMO, eventually shrink down if the
1472 * TCEs overlap
1473 *
1474 * alloc_bottom is set to the top of kernel/initrd
1475 *
1476 * from there, allocations are done this way : rtas is allocated
1477 * topmost, and the device-tree is allocated from the bottom. We try
1478 * to grow the device-tree allocation as we progress. If we can't,
1479 * then we fail, we don't currently have a facility to restart
1480 * elsewhere, but that shouldn't be necessary.
1481 *
1482 * Note that calls to reserve_mem have to be done explicitly, memory
1483 * allocated with either alloc_up or alloc_down isn't automatically
1484 * reserved.
1485 */
1486
1487
1488/*
1489 * Allocates memory in the RMO upward from the kernel/initrd
1490 *
1491 * When align is 0, this is a special case, it means to allocate in place
1492 * at the current location of alloc_bottom or fail (that is basically
1493 * extending the previous allocation). Used for the device-tree flattening
1494 */
1495static unsigned long __init alloc_up(unsigned long size, unsigned long align)
1496{
1497 unsigned long base = alloc_bottom;
1498 unsigned long addr = 0;
1499
1500 if (align)
1501 base = ALIGN(base, align);
1502 prom_debug("%s(%lx, %lx)\n", __func__, size, align);
1503 if (ram_top == 0)
1504 prom_panic("alloc_up() called with mem not initialized\n");
1505
1506 if (align)
1507 base = ALIGN(alloc_bottom, align);
1508 else
1509 base = alloc_bottom;
1510
1511 for(; (base + size) <= alloc_top;
1512 base = ALIGN(base + 0x100000, align)) {
1513 prom_debug(" trying: 0x%lx\n\r", base);
1514 addr = (unsigned long)prom_claim(base, size, 0);
1515 if (addr != PROM_ERROR && addr != 0)
1516 break;
1517 addr = 0;
1518 if (align == 0)
1519 break;
1520 }
1521 if (addr == 0)
1522 return 0;
1523 alloc_bottom = addr + size;
1524
1525 prom_debug(" -> %lx\n", addr);
1526 prom_debug(" alloc_bottom : %lx\n", alloc_bottom);
1527 prom_debug(" alloc_top : %lx\n", alloc_top);
1528 prom_debug(" alloc_top_hi : %lx\n", alloc_top_high);
1529 prom_debug(" rmo_top : %lx\n", rmo_top);
1530 prom_debug(" ram_top : %lx\n", ram_top);
1531
1532 return addr;
1533}
1534
1535/*
1536 * Allocates memory downward, either from top of RMO, or if highmem
1537 * is set, from the top of RAM. Note that this one doesn't handle
1538 * failures. It does claim memory if highmem is not set.
1539 */
1540static unsigned long __init alloc_down(unsigned long size, unsigned long align,
1541 int highmem)
1542{
1543 unsigned long base, addr = 0;
1544
1545 prom_debug("%s(%lx, %lx, %s)\n", __func__, size, align,
1546 highmem ? "(high)" : "(low)");
1547 if (ram_top == 0)
1548 prom_panic("alloc_down() called with mem not initialized\n");
1549
1550 if (highmem) {
1551 /* Carve out storage for the TCE table. */
1552 addr = ALIGN_DOWN(alloc_top_high - size, align);
1553 if (addr <= alloc_bottom)
1554 return 0;
1555 /* Will we bump into the RMO ? If yes, check out that we
1556 * didn't overlap existing allocations there, if we did,
1557 * we are dead, we must be the first in town !
1558 */
1559 if (addr < rmo_top) {
1560 /* Good, we are first */
1561 if (alloc_top == rmo_top)
1562 alloc_top = rmo_top = addr;
1563 else
1564 return 0;
1565 }
1566 alloc_top_high = addr;
1567 goto bail;
1568 }
1569
1570 base = ALIGN_DOWN(alloc_top - size, align);
1571 for (; base > alloc_bottom;
1572 base = ALIGN_DOWN(base - 0x100000, align)) {
1573 prom_debug(" trying: 0x%lx\n\r", base);
1574 addr = (unsigned long)prom_claim(base, size, 0);
1575 if (addr != PROM_ERROR && addr != 0)
1576 break;
1577 addr = 0;
1578 }
1579 if (addr == 0)
1580 return 0;
1581 alloc_top = addr;
1582
1583 bail:
1584 prom_debug(" -> %lx\n", addr);
1585 prom_debug(" alloc_bottom : %lx\n", alloc_bottom);
1586 prom_debug(" alloc_top : %lx\n", alloc_top);
1587 prom_debug(" alloc_top_hi : %lx\n", alloc_top_high);
1588 prom_debug(" rmo_top : %lx\n", rmo_top);
1589 prom_debug(" ram_top : %lx\n", ram_top);
1590
1591 return addr;
1592}
1593
1594/*
1595 * Parse a "reg" cell
1596 */
1597static unsigned long __init prom_next_cell(int s, cell_t **cellp)
1598{
1599 cell_t *p = *cellp;
1600 unsigned long r = 0;
1601
1602 /* Ignore more than 2 cells */
1603 while (s > sizeof(unsigned long) / 4) {
1604 p++;
1605 s--;
1606 }
1607 r = be32_to_cpu(*p++);
1608#ifdef CONFIG_PPC64
1609 if (s > 1) {
1610 r <<= 32;
1611 r |= be32_to_cpu(*(p++));
1612 }
1613#endif
1614 *cellp = p;
1615 return r;
1616}
1617
1618/*
1619 * Very dumb function for adding to the memory reserve list, but
1620 * we don't need anything smarter at this point
1621 *
1622 * XXX Eventually check for collisions. They should NEVER happen.
1623 * If problems seem to show up, it would be a good start to track
1624 * them down.
1625 */
1626static void __init reserve_mem(u64 base, u64 size)
1627{
1628 u64 top = base + size;
1629 unsigned long cnt = mem_reserve_cnt;
1630
1631 if (size == 0)
1632 return;
1633
1634 /* We need to always keep one empty entry so that we
1635 * have our terminator with "size" set to 0 since we are
1636 * dumb and just copy this entire array to the boot params
1637 */
1638 base = ALIGN_DOWN(base, PAGE_SIZE);
1639 top = ALIGN(top, PAGE_SIZE);
1640 size = top - base;
1641
1642 if (cnt >= (MEM_RESERVE_MAP_SIZE - 1))
1643 prom_panic("Memory reserve map exhausted !\n");
1644 mem_reserve_map[cnt].base = cpu_to_be64(base);
1645 mem_reserve_map[cnt].size = cpu_to_be64(size);
1646 mem_reserve_cnt = cnt + 1;
1647}
1648
1649/*
1650 * Initialize memory allocation mechanism, parse "memory" nodes and
1651 * obtain that way the top of memory and RMO to setup out local allocator
1652 */
1653static void __init prom_init_mem(void)
1654{
1655 phandle node;
1656 char type[64];
1657 unsigned int plen;
1658 cell_t *p, *endp;
1659 __be32 val;
1660 u32 rac, rsc;
1661
1662 /*
1663 * We iterate the memory nodes to find
1664 * 1) top of RMO (first node)
1665 * 2) top of memory
1666 */
1667 val = cpu_to_be32(2);
1668 prom_getprop(prom.root, "#address-cells", &val, sizeof(val));
1669 rac = be32_to_cpu(val);
1670 val = cpu_to_be32(1);
1671 prom_getprop(prom.root, "#size-cells", &val, sizeof(rsc));
1672 rsc = be32_to_cpu(val);
1673 prom_debug("root_addr_cells: %x\n", rac);
1674 prom_debug("root_size_cells: %x\n", rsc);
1675
1676 prom_debug("scanning memory:\n");
1677
1678 for (node = 0; prom_next_node(&node); ) {
1679 type[0] = 0;
1680 prom_getprop(node, "device_type", type, sizeof(type));
1681
1682 if (type[0] == 0) {
1683 /*
1684 * CHRP Longtrail machines have no device_type
1685 * on the memory node, so check the name instead...
1686 */
1687 prom_getprop(node, "name", type, sizeof(type));
1688 }
1689 if (prom_strcmp(type, "memory"))
1690 continue;
1691
1692 plen = prom_getprop(node, "reg", regbuf, sizeof(regbuf));
1693 if (plen > sizeof(regbuf)) {
1694 prom_printf("memory node too large for buffer !\n");
1695 plen = sizeof(regbuf);
1696 }
1697 p = regbuf;
1698 endp = p + (plen / sizeof(cell_t));
1699
1700#ifdef DEBUG_PROM
1701 memset(prom_scratch, 0, sizeof(prom_scratch));
1702 call_prom("package-to-path", 3, 1, node, prom_scratch,
1703 sizeof(prom_scratch) - 1);
1704 prom_debug(" node %s :\n", prom_scratch);
1705#endif /* DEBUG_PROM */
1706
1707 while ((endp - p) >= (rac + rsc)) {
1708 unsigned long base, size;
1709
1710 base = prom_next_cell(rac, &p);
1711 size = prom_next_cell(rsc, &p);
1712
1713 if (size == 0)
1714 continue;
1715 prom_debug(" %lx %lx\n", base, size);
1716 if (base == 0 && (of_platform & PLATFORM_LPAR))
1717 rmo_top = size;
1718 if ((base + size) > ram_top)
1719 ram_top = base + size;
1720 }
1721 }
1722
1723 alloc_bottom = PAGE_ALIGN((unsigned long)&_end + 0x4000);
1724
1725 /*
1726 * If prom_memory_limit is set we reduce the upper limits *except* for
1727 * alloc_top_high. This must be the real top of RAM so we can put
1728 * TCE's up there.
1729 */
1730
1731 alloc_top_high = ram_top;
1732
1733 if (prom_memory_limit) {
1734 if (prom_memory_limit <= alloc_bottom) {
1735 prom_printf("Ignoring mem=%lx <= alloc_bottom.\n",
1736 prom_memory_limit);
1737 prom_memory_limit = 0;
1738 } else if (prom_memory_limit >= ram_top) {
1739 prom_printf("Ignoring mem=%lx >= ram_top.\n",
1740 prom_memory_limit);
1741 prom_memory_limit = 0;
1742 } else {
1743 ram_top = prom_memory_limit;
1744 rmo_top = min(rmo_top, prom_memory_limit);
1745 }
1746 }
1747
1748 /*
1749 * Setup our top alloc point, that is top of RMO or top of
1750 * segment 0 when running non-LPAR.
1751 * Some RS64 machines have buggy firmware where claims up at
1752 * 1GB fail. Cap at 768MB as a workaround.
1753 * Since 768MB is plenty of room, and we need to cap to something
1754 * reasonable on 32-bit, cap at 768MB on all machines.
1755 */
1756 if (!rmo_top)
1757 rmo_top = ram_top;
1758 rmo_top = min(0x30000000ul, rmo_top);
1759 alloc_top = rmo_top;
1760 alloc_top_high = ram_top;
1761
1762 /*
1763 * Check if we have an initrd after the kernel but still inside
1764 * the RMO. If we do move our bottom point to after it.
1765 */
1766 if (prom_initrd_start &&
1767 prom_initrd_start < rmo_top &&
1768 prom_initrd_end > alloc_bottom)
1769 alloc_bottom = PAGE_ALIGN(prom_initrd_end);
1770
1771 prom_printf("memory layout at init:\n");
1772 prom_printf(" memory_limit : %lx (16 MB aligned)\n",
1773 prom_memory_limit);
1774 prom_printf(" alloc_bottom : %lx\n", alloc_bottom);
1775 prom_printf(" alloc_top : %lx\n", alloc_top);
1776 prom_printf(" alloc_top_hi : %lx\n", alloc_top_high);
1777 prom_printf(" rmo_top : %lx\n", rmo_top);
1778 prom_printf(" ram_top : %lx\n", ram_top);
1779}
1780
1781static void __init prom_close_stdin(void)
1782{
1783 __be32 val;
1784 ihandle stdin;
1785
1786 if (prom_getprop(prom.chosen, "stdin", &val, sizeof(val)) > 0) {
1787 stdin = be32_to_cpu(val);
1788 call_prom("close", 1, 0, stdin);
1789 }
1790}
1791
1792#ifdef CONFIG_PPC_SVM
1793static int __init prom_rtas_hcall(uint64_t args)
1794{
1795 register uint64_t arg1 asm("r3") = H_RTAS;
1796 register uint64_t arg2 asm("r4") = args;
1797
1798 asm volatile("sc 1\n" : "=r" (arg1) :
1799 "r" (arg1),
1800 "r" (arg2) :);
1801 srr_regs_clobbered();
1802
1803 return arg1;
1804}
1805
1806static struct rtas_args __prombss os_term_args;
1807
1808static void __init prom_rtas_os_term(char *str)
1809{
1810 phandle rtas_node;
1811 __be32 val;
1812 u32 token;
1813
1814 prom_debug("%s: start...\n", __func__);
1815 rtas_node = call_prom("finddevice", 1, 1, ADDR("/rtas"));
1816 prom_debug("rtas_node: %x\n", rtas_node);
1817 if (!PHANDLE_VALID(rtas_node))
1818 return;
1819
1820 val = 0;
1821 prom_getprop(rtas_node, "ibm,os-term", &val, sizeof(val));
1822 token = be32_to_cpu(val);
1823 prom_debug("ibm,os-term: %x\n", token);
1824 if (token == 0)
1825 prom_panic("Could not get token for ibm,os-term\n");
1826 os_term_args.token = cpu_to_be32(token);
1827 os_term_args.nargs = cpu_to_be32(1);
1828 os_term_args.nret = cpu_to_be32(1);
1829 os_term_args.args[0] = cpu_to_be32(__pa(str));
1830 prom_rtas_hcall((uint64_t)&os_term_args);
1831}
1832#endif /* CONFIG_PPC_SVM */
1833
1834/*
1835 * Allocate room for and instantiate RTAS
1836 */
1837static void __init prom_instantiate_rtas(void)
1838{
1839 phandle rtas_node;
1840 ihandle rtas_inst;
1841 u32 base, entry = 0;
1842 __be32 val;
1843 u32 size = 0;
1844
1845 prom_debug("prom_instantiate_rtas: start...\n");
1846
1847 rtas_node = call_prom("finddevice", 1, 1, ADDR("/rtas"));
1848 prom_debug("rtas_node: %x\n", rtas_node);
1849 if (!PHANDLE_VALID(rtas_node))
1850 return;
1851
1852 val = 0;
1853 prom_getprop(rtas_node, "rtas-size", &val, sizeof(size));
1854 size = be32_to_cpu(val);
1855 if (size == 0)
1856 return;
1857
1858 base = alloc_down(size, PAGE_SIZE, 0);
1859 if (base == 0)
1860 prom_panic("Could not allocate memory for RTAS\n");
1861
1862 rtas_inst = call_prom("open", 1, 1, ADDR("/rtas"));
1863 if (!IHANDLE_VALID(rtas_inst)) {
1864 prom_printf("opening rtas package failed (%x)\n", rtas_inst);
1865 return;
1866 }
1867
1868 prom_printf("instantiating rtas at 0x%x...", base);
1869
1870 if (call_prom_ret("call-method", 3, 2, &entry,
1871 ADDR("instantiate-rtas"),
1872 rtas_inst, base) != 0
1873 || entry == 0) {
1874 prom_printf(" failed\n");
1875 return;
1876 }
1877 prom_printf(" done\n");
1878
1879 reserve_mem(base, size);
1880
1881 val = cpu_to_be32(base);
1882 prom_setprop(rtas_node, "/rtas", "linux,rtas-base",
1883 &val, sizeof(val));
1884 val = cpu_to_be32(entry);
1885 prom_setprop(rtas_node, "/rtas", "linux,rtas-entry",
1886 &val, sizeof(val));
1887
1888 /* Check if it supports "query-cpu-stopped-state" */
1889 if (prom_getprop(rtas_node, "query-cpu-stopped-state",
1890 &val, sizeof(val)) != PROM_ERROR)
1891 rtas_has_query_cpu_stopped = true;
1892
1893 prom_debug("rtas base = 0x%x\n", base);
1894 prom_debug("rtas entry = 0x%x\n", entry);
1895 prom_debug("rtas size = 0x%x\n", size);
1896
1897 prom_debug("prom_instantiate_rtas: end...\n");
1898}
1899
1900#ifdef CONFIG_PPC64
1901/*
1902 * Allocate room for and instantiate Stored Measurement Log (SML)
1903 */
1904static void __init prom_instantiate_sml(void)
1905{
1906 phandle ibmvtpm_node;
1907 ihandle ibmvtpm_inst;
1908 u32 entry = 0, size = 0, succ = 0;
1909 u64 base;
1910 __be32 val;
1911
1912 prom_debug("prom_instantiate_sml: start...\n");
1913
1914 ibmvtpm_node = call_prom("finddevice", 1, 1, ADDR("/vdevice/vtpm"));
1915 prom_debug("ibmvtpm_node: %x\n", ibmvtpm_node);
1916 if (!PHANDLE_VALID(ibmvtpm_node))
1917 return;
1918
1919 ibmvtpm_inst = call_prom("open", 1, 1, ADDR("/vdevice/vtpm"));
1920 if (!IHANDLE_VALID(ibmvtpm_inst)) {
1921 prom_printf("opening vtpm package failed (%x)\n", ibmvtpm_inst);
1922 return;
1923 }
1924
1925 if (prom_getprop(ibmvtpm_node, "ibm,sml-efi-reformat-supported",
1926 &val, sizeof(val)) != PROM_ERROR) {
1927 if (call_prom_ret("call-method", 2, 2, &succ,
1928 ADDR("reformat-sml-to-efi-alignment"),
1929 ibmvtpm_inst) != 0 || succ == 0) {
1930 prom_printf("Reformat SML to EFI alignment failed\n");
1931 return;
1932 }
1933
1934 if (call_prom_ret("call-method", 2, 2, &size,
1935 ADDR("sml-get-allocated-size"),
1936 ibmvtpm_inst) != 0 || size == 0) {
1937 prom_printf("SML get allocated size failed\n");
1938 return;
1939 }
1940 } else {
1941 if (call_prom_ret("call-method", 2, 2, &size,
1942 ADDR("sml-get-handover-size"),
1943 ibmvtpm_inst) != 0 || size == 0) {
1944 prom_printf("SML get handover size failed\n");
1945 return;
1946 }
1947 }
1948
1949 base = alloc_down(size, PAGE_SIZE, 0);
1950 if (base == 0)
1951 prom_panic("Could not allocate memory for sml\n");
1952
1953 prom_printf("instantiating sml at 0x%llx...", base);
1954
1955 memset((void *)base, 0, size);
1956
1957 if (call_prom_ret("call-method", 4, 2, &entry,
1958 ADDR("sml-handover"),
1959 ibmvtpm_inst, size, base) != 0 || entry == 0) {
1960 prom_printf("SML handover failed\n");
1961 return;
1962 }
1963 prom_printf(" done\n");
1964
1965 reserve_mem(base, size);
1966
1967 prom_setprop(ibmvtpm_node, "/vdevice/vtpm", "linux,sml-base",
1968 &base, sizeof(base));
1969 prom_setprop(ibmvtpm_node, "/vdevice/vtpm", "linux,sml-size",
1970 &size, sizeof(size));
1971
1972 prom_debug("sml base = 0x%llx\n", base);
1973 prom_debug("sml size = 0x%x\n", size);
1974
1975 prom_debug("prom_instantiate_sml: end...\n");
1976}
1977
1978/*
1979 * Allocate room for and initialize TCE tables
1980 */
1981#ifdef __BIG_ENDIAN__
1982static void __init prom_initialize_tce_table(void)
1983{
1984 phandle node;
1985 ihandle phb_node;
1986 char compatible[64], type[64], model[64];
1987 char *path = prom_scratch;
1988 u64 base, align;
1989 u32 minalign, minsize;
1990 u64 tce_entry, *tce_entryp;
1991 u64 local_alloc_top, local_alloc_bottom;
1992 u64 i;
1993
1994 if (prom_iommu_off)
1995 return;
1996
1997 prom_debug("starting prom_initialize_tce_table\n");
1998
1999 /* Cache current top of allocs so we reserve a single block */
2000 local_alloc_top = alloc_top_high;
2001 local_alloc_bottom = local_alloc_top;
2002
2003 /* Search all nodes looking for PHBs. */
2004 for (node = 0; prom_next_node(&node); ) {
2005 compatible[0] = 0;
2006 type[0] = 0;
2007 model[0] = 0;
2008 prom_getprop(node, "compatible",
2009 compatible, sizeof(compatible));
2010 prom_getprop(node, "device_type", type, sizeof(type));
2011 prom_getprop(node, "model", model, sizeof(model));
2012
2013 if ((type[0] == 0) || (prom_strstr(type, "pci") == NULL))
2014 continue;
2015
2016 /* Keep the old logic intact to avoid regression. */
2017 if (compatible[0] != 0) {
2018 if ((prom_strstr(compatible, "python") == NULL) &&
2019 (prom_strstr(compatible, "Speedwagon") == NULL) &&
2020 (prom_strstr(compatible, "Winnipeg") == NULL))
2021 continue;
2022 } else if (model[0] != 0) {
2023 if ((prom_strstr(model, "ython") == NULL) &&
2024 (prom_strstr(model, "peedwagon") == NULL) &&
2025 (prom_strstr(model, "innipeg") == NULL))
2026 continue;
2027 }
2028
2029 if (prom_getprop(node, "tce-table-minalign", &minalign,
2030 sizeof(minalign)) == PROM_ERROR)
2031 minalign = 0;
2032 if (prom_getprop(node, "tce-table-minsize", &minsize,
2033 sizeof(minsize)) == PROM_ERROR)
2034 minsize = 4UL << 20;
2035
2036 /*
2037 * Even though we read what OF wants, we just set the table
2038 * size to 4 MB. This is enough to map 2GB of PCI DMA space.
2039 * By doing this, we avoid the pitfalls of trying to DMA to
2040 * MMIO space and the DMA alias hole.
2041 */
2042 minsize = 4UL << 20;
2043
2044 /* Align to the greater of the align or size */
2045 align = max(minalign, minsize);
2046 base = alloc_down(minsize, align, 1);
2047 if (base == 0)
2048 prom_panic("ERROR, cannot find space for TCE table.\n");
2049 if (base < local_alloc_bottom)
2050 local_alloc_bottom = base;
2051
2052 /* It seems OF doesn't null-terminate the path :-( */
2053 memset(path, 0, sizeof(prom_scratch));
2054 /* Call OF to setup the TCE hardware */
2055 if (call_prom("package-to-path", 3, 1, node,
2056 path, sizeof(prom_scratch) - 1) == PROM_ERROR) {
2057 prom_printf("package-to-path failed\n");
2058 }
2059
2060 /* Save away the TCE table attributes for later use. */
2061 prom_setprop(node, path, "linux,tce-base", &base, sizeof(base));
2062 prom_setprop(node, path, "linux,tce-size", &minsize, sizeof(minsize));
2063
2064 prom_debug("TCE table: %s\n", path);
2065 prom_debug("\tnode = 0x%x\n", node);
2066 prom_debug("\tbase = 0x%llx\n", base);
2067 prom_debug("\tsize = 0x%x\n", minsize);
2068
2069 /* Initialize the table to have a one-to-one mapping
2070 * over the allocated size.
2071 */
2072 tce_entryp = (u64 *)base;
2073 for (i = 0; i < (minsize >> 3) ;tce_entryp++, i++) {
2074 tce_entry = (i << PAGE_SHIFT);
2075 tce_entry |= 0x3;
2076 *tce_entryp = tce_entry;
2077 }
2078
2079 prom_printf("opening PHB %s", path);
2080 phb_node = call_prom("open", 1, 1, path);
2081 if (phb_node == 0)
2082 prom_printf("... failed\n");
2083 else
2084 prom_printf("... done\n");
2085
2086 call_prom("call-method", 6, 0, ADDR("set-64-bit-addressing"),
2087 phb_node, -1, minsize,
2088 (u32) base, (u32) (base >> 32));
2089 call_prom("close", 1, 0, phb_node);
2090 }
2091
2092 reserve_mem(local_alloc_bottom, local_alloc_top - local_alloc_bottom);
2093
2094 /* These are only really needed if there is a memory limit in
2095 * effect, but we don't know so export them always. */
2096 prom_tce_alloc_start = local_alloc_bottom;
2097 prom_tce_alloc_end = local_alloc_top;
2098
2099 /* Flag the first invalid entry */
2100 prom_debug("ending prom_initialize_tce_table\n");
2101}
2102#endif /* __BIG_ENDIAN__ */
2103#endif /* CONFIG_PPC64 */
2104
2105/*
2106 * With CHRP SMP we need to use the OF to start the other processors.
