xref: /linux/arch/sparc/kernel/mdesc.c (revision c79c3c34f75d72a066e292b10aa50fc758c97c89)
1 // SPDX-License-Identifier: GPL-2.0
2 /* mdesc.c: Sun4V machine description handling.
3  *
4  * Copyright (C) 2007, 2008 David S. Miller <davem@davemloft.net>
5  */
6 #include <linux/kernel.h>
7 #include <linux/types.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/memblock.h>
14 #include <linux/export.h>
15 #include <linux/refcount.h>
16 
17 #include <asm/cpudata.h>
18 #include <asm/hypervisor.h>
19 #include <asm/mdesc.h>
20 #include <asm/prom.h>
21 #include <linux/uaccess.h>
22 #include <asm/oplib.h>
23 #include <asm/smp.h>
24 #include <asm/adi.h>
25 
26 /* Unlike the OBP device tree, the machine description is a full-on
27  * DAG.  An arbitrary number of ARCs are possible from one
28  * node to other nodes and thus we can't use the OBP device_node
29  * data structure to represent these nodes inside of the kernel.
30  *
31  * Actually, it isn't even a DAG, because there are back pointers
32  * which create cycles in the graph.
33  *
34  * mdesc_hdr and mdesc_elem describe the layout of the data structure
35  * we get from the Hypervisor.
36  */
37 struct mdesc_hdr {
38 	u32	version; /* Transport version */
39 	u32	node_sz; /* node block size */
40 	u32	name_sz; /* name block size */
41 	u32	data_sz; /* data block size */
42 } __attribute__((aligned(16)));
43 
44 struct mdesc_elem {
45 	u8	tag;
46 #define MD_LIST_END	0x00
47 #define MD_NODE		0x4e
48 #define MD_NODE_END	0x45
49 #define MD_NOOP		0x20
50 #define MD_PROP_ARC	0x61
51 #define MD_PROP_VAL	0x76
52 #define MD_PROP_STR	0x73
53 #define MD_PROP_DATA	0x64
54 	u8	name_len;
55 	u16	resv;
56 	u32	name_offset;
57 	union {
58 		struct {
59 			u32	data_len;
60 			u32	data_offset;
61 		} data;
62 		u64	val;
63 	} d;
64 };
65 
66 struct mdesc_mem_ops {
67 	struct mdesc_handle *(*alloc)(unsigned int mdesc_size);
68 	void (*free)(struct mdesc_handle *handle);
69 };
70 
71 struct mdesc_handle {
72 	struct list_head	list;
73 	struct mdesc_mem_ops	*mops;
74 	void			*self_base;
75 	refcount_t		refcnt;
76 	unsigned int		handle_size;
77 	struct mdesc_hdr	mdesc;
78 };
79 
80 typedef int (*mdesc_node_info_get_f)(struct mdesc_handle *, u64,
81 				     union md_node_info *);
82 typedef void (*mdesc_node_info_rel_f)(union md_node_info *);
83 typedef bool (*mdesc_node_match_f)(union md_node_info *, union md_node_info *);
84 
85 struct md_node_ops {
86 	char			*name;
87 	mdesc_node_info_get_f	get_info;
88 	mdesc_node_info_rel_f	rel_info;
89 	mdesc_node_match_f	node_match;
90 };
91 
92 static int get_vdev_port_node_info(struct mdesc_handle *md, u64 node,
93 				   union md_node_info *node_info);
94 static void rel_vdev_port_node_info(union md_node_info *node_info);
95 static bool vdev_port_node_match(union md_node_info *a_node_info,
96 				 union md_node_info *b_node_info);
97 
98 static int get_ds_port_node_info(struct mdesc_handle *md, u64 node,
99 				 union md_node_info *node_info);
100 static void rel_ds_port_node_info(union md_node_info *node_info);
101 static bool ds_port_node_match(union md_node_info *a_node_info,
102 			       union md_node_info *b_node_info);
103 
104 /* supported node types which can be registered */
105 static struct md_node_ops md_node_ops_table[] = {
106 	{"virtual-device-port", get_vdev_port_node_info,
107 	 rel_vdev_port_node_info, vdev_port_node_match},
108 	{"domain-services-port", get_ds_port_node_info,
