xref: /openbmc/linux/arch/powerpc/kernel/prom.c (revision 66127f0d)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * Procedures for creating, accessing and interpreting the device tree.
4  *
5  * Paul Mackerras	August 1996.
6  * Copyright (C) 1996-2005 Paul Mackerras.
7  *
8  *  Adapted for 64bit PowerPC by Dave Engebretsen and Peter Bergner.
9  *    {engebret|bergner}@us.ibm.com
10  */
11 
12 #undef DEBUG
13 
14 #include <linux/kernel.h>
15 #include <linux/string.h>
16 #include <linux/init.h>
17 #include <linux/threads.h>
18 #include <linux/spinlock.h>
19 #include <linux/types.h>
20 #include <linux/pci.h>
21 #include <linux/delay.h>
22 #include <linux/initrd.h>
23 #include <linux/bitops.h>
24 #include <linux/export.h>
25 #include <linux/kexec.h>
26 #include <linux/irq.h>
27 #include <linux/memblock.h>
28 #include <linux/of.h>
29 #include <linux/of_fdt.h>
30 #include <linux/libfdt.h>
31 #include <linux/cpu.h>
32 #include <linux/pgtable.h>
33 #include <linux/seq_buf.h>
34 
35 #include <asm/rtas.h>
36 #include <asm/page.h>
37 #include <asm/processor.h>
38 #include <asm/irq.h>
39 #include <asm/io.h>
40 #include <asm/kdump.h>
41 #include <asm/smp.h>
42 #include <asm/mmu.h>
43 #include <asm/paca.h>
44 #include <asm/powernv.h>
45 #include <asm/iommu.h>
46 #include <asm/btext.h>
47 #include <asm/sections.h>
48 #include <asm/setup.h>
49 #include <asm/pci-bridge.h>
50 #include <asm/kexec.h>
51 #include <asm/opal.h>
52 #include <asm/fadump.h>
53 #include <asm/epapr_hcalls.h>
54 #include <asm/firmware.h>
55 #include <asm/dt_cpu_ftrs.h>
56 #include <asm/drmem.h>
57 #include <asm/ultravisor.h>
58 #include <asm/prom.h>
59 #include <asm/plpks.h>
60 
61 #include <mm/mmu_decl.h>
62 
63 #ifdef DEBUG
64 #define DBG(fmt...) printk(KERN_ERR fmt)
65 #else
66 #define DBG(fmt...)
67 #endif
68 
69 int *chip_id_lookup_table;
70 
71 #ifdef CONFIG_PPC64
72 int __initdata iommu_is_off;
73 int __initdata iommu_force_on;
74 unsigned long tce_alloc_start, tce_alloc_end;
75 u64 ppc64_rma_size;
76 unsigned int boot_cpu_node_count __ro_after_init;
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(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 	    !memblock_is_memory(start + size - 1) ||
129 	    overlaps_crashkernel(start, size) || overlaps_initrd(start, size)) {
130 		p = memblock_alloc_raw(size, PAGE_SIZE);
131 		if (!p)
132 			panic("Failed to allocate %lu bytes to move device tree\n",
133 			      size);
134 		memcpy(p, initial_boot_params, size);
135 		initial_boot_params = p;
136 		DBG("Moved device tree to 0x%px\n", p);
137 	}
138 
139 	DBG("<- move_device_tree\n");
140 }
141 
142 /*
143  * ibm,pa/pi-features is a per-cpu property that contains a string of
144  * attribute descriptors, each of which has a 2 byte header plus up
145  * to 254 bytes worth of processor attribute bits.  First header
146  * byte specifies the number of bytes following the header.
147  * Second header byte is an "attribute-specifier" type, of which
148  * zero is the only currently-defined value.
149  * Implementation:  Pass in the byte and bit offset for the feature
150  * that we are interested in.  The function will return -1 if the
151  * pa-features property is missing, or a 1/0 to indicate if the feature
152  * is supported/not supported.  Note that the bit numbers are
153  * big-endian to match the definition in PAPR.
