Loading...
1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * Procedures for creating, accessing and interpreting the device tree.
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
13
14#include <linux/kernel.h>
15#include <linux/string.h>
16#include <linux/init.h>
17#include <linux/threads.h>
18#include <linux/spinlock.h>
19#include <linux/types.h>
20#include <linux/pci.h>
21#include <linux/delay.h>
22#include <linux/initrd.h>
23#include <linux/bitops.h>
24#include <linux/export.h>
25#include <linux/kexec.h>
26#include <linux/irq.h>
27#include <linux/memblock.h>
28#include <linux/of.h>
29#include <linux/of_fdt.h>
30#include <linux/libfdt.h>
31#include <linux/cpu.h>
32#include <linux/pgtable.h>
33#include <linux/seq_buf.h>
34
35#include <asm/rtas.h>
36#include <asm/page.h>
37#include <asm/processor.h>
38#include <asm/irq.h>
39#include <asm/io.h>
40#include <asm/kdump.h>
41#include <asm/smp.h>
42#include <asm/mmu.h>
43#include <asm/paca.h>
44#include <asm/powernv.h>
45#include <asm/iommu.h>
46#include <asm/btext.h>
47#include <asm/sections.h>
48#include <asm/setup.h>
49#include <asm/pci-bridge.h>
50#include <asm/kexec.h>
51#include <asm/opal.h>
52#include <asm/fadump.h>
53#include <asm/epapr_hcalls.h>
54#include <asm/firmware.h>
55#include <asm/dt_cpu_ftrs.h>
56#include <asm/drmem.h>
57#include <asm/ultravisor.h>
58#include <asm/prom.h>
59
60#include <mm/mmu_decl.h>
61
62#ifdef DEBUG
63#define DBG(fmt...) printk(KERN_ERR fmt)
64#else
65#define DBG(fmt...)
66#endif
67
68int *chip_id_lookup_table;
69
70#ifdef CONFIG_PPC64
71int __initdata iommu_is_off;
72int __initdata iommu_force_on;
73unsigned long tce_alloc_start, tce_alloc_end;
74u64 ppc64_rma_size;
75unsigned int boot_cpu_node_count __ro_after_init;
76#endif
77static phys_addr_t first_memblock_size;
78static int __initdata boot_cpu_count;
79
80static int __init early_parse_mem(char *p)
81{
82 if (!p)
83 return 1;
84
85 memory_limit = PAGE_ALIGN(memparse(p, &p));
86 DBG("memory limit = 0x%llx\n", memory_limit);
87
88 return 0;
89}
90early_param("mem", early_parse_mem);
91
92/*
93 * overlaps_initrd - check for overlap with page aligned extension of
94 * initrd.
95 */
96static inline int overlaps_initrd(unsigned long start, unsigned long size)
97{
98#ifdef CONFIG_BLK_DEV_INITRD
99 if (!initrd_start)
100 return 0;
101
102 return (start + size) > ALIGN_DOWN(initrd_start, PAGE_SIZE) &&
103 start <= ALIGN(initrd_end, PAGE_SIZE);
104#else
105 return 0;
106#endif
107}
108
109/**
110 * move_device_tree - move tree to an unused area, if needed.
111 *
112 * The device tree may be allocated beyond our memory limit, or inside the
113 * crash kernel region for kdump, or within the page aligned range of initrd.
114 * If so, move it out of the way.
115 */
116static void __init move_device_tree(void)
117{
118 unsigned long start, size;
119 void *p;
120
121 DBG("-> move_device_tree\n");
122
123 start = __pa(initial_boot_params);
124 size = fdt_totalsize(initial_boot_params);
125
126 if ((memory_limit && (start + size) > PHYSICAL_START + memory_limit) ||
127 !memblock_is_memory(start + size - 1) ||
128 overlaps_crashkernel(start, size) || overlaps_initrd(start, size)) {
129 p = memblock_alloc_raw(size, PAGE_SIZE);
130 if (!p)
131 panic("Failed to allocate %lu bytes to move device tree\n",
132 size);
133 memcpy(p, initial_boot_params, size);
134 initial_boot_params = p;
135 DBG("Moved device tree to 0x%px\n", p);
136 }
137
138 DBG("<- move_device_tree\n");
139}
140
141/*
142 * ibm,pa/pi-features is a per-cpu property that contains a string of
143 * attribute descriptors, each of which has a 2 byte header plus up
144 * to 254 bytes worth of processor attribute bits. First header
145 * byte specifies the number of bytes following the header.
146 * Second header byte is an "attribute-specifier" type, of which
147 * zero is the only currently-defined value.
148 * Implementation: Pass in the byte and bit offset for the feature
149 * that we are interested in. The function will return -1 if the
150 * pa-features property is missing, or a 1/0 to indicate if the feature
151 * is supported/not supported. Note that the bit numbers are
152 * big-endian to match the definition in PAPR.
153 */
154struct ibm_feature {
155 unsigned long cpu_features; /* CPU_FTR_xxx bit */
156 unsigned long mmu_features; /* MMU_FTR_xxx bit */
157 unsigned int cpu_user_ftrs; /* PPC_FEATURE_xxx bit */
158 unsigned int cpu_user_ftrs2; /* PPC_FEATURE2_xxx bit */
159 unsigned char pabyte; /* byte number in ibm,pa/pi-features */
160 unsigned char pabit; /* bit number (big-endian) */
161 unsigned char invert; /* if 1, pa bit set => clear feature */
162};
163
164static struct ibm_feature ibm_pa_features[] __initdata = {
165 { .pabyte = 0, .pabit = 0, .cpu_user_ftrs = PPC_FEATURE_HAS_MMU },
166 { .pabyte = 0, .pabit = 1, .cpu_user_ftrs = PPC_FEATURE_HAS_FPU },
167 { .pabyte = 0, .pabit = 3, .cpu_features = CPU_FTR_CTRL },
168 { .pabyte = 0, .pabit = 6, .cpu_features = CPU_FTR_NOEXECUTE },
169 { .pabyte = 1, .pabit = 2, .mmu_features = MMU_FTR_CI_LARGE_PAGE },
170#ifdef CONFIG_PPC_RADIX_MMU
171 { .pabyte = 40, .pabit = 0, .mmu_features = MMU_FTR_TYPE_RADIX | MMU_FTR_GTSE },
172#endif
173 { .pabyte = 5, .pabit = 0, .cpu_features = CPU_FTR_REAL_LE,
174 .cpu_user_ftrs = PPC_FEATURE_TRUE_LE },
175 /*
176 * If the kernel doesn't support TM (ie CONFIG_PPC_TRANSACTIONAL_MEM=n),
177 * we don't want to turn on TM here, so we use the *_COMP versions
178 * which are 0 if the kernel doesn't support TM.
179 */
180 { .pabyte = 22, .pabit = 0, .cpu_features = CPU_FTR_TM_COMP,
181 .cpu_user_ftrs2 = PPC_FEATURE2_HTM_COMP | PPC_FEATURE2_HTM_NOSC_COMP },
182
183 { .pabyte = 64, .pabit = 0, .cpu_features = CPU_FTR_DAWR1 },
184};
185
186/*
187 * ibm,pi-features property provides the support of processor specific
188 * options not described in ibm,pa-features. Right now use byte 0, bit 3
189 * which indicates the occurrence of DSI interrupt when the paste operation
190 * on the suspended NX window.
