xref: /openbmc/linux/arch/powerpc/kernel/prom.c (revision 1bcca2b1)
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 
34 #include <asm/rtas.h>
35 #include <asm/page.h>
36 #include <asm/processor.h>
37 #include <asm/irq.h>
38 #include <asm/io.h>
39 #include <asm/kdump.h>
40 #include <asm/smp.h>
41 #include <asm/mmu.h>
42 #include <asm/paca.h>
43 #include <asm/powernv.h>
44 #include <asm/iommu.h>
45 #include <asm/btext.h>
46 #include <asm/sections.h>
47 #include <asm/machdep.h>
48 #include <asm/pci-bridge.h>
49 #include <asm/kexec.h>
50 #include <asm/opal.h>
51 #include <asm/fadump.h>
52 #include <asm/epapr_hcalls.h>
53 #include <asm/firmware.h>
54 #include <asm/dt_cpu_ftrs.h>
55 #include <asm/drmem.h>
56 #include <asm/ultravisor.h>
57 
58 #include <mm/mmu_decl.h>
59 
60 #ifdef DEBUG
61 #define DBG(fmt...) printk(KERN_ERR fmt)
62 #else
63 #define DBG(fmt...)
64 #endif
65 
66 int *chip_id_lookup_table;
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 | MMU_FTR_GTSE },
167 #endif
168 	{ .pabyte = 5,  .pabit = 0, .cpu_features  = CPU_FTR_REAL_LE,
169 				    .cpu_user_ftrs = PPC_FEATURE_TRUE_LE },
170 	/*
171 	 * If the kernel doesn't support TM (ie CONFIG_PPC_TRANSACTIONAL_MEM=n),
172 	 * we don't want to turn on TM here, so we use the *_COMP versions
173 	 * which are 0 if the kernel doesn't support TM.
174 	 */
175 	{ .pabyte = 22, .pabit = 0, .cpu_features = CPU_FTR_TM_COMP,
176 	  .cpu_user_ftrs2 = PPC_FEATURE2_HTM_COMP | PPC_FEATURE2_HTM_NOSC_COMP },
177 
178 	{ .pabyte = 64, .pabit = 0, .cpu_features = CPU_FTR_DAWR1 },
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_64S_HASH_MMU
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 __init 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 	int i;
295 	struct feature_property *fp = feature_properties;
296 	const __be32 *prop;
297 
298 	for (i = 0; i < (int)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 		if (be32_to_cpu(intserv[i]) ==
336 			fdt_boot_cpuid_phys(initial_boot_params)) {
337 			found = boot_cpu_count;
338 			found_thread = i;
339 		}
340 #ifdef CONFIG_SMP
341 		/* logical cpu id is always 0 on UP kernels */
342 		boot_cpu_count++;
343 #endif
344 	}
345 
346 	/* Not the boot CPU */
347 	if (found < 0)
348 		return 0;
349 
350 	DBG("boot cpu: logical %d physical %d\n", found,
351 	    be32_to_cpu(intserv[found_thread]));
352 	boot_cpuid = found;
353 
354 	// Pass the boot CPU's hard CPU id back to our caller
355 	*((u32 *)data) = be32_to_cpu(intserv[found_thread]);
356 
357 	/*
358 	 * PAPR defines "logical" PVR values for cpus that
359 	 * meet various levels of the architecture:
360 	 * 0x0f000001	Architecture version 2.04
361 	 * 0x0f000002	Architecture version 2.05
362 	 * If the cpu-version property in the cpu node contains
363 	 * such a value, we call identify_cpu again with the
364 	 * logical PVR value in order to use the cpu feature
365 	 * bits appropriate for the architecture level.
366 	 *
367 	 * A POWER6 partition in "POWER6 architected" mode
368 	 * uses the 0x0f000002 PVR value; in POWER5+ mode
369 	 * it uses 0x0f000001.
370 	 *
371 	 * If we're using device tree CPU feature discovery then we don't
372 	 * support the cpu-version property, and it's the responsibility of the
373 	 * firmware/hypervisor to provide the correct feature set for the
374 	 * architecture level via the ibm,powerpc-cpu-features binding.
