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 <stdarg.h>
15#include <linux/kernel.h>
16#include <linux/string.h>
17#include <linux/init.h>
18#include <linux/threads.h>
19#include <linux/spinlock.h>
20#include <linux/types.h>
21#include <linux/pci.h>
22#include <linux/delay.h>
23#include <linux/initrd.h>
24#include <linux/bitops.h>
25#include <linux/export.h>
26#include <linux/kexec.h>
27#include <linux/irq.h>
28#include <linux/memblock.h>
29#include <linux/of.h>
30#include <linux/of_fdt.h>
31#include <linux/libfdt.h>
32#include <linux/cpu.h>
33#include <linux/pgtable.h>
34
35#include <asm/prom.h>
36#include <asm/rtas.h>
37#include <asm/page.h>
38#include <asm/processor.h>
39#include <asm/irq.h>
40#include <asm/io.h>
41#include <asm/kdump.h>
42#include <asm/smp.h>
43#include <asm/mmu.h>
44#include <asm/paca.h>
45#include <asm/powernv.h>
46#include <asm/iommu.h>
47#include <asm/btext.h>
48#include <asm/sections.h>
49#include <asm/machdep.h>
50#include <asm/pci-bridge.h>
51#include <asm/kexec.h>
52#include <asm/opal.h>
53#include <asm/fadump.h>
54#include <asm/epapr_hcalls.h>
55#include <asm/firmware.h>
56#include <asm/dt_cpu_ftrs.h>
57#include <asm/drmem.h>
58#include <asm/ultravisor.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;
75#endif
76static phys_addr_t first_memblock_size;
77static int __initdata boot_cpu_count;
78
79static int __init early_parse_mem(char *p)
80{
81 if (!p)
82 return 1;
83
84 memory_limit = PAGE_ALIGN(memparse(p, &p));
85 DBG("memory limit = 0x%llx\n", memory_limit);
86
87 return 0;
88}
89early_param("mem", early_parse_mem);
90
91/*
92 * overlaps_initrd - check for overlap with page aligned extension of
93 * initrd.
94 */
95static inline int overlaps_initrd(unsigned long start, unsigned long size)
96{
97#ifdef CONFIG_BLK_DEV_INITRD
98 if (!initrd_start)
99 return 0;
100
101 return (start + size) > ALIGN_DOWN(initrd_start, PAGE_SIZE) &&
102 start <= ALIGN(initrd_end, PAGE_SIZE);
103#else
104 return 0;
105#endif
106}
107
108/**
109 * move_device_tree - move tree to an unused area, if needed.
110 *
111 * The device tree may be allocated beyond our memory limit, or inside the
112 * crash kernel region for kdump, or within the page aligned range of initrd.
113 * If so, move it out of the way.
114 */
115static void __init move_device_tree(void)
116{
117 unsigned long start, size;
118 void *p;
119
120 DBG("-> move_device_tree\n");
121
122 start = __pa(initial_boot_params);
123 size = fdt_totalsize(initial_boot_params);
124
125 if ((memory_limit && (start + size) > PHYSICAL_START + memory_limit) ||
126 !memblock_is_memory(start + size - 1) ||
127 overlaps_crashkernel(start, size) || overlaps_initrd(start, size)) {
128 p = memblock_alloc_raw(size, PAGE_SIZE);
129 if (!p)
130 panic("Failed to allocate %lu bytes to move device tree\n",
131 size);
132 memcpy(p, initial_boot_params, size);
133 initial_boot_params = p;
134 DBG("Moved device tree to 0x%px\n", p);
135 }
136
137 DBG("<- move_device_tree\n");
138}
139
140/*
141 * ibm,pa-features is a per-cpu property that contains a string of
142 * attribute descriptors, each of which has a 2 byte header plus up
143 * to 254 bytes worth of processor attribute bits. First header
144 * byte specifies the number of bytes following the header.
145 * Second header byte is an "attribute-specifier" type, of which
146 * zero is the only currently-defined value.
147 * Implementation: Pass in the byte and bit offset for the feature
148 * that we are interested in. The function will return -1 if the
149 * pa-features property is missing, or a 1/0 to indicate if the feature
150 * is supported/not supported. Note that the bit numbers are
151 * big-endian to match the definition in PAPR.
152 */
153static struct ibm_pa_feature {
154 unsigned long cpu_features; /* CPU_FTR_xxx bit */
155 unsigned long mmu_features; /* MMU_FTR_xxx bit */
156 unsigned int cpu_user_ftrs; /* PPC_FEATURE_xxx bit */
157 unsigned int cpu_user_ftrs2; /* PPC_FEATURE2_xxx bit */
158 unsigned char pabyte; /* byte number in ibm,pa-features */
159 unsigned char pabit; /* bit number (big-endian) */
160 unsigned char invert; /* if 1, pa bit set => clear feature */
161} ibm_pa_features[] __initdata = {
162 { .pabyte = 0, .pabit = 0, .cpu_user_ftrs = PPC_FEATURE_HAS_MMU },
163 { .pabyte = 0, .pabit = 1, .cpu_user_ftrs = PPC_FEATURE_HAS_FPU },
164 { .pabyte = 0, .pabit = 3, .cpu_features = CPU_FTR_CTRL },
165 { .pabyte = 0, .pabit = 6, .cpu_features = CPU_FTR_NOEXECUTE },
166 { .pabyte = 1, .pabit = 2, .mmu_features = MMU_FTR_CI_LARGE_PAGE },
167#ifdef CONFIG_PPC_RADIX_MMU
168 { .pabyte = 40, .pabit = 0, .mmu_features = MMU_FTR_TYPE_RADIX | MMU_FTR_GTSE },
169#endif
170 { .pabyte = 5, .pabit = 0, .cpu_features = CPU_FTR_REAL_LE,
171 .cpu_user_ftrs = PPC_FEATURE_TRUE_LE },
172 /*
173 * If the kernel doesn't support TM (ie CONFIG_PPC_TRANSACTIONAL_MEM=n),
174 * we don't want to turn on TM here, so we use the *_COMP versions
175 * which are 0 if the kernel doesn't support TM.
176 */
177 { .pabyte = 22, .pabit = 0, .cpu_features = CPU_FTR_TM_COMP,
178 .cpu_user_ftrs2 = PPC_FEATURE2_HTM_COMP | PPC_FEATURE2_HTM_NOSC_COMP },
179
180 { .pabyte = 64, .pabit = 0, .cpu_features = CPU_FTR_DAWR1 },
181};
182
183static void __init scan_features(unsigned long node, const unsigned char *ftrs,
184 unsigned long tablelen,
185 struct ibm_pa_feature *fp,
186 unsigned long ft_size)
187{
188 unsigned long i, len, bit;
189
190 /* find descriptor with type == 0 */
191 for (;;) {
192 if (tablelen < 3)
193 return;
194 len = 2 + ftrs[0];
195 if (tablelen < len)
196 return; /* descriptor 0 not found */
197 if (ftrs[1] == 0)
198 break;
199 tablelen -= len;
200 ftrs += len;
201 }
202
203 /* loop over bits we know about */
204 for (i = 0; i < ft_size; ++i, ++fp) {
205 if (fp->pabyte >= ftrs[0])
206 continue;
207 bit = (ftrs[2 + fp->pabyte] >> (7 - fp->pabit)) & 1;
208 if (bit ^ fp->invert) {
209 cur_cpu_spec->cpu_features |= fp->cpu_features;
210 cur_cpu_spec->cpu_user_features |= fp->cpu_user_ftrs;
211 cur_cpu_spec->cpu_user_features2 |= fp->cpu_user_ftrs2;
212 cur_cpu_spec->mmu_features |= fp->mmu_features;
213 } else {
214 cur_cpu_spec->cpu_features &= ~fp->cpu_features;
215 cur_cpu_spec->cpu_user_features &= ~fp->cpu_user_ftrs;
216 cur_cpu_spec->cpu_user_features2 &= ~fp->cpu_user_ftrs2;
217 cur_cpu_spec->mmu_features &= ~fp->mmu_features;
218 }
219 }
220}
221
222static void __init check_cpu_pa_features(unsigned long node)
223{
224 const unsigned char *pa_ftrs;
225 int tablelen;
226
227 pa_ftrs = of_get_flat_dt_prop(node, "ibm,pa-features", &tablelen);
228 if (pa_ftrs == NULL)
229 return;
230
231 scan_features(node, pa_ftrs, tablelen,
232 ibm_pa_features, ARRAY_SIZE(ibm_pa_features));
233}
234
235#ifdef CONFIG_PPC_BOOK3S_64
236static void __init init_mmu_slb_size(unsigned long node)
237{
238 const __be32 *slb_size_ptr;
239
240 slb_size_ptr = of_get_flat_dt_prop(node, "slb-size", NULL) ? :
241 of_get_flat_dt_prop(node, "ibm,slb-size", NULL);
242
243 if (slb_size_ptr)
244 mmu_slb_size = be32_to_cpup(slb_size_ptr);
245}
246#else
247#define init_mmu_slb_size(node) do { } while(0)
248#endif
249
250static struct feature_property {
251 const char *name;
252 u32 min_value;
253 unsigned long cpu_feature;
254 unsigned long cpu_user_ftr;
255} feature_properties[] __initdata = {
256#ifdef CONFIG_ALTIVEC
257 {"altivec", 0, CPU_FTR_ALTIVEC, PPC_FEATURE_HAS_ALTIVEC},
258 {"ibm,vmx", 1, CPU_FTR_ALTIVEC, PPC_FEATURE_HAS_ALTIVEC},
259#endif /* CONFIG_ALTIVEC */
260#ifdef CONFIG_VSX
261 /* Yes, this _really_ is ibm,vmx == 2 to enable VSX */
262 {"ibm,vmx", 2, CPU_FTR_VSX, PPC_FEATURE_HAS_VSX},
263#endif /* CONFIG_VSX */
264#ifdef CONFIG_PPC64
265 {"ibm,dfp", 1, 0, PPC_FEATURE_HAS_DFP},
266 {"ibm,purr", 1, CPU_FTR_PURR, 0},
267 {"ibm,spurr", 1, CPU_FTR_SPURR, 0},
268#endif /* CONFIG_PPC64 */
269};
270
271#if defined(CONFIG_44x) && defined(CONFIG_PPC_FPU)
272static __init void identical_pvr_fixup(unsigned long node)
273{
274 unsigned int pvr;
275 const char *model = of_get_flat_dt_prop(node, "model", NULL);
276
277 /*
278 * Since 440GR(x)/440EP(x) processors have the same pvr,
279 * we check the node path and set bit 28 in the cur_cpu_spec
280 * pvr for EP(x) processor version. This bit is always 0 in
281 * the "real" pvr. Then we call identify_cpu again with
282 * the new logical pvr to enable FPU support.
