Linux Audio

Check our new training course

Loading...
v4.17
 
  1/*
  2 * Procedures for creating, accessing and interpreting the device tree.
  3 *
  4 * Paul Mackerras	August 1996.
  5 * Copyright (C) 1996-2005 Paul Mackerras.
  6 * 
  7 *  Adapted for 64bit PowerPC by Dave Engebretsen and Peter Bergner.
  8 *    {engebret|bergner}@us.ibm.com 
  9 *
 10 *      This program is free software; you can redistribute it and/or
 11 *      modify it under the terms of the GNU General Public License
 12 *      as published by the Free Software Foundation; either version
 13 *      2 of the License, or (at your option) any later version.
 14 */
 15
 16#undef DEBUG
 17
 18#include <stdarg.h>
 19#include <linux/kernel.h>
 20#include <linux/string.h>
 21#include <linux/init.h>
 22#include <linux/threads.h>
 23#include <linux/spinlock.h>
 24#include <linux/types.h>
 25#include <linux/pci.h>
 26#include <linux/stringify.h>
 27#include <linux/delay.h>
 28#include <linux/initrd.h>
 29#include <linux/bitops.h>
 30#include <linux/export.h>
 31#include <linux/kexec.h>
 32#include <linux/irq.h>
 33#include <linux/memblock.h>
 34#include <linux/of.h>
 35#include <linux/of_fdt.h>
 36#include <linux/libfdt.h>
 37#include <linux/cpu.h>
 38
 39#include <asm/prom.h>
 40#include <asm/rtas.h>
 41#include <asm/page.h>
 42#include <asm/processor.h>
 43#include <asm/irq.h>
 44#include <asm/io.h>
 45#include <asm/kdump.h>
 46#include <asm/smp.h>
 47#include <asm/mmu.h>
 48#include <asm/paca.h>
 49#include <asm/pgtable.h>
 50#include <asm/powernv.h>
 51#include <asm/iommu.h>
 52#include <asm/btext.h>
 53#include <asm/sections.h>
 54#include <asm/machdep.h>
 55#include <asm/pci-bridge.h>
 56#include <asm/kexec.h>
 57#include <asm/opal.h>
 58#include <asm/fadump.h>
 59#include <asm/epapr_hcalls.h>
 60#include <asm/firmware.h>
 61#include <asm/dt_cpu_ftrs.h>
 62#include <asm/drmem.h>
 
 63
 64#include <mm/mmu_decl.h>
 65
 66#ifdef DEBUG
 67#define DBG(fmt...) printk(KERN_ERR fmt)
 68#else
 69#define DBG(fmt...)
 70#endif
 71
 72#ifdef CONFIG_PPC64
 73int __initdata iommu_is_off;
 74int __initdata iommu_force_on;
 75unsigned long tce_alloc_start, tce_alloc_end;
 76u64 ppc64_rma_size;
 77#endif
 78static phys_addr_t first_memblock_size;
 79static int __initdata boot_cpu_count;
 80
 81static 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}
 91early_param("mem", early_parse_mem);
 92
 93/*
 94 * overlaps_initrd - check for overlap with page aligned extension of
 95 * initrd.
 96 */
 97static 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_UP(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 */
117static 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			overlaps_crashkernel(start, size) ||
129			overlaps_initrd(start, size)) {
130		p = __va(memblock_alloc(size, PAGE_SIZE));
 
 
 
131		memcpy(p, initial_boot_params, size);
132		initial_boot_params = p;
133		DBG("Moved device tree to 0x%p\n", p);
134	}
135
136	DBG("<- move_device_tree\n");
137}
138
139/*
140 * ibm,pa-features is a per-cpu property that contains a string of
141 * attribute descriptors, each of which has a 2 byte header plus up
142 * to 254 bytes worth of processor attribute bits.  First header
143 * byte specifies the number of bytes following the header.
144 * Second header byte is an "attribute-specifier" type, of which
145 * zero is the only currently-defined value.
146 * Implementation:  Pass in the byte and bit offset for the feature
147 * that we are interested in.  The function will return -1 if the
148 * pa-features property is missing, or a 1/0 to indicate if the feature
149 * is supported/not supported.  Note that the bit numbers are
150 * big-endian to match the definition in PAPR.
151 */
152static struct ibm_pa_feature {
153	unsigned long	cpu_features;	/* CPU_FTR_xxx bit */
154	unsigned long	mmu_features;	/* MMU_FTR_xxx bit */
155	unsigned int	cpu_user_ftrs;	/* PPC_FEATURE_xxx bit */
156	unsigned int	cpu_user_ftrs2;	/* PPC_FEATURE2_xxx bit */
157	unsigned char	pabyte;		/* byte number in ibm,pa-features */
158	unsigned char	pabit;		/* bit number (big-endian) */
159	unsigned char	invert;		/* if 1, pa bit set => clear feature */
160} ibm_pa_features[] __initdata = {
161	{ .pabyte = 0,  .pabit = 0, .cpu_user_ftrs = PPC_FEATURE_HAS_MMU },
162	{ .pabyte = 0,  .pabit = 1, .cpu_user_ftrs = PPC_FEATURE_HAS_FPU },
163	{ .pabyte = 0,  .pabit = 3, .cpu_features  = CPU_FTR_CTRL },
164	{ .pabyte = 0,  .pabit = 6, .cpu_features  = CPU_FTR_NOEXECUTE },
165	{ .pabyte = 1,  .pabit = 2, .mmu_features  = MMU_FTR_CI_LARGE_PAGE },
166#ifdef CONFIG_PPC_RADIX_MMU
167	{ .pabyte = 40, .pabit = 0, .mmu_features  = MMU_FTR_TYPE_RADIX },
168#endif
169	{ .pabyte = 1,  .pabit = 1, .invert = 1, .cpu_features = CPU_FTR_NODSISRALIGN },
170	{ .pabyte = 5,  .pabit = 0, .cpu_features  = CPU_FTR_REAL_LE,
171				    .cpu_user_ftrs = PPC_FEATURE_TRUE_LE },
172	/*
173	 * If the kernel doesn't support TM (ie CONFIG_PPC_TRANSACTIONAL_MEM=n),
174	 * we don't want to turn on TM here, so we use the *_COMP versions
175	 * which are 0 if the kernel doesn't support TM.
