xref: /openbmc/linux/arch/powerpc/kernel/prom.c (revision a790cc3a)
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/setup.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 #include <asm/prom.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 int *chip_id_lookup_table;
68 
69 #ifdef CONFIG_PPC64
70 int __initdata iommu_is_off;
71 int __initdata iommu_force_on;
72 unsigned long tce_alloc_start, tce_alloc_end;
73 u64 ppc64_rma_size;
74 #endif
75 static phys_addr_t first_memblock_size;
76 static int __initdata boot_cpu_count;
77 
78 static int __init early_parse_mem(char *p)
79 {
80 	if (!p)
81 		return 1;
82 
83 	memory_limit = PAGE_ALIGN(memparse(p, &p));
84 	DBG("memory limit = 0x%llx\n", memory_limit);
85 
86 	return 0;
87 }
88 early_param("mem", early_parse_mem);
89 
90 /*
91  * overlaps_initrd - check for overlap with page aligned extension of
92  * initrd.
93  */
94 static inline int overlaps_initrd(unsigned long start, unsigned long size)
95 {
96 #ifdef CONFIG_BLK_DEV_INITRD
97 	if (!initrd_start)
98 		return 0;
99 
100 	return	(start + size) > ALIGN_DOWN(initrd_start, PAGE_SIZE) &&
101 			start <= ALIGN(initrd_end, PAGE_SIZE);
102 #else
103 	return 0;
104 #endif
105 }
106 
107 /**
108  * move_device_tree - move tree to an unused area, if needed.
109  *
110  * The device tree may be allocated beyond our memory limit, or inside the
111  * crash kernel region for kdump, or within the page aligned range of initrd.
112  * If so, move it out of the way.
113  */
114 static void __init move_device_tree(void)
115 {
116 	unsigned long start, size;
117 	void *p;
118 
119 	DBG("-> move_device_tree\n");
120 
121 	start = __pa(initial_boot_params);
122 	size = fdt_totalsize(initial_boot_params);
123 
124 	if ((memory_limit && (start + size) > PHYSICAL_START + memory_limit) ||
125 	    !memblock_is_memory(start + size - 1) ||
126 	    overlaps_crashkernel(start, size) || overlaps_initrd(start, size)) {
127 		p = memblock_alloc_raw(size, PAGE_SIZE);
128 		if (!p)
129 			panic("Failed to allocate %lu bytes to move device tree\n",
130 			      size);
131 		memcpy(p, initial_boot_params, size);
132 		initial_boot_params = p;
133 		DBG("Moved device tree to 0x%px\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 | MMU_FTR_GTSE },
168 #endif
169 	{ .pabyte = 5,  .pabit = 0, .cpu_features  = CPU_FTR_REAL_LE,
170 				    .cpu_user_ftrs = PPC_FEATURE_TRUE_LE },
171 	/*
172 	 * If the kernel doesn't support TM (ie CONFIG_PPC_TRANSACTIONAL_MEM=n),
173 	 * we don't want to turn on TM here, so we use the *_COMP versions
174 	 * which are 0 if the kernel doesn't support TM.
