xref: /openbmc/linux/arch/powerpc/kernel/prom.c (revision 62975d27)
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
69 int __initdata iommu_is_off;
70 int __initdata iommu_force_on;
71 unsigned long tce_alloc_start, tce_alloc_end;
72 u64 ppc64_rma_size;
73 #endif
74 static phys_addr_t first_memblock_size;
75 static int __initdata boot_cpu_count;
76 
77 static 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 }
87 early_param("mem", early_parse_mem);
88 
89 /*
90  * overlaps_initrd - check for overlap with page aligned extension of
91  * initrd.
92  */
93 static 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  */
113 static 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  */
151 static 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 },
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 
180 static void __init scan_features(unsigned long node, const unsigned char *ftrs,
181 				 unsigned long tablelen,
182 				 struct ibm_pa_feature *fp,
183 				 unsigned long ft_size)
184 {
185 	unsigned long i, len, bit;
186 
187 	/* find descriptor with type == 0 */
188 	for (;;) {
189 		if (tablelen < 3)
190 			return;
191 		len = 2 + ftrs[0];
192 		if (tablelen < len)
193 			return;		/* descriptor 0 not found */
194 		if (ftrs[1] == 0)
195 			break;
196 		tablelen -= len;
197 		ftrs += len;
198 	}
199 
200 	/* loop over bits we know about */
201 	for (i = 0; i < ft_size; ++i, ++fp) {
202 		if (fp->pabyte >= ftrs[0])
203 			continue;
204 		bit = (ftrs[2 + fp->pabyte] >> (7 - fp->pabit)) & 1;
205 		if (bit ^ fp->invert) {
206 			cur_cpu_spec->cpu_features |= fp->cpu_features;
207 			cur_cpu_spec->cpu_user_features |= fp->cpu_user_ftrs;
208 			cur_cpu_spec->cpu_user_features2 |= fp->cpu_user_ftrs2;
209 			cur_cpu_spec->mmu_features |= fp->mmu_features;
210 		} else {
211 			cur_cpu_spec->cpu_features &= ~fp->cpu_features;
212 			cur_cpu_spec->cpu_user_features &= ~fp->cpu_user_ftrs;
213 			cur_cpu_spec->cpu_user_features2 &= ~fp->cpu_user_ftrs2;
214 			cur_cpu_spec->mmu_features &= ~fp->mmu_features;
215 		}
216 	}
217 }
218 
219 static void __init check_cpu_pa_features(unsigned long node)
220 {
221 	const unsigned char *pa_ftrs;
222 	int tablelen;
223 
224 	pa_ftrs = of_get_flat_dt_prop(node, "ibm,pa-features", &tablelen);
225 	if (pa_ftrs == NULL)
226 		return;
227 
228 	scan_features(node, pa_ftrs, tablelen,
229 		      ibm_pa_features, ARRAY_SIZE(ibm_pa_features));
230 }
231 
232 #ifdef CONFIG_PPC_BOOK3S_64
233 static void __init init_mmu_slb_size(unsigned long node)
234 {
235 	const __be32 *slb_size_ptr;
236 
237 	slb_size_ptr = of_get_flat_dt_prop(node, "slb-size", NULL) ? :
238 			of_get_flat_dt_prop(node, "ibm,slb-size", NULL);
239 
240 	if (slb_size_ptr)
241 		mmu_slb_size = be32_to_cpup(slb_size_ptr);
242 }
243 #else
244 #define init_mmu_slb_size(node) do { } while(0)
245 #endif
246 
247 static struct feature_property {
248 	const char *name;
249 	u32 min_value;
250 	unsigned long cpu_feature;
251 	unsigned long cpu_user_ftr;
252 } feature_properties[] __initdata = {
253 #ifdef CONFIG_ALTIVEC
254 	{"altivec", 0, CPU_FTR_ALTIVEC, PPC_FEATURE_HAS_ALTIVEC},
255 	{"ibm,vmx", 1, CPU_FTR_ALTIVEC, PPC_FEATURE_HAS_ALTIVEC},
256 #endif /* CONFIG_ALTIVEC */
257 #ifdef CONFIG_VSX
258 	/* Yes, this _really_ is ibm,vmx == 2 to enable VSX */
259 	{"ibm,vmx", 2, CPU_FTR_VSX, PPC_FEATURE_HAS_VSX},
260 #endif /* CONFIG_VSX */
261 #ifdef CONFIG_PPC64
262 	{"ibm,dfp", 1, 0, PPC_FEATURE_HAS_DFP},
263 	{"ibm,purr", 1, CPU_FTR_PURR, 0},
264 	{"ibm,spurr", 1, CPU_FTR_SPURR, 0},
265 #endif /* CONFIG_PPC64 */
266 };
267 
268 #if defined(CONFIG_44x) && defined(CONFIG_PPC_FPU)
269 static inline void identical_pvr_fixup(unsigned long node)
270 {
271 	unsigned int pvr;
272 	const char *model = of_get_flat_dt_prop(node, "model", NULL);
273 
274 	/*
275 	 * Since 440GR(x)/440EP(x) processors have the same pvr,
276 	 * we check the node path and set bit 28 in the cur_cpu_spec
277 	 * pvr for EP(x) processor version. This bit is always 0 in
278 	 * the "real" pvr. Then we call identify_cpu again with
279 	 * the new logical pvr to enable FPU support.
