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 
60 #include <mm/mmu_decl.h>
61 
62 #ifdef DEBUG
63 #define DBG(fmt...) printk(KERN_ERR fmt)
64 #else
65 #define DBG(fmt...)
66 #endif
67 
68 int *chip_id_lookup_table;
69 
70 #ifdef CONFIG_PPC64
71 int __initdata iommu_is_off;
72 int __initdata iommu_force_on;
73 unsigned long tce_alloc_start, tce_alloc_end;
74 u64 ppc64_rma_size;
75 #endif
76 static phys_addr_t first_memblock_size;
77 static int __initdata boot_cpu_count;
78 
early_parse_mem(char * p)79 static int __init early_parse_mem(char *p)
80 {
81 	if (!p)
82 		return 1;
83 
84 	memory_limit = PAGE_ALIGN(memparse(p, &p));
85 	DBG("memory limit = 0x%llx\n", memory_limit);
86 
87 	return 0;
88 }
89 early_param("mem", early_parse_mem);
90 
91 /*
92  * overlaps_initrd - check for overlap with page aligned extension of
93  * initrd.
94  */
overlaps_initrd(unsigned long start,unsigned long size)95 static inline int overlaps_initrd(unsigned long start, unsigned long size)
96 {
97 #ifdef CONFIG_BLK_DEV_INITRD
98 	if (!initrd_start)
99 		return 0;
100 
101 	return	(start + size) > ALIGN_DOWN(initrd_start, PAGE_SIZE) &&
102 			start <= ALIGN(initrd_end, PAGE_SIZE);
103 #else
104 	return 0;
105 #endif
106 }
107 
108 /**
109  * move_device_tree - move tree to an unused area, if needed.
110  *
111  * The device tree may be allocated beyond our memory limit, or inside the
112  * crash kernel region for kdump, or within the page aligned range of initrd.
113  * If so, move it out of the way.
114  */
move_device_tree(void)115 static void __init move_device_tree(void)
116 {
117 	unsigned long start, size;
118 	void *p;
119 
120 	DBG("-> move_device_tree\n");
121 
122 	start = __pa(initial_boot_params);
123 	size = fdt_totalsize(initial_boot_params);
124 
125 	if ((memory_limit && (start + size) > PHYSICAL_START + memory_limit) ||
126 	    !memblock_is_memory(start + size - 1) ||
127 	    overlaps_crashkernel(start, size) || overlaps_initrd(start, size)) {
128 		p = memblock_alloc_raw(size, PAGE_SIZE);
129 		if (!p)
130 			panic("Failed to allocate %lu bytes to move device tree\n",
131 			      size);
132 		memcpy(p, initial_boot_params, size);
133 		initial_boot_params = p;
134 		DBG("Moved device tree to 0x%px\n", p);
135 	}
136 
137 	DBG("<- move_device_tree\n");
138 }
139 
140 /*
141  * ibm,pa/pi-features is a per-cpu property that contains a string of
142  * attribute descriptors, each of which has a 2 byte header plus up
143  * to 254 bytes worth of processor attribute bits.  First header
144  * byte specifies the number of bytes following the header.
145  * Second header byte is an "attribute-specifier" type, of which
146  * zero is the only currently-defined value.
147  * Implementation:  Pass in the byte and bit offset for the feature
148  * that we are interested in.  The function will return -1 if the
149  * pa-features property is missing, or a 1/0 to indicate if the feature
150  * is supported/not supported.  Note that the bit numbers are
151  * big-endian to match the definition in PAPR.
152  */
153 struct ibm_feature {
154 	unsigned long	cpu_features;	/* CPU_FTR_xxx bit */
155 	unsigned long	mmu_features;	/* MMU_FTR_xxx bit */
156 	unsigned int	cpu_user_ftrs;	/* PPC_FEATURE_xxx bit */
157 	unsigned int	cpu_user_ftrs2;	/* PPC_FEATURE2_xxx bit */
158 	unsigned char	pabyte;		/* byte number in ibm,pa/pi-features */
159 	unsigned char	pabit;		/* bit number (big-endian) */
160 	unsigned char	invert;		/* if 1, pa bit set => clear feature */
161 };
162 
163 static struct ibm_feature ibm_pa_features[] __initdata = {
164 	{ .pabyte = 0,  .pabit = 0, .cpu_user_ftrs = PPC_FEATURE_HAS_MMU },
165 	{ .pabyte = 0,  .pabit = 1, .cpu_user_ftrs = PPC_FEATURE_HAS_FPU },
166 	{ .pabyte = 0,  .pabit = 3, .cpu_features  = CPU_FTR_CTRL },
167 	{ .pabyte = 0,  .pabit = 6, .cpu_features  = CPU_FTR_NOEXECUTE },
168 	{ .pabyte = 1,  .pabit = 2, .mmu_features  = MMU_FTR_CI_LARGE_PAGE },
169 #ifdef CONFIG_PPC_RADIX_MMU
170 	{ .pabyte = 40, .pabit = 0, .mmu_features  = MMU_FTR_TYPE_RADIX | MMU_FTR_GTSE },
171 #endif
172 	{ .pabyte = 5,  .pabit = 0, .cpu_features  = CPU_FTR_REAL_LE,
173 				    .cpu_user_ftrs = PPC_FEATURE_TRUE_LE },
174 	/*
175 	 * If the kernel doesn't support TM (ie CONFIG_PPC_TRANSACTIONAL_MEM=n),
176 	 * we don't want to turn on TM here, so we use the *_COMP versions
177 	 * which are 0 if the kernel doesn't support TM.