175 	 */
176 	{ .pabyte = 22, .pabit = 0, .cpu_features = CPU_FTR_TM_COMP,
177 	  .cpu_user_ftrs2 = PPC_FEATURE2_HTM_COMP | PPC_FEATURE2_HTM_NOSC_COMP },
178 
179 	{ .pabyte = 64, .pabit = 0, .cpu_features = CPU_FTR_DAWR1 },
180 };
181 
182 static void __init scan_features(unsigned long node, const unsigned char *ftrs,
183 				 unsigned long tablelen,
184 				 struct ibm_pa_feature *fp,
185 				 unsigned long ft_size)
186 {
187 	unsigned long i, len, bit;
188 
189 	/* find descriptor with type == 0 */
190 	for (;;) {
191 		if (tablelen < 3)
192 			return;
193 		len = 2 + ftrs[0];
194 		if (tablelen < len)
195 			return;		/* descriptor 0 not found */
196 		if (ftrs[1] == 0)
197 			break;
198 		tablelen -= len;
199 		ftrs += len;
200 	}
201 
202 	/* loop over bits we know about */
203 	for (i = 0; i < ft_size; ++i, ++fp) {
204 		if (fp->pabyte >= ftrs[0])
205 			continue;
206 		bit = (ftrs[2 + fp->pabyte] >> (7 - fp->pabit)) & 1;
207 		if (bit ^ fp->invert) {
208 			cur_cpu_spec->cpu_features |= fp->cpu_features;
209 			cur_cpu_spec->cpu_user_features |= fp->cpu_user_ftrs;
210 			cur_cpu_spec->cpu_user_features2 |= fp->cpu_user_ftrs2;
211 			cur_cpu_spec->mmu_features |= fp->mmu_features;
212 		} else {
213 			cur_cpu_spec->cpu_features &= ~fp->cpu_features;
214 			cur_cpu_spec->cpu_user_features &= ~fp->cpu_user_ftrs;
215 			cur_cpu_spec->cpu_user_features2 &= ~fp->cpu_user_ftrs2;
216 			cur_cpu_spec->mmu_features &= ~fp->mmu_features;
217 		}
218 	}
219 }
220 
221 static void __init check_cpu_pa_features(unsigned long node)
222 {
223 	const unsigned char *pa_ftrs;
224 	int tablelen;
225 
226 	pa_ftrs = of_get_flat_dt_prop(node, "ibm,pa-features", &tablelen);
227 	if (pa_ftrs == NULL)
228 		return;
229 
230 	scan_features(node, pa_ftrs, tablelen,
231 		      ibm_pa_features, ARRAY_SIZE(ibm_pa_features));
232 }
233 
234 #ifdef CONFIG_PPC_64S_HASH_MMU
235 static void __init init_mmu_slb_size(unsigned long node)
236 {
237 	const __be32 *slb_size_ptr;
238 
239 	slb_size_ptr = of_get_flat_dt_prop(node, "slb-size", NULL) ? :
240 			of_get_flat_dt_prop(node, "ibm,slb-size", NULL);
241 
242 	if (slb_size_ptr)
243 		mmu_slb_size = be32_to_cpup(slb_size_ptr);
244 }
245 #else
246 #define init_mmu_slb_size(node) do { } while(0)
247 #endif
248 
249 static struct feature_property {
250 	const char *name;
251 	u32 min_value;
252 	unsigned long cpu_feature;
253 	unsigned long cpu_user_ftr;
254 } feature_properties[] __initdata = {
255 #ifdef CONFIG_ALTIVEC
256 	{"altivec", 0, CPU_FTR_ALTIVEC, PPC_FEATURE_HAS_ALTIVEC},
257 	{"ibm,vmx", 1, CPU_FTR_ALTIVEC, PPC_FEATURE_HAS_ALTIVEC},
258 #endif /* CONFIG_ALTIVEC */
259 #ifdef CONFIG_VSX
260 	/* Yes, this _really_ is ibm,vmx == 2 to enable VSX */
261 	{"ibm,vmx", 2, CPU_FTR_VSX, PPC_FEATURE_HAS_VSX},
262 #endif /* CONFIG_VSX */
263 #ifdef CONFIG_PPC64
264 	{"ibm,dfp", 1, 0, PPC_FEATURE_HAS_DFP},
265 	{"ibm,purr", 1, CPU_FTR_PURR, 0},
266 	{"ibm,spurr", 1, CPU_FTR_SPURR, 0},
267 #endif /* CONFIG_PPC64 */
268 };
269 
270 #if defined(CONFIG_44x) && defined(CONFIG_PPC_FPU)
271 static __init void identical_pvr_fixup(unsigned long node)
272 {
273 	unsigned int pvr;
274 	const char *model = of_get_flat_dt_prop(node, "model", NULL);
275 
276 	/*
277 	 * Since 440GR(x)/440EP(x) processors have the same pvr,
278 	 * we check the node path and set bit 28 in the cur_cpu_spec
279 	 * pvr for EP(x) processor version. This bit is always 0 in
280 	 * the "real" pvr. Then we call identify_cpu again with
281 	 * the new logical pvr to enable FPU support.