280 	 */
281 	if (model && strstr(model, "440EP")) {
282 		pvr = cur_cpu_spec->pvr_value | 0x8;
283 		identify_cpu(0, pvr);
284 		DBG("Using logical pvr %x for %s\n", pvr, model);
285 	}
286 }
287 #else
288 #define identical_pvr_fixup(node) do { } while(0)
289 #endif
290 
291 static void __init check_cpu_feature_properties(unsigned long node)
292 {
293 	int i;
294 	struct feature_property *fp = feature_properties;
295 	const __be32 *prop;
296 
297 	for (i = 0; i < (int)ARRAY_SIZE(feature_properties); ++i, ++fp) {
298 		prop = of_get_flat_dt_prop(node, fp->name, NULL);
299 		if (prop && be32_to_cpup(prop) >= fp->min_value) {
300 			cur_cpu_spec->cpu_features |= fp->cpu_feature;
301 			cur_cpu_spec->cpu_user_features |= fp->cpu_user_ftr;
302 		}
303 	}
304 }
305 
306 static int __init early_init_dt_scan_cpus(unsigned long node,
307 					  const char *uname, int depth,
308 					  void *data)
309 {
310 	const char *type = of_get_flat_dt_prop(node, "device_type", NULL);
311 	const __be32 *prop;
312 	const __be32 *intserv;
313 	int i, nthreads;
314 	int len;
315 	int found = -1;
316 	int found_thread = 0;
317 
318 	/* We are scanning "cpu" nodes only */
319 	if (type == NULL || strcmp(type, "cpu") != 0)
320 		return 0;
321 
322 	/* Get physical cpuid */
323 	intserv = of_get_flat_dt_prop(node, "ibm,ppc-interrupt-server#s", &len);
324 	if (!intserv)
325 		intserv = of_get_flat_dt_prop(node, "reg", &len);
326 
327 	nthreads = len / sizeof(int);
328 
329 	/*
330 	 * Now see if any of these threads match our boot cpu.
331 	 * NOTE: This must match the parsing done in smp_setup_cpu_maps.
332 	 */
333 	for (i = 0; i < nthreads; i++) {
334 		if (be32_to_cpu(intserv[i]) ==
335 			fdt_boot_cpuid_phys(initial_boot_params)) {
336 			found = boot_cpu_count;
337 			found_thread = i;
338 		}
339 #ifdef CONFIG_SMP
340 		/* logical cpu id is always 0 on UP kernels */
341 		boot_cpu_count++;
342 #endif
343 	}
344 
345 	/* Not the boot CPU */
346 	if (found < 0)
347 		return 0;
348 
349 	DBG("boot cpu: logical %d physical %d\n", found,
350 	    be32_to_cpu(intserv[found_thread]));
351 	boot_cpuid = found;
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 	 * If we're using device tree CPU feature discovery then we don't
368 	 * support the cpu-version property, and it's the responsibility of the
369 	 * firmware/hypervisor to provide the correct feature set for the
370 	 * architecture level via the ibm,powerpc-cpu-features binding.