191 */
192static struct ibm_feature ibm_pi_features[] __initdata = {
193 { .pabyte = 0, .pabit = 3, .mmu_features = MMU_FTR_NX_DSI },
194};
195
196static void __init scan_features(unsigned long node, const unsigned char *ftrs,
197 unsigned long tablelen,
198 struct ibm_feature *fp,
199 unsigned long ft_size)
200{
201 unsigned long i, len, bit;
202
203 /* find descriptor with type == 0 */
204 for (;;) {
205 if (tablelen < 3)
206 return;
207 len = 2 + ftrs[0];
208 if (tablelen < len)
209 return; /* descriptor 0 not found */
210 if (ftrs[1] == 0)
211 break;
212 tablelen -= len;
213 ftrs += len;
214 }
215
216 /* loop over bits we know about */
217 for (i = 0; i < ft_size; ++i, ++fp) {
218 if (fp->pabyte >= ftrs[0])
219 continue;
220 bit = (ftrs[2 + fp->pabyte] >> (7 - fp->pabit)) & 1;
221 if (bit ^ fp->invert) {
222 cur_cpu_spec->cpu_features |= fp->cpu_features;
223 cur_cpu_spec->cpu_user_features |= fp->cpu_user_ftrs;
224 cur_cpu_spec->cpu_user_features2 |= fp->cpu_user_ftrs2;
225 cur_cpu_spec->mmu_features |= fp->mmu_features;
226 } else {
227 cur_cpu_spec->cpu_features &= ~fp->cpu_features;
228 cur_cpu_spec->cpu_user_features &= ~fp->cpu_user_ftrs;
229 cur_cpu_spec->cpu_user_features2 &= ~fp->cpu_user_ftrs2;
230 cur_cpu_spec->mmu_features &= ~fp->mmu_features;
231 }
232 }
233}
234
235static void __init check_cpu_features(unsigned long node, char *name,
236 struct ibm_feature *fp,
237 unsigned long size)
238{
239 const unsigned char *pa_ftrs;
240 int tablelen;
241
242 pa_ftrs = of_get_flat_dt_prop(node, name, &tablelen);
243 if (pa_ftrs == NULL)
244 return;
245
246 scan_features(node, pa_ftrs, tablelen, fp, size);
247}
248
249#ifdef CONFIG_PPC_64S_HASH_MMU
250static void __init init_mmu_slb_size(unsigned long node)
251{
252 const __be32 *slb_size_ptr;
253
254 slb_size_ptr = of_get_flat_dt_prop(node, "slb-size", NULL) ? :
255 of_get_flat_dt_prop(node, "ibm,slb-size", NULL);
256
257 if (slb_size_ptr)
258 mmu_slb_size = be32_to_cpup(slb_size_ptr);
259}
260#else
261#define init_mmu_slb_size(node) do { } while(0)
262#endif
263
264static struct feature_property {
265 const char *name;
266 u32 min_value;
267 unsigned long cpu_feature;
268 unsigned long cpu_user_ftr;
269} feature_properties[] __initdata = {
270#ifdef CONFIG_ALTIVEC
271 {"altivec", 0, CPU_FTR_ALTIVEC, PPC_FEATURE_HAS_ALTIVEC},
272 {"ibm,vmx", 1, CPU_FTR_ALTIVEC, PPC_FEATURE_HAS_ALTIVEC},
273#endif /* CONFIG_ALTIVEC */
274#ifdef CONFIG_VSX
275 /* Yes, this _really_ is ibm,vmx == 2 to enable VSX */
276 {"ibm,vmx", 2, CPU_FTR_VSX, PPC_FEATURE_HAS_VSX},
277#endif /* CONFIG_VSX */
278#ifdef CONFIG_PPC64
279 {"ibm,dfp", 1, 0, PPC_FEATURE_HAS_DFP},
280 {"ibm,purr", 1, CPU_FTR_PURR, 0},
281 {"ibm,spurr", 1, CPU_FTR_SPURR, 0},
282#endif /* CONFIG_PPC64 */
283};
284
285#if defined(CONFIG_44x) && defined(CONFIG_PPC_FPU)
286static __init void identical_pvr_fixup(unsigned long node)
287{
288 unsigned int pvr;
289 const char *model = of_get_flat_dt_prop(node, "model", NULL);
290
291 /*
292 * Since 440GR(x)/440EP(x) processors have the same pvr,
293 * we check the node path and set bit 28 in the cur_cpu_spec
294 * pvr for EP(x) processor version. This bit is always 0 in
295 * the "real" pvr. Then we call identify_cpu again with
296 * the new logical pvr to enable FPU support.
297 */
298 if (model && strstr(model, "440EP")) {
299 pvr = cur_cpu_spec->pvr_value | 0x8;
300 identify_cpu(0, pvr);
301 DBG("Using logical pvr %x for %s\n", pvr, model);
302 }
303}
304#else
305#define identical_pvr_fixup(node) do { } while(0)
306#endif
307
308static void __init check_cpu_feature_properties(unsigned long node)
309{
310 int i;
311 struct feature_property *fp = feature_properties;
312 const __be32 *prop;
313
314 for (i = 0; i < (int)ARRAY_SIZE(feature_properties); ++i, ++fp) {
315 prop = of_get_flat_dt_prop(node, fp->name, NULL);
316 if (prop && be32_to_cpup(prop) >= fp->min_value) {
317 cur_cpu_spec->cpu_features |= fp->cpu_feature;
318 cur_cpu_spec->cpu_user_features |= fp->cpu_user_ftr;
319 }
320 }
321}
322
323static int __init early_init_dt_scan_cpus(unsigned long node,
324 const char *uname, int depth,
325 void *data)
326{
327 const char *type = of_get_flat_dt_prop(node, "device_type", NULL);
328 const __be32 *prop;
329 const __be32 *intserv;
330 int i, nthreads;
331 int len;
332 int found = -1;
333 int found_thread = 0;
334
335 /* We are scanning "cpu" nodes only */
336 if (type == NULL || strcmp(type, "cpu") != 0)
337 return 0;
338
339 if (IS_ENABLED(CONFIG_PPC64))
340 boot_cpu_node_count++;
341
342 /* Get physical cpuid */
343 intserv = of_get_flat_dt_prop(node, "ibm,ppc-interrupt-server#s", &len);
344 if (!intserv)
345 intserv = of_get_flat_dt_prop(node, "reg", &len);
346
347 nthreads = len / sizeof(int);
348
349 /*
350 * Now see if any of these threads match our boot cpu.
351 * NOTE: This must match the parsing done in smp_setup_cpu_maps.
352 */
353 for (i = 0; i < nthreads; i++) {
354 if (be32_to_cpu(intserv[i]) ==
355 fdt_boot_cpuid_phys(initial_boot_params)) {
356 found = boot_cpu_count;
357 found_thread = i;
358 }
359#ifdef CONFIG_SMP
360 /* logical cpu id is always 0 on UP kernels */
361 boot_cpu_count++;
362#endif
363 }
364
365 /* Not the boot CPU */
366 if (found < 0)
367 return 0;
368
369 DBG("boot cpu: logical %d physical %d\n", found,
370 be32_to_cpu(intserv[found_thread]));
371 boot_cpuid = found;
372
373 // Pass the boot CPU's hard CPU id back to our caller
374 *((u32 *)data) = be32_to_cpu(intserv[found_thread]);
375
376 /*
377 * PAPR defines "logical" PVR values for cpus that
378 * meet various levels of the architecture:
379 * 0x0f000001 Architecture version 2.04
380 * 0x0f000002 Architecture version 2.05
381 * If the cpu-version property in the cpu node contains
382 * such a value, we call identify_cpu again with the
383 * logical PVR value in order to use the cpu feature
384 * bits appropriate for the architecture level.
385 *
386 * A POWER6 partition in "POWER6 architected" mode
387 * uses the 0x0f000002 PVR value; in POWER5+ mode
388 * it uses 0x0f000001.
389 *
390 * If we're using device tree CPU feature discovery then we don't
391 * support the cpu-version property, and it's the responsibility of the
392 * firmware/hypervisor to provide the correct feature set for the
393 * architecture level via the ibm,powerpc-cpu-features binding.
394 */
395 if (!dt_cpu_ftrs_in_use()) {
396 prop = of_get_flat_dt_prop(node, "cpu-version", NULL);
397 if (prop && (be32_to_cpup(prop) & 0xff000000) == 0x0f000000) {
398 identify_cpu(0, be32_to_cpup(prop));
399 seq_buf_printf(&ppc_hw_desc, "0x%04x ", be32_to_cpup(prop));
400 }
401
402 check_cpu_feature_properties(node);
403 check_cpu_features(node, "ibm,pa-features", ibm_pa_features,
404 ARRAY_SIZE(ibm_pa_features));
405 check_cpu_features(node, "ibm,pi-features", ibm_pi_features,
406 ARRAY_SIZE(ibm_pi_features));
407 }
408
409 identical_pvr_fixup(node);
410 init_mmu_slb_size(node);
411
412#ifdef CONFIG_PPC64
413 if (nthreads == 1)
414 cur_cpu_spec->cpu_features &= ~CPU_FTR_SMT;
415 else if (!dt_cpu_ftrs_in_use())
416 cur_cpu_spec->cpu_features |= CPU_FTR_SMT;
417#endif
418
419 return 0;
420}
421
422static int __init early_init_dt_scan_chosen_ppc(unsigned long node,
423 const char *uname,
424 int depth, void *data)
425{
426 const unsigned long *lprop; /* All these set by kernel, so no need to convert endian */
427
428 /* Use common scan routine to determine if this is the chosen node */
429 if (early_init_dt_scan_chosen(data) < 0)
430 return 0;
431
432#ifdef CONFIG_PPC64
433 /* check if iommu is forced on or off */
434 if (of_get_flat_dt_prop(node, "linux,iommu-off", NULL) != NULL)
435 iommu_is_off = 1;
436 if (of_get_flat_dt_prop(node, "linux,iommu-force-on", NULL) != NULL)
437 iommu_force_on = 1;
438#endif
439
440 /* mem=x on the command line is the preferred mechanism */
441 lprop = of_get_flat_dt_prop(node, "linux,memory-limit", NULL);
442 if (lprop)
443 memory_limit = *lprop;
444
445#ifdef CONFIG_PPC64
446 lprop = of_get_flat_dt_prop(node, "linux,tce-alloc-start", NULL);
447 if (lprop)
448 tce_alloc_start = *lprop;
449 lprop = of_get_flat_dt_prop(node, "linux,tce-alloc-end", NULL);
450 if (lprop)
451 tce_alloc_end = *lprop;
452#endif
453
454#ifdef CONFIG_KEXEC_CORE
455 lprop = of_get_flat_dt_prop(node, "linux,crashkernel-base", NULL);
456 if (lprop)
457 crashk_res.start = *lprop;
458
459 lprop = of_get_flat_dt_prop(node, "linux,crashkernel-size", NULL);
460 if (lprop)
461 crashk_res.end = crashk_res.start + *lprop - 1;
462#endif
463
464 /* break now */
465 return 1;
466}
467
468/*
469 * Compare the range against max mem limit and update
470 * size if it cross the limit.