282 	 */
283 	if (model && strstr(model, "440EP")) {
284 		pvr = cur_cpu_spec->pvr_value | 0x8;
285 		identify_cpu(0, pvr);
286 		DBG("Using logical pvr %x for %s\n", pvr, model);
287 	}
288 }
289 #else
290 #define identical_pvr_fixup(node) do { } while(0)
291 #endif
292 
293 static void __init check_cpu_feature_properties(unsigned long node)
294 {
295 	int i;
296 	struct feature_property *fp = feature_properties;
297 	const __be32 *prop;
298 
299 	for (i = 0; i < (int)ARRAY_SIZE(feature_properties); ++i, ++fp) {
300 		prop = of_get_flat_dt_prop(node, fp->name, NULL);
301 		if (prop && be32_to_cpup(prop) >= fp->min_value) {
302 			cur_cpu_spec->cpu_features |= fp->cpu_feature;
303 			cur_cpu_spec->cpu_user_features |= fp->cpu_user_ftr;
304 		}
305 	}
306 }
307 
308 static int __init early_init_dt_scan_cpus(unsigned long node,
309 					  const char *uname, int depth,
310 					  void *data)
311 {
312 	const char *type = of_get_flat_dt_prop(node, "device_type", NULL);
313 	const __be32 *prop;
314 	const __be32 *intserv;
315 	int i, nthreads;
316 	int len;
317 	int found = -1;
318 	int found_thread = 0;
319 
320 	/* We are scanning "cpu" nodes only */
321 	if (type == NULL || strcmp(type, "cpu") != 0)
322 		return 0;
323 
324 	/* Get physical cpuid */
325 	intserv = of_get_flat_dt_prop(node, "ibm,ppc-interrupt-server#s", &len);
326 	if (!intserv)
327 		intserv = of_get_flat_dt_prop(node, "reg", &len);
328 
329 	nthreads = len / sizeof(int);
330 
331 	/*
332 	 * Now see if any of these threads match our boot cpu.
333 	 * NOTE: This must match the parsing done in smp_setup_cpu_maps.
334 	 */
335 	for (i = 0; i < nthreads; i++) {
336 		if (be32_to_cpu(intserv[i]) ==
337 			fdt_boot_cpuid_phys(initial_boot_params)) {
338 			found = boot_cpu_count;
339 			found_thread = i;
340 		}
341 #ifdef CONFIG_SMP
342 		/* logical cpu id is always 0 on UP kernels */
343 		boot_cpu_count++;
344 #endif
345 	}
346 
347 	/* Not the boot CPU */
348 	if (found < 0)
349 		return 0;
350 
351 	DBG("boot cpu: logical %d physical %d\n", found,
352 	    be32_to_cpu(intserv[found_thread]));
353 	boot_cpuid = found;
354 
355 	// Pass the boot CPU's hard CPU id back to our caller
356 	*((u32 *)data) = be32_to_cpu(intserv[found_thread]);
357 
358 	/*
359 	 * PAPR defines "logical" PVR values for cpus that
360 	 * meet various levels of the architecture:
361 	 * 0x0f000001	Architecture version 2.04
362 	 * 0x0f000002	Architecture version 2.05
363 	 * If the cpu-version property in the cpu node contains
364 	 * such a value, we call identify_cpu again with the
365 	 * logical PVR value in order to use the cpu feature
366 	 * bits appropriate for the architecture level.
367 	 *
368 	 * A POWER6 partition in "POWER6 architected" mode
369 	 * uses the 0x0f000002 PVR value; in POWER5+ mode
370 	 * it uses 0x0f000001.
371 	 *
372 	 * If we're using device tree CPU feature discovery then we don't
373 	 * support the cpu-version property, and it's the responsibility of the
374 	 * firmware/hypervisor to provide the correct feature set for the
375 	 * architecture level via the ibm,powerpc-cpu-features binding.
