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 #include <linux/pgtable.h>
34
35 #include <asm/prom.h>
36 #include <asm/rtas.h>
37 #include <asm/page.h>
38 #include <asm/processor.h>
39 #include <asm/irq.h>
40 #include <asm/io.h>
41 #include <asm/kdump.h>
42 #include <asm/smp.h>
43 #include <asm/mmu.h>
44 #include <asm/paca.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
69 int __initdata iommu_is_off;
70 int __initdata iommu_force_on;
71 unsigned long tce_alloc_start, tce_alloc_end;
72 u64 ppc64_rma_size;
73 #endif
74 static phys_addr_t first_memblock_size;
75 static int __initdata boot_cpu_count;
76
early_parse_mem(char * p)77 static 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 }
87 early_param("mem", early_parse_mem);
88
89 /*
90 * overlaps_initrd - check for overlap with page aligned extension of
91 * initrd.
92 */
overlaps_initrd(unsigned long start,unsigned long size)93 static 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(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 */
move_device_tree(void)113 static 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 */
151 static 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 | MMU_FTR_GTSE },
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 { .pabyte = 64, .pabit = 0, .cpu_features = CPU_FTR_DAWR1 },
180 };
181
scan_features(unsigned long node,const unsigned char * ftrs,unsigned long tablelen,struct ibm_pa_feature * fp,unsigned long ft_size)182 static void __init scan_features(unsigned long node, const unsigned char *ftrs,
183 unsigned long tablelen,
184 struct ibm_pa_feature *fp,
185 unsigned long ft_size)
186 {
187 unsigned long i, len, bit;
188
189 /* find descriptor with type == 0 */
190 for (;;) {
191 if (tablelen < 3)
192 return;
193 len = 2 + ftrs[0];
194 if (tablelen < len)
195 return; /* descriptor 0 not found */
196 if (ftrs[1] == 0)
197 break;
198 tablelen -= len;
199 ftrs += len;
200 }
201
202 /* loop over bits we know about */
203 for (i = 0; i < ft_size; ++i, ++fp) {
204 if (fp->pabyte >= ftrs[0])
205 continue;
206 bit = (ftrs[2 + fp->pabyte] >> (7 - fp->pabit)) & 1;
207 if (bit ^ fp->invert) {
208 cur_cpu_spec->cpu_features |= fp->cpu_features;
209 cur_cpu_spec->cpu_user_features |= fp->cpu_user_ftrs;
210 cur_cpu_spec->cpu_user_features2 |= fp->cpu_user_ftrs2;
211 cur_cpu_spec->mmu_features |= fp->mmu_features;
212 } else {
213 cur_cpu_spec->cpu_features &= ~fp->cpu_features;
214 cur_cpu_spec->cpu_user_features &= ~fp->cpu_user_ftrs;
215 cur_cpu_spec->cpu_user_features2 &= ~fp->cpu_user_ftrs2;
216 cur_cpu_spec->mmu_features &= ~fp->mmu_features;
217 }
218 }
219 }
220
check_cpu_pa_features(unsigned long node)221 static void __init check_cpu_pa_features(unsigned long node)
222 {
223 const unsigned char *pa_ftrs;
224 int tablelen;
225
226 pa_ftrs = of_get_flat_dt_prop(node, "ibm,pa-features", &tablelen);
227 if (pa_ftrs == NULL)
228 return;
229
230 scan_features(node, pa_ftrs, tablelen,
231 ibm_pa_features, ARRAY_SIZE(ibm_pa_features));
232 }
233
234 #ifdef CONFIG_PPC_BOOK3S_64
init_mmu_slb_size(unsigned long node)235 static void __init init_mmu_slb_size(unsigned long node)
236 {
237 const __be32 *slb_size_ptr;
238
239 slb_size_ptr = of_get_flat_dt_prop(node, "slb-size", NULL) ? :
240 of_get_flat_dt_prop(node, "ibm,slb-size", NULL);
241
242 if (slb_size_ptr)
243 mmu_slb_size = be32_to_cpup(slb_size_ptr);
244 }
245 #else
246 #define init_mmu_slb_size(node) do { } while(0)
247 #endif
248
249 static struct feature_property {
250 const char *name;
251 u32 min_value;
252 unsigned long cpu_feature;
253 unsigned long cpu_user_ftr;
254 } feature_properties[] __initdata = {