109 	 rel_ds_port_node_info, ds_port_node_match},
110 	{NULL, NULL, NULL, NULL}
111 };
112 
113 static void mdesc_get_node_ops(const char *node_name,
114 			       mdesc_node_info_get_f *get_info_f,
115 			       mdesc_node_info_rel_f *rel_info_f,
116 			       mdesc_node_match_f *match_f)
117 {
118 	int i;
119 
120 	if (get_info_f)
121 		*get_info_f = NULL;
122 
123 	if (rel_info_f)
124 		*rel_info_f = NULL;
125 
126 	if (match_f)
127 		*match_f = NULL;
128 
129 	if (!node_name)
130 		return;
131 
132 	for (i = 0; md_node_ops_table[i].name != NULL; i++) {
133 		if (strcmp(md_node_ops_table[i].name, node_name) == 0) {
134 			if (get_info_f)
135 				*get_info_f = md_node_ops_table[i].get_info;
136 
137 			if (rel_info_f)
138 				*rel_info_f = md_node_ops_table[i].rel_info;
139 
140 			if (match_f)
141 				*match_f = md_node_ops_table[i].node_match;
142 
143 			break;
144 		}
145 	}
146 }
147 
148 static void mdesc_handle_init(struct mdesc_handle *hp,
149 			      unsigned int handle_size,
150 			      void *base)
151 {
152 	BUG_ON(((unsigned long)&hp->mdesc) & (16UL - 1));
153 
154 	memset(hp, 0, handle_size);
155 	INIT_LIST_HEAD(&hp->list);
156 	hp->self_base = base;
157 	refcount_set(&hp->refcnt, 1);
158 	hp->handle_size = handle_size;
159 }
160 
161 static struct mdesc_handle * __init mdesc_memblock_alloc(unsigned int mdesc_size)
162 {
163 	unsigned int handle_size, alloc_size;
164 	struct mdesc_handle *hp;
165 	unsigned long paddr;
166 
167 	handle_size = (sizeof(struct mdesc_handle) -
168 		       sizeof(struct mdesc_hdr) +
169 		       mdesc_size);
170 	alloc_size = PAGE_ALIGN(handle_size);
171 
172 	paddr = memblock_phys_alloc(alloc_size, PAGE_SIZE);
173 
174 	hp = NULL;
175 	if (paddr) {
176 		hp = __va(paddr);
177 		mdesc_handle_init(hp, handle_size, hp);
178 	}
179 	return hp;
180 }
181 
182 static void __init mdesc_memblock_free(struct mdesc_handle *hp)
183 {
184 	unsigned int alloc_size;
185 	unsigned long start;
186 
187 	BUG_ON(refcount_read(&hp->refcnt) != 0);
188 	BUG_ON(!list_empty(&hp->list));
189 
190 	alloc_size = PAGE_ALIGN(hp->handle_size);
191 	start = __pa(hp);
192 	memblock_free_late(start, alloc_size);
193 }
194 
195 static struct mdesc_mem_ops memblock_mdesc_ops = {
196 	.alloc = mdesc_memblock_alloc,
197 	.free  = mdesc_memblock_free,
198 };
199 
200 static struct mdesc_handle *mdesc_kmalloc(unsigned int mdesc_size)
201 {
202 	unsigned int handle_size;
203 	struct mdesc_handle *hp;
204 	unsigned long addr;
205 	void *base;
206 
207 	handle_size = (sizeof(struct mdesc_handle) -
208 		       sizeof(struct mdesc_hdr) +
209 		       mdesc_size);
210 	base = kmalloc(handle_size + 15, GFP_KERNEL | __GFP_RETRY_MAYFAIL);
211 	if (!base)
212 		return NULL;
213 
214 	addr = (unsigned long)base;
215 	addr = (addr + 15UL) & ~15UL;
216 	hp = (struct mdesc_handle *) addr;
217 
218 	mdesc_handle_init(hp, handle_size, base);
219 
220 	return hp;
221 }
222 
223 static void mdesc_kfree(struct mdesc_handle *hp)
224 {
225 	BUG_ON(refcount_read(&hp->refcnt) != 0);
226 	BUG_ON(!list_empty(&hp->list));
227 
228 	kfree(hp->self_base);
229 }
230 
231 static struct mdesc_mem_ops kmalloc_mdesc_memops = {
232 	.alloc = mdesc_kmalloc,
233 	.free  = mdesc_kfree,
234 };
235 
236 static struct mdesc_handle *mdesc_alloc(unsigned int mdesc_size,
237 					struct mdesc_mem_ops *mops)
238 {
239 	struct mdesc_handle *hp = mops->alloc(mdesc_size);
240 
241 	if (hp)