154  */
155 struct ibm_feature {
156 	unsigned long	cpu_features;	/* CPU_FTR_xxx bit */
157 	unsigned long	mmu_features;	/* MMU_FTR_xxx bit */
158 	unsigned int	cpu_user_ftrs;	/* PPC_FEATURE_xxx bit */
159 	unsigned int	cpu_user_ftrs2;	/* PPC_FEATURE2_xxx bit */
160 	unsigned char	pabyte;		/* byte number in ibm,pa/pi-features */
161 	unsigned char	pabit;		/* bit number (big-endian) */
162 	unsigned char	invert;		/* if 1, pa bit set => clear feature */
163 };
164 
165 static struct ibm_feature ibm_pa_features[] __initdata = {
166 	{ .pabyte = 0,  .pabit = 0, .cpu_user_ftrs = PPC_FEATURE_HAS_MMU },
167 	{ .pabyte = 0,  .pabit = 1, .cpu_user_ftrs = PPC_FEATURE_HAS_FPU },
168 	{ .pabyte = 0,  .pabit = 3, .cpu_features  = CPU_FTR_CTRL },
169 	{ .pabyte = 0,  .pabit = 6, .cpu_features  = CPU_FTR_NOEXECUTE },
170 	{ .pabyte = 1,  .pabit = 2, .mmu_features  = MMU_FTR_CI_LARGE_PAGE },
171 #ifdef CONFIG_PPC_RADIX_MMU
172 	{ .pabyte = 40, .pabit = 0, .mmu_features  = MMU_FTR_TYPE_RADIX | MMU_FTR_GTSE },
173 #endif
174 	{ .pabyte = 5,  .pabit = 0, .cpu_features  = CPU_FTR_REAL_LE,
175 				    .cpu_user_ftrs = PPC_FEATURE_TRUE_LE },
176 	/*
177 	 * If the kernel doesn't support TM (ie CONFIG_PPC_TRANSACTIONAL_MEM=n),
178 	 * we don't want to turn on TM here, so we use the *_COMP versions
179 	 * which are 0 if the kernel doesn't support TM.
180 	 */
181 	{ .pabyte = 22, .pabit = 0, .cpu_features = CPU_FTR_TM_COMP,
182 	  .cpu_user_ftrs2 = PPC_FEATURE2_HTM_COMP | PPC_FEATURE2_HTM_NOSC_COMP },
183 
184 	{ .pabyte = 64, .pabit = 0, .cpu_features = CPU_FTR_DAWR1 },
185 	{ .pabyte = 68, .pabit = 5, .cpu_features = CPU_FTR_DEXCR_NPHIE },
186 };
187 
188 /*
189  * ibm,pi-features property provides the support of processor specific
190  * options not described in ibm,pa-features. Right now use byte 0, bit 3
191  * which indicates the occurrence of DSI interrupt when the paste operation
192  * on the suspended NX window.
193  */
194 static struct ibm_feature ibm_pi_features[] __initdata = {
195 	{ .pabyte = 0, .pabit = 3, .mmu_features  = MMU_FTR_NX_DSI },
196 };
197 
198 static void __init scan_features(unsigned long node, const unsigned char *ftrs,
199 				 unsigned long tablelen,
200 				 struct ibm_feature *fp,
201 				 unsigned long ft_size)
202 {
203 	unsigned long i, len, bit;
204 
205 	/* find descriptor with type == 0 */
206 	for (;;) {
207 		if (tablelen < 3)
208 			return;
209 		len = 2 + ftrs[0];
210 		if (tablelen < len)
211 			return;		/* descriptor 0 not found */
212 		if (ftrs[1] == 0)
213 			break;
214 		tablelen -= len;
215 		ftrs += len;
216 	}
217 
218 	/* loop over bits we know about */
219 	for (i = 0; i < ft_size; ++i, ++fp) {
220 		if (fp->pabyte >= ftrs[0])
221 			continue;
222 		bit = (ftrs[2 + fp->pabyte] >> (7 - fp->pabit)) & 1;
223 		if (bit ^ fp->invert) {
224 			cur_cpu_spec->cpu_features |= fp->cpu_features;
225 			cur_cpu_spec->cpu_user_features |= fp->cpu_user_ftrs;
226 			cur_cpu_spec->cpu_user_features2 |= fp->cpu_user_ftrs2;
227 			cur_cpu_spec->mmu_features |= fp->mmu_features;
228 		} else {
229 			cur_cpu_spec->cpu_features &= ~fp->cpu_features;
230 			cur_cpu_spec->cpu_user_features &= ~fp->cpu_user_ftrs;
231 			cur_cpu_spec->cpu_user_features2 &= ~fp->cpu_user_ftrs2;
232 			cur_cpu_spec->mmu_features &= ~fp->mmu_features;
233 		}
234 	}
235 }
236 
237 static void __init check_cpu_features(unsigned long node, char *name,
238 				      struct ibm_feature *fp,
239 				      unsigned long size)
240 {
241 	const unsigned char *pa_ftrs;
242 	int tablelen;
243 