255 #ifdef CONFIG_ALTIVEC
256 {"altivec", 0, CPU_FTR_ALTIVEC, PPC_FEATURE_HAS_ALTIVEC},
257 {"ibm,vmx", 1, CPU_FTR_ALTIVEC, PPC_FEATURE_HAS_ALTIVEC},
258 #endif /* CONFIG_ALTIVEC */
259 #ifdef CONFIG_VSX
260 /* Yes, this _really_ is ibm,vmx == 2 to enable VSX */
261 {"ibm,vmx", 2, CPU_FTR_VSX, PPC_FEATURE_HAS_VSX},
262 #endif /* CONFIG_VSX */
263 #ifdef CONFIG_PPC64
264 {"ibm,dfp", 1, 0, PPC_FEATURE_HAS_DFP},
265 {"ibm,purr", 1, CPU_FTR_PURR, 0},
266 {"ibm,spurr", 1, CPU_FTR_SPURR, 0},
267 #endif /* CONFIG_PPC64 */
268 };
269
270 #if defined(CONFIG_44x) && defined(CONFIG_PPC_FPU)
identical_pvr_fixup(unsigned long node)271 static inline void identical_pvr_fixup(unsigned long node)
272 {
273 unsigned int pvr;
274 const char *model = of_get_flat_dt_prop(node, "model", NULL);
275
276 /*
277 * Since 440GR(x)/440EP(x) processors have the same pvr,
278 * we check the node path and set bit 28 in the cur_cpu_spec
279 * pvr for EP(x) processor version. This bit is always 0 in
280 * the "real" pvr. Then we call identify_cpu again with
281 * the new logical pvr to enable FPU support.
282 */
283 if (model && strstr(model, "440EP")) {
284 pvr = cur_cpu_spec->pvr_value | 0x8;
285 identify_cpu(0, pvr);
286 DBG("Using logical pvr %x for %s\n", pvr, model);
287 }
288 }
289 #else
290 #define identical_pvr_fixup(node) do { } while(0)
291 #endif
292
check_cpu_feature_properties(unsigned long node)293 static void __init check_cpu_feature_properties(unsigned long node)
294 {
295 int i;
296 struct feature_property *fp = feature_properties;
297 const __be32 *prop;
298
299 for (i = 0; i < (int)ARRAY_SIZE(feature_properties); ++i, ++fp) {
300 prop = of_get_flat_dt_prop(node, fp->name, NULL);
301 if (prop && be32_to_cpup(prop) >= fp->min_value) {
302 cur_cpu_spec->cpu_features |= fp->cpu_feature;
303 cur_cpu_spec->cpu_user_features |= fp->cpu_user_ftr;
304 }
305 }
306 }
307
early_init_dt_scan_cpus(unsigned long node,const char * uname,int depth,void * data)308 static int __init early_init_dt_scan_cpus(unsigned long node,
309 const char *uname, int depth,
310 void *data)
311 {
312 const char *type = of_get_flat_dt_prop(node, "device_type", NULL);
313 const __be32 *prop;
314 const __be32 *intserv;
315 int i, nthreads;
316 int len;
317 int found = -1;
318 int found_thread = 0;
319
320 /* We are scanning "cpu" nodes only */
321 if (type == NULL || strcmp(type, "cpu") != 0)
322 return 0;
323
324 /* Get physical cpuid */
325 intserv = of_get_flat_dt_prop(node, "ibm,ppc-interrupt-server#s", &len);
326 if (!intserv)
327 intserv = of_get_flat_dt_prop(node, "reg", &len);
328
329 nthreads = len / sizeof(int);
330
331 /*
332 * Now see if any of these threads match our boot cpu.
333 * NOTE: This must match the parsing done in smp_setup_cpu_maps.
334 */
335 for (i = 0; i < nthreads; i++) {
336 if (be32_to_cpu(intserv[i]) ==
337 fdt_boot_cpuid_phys(initial_boot_params)) {
338 found = boot_cpu_count;
339 found_thread = i;
340 }
341 #ifdef CONFIG_SMP
342 /* logical cpu id is always 0 on UP kernels */
343 boot_cpu_count++;
344 #endif
345 }
346
347 /* Not the boot CPU */
348 if (found < 0)
349 return 0;
350
351 DBG("boot cpu: logical %d physical %d\n", found,
352 be32_to_cpu(intserv[found_thread]));
353 boot_cpuid = found;
354
355 /*
356 * PAPR defines "logical" PVR values for cpus that
357 * meet various levels of the architecture:
358 * 0x0f000001 Architecture version 2.04
359 * 0x0f000002 Architecture version 2.05
360 * If the cpu-version property in the cpu node contains
361 * such a value, we call identify_cpu again with the
362 * logical PVR value in order to use the cpu feature
363 * bits appropriate for the architecture level.
