xref: /openbmc/linux/arch/sparc/kernel/mdesc.c (revision 92b19ff5)
1 /* mdesc.c: Sun4V machine description handling.
2  *
3  * Copyright (C) 2007, 2008 David S. Miller <davem@davemloft.net>
4  */
5 #include <linux/kernel.h>
6 #include <linux/types.h>
7 #include <linux/memblock.h>
8 #include <linux/log2.h>
9 #include <linux/list.h>
10 #include <linux/slab.h>
11 #include <linux/mm.h>
12 #include <linux/miscdevice.h>
13 #include <linux/bootmem.h>
14 #include <linux/export.h>
15 
16 #include <asm/cpudata.h>
17 #include <asm/hypervisor.h>
18 #include <asm/mdesc.h>
19 #include <asm/prom.h>
20 #include <asm/uaccess.h>
21 #include <asm/oplib.h>
22 #include <asm/smp.h>
23 
24 /* Unlike the OBP device tree, the machine description is a full-on
25  * DAG.  An arbitrary number of ARCs are possible from one
26  * node to other nodes and thus we can't use the OBP device_node
27  * data structure to represent these nodes inside of the kernel.
28  *
29  * Actually, it isn't even a DAG, because there are back pointers
30  * which create cycles in the graph.
31  *
32  * mdesc_hdr and mdesc_elem describe the layout of the data structure
33  * we get from the Hypervisor.
34  */
35 struct mdesc_hdr {
36 	u32	version; /* Transport version */
37 	u32	node_sz; /* node block size */
38 	u32	name_sz; /* name block size */
39 	u32	data_sz; /* data block size */
40 } __attribute__((aligned(16)));
41 
42 struct mdesc_elem {
43 	u8	tag;
44 #define MD_LIST_END	0x00
45 #define MD_NODE		0x4e
46 #define MD_NODE_END	0x45
47 #define MD_NOOP		0x20
48 #define MD_PROP_ARC	0x61
49 #define MD_PROP_VAL	0x76
50 #define MD_PROP_STR	0x73
51 #define MD_PROP_DATA	0x64
52 	u8	name_len;
53 	u16	resv;
54 	u32	name_offset;
55 	union {
56 		struct {
57 			u32	data_len;
58 			u32	data_offset;
59 		} data;
60 		u64	val;
61 	} d;
62 };
63 
64 struct mdesc_mem_ops {
65 	struct mdesc_handle *(*alloc)(unsigned int mdesc_size);
66 	void (*free)(struct mdesc_handle *handle);
67 };
68 
69 struct mdesc_handle {
70 	struct list_head	list;
71 	struct mdesc_mem_ops	*mops;
72 	void			*self_base;
73 	atomic_t		refcnt;
74 	unsigned int		handle_size;
75 	struct mdesc_hdr	mdesc;
76 };
77 
78 static void mdesc_handle_init(struct mdesc_handle *hp,
79 			      unsigned int handle_size,
80 			      void *base)
81 {
82 	BUG_ON(((unsigned long)&hp->mdesc) & (16UL - 1));
83 
84 	memset(hp, 0, handle_size);
85 	INIT_LIST_HEAD(&hp->list);
86 	hp->self_base = base;
87 	atomic_set(&hp->refcnt, 1);
88 	hp->handle_size = handle_size;
89 }
90 
91 static struct mdesc_handle * __init mdesc_memblock_alloc(unsigned int mdesc_size)
92 {
93 	unsigned int handle_size, alloc_size;
94 	struct mdesc_handle *hp;
95 	unsigned long paddr;
96 
97 	handle_size = (sizeof(struct mdesc_handle) -
98 		       sizeof(struct mdesc_hdr) +
99 		       mdesc_size);
100 	alloc_size = PAGE_ALIGN(handle_size);
101 
102 	paddr = memblock_alloc(alloc_size, PAGE_SIZE);
103 
104 	hp = NULL;
105 	if (paddr) {
106 		hp = __va(paddr);
107 		mdesc_handle_init(hp, handle_size, hp);
108 	}
109 	return hp;
110 }
111 
112 static void __init mdesc_memblock_free(struct mdesc_handle *hp)
113 {
114 	unsigned int alloc_size;
115 	unsigned long start;
116 
117 	BUG_ON(atomic_read(&hp->refcnt) != 0);
118 	BUG_ON(!list_empty(&hp->list));
119 
120 	alloc_size = PAGE_ALIGN(hp->handle_size);
121 	start = __pa(hp);
122 	free_bootmem_late(start, alloc_size);
123 }
124 
125 static struct mdesc_mem_ops memblock_mdesc_ops = {
126 	.alloc = mdesc_memblock_alloc,
127 	.free  = mdesc_memblock_free,
128 };
129 
130 static struct mdesc_handle *mdesc_kmalloc(unsigned int mdesc_size)
131 {
132 	unsigned int handle_size;
133 	struct mdesc_handle *hp;
134 	unsigned long addr;
135 	void *base;
136 
137 	handle_size = (sizeof(struct mdesc_handle) -
138 		       sizeof(struct mdesc_hdr) +
139 		       mdesc_size);
140 
141 	/*
142 	 * Allocation has to succeed because mdesc update would be missed
143 	 * and such events are not retransmitted.
