xref: /openbmc/linux/arch/sparc/kernel/mdesc.c (revision 711aab1d)
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 <linux/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 typedef int (*mdesc_node_info_get_f)(struct mdesc_handle *, u64,
79 				     union md_node_info *);
80 typedef void (*mdesc_node_info_rel_f)(union md_node_info *);
81 typedef bool (*mdesc_node_match_f)(union md_node_info *, union md_node_info *);
82 
83 struct md_node_ops {
84 	char			*name;
85 	mdesc_node_info_get_f	get_info;
86 	mdesc_node_info_rel_f	rel_info;
87 	mdesc_node_match_f	node_match;
88 };
89 
90 static int get_vdev_port_node_info(struct mdesc_handle *md, u64 node,
91 				   union md_node_info *node_info);
92 static void rel_vdev_port_node_info(union md_node_info *node_info);
93 static bool vdev_port_node_match(union md_node_info *a_node_info,
94 				 union md_node_info *b_node_info);
95 
96 static int get_ds_port_node_info(struct mdesc_handle *md, u64 node,
97 				 union md_node_info *node_info);
98 static void rel_ds_port_node_info(union md_node_info *node_info);
99 static bool ds_port_node_match(union md_node_info *a_node_info,
100 			       union md_node_info *b_node_info);
101 
102 /* supported node types which can be registered */
103 static struct md_node_ops md_node_ops_table[] = {
104 	{"virtual-device-port", get_vdev_port_node_info,
105 	 rel_vdev_port_node_info, vdev_port_node_match},
106 	{"domain-services-port", get_ds_port_node_info,
107 	 rel_ds_port_node_info, ds_port_node_match},
108 	{NULL, NULL, NULL, NULL}
109 };
110 
111 static void mdesc_get_node_ops(const char *node_name,
112 			       mdesc_node_info_get_f *get_info_f,
113 			       mdesc_node_info_rel_f *rel_info_f,
114 			       mdesc_node_match_f *match_f)
115 {
116 	int i;
117 
118 	if (get_info_f)
119 		*get_info_f = NULL;
120 
121 	if (rel_info_f)
122 		*rel_info_f = NULL;
123 
124 	if (match_f)
125 		*match_f = NULL;
126 
127 	if (!node_name)
128 		return;
129 
130 	for (i = 0; md_node_ops_table[i].name != NULL; i++) {
131 		if (strcmp(md_node_ops_table[i].name, node_name) == 0) {
132 			if (get_info_f)
133 				*get_info_f = md_node_ops_table[i].get_info;
134 
135 			if (rel_info_f)
136 				*rel_info_f = md_node_ops_table[i].rel_info;
137 
138 			if (match_f)
139 				*match_f = md_node_ops_table[i].node_match;
140 
141 			break;
142 		}
143 	}
144 }
145 
146 static void mdesc_handle_init(struct mdesc_handle *hp,
147 			      unsigned int handle_size,
148 			      void *base)
149 {
150 	BUG_ON(((unsigned long)&hp->mdesc) & (16UL - 1));
151 
152 	memset(hp, 0, handle_size);
153 	INIT_LIST_HEAD(&hp->list);
154 	hp->self_base = base;
155 	atomic_set(&hp->refcnt, 1);
156 	hp->handle_size = handle_size;
157 }
158 
159 static struct mdesc_handle * __init mdesc_memblock_alloc(unsigned int mdesc_size)
160 {
161 	unsigned int handle_size, alloc_size;
162 	struct mdesc_handle *hp;
163 	unsigned long paddr;
164 
165 	handle_size = (sizeof(struct mdesc_handle) -
166 		       sizeof(struct mdesc_hdr) +
167 		       mdesc_size);
168 	alloc_size = PAGE_ALIGN(handle_size);
169 
170 	paddr = memblock_alloc(alloc_size, PAGE_SIZE);
171 
172 	hp = NULL;
173 	if (paddr) {
174 		hp = __va(paddr);
175 		mdesc_handle_init(hp, handle_size, hp);
176 	}
177 	return hp;
178 }
179 
180 static void __init mdesc_memblock_free(struct mdesc_handle *hp)
181 {
182 	unsigned int alloc_size;
183 	unsigned long start;
184 
185 	BUG_ON(atomic_read(&hp->refcnt) != 0);
186 	BUG_ON(!list_empty(&hp->list));
187 
188 	alloc_size = PAGE_ALIGN(hp->handle_size);
189 	start = __pa(hp);
190 	free_bootmem_late(start, alloc_size);
191 }
192 
193 static struct mdesc_mem_ops memblock_mdesc_ops = {
194 	.alloc = mdesc_memblock_alloc,
195 	.free  = mdesc_memblock_free,
196 };
197 
198 static struct mdesc_handle *mdesc_kmalloc(unsigned int mdesc_size)
199 {
200 	unsigned int handle_size;
201 	struct mdesc_handle *hp;
202 	unsigned long addr;
203 	void *base;
204 
205 	handle_size = (sizeof(struct mdesc_handle) -
206 		       sizeof(struct mdesc_hdr) +
207 		       mdesc_size);
208 	base = kmalloc(handle_size + 15, GFP_KERNEL | __GFP_RETRY_MAYFAIL);
209 	if (!base)
210 		return NULL;
211 
212 	addr = (unsigned long)base;
213 	addr = (addr + 15UL) & ~15UL;
214 	hp = (struct mdesc_handle *) addr;
215 
216 	mdesc_handle_init(hp, handle_size, base);
217 
218 	return hp;
219 }
220 
221 static void mdesc_kfree(struct mdesc_handle *hp)
222 {
223 	BUG_ON(atomic_read(&hp->refcnt) != 0);
224 	BUG_ON(!list_empty(&hp->list));
225 
226 	kfree(hp->self_base);
227 }
228 
229 static struct mdesc_mem_ops kmalloc_mdesc_memops = {
230 	.alloc = mdesc_kmalloc,
231 	.free  = mdesc_kfree,
