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