376 	 */
377 	if (!dt_cpu_ftrs_in_use()) {
378 		prop = of_get_flat_dt_prop(node, "cpu-version", NULL);
379 		if (prop && (be32_to_cpup(prop) & 0xff000000) == 0x0f000000)
380 			identify_cpu(0, be32_to_cpup(prop));
381 
382 		check_cpu_feature_properties(node);
383 		check_cpu_pa_features(node);
384 	}
385 
386 	identical_pvr_fixup(node);
387 	init_mmu_slb_size(node);
388 
389 #ifdef CONFIG_PPC64
390 	if (nthreads == 1)
391 		cur_cpu_spec->cpu_features &= ~CPU_FTR_SMT;
392 	else if (!dt_cpu_ftrs_in_use())
393 		cur_cpu_spec->cpu_features |= CPU_FTR_SMT;
394 #endif
395 
396 	return 0;
397 }
398 
399 static int __init early_init_dt_scan_chosen_ppc(unsigned long node,
400 						const char *uname,
401 						int depth, void *data)
402 {
403 	const unsigned long *lprop; /* All these set by kernel, so no need to convert endian */
404 
405 	/* Use common scan routine to determine if this is the chosen node */
406 	if (early_init_dt_scan_chosen(data) < 0)
407 		return 0;
408 
409 #ifdef CONFIG_PPC64
410 	/* check if iommu is forced on or off */
411 	if (of_get_flat_dt_prop(node, "linux,iommu-off", NULL) != NULL)
412 		iommu_is_off = 1;
413 	if (of_get_flat_dt_prop(node, "linux,iommu-force-on", NULL) != NULL)
414 		iommu_force_on = 1;
415 #endif
416 
417 	/* mem=x on the command line is the preferred mechanism */
418 	lprop = of_get_flat_dt_prop(node, "linux,memory-limit", NULL);
419 	if (lprop)
420 		memory_limit = *lprop;
421 
422 #ifdef CONFIG_PPC64
423 	lprop = of_get_flat_dt_prop(node, "linux,tce-alloc-start", NULL);
424 	if (lprop)
425 		tce_alloc_start = *lprop;
426 	lprop = of_get_flat_dt_prop(node, "linux,tce-alloc-end", NULL);
427 	if (lprop)
428 		tce_alloc_end = *lprop;
429 #endif
430 
431 #ifdef CONFIG_KEXEC_CORE
432 	lprop = of_get_flat_dt_prop(node, "linux,crashkernel-base", NULL);
433 	if (lprop)
434 		crashk_res.start = *lprop;
435 
436 	lprop = of_get_flat_dt_prop(node, "linux,crashkernel-size", NULL);
437 	if (lprop)
438 		crashk_res.end = crashk_res.start + *lprop - 1;
439 #endif
440 
441 	/* break now */
442 	return 1;
443 }
444 
445 /*
446  * Compare the range against max mem limit and update
447  * size if it cross the limit.
448  */
449 
450 #ifdef CONFIG_SPARSEMEM
451 static bool __init validate_mem_limit(u64 base, u64 *size)
452 {
453 	u64 max_mem = 1UL << (MAX_PHYSMEM_BITS);
454 
455 	if (base >= max_mem)
456 		return false;
457 	if ((base + *size) > max_mem)
458 		*size = max_mem - base;
459 	return true;
460 }
461 #else
462 static bool __init validate_mem_limit(u64 base, u64 *size)
463 {
464 	return true;
465 }
466 #endif
467 
468 #ifdef CONFIG_PPC_PSERIES
469 /*
470  * Interpret the ibm dynamic reconfiguration memory LMBs.
471  * This contains a list of memory blocks along with NUMA affinity
472  * information.
473  */
474 static int  __init early_init_drmem_lmb(struct drmem_lmb *lmb,
475 					const __be32 **usm,
476 					void *data)
477 {
478 	u64 base, size;
479 	int is_kexec_kdump = 0, rngs;
480 
481 	base = lmb->base_addr;
482 	size = drmem_lmb_size();
483 	rngs = 1;
484 
485 	/*
486 	 * Skip this block if the reserved bit is set in flags
487 	 * or if the block is not assigned to this partition.
