xref: /openbmc/linux/arch/powerpc/kernel/prom.c (revision bc05aa6e)
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 #include <linux/cpu.h>
38 
39 #include <asm/prom.h>
40 #include <asm/rtas.h>
41 #include <asm/page.h>
42 #include <asm/processor.h>
43 #include <asm/irq.h>
44 #include <asm/io.h>
45 #include <asm/kdump.h>
46 #include <asm/smp.h>
47 #include <asm/mmu.h>
48 #include <asm/paca.h>
49 #include <asm/pgtable.h>
50 #include <asm/powernv.h>
51 #include <asm/iommu.h>
52 #include <asm/btext.h>
53 #include <asm/sections.h>
54 #include <asm/machdep.h>
55 #include <asm/pci-bridge.h>
56 #include <asm/kexec.h>
57 #include <asm/opal.h>
58 #include <asm/fadump.h>
59 #include <asm/epapr_hcalls.h>
60 #include <asm/firmware.h>
61 #include <asm/dt_cpu_ftrs.h>
62 #include <asm/drmem.h>
63 
64 #include <mm/mmu_decl.h>
65 
66 #ifdef DEBUG
67 #define DBG(fmt...) printk(KERN_ERR fmt)
68 #else
69 #define DBG(fmt...)
70 #endif
71 
72 #ifdef CONFIG_PPC64
73 int __initdata iommu_is_off;
74 int __initdata iommu_force_on;
75 unsigned long tce_alloc_start, tce_alloc_end;
76 u64 ppc64_rma_size;
77 #endif
78 static phys_addr_t first_memblock_size;
79 static int __initdata boot_cpu_count;
80 
81 static int __init early_parse_mem(char *p)
82 {
83 	if (!p)
84 		return 1;
85 
86 	memory_limit = PAGE_ALIGN(memparse(p, &p));
87 	DBG("memory limit = 0x%llx\n", memory_limit);
88 
89 	return 0;
90 }
91 early_param("mem", early_parse_mem);
92 
93 /*
94  * overlaps_initrd - check for overlap with page aligned extension of
95  * initrd.
96  */
97 static inline int overlaps_initrd(unsigned long start, unsigned long size)
98 {
99 #ifdef CONFIG_BLK_DEV_INITRD
100 	if (!initrd_start)
101 		return 0;
102 
103 	return	(start + size) > _ALIGN_DOWN(initrd_start, PAGE_SIZE) &&
104 			start <= _ALIGN_UP(initrd_end, PAGE_SIZE);
105 #else
106 	return 0;
107 #endif
108 }
109 
110 /**
111  * move_device_tree - move tree to an unused area, if needed.
112  *
113  * The device tree may be allocated beyond our memory limit, or inside the
114  * crash kernel region for kdump, or within the page aligned range of initrd.
115  * If so, move it out of the way.
116  */
117 static void __init move_device_tree(void)
118 {
119 	unsigned long start, size;
120 	void *p;
121 
122 	DBG("-> move_device_tree\n");
123 
124 	start = __pa(initial_boot_params);
125 	size = fdt_totalsize(initial_boot_params);
126 
127 	if ((memory_limit && (start + size) > PHYSICAL_START + memory_limit) ||
128 			overlaps_crashkernel(start, size) ||
129 			overlaps_initrd(start, size)) {
130 		p = __va(memblock_alloc(size, PAGE_SIZE));
131 		memcpy(p, initial_boot_params, size);
132 		initial_boot_params = p;
133 		DBG("Moved device tree to 0x%p\n", p);
134 	}
135 
136 	DBG("<- move_device_tree\n");
137 }
138 
139 /*
140  * ibm,pa-features is a per-cpu property that contains a string of
141  * attribute descriptors, each of which has a 2 byte header plus up
142  * to 254 bytes worth of processor attribute bits.  First header
143  * byte specifies the number of bytes following the header.
144  * Second header byte is an "attribute-specifier" type, of which
145  * zero is the only currently-defined value.
146  * Implementation:  Pass in the byte and bit offset for the feature
147  * that we are interested in.  The function will return -1 if the
148  * pa-features property is missing, or a 1/0 to indicate if the feature
149  * is supported/not supported.  Note that the bit numbers are
150  * big-endian to match the definition in PAPR.
