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