232 };
233 
234 static struct mdesc_handle *mdesc_alloc(unsigned int mdesc_size,
235 					struct mdesc_mem_ops *mops)
236 {
237 	struct mdesc_handle *hp = mops->alloc(mdesc_size);
238 
239 	if (hp)
240 		hp->mops = mops;
241 
242 	return hp;
243 }
244 
245 static void mdesc_free(struct mdesc_handle *hp)
246 {
247 	hp->mops->free(hp);
248 }
249 
250 static struct mdesc_handle *cur_mdesc;
251 static LIST_HEAD(mdesc_zombie_list);
252 static DEFINE_SPINLOCK(mdesc_lock);
253 
254 struct mdesc_handle *mdesc_grab(void)
255 {
256 	struct mdesc_handle *hp;
257 	unsigned long flags;
258 
259 	spin_lock_irqsave(&mdesc_lock, flags);
260 	hp = cur_mdesc;
261 	if (hp)
262 		atomic_inc(&hp->refcnt);
263 	spin_unlock_irqrestore(&mdesc_lock, flags);
264 
265 	return hp;
266 }
267 EXPORT_SYMBOL(mdesc_grab);
268 
269 void mdesc_release(struct mdesc_handle *hp)
270 {
271 	unsigned long flags;
272 
273 	spin_lock_irqsave(&mdesc_lock, flags);
274 	if (atomic_dec_and_test(&hp->refcnt)) {
275 		list_del_init(&hp->list);
276 		hp->mops->free(hp);
277 	}
278 	spin_unlock_irqrestore(&mdesc_lock, flags);
279 }
280 EXPORT_SYMBOL(mdesc_release);
281 
282 static DEFINE_MUTEX(mdesc_mutex);
283 static struct mdesc_notifier_client *client_list;
284 
285 void mdesc_register_notifier(struct mdesc_notifier_client *client)
286 {
287 	bool supported = false;
288 	u64 node;
289 	int i;
290 
291 	mutex_lock(&mdesc_mutex);
292 
293 	/* check to see if the node is supported for registration */
294 	for (i = 0; md_node_ops_table[i].name != NULL; i++) {
295 		if (strcmp(md_node_ops_table[i].name, client->node_name) == 0) {
296 			supported = true;
297 			break;
298 		}
299 	}
300 
301 	if (!supported) {
302 		pr_err("MD: %s node not supported\n", client->node_name);
303 		mutex_unlock(&mdesc_mutex);
304 		return;
305 	}
306 
307 	client->next = client_list;
308 	client_list = client;
309 
310 	mdesc_for_each_node_by_name(cur_mdesc, node, client->node_name)
311 		client->add(cur_mdesc, node, client->node_name);
312 
313 	mutex_unlock(&mdesc_mutex);
314 }
315 
316 static const u64 *parent_cfg_handle(struct mdesc_handle *hp, u64 node)
317 {
318 	const u64 *id;
319 	u64 a;
320 
321 	id = NULL;
322 	mdesc_for_each_arc(a, hp, node, MDESC_ARC_TYPE_BACK) {
323 		u64 target;
324 
325 		target = mdesc_arc_target(hp, a);
326 		id = mdesc_get_property(hp, target,
327 					"cfg-handle", NULL);
328 		if (id)
329 			break;
330 	}
331 
332 	return id;
333 }
334 
335 static int get_vdev_port_node_info(struct mdesc_handle *md, u64 node,
336 				   union md_node_info *node_info)
337 {
338 	const u64 *parent_cfg_hdlp;
339 	const char *name;
340 	const u64 *idp;
341 
342 	/*
343 	 * Virtual device nodes are distinguished by:
344 	 * 1. "id" property
345 	 * 2. "name" property
346 	 * 3. parent node "cfg-handle" property
347 	 */
348 	idp = mdesc_get_property(md, node, "id", NULL);
349 	name = mdesc_get_property(md, node, "name", NULL);
350 	parent_cfg_hdlp = parent_cfg_handle(md, node);
351 
352 	if (!idp || !name || !parent_cfg_hdlp)
353 		return -1;
354 
355 	node_info->vdev_port.id = *idp;
356 	node_info->vdev_port.name = kstrdup_const(name, GFP_KERNEL);
357 	node_info->vdev_port.parent_cfg_hdl = *parent_cfg_hdlp;
358 
359 	return 0;
360 }
361 
362 static void rel_vdev_port_node_info(union md_node_info *node_info)
363 {
364 	if (node_info && node_info->vdev_port.name) {
365 		kfree_const(node_info->vdev_port.name);
366 		node_info->vdev_port.name = NULL;
367 	}
368 }
369 
370 static bool vdev_port_node_match(union md_node_info *a_node_info,
371 				 union md_node_info *b_node_info)
372 {
373 	if (a_node_info->vdev_port.id != b_node_info->vdev_port.id)
374 		return false;
375 
376 	if (a_node_info->vdev_port.parent_cfg_hdl !=
377 	    b_node_info->vdev_port.parent_cfg_hdl)
378 		return false;
379 
380 	if (strncmp(a_node_info->vdev_port.name,
381 		    b_node_info->vdev_port.name, MDESC_MAX_STR_LEN) != 0)
382 		return false;
383 
384 	return true;
385 }
386 
387 static int get_ds_port_node_info(struct mdesc_handle *md, u64 node,
388 				 union md_node_info *node_info)
389 {
390 	const u64 *idp;
391 
392 	/* DS port nodes use the "id" property to distinguish them */
393 	idp = mdesc_get_property(md, node, "id", NULL);
394 	if (!idp)
395 		return -1;
396 
397 	node_info->ds_port.id = *idp;
398 
399 	return 0;
400 }
401 
402 static void rel_ds_port_node_info(union md_node_info *node_info)
403 {
404 }
405 
406 static bool ds_port_node_match(union md_node_info *a_node_info,
407 			       union md_node_info *b_node_info)
408 {
409 	if (a_node_info->ds_port.id != b_node_info->ds_port.id)
410 		return false;
411 
412 	return true;
413 }
414 