244 	pa_ftrs = of_get_flat_dt_prop(node, name, &tablelen);
245 	if (pa_ftrs == NULL)
246 		return;
247 
248 	scan_features(node, pa_ftrs, tablelen, fp, size);
249 }
250 
251 #ifdef CONFIG_PPC_64S_HASH_MMU
252 static void __init init_mmu_slb_size(unsigned long node)
253 {
254 	const __be32 *slb_size_ptr;
255 
256 	slb_size_ptr = of_get_flat_dt_prop(node, "slb-size", NULL) ? :
257 			of_get_flat_dt_prop(node, "ibm,slb-size", NULL);
258 
259 	if (slb_size_ptr)
260 		mmu_slb_size = be32_to_cpup(slb_size_ptr);
261 }
262 #else
263 #define init_mmu_slb_size(node) do { } while(0)
264 #endif
265 
266 static struct feature_property {
267 	const char *name;
268 	u32 min_value;
269 	unsigned long cpu_feature;
270 	unsigned long cpu_user_ftr;
271 } feature_properties[] __initdata = {
272 #ifdef CONFIG_ALTIVEC
273 	{"altivec", 0, CPU_FTR_ALTIVEC, PPC_FEATURE_HAS_ALTIVEC},
274 	{"ibm,vmx", 1, CPU_FTR_ALTIVEC, PPC_FEATURE_HAS_ALTIVEC},
275 #endif /* CONFIG_ALTIVEC */
276 #ifdef CONFIG_VSX
277 	/* Yes, this _really_ is ibm,vmx == 2 to enable VSX */
278 	{"ibm,vmx", 2, CPU_FTR_VSX, PPC_FEATURE_HAS_VSX},
279 #endif /* CONFIG_VSX */
280 #ifdef CONFIG_PPC64
281 	{"ibm,dfp", 1, 0, PPC_FEATURE_HAS_DFP},
282 	{"ibm,purr", 1, CPU_FTR_PURR, 0},
283 	{"ibm,spurr", 1, CPU_FTR_SPURR, 0},
284 #endif /* CONFIG_PPC64 */
285 };
286 
287 #if defined(CONFIG_44x) && defined(CONFIG_PPC_FPU)
288 static __init void identical_pvr_fixup(unsigned long node)
289 {
290 	unsigned int pvr;
291 	const char *model = of_get_flat_dt_prop(node, "model", NULL);
292 
293 	/*
294 	 * Since 440GR(x)/440EP(x) processors have the same pvr,
295 	 * we check the node path and set bit 28 in the cur_cpu_spec
296 	 * pvr for EP(x) processor version. This bit is always 0 in
297 	 * the "real" pvr. Then we call identify_cpu again with
298 	 * the new logical pvr to enable FPU support.
299 	 */
300 	if (model && strstr(model, "440EP")) {
301 		pvr = cur_cpu_spec->pvr_value | 0x8;
302 		identify_cpu(0, pvr);
303 		DBG("Using logical pvr %x for %s\n", pvr, model);
304 	}
305 }
306 #else
307 #define identical_pvr_fixup(node) do { } while(0)
308 #endif
309 
310 static void __init check_cpu_feature_properties(unsigned long node)
311 {
312 	int i;
313 	struct feature_property *fp = feature_properties;
314 	const __be32 *prop;
315 
316 	for (i = 0; i < (int)ARRAY_SIZE(feature_properties); ++i, ++fp) {
317 		prop = of_get_flat_dt_prop(node, fp->name, NULL);
318 		if (prop && be32_to_cpup(prop) >= fp->min_value) {
319 			cur_cpu_spec->cpu_features |= fp->cpu_feature;
320 			cur_cpu_spec->cpu_user_features |= fp->cpu_user_ftr;
321 		}
322 	}
323 }
324 
325 static int __init early_init_dt_scan_cpus(unsigned long node,
326 					  const char *uname, int depth,
327 					  void *data)
328 {
329 	const char *type = of_get_flat_dt_prop(node, "device_type", NULL);
330 	const __be32 *cpu_version = NULL;
331 	const __be32 *prop;
332 	const __be32 *intserv;
333 	int i, nthreads;
334 	int len;
335 	int found = -1;
336 	int found_thread = 0;
337 
338 	/* We are scanning "cpu" nodes only */
339 	if (type == NULL || strcmp(type, "cpu") != 0)
340 		return 0;
341 
342 	if (IS_ENABLED(CONFIG_PPC64))
343 		boot_cpu_node_count++;
344 
345 	/* Get physical cpuid */
346 	intserv = of_get_flat_dt_prop(node, "ibm,ppc-interrupt-server#s", &len);
347 	if (!intserv)
348 		intserv = of_get_flat_dt_prop(node, "reg", &len);
349 
350 	nthreads = len / sizeof(int);
351 
352 	/*
353 	 * Now see if any of these threads match our boot cpu.
354 	 * NOTE: This must match the parsing done in smp_setup_cpu_maps.