371 	 */
372 	if (!dt_cpu_ftrs_in_use()) {
373 		prop = of_get_flat_dt_prop(node, "cpu-version", NULL);
374 		if (prop && (be32_to_cpup(prop) & 0xff000000) == 0x0f000000)
375 			identify_cpu(0, be32_to_cpup(prop));
376 
377 		check_cpu_feature_properties(node);
378 		check_cpu_pa_features(node);
379 	}
380 
381 	identical_pvr_fixup(node);
382 	init_mmu_slb_size(node);
383 
384 #ifdef CONFIG_PPC64
385 	if (nthreads == 1)
386 		cur_cpu_spec->cpu_features &= ~CPU_FTR_SMT;
387 	else if (!dt_cpu_ftrs_in_use())
388 		cur_cpu_spec->cpu_features |= CPU_FTR_SMT;
389 	allocate_paca(boot_cpuid);
390 #endif
391 	set_hard_smp_processor_id(found, be32_to_cpu(intserv[found_thread]));
392 
393 	return 0;
394 }
395 
396 static int __init early_init_dt_scan_chosen_ppc(unsigned long node,
397 						const char *uname,
398 						int depth, void *data)
399 {
400 	const unsigned long *lprop; /* All these set by kernel, so no need to convert endian */
401 
402 	/* Use common scan routine to determine if this is the chosen node */
403 	if (early_init_dt_scan_chosen(node, uname, depth, data) == 0)
404 		return 0;
405 
406 #ifdef CONFIG_PPC64
407 	/* check if iommu is forced on or off */
408 	if (of_get_flat_dt_prop(node, "linux,iommu-off", NULL) != NULL)
409 		iommu_is_off = 1;
410 	if (of_get_flat_dt_prop(node, "linux,iommu-force-on", NULL) != NULL)
411 		iommu_force_on = 1;
412 #endif
413 
414 	/* mem=x on the command line is the preferred mechanism */
415 	lprop = of_get_flat_dt_prop(node, "linux,memory-limit", NULL);
416 	if (lprop)
417 		memory_limit = *lprop;
418 
419 #ifdef CONFIG_PPC64
420 	lprop = of_get_flat_dt_prop(node, "linux,tce-alloc-start", NULL);
421 	if (lprop)
422 		tce_alloc_start = *lprop;
423 	lprop = of_get_flat_dt_prop(node, "linux,tce-alloc-end", NULL);
424 	if (lprop)
425 		tce_alloc_end = *lprop;
426 #endif
427 
428 #ifdef CONFIG_KEXEC_CORE
429 	lprop = of_get_flat_dt_prop(node, "linux,crashkernel-base", NULL);
430 	if (lprop)
431 		crashk_res.start = *lprop;
432 
433 	lprop = of_get_flat_dt_prop(node, "linux,crashkernel-size", NULL);
434 	if (lprop)
435 		crashk_res.end = crashk_res.start + *lprop - 1;
436 #endif
437 
438 	/* break now */
439 	return 1;
440 }
441 
442 /*
443  * Compare the range against max mem limit and update
444  * size if it cross the limit.
445  */
446 
447 #ifdef CONFIG_SPARSEMEM
448 static bool validate_mem_limit(u64 base, u64 *size)
449 {
450 	u64 max_mem = 1UL << (MAX_PHYSMEM_BITS);
451 
452 	if (base >= max_mem)
453 		return false;
454 	if ((base + *size) > max_mem)
455 		*size = max_mem - base;
456 	return true;
457 }
458 #else
459 static bool validate_mem_limit(u64 base, u64 *size)
460 {
461 	return true;
462 }
463 #endif
464 
465 #ifdef CONFIG_PPC_PSERIES
466 /*
467  * Interpret the ibm dynamic reconfiguration memory LMBs.
468  * This contains a list of memory blocks along with NUMA affinity
469  * information.
470  */
471 static void __init early_init_drmem_lmb(struct drmem_lmb *lmb,
472 					const __be32 **usm)
473 {
474 	u64 base, size;
475 	int is_kexec_kdump = 0, rngs;
476 
477 	base = lmb->base_addr;
478 	size = drmem_lmb_size();
479 	rngs = 1;
480 
481 	/*
482 	 * Skip this block if the reserved bit is set in flags
483 	 * or if the block is not assigned to this partition.