415 /* Run 'func' on nodes which are in A but not in B.  */
416 static void invoke_on_missing(const char *name,
417 			      struct mdesc_handle *a,
418 			      struct mdesc_handle *b,
419 			      void (*func)(struct mdesc_handle *, u64,
420 					   const char *node_name))
421 {
422 	mdesc_node_info_get_f get_info_func;
423 	mdesc_node_info_rel_f rel_info_func;
424 	mdesc_node_match_f node_match_func;
425 	union md_node_info a_node_info;
426 	union md_node_info b_node_info;
427 	bool found;
428 	u64 a_node;
429 	u64 b_node;
430 	int rv;
431 
432 	/*
433 	 * Find the get_info, rel_info and node_match ops for the given
434 	 * node name
435 	 */
436 	mdesc_get_node_ops(name, &get_info_func, &rel_info_func,
437 			   &node_match_func);
438 
439 	/* If we didn't find a match, the node type is not supported */
440 	if (!get_info_func || !rel_info_func || !node_match_func) {
441 		pr_err("MD: %s node type is not supported\n", name);
442 		return;
443 	}
444 
445 	mdesc_for_each_node_by_name(a, a_node, name) {
446 		found = false;
447 
448 		rv = get_info_func(a, a_node, &a_node_info);
449 		if (rv != 0) {
450 			pr_err("MD: Cannot find 1 or more required match properties for %s node.\n",
451 			       name);
452 			continue;
453 		}
454 
455 		/* Check each node in B for node matching a_node */
456 		mdesc_for_each_node_by_name(b, b_node, name) {
457 			rv = get_info_func(b, b_node, &b_node_info);
458 			if (rv != 0)
459 				continue;
460 
461 			if (node_match_func(&a_node_info, &b_node_info)) {
462 				found = true;
463 				rel_info_func(&b_node_info);
464 				break;
465 			}
466 
467 			rel_info_func(&b_node_info);
468 		}
469 
470 		rel_info_func(&a_node_info);
471 
472 		if (!found)
473 			func(a, a_node, name);
474 	}
475 }
476 
477 static void notify_one(struct mdesc_notifier_client *p,
478 		       struct mdesc_handle *old_hp,
479 		       struct mdesc_handle *new_hp)
480 {
481 	invoke_on_missing(p->node_name, old_hp, new_hp, p->remove);
482 	invoke_on_missing(p->node_name, new_hp, old_hp, p->add);
483 }
484 
485 static void mdesc_notify_clients(struct mdesc_handle *old_hp,
486 				 struct mdesc_handle *new_hp)
487 {
488 	struct mdesc_notifier_client *p = client_list;
489 
490 	while (p) {
491 		notify_one(p, old_hp, new_hp);
492 		p = p->next;
493 	}
494 }
495 
496 void mdesc_update(void)
497 {
498 	unsigned long len, real_len, status;
499 	struct mdesc_handle *hp, *orig_hp;
500 	unsigned long flags;
501 
502 	mutex_lock(&mdesc_mutex);
503 
504 	(void) sun4v_mach_desc(0UL, 0UL, &len);
505 
506 	hp = mdesc_alloc(len, &kmalloc_mdesc_memops);
507 	if (!hp) {
508 		printk(KERN_ERR "MD: mdesc alloc fails\n");
509 		goto out;
510 	}
511 
512 	status = sun4v_mach_desc(__pa(&hp->mdesc), len, &real_len);
513 	if (status != HV_EOK || real_len > len) {
514 		printk(KERN_ERR "MD: mdesc reread fails with %lu\n",
515 		       status);
516 		atomic_dec(&hp->refcnt);
517 		mdesc_free(hp);
518 		goto out;
519 	}
520 
521 	spin_lock_irqsave(&mdesc_lock, flags);
522 	orig_hp = cur_mdesc;
523 	cur_mdesc = hp;
524 	spin_unlock_irqrestore(&mdesc_lock, flags);
525 
526 	mdesc_notify_clients(orig_hp, hp);
527 
528 	spin_lock_irqsave(&mdesc_lock, flags);
529 	if (atomic_dec_and_test(&orig_hp->refcnt))
530 		mdesc_free(orig_hp);
531 	else
532 		list_add(&orig_hp->list, &mdesc_zombie_list);
533 	spin_unlock_irqrestore(&mdesc_lock, flags);
534 
535 out:
536 	mutex_unlock(&mdesc_mutex);
537 }
538 
539 u64 mdesc_get_node(struct mdesc_handle *hp, const char *node_name,
540 		   union md_node_info *node_info)
541 {
542 	mdesc_node_info_get_f get_info_func;
543 	mdesc_node_info_rel_f rel_info_func;
544 	mdesc_node_match_f node_match_func;
545 	union md_node_info hp_node_info;
546 	u64 hp_node;
547 	int rv;
548 
549 	if (hp == NULL || node_name == NULL || node_info == NULL)
550 		return MDESC_NODE_NULL;
551 
552 	/* Find the ops for the given node name */
553 	mdesc_get_node_ops(node_name, &get_info_func, &rel_info_func,
554 			   &node_match_func);
555 
556 	/* If we didn't find ops for the given node name, it is not supported */
557 	if (!get_info_func || !rel_info_func || !node_match_func) {
558 		pr_err("MD: %s node is not supported\n", node_name);
559 		return -EINVAL;
560 	}
561 
562 	mdesc_for_each_node_by_name(hp, hp_node, node_name) {
563 		rv = get_info_func(hp, hp_node, &hp_node_info);
564 		if (rv != 0)
565 			continue;
566 
567 		if (node_match_func(node_info, &hp_node_info))
568 			break;
569 
570 		rel_info_func(&hp_node_info);
571 	}
572 
573 	rel_info_func(&hp_node_info);
574 
575 	return hp_node;
576 }
577 EXPORT_SYMBOL(mdesc_get_node);
578 
579 int mdesc_get_node_info(struct mdesc_handle *hp, u64 node,