471 */
472
473#ifdef CONFIG_SPARSEMEM
474static bool __init validate_mem_limit(u64 base, u64 *size)
475{
476 u64 max_mem = 1UL << (MAX_PHYSMEM_BITS);
477
478 if (base >= max_mem)
479 return false;
480 if ((base + *size) > max_mem)
481 *size = max_mem - base;
482 return true;
483}
484#else
485static bool __init validate_mem_limit(u64 base, u64 *size)
486{
487 return true;
488}
489#endif
490
491#ifdef CONFIG_PPC_PSERIES
492/*
493 * Interpret the ibm dynamic reconfiguration memory LMBs.
494 * This contains a list of memory blocks along with NUMA affinity
495 * information.
496 */
497static int __init early_init_drmem_lmb(struct drmem_lmb *lmb,
498 const __be32 **usm,
499 void *data)
500{
501 u64 base, size;
502 int is_kexec_kdump = 0, rngs;
503
504 base = lmb->base_addr;
505 size = drmem_lmb_size();
506 rngs = 1;
507
508 /*
509 * Skip this block if the reserved bit is set in flags
510 * or if the block is not assigned to this partition.
511 */
512 if ((lmb->flags & DRCONF_MEM_RESERVED) ||
513 !(lmb->flags & DRCONF_MEM_ASSIGNED))
514 return 0;
515
516 if (*usm)
517 is_kexec_kdump = 1;
518
519 if (is_kexec_kdump) {
520 /*
521 * For each memblock in ibm,dynamic-memory, a
522 * corresponding entry in linux,drconf-usable-memory
523 * property contains a counter 'p' followed by 'p'
524 * (base, size) duple. Now read the counter from
525 * linux,drconf-usable-memory property
526 */
527 rngs = dt_mem_next_cell(dt_root_size_cells, usm);
528 if (!rngs) /* there are no (base, size) duple */
529 return 0;
530 }
531
532 do {
533 if (is_kexec_kdump) {
534 base = dt_mem_next_cell(dt_root_addr_cells, usm);
535 size = dt_mem_next_cell(dt_root_size_cells, usm);
536 }
537
538 if (iommu_is_off) {
539 if (base >= 0x80000000ul)
540 continue;
541 if ((base + size) > 0x80000000ul)
542 size = 0x80000000ul - base;
543 }
544
545 if (!validate_mem_limit(base, &size))
546 continue;
547
548 DBG("Adding: %llx -> %llx\n", base, size);
549 memblock_add(base, size);
550
551 if (lmb->flags & DRCONF_MEM_HOTREMOVABLE)
552 memblock_mark_hotplug(base, size);
553 } while (--rngs);
554
555 return 0;
556}
557#endif /* CONFIG_PPC_PSERIES */
558
559static int __init early_init_dt_scan_memory_ppc(void)
560{
561#ifdef CONFIG_PPC_PSERIES
562 const void *fdt = initial_boot_params;
563 int node = fdt_path_offset(fdt, "/ibm,dynamic-reconfiguration-memory");
564
565 if (node > 0)
566 walk_drmem_lmbs_early(node, NULL, early_init_drmem_lmb);
567
568#endif
569
570 return early_init_dt_scan_memory();
571}
572
573/*
574 * For a relocatable kernel, we need to get the memstart_addr first,
575 * then use it to calculate the virtual kernel start address. This has
576 * to happen at a very early stage (before machine_init). In this case,
577 * we just want to get the memstart_address and would not like to mess the
578 * memblock at this stage. So introduce a variable to skip the memblock_add()
579 * for this reason.
580 */
581#ifdef CONFIG_RELOCATABLE
582static int add_mem_to_memblock = 1;
583#else
584#define add_mem_to_memblock 1
585#endif
586
587void __init early_init_dt_add_memory_arch(u64 base, u64 size)
588{
589#ifdef CONFIG_PPC64
590 if (iommu_is_off) {
591 if (base >= 0x80000000ul)
592 return;
593 if ((base + size) > 0x80000000ul)
594 size = 0x80000000ul - base;
595 }
596#endif
597 /* Keep track of the beginning of memory -and- the size of
598 * the very first block in the device-tree as it represents
599 * the RMA on ppc64 server
600 */
601 if (base < memstart_addr) {
602 memstart_addr = base;
603 first_memblock_size = size;
604 }
605
606 /* Add the chunk to the MEMBLOCK list */
607 if (add_mem_to_memblock) {
608 if (validate_mem_limit(base, &size))
609 memblock_add(base, size);
610 }
611}
612
613static void __init early_reserve_mem_dt(void)
614{
615 unsigned long i, dt_root;
616 int len;
617 const __be32 *prop;
618
619 early_init_fdt_reserve_self();
620 early_init_fdt_scan_reserved_mem();
621
622 dt_root = of_get_flat_dt_root();
623
624 prop = of_get_flat_dt_prop(dt_root, "reserved-ranges", &len);
625
626 if (!prop)
627 return;
628
629 DBG("Found new-style reserved-ranges\n");
630
631 /* Each reserved range is an (address,size) pair, 2 cells each,
632 * totalling 4 cells per range. */
633 for (i = 0; i < len / (sizeof(*prop) * 4); i++) {
634 u64 base, size;
635
636 base = of_read_number(prop + (i * 4) + 0, 2);
637 size = of_read_number(prop + (i * 4) + 2, 2);
638
639 if (size) {
640 DBG("reserving: %llx -> %llx\n", base, size);
641 memblock_reserve(base, size);
642 }
643 }
644}
645
646static void __init early_reserve_mem(void)
647{
648 __be64 *reserve_map;
649
650 reserve_map = (__be64 *)(((unsigned long)initial_boot_params) +
651 fdt_off_mem_rsvmap(initial_boot_params));
652
653 /* Look for the new "reserved-regions" property in the DT */
654 early_reserve_mem_dt();
655
656#ifdef CONFIG_BLK_DEV_INITRD
657 /* Then reserve the initrd, if any */
658 if (initrd_start && (initrd_end > initrd_start)) {
659 memblock_reserve(ALIGN_DOWN(__pa(initrd_start), PAGE_SIZE),
660 ALIGN(initrd_end, PAGE_SIZE) -
661 ALIGN_DOWN(initrd_start, PAGE_SIZE));
662 }
663#endif /* CONFIG_BLK_DEV_INITRD */
664
665 if (!IS_ENABLED(CONFIG_PPC32))
666 return;
667
668 /*
669 * Handle the case where we might be booting from an old kexec
670 * image that setup the mem_rsvmap as pairs of 32-bit values
671 */
672 if (be64_to_cpup(reserve_map) > 0xffffffffull) {
673 u32 base_32, size_32;
674 __be32 *reserve_map_32 = (__be32 *)reserve_map;
675
676 DBG("Found old 32-bit reserve map\n");
677
678 while (1) {
679 base_32 = be32_to_cpup(reserve_map_32++);
680 size_32 = be32_to_cpup(reserve_map_32++);
681 if (size_32 == 0)
682 break;
683 DBG("reserving: %x -> %x\n", base_32, size_32);
684 memblock_reserve(base_32, size_32);
685 }
686 return;
687 }
688}
689
690#ifdef CONFIG_PPC_TRANSACTIONAL_MEM
691static bool tm_disabled __initdata;
692
693static int __init parse_ppc_tm(char *str)
694{
695 bool res;
696
697 if (kstrtobool(str, &res))
698 return -EINVAL;
699
700 tm_disabled = !res;
701
702 return 0;
703}
704early_param("ppc_tm", parse_ppc_tm);
705
706static void __init tm_init(void)
707{
708 if (tm_disabled) {
709 pr_info("Disabling hardware transactional memory (HTM)\n");
710 cur_cpu_spec->cpu_user_features2 &=
711 ~(PPC_FEATURE2_HTM_NOSC | PPC_FEATURE2_HTM);
712 cur_cpu_spec->cpu_features &= ~CPU_FTR_TM;
713 return;
714 }
715
716 pnv_tm_init();
717}
718#else
719static void tm_init(void) { }
720#endif /* CONFIG_PPC_TRANSACTIONAL_MEM */
721
722static int __init
723early_init_dt_scan_model(unsigned long node, const char *uname,
724 int depth, void *data)
725{
726 const char *prop;
727
728 if (depth != 0)
729 return 0;
730
731 prop = of_get_flat_dt_prop(node, "model", NULL);
732 if (prop)
733 seq_buf_printf(&ppc_hw_desc, "%s ", prop);
734
735 /* break now */
736 return 1;
737}
738
739#ifdef CONFIG_PPC64
740static void __init save_fscr_to_task(void)
741{
742 /*
743 * Ensure the init_task (pid 0, aka swapper) uses the value of FSCR we
744 * have configured via the device tree features or via __init_FSCR().