375 	 */
376 	if (!dt_cpu_ftrs_in_use()) {
377 		prop = of_get_flat_dt_prop(node, "cpu-version", NULL);
378 		if (prop && (be32_to_cpup(prop) & 0xff000000) == 0x0f000000)
379 			identify_cpu(0, be32_to_cpup(prop));
380 
381 		check_cpu_feature_properties(node);
382 		check_cpu_pa_features(node);
383 	}
384 
385 	identical_pvr_fixup(node);
386 	init_mmu_slb_size(node);
387 
388 #ifdef CONFIG_PPC64
389 	if (nthreads == 1)
390 		cur_cpu_spec->cpu_features &= ~CPU_FTR_SMT;
391 	else if (!dt_cpu_ftrs_in_use())
392 		cur_cpu_spec->cpu_features |= CPU_FTR_SMT;
393 #endif
394 
395 	return 0;
396 }
397 
398 static int __init early_init_dt_scan_chosen_ppc(unsigned long node,
399 						const char *uname,
400 						int depth, void *data)
401 {
402 	const unsigned long *lprop; /* All these set by kernel, so no need to convert endian */
403 
404 	/* Use common scan routine to determine if this is the chosen node */
405 	if (early_init_dt_scan_chosen(data) < 0)
406 		return 0;
407 
408 #ifdef CONFIG_PPC64
409 	/* check if iommu is forced on or off */
410 	if (of_get_flat_dt_prop(node, "linux,iommu-off", NULL) != NULL)
411 		iommu_is_off = 1;
412 	if (of_get_flat_dt_prop(node, "linux,iommu-force-on", NULL) != NULL)
413 		iommu_force_on = 1;
414 #endif
415 
416 	/* mem=x on the command line is the preferred mechanism */
417 	lprop = of_get_flat_dt_prop(node, "linux,memory-limit", NULL);
418 	if (lprop)
419 		memory_limit = *lprop;
420 
421 #ifdef CONFIG_PPC64
422 	lprop = of_get_flat_dt_prop(node, "linux,tce-alloc-start", NULL);
423 	if (lprop)
424 		tce_alloc_start = *lprop;
425 	lprop = of_get_flat_dt_prop(node, "linux,tce-alloc-end", NULL);
426 	if (lprop)
427 		tce_alloc_end = *lprop;
428 #endif
429 
430 #ifdef CONFIG_KEXEC_CORE
431 	lprop = of_get_flat_dt_prop(node, "linux,crashkernel-base", NULL);
432 	if (lprop)
433 		crashk_res.start = *lprop;
434 
435 	lprop = of_get_flat_dt_prop(node, "linux,crashkernel-size", NULL);
436 	if (lprop)
437 		crashk_res.end = crashk_res.start + *lprop - 1;
438 #endif
439 
440 	/* break now */
441 	return 1;
442 }
443 
444 /*
445  * Compare the range against max mem limit and update
446  * size if it cross the limit.
447  */
448 
449 #ifdef CONFIG_SPARSEMEM
450 static bool __init validate_mem_limit(u64 base, u64 *size)
451 {
452 	u64 max_mem = 1UL << (MAX_PHYSMEM_BITS);
453 
454 	if (base >= max_mem)
455 		return false;
456 	if ((base + *size) > max_mem)
457 		*size = max_mem - base;
458 	return true;
459 }
460 #else
461 static bool __init validate_mem_limit(u64 base, u64 *size)
462 {
463 	return true;
464 }
465 #endif
466 
467 #ifdef CONFIG_PPC_PSERIES
468 /*
469  * Interpret the ibm dynamic reconfiguration memory LMBs.
470  * This contains a list of memory blocks along with NUMA affinity
471  * information.
472  */
473 static int  __init early_init_drmem_lmb(struct drmem_lmb *lmb,
474 					const __be32 **usm,
475 					void *data)
476 {
477 	u64 base, size;
478 	int is_kexec_kdump = 0, rngs;
479 
480 	base = lmb->base_addr;
481 	size = drmem_lmb_size();
482 	rngs = 1;
483 
484 	/*
485 	 * Skip this block if the reserved bit is set in flags
486 	 * or if the block is not assigned to this partition.