364 *
365 * A POWER6 partition in "POWER6 architected" mode
366 * uses the 0x0f000002 PVR value; in POWER5+ mode
367 * it uses 0x0f000001.
368 *
369 * If we're using device tree CPU feature discovery then we don't
370 * support the cpu-version property, and it's the responsibility of the
371 * firmware/hypervisor to provide the correct feature set for the
372 * architecture level via the ibm,powerpc-cpu-features binding.
373 */
374 if (!dt_cpu_ftrs_in_use()) {
375 prop = of_get_flat_dt_prop(node, "cpu-version", NULL);
376 if (prop && (be32_to_cpup(prop) & 0xff000000) == 0x0f000000)
377 identify_cpu(0, be32_to_cpup(prop));
378
379 check_cpu_feature_properties(node);
380 check_cpu_pa_features(node);
381 }
382
383 identical_pvr_fixup(node);
384 init_mmu_slb_size(node);
385
386 #ifdef CONFIG_PPC64
387 if (nthreads == 1)
388 cur_cpu_spec->cpu_features &= ~CPU_FTR_SMT;
389 else if (!dt_cpu_ftrs_in_use())
390 cur_cpu_spec->cpu_features |= CPU_FTR_SMT;
391 allocate_paca(boot_cpuid);
392 #endif
393 set_hard_smp_processor_id(found, be32_to_cpu(intserv[found_thread]));
394
395 return 0;
396 }
397
early_init_dt_scan_chosen_ppc(unsigned long node,const char * uname,int depth,void * data)398 static int __init early_init_dt_scan_chosen_ppc(unsigned long node,
399 const char *uname,
400 int depth, void *data)
401 {
402 const unsigned long *lprop; /* All these set by kernel, so no need to convert endian */
403
404 /* Use common scan routine to determine if this is the chosen node */
405 if (early_init_dt_scan_chosen(node, uname, depth, data) == 0)
406 return 0;
407
408 #ifdef CONFIG_PPC64
409 /* check if iommu is forced on or off */
410 if (of_get_flat_dt_prop(node, "linux,iommu-off", NULL) != NULL)
411 iommu_is_off = 1;
412 if (of_get_flat_dt_prop(node, "linux,iommu-force-on", NULL) != NULL)
413 iommu_force_on = 1;
414 #endif
415
416 /* mem=x on the command line is the preferred mechanism */
417 lprop = of_get_flat_dt_prop(node, "linux,memory-limit", NULL);
418 if (lprop)
419 memory_limit = *lprop;
420
421 #ifdef CONFIG_PPC64
422 lprop = of_get_flat_dt_prop(node, "linux,tce-alloc-start", NULL);
423 if (lprop)
424 tce_alloc_start = *lprop;
425 lprop = of_get_flat_dt_prop(node, "linux,tce-alloc-end", NULL);
426 if (lprop)
427 tce_alloc_end = *lprop;
428 #endif
429
430 #ifdef CONFIG_KEXEC_CORE
431 lprop = of_get_flat_dt_prop(node, "linux,crashkernel-base", NULL);
432 if (lprop)
433 crashk_res.start = *lprop;
434
435 lprop = of_get_flat_dt_prop(node, "linux,crashkernel-size", NULL);
436 if (lprop)
437 crashk_res.end = crashk_res.start + *lprop - 1;
438 #endif
439
440 /* break now */
441 return 1;
442 }
443
444 /*
445 * Compare the range against max mem limit and update
446 * size if it cross the limit.
447 */
448
449 #ifdef CONFIG_SPARSEMEM
validate_mem_limit(u64 base,u64 * size)450 static bool validate_mem_limit(u64 base, u64 *size)
451 {
452 u64 max_mem = 1UL << (MAX_PHYSMEM_BITS);
453
454 if (base >= max_mem)
455 return false;
456 if ((base + *size) > max_mem)
457 *size = max_mem - base;
458 return true;
459 }
460 #else
validate_mem_limit(u64 base,u64 * size)461 static bool validate_mem_limit(u64 base, u64 *size)
462 {
463 return true;
464 }
465 #endif
466
467 #ifdef CONFIG_PPC_PSERIES
468 /*
469 * Interpret the ibm dynamic reconfiguration memory LMBs.