144 	 */
145 	base = kmalloc(handle_size + 15, GFP_KERNEL | __GFP_NOFAIL);
146 	addr = (unsigned long)base;
147 	addr = (addr + 15UL) & ~15UL;
148 	hp = (struct mdesc_handle *) addr;
149 
150 	mdesc_handle_init(hp, handle_size, base);
151 
152 	return hp;
153 }
154 
155 static void mdesc_kfree(struct mdesc_handle *hp)
156 {
157 	BUG_ON(atomic_read(&hp->refcnt) != 0);
158 	BUG_ON(!list_empty(&hp->list));
159 
160 	kfree(hp->self_base);
161 }
162 
163 static struct mdesc_mem_ops kmalloc_mdesc_memops = {
164 	.alloc = mdesc_kmalloc,
165 	.free  = mdesc_kfree,
166 };
167 
168 static struct mdesc_handle *mdesc_alloc(unsigned int mdesc_size,
169 					struct mdesc_mem_ops *mops)
170 {
171 	struct mdesc_handle *hp = mops->alloc(mdesc_size);
172 
173 	if (hp)
174 		hp->mops = mops;
175 
176 	return hp;
177 }
178 
179 static void mdesc_free(struct mdesc_handle *hp)
180 {
181 	hp->mops->free(hp);
182 }
183 
184 static struct mdesc_handle *cur_mdesc;
185 static LIST_HEAD(mdesc_zombie_list);
186 static DEFINE_SPINLOCK(mdesc_lock);
187 
188 struct mdesc_handle *mdesc_grab(void)
189 {
190 	struct mdesc_handle *hp;
191 	unsigned long flags;
192 
193 	spin_lock_irqsave(&mdesc_lock, flags);
194 	hp = cur_mdesc;
195 	if (hp)
196 		atomic_inc(&hp->refcnt);
197 	spin_unlock_irqrestore(&mdesc_lock, flags);
198 
199 	return hp;
200 }
201 EXPORT_SYMBOL(mdesc_grab);
202 
203 void mdesc_release(struct mdesc_handle *hp)
204 {
205 	unsigned long flags;
206 
207 	spin_lock_irqsave(&mdesc_lock, flags);
208 	if (atomic_dec_and_test(&hp->refcnt)) {
209 		list_del_init(&hp->list);
210 		hp->mops->free(hp);
211 	}
212 	spin_unlock_irqrestore(&mdesc_lock, flags);
213 }
214 EXPORT_SYMBOL(mdesc_release);
215 
216 static DEFINE_MUTEX(mdesc_mutex);
217 static struct mdesc_notifier_client *client_list;
218 
219 void mdesc_register_notifier(struct mdesc_notifier_client *client)
220 {
221 	u64 node;
222 
223 	mutex_lock(&mdesc_mutex);
224 	client->next = client_list;
225 	client_list = client;
226 
227 	mdesc_for_each_node_by_name(cur_mdesc, node, client->node_name)
228 		client->add(cur_mdesc, node);
229 
230 	mutex_unlock(&mdesc_mutex);
231 }
232 
233 static const u64 *parent_cfg_handle(struct mdesc_handle *hp, u64 node)
234 {
235 	const u64 *id;
236 	u64 a;
237 
238 	id = NULL;
239 	mdesc_for_each_arc(a, hp, node, MDESC_ARC_TYPE_BACK) {
240 		u64 target;
241 
242 		target = mdesc_arc_target(hp, a);
243 		id = mdesc_get_property(hp, target,
244 					"cfg-handle", NULL);
245 		if (id)
246 			break;
247 	}
248 
249 	return id;
250 }
251 
252 /* Run 'func' on nodes which are in A but not in B.  */
253 static void invoke_on_missing(const char *name,
254 			      struct mdesc_handle *a,
255 			      struct mdesc_handle *b,
256 			      void (*func)(struct mdesc_handle *, u64))
257 {
258 	u64 node;
259 
260 	mdesc_for_each_node_by_name(a, node, name) {
261 		int found = 0, is_vdc_port = 0;
262 		const char *name_prop;
263 		const u64 *id;
264 		u64 fnode;
265 
266 		name_prop = mdesc_get_property(a, node, "name", NULL);
267 		if (name_prop && !strcmp(name_prop, "vdc-port")) {
268 			is_vdc_port = 1;
269 			id = parent_cfg_handle(a, node);
270 		} else
271 			id = mdesc_get_property(a, node, "id", NULL);
272 
273 		if (!id) {
274 			printk(KERN_ERR "MD: Cannot find ID for %s node.\n",
275 			       (name_prop ? name_prop : name));
276 			continue;
277 		}
278 
279 		mdesc_for_each_node_by_name(b, fnode, name) {
280 			const u64 *fid;
281 
282 			if (is_vdc_port) {
283 				name_prop = mdesc_get_property(b, fnode,
284 							       "name", NULL);
285 				if (!name_prop ||
286 				    strcmp(name_prop, "vdc-port"))
287 					continue;