242 		hp->mops = mops;
243 
244 	return hp;
245 }
246 
247 static void mdesc_free(struct mdesc_handle *hp)
248 {
249 	hp->mops->free(hp);
250 }
251 
252 static struct mdesc_handle *cur_mdesc;
253 static LIST_HEAD(mdesc_zombie_list);
254 static DEFINE_SPINLOCK(mdesc_lock);
255 
256 struct mdesc_handle *mdesc_grab(void)
257 {
258 	struct mdesc_handle *hp;
259 	unsigned long flags;
260 
261 	spin_lock_irqsave(&mdesc_lock, flags);
262 	hp = cur_mdesc;
263 	if (hp)
264 		refcount_inc(&hp->refcnt);
265 	spin_unlock_irqrestore(&mdesc_lock, flags);
266 
267 	return hp;
268 }
269 EXPORT_SYMBOL(mdesc_grab);
270 
271 void mdesc_release(struct mdesc_handle *hp)
272 {
273 	unsigned long flags;
274 
275 	spin_lock_irqsave(&mdesc_lock, flags);
276 	if (refcount_dec_and_test(&hp->refcnt)) {
277 		list_del_init(&hp->list);
278 		hp->mops->free(hp);
279 	}
280 	spin_unlock_irqrestore(&mdesc_lock, flags);
281 }
282 EXPORT_SYMBOL(mdesc_release);
283 
284 static DEFINE_MUTEX(mdesc_mutex);
285 static struct mdesc_notifier_client *client_list;
286 
287 void mdesc_register_notifier(struct mdesc_notifier_client *client)
288 {
289 	bool supported = false;
290 	u64 node;
291 	int i;
292 
293 	mutex_lock(&mdesc_mutex);
294 
295 	/* check to see if the node is supported for registration */
296 	for (i = 0; md_node_ops_table[i].name != NULL; i++) {
297 		if (strcmp(md_node_ops_table[i].name, client->node_name) == 0) {
298 			supported = true;
299 			break;
300 		}
301 	}
302 
303 	if (!supported) {
304 		pr_err("MD: %s node not supported\n", client->node_name);
305 		mutex_unlock(&mdesc_mutex);
306 		return;
307 	}
308 
309 	client->next = client_list;
310 	client_list = client;
311 
312 	mdesc_for_each_node_by_name(cur_mdesc, node, client->node_name)
313 		client->add(cur_mdesc, node, client->node_name);
314 
315 	mutex_unlock(&mdesc_mutex);
316 }
317 
318 static const u64 *parent_cfg_handle(struct mdesc_handle *hp, u64 node)
319 {
320 	const u64 *id;
321 	u64 a;
322 
323 	id = NULL;
324 	mdesc_for_each_arc(a, hp, node, MDESC_ARC_TYPE_BACK) {
325 		u64 target;
326 
327 		target = mdesc_arc_target(hp, a);
328 		id = mdesc_get_property(hp, target,
329 					"cfg-handle", NULL);
330 		if (id)
331 			break;
332 	}
333 
334 	return id;
335 }
336 
337 static int get_vdev_port_node_info(struct mdesc_handle *md, u64 node,
338 				   union md_node_info *node_info)
339 {
340 	const u64 *parent_cfg_hdlp;
341 	const char *name;
342 	const u64 *idp;
343 
344 	/*
345 	 * Virtual device nodes are distinguished by:
346 	 * 1. "id" property
347 	 * 2. "name" property
348 	 * 3. parent node "cfg-handle" property
349 	 */
350 	idp = mdesc_get_property(md, node, "id", NULL);
351 	name = mdesc_get_property(md, node, "name", NULL);
352 	parent_cfg_hdlp = parent_cfg_handle(md, node);
353 
354 	if (!idp || !name || !parent_cfg_hdlp)
355 		return -1;
356 
357 	node_info->vdev_port.id = *idp;
358 	node_info->vdev_port.name = kstrdup_const(name, GFP_KERNEL);
359 	if (!node_info->vdev_port.name)
360 		return -1;
361 	node_info->vdev_port.parent_cfg_hdl = *parent_cfg_hdlp;
362 
363 	return 0;
364 }
365 
366 static void rel_vdev_port_node_info(union md_node_info *node_info)
367 {
368 	if (node_info && node_info->vdev_port.name) {
369 		kfree_const(node_info->vdev_port.name);
370 		node_info->vdev_port.name = NULL;
371 	}
372 }
373 
374 static bool vdev_port_node_match(union md_node_info *a_node_info,