176	 */
177	{ .pabyte = 22, .pabit = 0, .cpu_features = CPU_FTR_TM_COMP,
178	  .cpu_user_ftrs2 = PPC_FEATURE2_HTM_COMP | PPC_FEATURE2_HTM_NOSC_COMP },
179};
180
181static void __init scan_features(unsigned long node, const unsigned char *ftrs,
182				 unsigned long tablelen,
183				 struct ibm_pa_feature *fp,
184				 unsigned long ft_size)
185{
186	unsigned long i, len, bit;
187
188	/* find descriptor with type == 0 */
189	for (;;) {
190		if (tablelen < 3)
191			return;
192		len = 2 + ftrs[0];
193		if (tablelen < len)
194			return;		/* descriptor 0 not found */
195		if (ftrs[1] == 0)
196			break;
197		tablelen -= len;
198		ftrs += len;
199	}
200
201	/* loop over bits we know about */
202	for (i = 0; i < ft_size; ++i, ++fp) {
203		if (fp->pabyte >= ftrs[0])
204			continue;
205		bit = (ftrs[2 + fp->pabyte] >> (7 - fp->pabit)) & 1;
206		if (bit ^ fp->invert) {
207			cur_cpu_spec->cpu_features |= fp->cpu_features;
208			cur_cpu_spec->cpu_user_features |= fp->cpu_user_ftrs;
209			cur_cpu_spec->cpu_user_features2 |= fp->cpu_user_ftrs2;
210			cur_cpu_spec->mmu_features |= fp->mmu_features;
211		} else {
212			cur_cpu_spec->cpu_features &= ~fp->cpu_features;
213			cur_cpu_spec->cpu_user_features &= ~fp->cpu_user_ftrs;
214			cur_cpu_spec->cpu_user_features2 &= ~fp->cpu_user_ftrs2;
215			cur_cpu_spec->mmu_features &= ~fp->mmu_features;
216		}
217	}
218}
219
220static void __init check_cpu_pa_features(unsigned long node)
221{
222	const unsigned char *pa_ftrs;
223	int tablelen;
224
225	pa_ftrs = of_get_flat_dt_prop(node, "ibm,pa-features", &tablelen);
226	if (pa_ftrs == NULL)
227		return;
228
229	scan_features(node, pa_ftrs, tablelen,
230		      ibm_pa_features, ARRAY_SIZE(ibm_pa_features));
231}
232
233#ifdef CONFIG_PPC_BOOK3S_64
234static void __init init_mmu_slb_size(unsigned long node)
235{
236	const __be32 *slb_size_ptr;
237
238	slb_size_ptr = of_get_flat_dt_prop(node, "slb-size", NULL) ? :
239			of_get_flat_dt_prop(node, "ibm,slb-size", NULL);
240
241	if (slb_size_ptr)
242		mmu_slb_size = be32_to_cpup(slb_size_ptr);
243}
244#else
245#define init_mmu_slb_size(node) do { } while(0)
246#endif
247
248static struct feature_property {
249	const char *name;
250	u32 min_value;
251	unsigned long cpu_feature;
252	unsigned long cpu_user_ftr;
253} feature_properties[] __initdata = {
254#ifdef CONFIG_ALTIVEC
255	{"altivec", 0, CPU_FTR_ALTIVEC, PPC_FEATURE_HAS_ALTIVEC},
256	{"ibm,vmx", 1, CPU_FTR_ALTIVEC, PPC_FEATURE_HAS_ALTIVEC},
257#endif /* CONFIG_ALTIVEC */
258#ifdef CONFIG_VSX
259	/* Yes, this _really_ is ibm,vmx == 2 to enable VSX */
260	{"ibm,vmx", 2, CPU_FTR_VSX, PPC_FEATURE_HAS_VSX},
261#endif /* CONFIG_VSX */
262#ifdef CONFIG_PPC64
263	{"ibm,dfp", 1, 0, PPC_FEATURE_HAS_DFP},
264	{"ibm,purr", 1, CPU_FTR_PURR, 0},
265	{"ibm,spurr", 1, CPU_FTR_SPURR, 0},
266#endif /* CONFIG_PPC64 */
267};
268
269#if defined(CONFIG_44x) && defined(CONFIG_PPC_FPU)
270static inline void identical_pvr_fixup(unsigned long node)
271{
272	unsigned int pvr;
273	const char *model = of_get_flat_dt_prop(node, "model", NULL);
274
275	/*
276	 * Since 440GR(x)/440EP(x) processors have the same pvr,
277	 * we check the node path and set bit 28 in the cur_cpu_spec
278	 * pvr for EP(x) processor version. This bit is always 0 in
279	 * the "real" pvr. Then we call identify_cpu again with
280	 * the new logical pvr to enable FPU support.
281	 */
282	if (model && strstr(model, "440EP")) {
283		pvr = cur_cpu_spec->pvr_value | 0x8;
284		identify_cpu(0, pvr);
285		DBG("Using logical pvr %x for %s\n", pvr, model);
286	}
287}
288#else
289#define identical_pvr_fixup(node) do { } while(0)
290#endif
291
292static void __init check_cpu_feature_properties(unsigned long node)
293{
294	int i;
295	struct feature_property *fp = feature_properties;
296	const __be32 *prop;
297
298	for (i = 0; i < (int)ARRAY_SIZE(feature_properties); ++i, ++fp) {
299		prop = of_get_flat_dt_prop(node, fp->name, NULL);
300		if (prop && be32_to_cpup(prop) >= fp->min_value) {
301			cur_cpu_spec->cpu_features |= fp->cpu_feature;
302			cur_cpu_spec->cpu_user_features |= fp->cpu_user_ftr;
303		}
304	}
305}
306
307static int __init early_init_dt_scan_cpus(unsigned long node,
308					  const char *uname, int depth,
309					  void *data)
310{
311	const char *type = of_get_flat_dt_prop(node, "device_type", NULL);
312	const __be32 *prop;
313	const __be32 *intserv;
314	int i, nthreads;
315	int len;
316	int found = -1;
317	int found_thread = 0;
318
319	/* We are scanning "cpu" nodes only */
320	if (type == NULL || strcmp(type, "cpu") != 0)
321		return 0;
322
323	/* Get physical cpuid */
324	intserv = of_get_flat_dt_prop(node, "ibm,ppc-interrupt-server#s", &len);
325	if (!intserv)
326		intserv = of_get_flat_dt_prop(node, "reg", &len);
327
328	nthreads = len / sizeof(int);
329
330	/*
331	 * Now see if any of these threads match our boot cpu.
332	 * NOTE: This must match the parsing done in smp_setup_cpu_maps.
333	 */
334	for (i = 0; i < nthreads; i++) {
335		/*
336		 * version 2 of the kexec param format adds the phys cpuid of
337		 * booted proc.
338		 */
339		if (fdt_version(initial_boot_params) >= 2) {
340			if (be32_to_cpu(intserv[i]) ==
341			    fdt_boot_cpuid_phys(initial_boot_params)) {
342				found = boot_cpu_count;
343				found_thread = i;
344			}
345		} else {
346			/*
347			 * Check if it's the boot-cpu, set it's hw index now,
348			 * unfortunately this format did not support booting
349			 * off secondary threads.
350			 */
351			if (of_get_flat_dt_prop(node,
352					"linux,boot-cpu", NULL) != NULL)
353				found = boot_cpu_count;
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	/*
370	 * PAPR defines "logical" PVR values for cpus that
371	 * meet various levels of the architecture:
372	 * 0x0f000001	Architecture version 2.04
373	 * 0x0f000002	Architecture version 2.05
374	 * If the cpu-version property in the cpu node contains
375	 * such a value, we call identify_cpu again with the
376	 * logical PVR value in order to use the cpu feature
377	 * bits appropriate for the architecture level.
378	 *
379	 * A POWER6 partition in "POWER6 architected" mode
380	 * uses the 0x0f000002 PVR value; in POWER5+ mode
381	 * it uses 0x0f000001.
382	 *
383	 * If we're using device tree CPU feature discovery then we don't
384	 * support the cpu-version property, and it's the responsibility of the
385	 * firmware/hypervisor to provide the correct feature set for the
386	 * architecture level via the ibm,powerpc-cpu-features binding.
387	 */
388	if (!dt_cpu_ftrs_in_use()) {
389		prop = of_get_flat_dt_prop(node, "cpu-version", NULL);
390		if (prop && (be32_to_cpup(prop) & 0xff000000) == 0x0f000000)
391			identify_cpu(0, be32_to_cpup(prop));
392
393		check_cpu_feature_properties(node);
394		check_cpu_pa_features(node);
395	}
396
397	identical_pvr_fixup(node);
398	init_mmu_slb_size(node);
399
400#ifdef CONFIG_PPC64
401	if (nthreads == 1)
402		cur_cpu_spec->cpu_features &= ~CPU_FTR_SMT;
403	else if (!dt_cpu_ftrs_in_use())
404		cur_cpu_spec->cpu_features |= CPU_FTR_SMT;
405	allocate_paca(boot_cpuid);
406#endif
407	set_hard_smp_processor_id(found, be32_to_cpu(intserv[found_thread]));
408
409	return 0;
410}
411
412static int __init early_init_dt_scan_chosen_ppc(unsigned long node,
413						const char *uname,
414						int depth, void *data)
415{
416	const unsigned long *lprop; /* All these set by kernel, so no need to convert endian */
417
418	/* Use common scan routine to determine if this is the chosen node */
419	if (early_init_dt_scan_chosen(node, uname, depth, data) == 0)
420		return 0;
421
422#ifdef CONFIG_PPC64
423	/* check if iommu is forced on or off */
424	if (of_get_flat_dt_prop(node, "linux,iommu-off", NULL) != NULL)
425		iommu_is_off = 1;
426	if (of_get_flat_dt_prop(node, "linux,iommu-force-on", NULL) != NULL)
427		iommu_force_on = 1;
428#endif
429
430	/* mem=x on the command line is the preferred mechanism */
431	lprop = of_get_flat_dt_prop(node, "linux,memory-limit", NULL);
432	if (lprop)
433		memory_limit = *lprop;
434
435#ifdef CONFIG_PPC64
436	lprop = of_get_flat_dt_prop(node, "linux,tce-alloc-start", NULL);
437	if (lprop)
438		tce_alloc_start = *lprop;
439	lprop = of_get_flat_dt_prop(node, "linux,tce-alloc-end", NULL);
440	if (lprop)
441		tce_alloc_end = *lprop;
442#endif
443
444#ifdef CONFIG_KEXEC_CORE
445	lprop = of_get_flat_dt_prop(node, "linux,crashkernel-base", NULL);
446	if (lprop)
447		crashk_res.start = *lprop;
448
449	lprop = of_get_flat_dt_prop(node, "linux,crashkernel-size", NULL);
450	if (lprop)
451		crashk_res.end = crashk_res.start + *lprop - 1;
452#endif
453
454	/* break now */
455	return 1;
456}
457
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
458#ifdef CONFIG_PPC_PSERIES
459/*
460 * Interpret the ibm dynamic reconfiguration memory LMBs.
