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