745 * That value will then be propagated to pid 1 (init) and all future
746 * processes.
747 */
748 if (early_cpu_has_feature(CPU_FTR_ARCH_207S))
749 init_task.thread.fscr = mfspr(SPRN_FSCR);
750}
751#else
752static inline void save_fscr_to_task(void) {}
753#endif
754
755
756void __init early_init_devtree(void *params)
757{
758 u32 boot_cpu_hwid;
759 phys_addr_t limit;
760
761 DBG(" -> early_init_devtree(%px)\n", params);
762
763 /* Too early to BUG_ON(), do it by hand */
764 if (!early_init_dt_verify(params))
765 panic("BUG: Failed verifying flat device tree, bad version?");
766
767 of_scan_flat_dt(early_init_dt_scan_model, NULL);
768
769#ifdef CONFIG_PPC_RTAS
770 /* Some machines might need RTAS info for debugging, grab it now. */
771 of_scan_flat_dt(early_init_dt_scan_rtas, NULL);
772#endif
773
774#ifdef CONFIG_PPC_POWERNV
775 /* Some machines might need OPAL info for debugging, grab it now. */
776 of_scan_flat_dt(early_init_dt_scan_opal, NULL);
777
778 /* Scan tree for ultravisor feature */
779 of_scan_flat_dt(early_init_dt_scan_ultravisor, NULL);
780#endif
781
782#if defined(CONFIG_FA_DUMP) || defined(CONFIG_PRESERVE_FA_DUMP)
783 /* scan tree to see if dump is active during last boot */
784 of_scan_flat_dt(early_init_dt_scan_fw_dump, NULL);
785#endif
786
787 /* Retrieve various informations from the /chosen node of the
788 * device-tree, including the platform type, initrd location and
789 * size, TCE reserve, and more ...
790 */
791 of_scan_flat_dt(early_init_dt_scan_chosen_ppc, boot_command_line);
792
793 /* Scan memory nodes and rebuild MEMBLOCKs */
794 early_init_dt_scan_root();
795 early_init_dt_scan_memory_ppc();
796
797 /*
798 * As generic code authors expect to be able to use static keys
799 * in early_param() handlers, we initialize the static keys just
800 * before parsing early params (it's fine to call jump_label_init()
801 * more than once).
802 */
803 jump_label_init();
804 parse_early_param();
805
806 /* make sure we've parsed cmdline for mem= before this */
807 if (memory_limit)
808 first_memblock_size = min_t(u64, first_memblock_size, memory_limit);
809 setup_initial_memory_limit(memstart_addr, first_memblock_size);
810 /* Reserve MEMBLOCK regions used by kernel, initrd, dt, etc... */
811 memblock_reserve(PHYSICAL_START, __pa(_end) - PHYSICAL_START);
812 /* If relocatable, reserve first 32k for interrupt vectors etc. */
813 if (PHYSICAL_START > MEMORY_START)
814 memblock_reserve(MEMORY_START, 0x8000);
815 reserve_kdump_trampoline();
816#if defined(CONFIG_FA_DUMP) || defined(CONFIG_PRESERVE_FA_DUMP)
817 /*
818 * If we fail to reserve memory for firmware-assisted dump then
819 * fallback to kexec based kdump.
820 */
821 if (fadump_reserve_mem() == 0)
822#endif
823 reserve_crashkernel();
824 early_reserve_mem();
825
826 /* Ensure that total memory size is page-aligned. */
827 limit = ALIGN(memory_limit ?: memblock_phys_mem_size(), PAGE_SIZE);
828 memblock_enforce_memory_limit(limit);
829
830#if defined(CONFIG_PPC_BOOK3S_64) && defined(CONFIG_PPC_4K_PAGES)
831 if (!early_radix_enabled())
832 memblock_cap_memory_range(0, 1UL << (H_MAX_PHYSMEM_BITS));
833#endif
834
835 memblock_allow_resize();
836 memblock_dump_all();
837
838 DBG("Phys. mem: %llx\n", (unsigned long long)memblock_phys_mem_size());
839
840 /* We may need to relocate the flat tree, do it now.
841 * FIXME .. and the initrd too? */
842 move_device_tree();
843
844 DBG("Scanning CPUs ...\n");
845
846 dt_cpu_ftrs_scan();
847
848 // We can now add the CPU name & PVR to the hardware description
849 seq_buf_printf(&ppc_hw_desc, "%s 0x%04lx ", cur_cpu_spec->cpu_name, mfspr(SPRN_PVR));
850
851 /* Retrieve CPU related informations from the flat tree
852 * (altivec support, boot CPU ID, ...)
853 */
854 of_scan_flat_dt(early_init_dt_scan_cpus, &boot_cpu_hwid);
855 if (boot_cpuid < 0) {
856 printk("Failed to identify boot CPU !\n");
857 BUG();
858 }
859
860 save_fscr_to_task();
861
862#if defined(CONFIG_SMP) && defined(CONFIG_PPC64)
863 /* We'll later wait for secondaries to check in; there are
864 * NCPUS-1 non-boot CPUs :-)
865 */
866 spinning_secondaries = boot_cpu_count - 1;
867#endif
868
869 mmu_early_init_devtree();
870
871 // NB. paca is not installed until later in early_setup()
872 allocate_paca_ptrs();
873 allocate_paca(boot_cpuid);
874 set_hard_smp_processor_id(boot_cpuid, boot_cpu_hwid);
875
876#ifdef CONFIG_PPC_POWERNV
877 /* Scan and build the list of machine check recoverable ranges */
878 of_scan_flat_dt(early_init_dt_scan_recoverable_ranges, NULL);
879#endif
880 epapr_paravirt_early_init();
881
882 /* Now try to figure out if we are running on LPAR and so on */
883 pseries_probe_fw_features();
884
885 /*
886 * Initialize pkey features and default AMR/IAMR values
887 */
888 pkey_early_init_devtree();
889
890#ifdef CONFIG_PPC_PS3
891 /* Identify PS3 firmware */
892 if (of_flat_dt_is_compatible(of_get_flat_dt_root(), "sony,ps3"))
893 powerpc_firmware_features |= FW_FEATURE_PS3_POSSIBLE;
894#endif
895
896 tm_init();
897
898 DBG(" <- early_init_devtree()\n");
899}
900
901#ifdef CONFIG_RELOCATABLE
902/*
903 * This function run before early_init_devtree, so we have to init
904 * initial_boot_params.
905 */
906void __init early_get_first_memblock_info(void *params, phys_addr_t *size)
907{
908 /* Setup flat device-tree pointer */
909 initial_boot_params = params;
910
911 /*
912 * Scan the memory nodes and set add_mem_to_memblock to 0 to avoid
913 * mess the memblock.
914 */
915 add_mem_to_memblock = 0;
916 early_init_dt_scan_root();
917 early_init_dt_scan_memory_ppc();
918 add_mem_to_memblock = 1;
919
920 if (size)
921 *size = first_memblock_size;
922}
923#endif
924
925/*******
926 *
927 * New implementation of the OF "find" APIs, return a refcounted
928 * object, call of_node_put() when done. The device tree and list
929 * are protected by a rw_lock.
930 *
931 * Note that property management will need some locking as well,
932 * this isn't dealt with yet.
933 *
934 *******/
935
936/**
937 * of_get_ibm_chip_id - Returns the IBM "chip-id" of a device
938 * @np: device node of the device
939 *
940 * This looks for a property "ibm,chip-id" in the node or any
941 * of its parents and returns its content, or -1 if it cannot
942 * be found.
943 */
944int of_get_ibm_chip_id(struct device_node *np)
945{
946 of_node_get(np);
947 while (np) {
948 u32 chip_id;
949
950 /*
951 * Skiboot may produce memory nodes that contain more than one
952 * cell in chip-id, we only read the first one here.
953 */
954 if (!of_property_read_u32(np, "ibm,chip-id", &chip_id)) {
955 of_node_put(np);
956 return chip_id;
957 }
958
959 np = of_get_next_parent(np);
960 }
961 return -1;
962}
963EXPORT_SYMBOL(of_get_ibm_chip_id);
964
965/**
966 * cpu_to_chip_id - Return the cpus chip-id
967 * @cpu: The logical cpu number.
968 *
969 * Return the value of the ibm,chip-id property corresponding to the given
970 * logical cpu number. If the chip-id can not be found, returns -1.