178 	 */
179 	{ .pabyte = 22, .pabit = 0, .cpu_features = CPU_FTR_TM_COMP,
180 	  .cpu_user_ftrs2 = PPC_FEATURE2_HTM_COMP | PPC_FEATURE2_HTM_NOSC_COMP },
181 
182 	{ .pabyte = 64, .pabit = 0, .cpu_features = CPU_FTR_DAWR1 },
183 };
184 
185 /*
186  * ibm,pi-features property provides the support of processor specific
187  * options not described in ibm,pa-features. Right now use byte 0, bit 3
188  * which indicates the occurrence of DSI interrupt when the paste operation
189  * on the suspended NX window.
190  */
191 static struct ibm_feature ibm_pi_features[] __initdata = {
192 	{ .pabyte = 0, .pabit = 3, .mmu_features  = MMU_FTR_NX_DSI },
193 };
194 
scan_features(unsigned long node,const unsigned char * ftrs,unsigned long tablelen,struct ibm_feature * fp,unsigned long ft_size)195 static void __init scan_features(unsigned long node, const unsigned char *ftrs,
196 				 unsigned long tablelen,
197 				 struct ibm_feature *fp,
198 				 unsigned long ft_size)
199 {
200 	unsigned long i, len, bit;
201 
202 	/* find descriptor with type == 0 */
203 	for (;;) {
204 		if (tablelen < 3)
205 			return;
206 		len = 2 + ftrs[0];
207 		if (tablelen < len)
208 			return;		/* descriptor 0 not found */
209 		if (ftrs[1] == 0)
210 			break;
211 		tablelen -= len;
212 		ftrs += len;
213 	}
214 
215 	/* loop over bits we know about */
216 	for (i = 0; i < ft_size; ++i, ++fp) {
217 		if (fp->pabyte >= ftrs[0])
218 			continue;
219 		bit = (ftrs[2 + fp->pabyte] >> (7 - fp->pabit)) & 1;
220 		if (bit ^ fp->invert) {
221 			cur_cpu_spec->cpu_features |= fp->cpu_features;
222 			cur_cpu_spec->cpu_user_features |= fp->cpu_user_ftrs;
223 			cur_cpu_spec->cpu_user_features2 |= fp->cpu_user_ftrs2;
224 			cur_cpu_spec->mmu_features |= fp->mmu_features;
225 		} else {
226 			cur_cpu_spec->cpu_features &= ~fp->cpu_features;
227 			cur_cpu_spec->cpu_user_features &= ~fp->cpu_user_ftrs;
228 			cur_cpu_spec->cpu_user_features2 &= ~fp->cpu_user_ftrs2;
229 			cur_cpu_spec->mmu_features &= ~fp->mmu_features;
230 		}
231 	}
232 }
233 
check_cpu_features(unsigned long node,char * name,struct ibm_feature * fp,unsigned long size)234 static void __init check_cpu_features(unsigned long node, char *name,
235 				      struct ibm_feature *fp,
236 				      unsigned long size)
237 {
238 	const unsigned char *pa_ftrs;
239 	int tablelen;
240 
241 	pa_ftrs = of_get_flat_dt_prop(node, name, &tablelen);
242 	if (pa_ftrs == NULL)
243 		return;
244 
245 	scan_features(node, pa_ftrs, tablelen, fp, size);
246 }
247 
248 #ifdef CONFIG_PPC_64S_HASH_MMU
init_mmu_slb_size(unsigned long node)249 static void __init init_mmu_slb_size(unsigned long node)
250 {
251 	const __be32 *slb_size_ptr;
252 
253 	slb_size_ptr = of_get_flat_dt_prop(node, "slb-size", NULL) ? :
254 			of_get_flat_dt_prop(node, "ibm,slb-size", NULL);
255 
256 	if (slb_size_ptr)
257 		mmu_slb_size = be32_to_cpup(slb_size_ptr);
258 }
259 #else
260 #define init_mmu_slb_size(node) do { } while(0)
261 #endif
262 
263 static struct feature_property {
264 	const char *name;
265 	u32 min_value;
266 	unsigned long cpu_feature;
267 	unsigned long cpu_user_ftr;
268 } feature_properties[] __initdata = {
269 #ifdef CONFIG_ALTIVEC
270 	{"altivec", 0, CPU_FTR_ALTIVEC, PPC_FEATURE_HAS_ALTIVEC},
271 	{"ibm,vmx", 1, CPU_FTR_ALTIVEC, PPC_FEATURE_HAS_ALTIVEC},
272 #endif /* CONFIG_ALTIVEC */
273 #ifdef CONFIG_VSX
274 	/* Yes, this _really_ is ibm,vmx == 2 to enable VSX */
275 	{"ibm,vmx", 2, CPU_FTR_VSX, PPC_FEATURE_HAS_VSX},
276 #endif /* CONFIG_VSX */
277 #ifdef CONFIG_PPC64