580 			const char *node_name, union md_node_info *node_info)
581 {
582 	mdesc_node_info_get_f get_info_func;
583 	int rv;
584 
585 	if (hp == NULL || node == MDESC_NODE_NULL ||
586 	    node_name == NULL || node_info == NULL)
587 		return -EINVAL;
588 
589 	/* Find the get_info op for the given node name */
590 	mdesc_get_node_ops(node_name, &get_info_func, NULL, NULL);
591 
592 	/* If we didn't find a get_info_func, the node name is not supported */
593 	if (get_info_func == NULL) {
594 		pr_err("MD: %s node is not supported\n", node_name);
595 		return -EINVAL;
596 	}
597 
598 	rv = get_info_func(hp, node, node_info);
599 	if (rv != 0) {
600 		pr_err("MD: Cannot find 1 or more required match properties for %s node.\n",
601 		       node_name);
602 		return -1;
603 	}
604 
605 	return 0;
606 }
607 EXPORT_SYMBOL(mdesc_get_node_info);
608 
609 static struct mdesc_elem *node_block(struct mdesc_hdr *mdesc)
610 {
611 	return (struct mdesc_elem *) (mdesc + 1);
612 }
613 
614 static void *name_block(struct mdesc_hdr *mdesc)
615 {
616 	return ((void *) node_block(mdesc)) + mdesc->node_sz;
617 }
618 
619 static void *data_block(struct mdesc_hdr *mdesc)
620 {
621 	return ((void *) name_block(mdesc)) + mdesc->name_sz;
622 }
623 
624 u64 mdesc_node_by_name(struct mdesc_handle *hp,
625 		       u64 from_node, const char *name)
626 {
627 	struct mdesc_elem *ep = node_block(&hp->mdesc);
628 	const char *names = name_block(&hp->mdesc);
629 	u64 last_node = hp->mdesc.node_sz / 16;
630 	u64 ret;
631 
632 	if (from_node == MDESC_NODE_NULL) {
633 		ret = from_node = 0;
634 	} else if (from_node >= last_node) {
635 		return MDESC_NODE_NULL;
636 	} else {
637 		ret = ep[from_node].d.val;
638 	}
639 
640 	while (ret < last_node) {
641 		if (ep[ret].tag != MD_NODE)
642 			return MDESC_NODE_NULL;
643 		if (!strcmp(names + ep[ret].name_offset, name))
644 			break;
645 		ret = ep[ret].d.val;
646 	}
647 	if (ret >= last_node)
648 		ret = MDESC_NODE_NULL;
649 	return ret;
650 }
651 EXPORT_SYMBOL(mdesc_node_by_name);
652 
653 const void *mdesc_get_property(struct mdesc_handle *hp, u64 node,
654 			       const char *name, int *lenp)
655 {
656 	const char *names = name_block(&hp->mdesc);
657 	u64 last_node = hp->mdesc.node_sz / 16;
658 	void *data = data_block(&hp->mdesc);
659 	struct mdesc_elem *ep;
660 
661 	if (node == MDESC_NODE_NULL || node >= last_node)
662 		return NULL;
663 
664 	ep = node_block(&hp->mdesc) + node;
665 	ep++;
666 	for (; ep->tag != MD_NODE_END; ep++) {
667 		void *val = NULL;
668 		int len = 0;
669 
670 		switch (ep->tag) {
671 		case MD_PROP_VAL:
672 			val = &ep->d.val;
673 			len = 8;
674 			break;
675 
676 		case MD_PROP_STR:
677 		case MD_PROP_DATA:
678 			val = data + ep->d.data.data_offset;
679 			len = ep->d.data.data_len;
680 			break;
681 
682 		default:
683 			break;
684 		}
685 		if (!val)
686 			continue;
687 
688 		if (!strcmp(names + ep->name_offset, name)) {
689 			if (lenp)
690 				*lenp = len;
691 			return val;
692 		}
693 	}
694 
695 	return NULL;
696 }
697 EXPORT_SYMBOL(mdesc_get_property);
698 
699 u64 mdesc_next_arc(struct mdesc_handle *hp, u64 from, const char *arc_type)
700 {
701 	struct mdesc_elem *ep, *base = node_block(&hp->mdesc);
702 	const char *names = name_block(&hp->mdesc);
703 	u64 last_node = hp->mdesc.node_sz / 16;
704 
705 	if (from == MDESC_NODE_NULL || from >= last_node)
706 		return MDESC_NODE_NULL;
707 
708 	ep = base + from;
709 
710 	ep++;
711 	for (; ep->tag != MD_NODE_END; ep++) {
712 		if (ep->tag != MD_PROP_ARC)
713 			continue;
714 
715 		if (strcmp(names + ep->name_offset, arc_type))
716 			continue;
717 
718 		return ep - base;
719 	}
720 
721 	return MDESC_NODE_NULL;
722 }
723 EXPORT_SYMBOL(mdesc_next_arc);
724 
725 u64 mdesc_arc_target(struct mdesc_handle *hp, u64 arc)
726 {
727 	struct mdesc_elem *ep, *base = node_block(&hp->mdesc);
728 
729 	ep = base + arc;
730 
731 	return ep->d.val;
732 }
733 EXPORT_SYMBOL(mdesc_arc_target);
734 
735 const char *mdesc_node_name(struct mdesc_handle *hp, u64 node)
736 {
737 	struct mdesc_elem *ep, *base = node_block(&hp->mdesc);
738 	const char *names = name_block(&hp->mdesc);
739 	u64 last_node = hp->mdesc.node_sz / 16;
740 
741 	if (node == MDESC_NODE_NULL || node >= last_node)
742 		return NULL;
743 
744 	ep = base + node;
745 	if (ep->tag != MD_NODE)
746 		return NULL;
747 
748 	return names + ep->name_offset;
749 }
750 EXPORT_SYMBOL(mdesc_node_name);
751 
752 static u64 max_cpus = 64;
753 
754 static void __init report_platform_properties(void)
755 {
756 	struct mdesc_handle *hp = mdesc_grab();
757 	u64 pn = mdesc_node_by_name(hp, MDESC_NODE_NULL, "platform");
758 	const char *s;
759 	const u64 *v;