484 	 */
485 	if ((lmb->flags & DRCONF_MEM_RESERVED) ||
486 	    !(lmb->flags & DRCONF_MEM_ASSIGNED))
487 		return;
488 
489 	if (*usm)
490 		is_kexec_kdump = 1;
491 
492 	if (is_kexec_kdump) {
493 		/*
494 		 * For each memblock in ibm,dynamic-memory, a
495 		 * corresponding entry in linux,drconf-usable-memory
496 		 * property contains a counter 'p' followed by 'p'
497 		 * (base, size) duple. Now read the counter from
498 		 * linux,drconf-usable-memory property
499 		 */
500 		rngs = dt_mem_next_cell(dt_root_size_cells, usm);
501 		if (!rngs) /* there are no (base, size) duple */
502 			return;
503 	}
504 
505 	do {
506 		if (is_kexec_kdump) {
507 			base = dt_mem_next_cell(dt_root_addr_cells, usm);
508 			size = dt_mem_next_cell(dt_root_size_cells, usm);
509 		}
510 
511 		if (iommu_is_off) {
512 			if (base >= 0x80000000ul)
513 				continue;
514 			if ((base + size) > 0x80000000ul)
515 				size = 0x80000000ul - base;
516 		}
517 
518 		if (!validate_mem_limit(base, &size))
519 			continue;
520 
521 		DBG("Adding: %llx -> %llx\n", base, size);
522 		memblock_add(base, size);
523 
524 		if (lmb->flags & DRCONF_MEM_HOTREMOVABLE)
525 			memblock_mark_hotplug(base, size);
526 	} while (--rngs);
527 }
528 #endif /* CONFIG_PPC_PSERIES */
529 
530 static int __init early_init_dt_scan_memory_ppc(unsigned long node,
531 						const char *uname,
532 						int depth, void *data)
533 {
534 #ifdef CONFIG_PPC_PSERIES
535 	if (depth == 1 &&
536 	    strcmp(uname, "ibm,dynamic-reconfiguration-memory") == 0) {
537 		walk_drmem_lmbs_early(node, early_init_drmem_lmb);
538 		return 0;
539 	}
540 #endif
541 
542 	return early_init_dt_scan_memory(node, uname, depth, data);
543 }
544 
545 /*
546  * For a relocatable kernel, we need to get the memstart_addr first,
547  * then use it to calculate the virtual kernel start address. This has
548  * to happen at a very early stage (before machine_init). In this case,
549  * we just want to get the memstart_address and would not like to mess the
550  * memblock at this stage. So introduce a variable to skip the memblock_add()
551  * for this reason.
552  */
553 #ifdef CONFIG_RELOCATABLE
554 static int add_mem_to_memblock = 1;
555 #else
556 #define add_mem_to_memblock 1
557 #endif
558 
559 void __init early_init_dt_add_memory_arch(u64 base, u64 size)
560 {
561 #ifdef CONFIG_PPC64
562 	if (iommu_is_off) {
563 		if (base >= 0x80000000ul)
564 			return;
565 		if ((base + size) > 0x80000000ul)
566 			size = 0x80000000ul - base;
567 	}
568 #endif
569 	/* Keep track of the beginning of memory -and- the size of
570 	 * the very first block in the device-tree as it represents
571 	 * the RMA on ppc64 server
572 	 */
573 	if (base < memstart_addr) {
574 		memstart_addr = base;
575 		first_memblock_size = size;
576 	}
577 
578 	/* Add the chunk to the MEMBLOCK list */
579 	if (add_mem_to_memblock) {
580 		if (validate_mem_limit(base, &size))
581 			memblock_add(base, size);
582 	}
583 }
584 
585 static void __init early_reserve_mem_dt(void)
586 {
587 	unsigned long i, dt_root;
588 	int len;
589 	const __be32 *prop;
590 