2107 * We can't wait until smp_boot_cpus (the OF is trashed by then)
2108 * so we have to put the processors into a holding pattern controlled
2109 * by the kernel (not OF) before we destroy the OF.
2110 *
2111 * This uses a chunk of low memory, puts some holding pattern
2112 * code there and sends the other processors off to there until
2113 * smp_boot_cpus tells them to do something. The holding pattern
2114 * checks that address until its cpu # is there, when it is that
2115 * cpu jumps to __secondary_start(). smp_boot_cpus() takes care
2116 * of setting those values.
2117 *
2118 * We also use physical address 0x4 here to tell when a cpu
2119 * is in its holding pattern code.
2120 *
2121 * -- Cort
2122 */
2123/*
2124 * We want to reference the copy of __secondary_hold_* in the
2125 * 0 - 0x100 address range
2126 */
2127#define LOW_ADDR(x) (((unsigned long) &(x)) & 0xff)
2128
2129static void __init prom_hold_cpus(void)
2130{
2131 unsigned long i;
2132 phandle node;
2133 char type[64];
2134 unsigned long *spinloop
2135 = (void *) LOW_ADDR(__secondary_hold_spinloop);
2136 unsigned long *acknowledge
2137 = (void *) LOW_ADDR(__secondary_hold_acknowledge);
2138 unsigned long secondary_hold = LOW_ADDR(__secondary_hold);
2139
2140 /*
2141 * On pseries, if RTAS supports "query-cpu-stopped-state",
2142 * we skip this stage, the CPUs will be started by the
2143 * kernel using RTAS.
2144 */
2145 if ((of_platform == PLATFORM_PSERIES ||
2146 of_platform == PLATFORM_PSERIES_LPAR) &&
2147 rtas_has_query_cpu_stopped) {
2148 prom_printf("prom_hold_cpus: skipped\n");
2149 return;
2150 }
2151
2152 prom_debug("prom_hold_cpus: start...\n");
2153 prom_debug(" 1) spinloop = 0x%lx\n", (unsigned long)spinloop);
2154 prom_debug(" 1) *spinloop = 0x%lx\n", *spinloop);
2155 prom_debug(" 1) acknowledge = 0x%lx\n",
2156 (unsigned long)acknowledge);
2157 prom_debug(" 1) *acknowledge = 0x%lx\n", *acknowledge);
2158 prom_debug(" 1) secondary_hold = 0x%lx\n", secondary_hold);
2159
2160 /* Set the common spinloop variable, so all of the secondary cpus
2161 * will block when they are awakened from their OF spinloop.
2162 * This must occur for both SMP and non SMP kernels, since OF will
2163 * be trashed when we move the kernel.
2164 */
2165 *spinloop = 0;
2166
2167 /* look for cpus */
2168 for (node = 0; prom_next_node(&node); ) {
2169 unsigned int cpu_no;
2170 __be32 reg;
2171
2172 type[0] = 0;
2173 prom_getprop(node, "device_type", type, sizeof(type));
2174 if (prom_strcmp(type, "cpu") != 0)
2175 continue;
2176
2177 /* Skip non-configured cpus. */
2178 if (prom_getprop(node, "status", type, sizeof(type)) > 0)
2179 if (prom_strcmp(type, "okay") != 0)
2180 continue;
2181
2182 reg = cpu_to_be32(-1); /* make sparse happy */
2183 prom_getprop(node, "reg", ®, sizeof(reg));
2184 cpu_no = be32_to_cpu(reg);
2185
2186 prom_debug("cpu hw idx = %u\n", cpu_no);
2187
2188 /* Init the acknowledge var which will be reset by
2189 * the secondary cpu when it awakens from its OF
2190 * spinloop.
2191 */
2192 *acknowledge = (unsigned long)-1;
2193
2194 if (cpu_no != prom.cpu) {
2195 /* Primary Thread of non-boot cpu or any thread */
2196 prom_printf("starting cpu hw idx %u... ", cpu_no);
2197 call_prom("start-cpu", 3, 0, node,
2198 secondary_hold, cpu_no);
2199
2200 for (i = 0; (i < 100000000) &&
2201 (*acknowledge == ((unsigned long)-1)); i++ )
2202 mb();
2203
2204 if (*acknowledge == cpu_no)
2205 prom_printf("done\n");
2206 else
2207 prom_printf("failed: %lx\n", *acknowledge);
2208 }
2209#ifdef CONFIG_SMP
2210 else
2211 prom_printf("boot cpu hw idx %u\n", cpu_no);
2212#endif /* CONFIG_SMP */
2213 }
2214
2215 prom_debug("prom_hold_cpus: end...\n");
2216}
2217
2218
2219static void __init prom_init_client_services(unsigned long pp)
2220{
2221 /* Get a handle to the prom entry point before anything else */
2222 prom_entry = pp;
2223
2224 /* get a handle for the stdout device */
2225 prom.chosen = call_prom("finddevice", 1, 1, ADDR("/chosen"));
2226 if (!PHANDLE_VALID(prom.chosen))
2227 prom_panic("cannot find chosen"); /* msg won't be printed :( */
2228
2229 /* get device tree root */
2230 prom.root = call_prom("finddevice", 1, 1, ADDR("/"));
2231 if (!PHANDLE_VALID(prom.root))
2232 prom_panic("cannot find device tree root"); /* msg won't be printed :( */
2233
2234 prom.mmumap = 0;
2235}
2236
2237#ifdef CONFIG_PPC32
2238/*
2239 * For really old powermacs, we need to map things we claim.
2240 * For that, we need the ihandle of the mmu.
2241 * Also, on the longtrail, we need to work around other bugs.
2242 */
2243static void __init prom_find_mmu(void)
2244{
2245 phandle oprom;
2246 char version[64];
2247
2248 oprom = call_prom("finddevice", 1, 1, ADDR("/openprom"));
2249 if (!PHANDLE_VALID(oprom))
2250 return;
2251 if (prom_getprop(oprom, "model", version, sizeof(version)) <= 0)
2252 return;
2253 version[sizeof(version) - 1] = 0;
2254 /* XXX might need to add other versions here */
2255 if (prom_strcmp(version, "Open Firmware, 1.0.5") == 0)
2256 of_workarounds = OF_WA_CLAIM;
2257 else if (prom_strncmp(version, "FirmWorks,3.", 12) == 0) {
2258 of_workarounds = OF_WA_CLAIM | OF_WA_LONGTRAIL;
2259 call_prom("interpret", 1, 1, "dev /memory 0 to allow-reclaim");
2260 } else
2261 return;
2262 prom.memory = call_prom("open", 1, 1, ADDR("/memory"));
2263 prom_getprop(prom.chosen, "mmu", &prom.mmumap,
2264 sizeof(prom.mmumap));
2265 prom.mmumap = be32_to_cpu(prom.mmumap);
2266 if (!IHANDLE_VALID(prom.memory) || !IHANDLE_VALID(prom.mmumap))
2267 of_workarounds &= ~OF_WA_CLAIM; /* hmmm */
2268}
2269#else
2270#define prom_find_mmu()
2271#endif
2272
2273static void __init prom_init_stdout(void)
2274{
2275 char *path = of_stdout_device;
2276 char type[16];
2277 phandle stdout_node;
2278 __be32 val;
2279
2280 if (prom_getprop(prom.chosen, "stdout", &val, sizeof(val)) <= 0)
2281 prom_panic("cannot find stdout");
2282
2283 prom.stdout = be32_to_cpu(val);
2284
2285 /* Get the full OF pathname of the stdout device */
2286 memset(path, 0, 256);
2287 call_prom("instance-to-path", 3, 1, prom.stdout, path, 255);
2288 prom_printf("OF stdout device is: %s\n", of_stdout_device);
2289 prom_setprop(prom.chosen, "/chosen", "linux,stdout-path",
2290 path, prom_strlen(path) + 1);
2291
2292 /* instance-to-package fails on PA-Semi */
2293 stdout_node = call_prom("instance-to-package", 1, 1, prom.stdout);
2294 if (stdout_node != PROM_ERROR) {
2295 val = cpu_to_be32(stdout_node);
2296
2297 /* If it's a display, note it */
2298 memset(type, 0, sizeof(type));
2299 prom_getprop(stdout_node, "device_type", type, sizeof(type));
2300 if (prom_strcmp(type, "display") == 0)
2301 prom_setprop(stdout_node, path, "linux,boot-display", NULL, 0);
2302 }
2303}
2304
2305static int __init prom_find_machine_type(void)
2306{
2307 static char compat[256] __prombss;
2308 int len, i = 0;
2309#ifdef CONFIG_PPC64
2310 phandle rtas;
2311 int x;
2312#endif
2313
2314 /* Look for a PowerMac or a Cell */
2315 len = prom_getprop(prom.root, "compatible",
2316 compat, sizeof(compat)-1);
2317 if (len > 0) {
2318 compat[len] = 0;
2319 while (i < len) {
2320 char *p = &compat[i];
2321 int sl = prom_strlen(p);
2322 if (sl == 0)
2323 break;
2324 if (prom_strstr(p, "Power Macintosh") ||
2325 prom_strstr(p, "MacRISC"))
2326 return PLATFORM_POWERMAC;
2327#ifdef CONFIG_PPC64
2328 /* We must make sure we don't detect the IBM Cell
2329 * blades as pSeries due to some firmware issues,
2330 * so we do it here.
2331 */
2332 if (prom_strstr(p, "IBM,CBEA") ||
2333 prom_strstr(p, "IBM,CPBW-1.0"))
2334 return PLATFORM_GENERIC;
2335#endif /* CONFIG_PPC64 */
2336 i += sl + 1;
2337 }
2338 }
2339#ifdef CONFIG_PPC64
2340 /* Try to figure out if it's an IBM pSeries or any other
2341 * PAPR compliant platform. We assume it is if :
2342 * - /device_type is "chrp" (please, do NOT use that for future
2343 * non-IBM designs !
2344 * - it has /rtas
2345 */
2346 len = prom_getprop(prom.root, "device_type",
2347 compat, sizeof(compat)-1);
2348 if (len <= 0)
2349 return PLATFORM_GENERIC;
2350 if (prom_strcmp(compat, "chrp"))
2351 return PLATFORM_GENERIC;
2352
2353 /* Default to pSeries. We need to know if we are running LPAR */
2354 rtas = call_prom("finddevice", 1, 1, ADDR("/rtas"));
2355 if (!PHANDLE_VALID(rtas))
2356 return PLATFORM_GENERIC;
2357 x = prom_getproplen(rtas, "ibm,hypertas-functions");
2358 if (x != PROM_ERROR) {
2359 prom_debug("Hypertas detected, assuming LPAR !\n");
2360 return PLATFORM_PSERIES_LPAR;
2361 }
2362 return PLATFORM_PSERIES;
2363#else
2364 return PLATFORM_GENERIC;
2365#endif
2366}
2367
2368static int __init prom_set_color(ihandle ih, int i, int r, int g, int b)
2369{
2370 return call_prom("call-method", 6, 1, ADDR("color!"), ih, i, b, g, r);
2371}
2372
2373/*
2374 * If we have a display that we don't know how to drive,
2375 * we will want to try to execute OF's open method for it
2376 * later. However, OF will probably fall over if we do that
2377 * we've taken over the MMU.
2378 * So we check whether we will need to open the display,
2379 * and if so, open it now.
2380 */
2381static void __init prom_check_displays(void)
2382{
2383 char type[16], *path;
2384 phandle node;
2385 ihandle ih;
2386 int i;
2387
2388 static const unsigned char default_colors[] __initconst = {
2389 0x00, 0x00, 0x00,
2390 0x00, 0x00, 0xaa,
2391 0x00, 0xaa, 0x00,
2392 0x00, 0xaa, 0xaa,
2393 0xaa, 0x00, 0x00,
2394 0xaa, 0x00, 0xaa,
2395 0xaa, 0xaa, 0x00,
2396 0xaa, 0xaa, 0xaa,
2397 0x55, 0x55, 0x55,
2398 0x55, 0x55, 0xff,
2399 0x55, 0xff, 0x55,
2400 0x55, 0xff, 0xff,
2401 0xff, 0x55, 0x55,
2402 0xff, 0x55, 0xff,
2403 0xff, 0xff, 0x55,
2404 0xff, 0xff, 0xff
2405 };
2406 const unsigned char *clut;
2407
2408 prom_debug("Looking for displays\n");
2409 for (node = 0; prom_next_node(&node); ) {
2410 memset(type, 0, sizeof(type));
2411 prom_getprop(node, "device_type", type, sizeof(type));
2412 if (prom_strcmp(type, "display") != 0)
2413 continue;
2414
2415 /* It seems OF doesn't null-terminate the path :-( */
2416 path = prom_scratch;
2417 memset(path, 0, sizeof(prom_scratch));
2418
2419 /*
2420 * leave some room at the end of the path for appending extra
2421 * arguments
2422 */
2423 if (call_prom("package-to-path", 3, 1, node, path,
2424 sizeof(prom_scratch) - 10) == PROM_ERROR)
2425 continue;
2426 prom_printf("found display : %s, opening... ", path);
2427
2428 ih = call_prom("open", 1, 1, path);
2429 if (ih == 0) {
2430 prom_printf("failed\n");
2431 continue;
2432 }
2433
2434 /* Success */
2435 prom_printf("done\n");
2436 prom_setprop(node, path, "linux,opened", NULL, 0);
2437
2438 /* Setup a usable color table when the appropriate
2439 * method is available. Should update this to set-colors */
2440 clut = default_colors;
2441 for (i = 0; i < 16; i++, clut += 3)
2442 if (prom_set_color(ih, i, clut[0], clut[1],
2443 clut[2]) != 0)
2444 break;
2445
2446#ifdef CONFIG_LOGO_LINUX_CLUT224
2447 clut = PTRRELOC(logo_linux_clut224.clut);
2448 for (i = 0; i < logo_linux_clut224.clutsize; i++, clut += 3)
2449 if (prom_set_color(ih, i + 32, clut[0], clut[1],
2450 clut[2]) != 0)
2451 break;
2452#endif /* CONFIG_LOGO_LINUX_CLUT224 */
2453
2454#ifdef CONFIG_PPC_EARLY_DEBUG_BOOTX
2455 if (prom_getprop(node, "linux,boot-display", NULL, 0) !=
2456 PROM_ERROR) {
2457 u32 width, height, pitch, addr;
2458
2459 prom_printf("Setting btext !\n");
2460
2461 if (prom_getprop(node, "width", &width, 4) == PROM_ERROR)
2462 return;
2463
2464 if (prom_getprop(node, "height", &height, 4) == PROM_ERROR)
2465 return;
2466
2467 if (prom_getprop(node, "linebytes", &pitch, 4) == PROM_ERROR)
2468 return;
2469
2470 if (prom_getprop(node, "address", &addr, 4) == PROM_ERROR)
2471 return;
2472
2473 prom_printf("W=%d H=%d LB=%d addr=0x%x\n",
2474 width, height, pitch, addr);
2475 btext_setup_display(width, height, 8, pitch, addr);
2476 btext_prepare_BAT();
2477 }
2478#endif /* CONFIG_PPC_EARLY_DEBUG_BOOTX */
2479 }
2480}
2481
2482
2483/* Return (relocated) pointer to this much memory: moves initrd if reqd. */
2484static void __init *make_room(unsigned long *mem_start, unsigned long *mem_end,
2485 unsigned long needed, unsigned long align)
2486{
2487 void *ret;
2488
2489 *mem_start = ALIGN(*mem_start, align);
2490 while ((*mem_start + needed) > *mem_end) {
2491 unsigned long room, chunk;
2492
2493 prom_debug("Chunk exhausted, claiming more at %lx...\n",
2494 alloc_bottom);
2495 room = alloc_top - alloc_bottom;
2496 if (room > DEVTREE_CHUNK_SIZE)
2497 room = DEVTREE_CHUNK_SIZE;
2498 if (room < PAGE_SIZE)
2499 prom_panic("No memory for flatten_device_tree "
2500 "(no room)\n");
2501 chunk = alloc_up(room, 0);
2502 if (chunk == 0)
2503 prom_panic("No memory for flatten_device_tree "
2504 "(claim failed)\n");
2505 *mem_end = chunk + room;
2506 }
2507
2508 ret = (void *)*mem_start;
2509 *mem_start += needed;
2510
2511 return ret;
2512}
2513
2514#define dt_push_token(token, mem_start, mem_end) do { \
2515 void *room = make_room(mem_start, mem_end, 4, 4); \
2516 *(__be32 *)room = cpu_to_be32(token); \
2517 } while(0)
2518
2519static unsigned long __init dt_find_string(char *str)
2520{
2521 char *s, *os;
2522
2523 s = os = (char *)dt_string_start;
2524 s += 4;
2525 while (s < (char *)dt_string_end) {
2526 if (prom_strcmp(s, str) == 0)
2527 return s - os;
2528 s += prom_strlen(s) + 1;
2529 }
2530 return 0;
2531}
2532
2533/*
2534 * The Open Firmware 1275 specification states properties must be 31 bytes or
2535 * less, however not all firmwares obey this. Make it 64 bytes to be safe.
2536 */
2537#define MAX_PROPERTY_NAME 64
2538
2539static void __init scan_dt_build_strings(phandle node,
2540 unsigned long *mem_start,
2541 unsigned long *mem_end)
2542{
2543 char *prev_name, *namep, *sstart;
2544 unsigned long soff;
2545 phandle child;
2546
2547 sstart = (char *)dt_string_start;
2548
2549 /* get and store all property names */
2550 prev_name = "";
2551 for (;;) {
2552 /* 64 is max len of name including nul. */
2553 namep = make_room(mem_start, mem_end, MAX_PROPERTY_NAME, 1);
2554 if (call_prom("nextprop", 3, 1, node, prev_name, namep) != 1) {
2555 /* No more nodes: unwind alloc */
2556 *mem_start = (unsigned long)namep;
2557 break;
2558 }
2559
2560 /* skip "name" */
2561 if (prom_strcmp(namep, "name") == 0) {
2562 *mem_start = (unsigned long)namep;
2563 prev_name = "name";
2564 continue;
2565 }
2566 /* get/create string entry */
2567 soff = dt_find_string(namep);
2568 if (soff != 0) {
2569 *mem_start = (unsigned long)namep;
2570 namep = sstart + soff;
2571 } else {
2572 /* Trim off some if we can */
2573 *mem_start = (unsigned long)namep + prom_strlen(namep) + 1;
2574 dt_string_end = *mem_start;
2575 }
2576 prev_name = namep;
2577 }
2578
2579 /* do all our children */
2580 child = call_prom("child", 1, 1, node);
2581 while (child != 0) {
2582 scan_dt_build_strings(child, mem_start, mem_end);
2583 child = call_prom("peer", 1, 1, child);
2584 }
2585}
2586
2587static void __init scan_dt_build_struct(phandle node, unsigned long *mem_start,
2588 unsigned long *mem_end)
2589{
2590 phandle child;
2591 char *namep, *prev_name, *sstart, *p, *ep, *lp, *path;
2592 unsigned long soff;
2593 unsigned char *valp;
2594 static char pname[MAX_PROPERTY_NAME] __prombss;
2595 int l, room, has_phandle = 0;
2596
2597 dt_push_token(OF_DT_BEGIN_NODE, mem_start, mem_end);
2598
2599 /* get the node's full name */
2600 namep = (char *)*mem_start;
2601 room = *mem_end - *mem_start;
2602 if (room > 255)
2603 room = 255;
2604 l = call_prom("package-to-path", 3, 1, node, namep, room);
2605 if (l >= 0) {
2606 /* Didn't fit? Get more room. */
2607 if (l >= room) {
2608 if (l >= *mem_end - *mem_start)
2609 namep = make_room(mem_start, mem_end, l+1, 1);
2610 call_prom("package-to-path", 3, 1, node, namep, l);
2611 }
2612 namep[l] = '\0';
2613
2614 /* Fixup an Apple bug where they have bogus \0 chars in the
2615 * middle of the path in some properties, and extract
2616 * the unit name (everything after the last '/').
2617 */
2618 for (lp = p = namep, ep = namep + l; p < ep; p++) {
2619 if (*p == '/')
2620 lp = namep;
2621 else if (*p != 0)
2622 *lp++ = *p;
2623 }
2624 *lp = 0;
2625 *mem_start = ALIGN((unsigned long)lp + 1, 4);
2626 }
2627
2628 /* get it again for debugging */
2629 path = prom_scratch;
2630 memset(path, 0, sizeof(prom_scratch));
2631 call_prom("package-to-path", 3, 1, node, path, sizeof(prom_scratch) - 1);
2632
2633 /* get and store all properties */
2634 prev_name = "";
2635 sstart = (char *)dt_string_start;
2636 for (;;) {
2637 if (call_prom("nextprop", 3, 1, node, prev_name,
2638 pname) != 1)
2639 break;
2640
2641 /* skip "name" */
2642 if (prom_strcmp(pname, "name") == 0) {
2643 prev_name = "name";
2644 continue;
2645 }
2646
2647 /* find string offset */
2648 soff = dt_find_string(pname);
2649 if (soff == 0) {
2650 prom_printf("WARNING: Can't find string index for"
2651 " <%s>, node %s\n", pname, path);
2652 break;
2653 }
2654 prev_name = sstart + soff;
2655
2656 /* get length */
2657 l = call_prom("getproplen", 2, 1, node, pname);
2658
2659 /* sanity checks */
2660 if (l == PROM_ERROR)
2661 continue;
2662
2663 /* push property head */
2664 dt_push_token(OF_DT_PROP, mem_start, mem_end);
2665 dt_push_token(l, mem_start, mem_end);
2666 dt_push_token(soff, mem_start, mem_end);
2667
2668 /* push property content */
2669 valp = make_room(mem_start, mem_end, l, 4);
2670 call_prom("getprop", 4, 1, node, pname, valp, l);
2671 *mem_start = ALIGN(*mem_start, 4);
2672
2673 if (!prom_strcmp(pname, "phandle"))
2674 has_phandle = 1;
2675 }
2676
2677 /* Add a "phandle" property if none already exist */
2678 if (!has_phandle) {
2679 soff = dt_find_string("phandle");
2680 if (soff == 0)
2681 prom_printf("WARNING: Can't find string index for <phandle> node %s\n", path);
2682 else {
2683 dt_push_token(OF_DT_PROP, mem_start, mem_end);
2684 dt_push_token(4, mem_start, mem_end);
2685 dt_push_token(soff, mem_start, mem_end);
2686 valp = make_room(mem_start, mem_end, 4, 4);
2687 *(__be32 *)valp = cpu_to_be32(node);
2688 }
2689 }
2690
2691 /* do all our children */
2692 child = call_prom("child", 1, 1, node);
2693 while (child != 0) {
2694 scan_dt_build_struct(child, mem_start, mem_end);
2695 child = call_prom("peer", 1, 1, child);
2696 }
2697
2698 dt_push_token(OF_DT_END_NODE, mem_start, mem_end);
2699}
2700
2701static void __init flatten_device_tree(void)
2702{
2703 phandle root;
2704 unsigned long mem_start, mem_end, room;
2705 struct boot_param_header *hdr;
2706 char *namep;
2707 u64 *rsvmap;
2708
2709 /*
2710 * Check how much room we have between alloc top & bottom (+/- a
2711 * few pages), crop to 1MB, as this is our "chunk" size
2712 */
2713 room = alloc_top - alloc_bottom - 0x4000;
2714 if (room > DEVTREE_CHUNK_SIZE)
2715 room = DEVTREE_CHUNK_SIZE;
2716 prom_debug("starting device tree allocs at %lx\n", alloc_bottom);
2717
2718 /* Now try to claim that */
2719 mem_start = (unsigned long)alloc_up(room, PAGE_SIZE);
2720 if (mem_start == 0)
2721 prom_panic("Can't allocate initial device-tree chunk\n");
2722 mem_end = mem_start + room;
2723
2724 /* Get root of tree */
2725 root = call_prom("peer", 1, 1, (phandle)0);
2726 if (root == (phandle)0)
2727 prom_panic ("couldn't get device tree root\n");
2728
2729 /* Build header and make room for mem rsv map */
2730 mem_start = ALIGN(mem_start, 4);
2731 hdr = make_room(&mem_start, &mem_end,
2732 sizeof(struct boot_param_header), 4);
2733 dt_header_start = (unsigned long)hdr;
2734 rsvmap = make_room(&mem_start, &mem_end, sizeof(mem_reserve_map), 8);
2735
2736 /* Start of strings */
2737 mem_start = PAGE_ALIGN(mem_start);
2738 dt_string_start = mem_start;
2739 mem_start += 4; /* hole */
2740
2741 /* Add "phandle" in there, we'll need it */
2742 namep = make_room(&mem_start, &mem_end, 16, 1);
2743 prom_strscpy_pad(namep, "phandle", sizeof("phandle"));
2744 mem_start = (unsigned long)namep + prom_strlen(namep) + 1;
2745
2746 /* Build string array */
2747 prom_printf("Building dt strings...\n");
2748 scan_dt_build_strings(root, &mem_start, &mem_end);
2749 dt_string_end = mem_start;
2750
2751 /* Build structure */
2752 mem_start = PAGE_ALIGN(mem_start);
2753 dt_struct_start = mem_start;
2754 prom_printf("Building dt structure...\n");
2755 scan_dt_build_struct(root, &mem_start, &mem_end);
2756 dt_push_token(OF_DT_END, &mem_start, &mem_end);
2757 dt_struct_end = PAGE_ALIGN(mem_start);
2758
2759 /* Finish header */
2760 hdr->boot_cpuid_phys = cpu_to_be32(prom.cpu);
2761 hdr->magic = cpu_to_be32(OF_DT_HEADER);
2762 hdr->totalsize = cpu_to_be32(dt_struct_end - dt_header_start);
2763 hdr->off_dt_struct = cpu_to_be32(dt_struct_start - dt_header_start);
2764 hdr->off_dt_strings = cpu_to_be32(dt_string_start - dt_header_start);
2765 hdr->dt_strings_size = cpu_to_be32(dt_string_end - dt_string_start);
2766 hdr->off_mem_rsvmap = cpu_to_be32(((unsigned long)rsvmap) - dt_header_start);
2767 hdr->version = cpu_to_be32(OF_DT_VERSION);
2768 /* Version 16 is not backward compatible */
2769 hdr->last_comp_version = cpu_to_be32(0x10);
2770
2771 /* Copy the reserve map in */
2772 memcpy(rsvmap, mem_reserve_map, sizeof(mem_reserve_map));
2773
2774#ifdef DEBUG_PROM
2775 {
2776 int i;
2777 prom_printf("reserved memory map:\n");
2778 for (i = 0; i < mem_reserve_cnt; i++)
2779 prom_printf(" %llx - %llx\n",
2780 be64_to_cpu(mem_reserve_map[i].base),
2781 be64_to_cpu(mem_reserve_map[i].size));
2782 }
2783#endif
2784 /* Bump mem_reserve_cnt to cause further reservations to fail
2785 * since it's too late.