488 	 */
489 	if ((lmb->flags & DRCONF_MEM_RESERVED) ||
490 	    !(lmb->flags & DRCONF_MEM_ASSIGNED))
491 		return 0;
492 
493 	if (*usm)
494 		is_kexec_kdump = 1;
495 
496 	if (is_kexec_kdump) {
497 		/*
498 		 * For each memblock in ibm,dynamic-memory, a
499 		 * corresponding entry in linux,drconf-usable-memory
500 		 * property contains a counter 'p' followed by 'p'
501 		 * (base, size) duple. Now read the counter from
502 		 * linux,drconf-usable-memory property
503 		 */
504 		rngs = dt_mem_next_cell(dt_root_size_cells, usm);
505 		if (!rngs) /* there are no (base, size) duple */
506 			return 0;
507 	}
508 
509 	do {
510 		if (is_kexec_kdump) {
511 			base = dt_mem_next_cell(dt_root_addr_cells, usm);
512 			size = dt_mem_next_cell(dt_root_size_cells, usm);
513 		}
514 
515 		if (iommu_is_off) {
516 			if (base >= 0x80000000ul)
517 				continue;
518 			if ((base + size) > 0x80000000ul)
519 				size = 0x80000000ul - base;
520 		}
521 
522 		if (!validate_mem_limit(base, &size))
523 			continue;
524 
525 		DBG("Adding: %llx -> %llx\n", base, size);
526 		memblock_add(base, size);
527 
528 		if (lmb->flags & DRCONF_MEM_HOTREMOVABLE)
529 			memblock_mark_hotplug(base, size);
530 	} while (--rngs);
531 
532 	return 0;
533 }
534 #endif /* CONFIG_PPC_PSERIES */
535 
536 static int __init early_init_dt_scan_memory_ppc(void)
537 {
538 #ifdef CONFIG_PPC_PSERIES
539 	const void *fdt = initial_boot_params;
540 	int node = fdt_path_offset(fdt, "/ibm,dynamic-reconfiguration-memory");
541 
542 	if (node > 0)
543 		walk_drmem_lmbs_early(node, NULL, early_init_drmem_lmb);
544 
545 #endif
546 
547 	return early_init_dt_scan_memory();
548 }
549 
550 /*
551  * For a relocatable kernel, we need to get the memstart_addr first,
552  * then use it to calculate the virtual kernel start address. This has
553  * to happen at a very early stage (before machine_init). In this case,
554  * we just want to get the memstart_address and would not like to mess the
555  * memblock at this stage. So introduce a variable to skip the memblock_add()
556  * for this reason.
557  */
558 #ifdef CONFIG_RELOCATABLE
559 static int add_mem_to_memblock = 1;
560 #else
561 #define add_mem_to_memblock 1
562 #endif
563 
564 void __init early_init_dt_add_memory_arch(u64 base, u64 size)
565 {
566 #ifdef CONFIG_PPC64
567 	if (iommu_is_off) {
568 		if (base >= 0x80000000ul)
569 			return;
570 		if ((base + size) > 0x80000000ul)
571 			size = 0x80000000ul - base;
572 	}
573 #endif
574 	/* Keep track of the beginning of memory -and- the size of
575 	 * the very first block in the device-tree as it represents
576 	 * the RMA on ppc64 server
577 	 */
578 	if (base < memstart_addr) {
579 		memstart_addr = base;
580 		first_memblock_size = size;
581 	}
582 
583 	/* Add the chunk to the MEMBLOCK list */
584 	if (add_mem_to_memblock) {
585 		if (validate_mem_limit(base, &size))
586 			memblock_add(base, size);
587 	}
588 }
589 
590 static void __init early_reserve_mem_dt(void)
591 {
592 	unsigned long i, dt_root;
593 	int len;
594 	const __be32 *prop;
595 
596 	early_init_fdt_reserve_self();
597 	early_init_fdt_scan_reserved_mem();
598 
599 	dt_root = of_get_flat_dt_root();
600 
601 	prop = of_get_flat_dt_prop(dt_root, "reserved-ranges", &len);
602 
603 	if (!prop)
604 		return;
605 
606 	DBG("Found new-style reserved-ranges\n");
607 
608 	/* Each reserved range is an (address,size) pair, 2 cells each,
609 	 * totalling 4 cells per range. */
610 	for (i = 0; i < len / (sizeof(*prop) * 4); i++) {
611 		u64 base, size;
612 
613 		base = of_read_number(prop + (i * 4) + 0, 2);
614 		size = of_read_number(prop + (i * 4) + 2, 2);
615 
616 		if (size) {
617 			DBG("reserving: %llx -> %llx\n", base, size);
618 			memblock_reserve(base, size);
619 		}
620 	}
621 }