288 				fid = parent_cfg_handle(b, fnode);
289 				if (!fid) {
290 					printk(KERN_ERR "MD: Cannot find ID "
291 					       "for vdc-port node.\n");
292 					continue;
293 				}
294 			} else
295 				fid = mdesc_get_property(b, fnode,
296 							 "id", NULL);
297 
298 			if (*id == *fid) {
299 				found = 1;
300 				break;
301 			}
302 		}
303 		if (!found)
304 			func(a, node);
305 	}
306 }
307 
308 static void notify_one(struct mdesc_notifier_client *p,
309 		       struct mdesc_handle *old_hp,
310 		       struct mdesc_handle *new_hp)
311 {
312 	invoke_on_missing(p->node_name, old_hp, new_hp, p->remove);
313 	invoke_on_missing(p->node_name, new_hp, old_hp, p->add);
314 }
315 
316 static void mdesc_notify_clients(struct mdesc_handle *old_hp,
317 				 struct mdesc_handle *new_hp)
318 {
319 	struct mdesc_notifier_client *p = client_list;
320 
321 	while (p) {
322 		notify_one(p, old_hp, new_hp);
323 		p = p->next;
324 	}
325 }
326 
327 void mdesc_update(void)
328 {
329 	unsigned long len, real_len, status;
330 	struct mdesc_handle *hp, *orig_hp;
331 	unsigned long flags;
332 
333 	mutex_lock(&mdesc_mutex);
334 
335 	(void) sun4v_mach_desc(0UL, 0UL, &len);
336 
337 	hp = mdesc_alloc(len, &kmalloc_mdesc_memops);
338 	if (!hp) {
339 		printk(KERN_ERR "MD: mdesc alloc fails\n");
340 		goto out;
341 	}
342 
343 	status = sun4v_mach_desc(__pa(&hp->mdesc), len, &real_len);
344 	if (status != HV_EOK || real_len > len) {
345 		printk(KERN_ERR "MD: mdesc reread fails with %lu\n",
346 		       status);
347 		atomic_dec(&hp->refcnt);
348 		mdesc_free(hp);
349 		goto out;
350 	}
351 
352 	spin_lock_irqsave(&mdesc_lock, flags);
353 	orig_hp = cur_mdesc;
354 	cur_mdesc = hp;
355 	spin_unlock_irqrestore(&mdesc_lock, flags);
356 
357 	mdesc_notify_clients(orig_hp, hp);
358 
359 	spin_lock_irqsave(&mdesc_lock, flags);
360 	if (atomic_dec_and_test(&orig_hp->refcnt))
361 		mdesc_free(orig_hp);
362 	else
363 		list_add(&orig_hp->list, &mdesc_zombie_list);
364 	spin_unlock_irqrestore(&mdesc_lock, flags);
365 
366 out:
367 	mutex_unlock(&mdesc_mutex);
368 }
369 
370 static struct mdesc_elem *node_block(struct mdesc_hdr *mdesc)
371 {
372 	return (struct mdesc_elem *) (mdesc + 1);
373 }
374 
375 static void *name_block(struct mdesc_hdr *mdesc)
376 {
377 	return ((void *) node_block(mdesc)) + mdesc->node_sz;
378 }
379 
380 static void *data_block(struct mdesc_hdr *mdesc)
381 {
382 	return ((void *) name_block(mdesc)) + mdesc->name_sz;
383 }
384 
385 u64 mdesc_node_by_name(struct mdesc_handle *hp,
386 		       u64 from_node, const char *name)
387 {
388 	struct mdesc_elem *ep = node_block(&hp->mdesc);
389 	const char *names = name_block(&hp->mdesc);
390 	u64 last_node = hp->mdesc.node_sz / 16;
391 	u64 ret;
392 
393 	if (from_node == MDESC_NODE_NULL) {
394 		ret = from_node = 0;
395 	} else if (from_node >= last_node) {
396 		return MDESC_NODE_NULL;
397 	} else {
398 		ret = ep[from_node].d.val;
399 	}
400 
401 	while (ret < last_node) {
402 		if (ep[ret].tag != MD_NODE)
403 			return MDESC_NODE_NULL;
404 		if (!strcmp(names + ep[ret].name_offset, name))
405 			break;
406 		ret = ep[ret].d.val;
407 	}
408 	if (ret >= last_node)
409 		ret = MDESC_NODE_NULL;
410 	return ret;
411 }
412 EXPORT_SYMBOL(mdesc_node_by_name);
413 
414 const void *mdesc_get_property(struct mdesc_handle *hp, u64 node,
415 			       const char *name, int *lenp)
416 {
417 	const char *names = name_block(&hp->mdesc);
418 	u64 last_node = hp->mdesc.node_sz / 16;
419 	void *data = data_block(&hp->mdesc);
420 	struct mdesc_elem *ep;
421 
422 	if (node == MDESC_NODE_NULL || node >= last_node)
423 		return NULL;
424 
425 	ep = node_block(&hp->mdesc) + node;
426 	ep++;