278 	{"ibm,dfp", 1, 0, PPC_FEATURE_HAS_DFP},
279 	{"ibm,purr", 1, CPU_FTR_PURR, 0},
280 	{"ibm,spurr", 1, CPU_FTR_SPURR, 0},
281 #endif /* CONFIG_PPC64 */
282 };
283 
284 #if defined(CONFIG_44x) && defined(CONFIG_PPC_FPU)
identical_pvr_fixup(unsigned long node)285 static __init void identical_pvr_fixup(unsigned long node)
286 {
287 	unsigned int pvr;
288 	const char *model = of_get_flat_dt_prop(node, "model", NULL);
289 
290 	/*
291 	 * Since 440GR(x)/440EP(x) processors have the same pvr,
292 	 * we check the node path and set bit 28 in the cur_cpu_spec
293 	 * pvr for EP(x) processor version. This bit is always 0 in
294 	 * the "real" pvr. Then we call identify_cpu again with
295 	 * the new logical pvr to enable FPU support.
296 	 */
297 	if (model && strstr(model, "440EP")) {
298 		pvr = cur_cpu_spec->pvr_value | 0x8;
299 		identify_cpu(0, pvr);
300 		DBG("Using logical pvr %x for %s\n", pvr, model);
301 	}
302 }
303 #else
304 #define identical_pvr_fixup(node) do { } while(0)
305 #endif
306 
check_cpu_feature_properties(unsigned long node)307 static void __init check_cpu_feature_properties(unsigned long node)
308 {
309 	int i;
310 	struct feature_property *fp = feature_properties;
311 	const __be32 *prop;
312 
313 	for (i = 0; i < (int)ARRAY_SIZE(feature_properties); ++i, ++fp) {
314 		prop = of_get_flat_dt_prop(node, fp->name, NULL);
315 		if (prop && be32_to_cpup(prop) >= fp->min_value) {
316 			cur_cpu_spec->cpu_features |= fp->cpu_feature;
317 			cur_cpu_spec->cpu_user_features |= fp->cpu_user_ftr;
318 		}
319 	}
320 }
321 
early_init_dt_scan_cpus(unsigned long node,const char * uname,int depth,void * data)322 static int __init early_init_dt_scan_cpus(unsigned long node,
323 					  const char *uname, int depth,
324 					  void *data)
325 {
326 	const char *type = of_get_flat_dt_prop(node, "device_type", NULL);
327 	const __be32 *prop;
328 	const __be32 *intserv;
329 	int i, nthreads;
330 	int len;
331 	int found = -1;
332 	int found_thread = 0;
333 
334 	/* We are scanning "cpu" nodes only */
335 	if (type == NULL || strcmp(type, "cpu") != 0)
336 		return 0;
337 
338 	/* Get physical cpuid */
339 	intserv = of_get_flat_dt_prop(node, "ibm,ppc-interrupt-server#s", &len);
340 	if (!intserv)
341 		intserv = of_get_flat_dt_prop(node, "reg", &len);
342 
343 	nthreads = len / sizeof(int);
344 
345 	/*
346 	 * Now see if any of these threads match our boot cpu.
347 	 * NOTE: This must match the parsing done in smp_setup_cpu_maps.
348 	 */
349 	for (i = 0; i < nthreads; i++) {
350 		if (be32_to_cpu(intserv[i]) ==
351 			fdt_boot_cpuid_phys(initial_boot_params)) {
352 			found = boot_cpu_count;
353 			found_thread = i;
354 		}
355 #ifdef CONFIG_SMP
356 		/* logical cpu id is always 0 on UP kernels */
357 		boot_cpu_count++;
358 #endif
359 	}
360 
361 	/* Not the boot CPU */
362 	if (found < 0)
363 		return 0;
364 
365 	DBG("boot cpu: logical %d physical %d\n", found,
366 	    be32_to_cpu(intserv[found_thread]));
367 	boot_cpuid = found;
368 
369 	// Pass the boot CPU's hard CPU id back to our caller
370 	*((u32 *)data) = be32_to_cpu(intserv[found_thread]);
371 
372 	/*
373 	 * PAPR defines "logical" PVR values for cpus that
374 	 * meet various levels of the architecture:
375 	 * 0x0f000001	Architecture version 2.04
376 	 * 0x0f000002	Architecture version 2.05
377 	 * If the cpu-version property in the cpu node contains
378 	 * such a value, we call identify_cpu again with the
379 	 * logical PVR value in order to use the cpu feature
380 	 * bits appropriate for the architecture level.