487 	 */
488 	if ((lmb->flags & DRCONF_MEM_RESERVED) ||
489 	    !(lmb->flags & DRCONF_MEM_ASSIGNED))
490 		return 0;
491 
492 	if (*usm)
493 		is_kexec_kdump = 1;
494 
495 	if (is_kexec_kdump) {
496 		/*
497 		 * For each memblock in ibm,dynamic-memory, a
498 		 * corresponding entry in linux,drconf-usable-memory
499 		 * property contains a counter 'p' followed by 'p'
500 		 * (base, size) duple. Now read the counter from
501 		 * linux,drconf-usable-memory property
502 		 */
503 		rngs = dt_mem_next_cell(dt_root_size_cells, usm);
504 		if (!rngs) /* there are no (base, size) duple */
505 			return 0;
506 	}
507 
508 	do {
509 		if (is_kexec_kdump) {
510 			base = dt_mem_next_cell(dt_root_addr_cells, usm);
511 			size = dt_mem_next_cell(dt_root_size_cells, usm);
512 		}
513 
514 		if (iommu_is_off) {
515 			if (base >= 0x80000000ul)
516 				continue;
517 			if ((base + size) > 0x80000000ul)
518 				size = 0x80000000ul - base;
519 		}
520 
521 		if (!validate_mem_limit(base, &size))
522 			continue;
523 
524 		DBG("Adding: %llx -> %llx\n", base, size);
525 		memblock_add(base, size);
526 
527 		if (lmb->flags & DRCONF_MEM_HOTREMOVABLE)
528 			memblock_mark_hotplug(base, size);
529 	} while (--rngs);
530 
531 	return 0;
532 }
533 #endif /* CONFIG_PPC_PSERIES */
534 
535 static int __init early_init_dt_scan_memory_ppc(void)
536 {
537 #ifdef CONFIG_PPC_PSERIES
538 	const void *fdt = initial_boot_params;
539 	int node = fdt_path_offset(fdt, "/ibm,dynamic-reconfiguration-memory");
540 
541 	if (node > 0)
542 		walk_drmem_lmbs_early(node, NULL, early_init_drmem_lmb);
543 
544 #endif
545 
546 	return early_init_dt_scan_memory();
547 }
548 
549 /*
550  * For a relocatable kernel, we need to get the memstart_addr first,
551  * then use it to calculate the virtual kernel start address. This has
552  * to happen at a very early stage (before machine_init). In this case,
553  * we just want to get the memstart_address and would not like to mess the
554  * memblock at this stage. So introduce a variable to skip the memblock_add()
555  * for this reason.
556  */
557 #ifdef CONFIG_RELOCATABLE
558 static int add_mem_to_memblock = 1;
559 #else
560 #define add_mem_to_memblock 1
561 #endif
562 
563 void __init early_init_dt_add_memory_arch(u64 base, u64 size)
564 {
565 #ifdef CONFIG_PPC64
566 	if (iommu_is_off) {
567 		if (base >= 0x80000000ul)
568 			return;
569 		if ((base + size) > 0x80000000ul)
570 			size = 0x80000000ul - base;
571 	}
572 #endif
573 	/* Keep track of the beginning of memory -and- the size of
574 	 * the very first block in the device-tree as it represents
575 	 * the RMA on ppc64 server
576 	 */
577 	if (base < memstart_addr) {
578 		memstart_addr = base;
579 		first_memblock_size = size;
580 	}
581 
582 	/* Add the chunk to the MEMBLOCK list */
583 	if (add_mem_to_memblock) {
584 		if (validate_mem_limit(base, &size))
585 			memblock_add(base, size);
586 	}
587 }
588 
589 static void __init early_reserve_mem_dt(void)
590 {
591 	unsigned long i, dt_root;
592 	int len;
593 	const __be32 *prop;
594 
595 	early_init_fdt_reserve_self();
596 	early_init_fdt_scan_reserved_mem();
597 
598 	dt_root = of_get_flat_dt_root();
599 
600 	prop = of_get_flat_dt_prop(dt_root, "reserved-ranges", &len);
601 
602 	if (!prop)
603 		return;
604 
605 	DBG("Found new-style reserved-ranges\n");
606 
607 	/* Each reserved range is an (address,size) pair, 2 cells each,
608 	 * totalling 4 cells per range. */
609 	for (i = 0; i < len / (sizeof(*prop) * 4); i++) {
610 		u64 base, size;
611 
612 		base = of_read_number(prop + (i * 4) + 0, 2);
613 		size = of_read_number(prop + (i * 4) + 2, 2);
614 
615 		if (size) {
616 			DBG("reserving: %llx -> %llx\n", base, size);
617 			memblock_reserve(base, size);
618 		}
619 	}
620 }