427 	for (; ep->tag != MD_NODE_END; ep++) {
428 		void *val = NULL;
429 		int len = 0;
430 
431 		switch (ep->tag) {
432 		case MD_PROP_VAL:
433 			val = &ep->d.val;
434 			len = 8;
435 			break;
436 
437 		case MD_PROP_STR:
438 		case MD_PROP_DATA:
439 			val = data + ep->d.data.data_offset;
440 			len = ep->d.data.data_len;
441 			break;
442 
443 		default:
444 			break;
445 		}
446 		if (!val)
447 			continue;
448 
449 		if (!strcmp(names + ep->name_offset, name)) {
450 			if (lenp)
451 				*lenp = len;
452 			return val;
453 		}
454 	}
455 
456 	return NULL;
457 }
458 EXPORT_SYMBOL(mdesc_get_property);
459 
460 u64 mdesc_next_arc(struct mdesc_handle *hp, u64 from, const char *arc_type)
461 {
462 	struct mdesc_elem *ep, *base = node_block(&hp->mdesc);
463 	const char *names = name_block(&hp->mdesc);
464 	u64 last_node = hp->mdesc.node_sz / 16;
465 
466 	if (from == MDESC_NODE_NULL || from >= last_node)
467 		return MDESC_NODE_NULL;
468 
469 	ep = base + from;
470 
471 	ep++;
472 	for (; ep->tag != MD_NODE_END; ep++) {
473 		if (ep->tag != MD_PROP_ARC)
474 			continue;
475 
476 		if (strcmp(names + ep->name_offset, arc_type))
477 			continue;
478 
479 		return ep - base;
480 	}
481 
482 	return MDESC_NODE_NULL;
483 }
484 EXPORT_SYMBOL(mdesc_next_arc);
485 
486 u64 mdesc_arc_target(struct mdesc_handle *hp, u64 arc)
487 {
488 	struct mdesc_elem *ep, *base = node_block(&hp->mdesc);
489 
490 	ep = base + arc;
491 
492 	return ep->d.val;
493 }
494 EXPORT_SYMBOL(mdesc_arc_target);
495 
496 const char *mdesc_node_name(struct mdesc_handle *hp, u64 node)
497 {
498 	struct mdesc_elem *ep, *base = node_block(&hp->mdesc);
499 	const char *names = name_block(&hp->mdesc);
500 	u64 last_node = hp->mdesc.node_sz / 16;
501 
502 	if (node == MDESC_NODE_NULL || node >= last_node)
503 		return NULL;
504 
505 	ep = base + node;
506 	if (ep->tag != MD_NODE)
507 		return NULL;
508 
509 	return names + ep->name_offset;
510 }
511 EXPORT_SYMBOL(mdesc_node_name);
512 
513 static u64 max_cpus = 64;
514 
515 static void __init report_platform_properties(void)
516 {
517 	struct mdesc_handle *hp = mdesc_grab();
518 	u64 pn = mdesc_node_by_name(hp, MDESC_NODE_NULL, "platform");
519 	const char *s;
520 	const u64 *v;
521 
522 	if (pn == MDESC_NODE_NULL) {
523 		prom_printf("No platform node in machine-description.\n");
524 		prom_halt();
525 	}
526 
527 	s = mdesc_get_property(hp, pn, "banner-name", NULL);
528 	printk("PLATFORM: banner-name [%s]\n", s);
529 	s = mdesc_get_property(hp, pn, "name", NULL);
530 	printk("PLATFORM: name [%s]\n", s);
531 
532 	v = mdesc_get_property(hp, pn, "hostid", NULL);
533 	if (v)
534 		printk("PLATFORM: hostid [%08llx]\n", *v);
535 	v = mdesc_get_property(hp, pn, "serial#", NULL);
536 	if (v)
537 		printk("PLATFORM: serial# [%08llx]\n", *v);
538 	v = mdesc_get_property(hp, pn, "stick-frequency", NULL);
539 	printk("PLATFORM: stick-frequency [%08llx]\n", *v);
540 	v = mdesc_get_property(hp, pn, "mac-address", NULL);
541 	if (v)
542 		printk("PLATFORM: mac-address [%llx]\n", *v);
543 	v = mdesc_get_property(hp, pn, "watchdog-resolution", NULL);
544 	if (v)
545 		printk("PLATFORM: watchdog-resolution [%llu ms]\n", *v);
546 	v = mdesc_get_property(hp, pn, "watchdog-max-timeout", NULL);
547 	if (v)
548 		printk("PLATFORM: watchdog-max-timeout [%llu ms]\n", *v);
549 	v = mdesc_get_property(hp, pn, "max-cpus", NULL);
550 	if (v) {
551 		max_cpus = *v;
552 		printk("PLATFORM: max-cpus [%llu]\n", max_cpus);
553 	}
554 
555 #ifdef CONFIG_SMP
556 	{
557 		int max_cpu, i;
558 
559 		if (v) {
560 			max_cpu = *v;
561 			if (max_cpu > NR_CPUS)
562 				max_cpu = NR_CPUS;
563 		} else {
564 			max_cpu = NR_CPUS;
565 		}