381 	 *
382 	 * A POWER6 partition in "POWER6 architected" mode
383 	 * uses the 0x0f000002 PVR value; in POWER5+ mode
384 	 * it uses 0x0f000001.
385 	 *
386 	 * If we're using device tree CPU feature discovery then we don't
387 	 * support the cpu-version property, and it's the responsibility of the
388 	 * firmware/hypervisor to provide the correct feature set for the
389 	 * architecture level via the ibm,powerpc-cpu-features binding.
390 	 */
391 	if (!dt_cpu_ftrs_in_use()) {
392 		prop = of_get_flat_dt_prop(node, "cpu-version", NULL);
393 		if (prop && (be32_to_cpup(prop) & 0xff000000) == 0x0f000000) {
394 			identify_cpu(0, be32_to_cpup(prop));
395 			seq_buf_printf(&ppc_hw_desc, "0x%04x ", be32_to_cpup(prop));
396 		}
397 
398 		check_cpu_feature_properties(node);
399 		check_cpu_features(node, "ibm,pa-features", ibm_pa_features,
400 				   ARRAY_SIZE(ibm_pa_features));
401 		check_cpu_features(node, "ibm,pi-features", ibm_pi_features,
402 				   ARRAY_SIZE(ibm_pi_features));
403 	}
404 
405 	identical_pvr_fixup(node);
406 	init_mmu_slb_size(node);
407 
408 #ifdef CONFIG_PPC64
409 	if (nthreads == 1)
410 		cur_cpu_spec->cpu_features &= ~CPU_FTR_SMT;
411 	else if (!dt_cpu_ftrs_in_use())
412 		cur_cpu_spec->cpu_features |= CPU_FTR_SMT;
413 #endif
414 
415 	return 0;
416 }
417 
early_init_dt_scan_chosen_ppc(unsigned long node,const char * uname,int depth,void * data)418 static int __init early_init_dt_scan_chosen_ppc(unsigned long node,
419 						const char *uname,
420 						int depth, void *data)
421 {
422 	const unsigned long *lprop; /* All these set by kernel, so no need to convert endian */
423 
424 	/* Use common scan routine to determine if this is the chosen node */
425 	if (early_init_dt_scan_chosen(data) < 0)
426 		return 0;
427 
428 #ifdef CONFIG_PPC64
429 	/* check if iommu is forced on or off */
430 	if (of_get_flat_dt_prop(node, "linux,iommu-off", NULL) != NULL)
431 		iommu_is_off = 1;
432 	if (of_get_flat_dt_prop(node, "linux,iommu-force-on", NULL) != NULL)
433 		iommu_force_on = 1;
434 #endif
435 
436 	/* mem=x on the command line is the preferred mechanism */
437 	lprop = of_get_flat_dt_prop(node, "linux,memory-limit", NULL);
438 	if (lprop)
439 		memory_limit = *lprop;
440 
441 #ifdef CONFIG_PPC64
442 	lprop = of_get_flat_dt_prop(node, "linux,tce-alloc-start", NULL);
443 	if (lprop)
444 		tce_alloc_start = *lprop;
445 	lprop = of_get_flat_dt_prop(node, "linux,tce-alloc-end", NULL);
446 	if (lprop)
447 		tce_alloc_end = *lprop;
448 #endif
449 
450 #ifdef CONFIG_KEXEC_CORE
451 	lprop = of_get_flat_dt_prop(node, "linux,crashkernel-base", NULL);
452 	if (lprop)
453 		crashk_res.start = *lprop;
454 
455 	lprop = of_get_flat_dt_prop(node, "linux,crashkernel-size", NULL);
456 	if (lprop)
457 		crashk_res.end = crashk_res.start + *lprop - 1;
458 #endif
459 
460 	/* break now */
461 	return 1;
462 }
463 
464 /*
465  * Compare the range against max mem limit and update
466  * size if it cross the limit.
467  */
468 
469 #ifdef CONFIG_SPARSEMEM
validate_mem_limit(u64 base,u64 * size)470 static bool __init validate_mem_limit(u64 base, u64 *size)
471 {
472 	u64 max_mem = 1UL << (MAX_PHYSMEM_BITS);
473 
474 	if (base >= max_mem)
475 		return false;
476 	if ((base + *size) > max_mem)
477 		*size = max_mem - base;
478 	return true;
479 }
480 #else
validate_mem_limit(u64 base,u64 * size)481 static bool __init validate_mem_limit(u64 base, u64 *size)
482 {
483 	return true;
484 }
485 #endif
486 
487 #ifdef CONFIG_PPC_PSERIES
488 /*
489  * Interpret the ibm dynamic reconfiguration memory LMBs.
490  * This contains a list of memory blocks along with NUMA affinity
491  * information.