355 	 */
356 	for (i = 0; i < nthreads; i++) {
357 		if (be32_to_cpu(intserv[i]) ==
358 			fdt_boot_cpuid_phys(initial_boot_params)) {
359 			found = boot_cpu_count;
360 			found_thread = i;
361 		}
362 #ifdef CONFIG_SMP
363 		/* logical cpu id is always 0 on UP kernels */
364 		boot_cpu_count++;
365 #endif
366 	}
367 
368 	/* Not the boot CPU */
369 	if (found < 0)
370 		return 0;
371 
372 	DBG("boot cpu: logical %d physical %d\n", found,
373 	    be32_to_cpu(intserv[found_thread]));
374 	boot_cpuid = found;
375 
376 	if (IS_ENABLED(CONFIG_PPC64))
377 		boot_cpu_hwid = be32_to_cpu(intserv[found_thread]);
378 
379 	if (nr_cpu_ids % nthreads != 0) {
380 		set_nr_cpu_ids(ALIGN(nr_cpu_ids, nthreads));
381 		pr_warn("nr_cpu_ids was not a multiple of threads_per_core, adjusted to %d\n",
382 			nr_cpu_ids);
383 	}
384 
385 	if (boot_cpuid >= nr_cpu_ids) {
386 		set_nr_cpu_ids(min(CONFIG_NR_CPUS, ALIGN(boot_cpuid + 1, nthreads)));
387 		pr_warn("Boot CPU %d >= nr_cpu_ids, adjusted nr_cpu_ids to %d\n",
388 			boot_cpuid, nr_cpu_ids);
389 	}
390 
391 	/*
392 	 * PAPR defines "logical" PVR values for cpus that
393 	 * meet various levels of the architecture:
394 	 * 0x0f000001	Architecture version 2.04
395 	 * 0x0f000002	Architecture version 2.05
396 	 * If the cpu-version property in the cpu node contains
397 	 * such a value, we call identify_cpu again with the
398 	 * logical PVR value in order to use the cpu feature
399 	 * bits appropriate for the architecture level.
400 	 *
401 	 * A POWER6 partition in "POWER6 architected" mode
402 	 * uses the 0x0f000002 PVR value; in POWER5+ mode
403 	 * it uses 0x0f000001.
404 	 *
405 	 * If we're using device tree CPU feature discovery then we don't
406 	 * support the cpu-version property, and it's the responsibility of the
407 	 * firmware/hypervisor to provide the correct feature set for the
408 	 * architecture level via the ibm,powerpc-cpu-features binding.
409 	 */
410 	if (!dt_cpu_ftrs_in_use()) {
411 		prop = of_get_flat_dt_prop(node, "cpu-version", NULL);
412 		if (prop && (be32_to_cpup(prop) & 0xff000000) == 0x0f000000) {
413 			identify_cpu(0, be32_to_cpup(prop));
414 			cpu_version = prop;
415 		}
416 
417 		check_cpu_feature_properties(node);
418 		check_cpu_features(node, "ibm,pa-features", ibm_pa_features,
419 				   ARRAY_SIZE(ibm_pa_features));
420 		check_cpu_features(node, "ibm,pi-features", ibm_pi_features,
421 				   ARRAY_SIZE(ibm_pi_features));
422 	}
423 
424 	identical_pvr_fixup(node);
425 
426 	// We can now add the CPU name & PVR to the hardware description
427 	seq_buf_printf(&ppc_hw_desc, "%s 0x%04lx ", cur_cpu_spec->cpu_name, mfspr(SPRN_PVR));
428 	if (cpu_version)
429 		seq_buf_printf(&ppc_hw_desc, "0x%04x ", be32_to_cpup(cpu_version));
430 
431 	init_mmu_slb_size(node);
432 
433 #ifdef CONFIG_PPC64
434 	if (nthreads == 1)
435 		cur_cpu_spec->cpu_features &= ~CPU_FTR_SMT;
436 	else if (!dt_cpu_ftrs_in_use())
437 		cur_cpu_spec->cpu_features |= CPU_FTR_SMT;
438 #endif
439 
440 	return 0;
441 }
442 
443 static int __init early_init_dt_scan_chosen_ppc(unsigned long node,
444 						const char *uname,
445 						int depth, void *data)
446 {
447 	const unsigned long *lprop; /* All these set by kernel, so no need to convert endian */
448 
449 	/* Use common scan routine to determine if this is the chosen node */
450 	if (early_init_dt_scan_chosen(data) < 0)
451 		return 0;
452 
453 #ifdef CONFIG_PPC64
454 	/* check if iommu is forced on or off */
455 	if (of_get_flat_dt_prop(node, "linux,iommu-off", NULL) != NULL)
456 		iommu_is_off = 1;
457 	if (of_get_flat_dt_prop(node, "linux,iommu-force-on", NULL) != NULL)
458 		iommu_force_on = 1;
459 #endif
460 
461 	/* mem=x on the command line is the preferred mechanism */
462 	lprop = of_get_flat_dt_prop(node, "linux,memory-limit", NULL);
463 	if (lprop)
464 		memory_limit = *lprop;
465 
466 #ifdef CONFIG_PPC64
467 	lprop = of_get_flat_dt_prop(node, "linux,tce-alloc-start", NULL);
468 	if (lprop)
469 		tce_alloc_start = *lprop;
470 	lprop = of_get_flat_dt_prop(node, "linux,tce-alloc-end", NULL);
471 	if (lprop)
472 		tce_alloc_end = *lprop;
473 #endif
474 
475 #ifdef CONFIG_KEXEC_CORE
476 	lprop = of_get_flat_dt_prop(node, "linux,crashkernel-base", NULL);
477 	if (lprop)
478 		crashk_res.start = *lprop;
479 
480 	lprop = of_get_flat_dt_prop(node, "linux,crashkernel-size", NULL);
481 	if (lprop)
482 		crashk_res.end = crashk_res.start + *lprop - 1;
483 #endif
484 
485 	/* break now */
486 	return 1;
487 }
488 
489 /*
490  * Compare the range against max mem limit and update
491  * size if it cross the limit.