283 */
284 if (model && strstr(model, "440EP")) {
285 pvr = cur_cpu_spec->pvr_value | 0x8;
286 identify_cpu(0, pvr);
287 DBG("Using logical pvr %x for %s\n", pvr, model);
288 }
289}
290#else
291#define identical_pvr_fixup(node) do { } while(0)
292#endif
293
294static void __init check_cpu_feature_properties(unsigned long node)
295{
296 int i;
297 struct feature_property *fp = feature_properties;
298 const __be32 *prop;
299
300 for (i = 0; i < (int)ARRAY_SIZE(feature_properties); ++i, ++fp) {
301 prop = of_get_flat_dt_prop(node, fp->name, NULL);
302 if (prop && be32_to_cpup(prop) >= fp->min_value) {
303 cur_cpu_spec->cpu_features |= fp->cpu_feature;
304 cur_cpu_spec->cpu_user_features |= fp->cpu_user_ftr;
305 }
306 }
307}
308
309static int __init early_init_dt_scan_cpus(unsigned long node,
310 const char *uname, int depth,
311 void *data)
312{
313 const char *type = of_get_flat_dt_prop(node, "device_type", NULL);
314 const __be32 *prop;
315 const __be32 *intserv;
316 int i, nthreads;
317 int len;
318 int found = -1;
319 int found_thread = 0;
320
321 /* We are scanning "cpu" nodes only */
322 if (type == NULL || strcmp(type, "cpu") != 0)
323 return 0;
324
325 /* Get physical cpuid */
326 intserv = of_get_flat_dt_prop(node, "ibm,ppc-interrupt-server#s", &len);
327 if (!intserv)
328 intserv = of_get_flat_dt_prop(node, "reg", &len);
329
330 nthreads = len / sizeof(int);
331
332 /*
333 * Now see if any of these threads match our boot cpu.
334 * NOTE: This must match the parsing done in smp_setup_cpu_maps.
335 */
336 for (i = 0; i < nthreads; i++) {
337 if (be32_to_cpu(intserv[i]) ==
338 fdt_boot_cpuid_phys(initial_boot_params)) {
339 found = boot_cpu_count;
340 found_thread = i;
341 }
342#ifdef CONFIG_SMP
343 /* logical cpu id is always 0 on UP kernels */
344 boot_cpu_count++;
345#endif
346 }
347
348 /* Not the boot CPU */
349 if (found < 0)
350 return 0;
351
352 DBG("boot cpu: logical %d physical %d\n", found,
353 be32_to_cpu(intserv[found_thread]));
354 boot_cpuid = found;
355
356 /*
357 * PAPR defines "logical" PVR values for cpus that
358 * meet various levels of the architecture:
359 * 0x0f000001 Architecture version 2.04
360 * 0x0f000002 Architecture version 2.05
361 * If the cpu-version property in the cpu node contains
362 * such a value, we call identify_cpu again with the
363 * logical PVR value in order to use the cpu feature
364 * bits appropriate for the architecture level.
365 *
366 * A POWER6 partition in "POWER6 architected" mode
367 * uses the 0x0f000002 PVR value; in POWER5+ mode
368 * it uses 0x0f000001.
369 *
370 * If we're using device tree CPU feature discovery then we don't
371 * support the cpu-version property, and it's the responsibility of the
372 * firmware/hypervisor to provide the correct feature set for the
373 * architecture level via the ibm,powerpc-cpu-features binding.
374 */
375 if (!dt_cpu_ftrs_in_use()) {
376 prop = of_get_flat_dt_prop(node, "cpu-version", NULL);
377 if (prop && (be32_to_cpup(prop) & 0xff000000) == 0x0f000000)
378 identify_cpu(0, be32_to_cpup(prop));
379
380 check_cpu_feature_properties(node);
381 check_cpu_pa_features(node);
382 }
383
384 identical_pvr_fixup(node);
385 init_mmu_slb_size(node);
386
387#ifdef CONFIG_PPC64
388 if (nthreads == 1)
389 cur_cpu_spec->cpu_features &= ~CPU_FTR_SMT;
390 else if (!dt_cpu_ftrs_in_use())
391 cur_cpu_spec->cpu_features |= CPU_FTR_SMT;
392 allocate_paca(boot_cpuid);
393#endif
394 set_hard_smp_processor_id(found, be32_to_cpu(intserv[found_thread]));
395
396 return 0;
397}
398
399static int __init early_init_dt_scan_chosen_ppc(unsigned long node,
400 const char *uname,
401 int depth, void *data)
402{
403 const unsigned long *lprop; /* All these set by kernel, so no need to convert endian */
404
405 /* Use common scan routine to determine if this is the chosen node */
406 if (early_init_dt_scan_chosen(node, uname, depth, data) == 0)
407 return 0;
408
409#ifdef CONFIG_PPC64
410 /* check if iommu is forced on or off */
411 if (of_get_flat_dt_prop(node, "linux,iommu-off", NULL) != NULL)
412 iommu_is_off = 1;
413 if (of_get_flat_dt_prop(node, "linux,iommu-force-on", NULL) != NULL)
414 iommu_force_on = 1;
415#endif
416
417 /* mem=x on the command line is the preferred mechanism */
418 lprop = of_get_flat_dt_prop(node, "linux,memory-limit", NULL);
419 if (lprop)
420 memory_limit = *lprop;
421
422#ifdef CONFIG_PPC64
423 lprop = of_get_flat_dt_prop(node, "linux,tce-alloc-start", NULL);
424 if (lprop)
425 tce_alloc_start = *lprop;
426 lprop = of_get_flat_dt_prop(node, "linux,tce-alloc-end", NULL);
427 if (lprop)
428 tce_alloc_end = *lprop;
429#endif
430
431#ifdef CONFIG_KEXEC_CORE
432 lprop = of_get_flat_dt_prop(node, "linux,crashkernel-base", NULL);
433 if (lprop)
434 crashk_res.start = *lprop;
435
436 lprop = of_get_flat_dt_prop(node, "linux,crashkernel-size", NULL);
437 if (lprop)
438 crashk_res.end = crashk_res.start + *lprop - 1;
439#endif
440
441 /* break now */
442 return 1;
443}
444
445/*
446 * Compare the range against max mem limit and update
447 * size if it cross the limit.