492  */
early_init_drmem_lmb(struct drmem_lmb * lmb,const __be32 ** usm,void * data)493 static int  __init early_init_drmem_lmb(struct drmem_lmb *lmb,
494 					const __be32 **usm,
495 					void *data)
496 {
497 	u64 base, size;
498 	int is_kexec_kdump = 0, rngs;
499 
500 	base = lmb->base_addr;
501 	size = drmem_lmb_size();
502 	rngs = 1;
503 
504 	/*
505 	 * Skip this block if the reserved bit is set in flags
506 	 * or if the block is not assigned to this partition.
507 	 */
508 	if ((lmb->flags & DRCONF_MEM_RESERVED) ||
509 	    !(lmb->flags & DRCONF_MEM_ASSIGNED))
510 		return 0;
511 
512 	if (*usm)
513 		is_kexec_kdump = 1;
514 
515 	if (is_kexec_kdump) {
516 		/*
517 		 * For each memblock in ibm,dynamic-memory, a
518 		 * corresponding entry in linux,drconf-usable-memory
519 		 * property contains a counter 'p' followed by 'p'
520 		 * (base, size) duple. Now read the counter from
521 		 * linux,drconf-usable-memory property
522 		 */
523 		rngs = dt_mem_next_cell(dt_root_size_cells, usm);
524 		if (!rngs) /* there are no (base, size) duple */
525 			return 0;
526 	}
527 
528 	do {
529 		if (is_kexec_kdump) {
530 			base = dt_mem_next_cell(dt_root_addr_cells, usm);
531 			size = dt_mem_next_cell(dt_root_size_cells, usm);
532 		}
533 
534 		if (iommu_is_off) {
535 			if (base >= 0x80000000ul)
536 				continue;
537 			if ((base + size) > 0x80000000ul)
538 				size = 0x80000000ul - base;
539 		}
540 
541 		if (!validate_mem_limit(base, &size))
542 			continue;
543 
544 		DBG("Adding: %llx -> %llx\n", base, size);
545 		memblock_add(base, size);
546 
547 		if (lmb->flags & DRCONF_MEM_HOTREMOVABLE)
548 			memblock_mark_hotplug(base, size);
549 	} while (--rngs);
550 
551 	return 0;
552 }
553 #endif /* CONFIG_PPC_PSERIES */
554 
early_init_dt_scan_memory_ppc(void)555 static int __init early_init_dt_scan_memory_ppc(void)
556 {
557 #ifdef CONFIG_PPC_PSERIES
558 	const void *fdt = initial_boot_params;
559 	int node = fdt_path_offset(fdt, "/ibm,dynamic-reconfiguration-memory");
560 
561 	if (node > 0)
562 		walk_drmem_lmbs_early(node, NULL, early_init_drmem_lmb);
563 
564 #endif
565 
566 	return early_init_dt_scan_memory();
567 }
568 
569 /*
570  * For a relocatable kernel, we need to get the memstart_addr first,
571  * then use it to calculate the virtual kernel start address. This has
572  * to happen at a very early stage (before machine_init). In this case,
573  * we just want to get the memstart_address and would not like to mess the
574  * memblock at this stage. So introduce a variable to skip the memblock_add()
575  * for this reason.
576  */
577 #ifdef CONFIG_RELOCATABLE
578 static int add_mem_to_memblock = 1;
579 #else
580 #define add_mem_to_memblock 1
581 #endif
582 
early_init_dt_add_memory_arch(u64 base,u64 size)583 void __init early_init_dt_add_memory_arch(u64 base, u64 size)
584 {
585 #ifdef CONFIG_PPC64
586 	if (iommu_is_off) {
587 		if (base >= 0x80000000ul)
588 			return;
589 		if ((base + size) > 0x80000000ul)
590 			size = 0x80000000ul - base;
591 	}
592 #endif
593 	/* Keep track of the beginning of memory -and- the size of
594 	 * the very first block in the device-tree as it represents
595 	 * the RMA on ppc64 server
596 	 */
597 	if (base < memstart_addr) {
598 		memstart_addr = base;
599 		first_memblock_size = size;
600 	}
601 
602 	/* Add the chunk to the MEMBLOCK list */
603 	if (add_mem_to_memblock) {
604 		if (validate_mem_limit(base, &size))
605 			memblock_add(base, size);
606 	}
607 }
608 
early_reserve_mem_dt(void)609 static void __init early_reserve_mem_dt(void)
610 {
611 	unsigned long i, dt_root;
612 	int len;
613 	const __be32 *prop;
614 
615 	early_init_fdt_reserve_self();