760 
761 	if (pn == MDESC_NODE_NULL) {
762 		prom_printf("No platform node in machine-description.\n");
763 		prom_halt();
764 	}
765 
766 	s = mdesc_get_property(hp, pn, "banner-name", NULL);
767 	printk("PLATFORM: banner-name [%s]\n", s);
768 	s = mdesc_get_property(hp, pn, "name", NULL);
769 	printk("PLATFORM: name [%s]\n", s);
770 
771 	v = mdesc_get_property(hp, pn, "hostid", NULL);
772 	if (v)
773 		printk("PLATFORM: hostid [%08llx]\n", *v);
774 	v = mdesc_get_property(hp, pn, "serial#", NULL);
775 	if (v)
776 		printk("PLATFORM: serial# [%08llx]\n", *v);
777 	v = mdesc_get_property(hp, pn, "stick-frequency", NULL);
778 	printk("PLATFORM: stick-frequency [%08llx]\n", *v);
779 	v = mdesc_get_property(hp, pn, "mac-address", NULL);
780 	if (v)
781 		printk("PLATFORM: mac-address [%llx]\n", *v);
782 	v = mdesc_get_property(hp, pn, "watchdog-resolution", NULL);
783 	if (v)
784 		printk("PLATFORM: watchdog-resolution [%llu ms]\n", *v);
785 	v = mdesc_get_property(hp, pn, "watchdog-max-timeout", NULL);
786 	if (v)
787 		printk("PLATFORM: watchdog-max-timeout [%llu ms]\n", *v);
788 	v = mdesc_get_property(hp, pn, "max-cpus", NULL);
789 	if (v) {
790 		max_cpus = *v;
791 		printk("PLATFORM: max-cpus [%llu]\n", max_cpus);
792 	}
793 
794 #ifdef CONFIG_SMP
795 	{
796 		int max_cpu, i;
797 
798 		if (v) {
799 			max_cpu = *v;
800 			if (max_cpu > NR_CPUS)
801 				max_cpu = NR_CPUS;
802 		} else {
803 			max_cpu = NR_CPUS;
804 		}
805 		for (i = 0; i < max_cpu; i++)
806 			set_cpu_possible(i, true);
807 	}
808 #endif
809 
810 	mdesc_release(hp);
811 }
812 
813 static void fill_in_one_cache(cpuinfo_sparc *c, struct mdesc_handle *hp, u64 mp)
814 {
815 	const u64 *level = mdesc_get_property(hp, mp, "level", NULL);
816 	const u64 *size = mdesc_get_property(hp, mp, "size", NULL);
817 	const u64 *line_size = mdesc_get_property(hp, mp, "line-size", NULL);
818 	const char *type;
819 	int type_len;
820 
821 	type = mdesc_get_property(hp, mp, "type", &type_len);
822 
823 	switch (*level) {
824 	case 1:
825 		if (of_find_in_proplist(type, "instn", type_len)) {
826 			c->icache_size = *size;
827 			c->icache_line_size = *line_size;
828 		} else if (of_find_in_proplist(type, "data", type_len)) {
829 			c->dcache_size = *size;
830 			c->dcache_line_size = *line_size;
831 		}
832 		break;
833 
834 	case 2:
835 		c->ecache_size = *size;
836 		c->ecache_line_size = *line_size;
837 		break;
838 
839 	default:
840 		break;
841 	}
842 
843 	if (*level == 1) {
844 		u64 a;
845 
846 		mdesc_for_each_arc(a, hp, mp, MDESC_ARC_TYPE_FWD) {
847 			u64 target = mdesc_arc_target(hp, a);
848 			const char *name = mdesc_node_name(hp, target);
849 
850 			if (!strcmp(name, "cache"))
851 				fill_in_one_cache(c, hp, target);
852 		}
853 	}
854 }
855 
856 static void find_back_node_value(struct mdesc_handle *hp, u64 node,
857 				 char *srch_val,
858 				 void (*func)(struct mdesc_handle *, u64, int),
859 				 u64 val, int depth)
860 {
861 	u64 arc;
862 
863 	/* Since we have an estimate of recursion depth, do a sanity check. */
864 	if (depth == 0)
865 		return;
866 
867 	mdesc_for_each_arc(arc, hp, node, MDESC_ARC_TYPE_BACK) {
868 		u64 n = mdesc_arc_target(hp, arc);
869 		const char *name = mdesc_node_name(hp, n);
870 
871 		if (!strcmp(srch_val, name))
872 			(*func)(hp, n, val);
873 
874 		find_back_node_value(hp, n, srch_val, func, val, depth-1);
875 	}
876 }
877 
878 static void __mark_core_id(struct mdesc_handle *hp, u64 node,
879 			   int core_id)
880 {
881 	const u64 *id = mdesc_get_property(hp, node, "id", NULL);
882 
883 	if (*id < num_possible_cpus())
884 		cpu_data(*id).core_id = core_id;
885 }
886 
887 static void __mark_max_cache_id(struct mdesc_handle *hp, u64 node,
888 				int max_cache_id)
889 {
890 	const u64 *id = mdesc_get_property(hp, node, "id", NULL);
891 
892 	if (*id < num_possible_cpus()) {
893 		cpu_data(*id).max_cache_id = max_cache_id;
894 
895 		/**
896 		 * On systems without explicit socket descriptions socket
897 		 * is max_cache_id
898 		 */
899 		cpu_data(*id).sock_id = max_cache_id;
900 	}
901 }
902 
903 static void mark_core_ids(struct mdesc_handle *hp, u64 mp,
904 			  int core_id)
905 {
906 	find_back_node_value(hp, mp, "cpu", __mark_core_id, core_id, 10);
907 }
908 
909 static void mark_max_cache_ids(struct mdesc_handle *hp, u64 mp,
910 			       int max_cache_id)
911 {
912 	find_back_node_value(hp, mp, "cpu", __mark_max_cache_id,
913 			     max_cache_id, 10);
914 }
915 
916 static void set_core_ids(struct mdesc_handle *hp)
917 {
918 	int idx;
919 	u64 mp;
920 
921 	idx = 1;
922 
923 	/* Identify unique cores by looking for cpus backpointed to by
924 	 * level 1 instruction caches.