448 */
449
450#ifdef CONFIG_SPARSEMEM
451static bool validate_mem_limit(u64 base, u64 *size)
452{
453 u64 max_mem = 1UL << (MAX_PHYSMEM_BITS);
454
455 if (base >= max_mem)
456 return false;
457 if ((base + *size) > max_mem)
458 *size = max_mem - base;
459 return true;
460}
461#else
462static bool validate_mem_limit(u64 base, u64 *size)
463{
464 return true;
465}
466#endif
467
468#ifdef CONFIG_PPC_PSERIES
469/*
470 * Interpret the ibm dynamic reconfiguration memory LMBs.
471 * This contains a list of memory blocks along with NUMA affinity
472 * information.
473 */
474static int __init early_init_drmem_lmb(struct drmem_lmb *lmb,
475 const __be32 **usm,
476 void *data)
477{
478 u64 base, size;
479 int is_kexec_kdump = 0, rngs;
480
481 base = lmb->base_addr;
482 size = drmem_lmb_size();
483 rngs = 1;
484
485 /*
486 * Skip this block if the reserved bit is set in flags
487 * or if the block is not assigned to this partition.
488 */
489 if ((lmb->flags & DRCONF_MEM_RESERVED) ||
490 !(lmb->flags & DRCONF_MEM_ASSIGNED))
491 return 0;
492
493 if (*usm)
494 is_kexec_kdump = 1;
495
496 if (is_kexec_kdump) {
497 /*
498 * For each memblock in ibm,dynamic-memory, a
499 * corresponding entry in linux,drconf-usable-memory
500 * property contains a counter 'p' followed by 'p'
501 * (base, size) duple. Now read the counter from
502 * linux,drconf-usable-memory property
503 */
504 rngs = dt_mem_next_cell(dt_root_size_cells, usm);
505 if (!rngs) /* there are no (base, size) duple */
506 return 0;
507 }
508
509 do {
510 if (is_kexec_kdump) {
511 base = dt_mem_next_cell(dt_root_addr_cells, usm);
512 size = dt_mem_next_cell(dt_root_size_cells, usm);
513 }
514
515 if (iommu_is_off) {
516 if (base >= 0x80000000ul)
517 continue;
518 if ((base + size) > 0x80000000ul)
519 size = 0x80000000ul - base;
520 }
521
522 if (!validate_mem_limit(base, &size))
523 continue;
524
525 DBG("Adding: %llx -> %llx\n", base, size);
526 memblock_add(base, size);
527
528 if (lmb->flags & DRCONF_MEM_HOTREMOVABLE)
529 memblock_mark_hotplug(base, size);
530 } while (--rngs);
531
532 return 0;
533}
534#endif /* CONFIG_PPC_PSERIES */
535
536static int __init early_init_dt_scan_memory_ppc(unsigned long node,
537 const char *uname,
538 int depth, void *data)
539{
540#ifdef CONFIG_PPC_PSERIES
541 if (depth == 1 &&
542 strcmp(uname, "ibm,dynamic-reconfiguration-memory") == 0) {
543 walk_drmem_lmbs_early(node, NULL, early_init_drmem_lmb);
544 return 0;
545 }
546#endif
547
548 return early_init_dt_scan_memory(node, uname, depth, data);
549}
550
551/*
552 * For a relocatable kernel, we need to get the memstart_addr first,
553 * then use it to calculate the virtual kernel start address. This has
554 * to happen at a very early stage (before machine_init). In this case,
555 * we just want to get the memstart_address and would not like to mess the
556 * memblock at this stage. So introduce a variable to skip the memblock_add()
557 * for this reason.
558 */
559#ifdef CONFIG_RELOCATABLE
560static int add_mem_to_memblock = 1;
561#else
562#define add_mem_to_memblock 1
563#endif
564
565void __init early_init_dt_add_memory_arch(u64 base, u64 size)
566{
567#ifdef CONFIG_PPC64
568 if (iommu_is_off) {
569 if (base >= 0x80000000ul)
570 return;
571 if ((base + size) > 0x80000000ul)
572 size = 0x80000000ul - base;
573 }
574#endif
575 /* Keep track of the beginning of memory -and- the size of
576 * the very first block in the device-tree as it represents
577 * the RMA on ppc64 server
578 */
579 if (base < memstart_addr) {
580 memstart_addr = base;
581 first_memblock_size = size;
582 }
583
584 /* Add the chunk to the MEMBLOCK list */
585 if (add_mem_to_memblock) {
586 if (validate_mem_limit(base, &size))
587 memblock_add(base, size);
588 }
589}
590
591static void __init early_reserve_mem_dt(void)
592{
593 unsigned long i, dt_root;
594 int len;
595 const __be32 *prop;
596
597 early_init_fdt_reserve_self();
598 early_init_fdt_scan_reserved_mem();
599
600 dt_root = of_get_flat_dt_root();
601
602 prop = of_get_flat_dt_prop(dt_root, "reserved-ranges", &len);
603
604 if (!prop)
605 return;
606
607 DBG("Found new-style reserved-ranges\n");
608
609 /* Each reserved range is an (address,size) pair, 2 cells each,
610 * totalling 4 cells per range. */
611 for (i = 0; i < len / (sizeof(*prop) * 4); i++) {
612 u64 base, size;
613
614 base = of_read_number(prop + (i * 4) + 0, 2);
615 size = of_read_number(prop + (i * 4) + 2, 2);
616
617 if (size) {
618 DBG("reserving: %llx -> %llx\n", base, size);
619 memblock_reserve(base, size);
620 }
621 }
622}
623
624static void __init early_reserve_mem(void)
625{
626 __be64 *reserve_map;
627
628 reserve_map = (__be64 *)(((unsigned long)initial_boot_params) +
629 fdt_off_mem_rsvmap(initial_boot_params));
630
631 /* Look for the new "reserved-regions" property in the DT */
632 early_reserve_mem_dt();
633
634#ifdef CONFIG_BLK_DEV_INITRD
635 /* Then reserve the initrd, if any */
636 if (initrd_start && (initrd_end > initrd_start)) {
637 memblock_reserve(ALIGN_DOWN(__pa(initrd_start), PAGE_SIZE),
638 ALIGN(initrd_end, PAGE_SIZE) -
639 ALIGN_DOWN(initrd_start, PAGE_SIZE));
640 }
641#endif /* CONFIG_BLK_DEV_INITRD */
642
643#ifdef CONFIG_PPC32
644 /*
645 * Handle the case where we might be booting from an old kexec
646 * image that setup the mem_rsvmap as pairs of 32-bit values
647 */
648 if (be64_to_cpup(reserve_map) > 0xffffffffull) {
649 u32 base_32, size_32;
650 __be32 *reserve_map_32 = (__be32 *)reserve_map;
651
652 DBG("Found old 32-bit reserve map\n");
653
654 while (1) {
655 base_32 = be32_to_cpup(reserve_map_32++);
656 size_32 = be32_to_cpup(reserve_map_32++);
657 if (size_32 == 0)
658 break;
659 DBG("reserving: %x -> %x\n", base_32, size_32);
660 memblock_reserve(base_32, size_32);
661 }
662 return;
663 }
664#endif
665}
666
667#ifdef CONFIG_PPC_TRANSACTIONAL_MEM
668static bool tm_disabled __initdata;
669
670static int __init parse_ppc_tm(char *str)
671{
672 bool res;
673
674 if (kstrtobool(str, &res))
675 return -EINVAL;
676
677 tm_disabled = !res;
678
679 return 0;
680}
681early_param("ppc_tm", parse_ppc_tm);
682
683static void __init tm_init(void)
684{
685 if (tm_disabled) {
686 pr_info("Disabling hardware transactional memory (HTM)\n");
687 cur_cpu_spec->cpu_user_features2 &=
688 ~(PPC_FEATURE2_HTM_NOSC | PPC_FEATURE2_HTM);
689 cur_cpu_spec->cpu_features &= ~CPU_FTR_TM;
690 return;
691 }
692
693 pnv_tm_init();
694}
695#else
696static void tm_init(void) { }
697#endif /* CONFIG_PPC_TRANSACTIONAL_MEM */
698
699#ifdef CONFIG_PPC64
700static void __init save_fscr_to_task(void)
701{
702 /*
703 * Ensure the init_task (pid 0, aka swapper) uses the value of FSCR we
704 * have configured via the device tree features or via __init_FSCR().