971 */
972int cpu_to_chip_id(int cpu)
973{
974 struct device_node *np;
975 int ret = -1, idx;
976
977 idx = cpu / threads_per_core;
978 if (chip_id_lookup_table && chip_id_lookup_table[idx] != -1)
979 return chip_id_lookup_table[idx];
980
981 np = of_get_cpu_node(cpu, NULL);
982 if (np) {
983 ret = of_get_ibm_chip_id(np);
984 of_node_put(np);
985
986 if (chip_id_lookup_table)
987 chip_id_lookup_table[idx] = ret;
988 }
989
990 return ret;
991}
992EXPORT_SYMBOL(cpu_to_chip_id);
993
994bool arch_match_cpu_phys_id(int cpu, u64 phys_id)
995{
996#ifdef CONFIG_SMP
997 /*
998 * Early firmware scanning must use this rather than
999 * get_hard_smp_processor_id because we don't have pacas allocated
1000 * until memory topology is discovered.
1001 */
1002 if (cpu_to_phys_id != NULL)
1003 return (int)phys_id == cpu_to_phys_id[cpu];
1004#endif
1005
1006 return (int)phys_id == get_hard_smp_processor_id(cpu);
1007}
1/*
2 * Procedures for creating, accessing and interpreting the device tree.
3 *
4 * Paul Mackerras August 1996.
5 * Copyright (C) 1996-2005 Paul Mackerras.
6 *
7 * Adapted for 64bit PowerPC by Dave Engebretsen and Peter Bergner.
8 * {engebret|bergner}@us.ibm.com
9 *
10 * This program is free software; you can redistribute it and/or
11 * modify it under the terms of the GNU General Public License
12 * as published by the Free Software Foundation; either version
13 * 2 of the License, or (at your option) any later version.
14 */
15
16#undef DEBUG
17
18#include <stdarg.h>
19#include <linux/kernel.h>
20#include <linux/string.h>
21#include <linux/init.h>
22#include <linux/threads.h>
23#include <linux/spinlock.h>
24#include <linux/types.h>
25#include <linux/pci.h>
26#include <linux/stringify.h>
27#include <linux/delay.h>
28#include <linux/initrd.h>
29#include <linux/bitops.h>
30#include <linux/export.h>
31#include <linux/kexec.h>
32#include <linux/debugfs.h>
33#include <linux/irq.h>
34#include <linux/memblock.h>
35#include <linux/of.h>
36#include <linux/of_fdt.h>
37
38#include <asm/prom.h>
39#include <asm/rtas.h>
40#include <asm/page.h>
41#include <asm/processor.h>
42#include <asm/irq.h>
43#include <asm/io.h>
44#include <asm/kdump.h>
45#include <asm/smp.h>
46#include <asm/mmu.h>
47#include <asm/paca.h>
48#include <asm/pgtable.h>
49#include <asm/pci.h>
50#include <asm/iommu.h>
51#include <asm/btext.h>
52#include <asm/sections.h>
53#include <asm/machdep.h>
54#include <asm/pci-bridge.h>
55#include <asm/kexec.h>
56#include <asm/opal.h>
57#include <asm/fadump.h>
58#include <asm/debug.h>
59
60#include <mm/mmu_decl.h>
61
62#ifdef DEBUG
63#define DBG(fmt...) printk(KERN_ERR fmt)
64#else
65#define DBG(fmt...)
66#endif
67
68#ifdef CONFIG_PPC64
69int __initdata iommu_is_off;
70int __initdata iommu_force_on;
71unsigned long tce_alloc_start, tce_alloc_end;
72u64 ppc64_rma_size;
73#endif
74static phys_addr_t first_memblock_size;
75static int __initdata boot_cpu_count;
76
77static int __init early_parse_mem(char *p)
78{
79 if (!p)
80 return 1;
81
82 memory_limit = PAGE_ALIGN(memparse(p, &p));
83 DBG("memory limit = 0x%llx\n", memory_limit);
84
85 return 0;
86}
87early_param("mem", early_parse_mem);
88
89/*
90 * overlaps_initrd - check for overlap with page aligned extension of
91 * initrd.
92 */
93static inline int overlaps_initrd(unsigned long start, unsigned long size)
94{
95#ifdef CONFIG_BLK_DEV_INITRD
96 if (!initrd_start)
97 return 0;
98
99 return (start + size) > _ALIGN_DOWN(initrd_start, PAGE_SIZE) &&
100 start <= _ALIGN_UP(initrd_end, PAGE_SIZE);
101#else
102 return 0;
103#endif
104}
105
106/**
107 * move_device_tree - move tree to an unused area, if needed.
108 *
109 * The device tree may be allocated beyond our memory limit, or inside the
110 * crash kernel region for kdump, or within the page aligned range of initrd.
111 * If so, move it out of the way.
112 */
113static void __init move_device_tree(void)
114{
115 unsigned long start, size;
116 void *p;
117
118 DBG("-> move_device_tree\n");
119
120 start = __pa(initial_boot_params);
121 size = be32_to_cpu(initial_boot_params->totalsize);
122
123 if ((memory_limit && (start + size) > PHYSICAL_START + memory_limit) ||
124 overlaps_crashkernel(start, size) ||
125 overlaps_initrd(start, size)) {
126 p = __va(memblock_alloc(size, PAGE_SIZE));
127 memcpy(p, initial_boot_params, size);
128 initial_boot_params = (struct boot_param_header *)p;
129 DBG("Moved device tree to 0x%p\n", p);
130 }
131
132 DBG("<- move_device_tree\n");
133}
134
135/*
136 * ibm,pa-features is a per-cpu property that contains a string of
137 * attribute descriptors, each of which has a 2 byte header plus up
138 * to 254 bytes worth of processor attribute bits. First header
139 * byte specifies the number of bytes following the header.
140 * Second header byte is an "attribute-specifier" type, of which
141 * zero is the only currently-defined value.
142 * Implementation: Pass in the byte and bit offset for the feature
143 * that we are interested in. The function will return -1 if the
144 * pa-features property is missing, or a 1/0 to indicate if the feature
145 * is supported/not supported. Note that the bit numbers are
146 * big-endian to match the definition in PAPR.
147 */
148static struct ibm_pa_feature {
149 unsigned long cpu_features; /* CPU_FTR_xxx bit */
150 unsigned long mmu_features; /* MMU_FTR_xxx bit */
151 unsigned int cpu_user_ftrs; /* PPC_FEATURE_xxx bit */
152 unsigned char pabyte; /* byte number in ibm,pa-features */
153 unsigned char pabit; /* bit number (big-endian) */
154 unsigned char invert; /* if 1, pa bit set => clear feature */
155} ibm_pa_features[] __initdata = {
156 {0, 0, PPC_FEATURE_HAS_MMU, 0, 0, 0},
157 {0, 0, PPC_FEATURE_HAS_FPU, 0, 1, 0},
158 {0, MMU_FTR_SLB, 0, 0, 2, 0},
159 {CPU_FTR_CTRL, 0, 0, 0, 3, 0},
160 {CPU_FTR_NOEXECUTE, 0, 0, 0, 6, 0},
161 {CPU_FTR_NODSISRALIGN, 0, 0, 1, 1, 1},
162 {0, MMU_FTR_CI_LARGE_PAGE, 0, 1, 2, 0},
163 {CPU_FTR_REAL_LE, PPC_FEATURE_TRUE_LE, 5, 0, 0},
164};
165
166static void __init scan_features(unsigned long node, unsigned char *ftrs,
167 unsigned long tablelen,
168 struct ibm_pa_feature *fp,
169 unsigned long ft_size)
170{
171 unsigned long i, len, bit;
172
173 /* find descriptor with type == 0 */
174 for (;;) {
175 if (tablelen < 3)
176 return;
177 len = 2 + ftrs[0];
178 if (tablelen < len)
179 return; /* descriptor 0 not found */
180 if (ftrs[1] == 0)
181 break;
182 tablelen -= len;
183 ftrs += len;
184 }
185
186 /* loop over bits we know about */
187 for (i = 0; i < ft_size; ++i, ++fp) {
188 if (fp->pabyte >= ftrs[0])
189 continue;
190 bit = (ftrs[2 + fp->pabyte] >> (7 - fp->pabit)) & 1;
191 if (bit ^ fp->invert) {
192 cur_cpu_spec->cpu_features |= fp->cpu_features;
193 cur_cpu_spec->cpu_user_features |= fp->cpu_user_ftrs;
194 cur_cpu_spec->mmu_features |= fp->mmu_features;
195 } else {
196 cur_cpu_spec->cpu_features &= ~fp->cpu_features;
197 cur_cpu_spec->cpu_user_features &= ~fp->cpu_user_ftrs;
198 cur_cpu_spec->mmu_features &= ~fp->mmu_features;
199 }
200 }
201}
202
203static void __init check_cpu_pa_features(unsigned long node)
204{
205 unsigned char *pa_ftrs;
206 unsigned long tablelen;
207