151  */
152 static struct ibm_pa_feature {
153 	unsigned long	cpu_features;	/* CPU_FTR_xxx bit */
154 	unsigned long	mmu_features;	/* MMU_FTR_xxx bit */
155 	unsigned int	cpu_user_ftrs;	/* PPC_FEATURE_xxx bit */
156 	unsigned int	cpu_user_ftrs2;	/* PPC_FEATURE2_xxx bit */
157 	unsigned char	pabyte;		/* byte number in ibm,pa-features */
158 	unsigned char	pabit;		/* bit number (big-endian) */
159 	unsigned char	invert;		/* if 1, pa bit set => clear feature */
160 } ibm_pa_features[] __initdata = {
161 	{ .pabyte = 0,  .pabit = 0, .cpu_user_ftrs = PPC_FEATURE_HAS_MMU },
162 	{ .pabyte = 0,  .pabit = 1, .cpu_user_ftrs = PPC_FEATURE_HAS_FPU },
163 	{ .pabyte = 0,  .pabit = 3, .cpu_features  = CPU_FTR_CTRL },
164 	{ .pabyte = 0,  .pabit = 6, .cpu_features  = CPU_FTR_NOEXECUTE },
165 	{ .pabyte = 1,  .pabit = 2, .mmu_features  = MMU_FTR_CI_LARGE_PAGE },
166 #ifdef CONFIG_PPC_RADIX_MMU
167 	{ .pabyte = 40, .pabit = 0, .mmu_features  = MMU_FTR_TYPE_RADIX },
168 #endif
169 	{ .pabyte = 1,  .pabit = 1, .invert = 1, .cpu_features = CPU_FTR_NODSISRALIGN },
170 	{ .pabyte = 5,  .pabit = 0, .cpu_features  = CPU_FTR_REAL_LE,
171 				    .cpu_user_ftrs = PPC_FEATURE_TRUE_LE },
172 	/*
173 	 * If the kernel doesn't support TM (ie CONFIG_PPC_TRANSACTIONAL_MEM=n),
174 	 * we don't want to turn on TM here, so we use the *_COMP versions
175 	 * which are 0 if the kernel doesn't support TM.
176 	 */
177 	{ .pabyte = 22, .pabit = 0, .cpu_features = CPU_FTR_TM_COMP,
178 	  .cpu_user_ftrs2 = PPC_FEATURE2_HTM_COMP | PPC_FEATURE2_HTM_NOSC_COMP },
179 };
180 
181 static void __init scan_features(unsigned long node, const unsigned char *ftrs,
182 				 unsigned long tablelen,
183 				 struct ibm_pa_feature *fp,
184 				 unsigned long ft_size)
185 {
186 	unsigned long i, len, bit;
187 
188 	/* find descriptor with type == 0 */
189 	for (;;) {
190 		if (tablelen < 3)
191 			return;
192 		len = 2 + ftrs[0];
193 		if (tablelen < len)
194 			return;		/* descriptor 0 not found */
195 		if (ftrs[1] == 0)
196 			break;
197 		tablelen -= len;
198 		ftrs += len;
199 	}
200 
201 	/* loop over bits we know about */
202 	for (i = 0; i < ft_size; ++i, ++fp) {
203 		if (fp->pabyte >= ftrs[0])
204 			continue;
205 		bit = (ftrs[2 + fp->pabyte] >> (7 - fp->pabit)) & 1;
206 		if (bit ^ fp->invert) {
207 			cur_cpu_spec->cpu_features |= fp->cpu_features;
208 			cur_cpu_spec->cpu_user_features |= fp->cpu_user_ftrs;
209 			cur_cpu_spec->cpu_user_features2 |= fp->cpu_user_ftrs2;
210 			cur_cpu_spec->mmu_features |= fp->mmu_features;
211 		} else {
212 			cur_cpu_spec->cpu_features &= ~fp->cpu_features;
213 			cur_cpu_spec->cpu_user_features &= ~fp->cpu_user_ftrs;
214 			cur_cpu_spec->cpu_user_features2 &= ~fp->cpu_user_ftrs2;
215 			cur_cpu_spec->mmu_features &= ~fp->mmu_features;
216 		}
217 	}
218 }
219 
220 static void __init check_cpu_pa_features(unsigned long node)
221 {
222 	const unsigned char *pa_ftrs;
223 	int tablelen;
224 
225 	pa_ftrs = of_get_flat_dt_prop(node, "ibm,pa-features", &tablelen);
226 	if (pa_ftrs == NULL)
227 		return;
228 
229 	scan_features(node, pa_ftrs, tablelen,
230 		      ibm_pa_features, ARRAY_SIZE(ibm_pa_features));
231 }
232 
233 #ifdef CONFIG_PPC_BOOK3S_64