925 	 */
926 	mdesc_for_each_node_by_name(hp, mp, "cache") {
927 		const u64 *level;
928 		const char *type;
929 		int len;
930 
931 		level = mdesc_get_property(hp, mp, "level", NULL);
932 		if (*level != 1)
933 			continue;
934 
935 		type = mdesc_get_property(hp, mp, "type", &len);
936 		if (!of_find_in_proplist(type, "instn", len))
937 			continue;
938 
939 		mark_core_ids(hp, mp, idx);
940 		idx++;
941 	}
942 }
943 
944 static int set_max_cache_ids_by_cache(struct mdesc_handle *hp, int level)
945 {
946 	u64 mp;
947 	int idx = 1;
948 	int fnd = 0;
949 
950 	/**
951 	 * Identify unique highest level of shared cache by looking for cpus
952 	 * backpointed to by shared level N caches.
953 	 */
954 	mdesc_for_each_node_by_name(hp, mp, "cache") {
955 		const u64 *cur_lvl;
956 
957 		cur_lvl = mdesc_get_property(hp, mp, "level", NULL);
958 		if (*cur_lvl != level)
959 			continue;
960 		mark_max_cache_ids(hp, mp, idx);
961 		idx++;
962 		fnd = 1;
963 	}
964 	return fnd;
965 }
966 
967 static void set_sock_ids_by_socket(struct mdesc_handle *hp, u64 mp)
968 {
969 	int idx = 1;
970 
971 	mdesc_for_each_node_by_name(hp, mp, "socket") {
972 		u64 a;
973 
974 		mdesc_for_each_arc(a, hp, mp, MDESC_ARC_TYPE_FWD) {
975 			u64 t = mdesc_arc_target(hp, a);
976 			const char *name;
977 			const u64 *id;
978 
979 			name = mdesc_node_name(hp, t);
980 			if (strcmp(name, "cpu"))
981 				continue;
982 
983 			id = mdesc_get_property(hp, t, "id", NULL);
984 			if (*id < num_possible_cpus())
985 				cpu_data(*id).sock_id = idx;
986 		}
987 		idx++;
988 	}
989 }
990 
991 static void set_sock_ids(struct mdesc_handle *hp)
992 {
993 	u64 mp;
994 
995 	/**
996 	 * Find the highest level of shared cache which pre-T7 is also
997 	 * the socket.
998 	 */
999 	if (!set_max_cache_ids_by_cache(hp, 3))
1000 		set_max_cache_ids_by_cache(hp, 2);
1001 
1002 	/* If machine description exposes sockets data use it.*/
1003 	mp = mdesc_node_by_name(hp, MDESC_NODE_NULL, "sockets");
1004 	if (mp != MDESC_NODE_NULL)
1005 		set_sock_ids_by_socket(hp, mp);
1006 }
1007 
1008 static void mark_proc_ids(struct mdesc_handle *hp, u64 mp, int proc_id)
1009 {
1010 	u64 a;
1011 
1012 	mdesc_for_each_arc(a, hp, mp, MDESC_ARC_TYPE_BACK) {
1013 		u64 t = mdesc_arc_target(hp, a);
1014 		const char *name;
1015 		const u64 *id;
1016 
1017 		name = mdesc_node_name(hp, t);
1018 		if (strcmp(name, "cpu"))
1019 			continue;
1020 
1021 		id = mdesc_get_property(hp, t, "id", NULL);
1022 		if (*id < NR_CPUS)
1023 			cpu_data(*id).proc_id = proc_id;
1024 	}
1025 }
1026 
1027 static void __set_proc_ids(struct mdesc_handle *hp, const char *exec_unit_name)
1028 {
1029 	int idx;
1030 	u64 mp;
1031 
1032 	idx = 0;
1033 	mdesc_for_each_node_by_name(hp, mp, exec_unit_name) {
1034 		const char *type;
1035 		int len;
1036 
1037 		type = mdesc_get_property(hp, mp, "type", &len);
1038 		if (!of_find_in_proplist(type, "int", len) &&
1039 		    !of_find_in_proplist(type, "integer", len))
1040 			continue;
1041 
1042 		mark_proc_ids(hp, mp, idx);
1043 		idx++;
1044 	}
1045 }
1046 
1047 static void set_proc_ids(struct mdesc_handle *hp)
1048 {
1049 	__set_proc_ids(hp, "exec_unit");
1050 	__set_proc_ids(hp, "exec-unit");
1051 }
1052 
1053 static void get_one_mondo_bits(const u64 *p, unsigned int *mask,
1054 			       unsigned long def, unsigned long max)
1055 {
1056 	u64 val;
1057 
1058 	if (!p)
1059 		goto use_default;
1060 	val = *p;
1061 
1062 	if (!val || val >= 64)
1063 		goto use_default;
1064 
1065 	if (val > max)