616 	early_init_fdt_scan_reserved_mem();
617 
618 	dt_root = of_get_flat_dt_root();
619 
620 	prop = of_get_flat_dt_prop(dt_root, "reserved-ranges", &len);
621 
622 	if (!prop)
623 		return;
624 
625 	DBG("Found new-style reserved-ranges\n");
626 
627 	/* Each reserved range is an (address,size) pair, 2 cells each,
628 	 * totalling 4 cells per range. */
629 	for (i = 0; i < len / (sizeof(*prop) * 4); i++) {
630 		u64 base, size;
631 
632 		base = of_read_number(prop + (i * 4) + 0, 2);
633 		size = of_read_number(prop + (i * 4) + 2, 2);
634 
635 		if (size) {
636 			DBG("reserving: %llx -> %llx\n", base, size);
637 			memblock_reserve(base, size);
638 		}
639 	}
640 }
641 
early_reserve_mem(void)642 static void __init early_reserve_mem(void)
643 {
644 	__be64 *reserve_map;
645 
646 	reserve_map = (__be64 *)(((unsigned long)initial_boot_params) +
647 			fdt_off_mem_rsvmap(initial_boot_params));
648 
649 	/* Look for the new "reserved-regions" property in the DT */
650 	early_reserve_mem_dt();
651 
652 #ifdef CONFIG_BLK_DEV_INITRD
653 	/* Then reserve the initrd, if any */
654 	if (initrd_start && (initrd_end > initrd_start)) {
655 		memblock_reserve(ALIGN_DOWN(__pa(initrd_start), PAGE_SIZE),
656 			ALIGN(initrd_end, PAGE_SIZE) -
657 			ALIGN_DOWN(initrd_start, PAGE_SIZE));
658 	}
659 #endif /* CONFIG_BLK_DEV_INITRD */
660 
661 	if (!IS_ENABLED(CONFIG_PPC32))
662 		return;
663 
664 	/*
665 	 * Handle the case where we might be booting from an old kexec
666 	 * image that setup the mem_rsvmap as pairs of 32-bit values
667 	 */
668 	if (be64_to_cpup(reserve_map) > 0xffffffffull) {
669 		u32 base_32, size_32;
670 		__be32 *reserve_map_32 = (__be32 *)reserve_map;
671 
672 		DBG("Found old 32-bit reserve map\n");
673 
674 		while (1) {
675 			base_32 = be32_to_cpup(reserve_map_32++);
676 			size_32 = be32_to_cpup(reserve_map_32++);
677 			if (size_32 == 0)
678 				break;
679 			DBG("reserving: %x -> %x\n", base_32, size_32);
680 			memblock_reserve(base_32, size_32);
681 		}
682 		return;
683 	}
684 }
685 
686 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
687 static bool tm_disabled __initdata;
688 
parse_ppc_tm(char * str)689 static int __init parse_ppc_tm(char *str)
690 {
691 	bool res;
692 
693 	if (kstrtobool(str, &res))
694 		return -EINVAL;
695 
696 	tm_disabled = !res;
697 
698 	return 0;
699 }
700 early_param("ppc_tm", parse_ppc_tm);
701 
tm_init(void)702 static void __init tm_init(void)
703 {
704 	if (tm_disabled) {
705 		pr_info("Disabling hardware transactional memory (HTM)\n");
706 		cur_cpu_spec->cpu_user_features2 &=
707 			~(PPC_FEATURE2_HTM_NOSC | PPC_FEATURE2_HTM);
708 		cur_cpu_spec->cpu_features &= ~CPU_FTR_TM;
709 		return;
710 	}
711 
712 	pnv_tm_init();
713 }
714 #else
tm_init(void)715 static void tm_init(void) { }
716 #endif /* CONFIG_PPC_TRANSACTIONAL_MEM */
717 
718 static int __init
early_init_dt_scan_model(unsigned long node,const char * uname,int depth,void * data)719 early_init_dt_scan_model(unsigned long node, const char *uname,
720 			 int depth, void *data)
721 {
722 	const char *prop;
723 
724 	if (depth != 0)
725 		return 0;
726 
727 	prop = of_get_flat_dt_prop(node, "model", NULL);
728 	if (prop)
729 		seq_buf_printf(&ppc_hw_desc, "%s ", prop);
730 
731 	/* break now */
732 	return 1;
733 }
734 
735 #ifdef CONFIG_PPC64
save_fscr_to_task(void)736 static void __init save_fscr_to_task(void)
737 {
738 	/*
739 	 * Ensure the init_task (pid 0, aka swapper) uses the value of FSCR we
740 	 * have configured via the device tree features or via __init_FSCR().
741 	 * That value will then be propagated to pid 1 (init) and all future
742 	 * processes.