492  */
493 
494 #ifdef CONFIG_SPARSEMEM
495 static bool __init validate_mem_limit(u64 base, u64 *size)
496 {
497 	u64 max_mem = 1UL << (MAX_PHYSMEM_BITS);
498 
499 	if (base >= max_mem)
500 		return false;
501 	if ((base + *size) > max_mem)
502 		*size = max_mem - base;
503 	return true;
504 }
505 #else
506 static bool __init validate_mem_limit(u64 base, u64 *size)
507 {
508 	return true;
509 }
510 #endif
511 
512 #ifdef CONFIG_PPC_PSERIES
513 /*
514  * Interpret the ibm dynamic reconfiguration memory LMBs.
515  * This contains a list of memory blocks along with NUMA affinity
516  * information.
517  */
518 static int  __init early_init_drmem_lmb(struct drmem_lmb *lmb,
519 					const __be32 **usm,
520 					void *data)
521 {
522 	u64 base, size;
523 	int is_kexec_kdump = 0, rngs;
524 
525 	base = lmb->base_addr;
526 	size = drmem_lmb_size();
527 	rngs = 1;
528 
529 	/*
530 	 * Skip this block if the reserved bit is set in flags
531 	 * or if the block is not assigned to this partition.
532 	 */
533 	if ((lmb->flags & DRCONF_MEM_RESERVED) ||
534 	    !(lmb->flags & DRCONF_MEM_ASSIGNED))
535 		return 0;
536 
537 	if (*usm)
538 		is_kexec_kdump = 1;
539 
540 	if (is_kexec_kdump) {
541 		/*
542 		 * For each memblock in ibm,dynamic-memory, a
543 		 * corresponding entry in linux,drconf-usable-memory
544 		 * property contains a counter 'p' followed by 'p'
545 		 * (base, size) duple. Now read the counter from
546 		 * linux,drconf-usable-memory property
547 		 */
548 		rngs = dt_mem_next_cell(dt_root_size_cells, usm);
549 		if (!rngs) /* there are no (base, size) duple */
550 			return 0;
551 	}
552 
553 	do {
554 		if (is_kexec_kdump) {
555 			base = dt_mem_next_cell(dt_root_addr_cells, usm);
556 			size = dt_mem_next_cell(dt_root_size_cells, usm);
557 		}
558 
559 		if (iommu_is_off) {
560 			if (base >= 0x80000000ul)
561 				continue;
562 			if ((base + size) > 0x80000000ul)
563 				size = 0x80000000ul - base;
564 		}
565 
566 		if (!validate_mem_limit(base, &size))
567 			continue;
568 
569 		DBG("Adding: %llx -> %llx\n", base, size);
570 		memblock_add(base, size);
571 
572 		if (lmb->flags & DRCONF_MEM_HOTREMOVABLE)
573 			memblock_mark_hotplug(base, size);
574 	} while (--rngs);
575 
576 	return 0;
577 }
578 #endif /* CONFIG_PPC_PSERIES */
579 
580 static int __init early_init_dt_scan_memory_ppc(void)
581 {
582 #ifdef CONFIG_PPC_PSERIES
583 	const void *fdt = initial_boot_params;
584 	int node = fdt_path_offset(fdt, "/ibm,dynamic-reconfiguration-memory");
585 
586 	if (node > 0)
587 		walk_drmem_lmbs_early(node, NULL, early_init_drmem_lmb);
588 
589 #endif
590 
591 	return early_init_dt_scan_memory();
592 }
593 
594 /*
595  * For a relocatable kernel, we need to get the memstart_addr first,
596  * then use it to calculate the virtual kernel start address. This has
597  * to happen at a very early stage (before machine_init). In this case,
598  * we just want to get the memstart_address and would not like to mess the
599  * memblock at this stage. So introduce a variable to skip the memblock_add()
600  * for this reason.