566 		for (i = 0; i < max_cpu; i++)
567 			set_cpu_possible(i, true);
568 	}
569 #endif
570 
571 	mdesc_release(hp);
572 }
573 
574 static void fill_in_one_cache(cpuinfo_sparc *c, struct mdesc_handle *hp, u64 mp)
575 {
576 	const u64 *level = mdesc_get_property(hp, mp, "level", NULL);
577 	const u64 *size = mdesc_get_property(hp, mp, "size", NULL);
578 	const u64 *line_size = mdesc_get_property(hp, mp, "line-size", NULL);
579 	const char *type;
580 	int type_len;
581 
582 	type = mdesc_get_property(hp, mp, "type", &type_len);
583 
584 	switch (*level) {
585 	case 1:
586 		if (of_find_in_proplist(type, "instn", type_len)) {
587 			c->icache_size = *size;
588 			c->icache_line_size = *line_size;
589 		} else if (of_find_in_proplist(type, "data", type_len)) {
590 			c->dcache_size = *size;
591 			c->dcache_line_size = *line_size;
592 		}
593 		break;
594 
595 	case 2:
596 		c->ecache_size = *size;
597 		c->ecache_line_size = *line_size;
598 		break;
599 
600 	default:
601 		break;
602 	}
603 
604 	if (*level == 1) {
605 		u64 a;
606 
607 		mdesc_for_each_arc(a, hp, mp, MDESC_ARC_TYPE_FWD) {
608 			u64 target = mdesc_arc_target(hp, a);
609 			const char *name = mdesc_node_name(hp, target);
610 
611 			if (!strcmp(name, "cache"))
612 				fill_in_one_cache(c, hp, target);
613 		}
614 	}
615 }
616 
617 static void find_back_node_value(struct mdesc_handle *hp, u64 node,
618 				 char *srch_val,
619 				 void (*func)(struct mdesc_handle *, u64, int),
620 				 u64 val, int depth)
621 {
622 	u64 arc;
623 
624 	/* Since we have an estimate of recursion depth, do a sanity check. */
625 	if (depth == 0)
626 		return;
627 
628 	mdesc_for_each_arc(arc, hp, node, MDESC_ARC_TYPE_BACK) {
629 		u64 n = mdesc_arc_target(hp, arc);
630 		const char *name = mdesc_node_name(hp, n);
631 
632 		if (!strcmp(srch_val, name))
633 			(*func)(hp, n, val);
634 
635 		find_back_node_value(hp, n, srch_val, func, val, depth-1);
636 	}
637 }
638 
639 static void __mark_core_id(struct mdesc_handle *hp, u64 node,
640 			   int core_id)
641 {
642 	const u64 *id = mdesc_get_property(hp, node, "id", NULL);
643 
644 	if (*id < num_possible_cpus())
645 		cpu_data(*id).core_id = core_id;
646 }
647 
648 static void __mark_sock_id(struct mdesc_handle *hp, u64 node,
649 			   int sock_id)
650 {
651 	const u64 *id = mdesc_get_property(hp, node, "id", NULL);
652 
653 	if (*id < num_possible_cpus())
654 		cpu_data(*id).sock_id = sock_id;
655 }
656 
657 static void mark_core_ids(struct mdesc_handle *hp, u64 mp,
658 			  int core_id)
659 {
660 	find_back_node_value(hp, mp, "cpu", __mark_core_id, core_id, 10);
661 }
662 
663 static void mark_sock_ids(struct mdesc_handle *hp, u64 mp,
664 			  int sock_id)
665 {
666 	find_back_node_value(hp, mp, "cpu", __mark_sock_id, sock_id, 10);
667 }
668 
669 static void set_core_ids(struct mdesc_handle *hp)
670 {
671 	int idx;
672 	u64 mp;
673 
674 	idx = 1;
675 
676 	/* Identify unique cores by looking for cpus backpointed to by
677 	 * level 1 instruction caches.
678 	 */
679 	mdesc_for_each_node_by_name(hp, mp, "cache") {
680 		const u64 *level;
681 		const char *type;
682 		int len;
683 
684 		level = mdesc_get_property(hp, mp, "level", NULL);
685 		if (*level != 1)
686 			continue;
687 
688 		type = mdesc_get_property(hp, mp, "type", &len);
689 		if (!of_find_in_proplist(type, "instn", len))
690 			continue;
691 
692 		mark_core_ids(hp, mp, idx);
693 		idx++;
694 	}
695 }
696 
697 static int set_sock_ids_by_cache(struct mdesc_handle *hp, int level)
698 {
699 	u64 mp;
700 	int idx = 1;
701 	int fnd = 0;
702 
703 	/* Identify unique sockets by looking for cpus backpointed to by
704 	 * shared level n caches.