2786 */
2787 mem_reserve_cnt = MEM_RESERVE_MAP_SIZE;
2788
2789 prom_printf("Device tree strings 0x%lx -> 0x%lx\n",
2790 dt_string_start, dt_string_end);
2791 prom_printf("Device tree struct 0x%lx -> 0x%lx\n",
2792 dt_struct_start, dt_struct_end);
2793}
2794
2795#ifdef CONFIG_PPC_CHRP
2796/*
2797 * Pegasos and BriQ lacks the "ranges" property in the isa node
2798 * Pegasos needs decimal IRQ 14/15, not hexadecimal
2799 * Pegasos has the IDE configured in legacy mode, but advertised as native
2800 */
2801static void __init fixup_device_tree_chrp(void)
2802{
2803 phandle ph;
2804 u32 prop[6];
2805 u32 rloc = 0x01006000; /* IO space; PCI device = 12 */
2806 char *name;
2807 int rc;
2808
2809 name = "/pci@80000000/isa@c";
2810 ph = call_prom("finddevice", 1, 1, ADDR(name));
2811 if (!PHANDLE_VALID(ph)) {
2812 name = "/pci@ff500000/isa@6";
2813 ph = call_prom("finddevice", 1, 1, ADDR(name));
2814 rloc = 0x01003000; /* IO space; PCI device = 6 */
2815 }
2816 if (PHANDLE_VALID(ph)) {
2817 rc = prom_getproplen(ph, "ranges");
2818 if (rc == 0 || rc == PROM_ERROR) {
2819 prom_printf("Fixing up missing ISA range on Pegasos...\n");
2820
2821 prop[0] = 0x1;
2822 prop[1] = 0x0;
2823 prop[2] = rloc;
2824 prop[3] = 0x0;
2825 prop[4] = 0x0;
2826 prop[5] = 0x00010000;
2827 prom_setprop(ph, name, "ranges", prop, sizeof(prop));
2828 }
2829 }
2830
2831 name = "/pci@80000000/ide@C,1";
2832 ph = call_prom("finddevice", 1, 1, ADDR(name));
2833 if (PHANDLE_VALID(ph)) {
2834 prom_printf("Fixing up IDE interrupt on Pegasos...\n");
2835 prop[0] = 14;
2836 prop[1] = 0x0;
2837 prom_setprop(ph, name, "interrupts", prop, 2*sizeof(u32));
2838 prom_printf("Fixing up IDE class-code on Pegasos...\n");
2839 rc = prom_getprop(ph, "class-code", prop, sizeof(u32));
2840 if (rc == sizeof(u32)) {
2841 prop[0] &= ~0x5;
2842 prom_setprop(ph, name, "class-code", prop, sizeof(u32));
2843 }
2844 }
2845}
2846#else
2847#define fixup_device_tree_chrp()
2848#endif
2849
2850#if defined(CONFIG_PPC64) && defined(CONFIG_PPC_PMAC)
2851static void __init fixup_device_tree_pmac64(void)
2852{
2853 phandle u3, i2c, mpic;
2854 u32 u3_rev;
2855 u32 interrupts[2];
2856 u32 parent;
2857
2858 /* Some G5s have a missing interrupt definition, fix it up here */
2859 u3 = call_prom("finddevice", 1, 1, ADDR("/u3@0,f8000000"));
2860 if (!PHANDLE_VALID(u3))
2861 return;
2862 i2c = call_prom("finddevice", 1, 1, ADDR("/u3@0,f8000000/i2c@f8001000"));
2863 if (!PHANDLE_VALID(i2c))
2864 return;
2865 mpic = call_prom("finddevice", 1, 1, ADDR("/u3@0,f8000000/mpic@f8040000"));
2866 if (!PHANDLE_VALID(mpic))
2867 return;
2868
2869 /* check if proper rev of u3 */
2870 if (prom_getprop(u3, "device-rev", &u3_rev, sizeof(u3_rev))
2871 == PROM_ERROR)
2872 return;
2873 if (u3_rev < 0x35 || u3_rev > 0x39)
2874 return;
2875 /* does it need fixup ? */
2876 if (prom_getproplen(i2c, "interrupts") > 0)
2877 return;
2878
2879 prom_printf("fixing up bogus interrupts for u3 i2c...\n");
2880
2881 /* interrupt on this revision of u3 is number 0 and level */
2882 interrupts[0] = 0;
2883 interrupts[1] = 1;
2884 prom_setprop(i2c, "/u3@0,f8000000/i2c@f8001000", "interrupts",
2885 &interrupts, sizeof(interrupts));
2886 parent = (u32)mpic;
2887 prom_setprop(i2c, "/u3@0,f8000000/i2c@f8001000", "interrupt-parent",
2888 &parent, sizeof(parent));
2889}
2890#else
2891#define fixup_device_tree_pmac64()
2892#endif
2893
2894#ifdef CONFIG_PPC_PMAC
2895static void __init fixup_device_tree_pmac(void)
2896{
2897 __be32 val = 1;
2898 char type[8];
2899 phandle node;
2900
2901 // Some pmacs are missing #size-cells on escc nodes
2902 for (node = 0; prom_next_node(&node); ) {
2903 type[0] = '\0';
2904 prom_getprop(node, "device_type", type, sizeof(type));
2905 if (prom_strcmp(type, "escc"))
2906 continue;
2907
2908 if (prom_getproplen(node, "#size-cells") != PROM_ERROR)
2909 continue;
2910
2911 prom_setprop(node, NULL, "#size-cells", &val, sizeof(val));
2912 }
2913}
2914#else
2915static inline void fixup_device_tree_pmac(void) { }
2916#endif
2917
2918#ifdef CONFIG_PPC_EFIKA
2919/*
2920 * The MPC5200 FEC driver requires an phy-handle property to tell it how
2921 * to talk to the phy. If the phy-handle property is missing, then this
2922 * function is called to add the appropriate nodes and link it to the
2923 * ethernet node.
2924 */
2925static void __init fixup_device_tree_efika_add_phy(void)
2926{
2927 u32 node;
2928 char prop[64];
2929 int rv;
2930
2931 /* Check if /builtin/ethernet exists - bail if it doesn't */
2932 node = call_prom("finddevice", 1, 1, ADDR("/builtin/ethernet"));
2933 if (!PHANDLE_VALID(node))
2934 return;
2935
2936 /* Check if the phy-handle property exists - bail if it does */
2937 rv = prom_getprop(node, "phy-handle", prop, sizeof(prop));
2938 if (rv <= 0)
2939 return;
2940
2941 /*
2942 * At this point the ethernet device doesn't have a phy described.
2943 * Now we need to add the missing phy node and linkage
2944 */
2945
2946 /* Check for an MDIO bus node - if missing then create one */
2947 node = call_prom("finddevice", 1, 1, ADDR("/builtin/mdio"));
2948 if (!PHANDLE_VALID(node)) {
2949 prom_printf("Adding Ethernet MDIO node\n");
2950 call_prom("interpret", 1, 1,
2951 " s\" /builtin\" find-device"
2952 " new-device"
2953 " 1 encode-int s\" #address-cells\" property"
2954 " 0 encode-int s\" #size-cells\" property"
2955 " s\" mdio\" device-name"
2956 " s\" fsl,mpc5200b-mdio\" encode-string"
2957 " s\" compatible\" property"
2958 " 0xf0003000 0x400 reg"
2959 " 0x2 encode-int"
2960 " 0x5 encode-int encode+"
2961 " 0x3 encode-int encode+"
2962 " s\" interrupts\" property"
2963 " finish-device");
2964 }
2965
2966 /* Check for a PHY device node - if missing then create one and
2967 * give it's phandle to the ethernet node */
2968 node = call_prom("finddevice", 1, 1,
2969 ADDR("/builtin/mdio/ethernet-phy"));
2970 if (!PHANDLE_VALID(node)) {
2971 prom_printf("Adding Ethernet PHY node\n");
2972 call_prom("interpret", 1, 1,
2973 " s\" /builtin/mdio\" find-device"
2974 " new-device"
2975 " s\" ethernet-phy\" device-name"
2976 " 0x10 encode-int s\" reg\" property"
2977 " my-self"
2978 " ihandle>phandle"
2979 " finish-device"
2980 " s\" /builtin/ethernet\" find-device"
2981 " encode-int"
2982 " s\" phy-handle\" property"
2983 " device-end");
2984 }
2985}
2986
2987static void __init fixup_device_tree_efika(void)
2988{
2989 int sound_irq[3] = { 2, 2, 0 };
2990 int bcomm_irq[3*16] = { 3,0,0, 3,1,0, 3,2,0, 3,3,0,
2991 3,4,0, 3,5,0, 3,6,0, 3,7,0,
2992 3,8,0, 3,9,0, 3,10,0, 3,11,0,
2993 3,12,0, 3,13,0, 3,14,0, 3,15,0 };
2994 u32 node;
2995 char prop[64];
2996 int rv, len;
2997
2998 /* Check if we're really running on a EFIKA */
2999 node = call_prom("finddevice", 1, 1, ADDR("/"));
3000 if (!PHANDLE_VALID(node))
3001 return;
3002
3003 rv = prom_getprop(node, "model", prop, sizeof(prop));
3004 if (rv == PROM_ERROR)
3005 return;
3006 if (prom_strcmp(prop, "EFIKA5K2"))
3007 return;
3008
3009 prom_printf("Applying EFIKA device tree fixups\n");
3010
3011 /* Claiming to be 'chrp' is death */
3012 node = call_prom("finddevice", 1, 1, ADDR("/"));
3013 rv = prom_getprop(node, "device_type", prop, sizeof(prop));
3014 if (rv != PROM_ERROR && (prom_strcmp(prop, "chrp") == 0))
3015 prom_setprop(node, "/", "device_type", "efika", sizeof("efika"));
3016
3017 /* CODEGEN,description is exposed in /proc/cpuinfo so
3018 fix that too */
3019 rv = prom_getprop(node, "CODEGEN,description", prop, sizeof(prop));
3020 if (rv != PROM_ERROR && (prom_strstr(prop, "CHRP")))
3021 prom_setprop(node, "/", "CODEGEN,description",
3022 "Efika 5200B PowerPC System",
3023 sizeof("Efika 5200B PowerPC System"));
3024
3025 /* Fixup bestcomm interrupts property */
3026 node = call_prom("finddevice", 1, 1, ADDR("/builtin/bestcomm"));
3027 if (PHANDLE_VALID(node)) {
3028 len = prom_getproplen(node, "interrupts");
3029 if (len == 12) {
3030 prom_printf("Fixing bestcomm interrupts property\n");
3031 prom_setprop(node, "/builtin/bestcom", "interrupts",
3032 bcomm_irq, sizeof(bcomm_irq));
3033 }
3034 }
3035
3036 /* Fixup sound interrupts property */
3037 node = call_prom("finddevice", 1, 1, ADDR("/builtin/sound"));
3038 if (PHANDLE_VALID(node)) {
3039 rv = prom_getprop(node, "interrupts", prop, sizeof(prop));
3040 if (rv == PROM_ERROR) {
3041 prom_printf("Adding sound interrupts property\n");
3042 prom_setprop(node, "/builtin/sound", "interrupts",
3043 sound_irq, sizeof(sound_irq));
3044 }
3045 }
3046
3047 /* Make sure ethernet phy-handle property exists */
3048 fixup_device_tree_efika_add_phy();
3049}
3050#else
3051#define fixup_device_tree_efika()
3052#endif
3053
3054#ifdef CONFIG_PPC_PASEMI_NEMO
3055/*
3056 * CFE supplied on Nemo is broken in several ways, biggest
3057 * problem is that it reassigns ISA interrupts to unused mpic ints.
3058 * Add an interrupt-controller property for the io-bridge to use
3059 * and correct the ints so we can attach them to an irq_domain
3060 */
3061static void __init fixup_device_tree_pasemi(void)
3062{
3063 u32 interrupts[2], parent, rval, val = 0;
3064 char *name, *pci_name;
3065 phandle iob, node;
3066
3067 /* Find the root pci node */
3068 name = "/pxp@0,e0000000";
3069 iob = call_prom("finddevice", 1, 1, ADDR(name));
3070 if (!PHANDLE_VALID(iob))
3071 return;
3072
3073 /* check if interrupt-controller node set yet */
3074 if (prom_getproplen(iob, "interrupt-controller") !=PROM_ERROR)
3075 return;
3076
3077 prom_printf("adding interrupt-controller property for SB600...\n");
3078
3079 prom_setprop(iob, name, "interrupt-controller", &val, 0);
3080
3081 pci_name = "/pxp@0,e0000000/pci@11";
3082 node = call_prom("finddevice", 1, 1, ADDR(pci_name));
3083 parent = ADDR(iob);
3084
3085 for( ; prom_next_node(&node); ) {
3086 /* scan each node for one with an interrupt */
3087 if (!PHANDLE_VALID(node))
3088 continue;
3089
3090 rval = prom_getproplen(node, "interrupts");
3091 if (rval == 0 || rval == PROM_ERROR)
3092 continue;
3093
3094 prom_getprop(node, "interrupts", &interrupts, sizeof(interrupts));
3095 if ((interrupts[0] < 212) || (interrupts[0] > 222))
3096 continue;
3097
3098 /* found a node, update both interrupts and interrupt-parent */
3099 if ((interrupts[0] >= 212) && (interrupts[0] <= 215))
3100 interrupts[0] -= 203;
3101 if ((interrupts[0] >= 216) && (interrupts[0] <= 220))
3102 interrupts[0] -= 213;
3103 if (interrupts[0] == 221)
3104 interrupts[0] = 14;
3105 if (interrupts[0] == 222)
3106 interrupts[0] = 8;
3107
3108 prom_setprop(node, pci_name, "interrupts", interrupts,
3109 sizeof(interrupts));
3110 prom_setprop(node, pci_name, "interrupt-parent", &parent,
3111 sizeof(parent));
3112 }
3113
3114 /*
3115 * The io-bridge has device_type set to 'io-bridge' change it to 'isa'
3116 * so that generic isa-bridge code can add the SB600 and its on-board
3117 * peripherals.
3118 */
3119 name = "/pxp@0,e0000000/io-bridge@0";
3120 iob = call_prom("finddevice", 1, 1, ADDR(name));
3121 if (!PHANDLE_VALID(iob))
3122 return;
3123
3124 /* device_type is already set, just change it. */
3125
3126 prom_printf("Changing device_type of SB600 node...\n");
3127
3128 prom_setprop(iob, name, "device_type", "isa", sizeof("isa"));
3129}
3130#else /* !CONFIG_PPC_PASEMI_NEMO */
3131static inline void fixup_device_tree_pasemi(void) { }
3132#endif
3133
3134static void __init fixup_device_tree(void)
3135{
3136 fixup_device_tree_chrp();
3137 fixup_device_tree_pmac();
3138 fixup_device_tree_pmac64();
3139 fixup_device_tree_efika();
3140 fixup_device_tree_pasemi();
3141}
3142
3143static void __init prom_find_boot_cpu(void)
3144{
3145 __be32 rval;
3146 ihandle prom_cpu;
3147 phandle cpu_pkg;
3148
3149 rval = 0;
3150 if (prom_getprop(prom.chosen, "cpu", &rval, sizeof(rval)) <= 0)
3151 return;
3152 prom_cpu = be32_to_cpu(rval);
3153
3154 cpu_pkg = call_prom("instance-to-package", 1, 1, prom_cpu);
3155
3156 if (!PHANDLE_VALID(cpu_pkg))
3157 return;
3158
3159 prom_getprop(cpu_pkg, "reg", &rval, sizeof(rval));
3160 prom.cpu = be32_to_cpu(rval);
3161
3162 prom_debug("Booting CPU hw index = %d\n", prom.cpu);
3163}
3164
3165static void __init prom_check_initrd(unsigned long r3, unsigned long r4)
3166{
3167#ifdef CONFIG_BLK_DEV_INITRD
3168 if (r3 && r4 && r4 != 0xdeadbeef) {
3169 __be64 val;
3170
3171 prom_initrd_start = is_kernel_addr(r3) ? __pa(r3) : r3;
3172 prom_initrd_end = prom_initrd_start + r4;
3173
3174 val = cpu_to_be64(prom_initrd_start);
3175 prom_setprop(prom.chosen, "/chosen", "linux,initrd-start",
3176 &val, sizeof(val));
3177 val = cpu_to_be64(prom_initrd_end);
3178 prom_setprop(prom.chosen, "/chosen", "linux,initrd-end",
3179 &val, sizeof(val));
3180
3181 reserve_mem(prom_initrd_start,
3182 prom_initrd_end - prom_initrd_start);
3183
3184 prom_debug("initrd_start=0x%lx\n", prom_initrd_start);
3185 prom_debug("initrd_end=0x%lx\n", prom_initrd_end);
3186 }
3187#endif /* CONFIG_BLK_DEV_INITRD */
3188}
3189
3190#ifdef CONFIG_PPC_SVM
3191/*
3192 * Perform the Enter Secure Mode ultracall.
3193 */
3194static int __init enter_secure_mode(unsigned long kbase, unsigned long fdt)
3195{
3196 register unsigned long r3 asm("r3") = UV_ESM;
3197 register unsigned long r4 asm("r4") = kbase;
3198 register unsigned long r5 asm("r5") = fdt;
3199
3200 asm volatile("sc 2" : "+r"(r3) : "r"(r4), "r"(r5));
3201
3202 return r3;
3203}
3204
3205/*
3206 * Call the Ultravisor to transfer us to secure memory if we have an ESM blob.
3207 */
3208static void __init setup_secure_guest(unsigned long kbase, unsigned long fdt)
3209{
3210 int ret;
3211
3212 if (!prom_svm_enable)
3213 return;
3214
3215 /* Switch to secure mode. */
3216 prom_printf("Switching to secure mode.\n");
3217
3218 /*
3219 * The ultravisor will do an integrity check of the kernel image but we
3220 * relocated it so the check will fail. Restore the original image by
3221 * relocating it back to the kernel virtual base address.
3222 */
3223 relocate(KERNELBASE);
3224
3225 ret = enter_secure_mode(kbase, fdt);
3226
3227 /* Relocate the kernel again. */
3228 relocate(kbase);
3229
3230 if (ret != U_SUCCESS) {
3231 prom_printf("Returned %d from switching to secure mode.\n", ret);
3232 prom_rtas_os_term("Switch to secure mode failed.\n");
3233 }
3234}
3235#else
3236static void __init setup_secure_guest(unsigned long kbase, unsigned long fdt)
3237{
3238}
3239#endif /* CONFIG_PPC_SVM */
3240
3241/*
3242 * We enter here early on, when the Open Firmware prom is still
3243 * handling exceptions and the MMU hash table for us.
3244 */
3245
3246unsigned long __init prom_init(unsigned long r3, unsigned long r4,
3247 unsigned long pp,
3248 unsigned long r6, unsigned long r7,
3249 unsigned long kbase)
3250{
3251 unsigned long hdr;
3252
3253#ifdef CONFIG_PPC32
3254 unsigned long offset = reloc_offset();
3255 reloc_got2(offset);
3256#endif
3257
3258 /*
3259 * First zero the BSS
3260 */
3261 memset(&__bss_start, 0, __bss_stop - __bss_start);
3262
3263 /*
3264 * Init interface to Open Firmware, get some node references,
3265 * like /chosen
3266 */
3267 prom_init_client_services(pp);
3268
3269 /*
3270 * See if this OF is old enough that we need to do explicit maps
3271 * and other workarounds
3272 */
3273 prom_find_mmu();
3274
3275 /*
3276 * Init prom stdout device
3277 */
3278 prom_init_stdout();
3279
3280 prom_printf("Preparing to boot %s", linux_banner);
3281
3282 /*
3283 * Get default machine type. At this point, we do not differentiate
3284 * between pSeries SMP and pSeries LPAR
3285 */
3286 of_platform = prom_find_machine_type();
3287 prom_printf("Detected machine type: %x\n", of_platform);
3288
3289#ifndef CONFIG_NONSTATIC_KERNEL
3290 /* Bail if this is a kdump kernel. */
3291 if (PHYSICAL_START > 0)
3292 prom_panic("Error: You can't boot a kdump kernel from OF!\n");
3293#endif
3294
3295 /*
3296 * Check for an initrd
3297 */
3298 prom_check_initrd(r3, r4);
3299
3300 /*
3301 * Do early parsing of command line
3302 */
3303 early_cmdline_parse();
3304
3305#ifdef CONFIG_PPC_PSERIES
3306 /*
3307 * On pSeries, inform the firmware about our capabilities
3308 */
3309 if (of_platform == PLATFORM_PSERIES ||
3310 of_platform == PLATFORM_PSERIES_LPAR)
3311 prom_send_capabilities();
3312#endif
3313
3314 /*
3315 * Copy the CPU hold code
3316 */
3317 if (of_platform != PLATFORM_POWERMAC)
3318 copy_and_flush(0, kbase, 0x100, 0);
3319
3320 /*
3321 * Initialize memory management within prom_init
3322 */
3323 prom_init_mem();
3324
3325 /*
3326 * Determine which cpu is actually running right _now_
3327 */
3328 prom_find_boot_cpu();
3329
3330 /*
3331 * Initialize display devices
3332 */
3333 prom_check_displays();
3334
3335#if defined(CONFIG_PPC64) && defined(__BIG_ENDIAN__)
3336 /*
3337 * Initialize IOMMU (TCE tables) on pSeries. Do that before anything else
3338 * that uses the allocator, we need to make sure we get the top of memory
3339 * available for us here...