208 pa_ftrs = of_get_flat_dt_prop(node, "ibm,pa-features", &tablelen);
209 if (pa_ftrs == NULL)
210 return;
211
212 scan_features(node, pa_ftrs, tablelen,
213 ibm_pa_features, ARRAY_SIZE(ibm_pa_features));
214}
215
216#ifdef CONFIG_PPC_STD_MMU_64
217static void __init check_cpu_slb_size(unsigned long node)
218{
219 __be32 *slb_size_ptr;
220
221 slb_size_ptr = of_get_flat_dt_prop(node, "slb-size", NULL);
222 if (slb_size_ptr != NULL) {
223 mmu_slb_size = be32_to_cpup(slb_size_ptr);
224 return;
225 }
226 slb_size_ptr = of_get_flat_dt_prop(node, "ibm,slb-size", NULL);
227 if (slb_size_ptr != NULL) {
228 mmu_slb_size = be32_to_cpup(slb_size_ptr);
229 }
230}
231#else
232#define check_cpu_slb_size(node) do { } while(0)
233#endif
234
235static struct feature_property {
236 const char *name;
237 u32 min_value;
238 unsigned long cpu_feature;
239 unsigned long cpu_user_ftr;
240} feature_properties[] __initdata = {
241#ifdef CONFIG_ALTIVEC
242 {"altivec", 0, CPU_FTR_ALTIVEC, PPC_FEATURE_HAS_ALTIVEC},
243 {"ibm,vmx", 1, CPU_FTR_ALTIVEC, PPC_FEATURE_HAS_ALTIVEC},
244#endif /* CONFIG_ALTIVEC */
245#ifdef CONFIG_VSX
246 /* Yes, this _really_ is ibm,vmx == 2 to enable VSX */
247 {"ibm,vmx", 2, CPU_FTR_VSX, PPC_FEATURE_HAS_VSX},
248#endif /* CONFIG_VSX */
249#ifdef CONFIG_PPC64
250 {"ibm,dfp", 1, 0, PPC_FEATURE_HAS_DFP},
251 {"ibm,purr", 1, CPU_FTR_PURR, 0},
252 {"ibm,spurr", 1, CPU_FTR_SPURR, 0},
253#endif /* CONFIG_PPC64 */
254};
255
256#if defined(CONFIG_44x) && defined(CONFIG_PPC_FPU)
257static inline void identical_pvr_fixup(unsigned long node)
258{
259 unsigned int pvr;
260 char *model = of_get_flat_dt_prop(node, "model", NULL);
261
262 /*
263 * Since 440GR(x)/440EP(x) processors have the same pvr,
264 * we check the node path and set bit 28 in the cur_cpu_spec
265 * pvr for EP(x) processor version. This bit is always 0 in
266 * the "real" pvr. Then we call identify_cpu again with
267 * the new logical pvr to enable FPU support.
268 */
269 if (model && strstr(model, "440EP")) {
270 pvr = cur_cpu_spec->pvr_value | 0x8;
271 identify_cpu(0, pvr);
272 DBG("Using logical pvr %x for %s\n", pvr, model);
273 }
274}
275#else
276#define identical_pvr_fixup(node) do { } while(0)
277#endif
278
279static void __init check_cpu_feature_properties(unsigned long node)
280{
281 unsigned long i;
282 struct feature_property *fp = feature_properties;
283 const __be32 *prop;
284
285 for (i = 0; i < ARRAY_SIZE(feature_properties); ++i, ++fp) {
286 prop = of_get_flat_dt_prop(node, fp->name, NULL);
287 if (prop && be32_to_cpup(prop) >= fp->min_value) {
288 cur_cpu_spec->cpu_features |= fp->cpu_feature;
289 cur_cpu_spec->cpu_user_features |= fp->cpu_user_ftr;
290 }
291 }
292}
293
294static int __init early_init_dt_scan_cpus(unsigned long node,
295 const char *uname, int depth,
296 void *data)
297{
298 char *type = of_get_flat_dt_prop(node, "device_type", NULL);
299 const __be32 *prop;
300 const __be32 *intserv;
301 int i, nthreads;
302 unsigned long len;
303 int found = -1;
304 int found_thread = 0;
305
306 /* We are scanning "cpu" nodes only */
307 if (type == NULL || strcmp(type, "cpu") != 0)
308 return 0;
309
310 /* Get physical cpuid */
311 intserv = of_get_flat_dt_prop(node, "ibm,ppc-interrupt-server#s", &len);
312 if (intserv) {
313 nthreads = len / sizeof(int);
314 } else {
315 intserv = of_get_flat_dt_prop(node, "reg", NULL);
316 nthreads = 1;
317 }
318
319 /*
320 * Now see if any of these threads match our boot cpu.
321 * NOTE: This must match the parsing done in smp_setup_cpu_maps.
322 */
323 for (i = 0; i < nthreads; i++) {
324 /*
325 * version 2 of the kexec param format adds the phys cpuid of
326 * booted proc.
327 */
328 if (be32_to_cpu(initial_boot_params->version) >= 2) {
329 if (be32_to_cpu(intserv[i]) ==
330 be32_to_cpu(initial_boot_params->boot_cpuid_phys)) {
331 found = boot_cpu_count;
332 found_thread = i;
333 }
334 } else {
335 /*
336 * Check if it's the boot-cpu, set it's hw index now,
337 * unfortunately this format did not support booting
338 * off secondary threads.
339 */
340 if (of_get_flat_dt_prop(node,
341 "linux,boot-cpu", NULL) != NULL)
342 found = boot_cpu_count;
343 }
344#ifdef CONFIG_SMP
345 /* logical cpu id is always 0 on UP kernels */
346 boot_cpu_count++;
347#endif
348 }
349
350 /* Not the boot CPU */
351 if (found < 0)
352 return 0;
353
354 DBG("boot cpu: logical %d physical %d\n", found,
355 be32_to_cpu(intserv[found_thread]));
356 boot_cpuid = found;
357 set_hard_smp_processor_id(found, be32_to_cpu(intserv[found_thread]));
358
359 /*
360 * PAPR defines "logical" PVR values for cpus that
361 * meet various levels of the architecture:
362 * 0x0f000001 Architecture version 2.04
363 * 0x0f000002 Architecture version 2.05
364 * If the cpu-version property in the cpu node contains
365 * such a value, we call identify_cpu again with the
366 * logical PVR value in order to use the cpu feature
367 * bits appropriate for the architecture level.
368 *
369 * A POWER6 partition in "POWER6 architected" mode
370 * uses the 0x0f000002 PVR value; in POWER5+ mode
371 * it uses 0x0f000001.
372 */
373 prop = of_get_flat_dt_prop(node, "cpu-version", NULL);
374 if (prop && (be32_to_cpup(prop) & 0xff000000) == 0x0f000000)
375 identify_cpu(0, be32_to_cpup(prop));
376
377 identical_pvr_fixup(node);
378
379 check_cpu_feature_properties(node);
380 check_cpu_pa_features(node);
381 check_cpu_slb_size(node);
382
383#ifdef CONFIG_PPC64
384 if (nthreads > 1)
385 cur_cpu_spec->cpu_features |= CPU_FTR_SMT;
386 else
387 cur_cpu_spec->cpu_features &= ~CPU_FTR_SMT;
388#endif
389 return 0;
390}
391
392int __init early_init_dt_scan_chosen_ppc(unsigned long node, const char *uname,
393 int depth, void *data)
394{
395 unsigned long *lprop; /* All these set by kernel, so no need to convert endian */
396
397 /* Use common scan routine to determine if this is the chosen node */
398 if (early_init_dt_scan_chosen(node, uname, depth, data) == 0)
399 return 0;
400
401#ifdef CONFIG_PPC64
402 /* check if iommu is forced on or off */
403 if (of_get_flat_dt_prop(node, "linux,iommu-off", NULL) != NULL)
404 iommu_is_off = 1;
405 if (of_get_flat_dt_prop(node, "linux,iommu-force-on", NULL) != NULL)
406 iommu_force_on = 1;
407#endif
408
409 /* mem=x on the command line is the preferred mechanism */
410 lprop = of_get_flat_dt_prop(node, "linux,memory-limit", NULL);
411 if (lprop)
412 memory_limit = *lprop;
413
414#ifdef CONFIG_PPC64
415 lprop = of_get_flat_dt_prop(node, "linux,tce-alloc-start", NULL);
416 if (lprop)
417 tce_alloc_start = *lprop;
418 lprop = of_get_flat_dt_prop(node, "linux,tce-alloc-end", NULL);
419 if (lprop)
420 tce_alloc_end = *lprop;
421#endif
422
423#ifdef CONFIG_KEXEC
424 lprop = of_get_flat_dt_prop(node, "linux,crashkernel-base", NULL);
425 if (lprop)
426 crashk_res.start = *lprop;
427
428 lprop = of_get_flat_dt_prop(node, "linux,crashkernel-size", NULL);
429 if (lprop)
430 crashk_res.end = crashk_res.start + *lprop - 1;
431#endif
432
433 /* break now */
434 return 1;
435}
436
437#ifdef CONFIG_PPC_PSERIES
438/*
439 * Interpret the ibm,dynamic-memory property in the
440 * /ibm,dynamic-reconfiguration-memory node.