705 	 */
706 	mdesc_for_each_node_by_name(hp, mp, "cache") {
707 		const u64 *cur_lvl;
708 
709 		cur_lvl = mdesc_get_property(hp, mp, "level", NULL);
710 		if (*cur_lvl != level)
711 			continue;
712 
713 		mark_sock_ids(hp, mp, idx);
714 		idx++;
715 		fnd = 1;
716 	}
717 	return fnd;
718 }
719 
720 static void set_sock_ids_by_socket(struct mdesc_handle *hp, u64 mp)
721 {
722 	int idx = 1;
723 
724 	mdesc_for_each_node_by_name(hp, mp, "socket") {
725 		u64 a;
726 
727 		mdesc_for_each_arc(a, hp, mp, MDESC_ARC_TYPE_FWD) {
728 			u64 t = mdesc_arc_target(hp, a);
729 			const char *name;
730 			const u64 *id;
731 
732 			name = mdesc_node_name(hp, t);
733 			if (strcmp(name, "cpu"))
734 				continue;
735 
736 			id = mdesc_get_property(hp, t, "id", NULL);
737 			if (*id < num_possible_cpus())
738 				cpu_data(*id).sock_id = idx;
739 		}
740 		idx++;
741 	}
742 }
743 
744 static void set_sock_ids(struct mdesc_handle *hp)
745 {
746 	u64 mp;
747 
748 	/* If machine description exposes sockets data use it.
749 	 * Otherwise fallback to use shared L3 or L2 caches.
750 	 */
751 	mp = mdesc_node_by_name(hp, MDESC_NODE_NULL, "sockets");
752 	if (mp != MDESC_NODE_NULL)
753 		return set_sock_ids_by_socket(hp, mp);
754 
755 	if (!set_sock_ids_by_cache(hp, 3))
756 		set_sock_ids_by_cache(hp, 2);
757 }
758 
759 static void mark_proc_ids(struct mdesc_handle *hp, u64 mp, int proc_id)
760 {
761 	u64 a;
762 
763 	mdesc_for_each_arc(a, hp, mp, MDESC_ARC_TYPE_BACK) {
764 		u64 t = mdesc_arc_target(hp, a);
765 		const char *name;
766 		const u64 *id;
767 
768 		name = mdesc_node_name(hp, t);
769 		if (strcmp(name, "cpu"))
770 			continue;
771 
772 		id = mdesc_get_property(hp, t, "id", NULL);
773 		if (*id < NR_CPUS)
774 			cpu_data(*id).proc_id = proc_id;
775 	}
776 }
777 
778 static void __set_proc_ids(struct mdesc_handle *hp, const char *exec_unit_name)
779 {
780 	int idx;
781 	u64 mp;
782 
783 	idx = 0;
784 	mdesc_for_each_node_by_name(hp, mp, exec_unit_name) {
785 		const char *type;
786 		int len;
787 
788 		type = mdesc_get_property(hp, mp, "type", &len);
789 		if (!of_find_in_proplist(type, "int", len) &&
790 		    !of_find_in_proplist(type, "integer", len))
791 			continue;
792 
793 		mark_proc_ids(hp, mp, idx);
794 		idx++;
795 	}
796 }
797 
798 static void set_proc_ids(struct mdesc_handle *hp)
799 {
800 	__set_proc_ids(hp, "exec_unit");
801 	__set_proc_ids(hp, "exec-unit");
802 }
803 
804 static void get_one_mondo_bits(const u64 *p, unsigned int *mask,
805 			       unsigned long def, unsigned long max)
806 {
807 	u64 val;
808 
809 	if (!p)
810 		goto use_default;
811 	val = *p;
812 
813 	if (!val || val >= 64)
814 		goto use_default;
815 
816 	if (val > max)
817 		val = max;
818 
819 	*mask = ((1U << val) * 64U) - 1U;
820 	return;
821 
822 use_default:
823 	*mask = ((1U << def) * 64U) - 1U;
824 }
825 
826 static void get_mondo_data(struct mdesc_handle *hp, u64 mp,
827 			   struct trap_per_cpu *tb)
828 {
829 	static int printed;
830 	const u64 *val;
831 
832 	val = mdesc_get_property(hp, mp, "q-cpu-mondo-#bits", NULL);