375 				 union md_node_info *b_node_info)
376 {
377 	if (a_node_info->vdev_port.id != b_node_info->vdev_port.id)
378 		return false;
379 
380 	if (a_node_info->vdev_port.parent_cfg_hdl !=
381 	    b_node_info->vdev_port.parent_cfg_hdl)
382 		return false;
383 
384 	if (strncmp(a_node_info->vdev_port.name,
385 		    b_node_info->vdev_port.name, MDESC_MAX_STR_LEN) != 0)
386 		return false;
387 
388 	return true;
389 }
390 
391 static int get_ds_port_node_info(struct mdesc_handle *md, u64 node,
392 				 union md_node_info *node_info)
393 {
394 	const u64 *idp;
395 
396 	/* DS port nodes use the "id" property to distinguish them */
397 	idp = mdesc_get_property(md, node, "id", NULL);
398 	if (!idp)
399 		return -1;
400 
401 	node_info->ds_port.id = *idp;
402 
403 	return 0;
404 }
405 
406 static void rel_ds_port_node_info(union md_node_info *node_info)
407 {
408 }
409 
410 static bool ds_port_node_match(union md_node_info *a_node_info,
411 			       union md_node_info *b_node_info)
412 {
413 	if (a_node_info->ds_port.id != b_node_info->ds_port.id)
414 		return false;
415 
416 	return true;
417 }
418 
419 /* Run 'func' on nodes which are in A but not in B.  */
420 static void invoke_on_missing(const char *name,
421 			      struct mdesc_handle *a,
422 			      struct mdesc_handle *b,
423 			      void (*func)(struct mdesc_handle *, u64,
424 					   const char *node_name))
425 {
426 	mdesc_node_info_get_f get_info_func;
427 	mdesc_node_info_rel_f rel_info_func;
428 	mdesc_node_match_f node_match_func;
429 	union md_node_info a_node_info;
430 	union md_node_info b_node_info;
431 	bool found;
432 	u64 a_node;
433 	u64 b_node;
434 	int rv;
435 
436 	/*
437 	 * Find the get_info, rel_info and node_match ops for the given
438 	 * node name
439 	 */
440 	mdesc_get_node_ops(name, &get_info_func, &rel_info_func,
441 			   &node_match_func);
442 
443 	/* If we didn't find a match, the node type is not supported */
444 	if (!get_info_func || !rel_info_func || !node_match_func) {
445 		pr_err("MD: %s node type is not supported\n", name);
446 		return;
447 	}
448 
449 	mdesc_for_each_node_by_name(a, a_node, name) {
450 		found = false;
451 
452 		rv = get_info_func(a, a_node, &a_node_info);
453 		if (rv != 0) {
454 			pr_err("MD: Cannot find 1 or more required match properties for %s node.\n",
455 			       name);
456 			continue;
457 		}
458 
459 		/* Check each node in B for node matching a_node */
460 		mdesc_for_each_node_by_name(b, b_node, name) {
461 			rv = get_info_func(b, b_node, &b_node_info);
462 			if (rv != 0)
463 				continue;
464 
465 			if (node_match_func(&a_node_info, &b_node_info)) {
466 				found = true;
467 				rel_info_func(&b_node_info);
468 				break;
469 			}
470 
471 			rel_info_func(&b_node_info);
472 		}
473 
474 		rel_info_func(&a_node_info);
475 
476 		if (!found)
477 			func(a, a_node, name);
478 	}
479 }
480 
481 static void notify_one(struct mdesc_notifier_client *p,
482 		       struct mdesc_handle *old_hp,
483 		       struct mdesc_handle *new_hp)
484 {
485 	invoke_on_missing(p->node_name, old_hp, new_hp, p->remove);
486 	invoke_on_missing(p->node_name, new_hp, old_hp, p->add);
487 }
488 
489 static void mdesc_notify_clients(struct mdesc_handle *old_hp,
490 				 struct mdesc_handle *new_hp)
491 {
492 	struct mdesc_notifier_client *p = client_list;
493 
494 	while (p) {
495 		notify_one(p, old_hp, new_hp);
496 		p = p->next;
497 	}
498 }
499 
500 void mdesc_update(void)