470 * This contains a list of memory blocks along with NUMA affinity
471 * information.
472 */
early_init_drmem_lmb(struct drmem_lmb * lmb,const __be32 ** usm,void * data)473 static int __init early_init_drmem_lmb(struct drmem_lmb *lmb,
474 const __be32 **usm,
475 void *data)
476 {
477 u64 base, size;
478 int is_kexec_kdump = 0, rngs;
479
480 base = lmb->base_addr;
481 size = drmem_lmb_size();
482 rngs = 1;
483
484 /*
485 * Skip this block if the reserved bit is set in flags
486 * or if the block is not assigned to this partition.
487 */
488 if ((lmb->flags & DRCONF_MEM_RESERVED) ||
489 !(lmb->flags & DRCONF_MEM_ASSIGNED))
490 return 0;
491
492 if (*usm)
493 is_kexec_kdump = 1;
494
495 if (is_kexec_kdump) {
496 /*
497 * For each memblock in ibm,dynamic-memory, a
498 * corresponding entry in linux,drconf-usable-memory
499 * property contains a counter 'p' followed by 'p'
500 * (base, size) duple. Now read the counter from
501 * linux,drconf-usable-memory property
502 */
503 rngs = dt_mem_next_cell(dt_root_size_cells, usm);
504 if (!rngs) /* there are no (base, size) duple */
505 return 0;
506 }
507
508 do {
509 if (is_kexec_kdump) {
510 base = dt_mem_next_cell(dt_root_addr_cells, usm);
511 size = dt_mem_next_cell(dt_root_size_cells, usm);
512 }
513
514 if (iommu_is_off) {
515 if (base >= 0x80000000ul)
516 continue;
517 if ((base + size) > 0x80000000ul)
518 size = 0x80000000ul - base;
519 }
520
521 if (!validate_mem_limit(base, &size))
522 continue;
523
524 DBG("Adding: %llx -> %llx\n", base, size);
525 memblock_add(base, size);
526
527 if (lmb->flags & DRCONF_MEM_HOTREMOVABLE)
528 memblock_mark_hotplug(base, size);
529 } while (--rngs);
530
531 return 0;
532 }
533 #endif /* CONFIG_PPC_PSERIES */
534
early_init_dt_scan_memory_ppc(unsigned long node,const char * uname,int depth,void * data)535 static int __init early_init_dt_scan_memory_ppc(unsigned long node,
536 const char *uname,
537 int depth, void *data)
538 {
539 #ifdef CONFIG_PPC_PSERIES
540 if (depth == 1 &&
541 strcmp(uname, "ibm,dynamic-reconfiguration-memory") == 0) {
542 walk_drmem_lmbs_early(node, NULL, early_init_drmem_lmb);
543 return 0;
544 }
545 #endif
546
547 return early_init_dt_scan_memory(node, uname, depth, data);
548 }
549
550 /*
551 * For a relocatable kernel, we need to get the memstart_addr first,
552 * then use it to calculate the virtual kernel start address. This has
553 * to happen at a very early stage (before machine_init). In this case,
554 * we just want to get the memstart_address and would not like to mess the
555 * memblock at this stage. So introduce a variable to skip the memblock_add()
556 * for this reason.