833 	get_one_mondo_bits(val, &tb->cpu_mondo_qmask, 7, ilog2(max_cpus * 2));
834 
835 	val = mdesc_get_property(hp, mp, "q-dev-mondo-#bits", NULL);
836 	get_one_mondo_bits(val, &tb->dev_mondo_qmask, 7, 8);
837 
838 	val = mdesc_get_property(hp, mp, "q-resumable-#bits", NULL);
839 	get_one_mondo_bits(val, &tb->resum_qmask, 6, 7);
840 
841 	val = mdesc_get_property(hp, mp, "q-nonresumable-#bits", NULL);
842 	get_one_mondo_bits(val, &tb->nonresum_qmask, 2, 2);
843 	if (!printed++) {
844 		pr_info("SUN4V: Mondo queue sizes "
845 			"[cpu(%u) dev(%u) r(%u) nr(%u)]\n",
846 			tb->cpu_mondo_qmask + 1,
847 			tb->dev_mondo_qmask + 1,
848 			tb->resum_qmask + 1,
849 			tb->nonresum_qmask + 1);
850 	}
851 }
852 
853 static void *mdesc_iterate_over_cpus(void *(*func)(struct mdesc_handle *, u64, int, void *), void *arg, cpumask_t *mask)
854 {
855 	struct mdesc_handle *hp = mdesc_grab();
856 	void *ret = NULL;
857 	u64 mp;
858 
859 	mdesc_for_each_node_by_name(hp, mp, "cpu") {
860 		const u64 *id = mdesc_get_property(hp, mp, "id", NULL);
861 		int cpuid = *id;
862 
863 #ifdef CONFIG_SMP
864 		if (cpuid >= NR_CPUS) {
865 			printk(KERN_WARNING "Ignoring CPU %d which is "
866 			       ">= NR_CPUS (%d)\n",
867 			       cpuid, NR_CPUS);
868 			continue;
869 		}
870 		if (!cpumask_test_cpu(cpuid, mask))
871 			continue;
872 #endif
873 
874 		ret = func(hp, mp, cpuid, arg);
875 		if (ret)
876 			goto out;
877 	}
878 out:
879 	mdesc_release(hp);
880 	return ret;
881 }
882 
883 static void *record_one_cpu(struct mdesc_handle *hp, u64 mp, int cpuid,
884 			    void *arg)
885 {
886 	ncpus_probed++;
887 #ifdef CONFIG_SMP
888 	set_cpu_present(cpuid, true);
889 #endif
890 	return NULL;
891 }
892 
893 void mdesc_populate_present_mask(cpumask_t *mask)
894 {
895 	if (tlb_type != hypervisor)
896 		return;
897 
898 	ncpus_probed = 0;
899 	mdesc_iterate_over_cpus(record_one_cpu, NULL, mask);
900 }
901 
902 static void * __init check_one_pgsz(struct mdesc_handle *hp, u64 mp, int cpuid, void *arg)
903 {
904 	const u64 *pgsz_prop = mdesc_get_property(hp, mp, "mmu-page-size-list", NULL);
905 	unsigned long *pgsz_mask = arg;
906 	u64 val;
907 
908 	val = (HV_PGSZ_MASK_8K | HV_PGSZ_MASK_64K |
909 	       HV_PGSZ_MASK_512K | HV_PGSZ_MASK_4MB);
910 	if (pgsz_prop)
911 		val = *pgsz_prop;
912 
913 	if (!*pgsz_mask)
914 		*pgsz_mask = val;
915 	else
916 		*pgsz_mask &= val;
917 	return NULL;
918 }
919 
920 void __init mdesc_get_page_sizes(cpumask_t *mask, unsigned long *pgsz_mask)
921 {
922 	*pgsz_mask = 0;
923 	mdesc_iterate_over_cpus(check_one_pgsz, pgsz_mask, mask);
924 }
925 
926 static void *fill_in_one_cpu(struct mdesc_handle *hp, u64 mp, int cpuid,
927 			     void *arg)
928 {
929 	const u64 *cfreq = mdesc_get_property(hp, mp, "clock-frequency", NULL);
930 	struct trap_per_cpu *tb;
931 	cpuinfo_sparc *c;
932 	u64 a;
933 
934 #ifndef CONFIG_SMP
935 	/* On uniprocessor we only want the values for the
936 	 * real physical cpu the kernel booted onto, however
937 	 * cpu_data() only has one entry at index 0.