234 static void __init init_mmu_slb_size(unsigned long node)
235 {
236 	const __be32 *slb_size_ptr;
237 
238 	slb_size_ptr = of_get_flat_dt_prop(node, "slb-size", NULL) ? :
239 			of_get_flat_dt_prop(node, "ibm,slb-size", NULL);
240 
241 	if (slb_size_ptr)
242 		mmu_slb_size = be32_to_cpup(slb_size_ptr);
243 }
244 #else
245 #define init_mmu_slb_size(node) do { } while(0)
246 #endif
247 
248 static struct feature_property {
249 	const char *name;
250 	u32 min_value;
251 	unsigned long cpu_feature;
252 	unsigned long cpu_user_ftr;
253 } feature_properties[] __initdata = {
254 #ifdef CONFIG_ALTIVEC
255 	{"altivec", 0, CPU_FTR_ALTIVEC, PPC_FEATURE_HAS_ALTIVEC},
256 	{"ibm,vmx", 1, CPU_FTR_ALTIVEC, PPC_FEATURE_HAS_ALTIVEC},
257 #endif /* CONFIG_ALTIVEC */
258 #ifdef CONFIG_VSX
259 	/* Yes, this _really_ is ibm,vmx == 2 to enable VSX */
260 	{"ibm,vmx", 2, CPU_FTR_VSX, PPC_FEATURE_HAS_VSX},
261 #endif /* CONFIG_VSX */
262 #ifdef CONFIG_PPC64
263 	{"ibm,dfp", 1, 0, PPC_FEATURE_HAS_DFP},
264 	{"ibm,purr", 1, CPU_FTR_PURR, 0},
265 	{"ibm,spurr", 1, CPU_FTR_SPURR, 0},
266 #endif /* CONFIG_PPC64 */
267 };
268 
269 #if defined(CONFIG_44x) && defined(CONFIG_PPC_FPU)
270 static inline void identical_pvr_fixup(unsigned long node)
271 {
272 	unsigned int pvr;
273 	const char *model = of_get_flat_dt_prop(node, "model", NULL);
274 
275 	/*
276 	 * Since 440GR(x)/440EP(x) processors have the same pvr,
277 	 * we check the node path and set bit 28 in the cur_cpu_spec
278 	 * pvr for EP(x) processor version. This bit is always 0 in
279 	 * the "real" pvr. Then we call identify_cpu again with
280 	 * the new logical pvr to enable FPU support.
281 	 */
282 	if (model && strstr(model, "440EP")) {
283 		pvr = cur_cpu_spec->pvr_value | 0x8;
284 		identify_cpu(0, pvr);
285 		DBG("Using logical pvr %x for %s\n", pvr, model);
286 	}
287 }
288 #else
289 #define identical_pvr_fixup(node) do { } while(0)
290 #endif
291 
292 static void __init check_cpu_feature_properties(unsigned long node)
293 {
294 	unsigned long i;
295 	struct feature_property *fp = feature_properties;
296 	const __be32 *prop;
297 
298 	for (i = 0; i < ARRAY_SIZE(feature_properties); ++i, ++fp) {
299 		prop = of_get_flat_dt_prop(node, fp->name, NULL);
300 		if (prop && be32_to_cpup(prop) >= fp->min_value) {
301 			cur_cpu_spec->cpu_features |= fp->cpu_feature;
302 			cur_cpu_spec->cpu_user_features |= fp->cpu_user_ftr;
303 		}
304 	}
305 }
306 
307 static int __init early_init_dt_scan_cpus(unsigned long node,
308 					  const char *uname, int depth,
309 					  void *data)
310 {
311 	const char *type = of_get_flat_dt_prop(node, "device_type", NULL);
312 	const __be32 *prop;
313 	const __be32 *intserv;
314 	int i, nthreads;
315 	int len;
316 	int found = -1;
317 	int found_thread = 0;
318 
319 	/* We are scanning "cpu" nodes only */
320 	if (type == NULL || strcmp(type, "cpu") != 0)
321 		return 0;
322 
323 	/* Get physical cpuid */
324 	intserv = of_get_flat_dt_prop(node, "ibm,ppc-interrupt-server#s", &len);
325 	if (!intserv)
326 		intserv = of_get_flat_dt_prop(node, "reg", &len);
327 
328 	nthreads = len / sizeof(int);
329 
330 	/*
331 	 * Now see if any of these threads match our boot cpu.