501 {
502 	unsigned long len, real_len, status;
503 	struct mdesc_handle *hp, *orig_hp;
504 	unsigned long flags;
505 
506 	mutex_lock(&mdesc_mutex);
507 
508 	(void) sun4v_mach_desc(0UL, 0UL, &len);
509 
510 	hp = mdesc_alloc(len, &kmalloc_mdesc_memops);
511 	if (!hp) {
512 		printk(KERN_ERR "MD: mdesc alloc fails\n");
513 		goto out;
514 	}
515 
516 	status = sun4v_mach_desc(__pa(&hp->mdesc), len, &real_len);
517 	if (status != HV_EOK || real_len > len) {
518 		printk(KERN_ERR "MD: mdesc reread fails with %lu\n",
519 		       status);
520 		refcount_dec(&hp->refcnt);
521 		mdesc_free(hp);
522 		goto out;
523 	}
524 
525 	spin_lock_irqsave(&mdesc_lock, flags);
526 	orig_hp = cur_mdesc;
527 	cur_mdesc = hp;
528 	spin_unlock_irqrestore(&mdesc_lock, flags);
529 
530 	mdesc_notify_clients(orig_hp, hp);
531 
532 	spin_lock_irqsave(&mdesc_lock, flags);
533 	if (refcount_dec_and_test(&orig_hp->refcnt))
534 		mdesc_free(orig_hp);
535 	else
536 		list_add(&orig_hp->list, &mdesc_zombie_list);
537 	spin_unlock_irqrestore(&mdesc_lock, flags);
538 
539 out:
540 	mutex_unlock(&mdesc_mutex);
541 }
542 
543 u64 mdesc_get_node(struct mdesc_handle *hp, const char *node_name,
544 		   union md_node_info *node_info)
545 {
546 	mdesc_node_info_get_f get_info_func;
547 	mdesc_node_info_rel_f rel_info_func;
548 	mdesc_node_match_f node_match_func;
549 	union md_node_info hp_node_info;
550 	u64 hp_node;
551 	int rv;
552 
553 	if (hp == NULL || node_name == NULL || node_info == NULL)
554 		return MDESC_NODE_NULL;
555 
556 	/* Find the ops for the given node name */
557 	mdesc_get_node_ops(node_name, &get_info_func, &rel_info_func,
558 			   &node_match_func);
559 
560 	/* If we didn't find ops for the given node name, it is not supported */
561 	if (!get_info_func || !rel_info_func || !node_match_func) {
562 		pr_err("MD: %s node is not supported\n", node_name);
563 		return -EINVAL;
564 	}
565 
566 	mdesc_for_each_node_by_name(hp, hp_node, node_name) {
567 		rv = get_info_func(hp, hp_node, &hp_node_info);
568 		if (rv != 0)
569 			continue;
570 
571 		if (node_match_func(node_info, &hp_node_info))
572 			break;
573 
574 		rel_info_func(&hp_node_info);
575 	}
576 
577 	rel_info_func(&hp_node_info);
578 
579 	return hp_node;
580 }
581 EXPORT_SYMBOL(mdesc_get_node);
582 
583 int mdesc_get_node_info(struct mdesc_handle *hp, u64 node,
584 			const char *node_name, union md_node_info *node_info)
585 {
586 	mdesc_node_info_get_f get_info_func;
587 	int rv;
588 
589 	if (hp == NULL || node == MDESC_NODE_NULL ||
590 	    node_name == NULL || node_info == NULL)
591 		return -EINVAL;
592 
593 	/* Find the get_info op for the given node name */
594 	mdesc_get_node_ops(node_name, &get_info_func, NULL, NULL);
595 
596 	/* If we didn't find a get_info_func, the node name is not supported */
597 	if (get_info_func == NULL) {
598 		pr_err("MD: %s node is not supported\n", node_name);
599 		return -EINVAL;
600 	}
601 
602 	rv = get_info_func(hp, node, node_info);
603 	if (rv != 0) {
604 		pr_err("MD: Cannot find 1 or more required match properties for %s node.\n",
605 		       node_name);
606 		return -1;
607 	}
608 
609 	return 0;
610 }
611 EXPORT_SYMBOL(mdesc_get_node_info);
612 
613 static struct mdesc_elem *node_block(struct mdesc_hdr *mdesc)
614 {
615 	return (struct mdesc_elem *) (mdesc + 1);
616 }
617 
618 static void *name_block(struct mdesc_hdr *mdesc)
619 {