601  */
602 #ifdef CONFIG_RELOCATABLE
603 static int add_mem_to_memblock = 1;
604 #else
605 #define add_mem_to_memblock 1
606 #endif
607 
608 void __init early_init_dt_add_memory_arch(u64 base, u64 size)
609 {
610 #ifdef CONFIG_PPC64
611 	if (iommu_is_off) {
612 		if (base >= 0x80000000ul)
613 			return;
614 		if ((base + size) > 0x80000000ul)
615 			size = 0x80000000ul - base;
616 	}
617 #endif
618 	/* Keep track of the beginning of memory -and- the size of
619 	 * the very first block in the device-tree as it represents
620 	 * the RMA on ppc64 server
621 	 */
622 	if (base < memstart_addr) {
623 		memstart_addr = base;
624 		first_memblock_size = size;
625 	}
626 
627 	/* Add the chunk to the MEMBLOCK list */
628 	if (add_mem_to_memblock) {
629 		if (validate_mem_limit(base, &size))
630 			memblock_add(base, size);
631 	}
632 }
633 
634 static void __init early_reserve_mem_dt(void)
635 {
636 	unsigned long i, dt_root;
637 	int len;
638 	const __be32 *prop;
639 
640 	early_init_fdt_reserve_self();
641 	early_init_fdt_scan_reserved_mem();
642 
643 	dt_root = of_get_flat_dt_root();
644 
645 	prop = of_get_flat_dt_prop(dt_root, "reserved-ranges", &len);
646 
647 	if (!prop)
648 		return;
649 
650 	DBG("Found new-style reserved-ranges\n");
651 
652 	/* Each reserved range is an (address,size) pair, 2 cells each,
653 	 * totalling 4 cells per range. */
654 	for (i = 0; i < len / (sizeof(*prop) * 4); i++) {
655 		u64 base, size;
656 
657 		base = of_read_number(prop + (i * 4) + 0, 2);
658 		size = of_read_number(prop + (i * 4) + 2, 2);
659 
660 		if (size) {
661 			DBG("reserving: %llx -> %llx\n", base, size);
662 			memblock_reserve(base, size);
663 		}
664 	}
665 }
666 
667 static void __init early_reserve_mem(void)
668 {
669 	__be64 *reserve_map;
670 
671 	reserve_map = (__be64 *)(((unsigned long)initial_boot_params) +
672 			fdt_off_mem_rsvmap(initial_boot_params));
673 
674 	/* Look for the new "reserved-regions" property in the DT */
675 	early_reserve_mem_dt();
676 
677 #ifdef CONFIG_BLK_DEV_INITRD
678 	/* Then reserve the initrd, if any */
679 	if (initrd_start && (initrd_end > initrd_start)) {
680 		memblock_reserve(ALIGN_DOWN(__pa(initrd_start), PAGE_SIZE),
681 			ALIGN(initrd_end, PAGE_SIZE) -
682 			ALIGN_DOWN(initrd_start, PAGE_SIZE));
683 	}
684 #endif /* CONFIG_BLK_DEV_INITRD */
685 
686 	if (!IS_ENABLED(CONFIG_PPC32))
687 		return;
688 
689 	/*
690 	 * Handle the case where we might be booting from an old kexec
691 	 * image that setup the mem_rsvmap as pairs of 32-bit values
692 	 */
693 	if (be64_to_cpup(reserve_map) > 0xffffffffull) {
694 		u32 base_32, size_32;
695 		__be32 *reserve_map_32 = (__be32 *)reserve_map;
696 
697 		DBG("Found old 32-bit reserve map\n");
698 
699 		while (1) {
700 			base_32 = be32_to_cpup(reserve_map_32++);
701 			size_32 = be32_to_cpup(reserve_map_32++);
702 			if (size_32 == 0)
703 				break;
704 			DBG("reserving: %x -> %x\n", base_32, size_32);
705 			memblock_reserve(base_32, size_32);
706 		}
707 		return;
708 	}
709 }
710 
711 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
712 static bool tm_disabled __initdata;
713 
714 static int __init parse_ppc_tm(char *str)
715 {
716 	bool res;
717 
718 	if (kstrtobool(str, &res))
719 		return -EINVAL;
720 
721 	tm_disabled = !res;
722 
723 	return 0;
724 }
725 early_param("ppc_tm", parse_ppc_tm);
726 
727 static void __init tm_init(void)
728 {
729 	if (tm_disabled) {
730 		pr_info("Disabling hardware transactional memory (HTM)\n");
731 		cur_cpu_spec->cpu_user_features2 &=
732 			~(PPC_FEATURE2_HTM_NOSC | PPC_FEATURE2_HTM);
733 		cur_cpu_spec->cpu_features &= ~CPU_FTR_TM;
734 		return;
735 	}
736 
737 	pnv_tm_init();
738 }
739 #else
740 static void tm_init(void) { }
741 #endif /* CONFIG_PPC_TRANSACTIONAL_MEM */
742 
743 static int __init
744 early_init_dt_scan_model(unsigned long node, const char *uname,
745 			 int depth, void *data)
746 {
747 	const char *prop;
748 
749 	if (depth != 0)
750 		return 0;
751 
752 	prop = of_get_flat_dt_prop(node, "model", NULL);
753 	if (prop)
754 		seq_buf_printf(&ppc_hw_desc, "%s ", prop);
755 
756 	/* break now */
757 	return 1;
758 }
759 
760 #ifdef CONFIG_PPC64
761 static void __init save_fscr_to_task(void)
762 {
763 	/*
764 	 * Ensure the init_task (pid 0, aka swapper) uses the value of FSCR we
765 	 * have configured via the device tree features or via __init_FSCR().
766 	 * That value will then be propagated to pid 1 (init) and all future
767 	 * processes.