461 * This contains a list of memory blocks along with NUMA affinity
462 * information.
463 */
464static void __init early_init_drmem_lmb(struct drmem_lmb *lmb,
465					const __be32 **usm)
466{
467	u64 base, size;
468	int is_kexec_kdump = 0, rngs;
469
470	base = lmb->base_addr;
471	size = drmem_lmb_size();
472	rngs = 1;
473
474	/*
475	 * Skip this block if the reserved bit is set in flags
476	 * or if the block is not assigned to this partition.
477	 */
478	if ((lmb->flags & DRCONF_MEM_RESERVED) ||
479	    !(lmb->flags & DRCONF_MEM_ASSIGNED))
480		return;
481
482	if (*usm)
483		is_kexec_kdump = 1;
484
485	if (is_kexec_kdump) {
486		/*
487		 * For each memblock in ibm,dynamic-memory, a
488		 * corresponding entry in linux,drconf-usable-memory
489		 * property contains a counter 'p' followed by 'p'
490		 * (base, size) duple. Now read the counter from
491		 * linux,drconf-usable-memory property
492		 */
493		rngs = dt_mem_next_cell(dt_root_size_cells, usm);
494		if (!rngs) /* there are no (base, size) duple */
495			return;
496	}
497
498	do {
499		if (is_kexec_kdump) {
500			base = dt_mem_next_cell(dt_root_addr_cells, usm);
501			size = dt_mem_next_cell(dt_root_size_cells, usm);
502		}
503
504		if (iommu_is_off) {
505			if (base >= 0x80000000ul)
506				continue;
507			if ((base + size) > 0x80000000ul)
508				size = 0x80000000ul - base;
509		}
510
511		DBG("Adding: %llx -> %llx\n", base, size);
512		memblock_add(base, size);
 
513	} while (--rngs);
514}
515#endif /* CONFIG_PPC_PSERIES */
516
517static int __init early_init_dt_scan_memory_ppc(unsigned long node,
518						const char *uname,
519						int depth, void *data)
520{
521#ifdef CONFIG_PPC_PSERIES
522	if (depth == 1 &&
523	    strcmp(uname, "ibm,dynamic-reconfiguration-memory") == 0) {
524		walk_drmem_lmbs_early(node, early_init_drmem_lmb);
525		return 0;
526	}
527#endif
528	
529	return early_init_dt_scan_memory(node, uname, depth, data);
530}
531
532/*
533 * For a relocatable kernel, we need to get the memstart_addr first,
534 * then use it to calculate the virtual kernel start address. This has
535 * to happen at a very early stage (before machine_init). In this case,
536 * we just want to get the memstart_address and would not like to mess the
537 * memblock at this stage. So introduce a variable to skip the memblock_add()
538 * for this reason.
539 */
540#ifdef CONFIG_RELOCATABLE
541static int add_mem_to_memblock = 1;
542#else
543#define add_mem_to_memblock 1
544#endif
545
546void __init early_init_dt_add_memory_arch(u64 base, u64 size)
547{
548#ifdef CONFIG_PPC64
549	if (iommu_is_off) {
550		if (base >= 0x80000000ul)
551			return;
552		if ((base + size) > 0x80000000ul)
553			size = 0x80000000ul - base;
554	}
555#endif
556	/* Keep track of the beginning of memory -and- the size of
557	 * the very first block in the device-tree as it represents
558	 * the RMA on ppc64 server
559	 */
560	if (base < memstart_addr) {
561		memstart_addr = base;
562		first_memblock_size = size;
563	}
564
565	/* Add the chunk to the MEMBLOCK list */
566	if (add_mem_to_memblock)
567		memblock_add(base, size);
 
 
568}
569
570static void __init early_reserve_mem_dt(void)
571{
572	unsigned long i, dt_root;
573	int len;
574	const __be32 *prop;
575
576	early_init_fdt_reserve_self();
577	early_init_fdt_scan_reserved_mem();
578
579	dt_root = of_get_flat_dt_root();
580
581	prop = of_get_flat_dt_prop(dt_root, "reserved-ranges", &len);
582
583	if (!prop)
584		return;
585
586	DBG("Found new-style reserved-ranges\n");
587
588	/* Each reserved range is an (address,size) pair, 2 cells each,
589	 * totalling 4 cells per range. */
590	for (i = 0; i < len / (sizeof(*prop) * 4); i++) {
591		u64 base, size;
592
593		base = of_read_number(prop + (i * 4) + 0, 2);
594		size = of_read_number(prop + (i * 4) + 2, 2);
595
596		if (size) {
597			DBG("reserving: %llx -> %llx\n", base, size);
598			memblock_reserve(base, size);
599		}
600	}
601}
602
603static void __init early_reserve_mem(void)
604{
605	__be64 *reserve_map;
606
607	reserve_map = (__be64 *)(((unsigned long)initial_boot_params) +
608			fdt_off_mem_rsvmap(initial_boot_params));
609
610	/* Look for the new "reserved-regions" property in the DT */
611	early_reserve_mem_dt();
612
613#ifdef CONFIG_BLK_DEV_INITRD
614	/* Then reserve the initrd, if any */
615	if (initrd_start && (initrd_end > initrd_start)) {
616		memblock_reserve(_ALIGN_DOWN(__pa(initrd_start), PAGE_SIZE),
617			_ALIGN_UP(initrd_end, PAGE_SIZE) -
618			_ALIGN_DOWN(initrd_start, PAGE_SIZE));
619	}
620#endif /* CONFIG_BLK_DEV_INITRD */
621
622#ifdef CONFIG_PPC32
623	/* 
624	 * Handle the case where we might be booting from an old kexec
625	 * image that setup the mem_rsvmap as pairs of 32-bit values
626	 */
627	if (be64_to_cpup(reserve_map) > 0xffffffffull) {
628		u32 base_32, size_32;
629		__be32 *reserve_map_32 = (__be32 *)reserve_map;
630
631		DBG("Found old 32-bit reserve map\n");
632
633		while (1) {
634			base_32 = be32_to_cpup(reserve_map_32++);
635			size_32 = be32_to_cpup(reserve_map_32++);
636			if (size_32 == 0)
637				break;
638			DBG("reserving: %x -> %x\n", base_32, size_32);
639			memblock_reserve(base_32, size_32);
640		}
641		return;
642	}
643#endif
644}
645
646#ifdef CONFIG_PPC_TRANSACTIONAL_MEM
647static bool tm_disabled __initdata;
648
649static int __init parse_ppc_tm(char *str)
650{
651	bool res;
652
653	if (kstrtobool(str, &res))
654		return -EINVAL;
655
656	tm_disabled = !res;
657
658	return 0;
659}
660early_param("ppc_tm", parse_ppc_tm);
661
662static void __init tm_init(void)
663{
664	if (tm_disabled) {
665		pr_info("Disabling hardware transactional memory (HTM)\n");
666		cur_cpu_spec->cpu_user_features2 &=
667			~(PPC_FEATURE2_HTM_NOSC | PPC_FEATURE2_HTM);
668		cur_cpu_spec->cpu_features &= ~CPU_FTR_TM;
669		return;
670	}
671
672	pnv_tm_init();
673}
674#else
675static void tm_init(void) { }
676#endif /* CONFIG_PPC_TRANSACTIONAL_MEM */
677
678void __init early_init_devtree(void *params)
679{
680	phys_addr_t limit;
681
682	DBG(" -> early_init_devtree(%p)\n", params);
683
684	/* Too early to BUG_ON(), do it by hand */
685	if (!early_init_dt_verify(params))
686		panic("BUG: Failed verifying flat device tree, bad version?");
687
688#ifdef CONFIG_PPC_RTAS
689	/* Some machines might need RTAS info for debugging, grab it now. */
690	of_scan_flat_dt(early_init_dt_scan_rtas, NULL);
691#endif
692
693#ifdef CONFIG_PPC_POWERNV
694	/* Some machines might need OPAL info for debugging, grab it now. */
695	of_scan_flat_dt(early_init_dt_scan_opal, NULL);
 
 
 
696#endif
697
698#ifdef CONFIG_FA_DUMP
699	/* scan tree to see if dump is active during last boot */
700	of_scan_flat_dt(early_init_dt_scan_fw_dump, NULL);
701#endif
702
703	/* Retrieve various informations from the /chosen node of the
704	 * device-tree, including the platform type, initrd location and
705	 * size, TCE reserve, and more ...