332 	 * NOTE: This must match the parsing done in smp_setup_cpu_maps.
333 	 */
334 	for (i = 0; i < nthreads; i++) {
335 		/*
336 		 * version 2 of the kexec param format adds the phys cpuid of
337 		 * booted proc.
338 		 */
339 		if (fdt_version(initial_boot_params) >= 2) {
340 			if (be32_to_cpu(intserv[i]) ==
341 			    fdt_boot_cpuid_phys(initial_boot_params)) {
342 				found = boot_cpu_count;
343 				found_thread = i;
344 			}
345 		} else {
346 			/*
347 			 * Check if it's the boot-cpu, set it's hw index now,
348 			 * unfortunately this format did not support booting
349 			 * off secondary threads.
350 			 */
351 			if (of_get_flat_dt_prop(node,
352 					"linux,boot-cpu", NULL) != NULL)
353 				found = boot_cpu_count;
354 		}
355 #ifdef CONFIG_SMP
356 		/* logical cpu id is always 0 on UP kernels */
357 		boot_cpu_count++;
358 #endif
359 	}
360 
361 	/* Not the boot CPU */
362 	if (found < 0)
363 		return 0;
364 
365 	DBG("boot cpu: logical %d physical %d\n", found,
366 	    be32_to_cpu(intserv[found_thread]));
367 	boot_cpuid = found;
368 	set_hard_smp_processor_id(found, be32_to_cpu(intserv[found_thread]));
369 
370 	/*
371 	 * PAPR defines "logical" PVR values for cpus that
372 	 * meet various levels of the architecture:
373 	 * 0x0f000001	Architecture version 2.04
374 	 * 0x0f000002	Architecture version 2.05
375 	 * If the cpu-version property in the cpu node contains
376 	 * such a value, we call identify_cpu again with the
377 	 * logical PVR value in order to use the cpu feature
378 	 * bits appropriate for the architecture level.
379 	 *
380 	 * A POWER6 partition in "POWER6 architected" mode
381 	 * uses the 0x0f000002 PVR value; in POWER5+ mode
382 	 * it uses 0x0f000001.
383 	 *
384 	 * If we're using device tree CPU feature discovery then we don't
385 	 * support the cpu-version property, and it's the responsibility of the
386 	 * firmware/hypervisor to provide the correct feature set for the
387 	 * architecture level via the ibm,powerpc-cpu-features binding.
388 	 */
389 	if (!dt_cpu_ftrs_in_use()) {
390 		prop = of_get_flat_dt_prop(node, "cpu-version", NULL);
391 		if (prop && (be32_to_cpup(prop) & 0xff000000) == 0x0f000000)
392 			identify_cpu(0, be32_to_cpup(prop));
393 
394 		check_cpu_feature_properties(node);
395 		check_cpu_pa_features(node);
396 	}
397 
398 	identical_pvr_fixup(node);
399 	init_mmu_slb_size(node);
400 
401 #ifdef CONFIG_PPC64
402 	if (nthreads == 1)
403 		cur_cpu_spec->cpu_features &= ~CPU_FTR_SMT;
404 	else if (!dt_cpu_ftrs_in_use())
405 		cur_cpu_spec->cpu_features |= CPU_FTR_SMT;
406 #endif
407 
408 	return 0;
409 }
410 
411 static int __init early_init_dt_scan_chosen_ppc(unsigned long node,
412 						const char *uname,
413 						int depth, void *data)
414 {
415 	const unsigned long *lprop; /* All these set by kernel, so no need to convert endian */
416 
417 	/* Use common scan routine to determine if this is the chosen node */
418 	if (early_init_dt_scan_chosen(node, uname, depth, data) == 0)
419 		return 0;
420 
421 #ifdef CONFIG_PPC64
422 	/* check if iommu is forced on or off */
423 	if (of_get_flat_dt_prop(node, "linux,iommu-off", NULL) != NULL)
424 		iommu_is_off = 1;
425 	if (of_get_flat_dt_prop(node, "linux,iommu-force-on", NULL) != NULL)
426 		iommu_force_on = 1;
427 #endif
428 