3340 */
3341 if (of_platform == PLATFORM_PSERIES)
3342 prom_initialize_tce_table();
3343#endif
3344
3345 /*
3346 * On non-powermacs, try to instantiate RTAS. PowerMacs don't
3347 * have a usable RTAS implementation.
3348 */
3349 if (of_platform != PLATFORM_POWERMAC)
3350 prom_instantiate_rtas();
3351
3352#ifdef CONFIG_PPC64
3353 /* instantiate sml */
3354 prom_instantiate_sml();
3355#endif
3356
3357 /*
3358 * On non-powermacs, put all CPUs in spin-loops.
3359 *
3360 * PowerMacs use a different mechanism to spin CPUs
3361 *
3362 * (This must be done after instantiating RTAS)
3363 */
3364 if (of_platform != PLATFORM_POWERMAC)
3365 prom_hold_cpus();
3366
3367 /*
3368 * Fill in some infos for use by the kernel later on
3369 */
3370 if (prom_memory_limit) {
3371 __be64 val = cpu_to_be64(prom_memory_limit);
3372 prom_setprop(prom.chosen, "/chosen", "linux,memory-limit",
3373 &val, sizeof(val));
3374 }
3375#ifdef CONFIG_PPC64
3376 if (prom_iommu_off)
3377 prom_setprop(prom.chosen, "/chosen", "linux,iommu-off",
3378 NULL, 0);
3379
3380 if (prom_iommu_force_on)
3381 prom_setprop(prom.chosen, "/chosen", "linux,iommu-force-on",
3382 NULL, 0);
3383
3384 if (prom_tce_alloc_start) {
3385 prom_setprop(prom.chosen, "/chosen", "linux,tce-alloc-start",
3386 &prom_tce_alloc_start,
3387 sizeof(prom_tce_alloc_start));
3388 prom_setprop(prom.chosen, "/chosen", "linux,tce-alloc-end",
3389 &prom_tce_alloc_end,
3390 sizeof(prom_tce_alloc_end));
3391 }
3392#endif
3393
3394 /*
3395 * Fixup any known bugs in the device-tree
3396 */
3397 fixup_device_tree();
3398
3399 /*
3400 * Now finally create the flattened device-tree
3401 */
3402 prom_printf("copying OF device tree...\n");
3403 flatten_device_tree();
3404
3405 /*
3406 * in case stdin is USB and still active on IBM machines...
3407 * Unfortunately quiesce crashes on some powermacs if we have
3408 * closed stdin already (in particular the powerbook 101).
3409 */
3410 if (of_platform != PLATFORM_POWERMAC)
3411 prom_close_stdin();
3412
3413 /*
3414 * Call OF "quiesce" method to shut down pending DMA's from
3415 * devices etc...
3416 */
3417 prom_printf("Quiescing Open Firmware ...\n");
3418 call_prom("quiesce", 0, 0);
3419
3420 /*
3421 * And finally, call the kernel passing it the flattened device
3422 * tree and NULL as r5, thus triggering the new entry point which
3423 * is common to us and kexec
3424 */
3425 hdr = dt_header_start;
3426
3427 prom_printf("Booting Linux via __start() @ 0x%lx ...\n", kbase);
3428 prom_debug("->dt_header_start=0x%lx\n", hdr);
3429
3430#ifdef CONFIG_PPC32
3431 reloc_got2(-offset);
3432#endif
3433
3434 /* Move to secure memory if we're supposed to be secure guests. */
3435 setup_secure_guest(kbase, hdr);
3436
3437 __start(hdr, kbase, 0, 0, 0, 0, 0);
3438
3439 return 0;
3440}
1/*
2 * Procedures for interfacing to Open Firmware.
3 *
4 * Paul Mackerras August 1996.
5 * Copyright (C) 1996-2005 Paul Mackerras.
6 *
7 * Adapted for 64bit PowerPC by Dave Engebretsen and Peter Bergner.
8 * {engebret|bergner}@us.ibm.com
9 *
10 * This program is free software; you can redistribute it and/or
11 * modify it under the terms of the GNU General Public License
12 * as published by the Free Software Foundation; either version
13 * 2 of the License, or (at your option) any later version.
14 */
15
16#undef DEBUG_PROM
17
18#include <stdarg.h>
19#include <linux/kernel.h>
20#include <linux/string.h>
21#include <linux/init.h>
22#include <linux/threads.h>
23#include <linux/spinlock.h>
24#include <linux/types.h>
25#include <linux/pci.h>
26#include <linux/proc_fs.h>
27#include <linux/stringify.h>
28#include <linux/delay.h>
29#include <linux/initrd.h>
30#include <linux/bitops.h>
31#include <asm/prom.h>
32#include <asm/rtas.h>
33#include <asm/page.h>
34#include <asm/processor.h>
35#include <asm/irq.h>
36#include <asm/io.h>
37#include <asm/smp.h>
38#include <asm/mmu.h>
39#include <asm/pgtable.h>
40#include <asm/pci.h>
41#include <asm/iommu.h>
42#include <asm/btext.h>
43#include <asm/sections.h>
44#include <asm/machdep.h>
45#include <asm/opal.h>
46
47#include <linux/linux_logo.h>
48
49/*
50 * Eventually bump that one up
51 */
52#define DEVTREE_CHUNK_SIZE 0x100000
53
54/*
55 * This is the size of the local memory reserve map that gets copied
56 * into the boot params passed to the kernel. That size is totally
57 * flexible as the kernel just reads the list until it encounters an
58 * entry with size 0, so it can be changed without breaking binary
59 * compatibility
60 */
61#define MEM_RESERVE_MAP_SIZE 8
62
63/*
64 * prom_init() is called very early on, before the kernel text
65 * and data have been mapped to KERNELBASE. At this point the code
66 * is running at whatever address it has been loaded at.
67 * On ppc32 we compile with -mrelocatable, which means that references
68 * to extern and static variables get relocated automatically.
69 * ppc64 objects are always relocatable, we just need to relocate the
70 * TOC.
71 *
72 * Because OF may have mapped I/O devices into the area starting at
73 * KERNELBASE, particularly on CHRP machines, we can't safely call
74 * OF once the kernel has been mapped to KERNELBASE. Therefore all
75 * OF calls must be done within prom_init().
76 *
77 * ADDR is used in calls to call_prom. The 4th and following
78 * arguments to call_prom should be 32-bit values.
79 * On ppc64, 64 bit values are truncated to 32 bits (and
80 * fortunately don't get interpreted as two arguments).
81 */
82#define ADDR(x) (u32)(unsigned long)(x)
83
84#ifdef CONFIG_PPC64
85#define OF_WORKAROUNDS 0
86#else
87#define OF_WORKAROUNDS of_workarounds
88int of_workarounds;
89#endif
90
91#define OF_WA_CLAIM 1 /* do phys/virt claim separately, then map */
92#define OF_WA_LONGTRAIL 2 /* work around longtrail bugs */
93
94#define PROM_BUG() do { \
95 prom_printf("kernel BUG at %s line 0x%x!\n", \
96 __FILE__, __LINE__); \
97 __asm__ __volatile__(".long " BUG_ILLEGAL_INSTR); \
98} while (0)
99
100#ifdef DEBUG_PROM
101#define prom_debug(x...) prom_printf(x)
102#else
103#define prom_debug(x...)
104#endif
105
106
107typedef u32 prom_arg_t;
108
109struct prom_args {
110 __be32 service;
111 __be32 nargs;
112 __be32 nret;
113 __be32 args[10];
114};
115
116struct prom_t {
117 ihandle root;
118 phandle chosen;
119 int cpu;
120 ihandle stdout;
121 ihandle mmumap;
122 ihandle memory;
123};
124
125struct mem_map_entry {
126 __be64 base;
127 __be64 size;
128};
129
130typedef __be32 cell_t;
131
132extern void __start(unsigned long r3, unsigned long r4, unsigned long r5,
133 unsigned long r6, unsigned long r7, unsigned long r8,
134 unsigned long r9);
135
136#ifdef CONFIG_PPC64
137extern int enter_prom(struct prom_args *args, unsigned long entry);
138#else
139static inline int enter_prom(struct prom_args *args, unsigned long entry)
140{
141 return ((int (*)(struct prom_args *))entry)(args);
142}
143#endif
144
145extern void copy_and_flush(unsigned long dest, unsigned long src,
146 unsigned long size, unsigned long offset);
147
148/* prom structure */
149static struct prom_t __initdata prom;
150
151static unsigned long prom_entry __initdata;
152
153#define PROM_SCRATCH_SIZE 256
154
155static char __initdata of_stdout_device[256];
156static char __initdata prom_scratch[PROM_SCRATCH_SIZE];
157
158static unsigned long __initdata dt_header_start;
159static unsigned long __initdata dt_struct_start, dt_struct_end;
160static unsigned long __initdata dt_string_start, dt_string_end;
161
162static unsigned long __initdata prom_initrd_start, prom_initrd_end;
163
164#ifdef CONFIG_PPC64
165static int __initdata prom_iommu_force_on;
166static int __initdata prom_iommu_off;
167static unsigned long __initdata prom_tce_alloc_start;
168static unsigned long __initdata prom_tce_alloc_end;
169#endif
170
171/* Platforms codes are now obsolete in the kernel. Now only used within this
172 * file and ultimately gone too. Feel free to change them if you need, they
173 * are not shared with anything outside of this file anymore
174 */
175#define PLATFORM_PSERIES 0x0100
176#define PLATFORM_PSERIES_LPAR 0x0101
177#define PLATFORM_LPAR 0x0001
178#define PLATFORM_POWERMAC 0x0400
179#define PLATFORM_GENERIC 0x0500
180#define PLATFORM_OPAL 0x0600
181
182static int __initdata of_platform;
183
184static char __initdata prom_cmd_line[COMMAND_LINE_SIZE];
185
186static unsigned long __initdata prom_memory_limit;
187
188static unsigned long __initdata alloc_top;
189static unsigned long __initdata alloc_top_high;
190static unsigned long __initdata alloc_bottom;
191static unsigned long __initdata rmo_top;
192static unsigned long __initdata ram_top;
193
194static struct mem_map_entry __initdata mem_reserve_map[MEM_RESERVE_MAP_SIZE];
195static int __initdata mem_reserve_cnt;
196
197static cell_t __initdata regbuf[1024];
198
199static bool rtas_has_query_cpu_stopped;
200
201
202/*
203 * Error results ... some OF calls will return "-1" on error, some
204 * will return 0, some will return either. To simplify, here are
205 * macros to use with any ihandle or phandle return value to check if
206 * it is valid
207 */
208
209#define PROM_ERROR (-1u)
210#define PHANDLE_VALID(p) ((p) != 0 && (p) != PROM_ERROR)
211#define IHANDLE_VALID(i) ((i) != 0 && (i) != PROM_ERROR)
212
213
214/* This is the one and *ONLY* place where we actually call open
215 * firmware.
216 */
217
218static int __init call_prom(const char *service, int nargs, int nret, ...)
219{
220 int i;
221 struct prom_args args;
222 va_list list;
223
224 args.service = cpu_to_be32(ADDR(service));
225 args.nargs = cpu_to_be32(nargs);
226 args.nret = cpu_to_be32(nret);
227
228 va_start(list, nret);
229 for (i = 0; i < nargs; i++)
230 args.args[i] = cpu_to_be32(va_arg(list, prom_arg_t));
231 va_end(list);
232
233 for (i = 0; i < nret; i++)
234 args.args[nargs+i] = 0;
235
236 if (enter_prom(&args, prom_entry) < 0)
237 return PROM_ERROR;
238
239 return (nret > 0) ? be32_to_cpu(args.args[nargs]) : 0;
240}
241
242static int __init call_prom_ret(const char *service, int nargs, int nret,
243 prom_arg_t *rets, ...)
244{
245 int i;
246 struct prom_args args;
247 va_list list;
248
249 args.service = cpu_to_be32(ADDR(service));
250 args.nargs = cpu_to_be32(nargs);
251 args.nret = cpu_to_be32(nret);
252
253 va_start(list, rets);
254 for (i = 0; i < nargs; i++)
255 args.args[i] = cpu_to_be32(va_arg(list, prom_arg_t));
256 va_end(list);
257
258 for (i = 0; i < nret; i++)
259 args.args[nargs+i] = 0;
260
261 if (enter_prom(&args, prom_entry) < 0)
262 return PROM_ERROR;
263
264 if (rets != NULL)
265 for (i = 1; i < nret; ++i)
266 rets[i-1] = be32_to_cpu(args.args[nargs+i]);
267
268 return (nret > 0) ? be32_to_cpu(args.args[nargs]) : 0;
269}
270
271
272static void __init prom_print(const char *msg)
273{
274 const char *p, *q;
275
276 if (prom.stdout == 0)
277 return;
278
279 for (p = msg; *p != 0; p = q) {
280 for (q = p; *q != 0 && *q != '\n'; ++q)
281 ;
282 if (q > p)
283 call_prom("write", 3, 1, prom.stdout, p, q - p);
284 if (*q == 0)
285 break;
286 ++q;
287 call_prom("write", 3, 1, prom.stdout, ADDR("\r\n"), 2);
288 }
289}
290
291
292static void __init prom_print_hex(unsigned long val)
293{
294 int i, nibbles = sizeof(val)*2;
295 char buf[sizeof(val)*2+1];
296
297 for (i = nibbles-1; i >= 0; i--) {
298 buf[i] = (val & 0xf) + '0';
299 if (buf[i] > '9')
300 buf[i] += ('a'-'0'-10);
301 val >>= 4;
302 }
303 buf[nibbles] = '\0';
304 call_prom("write", 3, 1, prom.stdout, buf, nibbles);
305}
306
307/* max number of decimal digits in an unsigned long */
308#define UL_DIGITS 21
309static void __init prom_print_dec(unsigned long val)
310{
311 int i, size;
312 char buf[UL_DIGITS+1];
313
314 for (i = UL_DIGITS-1; i >= 0; i--) {
315 buf[i] = (val % 10) + '0';
316 val = val/10;
317 if (val == 0)
318 break;
319 }
320 /* shift stuff down */
321 size = UL_DIGITS - i;
322 call_prom("write", 3, 1, prom.stdout, buf+i, size);
323}
324
325static void __init prom_printf(const char *format, ...)
326{
327 const char *p, *q, *s;
328 va_list args;
329 unsigned long v;
330 long vs;
331
332 va_start(args, format);
333 for (p = format; *p != 0; p = q) {
334 for (q = p; *q != 0 && *q != '\n' && *q != '%'; ++q)
335 ;
336 if (q > p)
337 call_prom("write", 3, 1, prom.stdout, p, q - p);
338 if (*q == 0)
339 break;
340 if (*q == '\n') {
341 ++q;
342 call_prom("write", 3, 1, prom.stdout,
343 ADDR("\r\n"), 2);
344 continue;
345 }
346 ++q;
347 if (*q == 0)
348 break;
349 switch (*q) {
350 case 's':
351 ++q;
352 s = va_arg(args, const char *);
353 prom_print(s);
354 break;
355 case 'x':
356 ++q;
357 v = va_arg(args, unsigned long);
358 prom_print_hex(v);
359 break;
360 case 'd':
361 ++q;
362 vs = va_arg(args, int);
363 if (vs < 0) {
364 prom_print("-");
365 vs = -vs;
366 }
367 prom_print_dec(vs);
368 break;
369 case 'l':
370 ++q;
371 if (*q == 0)
372 break;
373 else if (*q == 'x') {
374 ++q;
375 v = va_arg(args, unsigned long);
376 prom_print_hex(v);
377 } else if (*q == 'u') { /* '%lu' */
378 ++q;
379 v = va_arg(args, unsigned long);
380 prom_print_dec(v);
381 } else if (*q == 'd') { /* %ld */
382 ++q;
383 vs = va_arg(args, long);
384 if (vs < 0) {
385 prom_print("-");
386 vs = -vs;
387 }
388 prom_print_dec(vs);
389 }
390 break;
391 }
392 }
393}
394
395
396static unsigned int __init prom_claim(unsigned long virt, unsigned long size,
397 unsigned long align)
398{
399
400 if (align == 0 && (OF_WORKAROUNDS & OF_WA_CLAIM)) {
401 /*
402 * Old OF requires we claim physical and virtual separately
403 * and then map explicitly (assuming virtual mode)
404 */
405 int ret;
406 prom_arg_t result;
407
408 ret = call_prom_ret("call-method", 5, 2, &result,
409 ADDR("claim"), prom.memory,
410 align, size, virt);
411 if (ret != 0 || result == -1)
412 return -1;
413 ret = call_prom_ret("call-method", 5, 2, &result,
414 ADDR("claim"), prom.mmumap,
415 align, size, virt);
416 if (ret != 0) {
417 call_prom("call-method", 4, 1, ADDR("release"),
418 prom.memory, size, virt);
419 return -1;
420 }
421 /* the 0x12 is M (coherence) + PP == read/write */
422 call_prom("call-method", 6, 1,
423 ADDR("map"), prom.mmumap, 0x12, size, virt, virt);
424 return virt;
425 }
426 return call_prom("claim", 3, 1, (prom_arg_t)virt, (prom_arg_t)size,
427 (prom_arg_t)align);
428}
429
430static void __init __attribute__((noreturn)) prom_panic(const char *reason)
431{
432 prom_print(reason);
433 /* Do not call exit because it clears the screen on pmac
434 * it also causes some sort of double-fault on early pmacs */
435 if (of_platform == PLATFORM_POWERMAC)
436 asm("trap\n");
437
438 /* ToDo: should put up an SRC here on pSeries */
439 call_prom("exit", 0, 0);
440
441 for (;;) /* should never get here */
442 ;
443}
444
445
446static int __init prom_next_node(phandle *nodep)
447{
448 phandle node;
449
450 if ((node = *nodep) != 0
451 && (*nodep = call_prom("child", 1, 1, node)) != 0)
452 return 1;
453 if ((*nodep = call_prom("peer", 1, 1, node)) != 0)
454 return 1;
455 for (;;) {
456 if ((node = call_prom("parent", 1, 1, node)) == 0)
457 return 0;
458 if ((*nodep = call_prom("peer", 1, 1, node)) != 0)
459 return 1;
460 }
461}
462
463static int inline prom_getprop(phandle node, const char *pname,
464 void *value, size_t valuelen)
465{
466 return call_prom("getprop", 4, 1, node, ADDR(pname),
467 (u32)(unsigned long) value, (u32) valuelen);
468}
469
470static int inline prom_getproplen(phandle node, const char *pname)
471{
472 return call_prom("getproplen", 2, 1, node, ADDR(pname));
473}
474
475static void add_string(char **str, const char *q)
476{
477 char *p = *str;
478
479 while (*q)
480 *p++ = *q++;
481 *p++ = ' ';
482 *str = p;
483}
484
485static char *tohex(unsigned int x)
486{
487 static char digits[] = "0123456789abcdef";
488 static char result[9];
489 int i;
490
491 result[8] = 0;
492 i = 8;
493 do {
494 --i;
495 result[i] = digits[x & 0xf];
496 x >>= 4;
497 } while (x != 0 && i > 0);
498 return &result[i];
499}
500
501static int __init prom_setprop(phandle node, const char *nodename,
502 const char *pname, void *value, size_t valuelen)
503{
504 char cmd[256], *p;
505
506 if (!(OF_WORKAROUNDS & OF_WA_LONGTRAIL))
507 return call_prom("setprop", 4, 1, node, ADDR(pname),
508 (u32)(unsigned long) value, (u32) valuelen);
509
510 /* gah... setprop doesn't work on longtrail, have to use interpret */
511 p = cmd;
512 add_string(&p, "dev");
513 add_string(&p, nodename);
514 add_string(&p, tohex((u32)(unsigned long) value));
515 add_string(&p, tohex(valuelen));
516 add_string(&p, tohex(ADDR(pname)));
517 add_string(&p, tohex(strlen(pname)));
518 add_string(&p, "property");
519 *p = 0;
520 return call_prom("interpret", 1, 1, (u32)(unsigned long) cmd);
521}
522
523/* We can't use the standard versions because of relocation headaches. */
524#define isxdigit(c) (('0' <= (c) && (c) <= '9') \
525 || ('a' <= (c) && (c) <= 'f') \
526 || ('A' <= (c) && (c) <= 'F'))
527
528#define isdigit(c) ('0' <= (c) && (c) <= '9')
529#define islower(c) ('a' <= (c) && (c) <= 'z')
530#define toupper(c) (islower(c) ? ((c) - 'a' + 'A') : (c))
531
532static unsigned long prom_strtoul(const char *cp, const char **endp)
533{
534 unsigned long result = 0, base = 10, value;
535
536 if (*cp == '0') {
537 base = 8;
538 cp++;
539 if (toupper(*cp) == 'X') {
540 cp++;
541 base = 16;
542 }
543 }
544
545 while (isxdigit(*cp) &&
546 (value = isdigit(*cp) ? *cp - '0' : toupper(*cp) - 'A' + 10) < base) {
547 result = result * base + value;
548 cp++;
549 }
550
551 if (endp)
552 *endp = cp;
553
554 return result;
555}
556
557static unsigned long prom_memparse(const char *ptr, const char **retptr)
558{
559 unsigned long ret = prom_strtoul(ptr, retptr);
560 int shift = 0;
561
562 /*
563 * We can't use a switch here because GCC *may* generate a
564 * jump table which won't work, because we're not running at
565 * the address we're linked at.
566 */
567 if ('G' == **retptr || 'g' == **retptr)
568 shift = 30;
569
570 if ('M' == **retptr || 'm' == **retptr)
571 shift = 20;
572
573 if ('K' == **retptr || 'k' == **retptr)
574 shift = 10;
575
576 if (shift) {
577 ret <<= shift;
578 (*retptr)++;
579 }
580
581 return ret;
582}
583
584/*
585 * Early parsing of the command line passed to the kernel, used for
586 * "mem=x" and the options that affect the iommu
587 */
588static void __init early_cmdline_parse(void)
589{
590 const char *opt;
591
592 char *p;
593 int l = 0;
594
595 prom_cmd_line[0] = 0;
596 p = prom_cmd_line;
597 if ((long)prom.chosen > 0)
598 l = prom_getprop(prom.chosen, "bootargs", p, COMMAND_LINE_SIZE-1);
599#ifdef CONFIG_CMDLINE
600 if (l <= 0 || p[0] == '\0') /* dbl check */
601 strlcpy(prom_cmd_line,
602 CONFIG_CMDLINE, sizeof(prom_cmd_line));
603#endif /* CONFIG_CMDLINE */
604 prom_printf("command line: %s\n", prom_cmd_line);
605
606#ifdef CONFIG_PPC64
607 opt = strstr(prom_cmd_line, "iommu=");
608 if (opt) {
609 prom_printf("iommu opt is: %s\n", opt);
610 opt += 6;
611 while (*opt && *opt == ' ')
612 opt++;
613 if (!strncmp(opt, "off", 3))
614 prom_iommu_off = 1;
615 else if (!strncmp(opt, "force", 5))
616 prom_iommu_force_on = 1;
617 }
618#endif
619 opt = strstr(prom_cmd_line, "mem=");
620 if (opt) {
621 opt += 4;
622 prom_memory_limit = prom_memparse(opt, (const char **)&opt);
623#ifdef CONFIG_PPC64
624 /* Align to 16 MB == size of ppc64 large page */
625 prom_memory_limit = ALIGN(prom_memory_limit, 0x1000000);
626#endif
627 }
628}
629
630#if defined(CONFIG_PPC_PSERIES) || defined(CONFIG_PPC_POWERNV)
631/*
632 * The architecture vector has an array of PVR mask/value pairs,
633 * followed by # option vectors - 1, followed by the option vectors.
634 *
635 * See prom.h for the definition of the bits specified in the
636 * architecture vector.
637 *
638 * Because the description vector contains a mix of byte and word
639 * values, we declare it as an unsigned char array, and use this
640 * macro to put word values in.