706	 */
707	of_scan_flat_dt(early_init_dt_scan_chosen_ppc, boot_command_line);
708
709	/* Scan memory nodes and rebuild MEMBLOCKs */
710	of_scan_flat_dt(early_init_dt_scan_root, NULL);
711	of_scan_flat_dt(early_init_dt_scan_memory_ppc, NULL);
712
713	parse_early_param();
714
715	/* make sure we've parsed cmdline for mem= before this */
716	if (memory_limit)
717		first_memblock_size = min_t(u64, first_memblock_size, memory_limit);
718	setup_initial_memory_limit(memstart_addr, first_memblock_size);
719	/* Reserve MEMBLOCK regions used by kernel, initrd, dt, etc... */
720	memblock_reserve(PHYSICAL_START, __pa(klimit) - PHYSICAL_START);
721	/* If relocatable, reserve first 32k for interrupt vectors etc. */
722	if (PHYSICAL_START > MEMORY_START)
723		memblock_reserve(MEMORY_START, 0x8000);
724	reserve_kdump_trampoline();
725#ifdef CONFIG_FA_DUMP
726	/*
727	 * If we fail to reserve memory for firmware-assisted dump then
728	 * fallback to kexec based kdump.
729	 */
730	if (fadump_reserve_mem() == 0)
731#endif
732		reserve_crashkernel();
733	early_reserve_mem();
734
735	/* Ensure that total memory size is page-aligned. */
736	limit = ALIGN(memory_limit ?: memblock_phys_mem_size(), PAGE_SIZE);
737	memblock_enforce_memory_limit(limit);
738
739	memblock_allow_resize();
740	memblock_dump_all();
741
742	DBG("Phys. mem: %llx\n", memblock_phys_mem_size());
743
744	/* We may need to relocate the flat tree, do it now.
745	 * FIXME .. and the initrd too? */
746	move_device_tree();
747
748	allocate_paca_ptrs();
749
750	DBG("Scanning CPUs ...\n");
751
752	dt_cpu_ftrs_scan();
753
754	/* Retrieve CPU related informations from the flat tree
755	 * (altivec support, boot CPU ID, ...)
756	 */
757	of_scan_flat_dt(early_init_dt_scan_cpus, NULL);
758	if (boot_cpuid < 0) {
759		printk("Failed to identify boot CPU !\n");
760		BUG();
761	}
762
763#if defined(CONFIG_SMP) && defined(CONFIG_PPC64)
764	/* We'll later wait for secondaries to check in; there are
765	 * NCPUS-1 non-boot CPUs  :-)
766	 */
767	spinning_secondaries = boot_cpu_count - 1;
768#endif
769
770	mmu_early_init_devtree();
771
772#ifdef CONFIG_PPC_POWERNV
773	/* Scan and build the list of machine check recoverable ranges */
774	of_scan_flat_dt(early_init_dt_scan_recoverable_ranges, NULL);
775#endif
776	epapr_paravirt_early_init();
777
778	/* Now try to figure out if we are running on LPAR and so on */
779	pseries_probe_fw_features();
780
781#ifdef CONFIG_PPC_PS3
782	/* Identify PS3 firmware */
783	if (of_flat_dt_is_compatible(of_get_flat_dt_root(), "sony,ps3"))
784		powerpc_firmware_features |= FW_FEATURE_PS3_POSSIBLE;
785#endif
786
787	tm_init();
788
789	DBG(" <- early_init_devtree()\n");
790}
791
792#ifdef CONFIG_RELOCATABLE
793/*
794 * This function run before early_init_devtree, so we have to init
795 * initial_boot_params.
796 */
797void __init early_get_first_memblock_info(void *params, phys_addr_t *size)
798{
799	/* Setup flat device-tree pointer */
800	initial_boot_params = params;
801
802	/*
803	 * Scan the memory nodes and set add_mem_to_memblock to 0 to avoid
804	 * mess the memblock.
805	 */
806	add_mem_to_memblock = 0;
807	of_scan_flat_dt(early_init_dt_scan_root, NULL);
808	of_scan_flat_dt(early_init_dt_scan_memory_ppc, NULL);
809	add_mem_to_memblock = 1;
810
811	if (size)
812		*size = first_memblock_size;
813}
814#endif
815
816/*******
817 *
818 * New implementation of the OF "find" APIs, return a refcounted
819 * object, call of_node_put() when done.  The device tree and list
820 * are protected by a rw_lock.
821 *
822 * Note that property management will need some locking as well,
823 * this isn't dealt with yet.
824 *
825 *******/
826
827/**
828 * of_get_ibm_chip_id - Returns the IBM "chip-id" of a device
829 * @np: device node of the device
830 *
831 * This looks for a property "ibm,chip-id" in the node or any
832 * of its parents and returns its content, or -1 if it cannot
833 * be found.
834 */
835int of_get_ibm_chip_id(struct device_node *np)
836{
837	of_node_get(np);
838	while (np) {
839		u32 chip_id;
840
841		/*
842		 * Skiboot may produce memory nodes that contain more than one
843		 * cell in chip-id, we only read the first one here.
844		 */
845		if (!of_property_read_u32(np, "ibm,chip-id", &chip_id)) {
846			of_node_put(np);
847			return chip_id;
848		}
849
850		np = of_get_next_parent(np);
851	}
852	return -1;
853}
854EXPORT_SYMBOL(of_get_ibm_chip_id);
855
856/**
857 * cpu_to_chip_id - Return the cpus chip-id
858 * @cpu: The logical cpu number.
859 *
860 * Return the value of the ibm,chip-id property corresponding to the given
861 * logical cpu number. If the chip-id can not be found, returns -1.
862 */
863int cpu_to_chip_id(int cpu)
864{
865	struct device_node *np;
866
867	np = of_get_cpu_node(cpu, NULL);
868	if (!np)
869		return -1;
870
871	of_node_put(np);
872	return of_get_ibm_chip_id(np);
873}
874EXPORT_SYMBOL(cpu_to_chip_id);
875
876bool arch_match_cpu_phys_id(int cpu, u64 phys_id)
877{
878#ifdef CONFIG_SMP
879	/*
880	 * Early firmware scanning must use this rather than
881	 * get_hard_smp_processor_id because we don't have pacas allocated
882	 * until memory topology is discovered.