429 	/* mem=x on the command line is the preferred mechanism */
430 	lprop = of_get_flat_dt_prop(node, "linux,memory-limit", NULL);
431 	if (lprop)
432 		memory_limit = *lprop;
433 
434 #ifdef CONFIG_PPC64
435 	lprop = of_get_flat_dt_prop(node, "linux,tce-alloc-start", NULL);
436 	if (lprop)
437 		tce_alloc_start = *lprop;
438 	lprop = of_get_flat_dt_prop(node, "linux,tce-alloc-end", NULL);
439 	if (lprop)
440 		tce_alloc_end = *lprop;
441 #endif
442 
443 #ifdef CONFIG_KEXEC_CORE
444 	lprop = of_get_flat_dt_prop(node, "linux,crashkernel-base", NULL);
445 	if (lprop)
446 		crashk_res.start = *lprop;
447 
448 	lprop = of_get_flat_dt_prop(node, "linux,crashkernel-size", NULL);
449 	if (lprop)
450 		crashk_res.end = crashk_res.start + *lprop - 1;
451 #endif
452 
453 	/* break now */
454 	return 1;
455 }
456 
457 #ifdef CONFIG_PPC_PSERIES
458 /*
459  * Interpret the ibm dynamic reconfiguration memory LMBs.
460  * This contains a list of memory blocks along with NUMA affinity
461  * information.
462  */
463 static void __init early_init_drmem_lmb(struct drmem_lmb *lmb,
464 					const __be32 **usm)
465 {
466 	u64 base, size;
467 	int is_kexec_kdump = 0, rngs;
468 
469 	base = lmb->base_addr;
470 	size = drmem_lmb_size();
471 	rngs = 1;
472 
473 	/*
474 	 * Skip this block if the reserved bit is set in flags
475 	 * or if the block is not assigned to this partition.
476 	 */
477 	if ((lmb->flags & DRCONF_MEM_RESERVED) ||
478 	    !(lmb->flags & DRCONF_MEM_ASSIGNED))
479 		return;
480 
481 	if (*usm)
482 		is_kexec_kdump = 1;
483 
484 	if (is_kexec_kdump) {
485 		/*
486 		 * For each memblock in ibm,dynamic-memory, a
487 		 * corresponding entry in linux,drconf-usable-memory
488 		 * property contains a counter 'p' followed by 'p'
489 		 * (base, size) duple. Now read the counter from
490 		 * linux,drconf-usable-memory property
491 		 */
492 		rngs = dt_mem_next_cell(dt_root_size_cells, usm);
493 		if (!rngs) /* there are no (base, size) duple */
494 			return;
495 	}
496 
497 	do {
498 		if (is_kexec_kdump) {
499 			base = dt_mem_next_cell(dt_root_addr_cells, usm);
500 			size = dt_mem_next_cell(dt_root_size_cells, usm);
501 		}
502 
503 		if (iommu_is_off) {
504 			if (base >= 0x80000000ul)
505 				continue;
506 			if ((base + size) > 0x80000000ul)
507 				size = 0x80000000ul - base;
508 		}
509 
510 		DBG("Adding: %llx -> %llx\n", base, size);
511 		memblock_add(base, size);
512 	} while (--rngs);
513 }
514 #endif /* CONFIG_PPC_PSERIES */
515 
516 static int __init early_init_dt_scan_memory_ppc(unsigned long node,
517 						const char *uname,
518 						int depth, void *data)
519 {
520 #ifdef CONFIG_PPC_PSERIES
521 	if (depth == 1 &&
522 	    strcmp(uname, "ibm,dynamic-reconfiguration-memory") == 0) {
523 		walk_drmem_lmbs_early(node, early_init_drmem_lmb);
524 		return 0;
525 	}
526 #endif
527 
528 	return early_init_dt_scan_memory(node, uname, depth, data);
529 }
530 
531 /*
532  * For a relocatable kernel, we need to get the memstart_addr first,
533  * then use it to calculate the virtual kernel start address. This has
534  * to happen at a very early stage (before machine_init). In this case,
535  * we just want to get the memstart_address and would not like to mess the
536  * memblock at this stage. So introduce a variable to skip the memblock_add()
537  * for this reason.