441 * This contains a list of memory blocks along with NUMA affinity
442 * information.
443 */
444static int __init early_init_dt_scan_drconf_memory(unsigned long node)
445{
446 __be32 *dm, *ls, *usm;
447 unsigned long l, n, flags;
448 u64 base, size, memblock_size;
449 unsigned int is_kexec_kdump = 0, rngs;
450
451 ls = of_get_flat_dt_prop(node, "ibm,lmb-size", &l);
452 if (ls == NULL || l < dt_root_size_cells * sizeof(__be32))
453 return 0;
454 memblock_size = dt_mem_next_cell(dt_root_size_cells, &ls);
455
456 dm = of_get_flat_dt_prop(node, "ibm,dynamic-memory", &l);
457 if (dm == NULL || l < sizeof(__be32))
458 return 0;
459
460 n = of_read_number(dm++, 1); /* number of entries */
461 if (l < (n * (dt_root_addr_cells + 4) + 1) * sizeof(__be32))
462 return 0;
463
464 /* check if this is a kexec/kdump kernel. */
465 usm = of_get_flat_dt_prop(node, "linux,drconf-usable-memory",
466 &l);
467 if (usm != NULL)
468 is_kexec_kdump = 1;
469
470 for (; n != 0; --n) {
471 base = dt_mem_next_cell(dt_root_addr_cells, &dm);
472 flags = of_read_number(&dm[3], 1);
473 /* skip DRC index, pad, assoc. list index, flags */
474 dm += 4;
475 /* skip this block if the reserved bit is set in flags (0x80)
476 or if the block is not assigned to this partition (0x8) */
477 if ((flags & 0x80) || !(flags & 0x8))
478 continue;
479 size = memblock_size;
480 rngs = 1;
481 if (is_kexec_kdump) {
482 /*
483 * For each memblock in ibm,dynamic-memory, a corresponding
484 * entry in linux,drconf-usable-memory property contains
485 * a counter 'p' followed by 'p' (base, size) duple.
486 * Now read the counter from
487 * linux,drconf-usable-memory property
488 */
489 rngs = dt_mem_next_cell(dt_root_size_cells, &usm);
490 if (!rngs) /* there are no (base, size) duple */
491 continue;
492 }
493 do {
494 if (is_kexec_kdump) {
495 base = dt_mem_next_cell(dt_root_addr_cells,
496 &usm);
497 size = dt_mem_next_cell(dt_root_size_cells,
498 &usm);
499 }
500 if (iommu_is_off) {
501 if (base >= 0x80000000ul)
502 continue;
503 if ((base + size) > 0x80000000ul)
504 size = 0x80000000ul - base;
505 }
506 memblock_add(base, size);
507 } while (--rngs);
508 }
509 memblock_dump_all();
510 return 0;
511}
512#else
513#define early_init_dt_scan_drconf_memory(node) 0
514#endif /* CONFIG_PPC_PSERIES */
515
516static int __init early_init_dt_scan_memory_ppc(unsigned long node,
517 const char *uname,
518 int depth, void *data)
519{
520 if (depth == 1 &&
521 strcmp(uname, "ibm,dynamic-reconfiguration-memory") == 0)
522 return early_init_dt_scan_drconf_memory(node);
523
524 return early_init_dt_scan_memory(node, uname, depth, data);
525}
526
527/*
528 * For a relocatable kernel, we need to get the memstart_addr first,
529 * then use it to calculate the virtual kernel start address. This has
530 * to happen at a very early stage (before machine_init). In this case,
531 * we just want to get the memstart_address and would not like to mess the
532 * memblock at this stage. So introduce a variable to skip the memblock_add()
533 * for this reason.
534 */
535#ifdef CONFIG_RELOCATABLE
536static int add_mem_to_memblock = 1;
537#else
538#define add_mem_to_memblock 1
539#endif
540
541void __init early_init_dt_add_memory_arch(u64 base, u64 size)
542{
543#ifdef CONFIG_PPC64
544 if (iommu_is_off) {
545 if (base >= 0x80000000ul)
546 return;
547 if ((base + size) > 0x80000000ul)
548 size = 0x80000000ul - base;
549 }
550#endif
551 /* Keep track of the beginning of memory -and- the size of
552 * the very first block in the device-tree as it represents
553 * the RMA on ppc64 server
554 */
555 if (base < memstart_addr) {
556 memstart_addr = base;
557 first_memblock_size = size;
558 }
559
560 /* Add the chunk to the MEMBLOCK list */
561 if (add_mem_to_memblock)
562 memblock_add(base, size);
563}
564
565static void __init early_reserve_mem_dt(void)
566{
567 unsigned long i, len, dt_root;
568 const __be32 *prop;
569
570 dt_root = of_get_flat_dt_root();
571
572 prop = of_get_flat_dt_prop(dt_root, "reserved-ranges", &len);
573
574 if (!prop)
575 return;
576
577 DBG("Found new-style reserved-ranges\n");
578
579 /* Each reserved range is an (address,size) pair, 2 cells each,
580 * totalling 4 cells per range. */
581 for (i = 0; i < len / (sizeof(*prop) * 4); i++) {
582 u64 base, size;
583
584 base = of_read_number(prop + (i * 4) + 0, 2);
585 size = of_read_number(prop + (i * 4) + 2, 2);
586
587 if (size) {
588 DBG("reserving: %llx -> %llx\n", base, size);
589 memblock_reserve(base, size);
590 }
591 }
592
593 early_init_fdt_scan_reserved_mem();
594}
595
596static void __init early_reserve_mem(void)
597{
598 u64 base, size;
599 __be64 *reserve_map;
600 unsigned long self_base;
601 unsigned long self_size;
602
603 reserve_map = (__be64 *)(((unsigned long)initial_boot_params) +
604 be32_to_cpu(initial_boot_params->off_mem_rsvmap));
605
606 /* before we do anything, lets reserve the dt blob */
607 self_base = __pa((unsigned long)initial_boot_params);
608 self_size = be32_to_cpu(initial_boot_params->totalsize);
609 memblock_reserve(self_base, self_size);
610
611 /* Look for the new "reserved-regions" property in the DT */
612 early_reserve_mem_dt();
613
614#ifdef CONFIG_BLK_DEV_INITRD
615 /* Then reserve the initrd, if any */
616 if (initrd_start && (initrd_end > initrd_start)) {
617 memblock_reserve(_ALIGN_DOWN(__pa(initrd_start), PAGE_SIZE),
618 _ALIGN_UP(initrd_end, PAGE_SIZE) -
619 _ALIGN_DOWN(initrd_start, PAGE_SIZE));
620 }
621#endif /* CONFIG_BLK_DEV_INITRD */
622
623#ifdef CONFIG_PPC32
624 /*
625 * Handle the case where we might be booting from an old kexec
626 * image that setup the mem_rsvmap as pairs of 32-bit values
627 */
628 if (be64_to_cpup(reserve_map) > 0xffffffffull) {
629 u32 base_32, size_32;
630 __be32 *reserve_map_32 = (__be32 *)reserve_map;
631
632 DBG("Found old 32-bit reserve map\n");
633
634 while (1) {
635 base_32 = be32_to_cpup(reserve_map_32++);
636 size_32 = be32_to_cpup(reserve_map_32++);
637 if (size_32 == 0)
638 break;
639 /* skip if the reservation is for the blob */
640 if (base_32 == self_base && size_32 == self_size)
641 continue;
642 DBG("reserving: %x -> %x\n", base_32, size_32);
643 memblock_reserve(base_32, size_32);
644 }
645 return;
646 }
647#endif
648 DBG("Processing reserve map\n");
649
650 /* Handle the reserve map in the fdt blob if it exists */
651 while (1) {
652 base = be64_to_cpup(reserve_map++);
653 size = be64_to_cpup(reserve_map++);
654 if (size == 0)
655 break;
656 DBG("reserving: %llx -> %llx\n", base, size);
657 memblock_reserve(base, size);
658 }
659}
660
661void __init early_init_devtree(void *params)
662{
663 phys_addr_t limit;
664
665 DBG(" -> early_init_devtree(%p)\n", params);
666
667 /* Setup flat device-tree pointer */
668 initial_boot_params = params;
669
670#ifdef CONFIG_PPC_RTAS
671 /* Some machines might need RTAS info for debugging, grab it now. */
672 of_scan_flat_dt(early_init_dt_scan_rtas, NULL);
673#endif
674
675#ifdef CONFIG_PPC_POWERNV
676 /* Some machines might need OPAL info for debugging, grab it now. */
677 of_scan_flat_dt(early_init_dt_scan_opal, NULL);
678#endif
679
680#ifdef CONFIG_FA_DUMP
681 /* scan tree to see if dump is active during last boot */
682 of_scan_flat_dt(early_init_dt_scan_fw_dump, NULL);
683#endif
684
685 /* Pre-initialize the cmd_line with the content of boot_commmand_line,
686 * which will be empty except when the content of the variable has
687 * been overriden by a bootloading mechanism. This happens typically
688 * with HAL takeover
689 */
690 strlcpy(cmd_line, boot_command_line, COMMAND_LINE_SIZE);
691
692 /* Retrieve various informations from the /chosen node of the
693 * device-tree, including the platform type, initrd location and
694 * size, TCE reserve, and more ...