622 
623 static void __init early_reserve_mem(void)
624 {
625 	__be64 *reserve_map;
626 
627 	reserve_map = (__be64 *)(((unsigned long)initial_boot_params) +
628 			fdt_off_mem_rsvmap(initial_boot_params));
629 
630 	/* Look for the new "reserved-regions" property in the DT */
631 	early_reserve_mem_dt();
632 
633 #ifdef CONFIG_BLK_DEV_INITRD
634 	/* Then reserve the initrd, if any */
635 	if (initrd_start && (initrd_end > initrd_start)) {
636 		memblock_reserve(ALIGN_DOWN(__pa(initrd_start), PAGE_SIZE),
637 			ALIGN(initrd_end, PAGE_SIZE) -
638 			ALIGN_DOWN(initrd_start, PAGE_SIZE));
639 	}
640 #endif /* CONFIG_BLK_DEV_INITRD */
641 
642 	if (!IS_ENABLED(CONFIG_PPC32))
643 		return;
644 
645 	/*
646 	 * Handle the case where we might be booting from an old kexec
647 	 * image that setup the mem_rsvmap as pairs of 32-bit values
648 	 */
649 	if (be64_to_cpup(reserve_map) > 0xffffffffull) {
650 		u32 base_32, size_32;
651 		__be32 *reserve_map_32 = (__be32 *)reserve_map;
652 
653 		DBG("Found old 32-bit reserve map\n");
654 
655 		while (1) {
656 			base_32 = be32_to_cpup(reserve_map_32++);
657 			size_32 = be32_to_cpup(reserve_map_32++);
658 			if (size_32 == 0)
659 				break;
660 			DBG("reserving: %x -> %x\n", base_32, size_32);
661 			memblock_reserve(base_32, size_32);
662 		}
663 		return;
664 	}
665 }
666 
667 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
668 static bool tm_disabled __initdata;
669 
670 static int __init parse_ppc_tm(char *str)
671 {
672 	bool res;
673 
674 	if (kstrtobool(str, &res))
675 		return -EINVAL;
676 
677 	tm_disabled = !res;
678 
679 	return 0;
680 }
681 early_param("ppc_tm", parse_ppc_tm);
682 
683 static void __init tm_init(void)
684 {
685 	if (tm_disabled) {
686 		pr_info("Disabling hardware transactional memory (HTM)\n");
687 		cur_cpu_spec->cpu_user_features2 &=
688 			~(PPC_FEATURE2_HTM_NOSC | PPC_FEATURE2_HTM);
689 		cur_cpu_spec->cpu_features &= ~CPU_FTR_TM;
690 		return;
691 	}
692 
693 	pnv_tm_init();
694 }
695 #else
696 static void tm_init(void) { }
697 #endif /* CONFIG_PPC_TRANSACTIONAL_MEM */
698 
699 #ifdef CONFIG_PPC64
700 static void __init save_fscr_to_task(void)
701 {
702 	/*
703 	 * Ensure the init_task (pid 0, aka swapper) uses the value of FSCR we
704 	 * have configured via the device tree features or via __init_FSCR().
705 	 * That value will then be propagated to pid 1 (init) and all future
706 	 * processes.
707 	 */
708 	if (early_cpu_has_feature(CPU_FTR_ARCH_207S))
709 		init_task.thread.fscr = mfspr(SPRN_FSCR);
710 }
711 #else
712 static inline void save_fscr_to_task(void) {}
713 #endif
714 
715 
716 void __init early_init_devtree(void *params)
717 {
718 	u32 boot_cpu_hwid;
719 	phys_addr_t limit;
720 
721 	DBG(" -> early_init_devtree(%px)\n", params);
722 
723 	/* Too early to BUG_ON(), do it by hand */
724 	if (!early_init_dt_verify(params))
725 		panic("BUG: Failed verifying flat device tree, bad version?");
726 
727 #ifdef CONFIG_PPC_RTAS
728 	/* Some machines might need RTAS info for debugging, grab it now. */
729 	of_scan_flat_dt(early_init_dt_scan_rtas, NULL);
730 #endif
731 
732 #ifdef CONFIG_PPC_POWERNV
733 	/* Some machines might need OPAL info for debugging, grab it now. */
734 	of_scan_flat_dt(early_init_dt_scan_opal, NULL);
735 
736 	/* Scan tree for ultravisor feature */
737 	of_scan_flat_dt(early_init_dt_scan_ultravisor, NULL);
738 #endif
739 
740 #if defined(CONFIG_FA_DUMP) || defined(CONFIG_PRESERVE_FA_DUMP)
741 	/* scan tree to see if dump is active during last boot */
742 	of_scan_flat_dt(early_init_dt_scan_fw_dump, NULL);
743 #endif
744 
745 	/* Retrieve various informations from the /chosen node of the
746 	 * device-tree, including the platform type, initrd location and
747 	 * size, TCE reserve, and more ...