557 */
558 #ifdef CONFIG_RELOCATABLE
559 static int add_mem_to_memblock = 1;
560 #else
561 #define add_mem_to_memblock 1
562 #endif
563
early_init_dt_add_memory_arch(u64 base,u64 size)564 void __init early_init_dt_add_memory_arch(u64 base, u64 size)
565 {
566 #ifdef CONFIG_PPC64
567 if (iommu_is_off) {
568 if (base >= 0x80000000ul)
569 return;
570 if ((base + size) > 0x80000000ul)
571 size = 0x80000000ul - base;
572 }
573 #endif
574 /* Keep track of the beginning of memory -and- the size of
575 * the very first block in the device-tree as it represents
576 * the RMA on ppc64 server
577 */
578 if (base < memstart_addr) {
579 memstart_addr = base;
580 first_memblock_size = size;
581 }
582
583 /* Add the chunk to the MEMBLOCK list */
584 if (add_mem_to_memblock) {
585 if (validate_mem_limit(base, &size))
586 memblock_add(base, size);
587 }
588 }
589
early_reserve_mem_dt(void)590 static void __init early_reserve_mem_dt(void)
591 {
592 unsigned long i, dt_root;
593 int len;
594 const __be32 *prop;
595
596 early_init_fdt_reserve_self();
597 early_init_fdt_scan_reserved_mem();
598
599 dt_root = of_get_flat_dt_root();
600
601 prop = of_get_flat_dt_prop(dt_root, "reserved-ranges", &len);
602
603 if (!prop)
604 return;
605
606 DBG("Found new-style reserved-ranges\n");
607
608 /* Each reserved range is an (address,size) pair, 2 cells each,
609 * totalling 4 cells per range. */
610 for (i = 0; i < len / (sizeof(*prop) * 4); i++) {
611 u64 base, size;
612
613 base = of_read_number(prop + (i * 4) + 0, 2);
614 size = of_read_number(prop + (i * 4) + 2, 2);
615
616 if (size) {
617 DBG("reserving: %llx -> %llx\n", base, size);
618 memblock_reserve(base, size);
619 }
620 }
621 }
622
early_reserve_mem(void)623 static void __init early_reserve_mem(void)
624 {
625 __be64 *reserve_map;
626
627 reserve_map = (__be64 *)(((unsigned long)initial_boot_params) +
628 fdt_off_mem_rsvmap(initial_boot_params));
629
630 /* Look for the new "reserved-regions" property in the DT */
631 early_reserve_mem_dt();
632
633 #ifdef CONFIG_BLK_DEV_INITRD
634 /* Then reserve the initrd, if any */
635 if (initrd_start && (initrd_end > initrd_start)) {
636 memblock_reserve(ALIGN_DOWN(__pa(initrd_start), PAGE_SIZE),
637 ALIGN(initrd_end, PAGE_SIZE) -
638 ALIGN_DOWN(initrd_start, PAGE_SIZE));
639 }
640 #endif /* CONFIG_BLK_DEV_INITRD */
641
642 #ifdef CONFIG_PPC32
643 /*
644 * Handle the case where we might be booting from an old kexec
645 * image that setup the mem_rsvmap as pairs of 32-bit values
646 */
647 if (be64_to_cpup(reserve_map) > 0xffffffffull) {
648 u32 base_32, size_32;
649 __be32 *reserve_map_32 = (__be32 *)reserve_map;
650
651 DBG("Found old 32-bit reserve map\n");
652
653 while (1) {
654 base_32 = be32_to_cpup(reserve_map_32++);
655 size_32 = be32_to_cpup(reserve_map_32++);
656 if (size_32 == 0)
657 break;
658 DBG("reserving: %x -> %x\n", base_32, size_32);
659 memblock_reserve(base_32, size_32);
660 }
661 return;
662 }
663 #endif
664 }
665
666 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
667 static bool tm_disabled __initdata;
668
parse_ppc_tm(char * str)669 static int __init parse_ppc_tm(char *str)
670 {
671 bool res;
672
673 if (kstrtobool(str, &res))
674 return -EINVAL;
675
676 tm_disabled = !res;
677
678 return 0;
679 }
680 early_param("ppc_tm", parse_ppc_tm);
681
tm_init(void)682 static void __init tm_init(void)
683 {
684 if (tm_disabled) {
685 pr_info("Disabling hardware transactional memory (HTM)\n");
686 cur_cpu_spec->cpu_user_features2 &=
687 ~(PPC_FEATURE2_HTM_NOSC | PPC_FEATURE2_HTM);
688 cur_cpu_spec->cpu_features &= ~CPU_FTR_TM;
689 return;
690 }
691
692 pnv_tm_init();
693 }
694 #else
tm_init(void)695 static void tm_init(void) { }
696 #endif /* CONFIG_PPC_TRANSACTIONAL_MEM */
697
698 #ifdef CONFIG_PPC64
save_fscr_to_task(void)699 static void __init save_fscr_to_task(void)
700 {
701 /*
702 * Ensure the init_task (pid 0, aka swapper) uses the value of FSCR we
703 * have configured via the device tree features or via __init_FSCR().
704 * That value will then be propagated to pid 1 (init) and all future
705 * processes.