538  */
539 #ifdef CONFIG_RELOCATABLE
540 static int add_mem_to_memblock = 1;
541 #else
542 #define add_mem_to_memblock 1
543 #endif
544 
545 void __init early_init_dt_add_memory_arch(u64 base, u64 size)
546 {
547 #ifdef CONFIG_PPC64
548 	if (iommu_is_off) {
549 		if (base >= 0x80000000ul)
550 			return;
551 		if ((base + size) > 0x80000000ul)
552 			size = 0x80000000ul - base;
553 	}
554 #endif
555 	/* Keep track of the beginning of memory -and- the size of
556 	 * the very first block in the device-tree as it represents
557 	 * the RMA on ppc64 server
558 	 */
559 	if (base < memstart_addr) {
560 		memstart_addr = base;
561 		first_memblock_size = size;
562 	}
563 
564 	/* Add the chunk to the MEMBLOCK list */
565 	if (add_mem_to_memblock)
566 		memblock_add(base, size);
567 }
568 
569 static void __init early_reserve_mem_dt(void)
570 {
571 	unsigned long i, dt_root;
572 	int len;
573 	const __be32 *prop;
574 
575 	early_init_fdt_reserve_self();
576 	early_init_fdt_scan_reserved_mem();
577 
578 	dt_root = of_get_flat_dt_root();
579 
580 	prop = of_get_flat_dt_prop(dt_root, "reserved-ranges", &len);
581 
582 	if (!prop)
583 		return;
584 
585 	DBG("Found new-style reserved-ranges\n");
586 
587 	/* Each reserved range is an (address,size) pair, 2 cells each,
588 	 * totalling 4 cells per range. */
589 	for (i = 0; i < len / (sizeof(*prop) * 4); i++) {
590 		u64 base, size;
591 
592 		base = of_read_number(prop + (i * 4) + 0, 2);
593 		size = of_read_number(prop + (i * 4) + 2, 2);
594 
595 		if (size) {
596 			DBG("reserving: %llx -> %llx\n", base, size);
597 			memblock_reserve(base, size);
598 		}
599 	}
600 }
601 
602 static void __init early_reserve_mem(void)
603 {
604 	__be64 *reserve_map;
605 
606 	reserve_map = (__be64 *)(((unsigned long)initial_boot_params) +
607 			fdt_off_mem_rsvmap(initial_boot_params));
608 
609 	/* Look for the new "reserved-regions" property in the DT */
610 	early_reserve_mem_dt();
611 
612 #ifdef CONFIG_BLK_DEV_INITRD
613 	/* Then reserve the initrd, if any */
614 	if (initrd_start && (initrd_end > initrd_start)) {
615 		memblock_reserve(_ALIGN_DOWN(__pa(initrd_start), PAGE_SIZE),
616 			_ALIGN_UP(initrd_end, PAGE_SIZE) -
617 			_ALIGN_DOWN(initrd_start, PAGE_SIZE));
618 	}
619 #endif /* CONFIG_BLK_DEV_INITRD */
620 
621 #ifdef CONFIG_PPC32
622 	/*
623 	 * Handle the case where we might be booting from an old kexec
624 	 * image that setup the mem_rsvmap as pairs of 32-bit values
625 	 */
626 	if (be64_to_cpup(reserve_map) > 0xffffffffull) {
627 		u32 base_32, size_32;
628 		__be32 *reserve_map_32 = (__be32 *)reserve_map;
629 
630 		DBG("Found old 32-bit reserve map\n");
631 
632 		while (1) {
633 			base_32 = be32_to_cpup(reserve_map_32++);
634 			size_32 = be32_to_cpup(reserve_map_32++);
635 			if (size_32 == 0)
636 				break;
637 			DBG("reserving: %x -> %x\n", base_32, size_32);
638 			memblock_reserve(base_32, size_32);
639 		}
640 		return;
641 	}
642 #endif
643 }
644 
645 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
646 static bool tm_disabled __initdata;
647 
648 static int __init parse_ppc_tm(char *str)
649 {
650 	bool res;
651 
652 	if (kstrtobool(str, &res))
653 		return -EINVAL;
654 
655 	tm_disabled = !res;