768 	 */
769 	if (early_cpu_has_feature(CPU_FTR_ARCH_207S))
770 		init_task.thread.fscr = mfspr(SPRN_FSCR);
771 }
772 #else
773 static inline void save_fscr_to_task(void) {}
774 #endif
775 
776 
777 void __init early_init_devtree(void *params)
778 {
779 	phys_addr_t limit;
780 
781 	DBG(" -> early_init_devtree(%px)\n", params);
782 
783 	/* Too early to BUG_ON(), do it by hand */
784 	if (!early_init_dt_verify(params, __pa(params)))
785 		panic("BUG: Failed verifying flat device tree, bad version?");
786 
787 	of_scan_flat_dt(early_init_dt_scan_model, NULL);
788 
789 #ifdef CONFIG_PPC_RTAS
790 	/* Some machines might need RTAS info for debugging, grab it now. */
791 	of_scan_flat_dt(early_init_dt_scan_rtas, NULL);
792 #endif
793 
794 #ifdef CONFIG_PPC_POWERNV
795 	/* Some machines might need OPAL info for debugging, grab it now. */
796 	of_scan_flat_dt(early_init_dt_scan_opal, NULL);
797 
798 	/* Scan tree for ultravisor feature */
799 	of_scan_flat_dt(early_init_dt_scan_ultravisor, NULL);
800 #endif
801 
802 #if defined(CONFIG_FA_DUMP) || defined(CONFIG_PRESERVE_FA_DUMP)
803 	/* scan tree to see if dump is active during last boot */
804 	of_scan_flat_dt(early_init_dt_scan_fw_dump, NULL);
805 #endif
806 
807 	/* Retrieve various informations from the /chosen node of the
808 	 * device-tree, including the platform type, initrd location and
809 	 * size, TCE reserve, and more ...
810 	 */
811 	of_scan_flat_dt(early_init_dt_scan_chosen_ppc, boot_command_line);
812 
813 	/* Scan memory nodes and rebuild MEMBLOCKs */
814 	early_init_dt_scan_root();
815 	early_init_dt_scan_memory_ppc();
816 
817 	/*
818 	 * As generic code authors expect to be able to use static keys
819 	 * in early_param() handlers, we initialize the static keys just
820 	 * before parsing early params (it's fine to call jump_label_init()
821 	 * more than once).
822 	 */
823 	jump_label_init();
824 	parse_early_param();
825 
826 	/* make sure we've parsed cmdline for mem= before this */
827 	if (memory_limit)
828 		first_memblock_size = min_t(u64, first_memblock_size, memory_limit);
829 	setup_initial_memory_limit(memstart_addr, first_memblock_size);
830 	/* Reserve MEMBLOCK regions used by kernel, initrd, dt, etc... */
831 	memblock_reserve(PHYSICAL_START, __pa(_end) - PHYSICAL_START);
832 	/* If relocatable, reserve first 32k for interrupt vectors etc. */
833 	if (PHYSICAL_START > MEMORY_START)
834 		memblock_reserve(MEMORY_START, 0x8000);
835 	reserve_kdump_trampoline();
836 #if defined(CONFIG_FA_DUMP) || defined(CONFIG_PRESERVE_FA_DUMP)
837 	/*
838 	 * If we fail to reserve memory for firmware-assisted dump then
839 	 * fallback to kexec based kdump.
840 	 */
841 	if (fadump_reserve_mem() == 0)
842 #endif
843 		reserve_crashkernel();
844 	early_reserve_mem();
845 
846 	/* Ensure that total memory size is page-aligned. */
847 	limit = ALIGN(memory_limit ?: memblock_phys_mem_size(), PAGE_SIZE);
848 	memblock_enforce_memory_limit(limit);
849 
850 #if defined(CONFIG_PPC_BOOK3S_64) && defined(CONFIG_PPC_4K_PAGES)
851 	if (!early_radix_enabled())
852 		memblock_cap_memory_range(0, 1UL << (H_MAX_PHYSMEM_BITS));
853 #endif
854 
855 	memblock_allow_resize();
856 	memblock_dump_all();
857 
858 	DBG("Phys. mem: %llx\n", (unsigned long long)memblock_phys_mem_size());
859 
860 	/* We may need to relocate the flat tree, do it now.
861 	 * FIXME .. and the initrd too? */
862 	move_device_tree();
863 
864 	DBG("Scanning CPUs ...\n");
865 
866 	dt_cpu_ftrs_scan();
867 
868 	/* Retrieve CPU related informations from the flat tree
869 	 * (altivec support, boot CPU ID, ...)