641 */
642#define W(x) ((x) >> 24) & 0xff, ((x) >> 16) & 0xff, \
643 ((x) >> 8) & 0xff, (x) & 0xff
644
645unsigned char ibm_architecture_vec[] = {
646 W(0xfffe0000), W(0x003a0000), /* POWER5/POWER5+ */
647 W(0xffff0000), W(0x003e0000), /* POWER6 */
648 W(0xffff0000), W(0x003f0000), /* POWER7 */
649 W(0xffff0000), W(0x004b0000), /* POWER8E */
650 W(0xffff0000), W(0x004d0000), /* POWER8 */
651 W(0xffffffff), W(0x0f000004), /* all 2.07-compliant */
652 W(0xffffffff), W(0x0f000003), /* all 2.06-compliant */
653 W(0xffffffff), W(0x0f000002), /* all 2.05-compliant */
654 W(0xfffffffe), W(0x0f000001), /* all 2.04-compliant and earlier */
655 6 - 1, /* 6 option vectors */
656
657 /* option vector 1: processor architectures supported */
658 3 - 2, /* length */
659 0, /* don't ignore, don't halt */
660 OV1_PPC_2_00 | OV1_PPC_2_01 | OV1_PPC_2_02 | OV1_PPC_2_03 |
661 OV1_PPC_2_04 | OV1_PPC_2_05 | OV1_PPC_2_06 | OV1_PPC_2_07,
662
663 /* option vector 2: Open Firmware options supported */
664 34 - 2, /* length */
665 OV2_REAL_MODE,
666 0, 0,
667 W(0xffffffff), /* real_base */
668 W(0xffffffff), /* real_size */
669 W(0xffffffff), /* virt_base */
670 W(0xffffffff), /* virt_size */
671 W(0xffffffff), /* load_base */
672 W(256), /* 256MB min RMA */
673 W(0xffffffff), /* full client load */
674 0, /* min RMA percentage of total RAM */
675 48, /* max log_2(hash table size) */
676
677 /* option vector 3: processor options supported */
678 3 - 2, /* length */
679 0, /* don't ignore, don't halt */
680 OV3_FP | OV3_VMX | OV3_DFP,
681
682 /* option vector 4: IBM PAPR implementation */
683 3 - 2, /* length */
684 0, /* don't halt */
685 OV4_MIN_ENT_CAP, /* minimum VP entitled capacity */
686
687 /* option vector 5: PAPR/OF options */
688 19 - 2, /* length */
689 0, /* don't ignore, don't halt */
690 OV5_FEAT(OV5_LPAR) | OV5_FEAT(OV5_SPLPAR) | OV5_FEAT(OV5_LARGE_PAGES) |
691 OV5_FEAT(OV5_DRCONF_MEMORY) | OV5_FEAT(OV5_DONATE_DEDICATE_CPU) |
692#ifdef CONFIG_PCI_MSI
693 /* PCIe/MSI support. Without MSI full PCIe is not supported */
694 OV5_FEAT(OV5_MSI),
695#else
696 0,
697#endif
698 0,
699#ifdef CONFIG_PPC_SMLPAR
700 OV5_FEAT(OV5_CMO) | OV5_FEAT(OV5_XCMO),
701#else
702 0,
703#endif
704 OV5_FEAT(OV5_TYPE1_AFFINITY) | OV5_FEAT(OV5_PRRN),
705 0,
706 0,
707 0,
708 /* WARNING: The offset of the "number of cores" field below
709 * must match by the macro below. Update the definition if
710 * the structure layout changes.
711 */
712#define IBM_ARCH_VEC_NRCORES_OFFSET 125
713 W(NR_CPUS), /* number of cores supported */
714 0,
715 0,
716 0,
717 0,
718 OV5_FEAT(OV5_PFO_HW_RNG) | OV5_FEAT(OV5_PFO_HW_ENCR) |
719 OV5_FEAT(OV5_PFO_HW_842),
720 OV5_FEAT(OV5_SUB_PROCESSORS),
721 /* option vector 6: IBM PAPR hints */
722 4 - 2, /* length */
723 0,
724 0,
725 OV6_LINUX,
726
727};
728
729/* Old method - ELF header with PT_NOTE sections only works on BE */
730#ifdef __BIG_ENDIAN__
731static struct fake_elf {
732 Elf32_Ehdr elfhdr;
733 Elf32_Phdr phdr[2];
734 struct chrpnote {
735 u32 namesz;
736 u32 descsz;
737 u32 type;
738 char name[8]; /* "PowerPC" */
739 struct chrpdesc {
740 u32 real_mode;
741 u32 real_base;
742 u32 real_size;
743 u32 virt_base;
744 u32 virt_size;
745 u32 load_base;
746 } chrpdesc;
747 } chrpnote;
748 struct rpanote {
749 u32 namesz;
750 u32 descsz;
751 u32 type;
752 char name[24]; /* "IBM,RPA-Client-Config" */
753 struct rpadesc {
754 u32 lpar_affinity;
755 u32 min_rmo_size;
756 u32 min_rmo_percent;
757 u32 max_pft_size;
758 u32 splpar;
759 u32 min_load;
760 u32 new_mem_def;
761 u32 ignore_me;
762 } rpadesc;
763 } rpanote;
764} fake_elf = {
765 .elfhdr = {
766 .e_ident = { 0x7f, 'E', 'L', 'F',
767 ELFCLASS32, ELFDATA2MSB, EV_CURRENT },
768 .e_type = ET_EXEC, /* yeah right */
769 .e_machine = EM_PPC,
770 .e_version = EV_CURRENT,
771 .e_phoff = offsetof(struct fake_elf, phdr),
772 .e_phentsize = sizeof(Elf32_Phdr),
773 .e_phnum = 2
774 },
775 .phdr = {
776 [0] = {
777 .p_type = PT_NOTE,
778 .p_offset = offsetof(struct fake_elf, chrpnote),
779 .p_filesz = sizeof(struct chrpnote)
780 }, [1] = {
781 .p_type = PT_NOTE,
782 .p_offset = offsetof(struct fake_elf, rpanote),
783 .p_filesz = sizeof(struct rpanote)
784 }
785 },
786 .chrpnote = {
787 .namesz = sizeof("PowerPC"),
788 .descsz = sizeof(struct chrpdesc),
789 .type = 0x1275,
790 .name = "PowerPC",
791 .chrpdesc = {
792 .real_mode = ~0U, /* ~0 means "don't care" */
793 .real_base = ~0U,
794 .real_size = ~0U,
795 .virt_base = ~0U,
796 .virt_size = ~0U,
797 .load_base = ~0U
798 },
799 },
800 .rpanote = {
801 .namesz = sizeof("IBM,RPA-Client-Config"),
802 .descsz = sizeof(struct rpadesc),
803 .type = 0x12759999,
804 .name = "IBM,RPA-Client-Config",
805 .rpadesc = {
806 .lpar_affinity = 0,
807 .min_rmo_size = 64, /* in megabytes */
808 .min_rmo_percent = 0,
809 .max_pft_size = 48, /* 2^48 bytes max PFT size */
810 .splpar = 1,
811 .min_load = ~0U,
812 .new_mem_def = 0
813 }
814 }
815};
816#endif /* __BIG_ENDIAN__ */
817
818static int __init prom_count_smt_threads(void)
819{
820 phandle node;
821 char type[64];
822 unsigned int plen;
823
824 /* Pick up th first CPU node we can find */
825 for (node = 0; prom_next_node(&node); ) {
826 type[0] = 0;
827 prom_getprop(node, "device_type", type, sizeof(type));
828
829 if (strcmp(type, "cpu"))
830 continue;
831 /*
832 * There is an entry for each smt thread, each entry being
833 * 4 bytes long. All cpus should have the same number of
834 * smt threads, so return after finding the first.
835 */
836 plen = prom_getproplen(node, "ibm,ppc-interrupt-server#s");
837 if (plen == PROM_ERROR)
838 break;
839 plen >>= 2;
840 prom_debug("Found %lu smt threads per core\n", (unsigned long)plen);
841
842 /* Sanity check */
843 if (plen < 1 || plen > 64) {
844 prom_printf("Threads per core %lu out of bounds, assuming 1\n",
845 (unsigned long)plen);
846 return 1;
847 }
848 return plen;
849 }
850 prom_debug("No threads found, assuming 1 per core\n");
851
852 return 1;
853
854}
855
856
857static void __init prom_send_capabilities(void)
858{
859 ihandle root;
860 prom_arg_t ret;
861 u32 cores;
862 unsigned char *ptcores;
863
864 root = call_prom("open", 1, 1, ADDR("/"));
865 if (root != 0) {
866 /* We need to tell the FW about the number of cores we support.
867 *
868 * To do that, we count the number of threads on the first core
869 * (we assume this is the same for all cores) and use it to
870 * divide NR_CPUS.
871 */
872
873 /* The core value may start at an odd address. If such a word
874 * access is made at a cache line boundary, this leads to an
875 * exception which may not be handled at this time.
876 * Forcing a per byte access to avoid exception.
877 */
878 ptcores = &ibm_architecture_vec[IBM_ARCH_VEC_NRCORES_OFFSET];
879 cores = 0;
880 cores |= ptcores[0] << 24;
881 cores |= ptcores[1] << 16;
882 cores |= ptcores[2] << 8;
883 cores |= ptcores[3];
884 if (cores != NR_CPUS) {
885 prom_printf("WARNING ! "
886 "ibm_architecture_vec structure inconsistent: %lu!\n",
887 cores);
888 } else {
889 cores = DIV_ROUND_UP(NR_CPUS, prom_count_smt_threads());
890 prom_printf("Max number of cores passed to firmware: %lu (NR_CPUS = %lu)\n",
891 cores, NR_CPUS);
892 ptcores[0] = (cores >> 24) & 0xff;
893 ptcores[1] = (cores >> 16) & 0xff;
894 ptcores[2] = (cores >> 8) & 0xff;
895 ptcores[3] = cores & 0xff;
896 }
897
898 /* try calling the ibm,client-architecture-support method */
899 prom_printf("Calling ibm,client-architecture-support...");
900 if (call_prom_ret("call-method", 3, 2, &ret,
901 ADDR("ibm,client-architecture-support"),
902 root,
903 ADDR(ibm_architecture_vec)) == 0) {
904 /* the call exists... */
905 if (ret)
906 prom_printf("\nWARNING: ibm,client-architecture"
907 "-support call FAILED!\n");
908 call_prom("close", 1, 0, root);
909 prom_printf(" done\n");
910 return;
911 }
912 call_prom("close", 1, 0, root);
913 prom_printf(" not implemented\n");
914 }
915
916#ifdef __BIG_ENDIAN__
917 {
918 ihandle elfloader;
919
920 /* no ibm,client-architecture-support call, try the old way */
921 elfloader = call_prom("open", 1, 1,
922 ADDR("/packages/elf-loader"));
923 if (elfloader == 0) {
924 prom_printf("couldn't open /packages/elf-loader\n");
925 return;
926 }
927 call_prom("call-method", 3, 1, ADDR("process-elf-header"),
928 elfloader, ADDR(&fake_elf));
929 call_prom("close", 1, 0, elfloader);
930 }
931#endif /* __BIG_ENDIAN__ */
932}
933#endif /* #if defined(CONFIG_PPC_PSERIES) || defined(CONFIG_PPC_POWERNV) */
934
935/*
936 * Memory allocation strategy... our layout is normally:
937 *
938 * at 14Mb or more we have vmlinux, then a gap and initrd. In some
939 * rare cases, initrd might end up being before the kernel though.
940 * We assume this won't override the final kernel at 0, we have no
941 * provision to handle that in this version, but it should hopefully
942 * never happen.
943 *
944 * alloc_top is set to the top of RMO, eventually shrink down if the
945 * TCEs overlap
946 *
947 * alloc_bottom is set to the top of kernel/initrd
948 *
949 * from there, allocations are done this way : rtas is allocated
950 * topmost, and the device-tree is allocated from the bottom. We try
951 * to grow the device-tree allocation as we progress. If we can't,
952 * then we fail, we don't currently have a facility to restart
953 * elsewhere, but that shouldn't be necessary.
954 *
955 * Note that calls to reserve_mem have to be done explicitly, memory
956 * allocated with either alloc_up or alloc_down isn't automatically
957 * reserved.
958 */
959
960
961/*
962 * Allocates memory in the RMO upward from the kernel/initrd
963 *
964 * When align is 0, this is a special case, it means to allocate in place
965 * at the current location of alloc_bottom or fail (that is basically
966 * extending the previous allocation). Used for the device-tree flattening
967 */
968static unsigned long __init alloc_up(unsigned long size, unsigned long align)
969{
970 unsigned long base = alloc_bottom;
971 unsigned long addr = 0;
972
973 if (align)
974 base = _ALIGN_UP(base, align);
975 prom_debug("alloc_up(%x, %x)\n", size, align);
976 if (ram_top == 0)
977 prom_panic("alloc_up() called with mem not initialized\n");
978
979 if (align)
980 base = _ALIGN_UP(alloc_bottom, align);
981 else
982 base = alloc_bottom;
983
984 for(; (base + size) <= alloc_top;
985 base = _ALIGN_UP(base + 0x100000, align)) {
986 prom_debug(" trying: 0x%x\n\r", base);
987 addr = (unsigned long)prom_claim(base, size, 0);
988 if (addr != PROM_ERROR && addr != 0)
989 break;
990 addr = 0;
991 if (align == 0)
992 break;
993 }
994 if (addr == 0)
995 return 0;
996 alloc_bottom = addr + size;
997
998 prom_debug(" -> %x\n", addr);
999 prom_debug(" alloc_bottom : %x\n", alloc_bottom);
1000 prom_debug(" alloc_top : %x\n", alloc_top);
1001 prom_debug(" alloc_top_hi : %x\n", alloc_top_high);
1002 prom_debug(" rmo_top : %x\n", rmo_top);
1003 prom_debug(" ram_top : %x\n", ram_top);
1004
1005 return addr;
1006}
1007
1008/*
1009 * Allocates memory downward, either from top of RMO, or if highmem
1010 * is set, from the top of RAM. Note that this one doesn't handle
1011 * failures. It does claim memory if highmem is not set.
1012 */
1013static unsigned long __init alloc_down(unsigned long size, unsigned long align,
1014 int highmem)
1015{
1016 unsigned long base, addr = 0;
1017
1018 prom_debug("alloc_down(%x, %x, %s)\n", size, align,
1019 highmem ? "(high)" : "(low)");
1020 if (ram_top == 0)
1021 prom_panic("alloc_down() called with mem not initialized\n");
1022
1023 if (highmem) {
1024 /* Carve out storage for the TCE table. */
1025 addr = _ALIGN_DOWN(alloc_top_high - size, align);
1026 if (addr <= alloc_bottom)
1027 return 0;
1028 /* Will we bump into the RMO ? If yes, check out that we
1029 * didn't overlap existing allocations there, if we did,
1030 * we are dead, we must be the first in town !
1031 */
1032 if (addr < rmo_top) {
1033 /* Good, we are first */
1034 if (alloc_top == rmo_top)
1035 alloc_top = rmo_top = addr;
1036 else
1037 return 0;
1038 }
1039 alloc_top_high = addr;
1040 goto bail;
1041 }
1042
1043 base = _ALIGN_DOWN(alloc_top - size, align);
1044 for (; base > alloc_bottom;
1045 base = _ALIGN_DOWN(base - 0x100000, align)) {
1046 prom_debug(" trying: 0x%x\n\r", base);
1047 addr = (unsigned long)prom_claim(base, size, 0);
1048 if (addr != PROM_ERROR && addr != 0)
1049 break;
1050 addr = 0;
1051 }
1052 if (addr == 0)
1053 return 0;
1054 alloc_top = addr;
1055
1056 bail:
1057 prom_debug(" -> %x\n", addr);
1058 prom_debug(" alloc_bottom : %x\n", alloc_bottom);
1059 prom_debug(" alloc_top : %x\n", alloc_top);
1060 prom_debug(" alloc_top_hi : %x\n", alloc_top_high);
1061 prom_debug(" rmo_top : %x\n", rmo_top);
1062 prom_debug(" ram_top : %x\n", ram_top);
1063
1064 return addr;
1065}
1066
1067/*
1068 * Parse a "reg" cell
1069 */
1070static unsigned long __init prom_next_cell(int s, cell_t **cellp)
1071{
1072 cell_t *p = *cellp;
1073 unsigned long r = 0;
1074
1075 /* Ignore more than 2 cells */
1076 while (s > sizeof(unsigned long) / 4) {
1077 p++;
1078 s--;
1079 }
1080 r = be32_to_cpu(*p++);
1081#ifdef CONFIG_PPC64
1082 if (s > 1) {
1083 r <<= 32;
1084 r |= be32_to_cpu(*(p++));
1085 }
1086#endif
1087 *cellp = p;
1088 return r;
1089}
1090
1091/*
1092 * Very dumb function for adding to the memory reserve list, but
1093 * we don't need anything smarter at this point
1094 *
1095 * XXX Eventually check for collisions. They should NEVER happen.
1096 * If problems seem to show up, it would be a good start to track
1097 * them down.
1098 */
1099static void __init reserve_mem(u64 base, u64 size)
1100{
1101 u64 top = base + size;
1102 unsigned long cnt = mem_reserve_cnt;
1103
1104 if (size == 0)
1105 return;
1106
1107 /* We need to always keep one empty entry so that we
1108 * have our terminator with "size" set to 0 since we are
1109 * dumb and just copy this entire array to the boot params
1110 */
1111 base = _ALIGN_DOWN(base, PAGE_SIZE);
1112 top = _ALIGN_UP(top, PAGE_SIZE);
1113 size = top - base;
1114
1115 if (cnt >= (MEM_RESERVE_MAP_SIZE - 1))
1116 prom_panic("Memory reserve map exhausted !\n");
1117 mem_reserve_map[cnt].base = cpu_to_be64(base);
1118 mem_reserve_map[cnt].size = cpu_to_be64(size);
1119 mem_reserve_cnt = cnt + 1;
1120}
1121
1122/*
1123 * Initialize memory allocation mechanism, parse "memory" nodes and
1124 * obtain that way the top of memory and RMO to setup out local allocator
1125 */
1126static void __init prom_init_mem(void)
1127{
1128 phandle node;
1129 char *path, type[64];
1130 unsigned int plen;
1131 cell_t *p, *endp;
1132 __be32 val;
1133 u32 rac, rsc;
1134
1135 /*
1136 * We iterate the memory nodes to find
1137 * 1) top of RMO (first node)
1138 * 2) top of memory
1139 */
1140 val = cpu_to_be32(2);
1141 prom_getprop(prom.root, "#address-cells", &val, sizeof(val));
1142 rac = be32_to_cpu(val);
1143 val = cpu_to_be32(1);
1144 prom_getprop(prom.root, "#size-cells", &val, sizeof(rsc));
1145 rsc = be32_to_cpu(val);
1146 prom_debug("root_addr_cells: %x\n", rac);
1147 prom_debug("root_size_cells: %x\n", rsc);
1148
1149 prom_debug("scanning memory:\n");
1150 path = prom_scratch;
1151
1152 for (node = 0; prom_next_node(&node); ) {
1153 type[0] = 0;
1154 prom_getprop(node, "device_type", type, sizeof(type));
1155
1156 if (type[0] == 0) {
1157 /*
1158 * CHRP Longtrail machines have no device_type
1159 * on the memory node, so check the name instead...
1160 */
1161 prom_getprop(node, "name", type, sizeof(type));
1162 }
1163 if (strcmp(type, "memory"))
1164 continue;
1165
1166 plen = prom_getprop(node, "reg", regbuf, sizeof(regbuf));
1167 if (plen > sizeof(regbuf)) {
1168 prom_printf("memory node too large for buffer !\n");
1169 plen = sizeof(regbuf);
1170 }
1171 p = regbuf;
1172 endp = p + (plen / sizeof(cell_t));
1173
1174#ifdef DEBUG_PROM
1175 memset(path, 0, PROM_SCRATCH_SIZE);
1176 call_prom("package-to-path", 3, 1, node, path, PROM_SCRATCH_SIZE-1);
1177 prom_debug(" node %s :\n", path);
1178#endif /* DEBUG_PROM */
1179
1180 while ((endp - p) >= (rac + rsc)) {
1181 unsigned long base, size;
1182
1183 base = prom_next_cell(rac, &p);
1184 size = prom_next_cell(rsc, &p);
1185
1186 if (size == 0)
1187 continue;
1188 prom_debug(" %x %x\n", base, size);
1189 if (base == 0 && (of_platform & PLATFORM_LPAR))
1190 rmo_top = size;
1191 if ((base + size) > ram_top)
1192 ram_top = base + size;
1193 }
1194 }
1195
1196 alloc_bottom = PAGE_ALIGN((unsigned long)&_end + 0x4000);
1197
1198 /*
1199 * If prom_memory_limit is set we reduce the upper limits *except* for
1200 * alloc_top_high. This must be the real top of RAM so we can put
1201 * TCE's up there.
1202 */
1203
1204 alloc_top_high = ram_top;
1205
1206 if (prom_memory_limit) {
1207 if (prom_memory_limit <= alloc_bottom) {
1208 prom_printf("Ignoring mem=%x <= alloc_bottom.\n",
1209 prom_memory_limit);
1210 prom_memory_limit = 0;
1211 } else if (prom_memory_limit >= ram_top) {
1212 prom_printf("Ignoring mem=%x >= ram_top.\n",
1213 prom_memory_limit);
1214 prom_memory_limit = 0;
1215 } else {
1216 ram_top = prom_memory_limit;
1217 rmo_top = min(rmo_top, prom_memory_limit);
1218 }
1219 }
1220
1221 /*
1222 * Setup our top alloc point, that is top of RMO or top of
1223 * segment 0 when running non-LPAR.
1224 * Some RS64 machines have buggy firmware where claims up at
1225 * 1GB fail. Cap at 768MB as a workaround.
1226 * Since 768MB is plenty of room, and we need to cap to something
1227 * reasonable on 32-bit, cap at 768MB on all machines.
1228 */
1229 if (!rmo_top)
1230 rmo_top = ram_top;
1231 rmo_top = min(0x30000000ul, rmo_top);
1232 alloc_top = rmo_top;
1233 alloc_top_high = ram_top;
1234
1235 /*
1236 * Check if we have an initrd after the kernel but still inside
1237 * the RMO. If we do move our bottom point to after it.
1238 */
1239 if (prom_initrd_start &&
1240 prom_initrd_start < rmo_top &&
1241 prom_initrd_end > alloc_bottom)
1242 alloc_bottom = PAGE_ALIGN(prom_initrd_end);
1243
1244 prom_printf("memory layout at init:\n");
1245 prom_printf(" memory_limit : %x (16 MB aligned)\n", prom_memory_limit);
1246 prom_printf(" alloc_bottom : %x\n", alloc_bottom);
1247 prom_printf(" alloc_top : %x\n", alloc_top);
1248 prom_printf(" alloc_top_hi : %x\n", alloc_top_high);
1249 prom_printf(" rmo_top : %x\n", rmo_top);
1250 prom_printf(" ram_top : %x\n", ram_top);
1251}
1252
1253static void __init prom_close_stdin(void)
1254{
1255 __be32 val;
1256 ihandle stdin;
1257
1258 if (prom_getprop(prom.chosen, "stdin", &val, sizeof(val)) > 0) {
1259 stdin = be32_to_cpu(val);
1260 call_prom("close", 1, 0, stdin);
1261 }
1262}
1263
1264#ifdef CONFIG_PPC_POWERNV
1265
1266#ifdef CONFIG_PPC_EARLY_DEBUG_OPAL
1267static u64 __initdata prom_opal_base;
1268static u64 __initdata prom_opal_entry;
1269#endif
1270
1271#ifdef __BIG_ENDIAN__
1272/* XXX Don't change this structure without updating opal-takeover.S */
1273static struct opal_secondary_data {
1274 s64 ack; /* 0 */
1275 u64 go; /* 8 */
1276 struct opal_takeover_args args; /* 16 */
1277} opal_secondary_data;
1278
1279static u64 __initdata prom_opal_align;
1280static u64 __initdata prom_opal_size;
1281static int __initdata prom_rtas_start_cpu;
1282static u64 __initdata prom_rtas_data;
1283static u64 __initdata prom_rtas_entry;
1284
1285extern char opal_secondary_entry;
1286
1287static void __init prom_query_opal(void)
1288{
1289 long rc;
1290
1291 /* We must not query for OPAL presence on a machine that
1292 * supports TNK takeover (970 blades), as this uses the same
1293 * h-call with different arguments and will crash
1294 */
1295 if (PHANDLE_VALID(call_prom("finddevice", 1, 1,
1296 ADDR("/tnk-memory-map")))) {
1297 prom_printf("TNK takeover detected, skipping OPAL check\n");
1298 return;
1299 }
1300
1301 prom_printf("Querying for OPAL presence... ");
1302
1303 rc = opal_query_takeover(&prom_opal_size,
1304 &prom_opal_align);
1305 prom_debug("(rc = %ld) ", rc);
1306 if (rc != 0) {
1307 prom_printf("not there.\n");
1308 return;
1309 }
1310 of_platform = PLATFORM_OPAL;
1311 prom_printf(" there !\n");
1312 prom_debug(" opal_size = 0x%lx\n", prom_opal_size);
1313 prom_debug(" opal_align = 0x%lx\n", prom_opal_align);
1314 if (prom_opal_align < 0x10000)
1315 prom_opal_align = 0x10000;
1316}
1317
1318static int __init prom_rtas_call(int token, int nargs, int nret,
1319 int *outputs, ...)