621 
622 static void __init early_reserve_mem(void)
623 {
624 	__be64 *reserve_map;
625 
626 	reserve_map = (__be64 *)(((unsigned long)initial_boot_params) +
627 			fdt_off_mem_rsvmap(initial_boot_params));
628 
629 	/* Look for the new "reserved-regions" property in the DT */
630 	early_reserve_mem_dt();
631 
632 #ifdef CONFIG_BLK_DEV_INITRD
633 	/* Then reserve the initrd, if any */
634 	if (initrd_start && (initrd_end > initrd_start)) {
635 		memblock_reserve(ALIGN_DOWN(__pa(initrd_start), PAGE_SIZE),
636 			ALIGN(initrd_end, PAGE_SIZE) -
637 			ALIGN_DOWN(initrd_start, PAGE_SIZE));
638 	}
639 #endif /* CONFIG_BLK_DEV_INITRD */
640 
641 	if (!IS_ENABLED(CONFIG_PPC32))
642 		return;
643 
644 	/*
645 	 * Handle the case where we might be booting from an old kexec
646 	 * image that setup the mem_rsvmap as pairs of 32-bit values
647 	 */
648 	if (be64_to_cpup(reserve_map) > 0xffffffffull) {
649 		u32 base_32, size_32;
650 		__be32 *reserve_map_32 = (__be32 *)reserve_map;
651 
652 		DBG("Found old 32-bit reserve map\n");
653 
654 		while (1) {
655 			base_32 = be32_to_cpup(reserve_map_32++);
656 			size_32 = be32_to_cpup(reserve_map_32++);
657 			if (size_32 == 0)
658 				break;
659 			DBG("reserving: %x -> %x\n", base_32, size_32);
660 			memblock_reserve(base_32, size_32);
661 		}
662 		return;
663 	}
664 }
665 
666 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
667 static bool tm_disabled __initdata;
668 
669 static int __init parse_ppc_tm(char *str)
670 {
671 	bool res;
672 
673 	if (kstrtobool(str, &res))
674 		return -EINVAL;
675 
676 	tm_disabled = !res;
677 
678 	return 0;
679 }
680 early_param("ppc_tm", parse_ppc_tm);
681 
682 static void __init tm_init(void)
683 {
684 	if (tm_disabled) {
685 		pr_info("Disabling hardware transactional memory (HTM)\n");
686 		cur_cpu_spec->cpu_user_features2 &=
687 			~(PPC_FEATURE2_HTM_NOSC | PPC_FEATURE2_HTM);
688 		cur_cpu_spec->cpu_features &= ~CPU_FTR_TM;
689 		return;
690 	}
691 
692 	pnv_tm_init();
693 }
694 #else
695 static void tm_init(void) { }
696 #endif /* CONFIG_PPC_TRANSACTIONAL_MEM */
697 
698 #ifdef CONFIG_PPC64
699 static void __init save_fscr_to_task(void)
700 {
701 	/*
702 	 * Ensure the init_task (pid 0, aka swapper) uses the value of FSCR we
703 	 * have configured via the device tree features or via __init_FSCR().
704 	 * That value will then be propagated to pid 1 (init) and all future
705 	 * processes.
706 	 */
707 	if (early_cpu_has_feature(CPU_FTR_ARCH_207S))
708 		init_task.thread.fscr = mfspr(SPRN_FSCR);
709 }
710 #else
711 static inline void save_fscr_to_task(void) {}
712 #endif
713 
714 
715 void __init early_init_devtree(void *params)
716 {
717 	u32 boot_cpu_hwid;
718 	phys_addr_t limit;
719 
720 	DBG(" -> early_init_devtree(%px)\n", params);
721 
722 	/* Too early to BUG_ON(), do it by hand */
723 	if (!early_init_dt_verify(params))
724 		panic("BUG: Failed verifying flat device tree, bad version?");
725 
726 #ifdef CONFIG_PPC_RTAS
727 	/* Some machines might need RTAS info for debugging, grab it now. */
728 	of_scan_flat_dt(early_init_dt_scan_rtas, NULL);
729 #endif
730 
731 #ifdef CONFIG_PPC_POWERNV
732 	/* Some machines might need OPAL info for debugging, grab it now. */
733 	of_scan_flat_dt(early_init_dt_scan_opal, NULL);
734 
735 	/* Scan tree for ultravisor feature */
736 	of_scan_flat_dt(early_init_dt_scan_ultravisor, NULL);
737 #endif
738 
739 #if defined(CONFIG_FA_DUMP) || defined(CONFIG_PRESERVE_FA_DUMP)
740 	/* scan tree to see if dump is active during last boot */
741 	of_scan_flat_dt(early_init_dt_scan_fw_dump, NULL);
742 #endif
743 
744 	/* Retrieve various informations from the /chosen node of the
745 	 * device-tree, including the platform type, initrd location and
746 	 * size, TCE reserve, and more ...