705 * That value will then be propagated to pid 1 (init) and all future
706 * processes.
707 */
708 if (early_cpu_has_feature(CPU_FTR_ARCH_207S))
709 init_task.thread.fscr = mfspr(SPRN_FSCR);
710}
711#else
712static inline void save_fscr_to_task(void) {}
713#endif
714
715
716void __init early_init_devtree(void *params)
717{
718 phys_addr_t limit;
719
720 DBG(" -> early_init_devtree(%px)\n", params);
721
722 /* Too early to BUG_ON(), do it by hand */
723 if (!early_init_dt_verify(params))
724 panic("BUG: Failed verifying flat device tree, bad version?");
725
726#ifdef CONFIG_PPC_RTAS
727 /* Some machines might need RTAS info for debugging, grab it now. */
728 of_scan_flat_dt(early_init_dt_scan_rtas, NULL);
729#endif
730
731#ifdef CONFIG_PPC_POWERNV
732 /* Some machines might need OPAL info for debugging, grab it now. */
733 of_scan_flat_dt(early_init_dt_scan_opal, NULL);
734
735 /* Scan tree for ultravisor feature */
736 of_scan_flat_dt(early_init_dt_scan_ultravisor, NULL);
737#endif
738
739#if defined(CONFIG_FA_DUMP) || defined(CONFIG_PRESERVE_FA_DUMP)
740 /* scan tree to see if dump is active during last boot */
741 of_scan_flat_dt(early_init_dt_scan_fw_dump, NULL);
742#endif
743
744 /* Retrieve various informations from the /chosen node of the
745 * device-tree, including the platform type, initrd location and
746 * size, TCE reserve, and more ...
747 */
748 of_scan_flat_dt(early_init_dt_scan_chosen_ppc, boot_command_line);
749
750 /* Scan memory nodes and rebuild MEMBLOCKs */
751 of_scan_flat_dt(early_init_dt_scan_root, NULL);
752 of_scan_flat_dt(early_init_dt_scan_memory_ppc, NULL);
753
754 parse_early_param();
755
756 /* make sure we've parsed cmdline for mem= before this */
757 if (memory_limit)
758 first_memblock_size = min_t(u64, first_memblock_size, memory_limit);
759 setup_initial_memory_limit(memstart_addr, first_memblock_size);
760 /* Reserve MEMBLOCK regions used by kernel, initrd, dt, etc... */
761 memblock_reserve(PHYSICAL_START, __pa(_end) - PHYSICAL_START);
762 /* If relocatable, reserve first 32k for interrupt vectors etc. */
763 if (PHYSICAL_START > MEMORY_START)
764 memblock_reserve(MEMORY_START, 0x8000);
765 reserve_kdump_trampoline();
766#if defined(CONFIG_FA_DUMP) || defined(CONFIG_PRESERVE_FA_DUMP)
767 /*
768 * If we fail to reserve memory for firmware-assisted dump then
769 * fallback to kexec based kdump.
770 */
771 if (fadump_reserve_mem() == 0)
772#endif
773 reserve_crashkernel();
774 early_reserve_mem();
775
776 /* Ensure that total memory size is page-aligned. */
777 limit = ALIGN(memory_limit ?: memblock_phys_mem_size(), PAGE_SIZE);
778 memblock_enforce_memory_limit(limit);
779
780#if defined(CONFIG_PPC_BOOK3S_64) && defined(CONFIG_PPC_4K_PAGES)
781 if (!early_radix_enabled())
782 memblock_cap_memory_range(0, 1UL << (H_MAX_PHYSMEM_BITS));
783#endif
784
785 memblock_allow_resize();
786 memblock_dump_all();
787
788 DBG("Phys. mem: %llx\n", (unsigned long long)memblock_phys_mem_size());
789
790 /* We may need to relocate the flat tree, do it now.
791 * FIXME .. and the initrd too? */
792 move_device_tree();
793
794 allocate_paca_ptrs();
795
796 DBG("Scanning CPUs ...\n");
797
798 dt_cpu_ftrs_scan();
799
800 /* Retrieve CPU related informations from the flat tree
801 * (altivec support, boot CPU ID, ...)
802 */
803 of_scan_flat_dt(early_init_dt_scan_cpus, NULL);
804 if (boot_cpuid < 0) {
805 printk("Failed to identify boot CPU !\n");
806 BUG();
807 }
808
809 save_fscr_to_task();
810
811#if defined(CONFIG_SMP) && defined(CONFIG_PPC64)
812 /* We'll later wait for secondaries to check in; there are
813 * NCPUS-1 non-boot CPUs :-)
814 */
815 spinning_secondaries = boot_cpu_count - 1;
816#endif
817
818 mmu_early_init_devtree();
819
820#ifdef CONFIG_PPC_POWERNV
821 /* Scan and build the list of machine check recoverable ranges */
822 of_scan_flat_dt(early_init_dt_scan_recoverable_ranges, NULL);
823#endif
824 epapr_paravirt_early_init();
825
826 /* Now try to figure out if we are running on LPAR and so on */
827 pseries_probe_fw_features();
828
829 /*
830 * Initialize pkey features and default AMR/IAMR values
831 */
832 pkey_early_init_devtree();
833
834#ifdef CONFIG_PPC_PS3
835 /* Identify PS3 firmware */
836 if (of_flat_dt_is_compatible(of_get_flat_dt_root(), "sony,ps3"))
837 powerpc_firmware_features |= FW_FEATURE_PS3_POSSIBLE;
838#endif
839
840 tm_init();
841
842 DBG(" <- early_init_devtree()\n");
843}
844
845#ifdef CONFIG_RELOCATABLE
846/*
847 * This function run before early_init_devtree, so we have to init
848 * initial_boot_params.
849 */
850void __init early_get_first_memblock_info(void *params, phys_addr_t *size)
851{
852 /* Setup flat device-tree pointer */
853 initial_boot_params = params;
854
855 /*
856 * Scan the memory nodes and set add_mem_to_memblock to 0 to avoid
857 * mess the memblock.
858 */
859 add_mem_to_memblock = 0;
860 of_scan_flat_dt(early_init_dt_scan_root, NULL);
861 of_scan_flat_dt(early_init_dt_scan_memory_ppc, NULL);
862 add_mem_to_memblock = 1;
863
864 if (size)
865 *size = first_memblock_size;
866}
867#endif
868
869/*******
870 *
871 * New implementation of the OF "find" APIs, return a refcounted
872 * object, call of_node_put() when done. The device tree and list
873 * are protected by a rw_lock.
874 *
875 * Note that property management will need some locking as well,
876 * this isn't dealt with yet.
877 *
878 *******/
879
880/**
881 * of_get_ibm_chip_id - Returns the IBM "chip-id" of a device
882 * @np: device node of the device
883 *
884 * This looks for a property "ibm,chip-id" in the node or any
885 * of its parents and returns its content, or -1 if it cannot
886 * be found.
887 */
888int of_get_ibm_chip_id(struct device_node *np)
889{
890 of_node_get(np);
891 while (np) {
892 u32 chip_id;
893
894 /*
895 * Skiboot may produce memory nodes that contain more than one
896 * cell in chip-id, we only read the first one here.
897 */
898 if (!of_property_read_u32(np, "ibm,chip-id", &chip_id)) {
899 of_node_put(np);
900 return chip_id;
901 }
902
903 np = of_get_next_parent(np);
904 }
905 return -1;
906}
907EXPORT_SYMBOL(of_get_ibm_chip_id);
908
909/**
910 * cpu_to_chip_id - Return the cpus chip-id
911 * @cpu: The logical cpu number.
912 *
913 * Return the value of the ibm,chip-id property corresponding to the given
914 * logical cpu number. If the chip-id can not be found, returns -1.
915 */
916int cpu_to_chip_id(int cpu)
917{
918 struct device_node *np;
919 int ret = -1, idx;
920
921 idx = cpu / threads_per_core;
922 if (chip_id_lookup_table && chip_id_lookup_table[idx] != -1)
923 return chip_id_lookup_table[idx];
924
925 np = of_get_cpu_node(cpu, NULL);
926 if (np) {
927 ret = of_get_ibm_chip_id(np);
928 of_node_put(np);
929
930 if (chip_id_lookup_table)
931 chip_id_lookup_table[idx] = ret;
932 }
933
934 return ret;
935}
936EXPORT_SYMBOL(cpu_to_chip_id);
937
938bool arch_match_cpu_phys_id(int cpu, u64 phys_id)
939{
940#ifdef CONFIG_SMP
941 /*
942 * Early firmware scanning must use this rather than
943 * get_hard_smp_processor_id because we don't have pacas allocated
944 * until memory topology is discovered.