1320{
1321 struct rtas_args rtas_args;
1322 va_list list;
1323 int i;
1324
1325 rtas_args.token = token;
1326 rtas_args.nargs = nargs;
1327 rtas_args.nret = nret;
1328 rtas_args.rets = (rtas_arg_t *)&(rtas_args.args[nargs]);
1329 va_start(list, outputs);
1330 for (i = 0; i < nargs; ++i)
1331 rtas_args.args[i] = va_arg(list, rtas_arg_t);
1332 va_end(list);
1333
1334 for (i = 0; i < nret; ++i)
1335 rtas_args.rets[i] = 0;
1336
1337 opal_enter_rtas(&rtas_args, prom_rtas_data,
1338 prom_rtas_entry);
1339
1340 if (nret > 1 && outputs != NULL)
1341 for (i = 0; i < nret-1; ++i)
1342 outputs[i] = rtas_args.rets[i+1];
1343 return (nret > 0)? rtas_args.rets[0]: 0;
1344}
1345
1346static void __init prom_opal_hold_cpus(void)
1347{
1348 int i, cnt, cpu, rc;
1349 long j;
1350 phandle node;
1351 char type[64];
1352 u32 servers[8];
1353 void *entry = (unsigned long *)&opal_secondary_entry;
1354 struct opal_secondary_data *data = &opal_secondary_data;
1355
1356 prom_debug("prom_opal_hold_cpus: start...\n");
1357 prom_debug(" - entry = 0x%x\n", entry);
1358 prom_debug(" - data = 0x%x\n", data);
1359
1360 data->ack = -1;
1361 data->go = 0;
1362
1363 /* look for cpus */
1364 for (node = 0; prom_next_node(&node); ) {
1365 type[0] = 0;
1366 prom_getprop(node, "device_type", type, sizeof(type));
1367 if (strcmp(type, "cpu") != 0)
1368 continue;
1369
1370 /* Skip non-configured cpus. */
1371 if (prom_getprop(node, "status", type, sizeof(type)) > 0)
1372 if (strcmp(type, "okay") != 0)
1373 continue;
1374
1375 cnt = prom_getprop(node, "ibm,ppc-interrupt-server#s", servers,
1376 sizeof(servers));
1377 if (cnt == PROM_ERROR)
1378 break;
1379 cnt >>= 2;
1380 for (i = 0; i < cnt; i++) {
1381 cpu = servers[i];
1382 prom_debug("CPU %d ... ", cpu);
1383 if (cpu == prom.cpu) {
1384 prom_debug("booted !\n");
1385 continue;
1386 }
1387 prom_debug("starting ... ");
1388
1389 /* Init the acknowledge var which will be reset by
1390 * the secondary cpu when it awakens from its OF
1391 * spinloop.
1392 */
1393 data->ack = -1;
1394 rc = prom_rtas_call(prom_rtas_start_cpu, 3, 1,
1395 NULL, cpu, entry, data);
1396 prom_debug("rtas rc=%d ...", rc);
1397
1398 for (j = 0; j < 100000000 && data->ack == -1; j++) {
1399 HMT_low();
1400 mb();
1401 }
1402 HMT_medium();
1403 if (data->ack != -1)
1404 prom_debug("done, PIR=0x%x\n", data->ack);
1405 else
1406 prom_debug("timeout !\n");
1407 }
1408 }
1409 prom_debug("prom_opal_hold_cpus: end...\n");
1410}
1411
1412static void __init prom_opal_takeover(void)
1413{
1414 struct opal_secondary_data *data = &opal_secondary_data;
1415 struct opal_takeover_args *args = &data->args;
1416 u64 align = prom_opal_align;
1417 u64 top_addr, opal_addr;
1418
1419 args->k_image = (u64)_stext;
1420 args->k_size = _end - _stext;
1421 args->k_entry = 0;
1422 args->k_entry2 = 0x60;
1423
1424 top_addr = _ALIGN_UP(args->k_size, align);
1425
1426 if (prom_initrd_start != 0) {
1427 args->rd_image = prom_initrd_start;
1428 args->rd_size = prom_initrd_end - args->rd_image;
1429 args->rd_loc = top_addr;
1430 top_addr = _ALIGN_UP(args->rd_loc + args->rd_size, align);
1431 }
1432
1433 /* Pickup an address for the HAL. We want to go really high
1434 * up to avoid problem with future kexecs. On the other hand
1435 * we don't want to be all over the TCEs on P5IOC2 machines
1436 * which are going to be up there too. We assume the machine
1437 * has plenty of memory, and we ask for the HAL for now to
1438 * be just below the 1G point, or above the initrd
1439 */
1440 opal_addr = _ALIGN_DOWN(0x40000000 - prom_opal_size, align);
1441 if (opal_addr < top_addr)
1442 opal_addr = top_addr;
1443 args->hal_addr = opal_addr;
1444
1445 /* Copy the command line to the kernel image */
1446 strlcpy(boot_command_line, prom_cmd_line,
1447 COMMAND_LINE_SIZE);
1448
1449 prom_debug(" k_image = 0x%lx\n", args->k_image);
1450 prom_debug(" k_size = 0x%lx\n", args->k_size);
1451 prom_debug(" k_entry = 0x%lx\n", args->k_entry);
1452 prom_debug(" k_entry2 = 0x%lx\n", args->k_entry2);
1453 prom_debug(" hal_addr = 0x%lx\n", args->hal_addr);
1454 prom_debug(" rd_image = 0x%lx\n", args->rd_image);
1455 prom_debug(" rd_size = 0x%lx\n", args->rd_size);
1456 prom_debug(" rd_loc = 0x%lx\n", args->rd_loc);
1457 prom_printf("Performing OPAL takeover,this can take a few minutes..\n");
1458 prom_close_stdin();
1459 mb();
1460 data->go = 1;
1461 for (;;)
1462 opal_do_takeover(args);
1463}
1464#endif /* __BIG_ENDIAN__ */
1465
1466/*
1467 * Allocate room for and instantiate OPAL
1468 */
1469static void __init prom_instantiate_opal(void)
1470{
1471 phandle opal_node;
1472 ihandle opal_inst;
1473 u64 base, entry;
1474 u64 size = 0, align = 0x10000;
1475 __be64 val64;
1476 u32 rets[2];
1477
1478 prom_debug("prom_instantiate_opal: start...\n");
1479
1480 opal_node = call_prom("finddevice", 1, 1, ADDR("/ibm,opal"));
1481 prom_debug("opal_node: %x\n", opal_node);
1482 if (!PHANDLE_VALID(opal_node))
1483 return;
1484
1485 val64 = 0;
1486 prom_getprop(opal_node, "opal-runtime-size", &val64, sizeof(val64));
1487 size = be64_to_cpu(val64);
1488 if (size == 0)
1489 return;
1490 val64 = 0;
1491 prom_getprop(opal_node, "opal-runtime-alignment", &val64,sizeof(val64));
1492 align = be64_to_cpu(val64);
1493
1494 base = alloc_down(size, align, 0);
1495 if (base == 0) {
1496 prom_printf("OPAL allocation failed !\n");
1497 return;
1498 }
1499
1500 opal_inst = call_prom("open", 1, 1, ADDR("/ibm,opal"));
1501 if (!IHANDLE_VALID(opal_inst)) {
1502 prom_printf("opening opal package failed (%x)\n", opal_inst);
1503 return;
1504 }
1505
1506 prom_printf("instantiating opal at 0x%x...", base);
1507
1508 if (call_prom_ret("call-method", 4, 3, rets,
1509 ADDR("load-opal-runtime"),
1510 opal_inst,
1511 base >> 32, base & 0xffffffff) != 0
1512 || (rets[0] == 0 && rets[1] == 0)) {
1513 prom_printf(" failed\n");
1514 return;
1515 }
1516 entry = (((u64)rets[0]) << 32) | rets[1];
1517
1518 prom_printf(" done\n");
1519
1520 reserve_mem(base, size);
1521
1522 prom_debug("opal base = 0x%x\n", base);
1523 prom_debug("opal align = 0x%x\n", align);
1524 prom_debug("opal entry = 0x%x\n", entry);
1525 prom_debug("opal size = 0x%x\n", (long)size);
1526
1527 prom_setprop(opal_node, "/ibm,opal", "opal-base-address",
1528 &base, sizeof(base));
1529 prom_setprop(opal_node, "/ibm,opal", "opal-entry-address",
1530 &entry, sizeof(entry));
1531
1532#ifdef CONFIG_PPC_EARLY_DEBUG_OPAL
1533 prom_opal_base = base;
1534 prom_opal_entry = entry;
1535#endif
1536 prom_debug("prom_instantiate_opal: end...\n");
1537}
1538
1539#endif /* CONFIG_PPC_POWERNV */
1540
1541/*
1542 * Allocate room for and instantiate RTAS
1543 */
1544static void __init prom_instantiate_rtas(void)
1545{
1546 phandle rtas_node;
1547 ihandle rtas_inst;
1548 u32 base, entry = 0;
1549 __be32 val;
1550 u32 size = 0;
1551
1552 prom_debug("prom_instantiate_rtas: start...\n");
1553
1554 rtas_node = call_prom("finddevice", 1, 1, ADDR("/rtas"));
1555 prom_debug("rtas_node: %x\n", rtas_node);
1556 if (!PHANDLE_VALID(rtas_node))
1557 return;
1558
1559 val = 0;
1560 prom_getprop(rtas_node, "rtas-size", &val, sizeof(size));
1561 size = be32_to_cpu(val);
1562 if (size == 0)
1563 return;
1564
1565 base = alloc_down(size, PAGE_SIZE, 0);
1566 if (base == 0)
1567 prom_panic("Could not allocate memory for RTAS\n");
1568
1569 rtas_inst = call_prom("open", 1, 1, ADDR("/rtas"));
1570 if (!IHANDLE_VALID(rtas_inst)) {
1571 prom_printf("opening rtas package failed (%x)\n", rtas_inst);
1572 return;
1573 }
1574
1575 prom_printf("instantiating rtas at 0x%x...", base);
1576
1577 if (call_prom_ret("call-method", 3, 2, &entry,
1578 ADDR("instantiate-rtas"),
1579 rtas_inst, base) != 0
1580 || entry == 0) {
1581 prom_printf(" failed\n");
1582 return;
1583 }
1584 prom_printf(" done\n");
1585
1586 reserve_mem(base, size);
1587
1588 val = cpu_to_be32(base);
1589 prom_setprop(rtas_node, "/rtas", "linux,rtas-base",
1590 &val, sizeof(val));
1591 val = cpu_to_be32(entry);
1592 prom_setprop(rtas_node, "/rtas", "linux,rtas-entry",
1593 &val, sizeof(val));
1594
1595 /* Check if it supports "query-cpu-stopped-state" */
1596 if (prom_getprop(rtas_node, "query-cpu-stopped-state",
1597 &val, sizeof(val)) != PROM_ERROR)
1598 rtas_has_query_cpu_stopped = true;
1599
1600#if defined(CONFIG_PPC_POWERNV) && defined(__BIG_ENDIAN__)
1601 /* PowerVN takeover hack */
1602 prom_rtas_data = base;
1603 prom_rtas_entry = entry;
1604 prom_getprop(rtas_node, "start-cpu", &prom_rtas_start_cpu, 4);
1605#endif
1606 prom_debug("rtas base = 0x%x\n", base);
1607 prom_debug("rtas entry = 0x%x\n", entry);
1608 prom_debug("rtas size = 0x%x\n", (long)size);
1609
1610 prom_debug("prom_instantiate_rtas: end...\n");
1611}
1612
1613#ifdef CONFIG_PPC64
1614/*
1615 * Allocate room for and instantiate Stored Measurement Log (SML)
1616 */
1617static void __init prom_instantiate_sml(void)
1618{
1619 phandle ibmvtpm_node;
1620 ihandle ibmvtpm_inst;
1621 u32 entry = 0, size = 0;
1622 u64 base;
1623
1624 prom_debug("prom_instantiate_sml: start...\n");
1625
1626 ibmvtpm_node = call_prom("finddevice", 1, 1, ADDR("/ibm,vtpm"));
1627 prom_debug("ibmvtpm_node: %x\n", ibmvtpm_node);
1628 if (!PHANDLE_VALID(ibmvtpm_node))
1629 return;
1630
1631 ibmvtpm_inst = call_prom("open", 1, 1, ADDR("/ibm,vtpm"));
1632 if (!IHANDLE_VALID(ibmvtpm_inst)) {
1633 prom_printf("opening vtpm package failed (%x)\n", ibmvtpm_inst);
1634 return;
1635 }
1636
1637 if (call_prom_ret("call-method", 2, 2, &size,
1638 ADDR("sml-get-handover-size"),
1639 ibmvtpm_inst) != 0 || size == 0) {
1640 prom_printf("SML get handover size failed\n");
1641 return;
1642 }
1643
1644 base = alloc_down(size, PAGE_SIZE, 0);
1645 if (base == 0)
1646 prom_panic("Could not allocate memory for sml\n");
1647
1648 prom_printf("instantiating sml at 0x%x...", base);
1649
1650 if (call_prom_ret("call-method", 4, 2, &entry,
1651 ADDR("sml-handover"),
1652 ibmvtpm_inst, size, base) != 0 || entry == 0) {
1653 prom_printf("SML handover failed\n");
1654 return;
1655 }
1656 prom_printf(" done\n");
1657
1658 reserve_mem(base, size);
1659
1660 prom_setprop(ibmvtpm_node, "/ibm,vtpm", "linux,sml-base",
1661 &base, sizeof(base));
1662 prom_setprop(ibmvtpm_node, "/ibm,vtpm", "linux,sml-size",
1663 &size, sizeof(size));
1664
1665 prom_debug("sml base = 0x%x\n", base);
1666 prom_debug("sml size = 0x%x\n", (long)size);
1667
1668 prom_debug("prom_instantiate_sml: end...\n");
1669}
1670
1671/*
1672 * Allocate room for and initialize TCE tables
1673 */
1674#ifdef __BIG_ENDIAN__
1675static void __init prom_initialize_tce_table(void)
1676{
1677 phandle node;
1678 ihandle phb_node;
1679 char compatible[64], type[64], model[64];
1680 char *path = prom_scratch;
1681 u64 base, align;
1682 u32 minalign, minsize;
1683 u64 tce_entry, *tce_entryp;
1684 u64 local_alloc_top, local_alloc_bottom;
1685 u64 i;
1686
1687 if (prom_iommu_off)
1688 return;
1689
1690 prom_debug("starting prom_initialize_tce_table\n");
1691
1692 /* Cache current top of allocs so we reserve a single block */
1693 local_alloc_top = alloc_top_high;
1694 local_alloc_bottom = local_alloc_top;
1695
1696 /* Search all nodes looking for PHBs. */
1697 for (node = 0; prom_next_node(&node); ) {
1698 compatible[0] = 0;
1699 type[0] = 0;
1700 model[0] = 0;
1701 prom_getprop(node, "compatible",
1702 compatible, sizeof(compatible));
1703 prom_getprop(node, "device_type", type, sizeof(type));
1704 prom_getprop(node, "model", model, sizeof(model));
1705
1706 if ((type[0] == 0) || (strstr(type, "pci") == NULL))
1707 continue;
1708
1709 /* Keep the old logic intact to avoid regression. */
1710 if (compatible[0] != 0) {
1711 if ((strstr(compatible, "python") == NULL) &&
1712 (strstr(compatible, "Speedwagon") == NULL) &&
1713 (strstr(compatible, "Winnipeg") == NULL))
1714 continue;
1715 } else if (model[0] != 0) {
1716 if ((strstr(model, "ython") == NULL) &&
1717 (strstr(model, "peedwagon") == NULL) &&
1718 (strstr(model, "innipeg") == NULL))
1719 continue;
1720 }
1721
1722 if (prom_getprop(node, "tce-table-minalign", &minalign,
1723 sizeof(minalign)) == PROM_ERROR)
1724 minalign = 0;
1725 if (prom_getprop(node, "tce-table-minsize", &minsize,
1726 sizeof(minsize)) == PROM_ERROR)
1727 minsize = 4UL << 20;
1728
1729 /*
1730 * Even though we read what OF wants, we just set the table
1731 * size to 4 MB. This is enough to map 2GB of PCI DMA space.
1732 * By doing this, we avoid the pitfalls of trying to DMA to
1733 * MMIO space and the DMA alias hole.
1734 *
1735 * On POWER4, firmware sets the TCE region by assuming
1736 * each TCE table is 8MB. Using this memory for anything
1737 * else will impact performance, so we always allocate 8MB.
1738 * Anton
1739 */
1740 if (pvr_version_is(PVR_POWER4) || pvr_version_is(PVR_POWER4p))
1741 minsize = 8UL << 20;
1742 else
1743 minsize = 4UL << 20;
1744
1745 /* Align to the greater of the align or size */
1746 align = max(minalign, minsize);
1747 base = alloc_down(minsize, align, 1);
1748 if (base == 0)
1749 prom_panic("ERROR, cannot find space for TCE table.\n");
1750 if (base < local_alloc_bottom)
1751 local_alloc_bottom = base;
1752
1753 /* It seems OF doesn't null-terminate the path :-( */
1754 memset(path, 0, PROM_SCRATCH_SIZE);
1755 /* Call OF to setup the TCE hardware */
1756 if (call_prom("package-to-path", 3, 1, node,
1757 path, PROM_SCRATCH_SIZE-1) == PROM_ERROR) {
1758 prom_printf("package-to-path failed\n");
1759 }
1760
1761 /* Save away the TCE table attributes for later use. */
1762 prom_setprop(node, path, "linux,tce-base", &base, sizeof(base));
1763 prom_setprop(node, path, "linux,tce-size", &minsize, sizeof(minsize));
1764
1765 prom_debug("TCE table: %s\n", path);
1766 prom_debug("\tnode = 0x%x\n", node);
1767 prom_debug("\tbase = 0x%x\n", base);
1768 prom_debug("\tsize = 0x%x\n", minsize);
1769
1770 /* Initialize the table to have a one-to-one mapping
1771 * over the allocated size.
1772 */
1773 tce_entryp = (u64 *)base;
1774 for (i = 0; i < (minsize >> 3) ;tce_entryp++, i++) {
1775 tce_entry = (i << PAGE_SHIFT);
1776 tce_entry |= 0x3;
1777 *tce_entryp = tce_entry;
1778 }
1779
1780 prom_printf("opening PHB %s", path);
1781 phb_node = call_prom("open", 1, 1, path);
1782 if (phb_node == 0)
1783 prom_printf("... failed\n");
1784 else
1785 prom_printf("... done\n");
1786
1787 call_prom("call-method", 6, 0, ADDR("set-64-bit-addressing"),
1788 phb_node, -1, minsize,
1789 (u32) base, (u32) (base >> 32));
1790 call_prom("close", 1, 0, phb_node);
1791 }
1792
1793 reserve_mem(local_alloc_bottom, local_alloc_top - local_alloc_bottom);
1794
1795 /* These are only really needed if there is a memory limit in
1796 * effect, but we don't know so export them always. */
1797 prom_tce_alloc_start = local_alloc_bottom;
1798 prom_tce_alloc_end = local_alloc_top;
1799
1800 /* Flag the first invalid entry */
1801 prom_debug("ending prom_initialize_tce_table\n");
1802}
1803#endif /* __BIG_ENDIAN__ */
1804#endif /* CONFIG_PPC64 */
1805
1806/*
1807 * With CHRP SMP we need to use the OF to start the other processors.
1808 * We can't wait until smp_boot_cpus (the OF is trashed by then)
1809 * so we have to put the processors into a holding pattern controlled
1810 * by the kernel (not OF) before we destroy the OF.
1811 *
1812 * This uses a chunk of low memory, puts some holding pattern
1813 * code there and sends the other processors off to there until
1814 * smp_boot_cpus tells them to do something. The holding pattern
1815 * checks that address until its cpu # is there, when it is that
1816 * cpu jumps to __secondary_start(). smp_boot_cpus() takes care
1817 * of setting those values.
1818 *
1819 * We also use physical address 0x4 here to tell when a cpu
1820 * is in its holding pattern code.
1821 *
1822 * -- Cort
1823 */
1824/*
1825 * We want to reference the copy of __secondary_hold_* in the
1826 * 0 - 0x100 address range
1827 */
1828#define LOW_ADDR(x) (((unsigned long) &(x)) & 0xff)
1829
1830static void __init prom_hold_cpus(void)
1831{
1832 unsigned long i;
1833 phandle node;
1834 char type[64];
1835 unsigned long *spinloop
1836 = (void *) LOW_ADDR(__secondary_hold_spinloop);
1837 unsigned long *acknowledge
1838 = (void *) LOW_ADDR(__secondary_hold_acknowledge);
1839 unsigned long secondary_hold = LOW_ADDR(__secondary_hold);
1840
1841 /*
1842 * On pseries, if RTAS supports "query-cpu-stopped-state",
1843 * we skip this stage, the CPUs will be started by the
1844 * kernel using RTAS.
1845 */
1846 if ((of_platform == PLATFORM_PSERIES ||
1847 of_platform == PLATFORM_PSERIES_LPAR) &&
1848 rtas_has_query_cpu_stopped) {
1849 prom_printf("prom_hold_cpus: skipped\n");
1850 return;
1851 }
1852
1853 prom_debug("prom_hold_cpus: start...\n");
1854 prom_debug(" 1) spinloop = 0x%x\n", (unsigned long)spinloop);
1855 prom_debug(" 1) *spinloop = 0x%x\n", *spinloop);
1856 prom_debug(" 1) acknowledge = 0x%x\n",
1857 (unsigned long)acknowledge);
1858 prom_debug(" 1) *acknowledge = 0x%x\n", *acknowledge);
1859 prom_debug(" 1) secondary_hold = 0x%x\n", secondary_hold);
1860
1861 /* Set the common spinloop variable, so all of the secondary cpus
1862 * will block when they are awakened from their OF spinloop.
1863 * This must occur for both SMP and non SMP kernels, since OF will
1864 * be trashed when we move the kernel.
1865 */
1866 *spinloop = 0;
1867
1868 /* look for cpus */
1869 for (node = 0; prom_next_node(&node); ) {
1870 unsigned int cpu_no;
1871 __be32 reg;
1872
1873 type[0] = 0;
1874 prom_getprop(node, "device_type", type, sizeof(type));
1875 if (strcmp(type, "cpu") != 0)
1876 continue;
1877
1878 /* Skip non-configured cpus. */
1879 if (prom_getprop(node, "status", type, sizeof(type)) > 0)
1880 if (strcmp(type, "okay") != 0)
1881 continue;
1882
1883 reg = cpu_to_be32(-1); /* make sparse happy */
1884 prom_getprop(node, "reg", ®, sizeof(reg));
1885 cpu_no = be32_to_cpu(reg);
1886
1887 prom_debug("cpu hw idx = %lu\n", cpu_no);
1888
1889 /* Init the acknowledge var which will be reset by
1890 * the secondary cpu when it awakens from its OF
1891 * spinloop.