1066 		val = max;
1067 
1068 	*mask = ((1U << val) * 64U) - 1U;
1069 	return;
1070 
1071 use_default:
1072 	*mask = ((1U << def) * 64U) - 1U;
1073 }
1074 
1075 static void get_mondo_data(struct mdesc_handle *hp, u64 mp,
1076 			   struct trap_per_cpu *tb)
1077 {
1078 	static int printed;
1079 	const u64 *val;
1080 
1081 	val = mdesc_get_property(hp, mp, "q-cpu-mondo-#bits", NULL);
1082 	get_one_mondo_bits(val, &tb->cpu_mondo_qmask, 7, ilog2(max_cpus * 2));
1083 
1084 	val = mdesc_get_property(hp, mp, "q-dev-mondo-#bits", NULL);
1085 	get_one_mondo_bits(val, &tb->dev_mondo_qmask, 7, 8);
1086 
1087 	val = mdesc_get_property(hp, mp, "q-resumable-#bits", NULL);
1088 	get_one_mondo_bits(val, &tb->resum_qmask, 6, 7);
1089 
1090 	val = mdesc_get_property(hp, mp, "q-nonresumable-#bits", NULL);
1091 	get_one_mondo_bits(val, &tb->nonresum_qmask, 2, 2);
1092 	if (!printed++) {
1093 		pr_info("SUN4V: Mondo queue sizes "
1094 			"[cpu(%u) dev(%u) r(%u) nr(%u)]\n",
1095 			tb->cpu_mondo_qmask + 1,
1096 			tb->dev_mondo_qmask + 1,
1097 			tb->resum_qmask + 1,
1098 			tb->nonresum_qmask + 1);
1099 	}
1100 }
1101 
1102 static void *mdesc_iterate_over_cpus(void *(*func)(struct mdesc_handle *, u64, int, void *), void *arg, cpumask_t *mask)
1103 {
1104 	struct mdesc_handle *hp = mdesc_grab();
1105 	void *ret = NULL;
1106 	u64 mp;
1107 
1108 	mdesc_for_each_node_by_name(hp, mp, "cpu") {
1109 		const u64 *id = mdesc_get_property(hp, mp, "id", NULL);
1110 		int cpuid = *id;
1111 
1112 #ifdef CONFIG_SMP
1113 		if (cpuid >= NR_CPUS) {
1114 			printk(KERN_WARNING "Ignoring CPU %d which is "
1115 			       ">= NR_CPUS (%d)\n",
1116 			       cpuid, NR_CPUS);
1117 			continue;
1118 		}
1119 		if (!cpumask_test_cpu(cpuid, mask))
1120 			continue;
1121 #endif
1122 
1123 		ret = func(hp, mp, cpuid, arg);
1124 		if (ret)
1125 			goto out;
1126 	}
1127 out:
1128 	mdesc_release(hp);
1129 	return ret;
1130 }
1131 
1132 static void *record_one_cpu(struct mdesc_handle *hp, u64 mp, int cpuid,
1133 			    void *arg)
1134 {
1135 	ncpus_probed++;
1136 #ifdef CONFIG_SMP
1137 	set_cpu_present(cpuid, true);
1138 #endif
1139 	return NULL;
1140 }
1141 
1142 void mdesc_populate_present_mask(cpumask_t *mask)
1143 {
1144 	if (tlb_type != hypervisor)
1145 		return;
1146 
1147 	ncpus_probed = 0;
1148 	mdesc_iterate_over_cpus(record_one_cpu, NULL, mask);
1149 }
1150 
1151 static void * __init check_one_pgsz(struct mdesc_handle *hp, u64 mp, int cpuid, void *arg)
1152 {
1153 	const u64 *pgsz_prop = mdesc_get_property(hp, mp, "mmu-page-size-list", NULL);
1154 	unsigned long *pgsz_mask = arg;
1155 	u64 val;
1156 
1157 	val = (HV_PGSZ_MASK_8K | HV_PGSZ_MASK_64K |
1158 	       HV_PGSZ_MASK_512K | HV_PGSZ_MASK_4MB);
1159 	if (pgsz_prop)
1160 		val = *pgsz_prop;
1161 
1162 	if (!*pgsz_mask)
1163 		*pgsz_mask = val;
1164 	else
1165 		*pgsz_mask &= val;
1166 	return NULL;
1167 }
1168 
1169 void __init mdesc_get_page_sizes(cpumask_t *mask, unsigned long *pgsz_mask)
1170 {
1171 	*pgsz_mask = 0;
1172 	mdesc_iterate_over_cpus(check_one_pgsz, pgsz_mask, mask);
1173 }
1174 
1175 static void *fill_in_one_cpu(struct mdesc_handle *hp, u64 mp, int cpuid,
1176 			     void *arg)
1177 {
1178 	const u64 *cfreq = mdesc_get_property(hp, mp, "clock-frequency", NULL);
1179 	struct trap_per_cpu *tb;
1180 	cpuinfo_sparc *c;
1181 	u64 a;
1182 
1183 #ifndef CONFIG_SMP
1184 	/* On uniprocessor we only want the values for the
1185 	 * real physical cpu the kernel booted onto, however
1186 	 * cpu_data() only has one entry at index 0.