945 */
946 if (cpu_to_phys_id != NULL)
947 return (int)phys_id == cpu_to_phys_id[cpu];
948#endif
949
950 return (int)phys_id == get_hard_smp_processor_id(cpu);
951}
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/module.h>
31#include <linux/kexec.h>
32#include <linux/debugfs.h>
33#include <linux/irq.h>
34#include <linux/memblock.h>
35
36#include <asm/prom.h>
37#include <asm/rtas.h>
38#include <asm/page.h>
39#include <asm/processor.h>
40#include <asm/irq.h>
41#include <asm/io.h>
42#include <asm/kdump.h>
43#include <asm/smp.h>
44#include <asm/system.h>
45#include <asm/mmu.h>
46#include <asm/paca.h>
47#include <asm/pgtable.h>
48#include <asm/pci.h>
49#include <asm/iommu.h>
50#include <asm/btext.h>
51#include <asm/sections.h>
52#include <asm/machdep.h>
53#include <asm/pSeries_reconfig.h>
54#include <asm/pci-bridge.h>
55#include <asm/phyp_dump.h>
56#include <asm/kexec.h>
57#include <mm/mmu_decl.h>
58
59#ifdef DEBUG
60#define DBG(fmt...) printk(KERN_ERR fmt)
61#else
62#define DBG(fmt...)
63#endif
64
65#ifdef CONFIG_PPC64
66int __initdata iommu_is_off;
67int __initdata iommu_force_on;
68unsigned long tce_alloc_start, tce_alloc_end;
69u64 ppc64_rma_size;
70#endif
71static phys_addr_t first_memblock_size;
72static int __initdata boot_cpu_count;
73
74static int __init early_parse_mem(char *p)
75{
76 if (!p)
77 return 1;
78
79 memory_limit = PAGE_ALIGN(memparse(p, &p));
80 DBG("memory limit = 0x%llx\n", (unsigned long long)memory_limit);
81
82 return 0;
83}
84early_param("mem", early_parse_mem);
85
86/*
87 * overlaps_initrd - check for overlap with page aligned extension of
88 * initrd.
89 */
90static inline int overlaps_initrd(unsigned long start, unsigned long size)
91{
92#ifdef CONFIG_BLK_DEV_INITRD
93 if (!initrd_start)
94 return 0;
95
96 return (start + size) > _ALIGN_DOWN(initrd_start, PAGE_SIZE) &&
97 start <= _ALIGN_UP(initrd_end, PAGE_SIZE);
98#else
99 return 0;
100#endif
101}
102
103/**
104 * move_device_tree - move tree to an unused area, if needed.
105 *
106 * The device tree may be allocated beyond our memory limit, or inside the
107 * crash kernel region for kdump, or within the page aligned range of initrd.
108 * If so, move it out of the way.
109 */
110static void __init move_device_tree(void)
111{
112 unsigned long start, size;
113 void *p;
114
115 DBG("-> move_device_tree\n");
116
117 start = __pa(initial_boot_params);
118 size = be32_to_cpu(initial_boot_params->totalsize);
119
120 if ((memory_limit && (start + size) > PHYSICAL_START + memory_limit) ||
121 overlaps_crashkernel(start, size) ||
122 overlaps_initrd(start, size)) {
123 p = __va(memblock_alloc(size, PAGE_SIZE));
124 memcpy(p, initial_boot_params, size);
125 initial_boot_params = (struct boot_param_header *)p;
126 DBG("Moved device tree to 0x%p\n", p);
127 }
128
129 DBG("<- move_device_tree\n");
130}
131
132/*
133 * ibm,pa-features is a per-cpu property that contains a string of
134 * attribute descriptors, each of which has a 2 byte header plus up
135 * to 254 bytes worth of processor attribute bits. First header
136 * byte specifies the number of bytes following the header.
137 * Second header byte is an "attribute-specifier" type, of which
138 * zero is the only currently-defined value.
139 * Implementation: Pass in the byte and bit offset for the feature
140 * that we are interested in. The function will return -1 if the
141 * pa-features property is missing, or a 1/0 to indicate if the feature
142 * is supported/not supported. Note that the bit numbers are
143 * big-endian to match the definition in PAPR.
144 */
145static struct ibm_pa_feature {
146 unsigned long cpu_features; /* CPU_FTR_xxx bit */
147 unsigned long mmu_features; /* MMU_FTR_xxx bit */
148 unsigned int cpu_user_ftrs; /* PPC_FEATURE_xxx bit */
149 unsigned char pabyte; /* byte number in ibm,pa-features */
150 unsigned char pabit; /* bit number (big-endian) */
151 unsigned char invert; /* if 1, pa bit set => clear feature */
152} ibm_pa_features[] __initdata = {
153 {0, 0, PPC_FEATURE_HAS_MMU, 0, 0, 0},
154 {0, 0, PPC_FEATURE_HAS_FPU, 0, 1, 0},
155 {0, MMU_FTR_SLB, 0, 0, 2, 0},
156 {CPU_FTR_CTRL, 0, 0, 0, 3, 0},
157 {CPU_FTR_NOEXECUTE, 0, 0, 0, 6, 0},
158 {CPU_FTR_NODSISRALIGN, 0, 0, 1, 1, 1},
159 {0, MMU_FTR_CI_LARGE_PAGE, 0, 1, 2, 0},
160 {CPU_FTR_REAL_LE, PPC_FEATURE_TRUE_LE, 5, 0, 0},
161};
162
163static void __init scan_features(unsigned long node, unsigned char *ftrs,
164 unsigned long tablelen,
165 struct ibm_pa_feature *fp,
166 unsigned long ft_size)
167{
168 unsigned long i, len, bit;
169
170 /* find descriptor with type == 0 */
171 for (;;) {
172 if (tablelen < 3)
173 return;
174 len = 2 + ftrs[0];
175 if (tablelen < len)
176 return; /* descriptor 0 not found */
177 if (ftrs[1] == 0)
178 break;
179 tablelen -= len;
180 ftrs += len;
181 }
182
183 /* loop over bits we know about */
184 for (i = 0; i < ft_size; ++i, ++fp) {
185 if (fp->pabyte >= ftrs[0])
186 continue;
187 bit = (ftrs[2 + fp->pabyte] >> (7 - fp->pabit)) & 1;
188 if (bit ^ fp->invert) {
189 cur_cpu_spec->cpu_features |= fp->cpu_features;
190 cur_cpu_spec->cpu_user_features |= fp->cpu_user_ftrs;
191 cur_cpu_spec->mmu_features |= fp->mmu_features;
192 } else {
193 cur_cpu_spec->cpu_features &= ~fp->cpu_features;
194 cur_cpu_spec->cpu_user_features &= ~fp->cpu_user_ftrs;
195 cur_cpu_spec->mmu_features &= ~fp->mmu_features;
196 }
197 }
198}
199
200static void __init check_cpu_pa_features(unsigned long node)
201{
202 unsigned char *pa_ftrs;
203 unsigned long tablelen;
204
205 pa_ftrs = of_get_flat_dt_prop(node, "ibm,pa-features", &tablelen);
206 if (pa_ftrs == NULL)
207 return;
208
209 scan_features(node, pa_ftrs, tablelen,
210 ibm_pa_features, ARRAY_SIZE(ibm_pa_features));
211}
212
213#ifdef CONFIG_PPC_STD_MMU_64
214static void __init check_cpu_slb_size(unsigned long node)
215{
216 u32 *slb_size_ptr;
217
218 slb_size_ptr = of_get_flat_dt_prop(node, "slb-size", NULL);
219 if (slb_size_ptr != NULL) {
220 mmu_slb_size = *slb_size_ptr;
221 return;
222 }
223 slb_size_ptr = of_get_flat_dt_prop(node, "ibm,slb-size", NULL);
224 if (slb_size_ptr != NULL) {
225 mmu_slb_size = *slb_size_ptr;
226 }
227}
228#else
229#define check_cpu_slb_size(node) do { } while(0)
230#endif
231
232static struct feature_property {
233 const char *name;
234 u32 min_value;
235 unsigned long cpu_feature;
236 unsigned long cpu_user_ftr;
237} feature_properties[] __initdata = {
238#ifdef CONFIG_ALTIVEC
239 {"altivec", 0, CPU_FTR_ALTIVEC, PPC_FEATURE_HAS_ALTIVEC},
240 {"ibm,vmx", 1, CPU_FTR_ALTIVEC, PPC_FEATURE_HAS_ALTIVEC},
241#endif /* CONFIG_ALTIVEC */
242#ifdef CONFIG_VSX
243 /* Yes, this _really_ is ibm,vmx == 2 to enable VSX */
244 {"ibm,vmx", 2, CPU_FTR_VSX, PPC_FEATURE_HAS_VSX},
245#endif /* CONFIG_VSX */
246#ifdef CONFIG_PPC64
247 {"ibm,dfp", 1, 0, PPC_FEATURE_HAS_DFP},
248 {"ibm,purr", 1, CPU_FTR_PURR, 0},
249 {"ibm,spurr", 1, CPU_FTR_SPURR, 0},
250#endif /* CONFIG_PPC64 */
251};
252
253#if defined(CONFIG_44x) && defined(CONFIG_PPC_FPU)
254static inline void identical_pvr_fixup(unsigned long node)
255{
256 unsigned int pvr;
257 char *model = of_get_flat_dt_prop(node, "model", NULL);
258
259 /*
260 * Since 440GR(x)/440EP(x) processors have the same pvr,
261 * we check the node path and set bit 28 in the cur_cpu_spec
262 * pvr for EP(x) processor version. This bit is always 0 in
263 * the "real" pvr. Then we call identify_cpu again with
264 * the new logical pvr to enable FPU support.