883	 */
884	if (cpu_to_phys_id != NULL)
885		return (int)phys_id == cpu_to_phys_id[cpu];
886#endif
887
888	return (int)phys_id == get_hard_smp_processor_id(cpu);
889}
v5.4
  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 <stdarg.h>
 15#include <linux/kernel.h>
 16#include <linux/string.h>
 17#include <linux/init.h>
 18#include <linux/threads.h>
 19#include <linux/spinlock.h>
 20#include <linux/types.h>
 21#include <linux/pci.h>
 
 22#include <linux/delay.h>
 23#include <linux/initrd.h>
 24#include <linux/bitops.h>
 25#include <linux/export.h>
 26#include <linux/kexec.h>
 27#include <linux/irq.h>
 28#include <linux/memblock.h>
 29#include <linux/of.h>
 30#include <linux/of_fdt.h>
 31#include <linux/libfdt.h>
 32#include <linux/cpu.h>
 33
 34#include <asm/prom.h>
 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/pgtable.h>
 45#include <asm/powernv.h>
 46#include <asm/iommu.h>
 47#include <asm/btext.h>
 48#include <asm/sections.h>
 49#include <asm/machdep.h>
 50#include <asm/pci-bridge.h>
 51#include <asm/kexec.h>
 52#include <asm/opal.h>
 53#include <asm/fadump.h>
 54#include <asm/epapr_hcalls.h>
 55#include <asm/firmware.h>
 56#include <asm/dt_cpu_ftrs.h>
 57#include <asm/drmem.h>
 58#include <asm/ultravisor.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#ifdef CONFIG_PPC64
 69int __initdata iommu_is_off;
 70int __initdata iommu_force_on;
 71unsigned long tce_alloc_start, tce_alloc_end;
 72u64 ppc64_rma_size;
 73#endif
 74static phys_addr_t first_memblock_size;
 75static int __initdata boot_cpu_count;
 76
 77static int __init early_parse_mem(char *p)
 78{
 79	if (!p)
 80		return 1;
 81
 82	memory_limit = PAGE_ALIGN(memparse(p, &p));
 83	DBG("memory limit = 0x%llx\n", memory_limit);
 84
 85	return 0;
 86}
 87early_param("mem", early_parse_mem);
 88
 89/*
 90 * overlaps_initrd - check for overlap with page aligned extension of
 91 * initrd.
 92 */
 93static inline int overlaps_initrd(unsigned long start, unsigned long size)
 94{
 95#ifdef CONFIG_BLK_DEV_INITRD
 96	if (!initrd_start)
 97		return 0;
 98
 99	return	(start + size) > _ALIGN_DOWN(initrd_start, PAGE_SIZE) &&
100			start <= _ALIGN_UP(initrd_end, PAGE_SIZE);
101#else
102	return 0;
103#endif
104}
105
106/**
107 * move_device_tree - move tree to an unused area, if needed.
108 *
109 * The device tree may be allocated beyond our memory limit, or inside the
110 * crash kernel region for kdump, or within the page aligned range of initrd.
111 * If so, move it out of the way.
112 */
113static void __init move_device_tree(void)
114{
115	unsigned long start, size;
116	void *p;
117
118	DBG("-> move_device_tree\n");
119
120	start = __pa(initial_boot_params);
121	size = fdt_totalsize(initial_boot_params);
122
123	if ((memory_limit && (start + size) > PHYSICAL_START + memory_limit) ||
124	    !memblock_is_memory(start + size - 1) ||
125	    overlaps_crashkernel(start, size) || overlaps_initrd(start, size)) {
126		p = memblock_alloc_raw(size, PAGE_SIZE);
127		if (!p)
128			panic("Failed to allocate %lu bytes to move device tree\n",
129			      size);
130		memcpy(p, initial_boot_params, size);
131		initial_boot_params = p;
132		DBG("Moved device tree to 0x%px\n", p);
133	}
134
135	DBG("<- move_device_tree\n");
136}
137
138/*
139 * ibm,pa-features is a per-cpu property that contains a string of
140 * attribute descriptors, each of which has a 2 byte header plus up
141 * to 254 bytes worth of processor attribute bits.  First header
142 * byte specifies the number of bytes following the header.
143 * Second header byte is an "attribute-specifier" type, of which
144 * zero is the only currently-defined value.
145 * Implementation:  Pass in the byte and bit offset for the feature
146 * that we are interested in.  The function will return -1 if the
147 * pa-features property is missing, or a 1/0 to indicate if the feature
148 * is supported/not supported.  Note that the bit numbers are
149 * big-endian to match the definition in PAPR.
150 */
151static struct ibm_pa_feature {
152	unsigned long	cpu_features;	/* CPU_FTR_xxx bit */
153	unsigned long	mmu_features;	/* MMU_FTR_xxx bit */
154	unsigned int	cpu_user_ftrs;	/* PPC_FEATURE_xxx bit */
155	unsigned int	cpu_user_ftrs2;	/* PPC_FEATURE2_xxx bit */
156	unsigned char	pabyte;		/* byte number in ibm,pa-features */
157	unsigned char	pabit;		/* bit number (big-endian) */
158	unsigned char	invert;		/* if 1, pa bit set => clear feature */
159} ibm_pa_features[] __initdata = {
160	{ .pabyte = 0,  .pabit = 0, .cpu_user_ftrs = PPC_FEATURE_HAS_MMU },
161	{ .pabyte = 0,  .pabit = 1, .cpu_user_ftrs = PPC_FEATURE_HAS_FPU },
162	{ .pabyte = 0,  .pabit = 3, .cpu_features  = CPU_FTR_CTRL },
163	{ .pabyte = 0,  .pabit = 6, .cpu_features  = CPU_FTR_NOEXECUTE },
164	{ .pabyte = 1,  .pabit = 2, .mmu_features  = MMU_FTR_CI_LARGE_PAGE },
165#ifdef CONFIG_PPC_RADIX_MMU
166	{ .pabyte = 40, .pabit = 0, .mmu_features  = MMU_FTR_TYPE_RADIX },
167#endif
168	{ .pabyte = 1,  .pabit = 1, .invert = 1, .cpu_features = CPU_FTR_NODSISRALIGN },
169	{ .pabyte = 5,  .pabit = 0, .cpu_features  = CPU_FTR_REAL_LE,
170				    .cpu_user_ftrs = PPC_FEATURE_TRUE_LE },
171	/*
172	 * If the kernel doesn't support TM (ie CONFIG_PPC_TRANSACTIONAL_MEM=n),
173	 * we don't want to turn on TM here, so we use the *_COMP versions
174	 * which are 0 if the kernel doesn't support TM.