656 
657 	return 0;
658 }
659 early_param("ppc_tm", parse_ppc_tm);
660 
661 static void __init tm_init(void)
662 {
663 	if (tm_disabled) {
664 		pr_info("Disabling hardware transactional memory (HTM)\n");
665 		cur_cpu_spec->cpu_user_features2 &=
666 			~(PPC_FEATURE2_HTM_NOSC | PPC_FEATURE2_HTM);
667 		cur_cpu_spec->cpu_features &= ~CPU_FTR_TM;
668 		return;
669 	}
670 
671 	pnv_tm_init();
672 }
673 #else
674 static void tm_init(void) { }
675 #endif /* CONFIG_PPC_TRANSACTIONAL_MEM */
676 
677 void __init early_init_devtree(void *params)
678 {
679 	phys_addr_t limit;
680 
681 	DBG(" -> early_init_devtree(%p)\n", params);
682 
683 	/* Too early to BUG_ON(), do it by hand */
684 	if (!early_init_dt_verify(params))
685 		panic("BUG: Failed verifying flat device tree, bad version?");
686 
687 #ifdef CONFIG_PPC_RTAS
688 	/* Some machines might need RTAS info for debugging, grab it now. */
689 	of_scan_flat_dt(early_init_dt_scan_rtas, NULL);
690 #endif
691 
692 #ifdef CONFIG_PPC_POWERNV
693 	/* Some machines might need OPAL info for debugging, grab it now. */
694 	of_scan_flat_dt(early_init_dt_scan_opal, NULL);
695 #endif
696 
697 #ifdef CONFIG_FA_DUMP
698 	/* scan tree to see if dump is active during last boot */
699 	of_scan_flat_dt(early_init_dt_scan_fw_dump, NULL);
700 #endif
701 
702 	/* Retrieve various informations from the /chosen node of the
703 	 * device-tree, including the platform type, initrd location and
704 	 * size, TCE reserve, and more ...
705 	 */
706 	of_scan_flat_dt(early_init_dt_scan_chosen_ppc, boot_command_line);
707 
708 	/* Scan memory nodes and rebuild MEMBLOCKs */
709 	of_scan_flat_dt(early_init_dt_scan_root, NULL);
710 	of_scan_flat_dt(early_init_dt_scan_memory_ppc, NULL);
711 
712 	parse_early_param();
713 
714 	/* make sure we've parsed cmdline for mem= before this */
715 	if (memory_limit)
716 		first_memblock_size = min_t(u64, first_memblock_size, memory_limit);
717 	setup_initial_memory_limit(memstart_addr, first_memblock_size);
718 	/* Reserve MEMBLOCK regions used by kernel, initrd, dt, etc... */
719 	memblock_reserve(PHYSICAL_START, __pa(klimit) - PHYSICAL_START);
720 	/* If relocatable, reserve first 32k for interrupt vectors etc. */
721 	if (PHYSICAL_START > MEMORY_START)
722 		memblock_reserve(MEMORY_START, 0x8000);
723 	reserve_kdump_trampoline();
724 #ifdef CONFIG_FA_DUMP
725 	/*
726 	 * If we fail to reserve memory for firmware-assisted dump then
727 	 * fallback to kexec based kdump.
728 	 */
729 	if (fadump_reserve_mem() == 0)
730 #endif
731 		reserve_crashkernel();
732 	early_reserve_mem();
733 
734 	/* Ensure that total memory size is page-aligned. */
735 	limit = ALIGN(memory_limit ?: memblock_phys_mem_size(), PAGE_SIZE);
736 	memblock_enforce_memory_limit(limit);
737 
738 	memblock_allow_resize();
739 	memblock_dump_all();
740 
741 	DBG("Phys. mem: %llx\n", memblock_phys_mem_size());
742 
743 	/* We may need to relocate the flat tree, do it now.
744 	 * FIXME .. and the initrd too? */
745 	move_device_tree();
746 
747 	allocate_pacas();
748 
749 	DBG("Scanning CPUs ...\n");
750 
751 	dt_cpu_ftrs_scan();
752 
753 	/* Retrieve CPU related informations from the flat tree
754 	 * (altivec support, boot CPU ID, ...)