265 */
266 if (model && strstr(model, "440EP")) {
267 pvr = cur_cpu_spec->pvr_value | 0x8;
268 identify_cpu(0, pvr);
269 DBG("Using logical pvr %x for %s\n", pvr, model);
270 }
271}
272#else
273#define identical_pvr_fixup(node) do { } while(0)
274#endif
275
276static void __init check_cpu_feature_properties(unsigned long node)
277{
278 unsigned long i;
279 struct feature_property *fp = feature_properties;
280 const u32 *prop;
281
282 for (i = 0; i < ARRAY_SIZE(feature_properties); ++i, ++fp) {
283 prop = of_get_flat_dt_prop(node, fp->name, NULL);
284 if (prop && *prop >= fp->min_value) {
285 cur_cpu_spec->cpu_features |= fp->cpu_feature;
286 cur_cpu_spec->cpu_user_features |= fp->cpu_user_ftr;
287 }
288 }
289}
290
291static int __init early_init_dt_scan_cpus(unsigned long node,
292 const char *uname, int depth,
293 void *data)
294{
295 char *type = of_get_flat_dt_prop(node, "device_type", NULL);
296 const u32 *prop;
297 const u32 *intserv;
298 int i, nthreads;
299 unsigned long len;
300 int found = -1;
301 int found_thread = 0;
302
303 /* We are scanning "cpu" nodes only */
304 if (type == NULL || strcmp(type, "cpu") != 0)
305 return 0;
306
307 /* Get physical cpuid */
308 intserv = of_get_flat_dt_prop(node, "ibm,ppc-interrupt-server#s", &len);
309 if (intserv) {
310 nthreads = len / sizeof(int);
311 } else {
312 intserv = of_get_flat_dt_prop(node, "reg", NULL);
313 nthreads = 1;
314 }
315
316 /*
317 * Now see if any of these threads match our boot cpu.
318 * NOTE: This must match the parsing done in smp_setup_cpu_maps.
319 */
320 for (i = 0; i < nthreads; i++) {
321 /*
322 * version 2 of the kexec param format adds the phys cpuid of
323 * booted proc.
324 */
325 if (initial_boot_params->version >= 2) {
326 if (intserv[i] == initial_boot_params->boot_cpuid_phys) {
327 found = boot_cpu_count;
328 found_thread = i;
329 }
330 } else {
331 /*
332 * Check if it's the boot-cpu, set it's hw index now,
333 * unfortunately this format did not support booting
334 * off secondary threads.
335 */
336 if (of_get_flat_dt_prop(node,
337 "linux,boot-cpu", NULL) != NULL)
338 found = boot_cpu_count;
339 }
340#ifdef CONFIG_SMP
341 /* logical cpu id is always 0 on UP kernels */
342 boot_cpu_count++;
343#endif
344 }
345
346 if (found >= 0) {
347 DBG("boot cpu: logical %d physical %d\n", found,
348 intserv[found_thread]);
349 boot_cpuid = found;
350 set_hard_smp_processor_id(found, intserv[found_thread]);
351
352 /*
353 * PAPR defines "logical" PVR values for cpus that
354 * meet various levels of the architecture:
355 * 0x0f000001 Architecture version 2.04
356 * 0x0f000002 Architecture version 2.05
357 * If the cpu-version property in the cpu node contains
358 * such a value, we call identify_cpu again with the
359 * logical PVR value in order to use the cpu feature
360 * bits appropriate for the architecture level.
361 *
362 * A POWER6 partition in "POWER6 architected" mode
363 * uses the 0x0f000002 PVR value; in POWER5+ mode
364 * it uses 0x0f000001.
365 */
366 prop = of_get_flat_dt_prop(node, "cpu-version", NULL);
367 if (prop && (*prop & 0xff000000) == 0x0f000000)
368 identify_cpu(0, *prop);
369
370 identical_pvr_fixup(node);
371 }
372
373 check_cpu_feature_properties(node);
374 check_cpu_pa_features(node);
375 check_cpu_slb_size(node);
376
377#ifdef CONFIG_PPC_PSERIES
378 if (nthreads > 1)
379 cur_cpu_spec->cpu_features |= CPU_FTR_SMT;
380 else
381 cur_cpu_spec->cpu_features &= ~CPU_FTR_SMT;
382#endif
383
384 return 0;
385}
386
387int __init early_init_dt_scan_chosen_ppc(unsigned long node, const char *uname,
388 int depth, void *data)
389{
390 unsigned long *lprop;
391
392 /* Use common scan routine to determine if this is the chosen node */
393 if (early_init_dt_scan_chosen(node, uname, depth, data) == 0)
394 return 0;
395
396#ifdef CONFIG_PPC64
397 /* check if iommu is forced on or off */
398 if (of_get_flat_dt_prop(node, "linux,iommu-off", NULL) != NULL)
399 iommu_is_off = 1;
400 if (of_get_flat_dt_prop(node, "linux,iommu-force-on", NULL) != NULL)
401 iommu_force_on = 1;
402#endif
403
404 /* mem=x on the command line is the preferred mechanism */
405 lprop = of_get_flat_dt_prop(node, "linux,memory-limit", NULL);
406 if (lprop)
407 memory_limit = *lprop;
408
409#ifdef CONFIG_PPC64
410 lprop = of_get_flat_dt_prop(node, "linux,tce-alloc-start", NULL);
411 if (lprop)
412 tce_alloc_start = *lprop;
413 lprop = of_get_flat_dt_prop(node, "linux,tce-alloc-end", NULL);
414 if (lprop)
415 tce_alloc_end = *lprop;
416#endif
417
418#ifdef CONFIG_KEXEC
419 lprop = of_get_flat_dt_prop(node, "linux,crashkernel-base", NULL);
420 if (lprop)
421 crashk_res.start = *lprop;
422
423 lprop = of_get_flat_dt_prop(node, "linux,crashkernel-size", NULL);
424 if (lprop)
425 crashk_res.end = crashk_res.start + *lprop - 1;
426#endif
427
428 /* break now */
429 return 1;
430}
431
432#ifdef CONFIG_PPC_PSERIES
433/*
434 * Interpret the ibm,dynamic-memory property in the
435 * /ibm,dynamic-reconfiguration-memory node.
436 * This contains a list of memory blocks along with NUMA affinity
437 * information.