591 	early_init_fdt_reserve_self();
592 	early_init_fdt_scan_reserved_mem();
593 
594 	dt_root = of_get_flat_dt_root();
595 
596 	prop = of_get_flat_dt_prop(dt_root, "reserved-ranges", &len);
597 
598 	if (!prop)
599 		return;
600 
601 	DBG("Found new-style reserved-ranges\n");
602 
603 	/* Each reserved range is an (address,size) pair, 2 cells each,
604 	 * totalling 4 cells per range. */
605 	for (i = 0; i < len / (sizeof(*prop) * 4); i++) {
606 		u64 base, size;
607 
608 		base = of_read_number(prop + (i * 4) + 0, 2);
609 		size = of_read_number(prop + (i * 4) + 2, 2);
610 
611 		if (size) {
612 			DBG("reserving: %llx -> %llx\n", base, size);
613 			memblock_reserve(base, size);
614 		}
615 	}
616 }
617 
618 static void __init early_reserve_mem(void)
619 {
620 	__be64 *reserve_map;
621 
622 	reserve_map = (__be64 *)(((unsigned long)initial_boot_params) +
623 			fdt_off_mem_rsvmap(initial_boot_params));
624 
625 	/* Look for the new "reserved-regions" property in the DT */
626 	early_reserve_mem_dt();
627 
628 #ifdef CONFIG_BLK_DEV_INITRD
629 	/* Then reserve the initrd, if any */
630 	if (initrd_start && (initrd_end > initrd_start)) {
631 		memblock_reserve(ALIGN_DOWN(__pa(initrd_start), PAGE_SIZE),
632 			ALIGN(initrd_end, PAGE_SIZE) -
633 			ALIGN_DOWN(initrd_start, PAGE_SIZE));
634 	}
635 #endif /* CONFIG_BLK_DEV_INITRD */
636 
637 #ifdef CONFIG_PPC32
638 	/*
639 	 * Handle the case where we might be booting from an old kexec
640 	 * image that setup the mem_rsvmap as pairs of 32-bit values
641 	 */
642 	if (be64_to_cpup(reserve_map) > 0xffffffffull) {
643 		u32 base_32, size_32;
644 		__be32 *reserve_map_32 = (__be32 *)reserve_map;
645 
646 		DBG("Found old 32-bit reserve map\n");
647 
648 		while (1) {
649 			base_32 = be32_to_cpup(reserve_map_32++);
650 			size_32 = be32_to_cpup(reserve_map_32++);
651 			if (size_32 == 0)
652 				break;
653 			DBG("reserving: %x -> %x\n", base_32, size_32);
654 			memblock_reserve(base_32, size_32);
655 		}
656 		return;
657 	}
658 #endif
659 }
660 
661 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
662 static bool tm_disabled __initdata;
663 
664 static int __init parse_ppc_tm(char *str)
665 {
666 	bool res;
667 
668 	if (kstrtobool(str, &res))
669 		return -EINVAL;
670 
671 	tm_disabled = !res;
672 
673 	return 0;
674 }
675 early_param("ppc_tm", parse_ppc_tm);
676 
677 static void __init tm_init(void)
678 {
679 	if (tm_disabled) {
680 		pr_info("Disabling hardware transactional memory (HTM)\n");
681 		cur_cpu_spec->cpu_user_features2 &=
682 			~(PPC_FEATURE2_HTM_NOSC | PPC_FEATURE2_HTM);
683 		cur_cpu_spec->cpu_features &= ~CPU_FTR_TM;
684 		return;
685 	}
686 
687 	pnv_tm_init();
688 }
689 #else
690 static void tm_init(void) { }
691 #endif /* CONFIG_PPC_TRANSACTIONAL_MEM */
692 
693 #ifdef CONFIG_PPC64
694 static void __init save_fscr_to_task(void)
695 {
696 	/*
697 	 * Ensure the init_task (pid 0, aka swapper) uses the value of FSCR we
698 	 * have configured via the device tree features or via __init_FSCR().
699 	 * That value will then be propagated to pid 1 (init) and all future
700 	 * processes.