1892 */
1893 *acknowledge = (unsigned long)-1;
1894
1895 if (cpu_no != prom.cpu) {
1896 /* Primary Thread of non-boot cpu or any thread */
1897 prom_printf("starting cpu hw idx %lu... ", cpu_no);
1898 call_prom("start-cpu", 3, 0, node,
1899 secondary_hold, cpu_no);
1900
1901 for (i = 0; (i < 100000000) &&
1902 (*acknowledge == ((unsigned long)-1)); i++ )
1903 mb();
1904
1905 if (*acknowledge == cpu_no)
1906 prom_printf("done\n");
1907 else
1908 prom_printf("failed: %x\n", *acknowledge);
1909 }
1910#ifdef CONFIG_SMP
1911 else
1912 prom_printf("boot cpu hw idx %lu\n", cpu_no);
1913#endif /* CONFIG_SMP */
1914 }
1915
1916 prom_debug("prom_hold_cpus: end...\n");
1917}
1918
1919
1920static void __init prom_init_client_services(unsigned long pp)
1921{
1922 /* Get a handle to the prom entry point before anything else */
1923 prom_entry = pp;
1924
1925 /* get a handle for the stdout device */
1926 prom.chosen = call_prom("finddevice", 1, 1, ADDR("/chosen"));
1927 if (!PHANDLE_VALID(prom.chosen))
1928 prom_panic("cannot find chosen"); /* msg won't be printed :( */
1929
1930 /* get device tree root */
1931 prom.root = call_prom("finddevice", 1, 1, ADDR("/"));
1932 if (!PHANDLE_VALID(prom.root))
1933 prom_panic("cannot find device tree root"); /* msg won't be printed :( */
1934
1935 prom.mmumap = 0;
1936}
1937
1938#ifdef CONFIG_PPC32
1939/*
1940 * For really old powermacs, we need to map things we claim.
1941 * For that, we need the ihandle of the mmu.
1942 * Also, on the longtrail, we need to work around other bugs.
1943 */
1944static void __init prom_find_mmu(void)
1945{
1946 phandle oprom;
1947 char version[64];
1948
1949 oprom = call_prom("finddevice", 1, 1, ADDR("/openprom"));
1950 if (!PHANDLE_VALID(oprom))
1951 return;
1952 if (prom_getprop(oprom, "model", version, sizeof(version)) <= 0)
1953 return;
1954 version[sizeof(version) - 1] = 0;
1955 /* XXX might need to add other versions here */
1956 if (strcmp(version, "Open Firmware, 1.0.5") == 0)
1957 of_workarounds = OF_WA_CLAIM;
1958 else if (strncmp(version, "FirmWorks,3.", 12) == 0) {
1959 of_workarounds = OF_WA_CLAIM | OF_WA_LONGTRAIL;
1960 call_prom("interpret", 1, 1, "dev /memory 0 to allow-reclaim");
1961 } else
1962 return;
1963 prom.memory = call_prom("open", 1, 1, ADDR("/memory"));
1964 prom_getprop(prom.chosen, "mmu", &prom.mmumap,
1965 sizeof(prom.mmumap));
1966 prom.mmumap = be32_to_cpu(prom.mmumap);
1967 if (!IHANDLE_VALID(prom.memory) || !IHANDLE_VALID(prom.mmumap))
1968 of_workarounds &= ~OF_WA_CLAIM; /* hmmm */
1969}
1970#else
1971#define prom_find_mmu()
1972#endif
1973
1974static void __init prom_init_stdout(void)
1975{
1976 char *path = of_stdout_device;
1977 char type[16];
1978 phandle stdout_node;
1979 __be32 val;
1980
1981 if (prom_getprop(prom.chosen, "stdout", &val, sizeof(val)) <= 0)
1982 prom_panic("cannot find stdout");
1983
1984 prom.stdout = be32_to_cpu(val);
1985
1986 /* Get the full OF pathname of the stdout device */
1987 memset(path, 0, 256);
1988 call_prom("instance-to-path", 3, 1, prom.stdout, path, 255);
1989 prom_printf("OF stdout device is: %s\n", of_stdout_device);
1990 prom_setprop(prom.chosen, "/chosen", "linux,stdout-path",
1991 path, strlen(path) + 1);
1992
1993 /* instance-to-package fails on PA-Semi */
1994 stdout_node = call_prom("instance-to-package", 1, 1, prom.stdout);
1995 if (stdout_node != PROM_ERROR) {
1996 val = cpu_to_be32(stdout_node);
1997 prom_setprop(prom.chosen, "/chosen", "linux,stdout-package",
1998 &val, sizeof(val));
1999
2000 /* If it's a display, note it */
2001 memset(type, 0, sizeof(type));
2002 prom_getprop(stdout_node, "device_type", type, sizeof(type));
2003 if (strcmp(type, "display") == 0)
2004 prom_setprop(stdout_node, path, "linux,boot-display", NULL, 0);
2005 }
2006}
2007
2008static int __init prom_find_machine_type(void)
2009{
2010 char compat[256];
2011 int len, i = 0;
2012#ifdef CONFIG_PPC64
2013 phandle rtas;
2014 int x;
2015#endif
2016
2017 /* Look for a PowerMac or a Cell */
2018 len = prom_getprop(prom.root, "compatible",
2019 compat, sizeof(compat)-1);
2020 if (len > 0) {
2021 compat[len] = 0;
2022 while (i < len) {
2023 char *p = &compat[i];
2024 int sl = strlen(p);
2025 if (sl == 0)
2026 break;
2027 if (strstr(p, "Power Macintosh") ||
2028 strstr(p, "MacRISC"))
2029 return PLATFORM_POWERMAC;
2030#ifdef CONFIG_PPC64
2031 /* We must make sure we don't detect the IBM Cell
2032 * blades as pSeries due to some firmware issues,
2033 * so we do it here.
2034 */
2035 if (strstr(p, "IBM,CBEA") ||
2036 strstr(p, "IBM,CPBW-1.0"))
2037 return PLATFORM_GENERIC;
2038#endif /* CONFIG_PPC64 */
2039 i += sl + 1;
2040 }
2041 }
2042#ifdef CONFIG_PPC64
2043 /* Try to detect OPAL */
2044 if (PHANDLE_VALID(call_prom("finddevice", 1, 1, ADDR("/ibm,opal"))))
2045 return PLATFORM_OPAL;
2046
2047 /* Try to figure out if it's an IBM pSeries or any other
2048 * PAPR compliant platform. We assume it is if :
2049 * - /device_type is "chrp" (please, do NOT use that for future
2050 * non-IBM designs !
2051 * - it has /rtas
2052 */
2053 len = prom_getprop(prom.root, "device_type",
2054 compat, sizeof(compat)-1);
2055 if (len <= 0)
2056 return PLATFORM_GENERIC;
2057 if (strcmp(compat, "chrp"))
2058 return PLATFORM_GENERIC;
2059
2060 /* Default to pSeries. We need to know if we are running LPAR */
2061 rtas = call_prom("finddevice", 1, 1, ADDR("/rtas"));
2062 if (!PHANDLE_VALID(rtas))
2063 return PLATFORM_GENERIC;
2064 x = prom_getproplen(rtas, "ibm,hypertas-functions");
2065 if (x != PROM_ERROR) {
2066 prom_debug("Hypertas detected, assuming LPAR !\n");
2067 return PLATFORM_PSERIES_LPAR;
2068 }
2069 return PLATFORM_PSERIES;
2070#else
2071 return PLATFORM_GENERIC;
2072#endif
2073}
2074
2075static int __init prom_set_color(ihandle ih, int i, int r, int g, int b)
2076{
2077 return call_prom("call-method", 6, 1, ADDR("color!"), ih, i, b, g, r);
2078}
2079
2080/*
2081 * If we have a display that we don't know how to drive,
2082 * we will want to try to execute OF's open method for it
2083 * later. However, OF will probably fall over if we do that
2084 * we've taken over the MMU.
2085 * So we check whether we will need to open the display,
2086 * and if so, open it now.
2087 */
2088static void __init prom_check_displays(void)
2089{
2090 char type[16], *path;
2091 phandle node;
2092 ihandle ih;
2093 int i;
2094
2095 static unsigned char default_colors[] = {
2096 0x00, 0x00, 0x00,
2097 0x00, 0x00, 0xaa,
2098 0x00, 0xaa, 0x00,
2099 0x00, 0xaa, 0xaa,
2100 0xaa, 0x00, 0x00,
2101 0xaa, 0x00, 0xaa,
2102 0xaa, 0xaa, 0x00,
2103 0xaa, 0xaa, 0xaa,
2104 0x55, 0x55, 0x55,
2105 0x55, 0x55, 0xff,
2106 0x55, 0xff, 0x55,
2107 0x55, 0xff, 0xff,
2108 0xff, 0x55, 0x55,
2109 0xff, 0x55, 0xff,
2110 0xff, 0xff, 0x55,
2111 0xff, 0xff, 0xff
2112 };
2113 const unsigned char *clut;
2114
2115 prom_debug("Looking for displays\n");
2116 for (node = 0; prom_next_node(&node); ) {
2117 memset(type, 0, sizeof(type));
2118 prom_getprop(node, "device_type", type, sizeof(type));
2119 if (strcmp(type, "display") != 0)
2120 continue;
2121
2122 /* It seems OF doesn't null-terminate the path :-( */
2123 path = prom_scratch;
2124 memset(path, 0, PROM_SCRATCH_SIZE);
2125
2126 /*
2127 * leave some room at the end of the path for appending extra
2128 * arguments
2129 */
2130 if (call_prom("package-to-path", 3, 1, node, path,
2131 PROM_SCRATCH_SIZE-10) == PROM_ERROR)
2132 continue;
2133 prom_printf("found display : %s, opening... ", path);
2134
2135 ih = call_prom("open", 1, 1, path);
2136 if (ih == 0) {
2137 prom_printf("failed\n");
2138 continue;
2139 }
2140
2141 /* Success */
2142 prom_printf("done\n");
2143 prom_setprop(node, path, "linux,opened", NULL, 0);
2144
2145 /* Setup a usable color table when the appropriate
2146 * method is available. Should update this to set-colors */
2147 clut = default_colors;
2148 for (i = 0; i < 16; i++, clut += 3)
2149 if (prom_set_color(ih, i, clut[0], clut[1],
2150 clut[2]) != 0)
2151 break;
2152
2153#ifdef CONFIG_LOGO_LINUX_CLUT224
2154 clut = PTRRELOC(logo_linux_clut224.clut);
2155 for (i = 0; i < logo_linux_clut224.clutsize; i++, clut += 3)
2156 if (prom_set_color(ih, i + 32, clut[0], clut[1],
2157 clut[2]) != 0)
2158 break;
2159#endif /* CONFIG_LOGO_LINUX_CLUT224 */
2160
2161#ifdef CONFIG_PPC_EARLY_DEBUG_BOOTX
2162 if (prom_getprop(node, "linux,boot-display", NULL, 0) !=
2163 PROM_ERROR) {
2164 u32 width, height, pitch, addr;
2165
2166 prom_printf("Setting btext !\n");
2167 prom_getprop(node, "width", &width, 4);
2168 prom_getprop(node, "height", &height, 4);
2169 prom_getprop(node, "linebytes", &pitch, 4);
2170 prom_getprop(node, "address", &addr, 4);
2171 prom_printf("W=%d H=%d LB=%d addr=0x%x\n",
2172 width, height, pitch, addr);
2173 btext_setup_display(width, height, 8, pitch, addr);
2174 }
2175#endif /* CONFIG_PPC_EARLY_DEBUG_BOOTX */
2176 }
2177}
2178
2179
2180/* Return (relocated) pointer to this much memory: moves initrd if reqd. */
2181static void __init *make_room(unsigned long *mem_start, unsigned long *mem_end,
2182 unsigned long needed, unsigned long align)
2183{
2184 void *ret;
2185
2186 *mem_start = _ALIGN(*mem_start, align);
2187 while ((*mem_start + needed) > *mem_end) {
2188 unsigned long room, chunk;
2189
2190 prom_debug("Chunk exhausted, claiming more at %x...\n",
2191 alloc_bottom);
2192 room = alloc_top - alloc_bottom;
2193 if (room > DEVTREE_CHUNK_SIZE)
2194 room = DEVTREE_CHUNK_SIZE;
2195 if (room < PAGE_SIZE)
2196 prom_panic("No memory for flatten_device_tree "
2197 "(no room)\n");
2198 chunk = alloc_up(room, 0);
2199 if (chunk == 0)
2200 prom_panic("No memory for flatten_device_tree "
2201 "(claim failed)\n");
2202 *mem_end = chunk + room;
2203 }
2204
2205 ret = (void *)*mem_start;
2206 *mem_start += needed;
2207
2208 return ret;
2209}
2210
2211#define dt_push_token(token, mem_start, mem_end) do { \
2212 void *room = make_room(mem_start, mem_end, 4, 4); \
2213 *(__be32 *)room = cpu_to_be32(token); \
2214 } while(0)
2215
2216static unsigned long __init dt_find_string(char *str)
2217{
2218 char *s, *os;
2219
2220 s = os = (char *)dt_string_start;
2221 s += 4;
2222 while (s < (char *)dt_string_end) {
2223 if (strcmp(s, str) == 0)
2224 return s - os;
2225 s += strlen(s) + 1;
2226 }
2227 return 0;
2228}
2229
2230/*
2231 * The Open Firmware 1275 specification states properties must be 31 bytes or
2232 * less, however not all firmwares obey this. Make it 64 bytes to be safe.
2233 */
2234#define MAX_PROPERTY_NAME 64
2235
2236static void __init scan_dt_build_strings(phandle node,
2237 unsigned long *mem_start,
2238 unsigned long *mem_end)
2239{
2240 char *prev_name, *namep, *sstart;
2241 unsigned long soff;
2242 phandle child;
2243
2244 sstart = (char *)dt_string_start;
2245
2246 /* get and store all property names */
2247 prev_name = "";
2248 for (;;) {
2249 /* 64 is max len of name including nul. */
2250 namep = make_room(mem_start, mem_end, MAX_PROPERTY_NAME, 1);
2251 if (call_prom("nextprop", 3, 1, node, prev_name, namep) != 1) {
2252 /* No more nodes: unwind alloc */
2253 *mem_start = (unsigned long)namep;
2254 break;
2255 }
2256
2257 /* skip "name" */
2258 if (strcmp(namep, "name") == 0) {
2259 *mem_start = (unsigned long)namep;
2260 prev_name = "name";
2261 continue;
2262 }
2263 /* get/create string entry */
2264 soff = dt_find_string(namep);
2265 if (soff != 0) {
2266 *mem_start = (unsigned long)namep;
2267 namep = sstart + soff;
2268 } else {
2269 /* Trim off some if we can */
2270 *mem_start = (unsigned long)namep + strlen(namep) + 1;
2271 dt_string_end = *mem_start;
2272 }
2273 prev_name = namep;
2274 }
2275
2276 /* do all our children */
2277 child = call_prom("child", 1, 1, node);
2278 while (child != 0) {
2279 scan_dt_build_strings(child, mem_start, mem_end);
2280 child = call_prom("peer", 1, 1, child);
2281 }
2282}
2283
2284static void __init scan_dt_build_struct(phandle node, unsigned long *mem_start,
2285 unsigned long *mem_end)
2286{
2287 phandle child;
2288 char *namep, *prev_name, *sstart, *p, *ep, *lp, *path;
2289 unsigned long soff;
2290 unsigned char *valp;
2291 static char pname[MAX_PROPERTY_NAME];
2292 int l, room, has_phandle = 0;
2293
2294 dt_push_token(OF_DT_BEGIN_NODE, mem_start, mem_end);
2295
2296 /* get the node's full name */
2297 namep = (char *)*mem_start;
2298 room = *mem_end - *mem_start;
2299 if (room > 255)
2300 room = 255;
2301 l = call_prom("package-to-path", 3, 1, node, namep, room);
2302 if (l >= 0) {
2303 /* Didn't fit? Get more room. */
2304 if (l >= room) {
2305 if (l >= *mem_end - *mem_start)
2306 namep = make_room(mem_start, mem_end, l+1, 1);
2307 call_prom("package-to-path", 3, 1, node, namep, l);
2308 }
2309 namep[l] = '\0';
2310
2311 /* Fixup an Apple bug where they have bogus \0 chars in the
2312 * middle of the path in some properties, and extract
2313 * the unit name (everything after the last '/').
2314 */
2315 for (lp = p = namep, ep = namep + l; p < ep; p++) {
2316 if (*p == '/')
2317 lp = namep;
2318 else if (*p != 0)
2319 *lp++ = *p;
2320 }
2321 *lp = 0;
2322 *mem_start = _ALIGN((unsigned long)lp + 1, 4);
2323 }
2324
2325 /* get it again for debugging */
2326 path = prom_scratch;
2327 memset(path, 0, PROM_SCRATCH_SIZE);
2328 call_prom("package-to-path", 3, 1, node, path, PROM_SCRATCH_SIZE-1);
2329
2330 /* get and store all properties */
2331 prev_name = "";
2332 sstart = (char *)dt_string_start;
2333 for (;;) {
2334 if (call_prom("nextprop", 3, 1, node, prev_name,
2335 pname) != 1)
2336 break;
2337
2338 /* skip "name" */
2339 if (strcmp(pname, "name") == 0) {
2340 prev_name = "name";
2341 continue;
2342 }
2343
2344 /* find string offset */
2345 soff = dt_find_string(pname);
2346 if (soff == 0) {
2347 prom_printf("WARNING: Can't find string index for"
2348 " <%s>, node %s\n", pname, path);
2349 break;
2350 }
2351 prev_name = sstart + soff;
2352
2353 /* get length */
2354 l = call_prom("getproplen", 2, 1, node, pname);
2355
2356 /* sanity checks */
2357 if (l == PROM_ERROR)
2358 continue;
2359
2360 /* push property head */
2361 dt_push_token(OF_DT_PROP, mem_start, mem_end);
2362 dt_push_token(l, mem_start, mem_end);
2363 dt_push_token(soff, mem_start, mem_end);
2364
2365 /* push property content */
2366 valp = make_room(mem_start, mem_end, l, 4);
2367 call_prom("getprop", 4, 1, node, pname, valp, l);
2368 *mem_start = _ALIGN(*mem_start, 4);
2369
2370 if (!strcmp(pname, "phandle"))
2371 has_phandle = 1;
2372 }
2373
2374 /* Add a "linux,phandle" property if no "phandle" property already
2375 * existed (can happen with OPAL)
2376 */
2377 if (!has_phandle) {
2378 soff = dt_find_string("linux,phandle");
2379 if (soff == 0)
2380 prom_printf("WARNING: Can't find string index for"
2381 " <linux-phandle> node %s\n", path);
2382 else {
2383 dt_push_token(OF_DT_PROP, mem_start, mem_end);
2384 dt_push_token(4, mem_start, mem_end);
2385 dt_push_token(soff, mem_start, mem_end);
2386 valp = make_room(mem_start, mem_end, 4, 4);
2387 *(__be32 *)valp = cpu_to_be32(node);
2388 }
2389 }
2390
2391 /* do all our children */
2392 child = call_prom("child", 1, 1, node);
2393 while (child != 0) {
2394 scan_dt_build_struct(child, mem_start, mem_end);
2395 child = call_prom("peer", 1, 1, child);
2396 }
2397
2398 dt_push_token(OF_DT_END_NODE, mem_start, mem_end);
2399}
2400
2401static void __init flatten_device_tree(void)
2402{
2403 phandle root;
2404 unsigned long mem_start, mem_end, room;
2405 struct boot_param_header *hdr;
2406 char *namep;
2407 u64 *rsvmap;
2408
2409 /*
2410 * Check how much room we have between alloc top & bottom (+/- a
2411 * few pages), crop to 1MB, as this is our "chunk" size
2412 */
2413 room = alloc_top - alloc_bottom - 0x4000;
2414 if (room > DEVTREE_CHUNK_SIZE)
2415 room = DEVTREE_CHUNK_SIZE;
2416 prom_debug("starting device tree allocs at %x\n", alloc_bottom);
2417
2418 /* Now try to claim that */
2419 mem_start = (unsigned long)alloc_up(room, PAGE_SIZE);
2420 if (mem_start == 0)
2421 prom_panic("Can't allocate initial device-tree chunk\n");
2422 mem_end = mem_start + room;
2423
2424 /* Get root of tree */
2425 root = call_prom("peer", 1, 1, (phandle)0);
2426 if (root == (phandle)0)
2427 prom_panic ("couldn't get device tree root\n");
2428
2429 /* Build header and make room for mem rsv map */
2430 mem_start = _ALIGN(mem_start, 4);
2431 hdr = make_room(&mem_start, &mem_end,
2432 sizeof(struct boot_param_header), 4);
2433 dt_header_start = (unsigned long)hdr;
2434 rsvmap = make_room(&mem_start, &mem_end, sizeof(mem_reserve_map), 8);
2435
2436 /* Start of strings */
2437 mem_start = PAGE_ALIGN(mem_start);
2438 dt_string_start = mem_start;
2439 mem_start += 4; /* hole */
2440
2441 /* Add "linux,phandle" in there, we'll need it */
2442 namep = make_room(&mem_start, &mem_end, 16, 1);
2443 strcpy(namep, "linux,phandle");
2444 mem_start = (unsigned long)namep + strlen(namep) + 1;
2445
2446 /* Build string array */
2447 prom_printf("Building dt strings...\n");
2448 scan_dt_build_strings(root, &mem_start, &mem_end);
2449 dt_string_end = mem_start;
2450
2451 /* Build structure */
2452 mem_start = PAGE_ALIGN(mem_start);
2453 dt_struct_start = mem_start;
2454 prom_printf("Building dt structure...\n");
2455 scan_dt_build_struct(root, &mem_start, &mem_end);
2456 dt_push_token(OF_DT_END, &mem_start, &mem_end);
2457 dt_struct_end = PAGE_ALIGN(mem_start);
2458
2459 /* Finish header */
2460 hdr->boot_cpuid_phys = cpu_to_be32(prom.cpu);
2461 hdr->magic = cpu_to_be32(OF_DT_HEADER);
2462 hdr->totalsize = cpu_to_be32(dt_struct_end - dt_header_start);
2463 hdr->off_dt_struct = cpu_to_be32(dt_struct_start - dt_header_start);
2464 hdr->off_dt_strings = cpu_to_be32(dt_string_start - dt_header_start);
2465 hdr->dt_strings_size = cpu_to_be32(dt_string_end - dt_string_start);
2466 hdr->off_mem_rsvmap = cpu_to_be32(((unsigned long)rsvmap) - dt_header_start);
2467 hdr->version = cpu_to_be32(OF_DT_VERSION);
2468 /* Version 16 is not backward compatible */
2469 hdr->last_comp_version = cpu_to_be32(0x10);
2470
2471 /* Copy the reserve map in */
2472 memcpy(rsvmap, mem_reserve_map, sizeof(mem_reserve_map));
2473
2474#ifdef DEBUG_PROM
2475 {
2476 int i;
2477 prom_printf("reserved memory map:\n");
2478 for (i = 0; i < mem_reserve_cnt; i++)
2479 prom_printf(" %x - %x\n",
2480 be64_to_cpu(mem_reserve_map[i].base),
2481 be64_to_cpu(mem_reserve_map[i].size));
2482 }
2483#endif
2484 /* Bump mem_reserve_cnt to cause further reservations to fail
2485 * since it's too late.
2486 */
2487 mem_reserve_cnt = MEM_RESERVE_MAP_SIZE;
2488
2489 prom_printf("Device tree strings 0x%x -> 0x%x\n",
2490 dt_string_start, dt_string_end);
2491 prom_printf("Device tree struct 0x%x -> 0x%x\n",
2492 dt_struct_start, dt_struct_end);
2493}
2494
2495#ifdef CONFIG_PPC_MAPLE
2496/* PIBS Version 1.05.0000 04/26/2005 has an incorrect /ht/isa/ranges property.