747 	 */
748 	of_scan_flat_dt(early_init_dt_scan_chosen_ppc, boot_command_line);
749 
750 	/* Scan memory nodes and rebuild MEMBLOCKs */
751 	early_init_dt_scan_root();
752 	early_init_dt_scan_memory_ppc();
753 
754 	parse_early_param();
755 
756 	/* make sure we've parsed cmdline for mem= before this */
757 	if (memory_limit)
758 		first_memblock_size = min_t(u64, first_memblock_size, memory_limit);
759 	setup_initial_memory_limit(memstart_addr, first_memblock_size);
760 	/* Reserve MEMBLOCK regions used by kernel, initrd, dt, etc... */
761 	memblock_reserve(PHYSICAL_START, __pa(_end) - PHYSICAL_START);
762 	/* If relocatable, reserve first 32k for interrupt vectors etc. */
763 	if (PHYSICAL_START > MEMORY_START)
764 		memblock_reserve(MEMORY_START, 0x8000);
765 	reserve_kdump_trampoline();
766 #if defined(CONFIG_FA_DUMP) || defined(CONFIG_PRESERVE_FA_DUMP)
767 	/*
768 	 * If we fail to reserve memory for firmware-assisted dump then
769 	 * fallback to kexec based kdump.
770 	 */
771 	if (fadump_reserve_mem() == 0)
772 #endif
773 		reserve_crashkernel();
774 	early_reserve_mem();
775 
776 	/* Ensure that total memory size is page-aligned. */
777 	limit = ALIGN(memory_limit ?: memblock_phys_mem_size(), PAGE_SIZE);
778 	memblock_enforce_memory_limit(limit);
779 
780 #if defined(CONFIG_PPC_BOOK3S_64) && defined(CONFIG_PPC_4K_PAGES)
781 	if (!early_radix_enabled())
782 		memblock_cap_memory_range(0, 1UL << (H_MAX_PHYSMEM_BITS));
783 #endif
784 
785 	memblock_allow_resize();
786 	memblock_dump_all();
787 
788 	DBG("Phys. mem: %llx\n", (unsigned long long)memblock_phys_mem_size());
789 
790 	/* We may need to relocate the flat tree, do it now.
791 	 * FIXME .. and the initrd too? */
792 	move_device_tree();
793 
794 	DBG("Scanning CPUs ...\n");
795 
796 	dt_cpu_ftrs_scan();
797 
798 	/* Retrieve CPU related informations from the flat tree
799 	 * (altivec support, boot CPU ID, ...)
800 	 */
801 	of_scan_flat_dt(early_init_dt_scan_cpus, &boot_cpu_hwid);
802 	if (boot_cpuid < 0) {
803 		printk("Failed to identify boot CPU !\n");
804 		BUG();
805 	}
806 
807 	save_fscr_to_task();
808 
809 #if defined(CONFIG_SMP) && defined(CONFIG_PPC64)
810 	/* We'll later wait for secondaries to check in; there are
811 	 * NCPUS-1 non-boot CPUs  :-)
812 	 */
813 	spinning_secondaries = boot_cpu_count - 1;
814 #endif
815 
816 	mmu_early_init_devtree();
817 
818 	// NB. paca is not installed until later in early_setup()
819 	allocate_paca_ptrs();
820 	allocate_paca(boot_cpuid);
821 	set_hard_smp_processor_id(boot_cpuid, boot_cpu_hwid);
822 
823 #ifdef CONFIG_PPC_POWERNV
824 	/* Scan and build the list of machine check recoverable ranges */
825 	of_scan_flat_dt(early_init_dt_scan_recoverable_ranges, NULL);
826 #endif
827 	epapr_paravirt_early_init();
828 
829 	/* Now try to figure out if we are running on LPAR and so on */
830 	pseries_probe_fw_features();
831 
832 	/*
833 	 * Initialize pkey features and default AMR/IAMR values
834 	 */
835 	pkey_early_init_devtree();
836 
837 #ifdef CONFIG_PPC_PS3
838 	/* Identify PS3 firmware */
839 	if (of_flat_dt_is_compatible(of_get_flat_dt_root(), "sony,ps3"))
840 		powerpc_firmware_features |= FW_FEATURE_PS3_POSSIBLE;
841 #endif
842 
843 	tm_init();
844 
845 	DBG(" <- early_init_devtree()\n");
846 }
847 
848 #ifdef CONFIG_RELOCATABLE
849 /*
850  * This function run before early_init_devtree, so we have to init
851  * initial_boot_params.