938 	 */
939 	if (cpuid != real_hard_smp_processor_id())
940 		return NULL;
941 	cpuid = 0;
942 #endif
943 
944 	c = &cpu_data(cpuid);
945 	c->clock_tick = *cfreq;
946 
947 	tb = &trap_block[cpuid];
948 	get_mondo_data(hp, mp, tb);
949 
950 	mdesc_for_each_arc(a, hp, mp, MDESC_ARC_TYPE_FWD) {
951 		u64 j, t = mdesc_arc_target(hp, a);
952 		const char *t_name;
953 
954 		t_name = mdesc_node_name(hp, t);
955 		if (!strcmp(t_name, "cache")) {
956 			fill_in_one_cache(c, hp, t);
957 			continue;
958 		}
959 
960 		mdesc_for_each_arc(j, hp, t, MDESC_ARC_TYPE_FWD) {
961 			u64 n = mdesc_arc_target(hp, j);
962 			const char *n_name;
963 
964 			n_name = mdesc_node_name(hp, n);
965 			if (!strcmp(n_name, "cache"))
966 				fill_in_one_cache(c, hp, n);
967 		}
968 	}
969 
970 	c->core_id = 0;
971 	c->proc_id = -1;
972 
973 	return NULL;
974 }
975 
976 void mdesc_fill_in_cpu_data(cpumask_t *mask)
977 {
978 	struct mdesc_handle *hp;
979 
980 	mdesc_iterate_over_cpus(fill_in_one_cpu, NULL, mask);
981 
982 	hp = mdesc_grab();
983 
984 	set_core_ids(hp);
985 	set_proc_ids(hp);
986 	set_sock_ids(hp);
987 
988 	mdesc_release(hp);
989 
990 	smp_fill_in_sib_core_maps();
991 }
992 
993 /* mdesc_open() - Grab a reference to mdesc_handle when /dev/mdesc is
994  * opened. Hold this reference until /dev/mdesc is closed to ensure
995  * mdesc data structure is not released underneath us. Store the
996  * pointer to mdesc structure in private_data for read and seek to use
997  */
998 static int mdesc_open(struct inode *inode, struct file *file)
999 {
1000 	struct mdesc_handle *hp = mdesc_grab();
1001 
1002 	if (!hp)
1003 		return -ENODEV;
1004 
1005 	file->private_data = hp;
1006 
1007 	return 0;
1008 }
1009 
1010 static ssize_t mdesc_read(struct file *file, char __user *buf,
1011 			  size_t len, loff_t *offp)
1012 {
1013 	struct mdesc_handle *hp = file->private_data;
1014 	unsigned char *mdesc;
1015 	int bytes_left, count = len;
1016 
1017 	if (*offp >= hp->handle_size)
1018 		return 0;
1019 
1020 	bytes_left = hp->handle_size - *offp;
1021 	if (count > bytes_left)
1022 		count = bytes_left;
1023 
1024 	mdesc = (unsigned char *)&hp->mdesc;
1025 	mdesc += *offp;
1026 	if (!copy_to_user(buf, mdesc, count)) {
1027 		*offp += count;
1028 		return count;
1029 	} else {
1030 		return -EFAULT;
1031 	}
1032 }
1033 
1034 static loff_t mdesc_llseek(struct file *file, loff_t offset, int whence)
1035 {
1036 	struct mdesc_handle *hp;
1037 
1038 	switch (whence) {
1039 	case SEEK_CUR:
1040 		offset += file->f_pos;
1041 		break;
1042 	case SEEK_SET:
1043 		break;
1044 	default:
1045 		return -EINVAL;
1046 	}
1047 
1048 	hp = file->private_data;
1049 	if (offset > hp->handle_size)
1050 		return -EINVAL;
1051 	else
1052 		file->f_pos = offset;
1053 
1054 	return offset;
1055 }
1056 
1057 /* mdesc_close() - /dev/mdesc is being closed, release the reference to
1058  * mdesc structure.
1059  */
1060 static int mdesc_close(struct inode *inode, struct file *file)
1061 {
1062 	mdesc_release(file->private_data);
1063 	return 0;
1064 }
1065 
1066 static const struct file_operations mdesc_fops = {
1067 	.open    = mdesc_open,
1068 	.read	 = mdesc_read,
1069 	.llseek  = mdesc_llseek,
1070 	.release = mdesc_close,
1071 	.owner	 = THIS_MODULE,
1072 };
1073 
1074 static struct miscdevice mdesc_misc = {
1075 	.minor	= MISC_DYNAMIC_MINOR,
1076 	.name	= "mdesc",
1077 	.fops	= &mdesc_fops,
1078 };
1079 
1080 static int __init mdesc_misc_init(void)
1081 {
1082 	return misc_register(&mdesc_misc);
1083 }
1084 
1085 __initcall(mdesc_misc_init);
1086 
1087 void __init sun4v_mdesc_init(void)
1088 {
1089 	struct mdesc_handle *hp;
1090 	unsigned long len, real_len, status;
1091 
1092 	(void) sun4v_mach_desc(0UL, 0UL, &len);
1093 
1094 	printk("MDESC: Size is %lu bytes.\n", len);
1095 
1096 	hp = mdesc_alloc(len, &memblock_mdesc_ops);
1097 	if (hp == NULL) {
1098 		prom_printf("MDESC: alloc of %lu bytes failed.\n", len);
1099 		prom_halt();
1100 	}
1101 
1102 	status = sun4v_mach_desc(__pa(&hp->mdesc), len, &real_len);
1103 	if (status != HV_EOK || real_len > len) {
1104 		prom_printf("sun4v_mach_desc fails, err(%lu), "
1105 			    "len(%lu), real_len(%lu)\n",
1106 			    status, len, real_len);
1107 		mdesc_free(hp);
1108 		prom_halt();
1109 	}
1110 
1111 	cur_mdesc = hp;
1112 
1113 	report_platform_properties();
1114 }
1115