1187 	 */
1188 	if (cpuid != real_hard_smp_processor_id())
1189 		return NULL;
1190 	cpuid = 0;
1191 #endif
1192 
1193 	c = &cpu_data(cpuid);
1194 	c->clock_tick = *cfreq;
1195 
1196 	tb = &trap_block[cpuid];
1197 	get_mondo_data(hp, mp, tb);
1198 
1199 	mdesc_for_each_arc(a, hp, mp, MDESC_ARC_TYPE_FWD) {
1200 		u64 j, t = mdesc_arc_target(hp, a);
1201 		const char *t_name;
1202 
1203 		t_name = mdesc_node_name(hp, t);
1204 		if (!strcmp(t_name, "cache")) {
1205 			fill_in_one_cache(c, hp, t);
1206 			continue;
1207 		}
1208 
1209 		mdesc_for_each_arc(j, hp, t, MDESC_ARC_TYPE_FWD) {
1210 			u64 n = mdesc_arc_target(hp, j);
1211 			const char *n_name;
1212 
1213 			n_name = mdesc_node_name(hp, n);
1214 			if (!strcmp(n_name, "cache"))
1215 				fill_in_one_cache(c, hp, n);
1216 		}
1217 	}
1218 
1219 	c->core_id = 0;
1220 	c->proc_id = -1;
1221 
1222 	return NULL;
1223 }
1224 
1225 void mdesc_fill_in_cpu_data(cpumask_t *mask)
1226 {
1227 	struct mdesc_handle *hp;
1228 
1229 	mdesc_iterate_over_cpus(fill_in_one_cpu, NULL, mask);
1230 
1231 	hp = mdesc_grab();
1232 
1233 	set_core_ids(hp);
1234 	set_proc_ids(hp);
1235 	set_sock_ids(hp);
1236 
1237 	mdesc_release(hp);
1238 
1239 	smp_fill_in_sib_core_maps();
1240 }
1241 
1242 /* mdesc_open() - Grab a reference to mdesc_handle when /dev/mdesc is
1243  * opened. Hold this reference until /dev/mdesc is closed to ensure
1244  * mdesc data structure is not released underneath us. Store the
1245  * pointer to mdesc structure in private_data for read and seek to use
1246  */
1247 static int mdesc_open(struct inode *inode, struct file *file)
1248 {
1249 	struct mdesc_handle *hp = mdesc_grab();
1250 
1251 	if (!hp)
1252 		return -ENODEV;
1253 
1254 	file->private_data = hp;
1255 
1256 	return 0;
1257 }
1258 
1259 static ssize_t mdesc_read(struct file *file, char __user *buf,
1260 			  size_t len, loff_t *offp)
1261 {
1262 	struct mdesc_handle *hp = file->private_data;
1263 	unsigned char *mdesc;
1264 	int bytes_left, count = len;
1265 
1266 	if (*offp >= hp->handle_size)
1267 		return 0;
1268 
1269 	bytes_left = hp->handle_size - *offp;
1270 	if (count > bytes_left)
1271 		count = bytes_left;
1272 
1273 	mdesc = (unsigned char *)&hp->mdesc;
1274 	mdesc += *offp;
1275 	if (!copy_to_user(buf, mdesc, count)) {
1276 		*offp += count;
1277 		return count;
1278 	} else {
1279 		return -EFAULT;
1280 	}
1281 }
1282 
1283 static loff_t mdesc_llseek(struct file *file, loff_t offset, int whence)
1284 {
1285 	struct mdesc_handle *hp = file->private_data;
1286 
1287 	return no_seek_end_llseek_size(file, offset, whence, hp->handle_size);
1288 }
1289 
1290 /* mdesc_close() - /dev/mdesc is being closed, release the reference to
1291  * mdesc structure.
1292  */
1293 static int mdesc_close(struct inode *inode, struct file *file)
1294 {
1295 	mdesc_release(file->private_data);
1296 	return 0;
1297 }
1298 
1299 static const struct file_operations mdesc_fops = {
1300 	.open    = mdesc_open,
1301 	.read	 = mdesc_read,
1302 	.llseek  = mdesc_llseek,
1303 	.release = mdesc_close,
1304 	.owner	 = THIS_MODULE,
1305 };
1306 
1307 static struct miscdevice mdesc_misc = {
1308 	.minor	= MISC_DYNAMIC_MINOR,
1309 	.name	= "mdesc",
1310 	.fops	= &mdesc_fops,
1311 };
1312 
1313 static int __init mdesc_misc_init(void)
1314 {
1315 	return misc_register(&mdesc_misc);
1316 }
1317 
1318 __initcall(mdesc_misc_init);
1319 
1320 void __init sun4v_mdesc_init(void)
1321 {
1322 	struct mdesc_handle *hp;
1323 	unsigned long len, real_len, status;
1324 
1325 	(void) sun4v_mach_desc(0UL, 0UL, &len);
1326 
1327 	printk("MDESC: Size is %lu bytes.\n", len);
1328 
1329 	hp = mdesc_alloc(len, &memblock_mdesc_ops);
1330 	if (hp == NULL) {
1331 		prom_printf("MDESC: alloc of %lu bytes failed.\n", len);
1332 		prom_halt();
1333 	}
1334 
1335 	status = sun4v_mach_desc(__pa(&hp->mdesc), len, &real_len);
1336 	if (status != HV_EOK || real_len > len) {
1337 		prom_printf("sun4v_mach_desc fails, err(%lu), "
1338 			    "len(%lu), real_len(%lu)\n",
1339 			    status, len, real_len);
1340 		mdesc_free(hp);
1341 		prom_halt();
1342 	}
1343 
1344 	cur_mdesc = hp;
1345 
1346 	report_platform_properties();
1347 }
1348