438 */
439static int __init early_init_dt_scan_drconf_memory(unsigned long node)
440{
441 __be32 *dm, *ls, *usm;
442 unsigned long l, n, flags;
443 u64 base, size, memblock_size;
444 unsigned int is_kexec_kdump = 0, rngs;
445
446 ls = of_get_flat_dt_prop(node, "ibm,lmb-size", &l);
447 if (ls == NULL || l < dt_root_size_cells * sizeof(__be32))
448 return 0;
449 memblock_size = dt_mem_next_cell(dt_root_size_cells, &ls);
450
451 dm = of_get_flat_dt_prop(node, "ibm,dynamic-memory", &l);
452 if (dm == NULL || l < sizeof(__be32))
453 return 0;
454
455 n = *dm++; /* number of entries */
456 if (l < (n * (dt_root_addr_cells + 4) + 1) * sizeof(__be32))
457 return 0;
458
459 /* check if this is a kexec/kdump kernel. */
460 usm = of_get_flat_dt_prop(node, "linux,drconf-usable-memory",
461 &l);
462 if (usm != NULL)
463 is_kexec_kdump = 1;
464
465 for (; n != 0; --n) {
466 base = dt_mem_next_cell(dt_root_addr_cells, &dm);
467 flags = dm[3];
468 /* skip DRC index, pad, assoc. list index, flags */
469 dm += 4;
470 /* skip this block if the reserved bit is set in flags (0x80)
471 or if the block is not assigned to this partition (0x8) */
472 if ((flags & 0x80) || !(flags & 0x8))
473 continue;
474 size = memblock_size;
475 rngs = 1;
476 if (is_kexec_kdump) {
477 /*
478 * For each memblock in ibm,dynamic-memory, a corresponding
479 * entry in linux,drconf-usable-memory property contains
480 * a counter 'p' followed by 'p' (base, size) duple.
481 * Now read the counter from
482 * linux,drconf-usable-memory property
483 */
484 rngs = dt_mem_next_cell(dt_root_size_cells, &usm);
485 if (!rngs) /* there are no (base, size) duple */
486 continue;
487 }
488 do {
489 if (is_kexec_kdump) {
490 base = dt_mem_next_cell(dt_root_addr_cells,
491 &usm);
492 size = dt_mem_next_cell(dt_root_size_cells,
493 &usm);
494 }
495 if (iommu_is_off) {
496 if (base >= 0x80000000ul)
497 continue;
498 if ((base + size) > 0x80000000ul)
499 size = 0x80000000ul - base;
500 }
501 memblock_add(base, size);
502 } while (--rngs);
503 }
504 memblock_dump_all();
505 return 0;
506}
507#else
508#define early_init_dt_scan_drconf_memory(node) 0
509#endif /* CONFIG_PPC_PSERIES */
510
511static int __init early_init_dt_scan_memory_ppc(unsigned long node,
512 const char *uname,
513 int depth, void *data)
514{
515 if (depth == 1 &&
516 strcmp(uname, "ibm,dynamic-reconfiguration-memory") == 0)
517 return early_init_dt_scan_drconf_memory(node);
518
519 return early_init_dt_scan_memory(node, uname, depth, data);
520}
521
522void __init early_init_dt_add_memory_arch(u64 base, u64 size)
523{
524#ifdef CONFIG_PPC64
525 if (iommu_is_off) {
526 if (base >= 0x80000000ul)
527 return;
528 if ((base + size) > 0x80000000ul)
529 size = 0x80000000ul - base;
530 }
531#endif
532 /* Keep track of the beginning of memory -and- the size of
533 * the very first block in the device-tree as it represents
534 * the RMA on ppc64 server
535 */
536 if (base < memstart_addr) {
537 memstart_addr = base;
538 first_memblock_size = size;
539 }
540
541 /* Add the chunk to the MEMBLOCK list */
542 memblock_add(base, size);
543}
544
545void * __init early_init_dt_alloc_memory_arch(u64 size, u64 align)
546{
547 return __va(memblock_alloc(size, align));
548}
549
550#ifdef CONFIG_BLK_DEV_INITRD
551void __init early_init_dt_setup_initrd_arch(unsigned long start,
552 unsigned long end)
553{
554 initrd_start = (unsigned long)__va(start);
555 initrd_end = (unsigned long)__va(end);
556 initrd_below_start_ok = 1;
557}
558#endif
559
560static void __init early_reserve_mem(void)
561{
562 u64 base, size;
563 u64 *reserve_map;
564 unsigned long self_base;
565 unsigned long self_size;
566
567 reserve_map = (u64 *)(((unsigned long)initial_boot_params) +
568 initial_boot_params->off_mem_rsvmap);
569
570 /* before we do anything, lets reserve the dt blob */
571 self_base = __pa((unsigned long)initial_boot_params);
572 self_size = initial_boot_params->totalsize;
573 memblock_reserve(self_base, self_size);
574
575#ifdef CONFIG_BLK_DEV_INITRD
576 /* then reserve the initrd, if any */
577 if (initrd_start && (initrd_end > initrd_start))
578 memblock_reserve(_ALIGN_DOWN(__pa(initrd_start), PAGE_SIZE),
579 _ALIGN_UP(initrd_end, PAGE_SIZE) -
580 _ALIGN_DOWN(initrd_start, PAGE_SIZE));
581#endif /* CONFIG_BLK_DEV_INITRD */
582
583#ifdef CONFIG_PPC32
584 /*
585 * Handle the case where we might be booting from an old kexec
586 * image that setup the mem_rsvmap as pairs of 32-bit values
587 */
588 if (*reserve_map > 0xffffffffull) {
589 u32 base_32, size_32;
590 u32 *reserve_map_32 = (u32 *)reserve_map;
591
592 while (1) {
593 base_32 = *(reserve_map_32++);
594 size_32 = *(reserve_map_32++);
595 if (size_32 == 0)
596 break;
597 /* skip if the reservation is for the blob */
598 if (base_32 == self_base && size_32 == self_size)
599 continue;
600 DBG("reserving: %x -> %x\n", base_32, size_32);
601 memblock_reserve(base_32, size_32);
602 }
603 return;
604 }
605#endif
606 while (1) {
607 base = *(reserve_map++);
608 size = *(reserve_map++);
609 if (size == 0)
610 break;
611 DBG("reserving: %llx -> %llx\n", base, size);
612 memblock_reserve(base, size);
613 }
614}
615
616#ifdef CONFIG_PHYP_DUMP
617/**
618 * phyp_dump_calculate_reserve_size() - reserve variable boot area 5% or arg
619 *
620 * Function to find the largest size we need to reserve
621 * during early boot process.
622 *
623 * It either looks for boot param and returns that OR
624 * returns larger of 256 or 5% rounded down to multiples of 256MB.
625 *
626 */
627static inline unsigned long phyp_dump_calculate_reserve_size(void)
628{
629 unsigned long tmp;
630
631 if (phyp_dump_info->reserve_bootvar)
632 return phyp_dump_info->reserve_bootvar;
633
634 /* divide by 20 to get 5% of value */
635 tmp = memblock_end_of_DRAM();
636 do_div(tmp, 20);
637
638 /* round it down in multiples of 256 */
639 tmp = tmp & ~0x0FFFFFFFUL;
640
641 return (tmp > PHYP_DUMP_RMR_END ? tmp : PHYP_DUMP_RMR_END);
642}
643
644/**
645 * phyp_dump_reserve_mem() - reserve all not-yet-dumped mmemory
646 *
647 * This routine may reserve memory regions in the kernel only
648 * if the system is supported and a dump was taken in last
649 * boot instance or if the hardware is supported and the
650 * scratch area needs to be setup. In other instances it returns
651 * without reserving anything. The memory in case of dump being
652 * active is freed when the dump is collected (by userland tools).
653 */
654static void __init phyp_dump_reserve_mem(void)
655{
656 unsigned long base, size;
657 unsigned long variable_reserve_size;
658
659 if (!phyp_dump_info->phyp_dump_configured) {
660 printk(KERN_ERR "Phyp-dump not supported on this hardware\n");
661 return;
662 }
663
664 if (!phyp_dump_info->phyp_dump_at_boot) {
665 printk(KERN_INFO "Phyp-dump disabled at boot time\n");
666 return;
667 }
668
669 variable_reserve_size = phyp_dump_calculate_reserve_size();
670
671 if (phyp_dump_info->phyp_dump_is_active) {
672 /* Reserve *everything* above RMR.Area freed by userland tools*/
673 base = variable_reserve_size;
674 size = memblock_end_of_DRAM() - base;
675
676 /* XXX crashed_ram_end is wrong, since it may be beyond
677 * the memory_limit, it will need to be adjusted. */
678 memblock_reserve(base, size);
679
680 phyp_dump_info->init_reserve_start = base;
681 phyp_dump_info->init_reserve_size = size;
682 } else {
683 size = phyp_dump_info->cpu_state_size +
684 phyp_dump_info->hpte_region_size +
685 variable_reserve_size;
686 base = memblock_end_of_DRAM() - size;
687 memblock_reserve(base, size);
688 phyp_dump_info->init_reserve_start = base;
689 phyp_dump_info->init_reserve_size = size;
690 }
691}
692#else
693static inline void __init phyp_dump_reserve_mem(void) {}
694#endif /* CONFIG_PHYP_DUMP && CONFIG_PPC_RTAS */
695
696void __init early_init_devtree(void *params)
697{
698 phys_addr_t limit;
699
700 DBG(" -> early_init_devtree(%p)\n", params);
701
702 /* Setup flat device-tree pointer */
703 initial_boot_params = params;
704
705#ifdef CONFIG_PPC_RTAS
706 /* Some machines might need RTAS info for debugging, grab it now. */
707 of_scan_flat_dt(early_init_dt_scan_rtas, NULL);
708#endif
709
710#ifdef CONFIG_PHYP_DUMP
711 /* scan tree to see if dump occurred during last boot */
712 of_scan_flat_dt(early_init_dt_scan_phyp_dump, NULL);
713#endif
714
715 /* Retrieve various informations from the /chosen node of the
716 * device-tree, including the platform type, initrd location and
717 * size, TCE reserve, and more ...