695 */
696 of_scan_flat_dt(early_init_dt_scan_chosen_ppc, cmd_line);
697
698 /* Scan memory nodes and rebuild MEMBLOCKs */
699 of_scan_flat_dt(early_init_dt_scan_root, NULL);
700 of_scan_flat_dt(early_init_dt_scan_memory_ppc, NULL);
701
702 /* Save command line for /proc/cmdline and then parse parameters */
703 strlcpy(boot_command_line, cmd_line, COMMAND_LINE_SIZE);
704 parse_early_param();
705
706 /* make sure we've parsed cmdline for mem= before this */
707 if (memory_limit)
708 first_memblock_size = min_t(u64, first_memblock_size, memory_limit);
709 setup_initial_memory_limit(memstart_addr, first_memblock_size);
710 /* Reserve MEMBLOCK regions used by kernel, initrd, dt, etc... */
711 memblock_reserve(PHYSICAL_START, __pa(klimit) - PHYSICAL_START);
712 /* If relocatable, reserve first 32k for interrupt vectors etc. */
713 if (PHYSICAL_START > MEMORY_START)
714 memblock_reserve(MEMORY_START, 0x8000);
715 reserve_kdump_trampoline();
716#ifdef CONFIG_FA_DUMP
717 /*
718 * If we fail to reserve memory for firmware-assisted dump then
719 * fallback to kexec based kdump.
720 */
721 if (fadump_reserve_mem() == 0)
722#endif
723 reserve_crashkernel();
724 early_reserve_mem();
725
726 /*
727 * Ensure that total memory size is page-aligned, because otherwise
728 * mark_bootmem() gets upset.
729 */
730 limit = ALIGN(memory_limit ?: memblock_phys_mem_size(), PAGE_SIZE);
731 memblock_enforce_memory_limit(limit);
732
733 memblock_allow_resize();
734 memblock_dump_all();
735
736 DBG("Phys. mem: %llx\n", memblock_phys_mem_size());
737
738 /* We may need to relocate the flat tree, do it now.
739 * FIXME .. and the initrd too? */
740 move_device_tree();
741
742 allocate_pacas();
743
744 DBG("Scanning CPUs ...\n");
745
746 /* Retrieve CPU related informations from the flat tree
747 * (altivec support, boot CPU ID, ...)
748 */
749 of_scan_flat_dt(early_init_dt_scan_cpus, NULL);
750 if (boot_cpuid < 0) {
751 printk("Failed to indentify boot CPU !\n");
752 BUG();
753 }
754
755#if defined(CONFIG_SMP) && defined(CONFIG_PPC64)
756 /* We'll later wait for secondaries to check in; there are
757 * NCPUS-1 non-boot CPUs :-)
758 */
759 spinning_secondaries = boot_cpu_count - 1;
760#endif
761
762#ifdef CONFIG_PPC_POWERNV
763 /* Scan and build the list of machine check recoverable ranges */
764 of_scan_flat_dt(early_init_dt_scan_recoverable_ranges, NULL);
765#endif
766
767 DBG(" <- early_init_devtree()\n");
768}
769
770#ifdef CONFIG_RELOCATABLE
771/*
772 * This function run before early_init_devtree, so we have to init
773 * initial_boot_params.
774 */
775void __init early_get_first_memblock_info(void *params, phys_addr_t *size)
776{
777 /* Setup flat device-tree pointer */
778 initial_boot_params = params;
779
780 /*
781 * Scan the memory nodes and set add_mem_to_memblock to 0 to avoid
782 * mess the memblock.
783 */
784 add_mem_to_memblock = 0;
785 of_scan_flat_dt(early_init_dt_scan_root, NULL);
786 of_scan_flat_dt(early_init_dt_scan_memory_ppc, NULL);
787 add_mem_to_memblock = 1;
788
789 if (size)
790 *size = first_memblock_size;
791}
792#endif
793
794/*******
795 *
796 * New implementation of the OF "find" APIs, return a refcounted
797 * object, call of_node_put() when done. The device tree and list
798 * are protected by a rw_lock.
799 *
800 * Note that property management will need some locking as well,
801 * this isn't dealt with yet.
802 *
803 *******/
804
805/**
806 * of_get_ibm_chip_id - Returns the IBM "chip-id" of a device
807 * @np: device node of the device
808 *
809 * This looks for a property "ibm,chip-id" in the node or any
810 * of its parents and returns its content, or -1 if it cannot
811 * be found.
812 */
813int of_get_ibm_chip_id(struct device_node *np)
814{
815 of_node_get(np);
816 while(np) {
817 struct device_node *old = np;
818 const __be32 *prop;
819
820 prop = of_get_property(np, "ibm,chip-id", NULL);
821 if (prop) {
822 of_node_put(np);
823 return be32_to_cpup(prop);
824 }
825 np = of_get_parent(np);
826 of_node_put(old);
827 }
828 return -1;
829}
830
831/**
832 * cpu_to_chip_id - Return the cpus chip-id
833 * @cpu: The logical cpu number.
834 *
835 * Return the value of the ibm,chip-id property corresponding to the given
836 * logical cpu number. If the chip-id can not be found, returns -1.
837 */
838int cpu_to_chip_id(int cpu)
839{
840 struct device_node *np;
841
842 np = of_get_cpu_node(cpu, NULL);
843 if (!np)
844 return -1;
845
846 of_node_put(np);
847 return of_get_ibm_chip_id(np);
848}
849EXPORT_SYMBOL(cpu_to_chip_id);
850
851#ifdef CONFIG_PPC_PSERIES
852/*
853 * Fix up the uninitialized fields in a new device node:
854 * name, type and pci-specific fields
855 */
856
857static int of_finish_dynamic_node(struct device_node *node)
858{
859 struct device_node *parent = of_get_parent(node);
860 int err = 0;
861 const phandle *ibm_phandle;
862
863 node->name = of_get_property(node, "name", NULL);
864 node->type = of_get_property(node, "device_type", NULL);
865
866 if (!node->name)
867 node->name = "<NULL>";
868 if (!node->type)
869 node->type = "<NULL>";
870
871 if (!parent) {
872 err = -ENODEV;
873 goto out;
874 }
875
876 /* We don't support that function on PowerMac, at least
877 * not yet
878 */
879 if (machine_is(powermac))
880 return -ENODEV;
881
882 /* fix up new node's phandle field */
883 if ((ibm_phandle = of_get_property(node, "ibm,phandle", NULL)))
884 node->phandle = *ibm_phandle;
885
886out:
887 of_node_put(parent);
888 return err;
889}
890
891static int prom_reconfig_notifier(struct notifier_block *nb,
892 unsigned long action, void *node)
893{
894 int err;
895
896 switch (action) {
897 case OF_RECONFIG_ATTACH_NODE:
898 err = of_finish_dynamic_node(node);
899 if (err < 0)
900 printk(KERN_ERR "finish_node returned %d\n", err);
901 break;
902 default:
903 err = 0;
904 break;
905 }
906 return notifier_from_errno(err);
907}
908
909static struct notifier_block prom_reconfig_nb = {
910 .notifier_call = prom_reconfig_notifier,
911 .priority = 10, /* This one needs to run first */
912};
913
914static int __init prom_reconfig_setup(void)
915{
916 return of_reconfig_notifier_register(&prom_reconfig_nb);
917}
918__initcall(prom_reconfig_setup);
919#endif
920
921bool arch_match_cpu_phys_id(int cpu, u64 phys_id)
922{
923 return (int)phys_id == get_hard_smp_processor_id(cpu);
924}
925
926#if defined(CONFIG_DEBUG_FS) && defined(DEBUG)
927static struct debugfs_blob_wrapper flat_dt_blob;
928
929static int __init export_flat_device_tree(void)
930{
931 struct dentry *d;
932
933 flat_dt_blob.data = initial_boot_params;
934 flat_dt_blob.size = be32_to_cpu(initial_boot_params->totalsize);
935
936 d = debugfs_create_blob("flat-device-tree", S_IFREG | S_IRUSR,
937 powerpc_debugfs_root, &flat_dt_blob);
938 if (!d)
939 return 1;
940
941 return 0;
942}
943__initcall(export_flat_device_tree);
944#endif