748 	 */
749 	of_scan_flat_dt(early_init_dt_scan_chosen_ppc, boot_command_line);
750 
751 	/* Scan memory nodes and rebuild MEMBLOCKs */
752 	early_init_dt_scan_root();
753 	early_init_dt_scan_memory_ppc();
754 
755 	/*
756 	 * As generic code authors expect to be able to use static keys
757 	 * in early_param() handlers, we initialize the static keys just
758 	 * before parsing early params (it's fine to call jump_label_init()
759 	 * more than once).
760 	 */
761 	jump_label_init();
762 	parse_early_param();
763 
764 	/* make sure we've parsed cmdline for mem= before this */
765 	if (memory_limit)
766 		first_memblock_size = min_t(u64, first_memblock_size, memory_limit);
767 	setup_initial_memory_limit(memstart_addr, first_memblock_size);
768 	/* Reserve MEMBLOCK regions used by kernel, initrd, dt, etc... */
769 	memblock_reserve(PHYSICAL_START, __pa(_end) - PHYSICAL_START);
770 	/* If relocatable, reserve first 32k for interrupt vectors etc. */
771 	if (PHYSICAL_START > MEMORY_START)
772 		memblock_reserve(MEMORY_START, 0x8000);
773 	reserve_kdump_trampoline();
774 #if defined(CONFIG_FA_DUMP) || defined(CONFIG_PRESERVE_FA_DUMP)
775 	/*
776 	 * If we fail to reserve memory for firmware-assisted dump then
777 	 * fallback to kexec based kdump.
778 	 */
779 	if (fadump_reserve_mem() == 0)
780 #endif
781 		reserve_crashkernel();
782 	early_reserve_mem();
783 
784 	/* Ensure that total memory size is page-aligned. */
785 	limit = ALIGN(memory_limit ?: memblock_phys_mem_size(), PAGE_SIZE);
786 	memblock_enforce_memory_limit(limit);
787 
788 #if defined(CONFIG_PPC_BOOK3S_64) && defined(CONFIG_PPC_4K_PAGES)
789 	if (!early_radix_enabled())
790 		memblock_cap_memory_range(0, 1UL << (H_MAX_PHYSMEM_BITS));
791 #endif
792 
793 	memblock_allow_resize();
794 	memblock_dump_all();
795 
796 	DBG("Phys. mem: %llx\n", (unsigned long long)memblock_phys_mem_size());
797 
798 	/* We may need to relocate the flat tree, do it now.
799 	 * FIXME .. and the initrd too? */
800 	move_device_tree();
801 
802 	DBG("Scanning CPUs ...\n");
803 
804 	dt_cpu_ftrs_scan();
805 
806 	/* Retrieve CPU related informations from the flat tree
807 	 * (altivec support, boot CPU ID, ...)
808 	 */
809 	of_scan_flat_dt(early_init_dt_scan_cpus, &boot_cpu_hwid);
810 	if (boot_cpuid < 0) {
811 		printk("Failed to identify boot CPU !\n");
812 		BUG();
813 	}
814 
815 	save_fscr_to_task();
816 
817 #if defined(CONFIG_SMP) && defined(CONFIG_PPC64)
818 	/* We'll later wait for secondaries to check in; there are
819 	 * NCPUS-1 non-boot CPUs  :-)
820 	 */
821 	spinning_secondaries = boot_cpu_count - 1;
822 #endif
823 
824 	mmu_early_init_devtree();
825 
826 	// NB. paca is not installed until later in early_setup()
827 	allocate_paca_ptrs();
828 	allocate_paca(boot_cpuid);
829 	set_hard_smp_processor_id(boot_cpuid, boot_cpu_hwid);
830 
831 #ifdef CONFIG_PPC_POWERNV
832 	/* Scan and build the list of machine check recoverable ranges */
833 	of_scan_flat_dt(early_init_dt_scan_recoverable_ranges, NULL);
834 #endif
835 	epapr_paravirt_early_init();
836 
837 	/* Now try to figure out if we are running on LPAR and so on */
838 	pseries_probe_fw_features();
839 
840 	/*
841 	 * Initialize pkey features and default AMR/IAMR values
842 	 */
843 	pkey_early_init_devtree();
844 
845 #ifdef CONFIG_PPC_PS3
846 	/* Identify PS3 firmware */
847 	if (of_flat_dt_is_compatible(of_get_flat_dt_root(), "sony,ps3"))
848 		powerpc_firmware_features |= FW_FEATURE_PS3_POSSIBLE;
849 #endif
850 
851 	tm_init();
852 
853 	DBG(" <- early_init_devtree()\n");
854 }
855 
856 #ifdef CONFIG_RELOCATABLE
857 /*
858  * This function run before early_init_devtree, so we have to init
859  * initial_boot_params.