175	 */
176	{ .pabyte = 22, .pabit = 0, .cpu_features = CPU_FTR_TM_COMP,
177	  .cpu_user_ftrs2 = PPC_FEATURE2_HTM_COMP | PPC_FEATURE2_HTM_NOSC_COMP },
178};
179
180static void __init scan_features(unsigned long node, const unsigned char *ftrs,
181				 unsigned long tablelen,
182				 struct ibm_pa_feature *fp,
183				 unsigned long ft_size)
184{
185	unsigned long i, len, bit;
186
187	/* find descriptor with type == 0 */
188	for (;;) {
189		if (tablelen < 3)
190			return;
191		len = 2 + ftrs[0];
192		if (tablelen < len)
193			return;		/* descriptor 0 not found */
194		if (ftrs[1] == 0)
195			break;
196		tablelen -= len;
197		ftrs += len;
198	}
199
200	/* loop over bits we know about */
201	for (i = 0; i < ft_size; ++i, ++fp) {
202		if (fp->pabyte >= ftrs[0])
203			continue;
204		bit = (ftrs[2 + fp->pabyte] >> (7 - fp->pabit)) & 1;
205		if (bit ^ fp->invert) {
206			cur_cpu_spec->cpu_features |= fp->cpu_features;
207			cur_cpu_spec->cpu_user_features |= fp->cpu_user_ftrs;
208			cur_cpu_spec->cpu_user_features2 |= fp->cpu_user_ftrs2;
209			cur_cpu_spec->mmu_features |= fp->mmu_features;
210		} else {
211			cur_cpu_spec->cpu_features &= ~fp->cpu_features;
212			cur_cpu_spec->cpu_user_features &= ~fp->cpu_user_ftrs;
213			cur_cpu_spec->cpu_user_features2 &= ~fp->cpu_user_ftrs2;
214			cur_cpu_spec->mmu_features &= ~fp->mmu_features;
215		}
216	}
217}
218
219static void __init check_cpu_pa_features(unsigned long node)
220{
221	const unsigned char *pa_ftrs;
222	int tablelen;
223
224	pa_ftrs = of_get_flat_dt_prop(node, "ibm,pa-features", &tablelen);
225	if (pa_ftrs == NULL)
226		return;
227
228	scan_features(node, pa_ftrs, tablelen,
229		      ibm_pa_features, ARRAY_SIZE(ibm_pa_features));
230}
231
232#ifdef CONFIG_PPC_BOOK3S_64
233static void __init init_mmu_slb_size(unsigned long node)
234{
235	const __be32 *slb_size_ptr;
236
237	slb_size_ptr = of_get_flat_dt_prop(node, "slb-size", NULL) ? :
238			of_get_flat_dt_prop(node, "ibm,slb-size", NULL);
239
240	if (slb_size_ptr)
241		mmu_slb_size = be32_to_cpup(slb_size_ptr);
242}
243#else
244#define init_mmu_slb_size(node) do { } while(0)
245#endif
246
247static struct feature_property {
248	const char *name;
249	u32 min_value;
250	unsigned long cpu_feature;
251	unsigned long cpu_user_ftr;
252} feature_properties[] __initdata = {
253#ifdef CONFIG_ALTIVEC
254	{"altivec", 0, CPU_FTR_ALTIVEC, PPC_FEATURE_HAS_ALTIVEC},
255	{"ibm,vmx", 1, CPU_FTR_ALTIVEC, PPC_FEATURE_HAS_ALTIVEC},
256#endif /* CONFIG_ALTIVEC */
257#ifdef CONFIG_VSX
258	/* Yes, this _really_ is ibm,vmx == 2 to enable VSX */
259	{"ibm,vmx", 2, CPU_FTR_VSX, PPC_FEATURE_HAS_VSX},
260#endif /* CONFIG_VSX */
261#ifdef CONFIG_PPC64
262	{"ibm,dfp", 1, 0, PPC_FEATURE_HAS_DFP},
263	{"ibm,purr", 1, CPU_FTR_PURR, 0},
264	{"ibm,spurr", 1, CPU_FTR_SPURR, 0},
265#endif /* CONFIG_PPC64 */
266};
267
268#if defined(CONFIG_44x) && defined(CONFIG_PPC_FPU)
269static inline void identical_pvr_fixup(unsigned long node)
270{
271	unsigned int pvr;
272	const char *model = of_get_flat_dt_prop(node, "model", NULL);
273
274	/*
275	 * Since 440GR(x)/440EP(x) processors have the same pvr,
276	 * we check the node path and set bit 28 in the cur_cpu_spec
277	 * pvr for EP(x) processor version. This bit is always 0 in
278	 * the "real" pvr. Then we call identify_cpu again with
279	 * the new logical pvr to enable FPU support.
280	 */
281	if (model && strstr(model, "440EP")) {
282		pvr = cur_cpu_spec->pvr_value | 0x8;
283		identify_cpu(0, pvr);
284		DBG("Using logical pvr %x for %s\n", pvr, model);
285	}
286}
287#else
288#define identical_pvr_fixup(node) do { } while(0)
289#endif
290
291static void __init check_cpu_feature_properties(unsigned long node)
292{
293	int i;
294	struct feature_property *fp = feature_properties;
295	const __be32 *prop;
296
297	for (i = 0; i < (int)ARRAY_SIZE(feature_properties); ++i, ++fp) {
298		prop = of_get_flat_dt_prop(node, fp->name, NULL);
299		if (prop && be32_to_cpup(prop) >= fp->min_value) {
300			cur_cpu_spec->cpu_features |= fp->cpu_feature;
301			cur_cpu_spec->cpu_user_features |= fp->cpu_user_ftr;
302		}
303	}
304}
305
306static int __init early_init_dt_scan_cpus(unsigned long node,
307					  const char *uname, int depth,
308					  void *data)
309{
310	const char *type = of_get_flat_dt_prop(node, "device_type", NULL);
311	const __be32 *prop;
312	const __be32 *intserv;
313	int i, nthreads;
314	int len;
315	int found = -1;
316	int found_thread = 0;
317
318	/* We are scanning "cpu" nodes only */
319	if (type == NULL || strcmp(type, "cpu") != 0)
320		return 0;
321
322	/* Get physical cpuid */
323	intserv = of_get_flat_dt_prop(node, "ibm,ppc-interrupt-server#s", &len);
324	if (!intserv)
325		intserv = of_get_flat_dt_prop(node, "reg", &len);
326
327	nthreads = len / sizeof(int);
328
329	/*
330	 * Now see if any of these threads match our boot cpu.
331	 * NOTE: This must match the parsing done in smp_setup_cpu_maps.
332	 */
333	for (i = 0; i < nthreads; i++) {
334		if (be32_to_cpu(intserv[i]) ==
335			fdt_boot_cpuid_phys(initial_boot_params)) {
336			found = boot_cpu_count;
337			found_thread = i;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
338		}
339#ifdef CONFIG_SMP
340		/* logical cpu id is always 0 on UP kernels */
341		boot_cpu_count++;
342#endif
343	}
344
345	/* Not the boot CPU */
346	if (found < 0)
347		return 0;
348
349	DBG("boot cpu: logical %d physical %d\n", found,
350	    be32_to_cpu(intserv[found_thread]));
351	boot_cpuid = found;
352
353	/*
354	 * PAPR defines "logical" PVR values for cpus that
355	 * meet various levels of the architecture:
356	 * 0x0f000001	Architecture version 2.04
357	 * 0x0f000002	Architecture version 2.05
358	 * If the cpu-version property in the cpu node contains
359	 * such a value, we call identify_cpu again with the
360	 * logical PVR value in order to use the cpu feature
361	 * bits appropriate for the architecture level.
362	 *
363	 * A POWER6 partition in "POWER6 architected" mode
364	 * uses the 0x0f000002 PVR value; in POWER5+ mode
365	 * it uses 0x0f000001.
366	 *
367	 * If we're using device tree CPU feature discovery then we don't
368	 * support the cpu-version property, and it's the responsibility of the
369	 * firmware/hypervisor to provide the correct feature set for the
370	 * architecture level via the ibm,powerpc-cpu-features binding.
371	 */
372	if (!dt_cpu_ftrs_in_use()) {
373		prop = of_get_flat_dt_prop(node, "cpu-version", NULL);
374		if (prop && (be32_to_cpup(prop) & 0xff000000) == 0x0f000000)
375			identify_cpu(0, be32_to_cpup(prop));
376
377		check_cpu_feature_properties(node);
378		check_cpu_pa_features(node);
379	}
380
381	identical_pvr_fixup(node);
382	init_mmu_slb_size(node);
383
384#ifdef CONFIG_PPC64
385	if (nthreads == 1)
386		cur_cpu_spec->cpu_features &= ~CPU_FTR_SMT;
387	else if (!dt_cpu_ftrs_in_use())
388		cur_cpu_spec->cpu_features |= CPU_FTR_SMT;
389	allocate_paca(boot_cpuid);
390#endif
391	set_hard_smp_processor_id(found, be32_to_cpu(intserv[found_thread]));
392
393	return 0;
394}
395
396static int __init early_init_dt_scan_chosen_ppc(unsigned long node,
397						const char *uname,
398						int depth, void *data)
399{
400	const unsigned long *lprop; /* All these set by kernel, so no need to convert endian */
401
402	/* Use common scan routine to determine if this is the chosen node */
403	if (early_init_dt_scan_chosen(node, uname, depth, data) == 0)
404		return 0;
405
406#ifdef CONFIG_PPC64
407	/* check if iommu is forced on or off */
408	if (of_get_flat_dt_prop(node, "linux,iommu-off", NULL) != NULL)
409		iommu_is_off = 1;
410	if (of_get_flat_dt_prop(node, "linux,iommu-force-on", NULL) != NULL)
411		iommu_force_on = 1;
412#endif
413
414	/* mem=x on the command line is the preferred mechanism */
415	lprop = of_get_flat_dt_prop(node, "linux,memory-limit", NULL);
416	if (lprop)
417		memory_limit = *lprop;
418
419#ifdef CONFIG_PPC64
420	lprop = of_get_flat_dt_prop(node, "linux,tce-alloc-start", NULL);
421	if (lprop)
422		tce_alloc_start = *lprop;
423	lprop = of_get_flat_dt_prop(node, "linux,tce-alloc-end", NULL);
424	if (lprop)
425		tce_alloc_end = *lprop;
426#endif
427
428#ifdef CONFIG_KEXEC_CORE
429	lprop = of_get_flat_dt_prop(node, "linux,crashkernel-base", NULL);
430	if (lprop)
431		crashk_res.start = *lprop;
432
433	lprop = of_get_flat_dt_prop(node, "linux,crashkernel-size", NULL);
434	if (lprop)
435		crashk_res.end = crashk_res.start + *lprop - 1;
436#endif
437
438	/* break now */
439	return 1;
440}
441
442/*
443 * Compare the range against max mem limit and update
444 * size if it cross the limit.