743 	 */
744 	if (early_cpu_has_feature(CPU_FTR_ARCH_207S))
745 		init_task.thread.fscr = mfspr(SPRN_FSCR);
746 }
747 #else
save_fscr_to_task(void)748 static inline void save_fscr_to_task(void) {}
749 #endif
750 
751 
early_init_devtree(void * params)752 void __init early_init_devtree(void *params)
753 {
754 	u32 boot_cpu_hwid;
755 	phys_addr_t limit;
756 
757 	DBG(" -> early_init_devtree(%px)\n", params);
758 
759 	/* Too early to BUG_ON(), do it by hand */
760 	if (!early_init_dt_verify(params))
761 		panic("BUG: Failed verifying flat device tree, bad version?");
762 
763 	of_scan_flat_dt(early_init_dt_scan_model, NULL);
764 
765 #ifdef CONFIG_PPC_RTAS
766 	/* Some machines might need RTAS info for debugging, grab it now. */
767 	of_scan_flat_dt(early_init_dt_scan_rtas, NULL);
768 #endif
769 
770 #ifdef CONFIG_PPC_POWERNV
771 	/* Some machines might need OPAL info for debugging, grab it now. */
772 	of_scan_flat_dt(early_init_dt_scan_opal, NULL);
773 
774 	/* Scan tree for ultravisor feature */
775 	of_scan_flat_dt(early_init_dt_scan_ultravisor, NULL);
776 #endif
777 
778 #if defined(CONFIG_FA_DUMP) || defined(CONFIG_PRESERVE_FA_DUMP)
779 	/* scan tree to see if dump is active during last boot */
780 	of_scan_flat_dt(early_init_dt_scan_fw_dump, NULL);
781 #endif
782 
783 	/* Retrieve various informations from the /chosen node of the
784 	 * device-tree, including the platform type, initrd location and
785 	 * size, TCE reserve, and more ...
786 	 */
787 	of_scan_flat_dt(early_init_dt_scan_chosen_ppc, boot_command_line);
788 
789 	/* Scan memory nodes and rebuild MEMBLOCKs */
790 	early_init_dt_scan_root();
791 	early_init_dt_scan_memory_ppc();
792 
793 	/*
794 	 * As generic code authors expect to be able to use static keys
795 	 * in early_param() handlers, we initialize the static keys just
796 	 * before parsing early params (it's fine to call jump_label_init()
797 	 * more than once).
798 	 */
799 	jump_label_init();
800 	parse_early_param();
801 
802 	/* make sure we've parsed cmdline for mem= before this */
803 	if (memory_limit)
804 		first_memblock_size = min_t(u64, first_memblock_size, memory_limit);
805 	setup_initial_memory_limit(memstart_addr, first_memblock_size);
806 	/* Reserve MEMBLOCK regions used by kernel, initrd, dt, etc... */
807 	memblock_reserve(PHYSICAL_START, __pa(_end) - PHYSICAL_START);
808 	/* If relocatable, reserve first 32k for interrupt vectors etc. */
809 	if (PHYSICAL_START > MEMORY_START)
810 		memblock_reserve(MEMORY_START, 0x8000);
811 	reserve_kdump_trampoline();
812 #if defined(CONFIG_FA_DUMP) || defined(CONFIG_PRESERVE_FA_DUMP)
813 	/*
814 	 * If we fail to reserve memory for firmware-assisted dump then
815 	 * fallback to kexec based kdump.
816 	 */
817 	if (fadump_reserve_mem() == 0)
818 #endif
819 		reserve_crashkernel();
820 	early_reserve_mem();
821 
822 	/* Ensure that total memory size is page-aligned. */
823 	limit = ALIGN(memory_limit ?: memblock_phys_mem_size(), PAGE_SIZE);
824 	memblock_enforce_memory_limit(limit);
825 
826 #if defined(CONFIG_PPC_BOOK3S_64) && defined(CONFIG_PPC_4K_PAGES)
827 	if (!early_radix_enabled())
828 		memblock_cap_memory_range(0, 1UL << (H_MAX_PHYSMEM_BITS));
829 #endif
830 
831 	memblock_allow_resize();
832 	memblock_dump_all();
833 
834 	DBG("Phys. mem: %llx\n", (unsigned long long)memblock_phys_mem_size());
835 
836 	/* We may need to relocate the flat tree, do it now.
837 	 * FIXME .. and the initrd too? */
838 	move_device_tree();
839 
840 	DBG("Scanning CPUs ...\n");
841 
842 	dt_cpu_ftrs_scan();
843 
844 	// We can now add the CPU name & PVR to the hardware description
845 	seq_buf_printf(&ppc_hw_desc, "%s 0x%04lx ", cur_cpu_spec->cpu_name, mfspr(SPRN_PVR));
846 
847 	/* Retrieve CPU related informations from the flat tree
848 	 * (altivec support, boot CPU ID, ...)