852  */
853 void __init early_get_first_memblock_info(void *params, phys_addr_t *size)
854 {
855 	/* Setup flat device-tree pointer */
856 	initial_boot_params = params;
857 
858 	/*
859 	 * Scan the memory nodes and set add_mem_to_memblock to 0 to avoid
860 	 * mess the memblock.
861 	 */
862 	add_mem_to_memblock = 0;
863 	early_init_dt_scan_root();
864 	early_init_dt_scan_memory_ppc();
865 	add_mem_to_memblock = 1;
866 
867 	if (size)
868 		*size = first_memblock_size;
869 }
870 #endif
871 
872 /*******
873  *
874  * New implementation of the OF "find" APIs, return a refcounted
875  * object, call of_node_put() when done.  The device tree and list
876  * are protected by a rw_lock.
877  *
878  * Note that property management will need some locking as well,
879  * this isn't dealt with yet.
880  *
881  *******/
882 
883 /**
884  * of_get_ibm_chip_id - Returns the IBM "chip-id" of a device
885  * @np: device node of the device
886  *
887  * This looks for a property "ibm,chip-id" in the node or any
888  * of its parents and returns its content, or -1 if it cannot
889  * be found.
890  */
891 int of_get_ibm_chip_id(struct device_node *np)
892 {
893 	of_node_get(np);
894 	while (np) {
895 		u32 chip_id;
896 
897 		/*
898 		 * Skiboot may produce memory nodes that contain more than one
899 		 * cell in chip-id, we only read the first one here.
900 		 */
901 		if (!of_property_read_u32(np, "ibm,chip-id", &chip_id)) {
902 			of_node_put(np);
903 			return chip_id;
904 		}
905 
906 		np = of_get_next_parent(np);
907 	}
908 	return -1;
909 }
910 EXPORT_SYMBOL(of_get_ibm_chip_id);
911 
912 /**
913  * cpu_to_chip_id - Return the cpus chip-id
914  * @cpu: The logical cpu number.
915  *
916  * Return the value of the ibm,chip-id property corresponding to the given
917  * logical cpu number. If the chip-id can not be found, returns -1.
918  */
919 int cpu_to_chip_id(int cpu)
920 {
921 	struct device_node *np;
922 	int ret = -1, idx;
923 
924 	idx = cpu / threads_per_core;
925 	if (chip_id_lookup_table && chip_id_lookup_table[idx] != -1)
926 		return chip_id_lookup_table[idx];
927 
928 	np = of_get_cpu_node(cpu, NULL);
929 	if (np) {
930 		ret = of_get_ibm_chip_id(np);
931 		of_node_put(np);
932 
933 		if (chip_id_lookup_table)
934 			chip_id_lookup_table[idx] = ret;
935 	}
936 
937 	return ret;
938 }
939 EXPORT_SYMBOL(cpu_to_chip_id);
940 
941 bool arch_match_cpu_phys_id(int cpu, u64 phys_id)
942 {
943 #ifdef CONFIG_SMP
944 	/*
945 	 * Early firmware scanning must use this rather than
946 	 * get_hard_smp_processor_id because we don't have pacas allocated
947 	 * until memory topology is discovered.
948 	 */
949 	if (cpu_to_phys_id != NULL)
950 		return (int)phys_id == cpu_to_phys_id[cpu];
951 #endif
952 
953 	return (int)phys_id == get_hard_smp_processor_id(cpu);
954 }
955