701 	 */
702 	if (early_cpu_has_feature(CPU_FTR_ARCH_207S))
703 		init_task.thread.fscr = mfspr(SPRN_FSCR);
704 }
705 #else
706 static inline void save_fscr_to_task(void) {};
707 #endif
708 
709 
710 void __init early_init_devtree(void *params)
711 {
712 	phys_addr_t limit;
713 
714 	DBG(" -> early_init_devtree(%px)\n", params);
715 
716 	/* Too early to BUG_ON(), do it by hand */
717 	if (!early_init_dt_verify(params))
718 		panic("BUG: Failed verifying flat device tree, bad version?");
719 
720 #ifdef CONFIG_PPC_RTAS
721 	/* Some machines might need RTAS info for debugging, grab it now. */
722 	of_scan_flat_dt(early_init_dt_scan_rtas, NULL);
723 #endif
724 
725 #ifdef CONFIG_PPC_POWERNV
726 	/* Some machines might need OPAL info for debugging, grab it now. */
727 	of_scan_flat_dt(early_init_dt_scan_opal, NULL);
728 
729 	/* Scan tree for ultravisor feature */
730 	of_scan_flat_dt(early_init_dt_scan_ultravisor, NULL);
731 #endif
732 
733 #if defined(CONFIG_FA_DUMP) || defined(CONFIG_PRESERVE_FA_DUMP)
734 	/* scan tree to see if dump is active during last boot */
735 	of_scan_flat_dt(early_init_dt_scan_fw_dump, NULL);
736 #endif
737 
738 	/* Retrieve various informations from the /chosen node of the
739 	 * device-tree, including the platform type, initrd location and
740 	 * size, TCE reserve, and more ...
741 	 */
742 	of_scan_flat_dt(early_init_dt_scan_chosen_ppc, boot_command_line);
743 
744 	/* Scan memory nodes and rebuild MEMBLOCKs */
745 	of_scan_flat_dt(early_init_dt_scan_root, NULL);
746 	of_scan_flat_dt(early_init_dt_scan_memory_ppc, NULL);
747 
748 	parse_early_param();
749 
750 	/* make sure we've parsed cmdline for mem= before this */
751 	if (memory_limit)
752 		first_memblock_size = min_t(u64, first_memblock_size, memory_limit);
753 	setup_initial_memory_limit(memstart_addr, first_memblock_size);
754 	/* Reserve MEMBLOCK regions used by kernel, initrd, dt, etc... */
755 	memblock_reserve(PHYSICAL_START, __pa(klimit) - PHYSICAL_START);
756 	/* If relocatable, reserve first 32k for interrupt vectors etc. */
757 	if (PHYSICAL_START > MEMORY_START)
758 		memblock_reserve(MEMORY_START, 0x8000);
759 	reserve_kdump_trampoline();
760 #if defined(CONFIG_FA_DUMP) || defined(CONFIG_PRESERVE_FA_DUMP)
761 	/*
762 	 * If we fail to reserve memory for firmware-assisted dump then
763 	 * fallback to kexec based kdump.
764 	 */
765 	if (fadump_reserve_mem() == 0)
766 #endif
767 		reserve_crashkernel();
768 	early_reserve_mem();
769 
770 	/* Ensure that total memory size is page-aligned. */
771 	limit = ALIGN(memory_limit ?: memblock_phys_mem_size(), PAGE_SIZE);
772 	memblock_enforce_memory_limit(limit);
773 
774 	memblock_allow_resize();
775 	memblock_dump_all();
776 
777 	DBG("Phys. mem: %llx\n", (unsigned long long)memblock_phys_mem_size());
778 
779 	/* We may need to relocate the flat tree, do it now.
780 	 * FIXME .. and the initrd too? */
781 	move_device_tree();
782 
783 	allocate_paca_ptrs();
784 
785 	DBG("Scanning CPUs ...\n");
786 
787 	dt_cpu_ftrs_scan();
788 
789 	/* Retrieve CPU related informations from the flat tree
790 	 * (altivec support, boot CPU ID, ...)