706 */
707 if (early_cpu_has_feature(CPU_FTR_ARCH_207S))
708 init_task.thread.fscr = mfspr(SPRN_FSCR);
709 }
710 #else
save_fscr_to_task(void)711 static inline void save_fscr_to_task(void) {};
712 #endif
713
714
early_init_devtree(void * params)715 void __init early_init_devtree(void *params)
716 {
717 phys_addr_t limit;
718
719 DBG(" -> early_init_devtree(%px)\n", params);
720
721 /* Too early to BUG_ON(), do it by hand */
722 if (!early_init_dt_verify(params))
723 panic("BUG: Failed verifying flat device tree, bad version?");
724
725 #ifdef CONFIG_PPC_RTAS
726 /* Some machines might need RTAS info for debugging, grab it now. */
727 of_scan_flat_dt(early_init_dt_scan_rtas, NULL);
728 #endif
729
730 #ifdef CONFIG_PPC_POWERNV
731 /* Some machines might need OPAL info for debugging, grab it now. */
732 of_scan_flat_dt(early_init_dt_scan_opal, NULL);
733
734 /* Scan tree for ultravisor feature */
735 of_scan_flat_dt(early_init_dt_scan_ultravisor, NULL);
736 #endif
737
738 #if defined(CONFIG_FA_DUMP) || defined(CONFIG_PRESERVE_FA_DUMP)
739 /* scan tree to see if dump is active during last boot */
740 of_scan_flat_dt(early_init_dt_scan_fw_dump, NULL);
741 #endif
742
743 /* Retrieve various informations from the /chosen node of the
744 * device-tree, including the platform type, initrd location and
745 * size, TCE reserve, and more ...
746 */
747 of_scan_flat_dt(early_init_dt_scan_chosen_ppc, boot_command_line);
748
749 /* Scan memory nodes and rebuild MEMBLOCKs */
750 of_scan_flat_dt(early_init_dt_scan_root, NULL);
751 of_scan_flat_dt(early_init_dt_scan_memory_ppc, NULL);
752
753 parse_early_param();
754
755 /* make sure we've parsed cmdline for mem= before this */
756 if (memory_limit)
757 first_memblock_size = min_t(u64, first_memblock_size, memory_limit);
758 setup_initial_memory_limit(memstart_addr, first_memblock_size);
759 /* Reserve MEMBLOCK regions used by kernel, initrd, dt, etc... */
760 memblock_reserve(PHYSICAL_START, __pa(klimit) - PHYSICAL_START);
761 /* If relocatable, reserve first 32k for interrupt vectors etc. */
762 if (PHYSICAL_START > MEMORY_START)
763 memblock_reserve(MEMORY_START, 0x8000);
764 reserve_kdump_trampoline();
765 #if defined(CONFIG_FA_DUMP) || defined(CONFIG_PRESERVE_FA_DUMP)
766 /*
767 * If we fail to reserve memory for firmware-assisted dump then
768 * fallback to kexec based kdump.
769 */
770 if (fadump_reserve_mem() == 0)
771 #endif
772 reserve_crashkernel();
773 early_reserve_mem();
774
775 /* Ensure that total memory size is page-aligned. */
776 limit = ALIGN(memory_limit ?: memblock_phys_mem_size(), PAGE_SIZE);
777 memblock_enforce_memory_limit(limit);
778
779 #if defined(CONFIG_PPC_BOOK3S_64) && defined(CONFIG_PPC_4K_PAGES)
780 if (!early_radix_enabled())
781 memblock_cap_memory_range(0, 1UL << (H_MAX_PHYSMEM_BITS));
782 #endif
783
784 memblock_allow_resize();
785 memblock_dump_all();
786
787 DBG("Phys. mem: %llx\n", (unsigned long long)memblock_phys_mem_size());
788
789 /* We may need to relocate the flat tree, do it now.
790 * FIXME .. and the initrd too? */
791 move_device_tree();
792
793 allocate_paca_ptrs();
794
795 DBG("Scanning CPUs ...\n");
796
797 dt_cpu_ftrs_scan();
798
799 /* Retrieve CPU related informations from the flat tree
800 * (altivec support, boot CPU ID, ...)