718 */
719 of_scan_flat_dt(early_init_dt_scan_chosen_ppc, cmd_line);
720
721 /* Scan memory nodes and rebuild MEMBLOCKs */
722 memblock_init();
723
724 of_scan_flat_dt(early_init_dt_scan_root, NULL);
725 of_scan_flat_dt(early_init_dt_scan_memory_ppc, NULL);
726 setup_initial_memory_limit(memstart_addr, first_memblock_size);
727
728 /* Save command line for /proc/cmdline and then parse parameters */
729 strlcpy(boot_command_line, cmd_line, COMMAND_LINE_SIZE);
730 parse_early_param();
731
732 /* Reserve MEMBLOCK regions used by kernel, initrd, dt, etc... */
733 memblock_reserve(PHYSICAL_START, __pa(klimit) - PHYSICAL_START);
734 /* If relocatable, reserve first 32k for interrupt vectors etc. */
735 if (PHYSICAL_START > MEMORY_START)
736 memblock_reserve(MEMORY_START, 0x8000);
737 reserve_kdump_trampoline();
738 reserve_crashkernel();
739 early_reserve_mem();
740 phyp_dump_reserve_mem();
741
742 limit = memory_limit;
743 if (! limit) {
744 phys_addr_t memsize;
745
746 /* Ensure that total memory size is page-aligned, because
747 * otherwise mark_bootmem() gets upset. */
748 memblock_analyze();
749 memsize = memblock_phys_mem_size();
750 if ((memsize & PAGE_MASK) != memsize)
751 limit = memsize & PAGE_MASK;
752 }
753 memblock_enforce_memory_limit(limit);
754
755 memblock_analyze();
756 memblock_dump_all();
757
758 DBG("Phys. mem: %llx\n", memblock_phys_mem_size());
759
760 /* We may need to relocate the flat tree, do it now.
761 * FIXME .. and the initrd too? */
762 move_device_tree();
763
764 allocate_pacas();
765
766 DBG("Scanning CPUs ...\n");
767
768 /* Retrieve CPU related informations from the flat tree
769 * (altivec support, boot CPU ID, ...)
770 */
771 of_scan_flat_dt(early_init_dt_scan_cpus, NULL);
772
773#if defined(CONFIG_SMP) && defined(CONFIG_PPC64)
774 /* We'll later wait for secondaries to check in; there are
775 * NCPUS-1 non-boot CPUs :-)
776 */
777 spinning_secondaries = boot_cpu_count - 1;
778#endif
779
780 DBG(" <- early_init_devtree()\n");
781}
782
783/*******
784 *
785 * New implementation of the OF "find" APIs, return a refcounted
786 * object, call of_node_put() when done. The device tree and list
787 * are protected by a rw_lock.
788 *
789 * Note that property management will need some locking as well,
790 * this isn't dealt with yet.
791 *
792 *******/
793
794/**
795 * of_find_next_cache_node - Find a node's subsidiary cache
796 * @np: node of type "cpu" or "cache"
797 *
798 * Returns a node pointer with refcount incremented, use
799 * of_node_put() on it when done. Caller should hold a reference
800 * to np.
801 */
802struct device_node *of_find_next_cache_node(struct device_node *np)
803{
804 struct device_node *child;
805 const phandle *handle;
806
807 handle = of_get_property(np, "l2-cache", NULL);
808 if (!handle)
809 handle = of_get_property(np, "next-level-cache", NULL);
810
811 if (handle)
812 return of_find_node_by_phandle(*handle);
813
814 /* OF on pmac has nodes instead of properties named "l2-cache"
815 * beneath CPU nodes.
816 */
817 if (!strcmp(np->type, "cpu"))
818 for_each_child_of_node(np, child)
819 if (!strcmp(child->type, "cache"))
820 return child;
821
822 return NULL;
823}
824
825#ifdef CONFIG_PPC_PSERIES
826/*
827 * Fix up the uninitialized fields in a new device node:
828 * name, type and pci-specific fields
829 */
830
831static int of_finish_dynamic_node(struct device_node *node)
832{
833 struct device_node *parent = of_get_parent(node);
834 int err = 0;
835 const phandle *ibm_phandle;
836
837 node->name = of_get_property(node, "name", NULL);
838 node->type = of_get_property(node, "device_type", NULL);
839
840 if (!node->name)
841 node->name = "<NULL>";
842 if (!node->type)
843 node->type = "<NULL>";
844
845 if (!parent) {
846 err = -ENODEV;
847 goto out;
848 }
849
850 /* We don't support that function on PowerMac, at least
851 * not yet
852 */
853 if (machine_is(powermac))
854 return -ENODEV;
855
856 /* fix up new node's phandle field */
857 if ((ibm_phandle = of_get_property(node, "ibm,phandle", NULL)))
858 node->phandle = *ibm_phandle;
859
860out:
861 of_node_put(parent);
862 return err;
863}
864
865static int prom_reconfig_notifier(struct notifier_block *nb,
866 unsigned long action, void *node)
867{
868 int err;
869
870 switch (action) {
871 case PSERIES_RECONFIG_ADD:
872 err = of_finish_dynamic_node(node);
873 if (err < 0)
874 printk(KERN_ERR "finish_node returned %d\n", err);
875 break;
876 default:
877 err = 0;
878 break;
879 }
880 return notifier_from_errno(err);
881}
882
883static struct notifier_block prom_reconfig_nb = {
884 .notifier_call = prom_reconfig_notifier,
885 .priority = 10, /* This one needs to run first */
886};
887
888static int __init prom_reconfig_setup(void)
889{
890 return pSeries_reconfig_notifier_register(&prom_reconfig_nb);
891}
892__initcall(prom_reconfig_setup);
893#endif
894
895/* Find the device node for a given logical cpu number, also returns the cpu
896 * local thread number (index in ibm,interrupt-server#s) if relevant and
897 * asked for (non NULL)
898 */
899struct device_node *of_get_cpu_node(int cpu, unsigned int *thread)
900{
901 int hardid;
902 struct device_node *np;
903
904 hardid = get_hard_smp_processor_id(cpu);
905
906 for_each_node_by_type(np, "cpu") {
907 const u32 *intserv;
908 unsigned int plen, t;
909
910 /* Check for ibm,ppc-interrupt-server#s. If it doesn't exist
911 * fallback to "reg" property and assume no threads
912 */
913 intserv = of_get_property(np, "ibm,ppc-interrupt-server#s",
914 &plen);
915 if (intserv == NULL) {
916 const u32 *reg = of_get_property(np, "reg", NULL);
917 if (reg == NULL)
918 continue;
919 if (*reg == hardid) {
920 if (thread)
921 *thread = 0;
922 return np;
923 }
924 } else {
925 plen /= sizeof(u32);
926 for (t = 0; t < plen; t++) {
927 if (hardid == intserv[t]) {
928 if (thread)
929 *thread = t;
930 return np;
931 }
932 }
933 }
934 }
935 return NULL;
936}
937EXPORT_SYMBOL(of_get_cpu_node);
938
939#if defined(CONFIG_DEBUG_FS) && defined(DEBUG)
940static struct debugfs_blob_wrapper flat_dt_blob;
941
942static int __init export_flat_device_tree(void)
943{
944 struct dentry *d;
945
946 flat_dt_blob.data = initial_boot_params;
947 flat_dt_blob.size = initial_boot_params->totalsize;
948
949 d = debugfs_create_blob("flat-device-tree", S_IFREG | S_IRUSR,
950 powerpc_debugfs_root, &flat_dt_blob);
951 if (!d)
952 return 1;
953
954 return 0;
955}
956__initcall(export_flat_device_tree);
957#endif