791 	 */
792 	of_scan_flat_dt(early_init_dt_scan_cpus, NULL);
793 	if (boot_cpuid < 0) {
794 		printk("Failed to identify boot CPU !\n");
795 		BUG();
796 	}
797 
798 	save_fscr_to_task();
799 
800 #if defined(CONFIG_SMP) && defined(CONFIG_PPC64)
801 	/* We'll later wait for secondaries to check in; there are
802 	 * NCPUS-1 non-boot CPUs  :-)
803 	 */
804 	spinning_secondaries = boot_cpu_count - 1;
805 #endif
806 
807 	mmu_early_init_devtree();
808 
809 #ifdef CONFIG_PPC_POWERNV
810 	/* Scan and build the list of machine check recoverable ranges */
811 	of_scan_flat_dt(early_init_dt_scan_recoverable_ranges, NULL);
812 #endif
813 	epapr_paravirt_early_init();
814 
815 	/* Now try to figure out if we are running on LPAR and so on */
816 	pseries_probe_fw_features();
817 
818 #ifdef CONFIG_PPC_PS3
819 	/* Identify PS3 firmware */
820 	if (of_flat_dt_is_compatible(of_get_flat_dt_root(), "sony,ps3"))
821 		powerpc_firmware_features |= FW_FEATURE_PS3_POSSIBLE;
822 #endif
823 
824 	tm_init();
825 
826 	DBG(" <- early_init_devtree()\n");
827 }
828 
829 #ifdef CONFIG_RELOCATABLE
830 /*
831  * This function run before early_init_devtree, so we have to init
832  * initial_boot_params.
833  */
834 void __init early_get_first_memblock_info(void *params, phys_addr_t *size)
835 {
836 	/* Setup flat device-tree pointer */
837 	initial_boot_params = params;
838 
839 	/*
840 	 * Scan the memory nodes and set add_mem_to_memblock to 0 to avoid
841 	 * mess the memblock.
842 	 */
843 	add_mem_to_memblock = 0;
844 	of_scan_flat_dt(early_init_dt_scan_root, NULL);
845 	of_scan_flat_dt(early_init_dt_scan_memory_ppc, NULL);
846 	add_mem_to_memblock = 1;
847 
848 	if (size)
849 		*size = first_memblock_size;
850 }
851 #endif
852 
853 /*******
854  *
855  * New implementation of the OF "find" APIs, return a refcounted
856  * object, call of_node_put() when done.  The device tree and list
857  * are protected by a rw_lock.
858  *
859  * Note that property management will need some locking as well,
860  * this isn't dealt with yet.
861  *
862  *******/
863 
864 /**
865  * of_get_ibm_chip_id - Returns the IBM "chip-id" of a device
866  * @np: device node of the device
867  *
868  * This looks for a property "ibm,chip-id" in the node or any
869  * of its parents and returns its content, or -1 if it cannot
870  * be found.
871  */
872 int of_get_ibm_chip_id(struct device_node *np)
873 {
874 	of_node_get(np);
875 	while (np) {
876 		u32 chip_id;
877 
878 		/*
879 		 * Skiboot may produce memory nodes that contain more than one
880 		 * cell in chip-id, we only read the first one here.
881 		 */
882 		if (!of_property_read_u32(np, "ibm,chip-id", &chip_id)) {
883 			of_node_put(np);
884 			return chip_id;
885 		}
886 
887 		np = of_get_next_parent(np);
888 	}
889 	return -1;
890 }
891 EXPORT_SYMBOL(of_get_ibm_chip_id);
892 
893 /**
894  * cpu_to_chip_id - Return the cpus chip-id
895  * @cpu: The logical cpu number.
896  *
897  * Return the value of the ibm,chip-id property corresponding to the given
898  * logical cpu number. If the chip-id can not be found, returns -1.
899  */
900 int cpu_to_chip_id(int cpu)
901 {
902 	struct device_node *np;
903 
904 	np = of_get_cpu_node(cpu, NULL);
905 	if (!np)
906 		return -1;
907 
908 	of_node_put(np);
909 	return of_get_ibm_chip_id(np);
910 }
911 EXPORT_SYMBOL(cpu_to_chip_id);
912 
913 bool arch_match_cpu_phys_id(int cpu, u64 phys_id)
914 {
915 #ifdef CONFIG_SMP
916 	/*
917 	 * Early firmware scanning must use this rather than
918 	 * get_hard_smp_processor_id because we don't have pacas allocated
919 	 * until memory topology is discovered.
920 	 */
921 	if (cpu_to_phys_id != NULL)
922 		return (int)phys_id == cpu_to_phys_id[cpu];
923 #endif
924 
925 	return (int)phys_id == get_hard_smp_processor_id(cpu);
926 }
927