849 	 */
850 	of_scan_flat_dt(early_init_dt_scan_cpus, &boot_cpu_hwid);
851 	if (boot_cpuid < 0) {
852 		printk("Failed to identify boot CPU !\n");
853 		BUG();
854 	}
855 
856 	save_fscr_to_task();
857 
858 #if defined(CONFIG_SMP) && defined(CONFIG_PPC64)
859 	/* We'll later wait for secondaries to check in; there are
860 	 * NCPUS-1 non-boot CPUs  :-)
861 	 */
862 	spinning_secondaries = boot_cpu_count - 1;
863 #endif
864 
865 	mmu_early_init_devtree();
866 
867 	// NB. paca is not installed until later in early_setup()
868 	allocate_paca_ptrs();
869 	allocate_paca(boot_cpuid);
870 	set_hard_smp_processor_id(boot_cpuid, boot_cpu_hwid);
871 
872 #ifdef CONFIG_PPC_POWERNV
873 	/* Scan and build the list of machine check recoverable ranges */
874 	of_scan_flat_dt(early_init_dt_scan_recoverable_ranges, NULL);
875 #endif
876 	epapr_paravirt_early_init();
877 
878 	/* Now try to figure out if we are running on LPAR and so on */
879 	pseries_probe_fw_features();
880 
881 	/*
882 	 * Initialize pkey features and default AMR/IAMR values
883 	 */
884 	pkey_early_init_devtree();
885 
886 #ifdef CONFIG_PPC_PS3
887 	/* Identify PS3 firmware */
888 	if (of_flat_dt_is_compatible(of_get_flat_dt_root(), "sony,ps3"))
889 		powerpc_firmware_features |= FW_FEATURE_PS3_POSSIBLE;
890 #endif
891 
892 	tm_init();
893 
894 	DBG(" <- early_init_devtree()\n");
895 }
896 
897 #ifdef CONFIG_RELOCATABLE
898 /*
899  * This function run before early_init_devtree, so we have to init
900  * initial_boot_params.
901  */
early_get_first_memblock_info(void * params,phys_addr_t * size)902 void __init early_get_first_memblock_info(void *params, phys_addr_t *size)
903 {
904 	/* Setup flat device-tree pointer */
905 	initial_boot_params = params;
906 
907 	/*
908 	 * Scan the memory nodes and set add_mem_to_memblock to 0 to avoid
909 	 * mess the memblock.
910 	 */
911 	add_mem_to_memblock = 0;
912 	early_init_dt_scan_root();
913 	early_init_dt_scan_memory_ppc();
914 	add_mem_to_memblock = 1;
915 
916 	if (size)
917 		*size = first_memblock_size;
918 }
919 #endif
920 
921 /*******
922  *
923  * New implementation of the OF "find" APIs, return a refcounted
924  * object, call of_node_put() when done.  The device tree and list
925  * are protected by a rw_lock.
926  *
927  * Note that property management will need some locking as well,
928  * this isn't dealt with yet.
929  *
930  *******/
931 
932 /**
933  * of_get_ibm_chip_id - Returns the IBM "chip-id" of a device
934  * @np: device node of the device
935  *
936  * This looks for a property "ibm,chip-id" in the node or any
937  * of its parents and returns its content, or -1 if it cannot
938  * be found.
939  */
of_get_ibm_chip_id(struct device_node * np)940 int of_get_ibm_chip_id(struct device_node *np)
941 {
942 	of_node_get(np);
943 	while (np) {
944 		u32 chip_id;
945 
946 		/*
947 		 * Skiboot may produce memory nodes that contain more than one
948 		 * cell in chip-id, we only read the first one here.
949 		 */
950 		if (!of_property_read_u32(np, "ibm,chip-id", &chip_id)) {
951 			of_node_put(np);
952 			return chip_id;
953 		}
954 
955 		np = of_get_next_parent(np);
956 	}
957 	return -1;
958 }
959 EXPORT_SYMBOL(of_get_ibm_chip_id);
960 
961 /**
962  * cpu_to_chip_id - Return the cpus chip-id
963  * @cpu: The logical cpu number.
964  *
965  * Return the value of the ibm,chip-id property corresponding to the given
966  * logical cpu number. If the chip-id can not be found, returns -1.
967  */
cpu_to_chip_id(int cpu)968 int cpu_to_chip_id(int cpu)
969 {
970 	struct device_node *np;
971 	int ret = -1, idx;
972 
973 	idx = cpu / threads_per_core;
974 	if (chip_id_lookup_table && chip_id_lookup_table[idx] != -1)
975 		return chip_id_lookup_table[idx];
976 
977 	np = of_get_cpu_node(cpu, NULL);
978 	if (np) {
979 		ret = of_get_ibm_chip_id(np);
980 		of_node_put(np);
981 
982 		if (chip_id_lookup_table)
983 			chip_id_lookup_table[idx] = ret;
984 	}
985 
986 	return ret;
987 }
988 EXPORT_SYMBOL(cpu_to_chip_id);
989 
arch_match_cpu_phys_id(int cpu,u64 phys_id)990 bool arch_match_cpu_phys_id(int cpu, u64 phys_id)
991 {
992 #ifdef CONFIG_SMP
993 	/*
994 	 * Early firmware scanning must use this rather than
995 	 * get_hard_smp_processor_id because we don't have pacas allocated
996 	 * until memory topology is discovered.
997 	 */
998 	if (cpu_to_phys_id != NULL)
999 		return (int)phys_id == cpu_to_phys_id[cpu];
1000 #endif
1001 
1002 	return (int)phys_id == get_hard_smp_processor_id(cpu);
1003 }
1004