860  */
861 void __init early_get_first_memblock_info(void *params, phys_addr_t *size)
862 {
863 	/* Setup flat device-tree pointer */
864 	initial_boot_params = params;
865 
866 	/*
867 	 * Scan the memory nodes and set add_mem_to_memblock to 0 to avoid
868 	 * mess the memblock.
869 	 */
870 	add_mem_to_memblock = 0;
871 	early_init_dt_scan_root();
872 	early_init_dt_scan_memory_ppc();
873 	add_mem_to_memblock = 1;
874 
875 	if (size)
876 		*size = first_memblock_size;
877 }
878 #endif
879 
880 /*******
881  *
882  * New implementation of the OF "find" APIs, return a refcounted
883  * object, call of_node_put() when done.  The device tree and list
884  * are protected by a rw_lock.
885  *
886  * Note that property management will need some locking as well,
887  * this isn't dealt with yet.
888  *
889  *******/
890 
891 /**
892  * of_get_ibm_chip_id - Returns the IBM "chip-id" of a device
893  * @np: device node of the device
894  *
895  * This looks for a property "ibm,chip-id" in the node or any
896  * of its parents and returns its content, or -1 if it cannot
897  * be found.
898  */
899 int of_get_ibm_chip_id(struct device_node *np)
900 {
901 	of_node_get(np);
902 	while (np) {
903 		u32 chip_id;
904 
905 		/*
906 		 * Skiboot may produce memory nodes that contain more than one
907 		 * cell in chip-id, we only read the first one here.
908 		 */
909 		if (!of_property_read_u32(np, "ibm,chip-id", &chip_id)) {
910 			of_node_put(np);
911 			return chip_id;
912 		}
913 
914 		np = of_get_next_parent(np);
915 	}
916 	return -1;
917 }
918 EXPORT_SYMBOL(of_get_ibm_chip_id);
919 
920 /**
921  * cpu_to_chip_id - Return the cpus chip-id
922  * @cpu: The logical cpu number.
923  *
924  * Return the value of the ibm,chip-id property corresponding to the given
925  * logical cpu number. If the chip-id can not be found, returns -1.
926  */
927 int cpu_to_chip_id(int cpu)
928 {
929 	struct device_node *np;
930 	int ret = -1, idx;
931 
932 	idx = cpu / threads_per_core;
933 	if (chip_id_lookup_table && chip_id_lookup_table[idx] != -1)
934 		return chip_id_lookup_table[idx];
935 
936 	np = of_get_cpu_node(cpu, NULL);
937 	if (np) {
938 		ret = of_get_ibm_chip_id(np);
939 		of_node_put(np);
940 
941 		if (chip_id_lookup_table)
942 			chip_id_lookup_table[idx] = ret;
943 	}
944 
945 	return ret;
946 }
947 EXPORT_SYMBOL(cpu_to_chip_id);
948 
949 bool arch_match_cpu_phys_id(int cpu, u64 phys_id)
950 {
951 #ifdef CONFIG_SMP
952 	/*
953 	 * Early firmware scanning must use this rather than
954 	 * get_hard_smp_processor_id because we don't have pacas allocated
955 	 * until memory topology is discovered.
956 	 */
957 	if (cpu_to_phys_id != NULL)
958 		return (int)phys_id == cpu_to_phys_id[cpu];
959 #endif
960 
961 	return (int)phys_id == get_hard_smp_processor_id(cpu);
962 }
963