445 */
446
447#ifdef CONFIG_SPARSEMEM
448static bool validate_mem_limit(u64 base, u64 *size)
449{
450	u64 max_mem = 1UL << (MAX_PHYSMEM_BITS);
451
452	if (base >= max_mem)
453		return false;
454	if ((base + *size) > max_mem)
455		*size = max_mem - base;
456	return true;
457}
458#else
459static bool validate_mem_limit(u64 base, u64 *size)
460{
461	return true;
462}
463#endif
464
465#ifdef CONFIG_PPC_PSERIES
466/*
467 * Interpret the ibm dynamic reconfiguration memory LMBs.
468 * This contains a list of memory blocks along with NUMA affinity
469 * information.
470 */
471static void __init early_init_drmem_lmb(struct drmem_lmb *lmb,
472					const __be32 **usm)
473{
474	u64 base, size;
475	int is_kexec_kdump = 0, rngs;
476
477	base = lmb->base_addr;
478	size = drmem_lmb_size();
479	rngs = 1;
480
481	/*
482	 * Skip this block if the reserved bit is set in flags
483	 * or if the block is not assigned to this partition.
484	 */
485	if ((lmb->flags & DRCONF_MEM_RESERVED) ||
486	    !(lmb->flags & DRCONF_MEM_ASSIGNED))
487		return;
488
489	if (*usm)
490		is_kexec_kdump = 1;
491
492	if (is_kexec_kdump) {
493		/*
494		 * For each memblock in ibm,dynamic-memory, a
495		 * corresponding entry in linux,drconf-usable-memory
496		 * property contains a counter 'p' followed by 'p'
497		 * (base, size) duple. Now read the counter from
498		 * linux,drconf-usable-memory property
499		 */
500		rngs = dt_mem_next_cell(dt_root_size_cells, usm);
501		if (!rngs) /* there are no (base, size) duple */
502			return;
503	}
504
505	do {
506		if (is_kexec_kdump) {
507			base = dt_mem_next_cell(dt_root_addr_cells, usm);
508			size = dt_mem_next_cell(dt_root_size_cells, usm);
509		}
510
511		if (iommu_is_off) {
512			if (base >= 0x80000000ul)
513				continue;
514			if ((base + size) > 0x80000000ul)
515				size = 0x80000000ul - base;
516		}
517
518		DBG("Adding: %llx -> %llx\n", base, size);
519		if (validate_mem_limit(base, &size))
520			memblock_add(base, size);
521	} while (--rngs);
522}
523#endif /* CONFIG_PPC_PSERIES */
524
525static int __init early_init_dt_scan_memory_ppc(unsigned long node,
526						const char *uname,
527						int depth, void *data)
528{
529#ifdef CONFIG_PPC_PSERIES
530	if (depth == 1 &&
531	    strcmp(uname, "ibm,dynamic-reconfiguration-memory") == 0) {
532		walk_drmem_lmbs_early(node, early_init_drmem_lmb);
533		return 0;
534	}
535#endif
536	
537	return early_init_dt_scan_memory(node, uname, depth, data);
538}
539
540/*
541 * For a relocatable kernel, we need to get the memstart_addr first,
542 * then use it to calculate the virtual kernel start address. This has
543 * to happen at a very early stage (before machine_init). In this case,
544 * we just want to get the memstart_address and would not like to mess the
545 * memblock at this stage. So introduce a variable to skip the memblock_add()
546 * for this reason.
547 */
548#ifdef CONFIG_RELOCATABLE
549static int add_mem_to_memblock = 1;
550#else
551#define add_mem_to_memblock 1
552#endif
553
554void __init early_init_dt_add_memory_arch(u64 base, u64 size)
555{
556#ifdef CONFIG_PPC64
557	if (iommu_is_off) {
558		if (base >= 0x80000000ul)
559			return;
560		if ((base + size) > 0x80000000ul)
561			size = 0x80000000ul - base;
562	}
563#endif
564	/* Keep track of the beginning of memory -and- the size of
565	 * the very first block in the device-tree as it represents
566	 * the RMA on ppc64 server
567	 */
568	if (base < memstart_addr) {
569		memstart_addr = base;
570		first_memblock_size = size;
571	}
572
573	/* Add the chunk to the MEMBLOCK list */
574	if (add_mem_to_memblock) {
575		if (validate_mem_limit(base, &size))
576			memblock_add(base, size);
577	}
578}
579
580static void __init early_reserve_mem_dt(void)
581{
582	unsigned long i, dt_root;
583	int len;
584	const __be32 *prop;
585
586	early_init_fdt_reserve_self();
587	early_init_fdt_scan_reserved_mem();
588
589	dt_root = of_get_flat_dt_root();
590
591	prop = of_get_flat_dt_prop(dt_root, "reserved-ranges", &len);
592
593	if (!prop)
594		return;
595
596	DBG("Found new-style reserved-ranges\n");
597
598	/* Each reserved range is an (address,size) pair, 2 cells each,
599	 * totalling 4 cells per range. */
600	for (i = 0; i < len / (sizeof(*prop) * 4); i++) {
601		u64 base, size;
602
603		base = of_read_number(prop + (i * 4) + 0, 2);
604		size = of_read_number(prop + (i * 4) + 2, 2);
605
606		if (size) {
607			DBG("reserving: %llx -> %llx\n", base, size);
608			memblock_reserve(base, size);
609		}
610	}
611}
612
613static void __init early_reserve_mem(void)
614{
615	__be64 *reserve_map;
616
617	reserve_map = (__be64 *)(((unsigned long)initial_boot_params) +
618			fdt_off_mem_rsvmap(initial_boot_params));
619
620	/* Look for the new "reserved-regions" property in the DT */
621	early_reserve_mem_dt();
622
623#ifdef CONFIG_BLK_DEV_INITRD
624	/* Then reserve the initrd, if any */
625	if (initrd_start && (initrd_end > initrd_start)) {
626		memblock_reserve(_ALIGN_DOWN(__pa(initrd_start), PAGE_SIZE),
627			_ALIGN_UP(initrd_end, PAGE_SIZE) -
628			_ALIGN_DOWN(initrd_start, PAGE_SIZE));
629	}
630#endif /* CONFIG_BLK_DEV_INITRD */
631
632#ifdef CONFIG_PPC32
633	/* 
634	 * Handle the case where we might be booting from an old kexec
635	 * image that setup the mem_rsvmap as pairs of 32-bit values
636	 */
637	if (be64_to_cpup(reserve_map) > 0xffffffffull) {
638		u32 base_32, size_32;
639		__be32 *reserve_map_32 = (__be32 *)reserve_map;
640
641		DBG("Found old 32-bit reserve map\n");
642
643		while (1) {
644			base_32 = be32_to_cpup(reserve_map_32++);
645			size_32 = be32_to_cpup(reserve_map_32++);
646			if (size_32 == 0)
647				break;
648			DBG("reserving: %x -> %x\n", base_32, size_32);
649			memblock_reserve(base_32, size_32);
650		}
651		return;
652	}
653#endif
654}
655
656#ifdef CONFIG_PPC_TRANSACTIONAL_MEM
657static bool tm_disabled __initdata;
658
659static int __init parse_ppc_tm(char *str)
660{
661	bool res;
662
663	if (kstrtobool(str, &res))
664		return -EINVAL;
665
666	tm_disabled = !res;
667
668	return 0;
669}
670early_param("ppc_tm", parse_ppc_tm);
671
672static void __init tm_init(void)
673{
674	if (tm_disabled) {
675		pr_info("Disabling hardware transactional memory (HTM)\n");
676		cur_cpu_spec->cpu_user_features2 &=
677			~(PPC_FEATURE2_HTM_NOSC | PPC_FEATURE2_HTM);
678		cur_cpu_spec->cpu_features &= ~CPU_FTR_TM;
679		return;
680	}
681
682	pnv_tm_init();
683}
684#else
685static void tm_init(void) { }
686#endif /* CONFIG_PPC_TRANSACTIONAL_MEM */
687
688void __init early_init_devtree(void *params)
689{
690	phys_addr_t limit;
691
692	DBG(" -> early_init_devtree(%px)\n", params);
693
694	/* Too early to BUG_ON(), do it by hand */
695	if (!early_init_dt_verify(params))
696		panic("BUG: Failed verifying flat device tree, bad version?");
697
698#ifdef CONFIG_PPC_RTAS
699	/* Some machines might need RTAS info for debugging, grab it now. */
700	of_scan_flat_dt(early_init_dt_scan_rtas, NULL);
701#endif
702
703#ifdef CONFIG_PPC_POWERNV
704	/* Some machines might need OPAL info for debugging, grab it now. */
705	of_scan_flat_dt(early_init_dt_scan_opal, NULL);
706
707	/* Scan tree for ultravisor feature */
708	of_scan_flat_dt(early_init_dt_scan_ultravisor, NULL);
709#endif
710
711#if defined(CONFIG_FA_DUMP) || defined(CONFIG_PRESERVE_FA_DUMP)
712	/* scan tree to see if dump is active during last boot */
713	of_scan_flat_dt(early_init_dt_scan_fw_dump, NULL);
714#endif
715
716	/* Retrieve various informations from the /chosen node of the
717	 * device-tree, including the platform type, initrd location and
718	 * size, TCE reserve, and more ...