2497 * The values are bad, and it doesn't even have the right number of cells. */
2498static void __init fixup_device_tree_maple(void)
2499{
2500 phandle isa;
2501 u32 rloc = 0x01002000; /* IO space; PCI device = 4 */
2502 u32 isa_ranges[6];
2503 char *name;
2504
2505 name = "/ht@0/isa@4";
2506 isa = call_prom("finddevice", 1, 1, ADDR(name));
2507 if (!PHANDLE_VALID(isa)) {
2508 name = "/ht@0/isa@6";
2509 isa = call_prom("finddevice", 1, 1, ADDR(name));
2510 rloc = 0x01003000; /* IO space; PCI device = 6 */
2511 }
2512 if (!PHANDLE_VALID(isa))
2513 return;
2514
2515 if (prom_getproplen(isa, "ranges") != 12)
2516 return;
2517 if (prom_getprop(isa, "ranges", isa_ranges, sizeof(isa_ranges))
2518 == PROM_ERROR)
2519 return;
2520
2521 if (isa_ranges[0] != 0x1 ||
2522 isa_ranges[1] != 0xf4000000 ||
2523 isa_ranges[2] != 0x00010000)
2524 return;
2525
2526 prom_printf("Fixing up bogus ISA range on Maple/Apache...\n");
2527
2528 isa_ranges[0] = 0x1;
2529 isa_ranges[1] = 0x0;
2530 isa_ranges[2] = rloc;
2531 isa_ranges[3] = 0x0;
2532 isa_ranges[4] = 0x0;
2533 isa_ranges[5] = 0x00010000;
2534 prom_setprop(isa, name, "ranges",
2535 isa_ranges, sizeof(isa_ranges));
2536}
2537
2538#define CPC925_MC_START 0xf8000000
2539#define CPC925_MC_LENGTH 0x1000000
2540/* The values for memory-controller don't have right number of cells */
2541static void __init fixup_device_tree_maple_memory_controller(void)
2542{
2543 phandle mc;
2544 u32 mc_reg[4];
2545 char *name = "/hostbridge@f8000000";
2546 u32 ac, sc;
2547
2548 mc = call_prom("finddevice", 1, 1, ADDR(name));
2549 if (!PHANDLE_VALID(mc))
2550 return;
2551
2552 if (prom_getproplen(mc, "reg") != 8)
2553 return;
2554
2555 prom_getprop(prom.root, "#address-cells", &ac, sizeof(ac));
2556 prom_getprop(prom.root, "#size-cells", &sc, sizeof(sc));
2557 if ((ac != 2) || (sc != 2))
2558 return;
2559
2560 if (prom_getprop(mc, "reg", mc_reg, sizeof(mc_reg)) == PROM_ERROR)
2561 return;
2562
2563 if (mc_reg[0] != CPC925_MC_START || mc_reg[1] != CPC925_MC_LENGTH)
2564 return;
2565
2566 prom_printf("Fixing up bogus hostbridge on Maple...\n");
2567
2568 mc_reg[0] = 0x0;
2569 mc_reg[1] = CPC925_MC_START;
2570 mc_reg[2] = 0x0;
2571 mc_reg[3] = CPC925_MC_LENGTH;
2572 prom_setprop(mc, name, "reg", mc_reg, sizeof(mc_reg));
2573}
2574#else
2575#define fixup_device_tree_maple()
2576#define fixup_device_tree_maple_memory_controller()
2577#endif
2578
2579#ifdef CONFIG_PPC_CHRP
2580/*
2581 * Pegasos and BriQ lacks the "ranges" property in the isa node
2582 * Pegasos needs decimal IRQ 14/15, not hexadecimal
2583 * Pegasos has the IDE configured in legacy mode, but advertised as native
2584 */
2585static void __init fixup_device_tree_chrp(void)
2586{
2587 phandle ph;
2588 u32 prop[6];
2589 u32 rloc = 0x01006000; /* IO space; PCI device = 12 */
2590 char *name;
2591 int rc;
2592
2593 name = "/pci@80000000/isa@c";
2594 ph = call_prom("finddevice", 1, 1, ADDR(name));
2595 if (!PHANDLE_VALID(ph)) {
2596 name = "/pci@ff500000/isa@6";
2597 ph = call_prom("finddevice", 1, 1, ADDR(name));
2598 rloc = 0x01003000; /* IO space; PCI device = 6 */
2599 }
2600 if (PHANDLE_VALID(ph)) {
2601 rc = prom_getproplen(ph, "ranges");
2602 if (rc == 0 || rc == PROM_ERROR) {
2603 prom_printf("Fixing up missing ISA range on Pegasos...\n");
2604
2605 prop[0] = 0x1;
2606 prop[1] = 0x0;
2607 prop[2] = rloc;
2608 prop[3] = 0x0;
2609 prop[4] = 0x0;
2610 prop[5] = 0x00010000;
2611 prom_setprop(ph, name, "ranges", prop, sizeof(prop));
2612 }
2613 }
2614
2615 name = "/pci@80000000/ide@C,1";
2616 ph = call_prom("finddevice", 1, 1, ADDR(name));
2617 if (PHANDLE_VALID(ph)) {
2618 prom_printf("Fixing up IDE interrupt on Pegasos...\n");
2619 prop[0] = 14;
2620 prop[1] = 0x0;
2621 prom_setprop(ph, name, "interrupts", prop, 2*sizeof(u32));
2622 prom_printf("Fixing up IDE class-code on Pegasos...\n");
2623 rc = prom_getprop(ph, "class-code", prop, sizeof(u32));
2624 if (rc == sizeof(u32)) {
2625 prop[0] &= ~0x5;
2626 prom_setprop(ph, name, "class-code", prop, sizeof(u32));
2627 }
2628 }
2629}
2630#else
2631#define fixup_device_tree_chrp()
2632#endif
2633
2634#if defined(CONFIG_PPC64) && defined(CONFIG_PPC_PMAC)
2635static void __init fixup_device_tree_pmac(void)
2636{
2637 phandle u3, i2c, mpic;
2638 u32 u3_rev;
2639 u32 interrupts[2];
2640 u32 parent;
2641
2642 /* Some G5s have a missing interrupt definition, fix it up here */
2643 u3 = call_prom("finddevice", 1, 1, ADDR("/u3@0,f8000000"));
2644 if (!PHANDLE_VALID(u3))
2645 return;
2646 i2c = call_prom("finddevice", 1, 1, ADDR("/u3@0,f8000000/i2c@f8001000"));
2647 if (!PHANDLE_VALID(i2c))
2648 return;
2649 mpic = call_prom("finddevice", 1, 1, ADDR("/u3@0,f8000000/mpic@f8040000"));
2650 if (!PHANDLE_VALID(mpic))
2651 return;
2652
2653 /* check if proper rev of u3 */
2654 if (prom_getprop(u3, "device-rev", &u3_rev, sizeof(u3_rev))
2655 == PROM_ERROR)
2656 return;
2657 if (u3_rev < 0x35 || u3_rev > 0x39)
2658 return;
2659 /* does it need fixup ? */
2660 if (prom_getproplen(i2c, "interrupts") > 0)
2661 return;
2662
2663 prom_printf("fixing up bogus interrupts for u3 i2c...\n");
2664
2665 /* interrupt on this revision of u3 is number 0 and level */
2666 interrupts[0] = 0;
2667 interrupts[1] = 1;
2668 prom_setprop(i2c, "/u3@0,f8000000/i2c@f8001000", "interrupts",
2669 &interrupts, sizeof(interrupts));
2670 parent = (u32)mpic;
2671 prom_setprop(i2c, "/u3@0,f8000000/i2c@f8001000", "interrupt-parent",
2672 &parent, sizeof(parent));
2673}
2674#else
2675#define fixup_device_tree_pmac()
2676#endif
2677
2678#ifdef CONFIG_PPC_EFIKA
2679/*
2680 * The MPC5200 FEC driver requires an phy-handle property to tell it how
2681 * to talk to the phy. If the phy-handle property is missing, then this
2682 * function is called to add the appropriate nodes and link it to the
2683 * ethernet node.
2684 */
2685static void __init fixup_device_tree_efika_add_phy(void)
2686{
2687 u32 node;
2688 char prop[64];
2689 int rv;
2690
2691 /* Check if /builtin/ethernet exists - bail if it doesn't */
2692 node = call_prom("finddevice", 1, 1, ADDR("/builtin/ethernet"));
2693 if (!PHANDLE_VALID(node))
2694 return;
2695
2696 /* Check if the phy-handle property exists - bail if it does */
2697 rv = prom_getprop(node, "phy-handle", prop, sizeof(prop));
2698 if (!rv)
2699 return;
2700
2701 /*
2702 * At this point the ethernet device doesn't have a phy described.
2703 * Now we need to add the missing phy node and linkage
2704 */
2705
2706 /* Check for an MDIO bus node - if missing then create one */
2707 node = call_prom("finddevice", 1, 1, ADDR("/builtin/mdio"));
2708 if (!PHANDLE_VALID(node)) {
2709 prom_printf("Adding Ethernet MDIO node\n");
2710 call_prom("interpret", 1, 1,
2711 " s\" /builtin\" find-device"
2712 " new-device"
2713 " 1 encode-int s\" #address-cells\" property"
2714 " 0 encode-int s\" #size-cells\" property"
2715 " s\" mdio\" device-name"
2716 " s\" fsl,mpc5200b-mdio\" encode-string"
2717 " s\" compatible\" property"
2718 " 0xf0003000 0x400 reg"
2719 " 0x2 encode-int"
2720 " 0x5 encode-int encode+"
2721 " 0x3 encode-int encode+"
2722 " s\" interrupts\" property"
2723 " finish-device");
2724 };
2725
2726 /* Check for a PHY device node - if missing then create one and
2727 * give it's phandle to the ethernet node */
2728 node = call_prom("finddevice", 1, 1,
2729 ADDR("/builtin/mdio/ethernet-phy"));
2730 if (!PHANDLE_VALID(node)) {
2731 prom_printf("Adding Ethernet PHY node\n");
2732 call_prom("interpret", 1, 1,
2733 " s\" /builtin/mdio\" find-device"
2734 " new-device"
2735 " s\" ethernet-phy\" device-name"
2736 " 0x10 encode-int s\" reg\" property"
2737 " my-self"
2738 " ihandle>phandle"
2739 " finish-device"
2740 " s\" /builtin/ethernet\" find-device"
2741 " encode-int"
2742 " s\" phy-handle\" property"
2743 " device-end");
2744 }
2745}
2746
2747static void __init fixup_device_tree_efika(void)
2748{
2749 int sound_irq[3] = { 2, 2, 0 };
2750 int bcomm_irq[3*16] = { 3,0,0, 3,1,0, 3,2,0, 3,3,0,
2751 3,4,0, 3,5,0, 3,6,0, 3,7,0,
2752 3,8,0, 3,9,0, 3,10,0, 3,11,0,
2753 3,12,0, 3,13,0, 3,14,0, 3,15,0 };
2754 u32 node;
2755 char prop[64];
2756 int rv, len;
2757
2758 /* Check if we're really running on a EFIKA */
2759 node = call_prom("finddevice", 1, 1, ADDR("/"));
2760 if (!PHANDLE_VALID(node))
2761 return;
2762
2763 rv = prom_getprop(node, "model", prop, sizeof(prop));
2764 if (rv == PROM_ERROR)
2765 return;
2766 if (strcmp(prop, "EFIKA5K2"))
2767 return;
2768
2769 prom_printf("Applying EFIKA device tree fixups\n");
2770
2771 /* Claiming to be 'chrp' is death */
2772 node = call_prom("finddevice", 1, 1, ADDR("/"));
2773 rv = prom_getprop(node, "device_type", prop, sizeof(prop));
2774 if (rv != PROM_ERROR && (strcmp(prop, "chrp") == 0))
2775 prom_setprop(node, "/", "device_type", "efika", sizeof("efika"));
2776
2777 /* CODEGEN,description is exposed in /proc/cpuinfo so
2778 fix that too */
2779 rv = prom_getprop(node, "CODEGEN,description", prop, sizeof(prop));
2780 if (rv != PROM_ERROR && (strstr(prop, "CHRP")))
2781 prom_setprop(node, "/", "CODEGEN,description",
2782 "Efika 5200B PowerPC System",
2783 sizeof("Efika 5200B PowerPC System"));
2784
2785 /* Fixup bestcomm interrupts property */
2786 node = call_prom("finddevice", 1, 1, ADDR("/builtin/bestcomm"));
2787 if (PHANDLE_VALID(node)) {
2788 len = prom_getproplen(node, "interrupts");
2789 if (len == 12) {
2790 prom_printf("Fixing bestcomm interrupts property\n");
2791 prom_setprop(node, "/builtin/bestcom", "interrupts",
2792 bcomm_irq, sizeof(bcomm_irq));
2793 }
2794 }
2795
2796 /* Fixup sound interrupts property */
2797 node = call_prom("finddevice", 1, 1, ADDR("/builtin/sound"));
2798 if (PHANDLE_VALID(node)) {
2799 rv = prom_getprop(node, "interrupts", prop, sizeof(prop));
2800 if (rv == PROM_ERROR) {
2801 prom_printf("Adding sound interrupts property\n");
2802 prom_setprop(node, "/builtin/sound", "interrupts",
2803 sound_irq, sizeof(sound_irq));
2804 }
2805 }
2806
2807 /* Make sure ethernet phy-handle property exists */
2808 fixup_device_tree_efika_add_phy();
2809}
2810#else
2811#define fixup_device_tree_efika()
2812#endif
2813
2814static void __init fixup_device_tree(void)
2815{
2816 fixup_device_tree_maple();
2817 fixup_device_tree_maple_memory_controller();
2818 fixup_device_tree_chrp();
2819 fixup_device_tree_pmac();
2820 fixup_device_tree_efika();
2821}
2822
2823static void __init prom_find_boot_cpu(void)
2824{
2825 __be32 rval;
2826 ihandle prom_cpu;
2827 phandle cpu_pkg;
2828
2829 rval = 0;
2830 if (prom_getprop(prom.chosen, "cpu", &rval, sizeof(rval)) <= 0)
2831 return;
2832 prom_cpu = be32_to_cpu(rval);
2833
2834 cpu_pkg = call_prom("instance-to-package", 1, 1, prom_cpu);
2835
2836 prom_getprop(cpu_pkg, "reg", &rval, sizeof(rval));
2837 prom.cpu = be32_to_cpu(rval);
2838
2839 prom_debug("Booting CPU hw index = %lu\n", prom.cpu);
2840}
2841
2842static void __init prom_check_initrd(unsigned long r3, unsigned long r4)
2843{
2844#ifdef CONFIG_BLK_DEV_INITRD
2845 if (r3 && r4 && r4 != 0xdeadbeef) {
2846 __be64 val;
2847
2848 prom_initrd_start = is_kernel_addr(r3) ? __pa(r3) : r3;
2849 prom_initrd_end = prom_initrd_start + r4;
2850
2851 val = cpu_to_be64(prom_initrd_start);
2852 prom_setprop(prom.chosen, "/chosen", "linux,initrd-start",
2853 &val, sizeof(val));
2854 val = cpu_to_be64(prom_initrd_end);
2855 prom_setprop(prom.chosen, "/chosen", "linux,initrd-end",
2856 &val, sizeof(val));
2857
2858 reserve_mem(prom_initrd_start,
2859 prom_initrd_end - prom_initrd_start);
2860
2861 prom_debug("initrd_start=0x%x\n", prom_initrd_start);
2862 prom_debug("initrd_end=0x%x\n", prom_initrd_end);
2863 }
2864#endif /* CONFIG_BLK_DEV_INITRD */
2865}
2866
2867#ifdef CONFIG_PPC64
2868#ifdef CONFIG_RELOCATABLE
2869static void reloc_toc(void)
2870{
2871}
2872
2873static void unreloc_toc(void)
2874{
2875}
2876#else
2877static void __reloc_toc(unsigned long offset, unsigned long nr_entries)
2878{
2879 unsigned long i;
2880 unsigned long *toc_entry;
2881
2882 /* Get the start of the TOC by using r2 directly. */
2883 asm volatile("addi %0,2,-0x8000" : "=b" (toc_entry));
2884
2885 for (i = 0; i < nr_entries; i++) {
2886 *toc_entry = *toc_entry + offset;
2887 toc_entry++;
2888 }
2889}
2890
2891static void reloc_toc(void)
2892{
2893 unsigned long offset = reloc_offset();
2894 unsigned long nr_entries =
2895 (__prom_init_toc_end - __prom_init_toc_start) / sizeof(long);
2896
2897 __reloc_toc(offset, nr_entries);
2898
2899 mb();
2900}
2901
2902static void unreloc_toc(void)
2903{
2904 unsigned long offset = reloc_offset();
2905 unsigned long nr_entries =
2906 (__prom_init_toc_end - __prom_init_toc_start) / sizeof(long);
2907
2908 mb();
2909
2910 __reloc_toc(-offset, nr_entries);
2911}
2912#endif
2913#endif
2914
2915/*
2916 * We enter here early on, when the Open Firmware prom is still
2917 * handling exceptions and the MMU hash table for us.
2918 */
2919
2920unsigned long __init prom_init(unsigned long r3, unsigned long r4,
2921 unsigned long pp,
2922 unsigned long r6, unsigned long r7,
2923 unsigned long kbase)
2924{
2925 unsigned long hdr;
2926
2927#ifdef CONFIG_PPC32
2928 unsigned long offset = reloc_offset();
2929 reloc_got2(offset);
2930#else
2931 reloc_toc();
2932#endif
2933
2934 /*
2935 * First zero the BSS
2936 */
2937 memset(&__bss_start, 0, __bss_stop - __bss_start);
2938
2939 /*
2940 * Init interface to Open Firmware, get some node references,
2941 * like /chosen
2942 */
2943 prom_init_client_services(pp);
2944
2945 /*
2946 * See if this OF is old enough that we need to do explicit maps
2947 * and other workarounds
2948 */
2949 prom_find_mmu();
2950
2951 /*
2952 * Init prom stdout device
2953 */
2954 prom_init_stdout();
2955
2956 prom_printf("Preparing to boot %s", linux_banner);
2957
2958 /*
2959 * Get default machine type. At this point, we do not differentiate
2960 * between pSeries SMP and pSeries LPAR
2961 */
2962 of_platform = prom_find_machine_type();
2963 prom_printf("Detected machine type: %x\n", of_platform);
2964
2965#ifndef CONFIG_NONSTATIC_KERNEL
2966 /* Bail if this is a kdump kernel. */
2967 if (PHYSICAL_START > 0)
2968 prom_panic("Error: You can't boot a kdump kernel from OF!\n");
2969#endif
2970
2971 /*
2972 * Check for an initrd
2973 */
2974 prom_check_initrd(r3, r4);
2975
2976#if defined(CONFIG_PPC_PSERIES) || defined(CONFIG_PPC_POWERNV)
2977 /*
2978 * On pSeries, inform the firmware about our capabilities
2979 */
2980 if (of_platform == PLATFORM_PSERIES ||
2981 of_platform == PLATFORM_PSERIES_LPAR)
2982 prom_send_capabilities();
2983#endif
2984
2985 /*
2986 * Copy the CPU hold code
2987 */
2988 if (of_platform != PLATFORM_POWERMAC)
2989 copy_and_flush(0, kbase, 0x100, 0);
2990
2991 /*
2992 * Do early parsing of command line
2993 */
2994 early_cmdline_parse();
2995
2996 /*
2997 * Initialize memory management within prom_init
2998 */
2999 prom_init_mem();
3000
3001 /*
3002 * Determine which cpu is actually running right _now_
3003 */
3004 prom_find_boot_cpu();
3005
3006 /*
3007 * Initialize display devices
3008 */
3009 prom_check_displays();
3010
3011#if defined(CONFIG_PPC64) && defined(__BIG_ENDIAN__)
3012 /*
3013 * Initialize IOMMU (TCE tables) on pSeries. Do that before anything else
3014 * that uses the allocator, we need to make sure we get the top of memory
3015 * available for us here...
3016 */
3017 if (of_platform == PLATFORM_PSERIES)
3018 prom_initialize_tce_table();
3019#endif
3020
3021 /*
3022 * On non-powermacs, try to instantiate RTAS. PowerMacs don't
3023 * have a usable RTAS implementation.
3024 */
3025 if (of_platform != PLATFORM_POWERMAC &&
3026 of_platform != PLATFORM_OPAL)
3027 prom_instantiate_rtas();
3028
3029#ifdef CONFIG_PPC_POWERNV
3030#ifdef __BIG_ENDIAN__
3031 /* Detect HAL and try instanciating it & doing takeover */
3032 if (of_platform == PLATFORM_PSERIES_LPAR) {
3033 prom_query_opal();
3034 if (of_platform == PLATFORM_OPAL) {
3035 prom_opal_hold_cpus();
3036 prom_opal_takeover();
3037 }
3038 } else
3039#endif /* __BIG_ENDIAN__ */
3040 if (of_platform == PLATFORM_OPAL)
3041 prom_instantiate_opal();
3042#endif /* CONFIG_PPC_POWERNV */
3043
3044#ifdef CONFIG_PPC64
3045 /* instantiate sml */
3046 prom_instantiate_sml();
3047#endif
3048
3049 /*
3050 * On non-powermacs, put all CPUs in spin-loops.
3051 *
3052 * PowerMacs use a different mechanism to spin CPUs
3053 *
3054 * (This must be done after instanciating RTAS)
3055 */
3056 if (of_platform != PLATFORM_POWERMAC &&
3057 of_platform != PLATFORM_OPAL)
3058 prom_hold_cpus();
3059
3060 /*
3061 * Fill in some infos for use by the kernel later on
3062 */
3063 if (prom_memory_limit) {
3064 __be64 val = cpu_to_be64(prom_memory_limit);
3065 prom_setprop(prom.chosen, "/chosen", "linux,memory-limit",
3066 &val, sizeof(val));
3067 }
3068#ifdef CONFIG_PPC64
3069 if (prom_iommu_off)
3070 prom_setprop(prom.chosen, "/chosen", "linux,iommu-off",
3071 NULL, 0);
3072
3073 if (prom_iommu_force_on)
3074 prom_setprop(prom.chosen, "/chosen", "linux,iommu-force-on",
3075 NULL, 0);
3076
3077 if (prom_tce_alloc_start) {
3078 prom_setprop(prom.chosen, "/chosen", "linux,tce-alloc-start",
3079 &prom_tce_alloc_start,
3080 sizeof(prom_tce_alloc_start));
3081 prom_setprop(prom.chosen, "/chosen", "linux,tce-alloc-end",
3082 &prom_tce_alloc_end,
3083 sizeof(prom_tce_alloc_end));
3084 }
3085#endif
3086
3087 /*
3088 * Fixup any known bugs in the device-tree
3089 */
3090 fixup_device_tree();
3091
3092 /*
3093 * Now finally create the flattened device-tree
3094 */
3095 prom_printf("copying OF device tree...\n");
3096 flatten_device_tree();
3097
3098 /*
3099 * in case stdin is USB and still active on IBM machines...
3100 * Unfortunately quiesce crashes on some powermacs if we have
3101 * closed stdin already (in particular the powerbook 101). It
3102 * appears that the OPAL version of OFW doesn't like it either.
3103 */
3104 if (of_platform != PLATFORM_POWERMAC &&
3105 of_platform != PLATFORM_OPAL)
3106 prom_close_stdin();
3107
3108 /*
3109 * Call OF "quiesce" method to shut down pending DMA's from
3110 * devices etc...
3111 */
3112 prom_printf("Calling quiesce...\n");
3113 call_prom("quiesce", 0, 0);
3114
3115 /*
3116 * And finally, call the kernel passing it the flattened device
3117 * tree and NULL as r5, thus triggering the new entry point which
3118 * is common to us and kexec
3119 */
3120 hdr = dt_header_start;
3121
3122 /* Don't print anything after quiesce under OPAL, it crashes OFW */
3123 if (of_platform != PLATFORM_OPAL) {
3124 prom_printf("returning from prom_init\n");
3125 prom_debug("->dt_header_start=0x%x\n", hdr);
3126 }
3127
3128#ifdef CONFIG_PPC32
3129 reloc_got2(-offset);
3130#else
3131 unreloc_toc();
3132#endif
3133
3134#ifdef CONFIG_PPC_EARLY_DEBUG_OPAL
3135 /* OPAL early debug gets the OPAL base & entry in r8 and r9 */
3136 __start(hdr, kbase, 0, 0, 0,
3137 prom_opal_base, prom_opal_entry);
3138#else
3139 __start(hdr, kbase, 0, 0, 0, 0, 0);
3140#endif
3141
3142 return 0;
3143}