870 	 */
871 	of_scan_flat_dt(early_init_dt_scan_cpus, NULL);
872 	if (boot_cpuid < 0) {
873 		printk("Failed to identify boot CPU !\n");
874 		BUG();
875 	}
876 
877 	save_fscr_to_task();
878 
879 #if defined(CONFIG_SMP) && defined(CONFIG_PPC64)
880 	/* We'll later wait for secondaries to check in; there are
881 	 * NCPUS-1 non-boot CPUs  :-)
882 	 */
883 	spinning_secondaries = boot_cpu_count - 1;
884 #endif
885 
886 	mmu_early_init_devtree();
887 
888 #ifdef CONFIG_PPC_POWERNV
889 	/* Scan and build the list of machine check recoverable ranges */
890 	of_scan_flat_dt(early_init_dt_scan_recoverable_ranges, NULL);
891 #endif
892 	epapr_paravirt_early_init();
893 
894 	/* Now try to figure out if we are running on LPAR and so on */
895 	pseries_probe_fw_features();
896 
897 	/*
898 	 * Initialize pkey features and default AMR/IAMR values
899 	 */
900 	pkey_early_init_devtree();
901 
902 #ifdef CONFIG_PPC_PS3
903 	/* Identify PS3 firmware */
904 	if (of_flat_dt_is_compatible(of_get_flat_dt_root(), "sony,ps3"))
905 		powerpc_firmware_features |= FW_FEATURE_PS3_POSSIBLE;
906 #endif
907 
908 	/* If kexec left a PLPKS password in the DT, get it and clear it */
909 	plpks_early_init_devtree();
910 
911 	tm_init();
912 
913 	DBG(" <- early_init_devtree()\n");
914 }
915 
916 #ifdef CONFIG_RELOCATABLE
917 /*
918  * This function run before early_init_devtree, so we have to init
919  * initial_boot_params.
920  */
921 void __init early_get_first_memblock_info(void *params, phys_addr_t *size)
922 {
923 	/* Setup flat device-tree pointer */
924 	initial_boot_params = params;
925 
926 	/*
927 	 * Scan the memory nodes and set add_mem_to_memblock to 0 to avoid
928 	 * mess the memblock.
929 	 */
930 	add_mem_to_memblock = 0;
931 	early_init_dt_scan_root();
932 	early_init_dt_scan_memory_ppc();
933 	add_mem_to_memblock = 1;
934 
935 	if (size)
936 		*size = first_memblock_size;
937 }
938 #endif
939 
940 /*******
941  *
942  * New implementation of the OF "find" APIs, return a refcounted
943  * object, call of_node_put() when done.  The device tree and list
944  * are protected by a rw_lock.
945  *
946  * Note that property management will need some locking as well,
947  * this isn't dealt with yet.
948  *
949  *******/
950 
951 /**
952  * of_get_ibm_chip_id - Returns the IBM "chip-id" of a device
953  * @np: device node of the device
954  *
955  * This looks for a property "ibm,chip-id" in the node or any
956  * of its parents and returns its content, or -1 if it cannot
957  * be found.
958  */
959 int of_get_ibm_chip_id(struct device_node *np)
960 {
961 	of_node_get(np);
962 	while (np) {
963 		u32 chip_id;
964 
965 		/*
966 		 * Skiboot may produce memory nodes that contain more than one
967 		 * cell in chip-id, we only read the first one here.
968 		 */
969 		if (!of_property_read_u32(np, "ibm,chip-id", &chip_id)) {
970 			of_node_put(np);
971 			return chip_id;
972 		}
973 
974 		np = of_get_next_parent(np);
975 	}
976 	return -1;
977 }
978 EXPORT_SYMBOL(of_get_ibm_chip_id);
979 
980 /**
981  * cpu_to_chip_id - Return the cpus chip-id
982  * @cpu: The logical cpu number.
983  *
984  * Return the value of the ibm,chip-id property corresponding to the given
985  * logical cpu number. If the chip-id can not be found, returns -1.
986  */
987 int cpu_to_chip_id(int cpu)
988 {
989 	struct device_node *np;
990 	int ret = -1, idx;
991 
992 	idx = cpu / threads_per_core;
993 	if (chip_id_lookup_table && chip_id_lookup_table[idx] != -1)
994 		return chip_id_lookup_table[idx];
995 
996 	np = of_get_cpu_node(cpu, NULL);
997 	if (np) {
998 		ret = of_get_ibm_chip_id(np);
999 		of_node_put(np);
1000 
1001 		if (chip_id_lookup_table)
1002 			chip_id_lookup_table[idx] = ret;
1003 	}
1004 
1005 	return ret;
1006 }
1007 EXPORT_SYMBOL(cpu_to_chip_id);
1008 
1009 bool arch_match_cpu_phys_id(int cpu, u64 phys_id)
1010 {
1011 #ifdef CONFIG_SMP
1012 	/*
1013 	 * Early firmware scanning must use this rather than
1014 	 * get_hard_smp_processor_id because we don't have pacas allocated
1015 	 * until memory topology is discovered.
1016 	 */
1017 	if (cpu_to_phys_id != NULL)
1018 		return (int)phys_id == cpu_to_phys_id[cpu];
1019 #endif
1020 
1021 	return (int)phys_id == get_hard_smp_processor_id(cpu);
1022 }
1023