755 	 */
756 	of_scan_flat_dt(early_init_dt_scan_cpus, NULL);
757 	if (boot_cpuid < 0) {
758 		printk("Failed to identify boot CPU !\n");
759 		BUG();
760 	}
761 
762 #if defined(CONFIG_SMP) && defined(CONFIG_PPC64)
763 	/* We'll later wait for secondaries to check in; there are
764 	 * NCPUS-1 non-boot CPUs  :-)
765 	 */
766 	spinning_secondaries = boot_cpu_count - 1;
767 #endif
768 
769 	mmu_early_init_devtree();
770 
771 #ifdef CONFIG_PPC_POWERNV
772 	/* Scan and build the list of machine check recoverable ranges */
773 	of_scan_flat_dt(early_init_dt_scan_recoverable_ranges, NULL);
774 #endif
775 	epapr_paravirt_early_init();
776 
777 	/* Now try to figure out if we are running on LPAR and so on */
778 	pseries_probe_fw_features();
779 
780 #ifdef CONFIG_PPC_PS3
781 	/* Identify PS3 firmware */
782 	if (of_flat_dt_is_compatible(of_get_flat_dt_root(), "sony,ps3"))
783 		powerpc_firmware_features |= FW_FEATURE_PS3_POSSIBLE;
784 #endif
785 
786 	tm_init();
787 
788 	DBG(" <- early_init_devtree()\n");
789 }
790 
791 #ifdef CONFIG_RELOCATABLE
792 /*
793  * This function run before early_init_devtree, so we have to init
794  * initial_boot_params.
795  */
796 void __init early_get_first_memblock_info(void *params, phys_addr_t *size)
797 {
798 	/* Setup flat device-tree pointer */
799 	initial_boot_params = params;
800 
801 	/*
802 	 * Scan the memory nodes and set add_mem_to_memblock to 0 to avoid
803 	 * mess the memblock.
804 	 */
805 	add_mem_to_memblock = 0;
806 	of_scan_flat_dt(early_init_dt_scan_root, NULL);
807 	of_scan_flat_dt(early_init_dt_scan_memory_ppc, NULL);
808 	add_mem_to_memblock = 1;
809 
810 	if (size)
811 		*size = first_memblock_size;
812 }
813 #endif
814 
815 /*******
816  *
817  * New implementation of the OF "find" APIs, return a refcounted
818  * object, call of_node_put() when done.  The device tree and list
819  * are protected by a rw_lock.
820  *
821  * Note that property management will need some locking as well,
822  * this isn't dealt with yet.
823  *
824  *******/
825 
826 /**
827  * of_get_ibm_chip_id - Returns the IBM "chip-id" of a device
828  * @np: device node of the device
829  *
830  * This looks for a property "ibm,chip-id" in the node or any
831  * of its parents and returns its content, or -1 if it cannot
832  * be found.
833  */
834 int of_get_ibm_chip_id(struct device_node *np)
835 {
836 	of_node_get(np);
837 	while (np) {
838 		u32 chip_id;
839 
840 		/*
841 		 * Skiboot may produce memory nodes that contain more than one
842 		 * cell in chip-id, we only read the first one here.
843 		 */
844 		if (!of_property_read_u32(np, "ibm,chip-id", &chip_id)) {
845 			of_node_put(np);
846 			return chip_id;
847 		}
848 
849 		np = of_get_next_parent(np);
850 	}
851 	return -1;
852 }
853 EXPORT_SYMBOL(of_get_ibm_chip_id);
854 
855 /**
856  * cpu_to_chip_id - Return the cpus chip-id
857  * @cpu: The logical cpu number.
858  *
859  * Return the value of the ibm,chip-id property corresponding to the given
860  * logical cpu number. If the chip-id can not be found, returns -1.
861  */
862 int cpu_to_chip_id(int cpu)
863 {
864 	struct device_node *np;
865 
866 	np = of_get_cpu_node(cpu, NULL);
867 	if (!np)
868 		return -1;
869 
870 	of_node_put(np);
871 	return of_get_ibm_chip_id(np);
872 }
873 EXPORT_SYMBOL(cpu_to_chip_id);
874 
875 bool arch_match_cpu_phys_id(int cpu, u64 phys_id)
876 {
877 	return (int)phys_id == get_hard_smp_processor_id(cpu);
878 }
879