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