801 */
802 of_scan_flat_dt(early_init_dt_scan_cpus, NULL);
803 if (boot_cpuid < 0) {
804 printk("Failed to identify boot CPU !\n");
805 BUG();
806 }
807
808 save_fscr_to_task();
809
810 #if defined(CONFIG_SMP) && defined(CONFIG_PPC64)
811 /* We'll later wait for secondaries to check in; there are
812 * NCPUS-1 non-boot CPUs :-)
813 */
814 spinning_secondaries = boot_cpu_count - 1;
815 #endif
816
817 mmu_early_init_devtree();
818
819 #ifdef CONFIG_PPC_POWERNV
820 /* Scan and build the list of machine check recoverable ranges */
821 of_scan_flat_dt(early_init_dt_scan_recoverable_ranges, NULL);
822 #endif
823 epapr_paravirt_early_init();
824
825 /* Now try to figure out if we are running on LPAR and so on */
826 pseries_probe_fw_features();
827
828 /*
829 * Initialize pkey features and default AMR/IAMR values
830 */
831 pkey_early_init_devtree();
832
833 #ifdef CONFIG_PPC_PS3
834 /* Identify PS3 firmware */
835 if (of_flat_dt_is_compatible(of_get_flat_dt_root(), "sony,ps3"))
836 powerpc_firmware_features |= FW_FEATURE_PS3_POSSIBLE;
837 #endif
838
839 tm_init();
840
841 DBG(" <- early_init_devtree()\n");
842 }
843
844 #ifdef CONFIG_RELOCATABLE
845 /*
846 * This function run before early_init_devtree, so we have to init
847 * initial_boot_params.
848 */
early_get_first_memblock_info(void * params,phys_addr_t * size)849 void __init early_get_first_memblock_info(void *params, phys_addr_t *size)
850 {
851 /* Setup flat device-tree pointer */
852 initial_boot_params = params;
853
854 /*
855 * Scan the memory nodes and set add_mem_to_memblock to 0 to avoid
856 * mess the memblock.
857 */
858 add_mem_to_memblock = 0;
859 of_scan_flat_dt(early_init_dt_scan_root, NULL);
860 of_scan_flat_dt(early_init_dt_scan_memory_ppc, NULL);
861 add_mem_to_memblock = 1;
862
863 if (size)
864 *size = first_memblock_size;
865 }
866 #endif
867
868 /*******
869 *
870 * New implementation of the OF "find" APIs, return a refcounted
871 * object, call of_node_put() when done. The device tree and list
872 * are protected by a rw_lock.
873 *
874 * Note that property management will need some locking as well,
875 * this isn't dealt with yet.
876 *
877 *******/
878
879 /**
880 * of_get_ibm_chip_id - Returns the IBM "chip-id" of a device
881 * @np: device node of the device
882 *
883 * This looks for a property "ibm,chip-id" in the node or any
884 * of its parents and returns its content, or -1 if it cannot
885 * be found.
886 */
of_get_ibm_chip_id(struct device_node * np)887 int of_get_ibm_chip_id(struct device_node *np)
888 {
889 of_node_get(np);
890 while (np) {
891 u32 chip_id;
892
893 /*
894 * Skiboot may produce memory nodes that contain more than one
895 * cell in chip-id, we only read the first one here.
896 */
897 if (!of_property_read_u32(np, "ibm,chip-id", &chip_id)) {
898 of_node_put(np);
899 return chip_id;
900 }
901
902 np = of_get_next_parent(np);
903 }
904 return -1;
905 }
906 EXPORT_SYMBOL(of_get_ibm_chip_id);
907
908 /**
909 * cpu_to_chip_id - Return the cpus chip-id
910 * @cpu: The logical cpu number.
911 *
912 * Return the value of the ibm,chip-id property corresponding to the given
913 * logical cpu number. If the chip-id can not be found, returns -1.
914 */
cpu_to_chip_id(int cpu)915 int cpu_to_chip_id(int cpu)
916 {
917 struct device_node *np;
918
919 np = of_get_cpu_node(cpu, NULL);
920 if (!np)
921 return -1;
922
923 of_node_put(np);
924 return of_get_ibm_chip_id(np);
925 }
926 EXPORT_SYMBOL(cpu_to_chip_id);
927
arch_match_cpu_phys_id(int cpu,u64 phys_id)928 bool arch_match_cpu_phys_id(int cpu, u64 phys_id)
929 {
930 #ifdef CONFIG_SMP
931 /*
932 * Early firmware scanning must use this rather than
933 * get_hard_smp_processor_id because we don't have pacas allocated
934 * until memory topology is discovered.
935 */
936 if (cpu_to_phys_id != NULL)
937 return (int)phys_id == cpu_to_phys_id[cpu];
938 #endif
939
940 return (int)phys_id == get_hard_smp_processor_id(cpu);
941 }
942