719	 */
720	of_scan_flat_dt(early_init_dt_scan_chosen_ppc, boot_command_line);
721
722	/* Scan memory nodes and rebuild MEMBLOCKs */
723	of_scan_flat_dt(early_init_dt_scan_root, NULL);
724	of_scan_flat_dt(early_init_dt_scan_memory_ppc, NULL);
725
726	parse_early_param();
727
728	/* make sure we've parsed cmdline for mem= before this */
729	if (memory_limit)
730		first_memblock_size = min_t(u64, first_memblock_size, memory_limit);
731	setup_initial_memory_limit(memstart_addr, first_memblock_size);
732	/* Reserve MEMBLOCK regions used by kernel, initrd, dt, etc... */
733	memblock_reserve(PHYSICAL_START, __pa(klimit) - PHYSICAL_START);
734	/* If relocatable, reserve first 32k for interrupt vectors etc. */
735	if (PHYSICAL_START > MEMORY_START)
736		memblock_reserve(MEMORY_START, 0x8000);
737	reserve_kdump_trampoline();
738#if defined(CONFIG_FA_DUMP) || defined(CONFIG_PRESERVE_FA_DUMP)
739	/*
740	 * If we fail to reserve memory for firmware-assisted dump then
741	 * fallback to kexec based kdump.
742	 */
743	if (fadump_reserve_mem() == 0)
744#endif
745		reserve_crashkernel();
746	early_reserve_mem();
747
748	/* Ensure that total memory size is page-aligned. */
749	limit = ALIGN(memory_limit ?: memblock_phys_mem_size(), PAGE_SIZE);
750	memblock_enforce_memory_limit(limit);
751
752	memblock_allow_resize();
753	memblock_dump_all();
754
755	DBG("Phys. mem: %llx\n", (unsigned long long)memblock_phys_mem_size());
756
757	/* We may need to relocate the flat tree, do it now.
758	 * FIXME .. and the initrd too? */
759	move_device_tree();
760
761	allocate_paca_ptrs();
762
763	DBG("Scanning CPUs ...\n");
764
765	dt_cpu_ftrs_scan();
766
767	/* Retrieve CPU related informations from the flat tree
768	 * (altivec support, boot CPU ID, ...)
769	 */
770	of_scan_flat_dt(early_init_dt_scan_cpus, NULL);
771	if (boot_cpuid < 0) {
772		printk("Failed to identify boot CPU !\n");
773		BUG();
774	}
775
776#if defined(CONFIG_SMP) && defined(CONFIG_PPC64)
777	/* We'll later wait for secondaries to check in; there are
778	 * NCPUS-1 non-boot CPUs  :-)
779	 */
780	spinning_secondaries = boot_cpu_count - 1;
781#endif
782
783	mmu_early_init_devtree();
784
785#ifdef CONFIG_PPC_POWERNV
786	/* Scan and build the list of machine check recoverable ranges */
787	of_scan_flat_dt(early_init_dt_scan_recoverable_ranges, NULL);
788#endif
789	epapr_paravirt_early_init();
790
791	/* Now try to figure out if we are running on LPAR and so on */
792	pseries_probe_fw_features();
793
794#ifdef CONFIG_PPC_PS3
795	/* Identify PS3 firmware */
796	if (of_flat_dt_is_compatible(of_get_flat_dt_root(), "sony,ps3"))
797		powerpc_firmware_features |= FW_FEATURE_PS3_POSSIBLE;
798#endif
799
800	tm_init();
801
802	DBG(" <- early_init_devtree()\n");
803}
804
805#ifdef CONFIG_RELOCATABLE
806/*
807 * This function run before early_init_devtree, so we have to init
808 * initial_boot_params.
809 */
810void __init early_get_first_memblock_info(void *params, phys_addr_t *size)
811{
812	/* Setup flat device-tree pointer */
813	initial_boot_params = params;
814
815	/*
816	 * Scan the memory nodes and set add_mem_to_memblock to 0 to avoid
817	 * mess the memblock.
818	 */
819	add_mem_to_memblock = 0;
820	of_scan_flat_dt(early_init_dt_scan_root, NULL);
821	of_scan_flat_dt(early_init_dt_scan_memory_ppc, NULL);
822	add_mem_to_memblock = 1;
823
824	if (size)
825		*size = first_memblock_size;
826}
827#endif
828
829/*******
830 *
831 * New implementation of the OF "find" APIs, return a refcounted
832 * object, call of_node_put() when done.  The device tree and list
833 * are protected by a rw_lock.
834 *
835 * Note that property management will need some locking as well,
836 * this isn't dealt with yet.
837 *
838 *******/
839
840/**
841 * of_get_ibm_chip_id - Returns the IBM "chip-id" of a device
842 * @np: device node of the device
843 *
844 * This looks for a property "ibm,chip-id" in the node or any
845 * of its parents and returns its content, or -1 if it cannot
846 * be found.
847 */
848int of_get_ibm_chip_id(struct device_node *np)
849{
850	of_node_get(np);
851	while (np) {
852		u32 chip_id;
853
854		/*
855		 * Skiboot may produce memory nodes that contain more than one
856		 * cell in chip-id, we only read the first one here.
857		 */
858		if (!of_property_read_u32(np, "ibm,chip-id", &chip_id)) {
859			of_node_put(np);
860			return chip_id;
861		}
862
863		np = of_get_next_parent(np);
864	}
865	return -1;
866}
867EXPORT_SYMBOL(of_get_ibm_chip_id);
868
869/**
870 * cpu_to_chip_id - Return the cpus chip-id
871 * @cpu: The logical cpu number.
872 *
873 * Return the value of the ibm,chip-id property corresponding to the given
874 * logical cpu number. If the chip-id can not be found, returns -1.
875 */
876int cpu_to_chip_id(int cpu)
877{
878	struct device_node *np;
879
880	np = of_get_cpu_node(cpu, NULL);
881	if (!np)
882		return -1;
883
884	of_node_put(np);
885	return of_get_ibm_chip_id(np);
886}
887EXPORT_SYMBOL(cpu_to_chip_id);
888
889bool arch_match_cpu_phys_id(int cpu, u64 phys_id)
890{
891#ifdef CONFIG_SMP
892	/*
893	 * Early firmware scanning must use this rather than
894	 * get_hard_smp_processor_id because we don't have pacas allocated
895	 * until memory topology is discovered.
896	 */
897	if (cpu_to_phys_id != NULL)
898		return (int)phys_id == cpu_to_phys_id[cpu];
899#endif
900
901	return (int)phys_id == get_hard_smp_processor_id(cpu);
902}