620 	return ((void *) node_block(mdesc)) + mdesc->node_sz;
621 }
622 
623 static void *data_block(struct mdesc_hdr *mdesc)
624 {
625 	return ((void *) name_block(mdesc)) + mdesc->name_sz;
626 }
627 
628 u64 mdesc_node_by_name(struct mdesc_handle *hp,
629 		       u64 from_node, const char *name)
630 {
631 	struct mdesc_elem *ep = node_block(&hp->mdesc);
632 	const char *names = name_block(&hp->mdesc);
633 	u64 last_node = hp->mdesc.node_sz / 16;
634 	u64 ret;
635 
636 	if (from_node == MDESC_NODE_NULL) {
637 		ret = from_node = 0;
638 	} else if (from_node >= last_node) {
639 		return MDESC_NODE_NULL;
640 	} else {
641 		ret = ep[from_node].d.val;
642 	}
643 
644 	while (ret < last_node) {
645 		if (ep[ret].tag != MD_NODE)
646 			return MDESC_NODE_NULL;
647 		if (!strcmp(names + ep[ret].name_offset, name))
648 			break;
649 		ret = ep[ret].d.val;
650 	}
651 	if (ret >= last_node)
652 		ret = MDESC_NODE_NULL;
653 	return ret;
654 }
655 EXPORT_SYMBOL(mdesc_node_by_name);
656 
657 const void *mdesc_get_property(struct mdesc_handle *hp, u64 node,
658 			       const char *name, int *lenp)
659 {
660 	const char *names = name_block(&hp->mdesc);
661 	u64 last_node = hp->mdesc.node_sz / 16;
662 	void *data = data_block(&hp->mdesc);
663 	struct mdesc_elem *ep;
664 
665 	if (node == MDESC_NODE_NULL || node >= last_node)
666 		return NULL;
667 
668 	ep = node_block(&hp->mdesc) + node;
669 	ep++;
670 	for (; ep->tag != MD_NODE_END; ep++) {
671 		void *val = NULL;
672 		int len = 0;
673 
674 		switch (ep->tag) {
675 		case MD_PROP_VAL:
676 			val = &ep->d.val;
677 			len = 8;
678 			break;
679 
680 		case MD_PROP_STR:
681 		case MD_PROP_DATA:
682 			val = data + ep->d.data.data_offset;
683 			len = ep->d.data.data_len;
684 			break;
685 
686 		default:
687 			break;
688 		}
689 		if (!val)
690 			continue;
691 
692 		if (!strcmp(names + ep->name_offset, name)) {
693 			if (lenp)
694 				*lenp = len;
695 			return val;
696 		}
697 	}
698 
699 	return NULL;
700 }
701 EXPORT_SYMBOL(mdesc_get_property);
702 
703 u64 mdesc_next_arc(struct mdesc_handle *hp, u64 from, const char *arc_type)
704 {
705 	struct mdesc_elem *ep, *base = node_block(&hp->mdesc);
706 	const char *names = name_block(&hp->mdesc);
707 	u64 last_node = hp->mdesc.node_sz / 16;
708 
709 	if (from == MDESC_NODE_NULL || from >= last_node)
710 		return MDESC_NODE_NULL;
711 
712 	ep = base + from;
713 
714 	ep++;
715 	for (; ep->tag != MD_NODE_END; ep++) {
716 		if (ep->tag != MD_PROP_ARC)
717 			continue;
718 
719 		if (strcmp(names + ep->name_offset, arc_type))
720 			continue;
721 
722 		return ep - base;
723 	}
724 
725 	return MDESC_NODE_NULL;
726 }
727 EXPORT_SYMBOL(mdesc_next_arc);
728 
729 u64 mdesc_arc_target(struct mdesc_handle *hp, u64 arc)
730 {
731 	struct mdesc_elem *ep, *base = node_block(&hp->mdesc);
732 
733 	ep = base + arc;
734 
735 	return ep->d.val;
736 }
737 EXPORT_SYMBOL(mdesc_arc_target);
738 
739 const char *mdesc_node_name(struct mdesc_handle *hp, u64 node)
740 {
741 	struct mdesc_elem *ep, *base = node_block(&hp->mdesc);
742 	const char *names = name_block(&hp->mdesc);
743 	u64 last_node = hp->mdesc.node_sz / 16;
744 
745 	if (node == MDESC_NODE_NULL || node >= last_node)
746 		return NULL;
747 
748 	ep = base + node;
749 	if (ep->tag != MD_NODE)
750 		return NULL;
751 
752 	return names + ep->name_offset;
753 }
754 EXPORT_SYMBOL(mdesc_node_name);
755 
756 static u64 max_cpus = 64;
757 
758 static void __init report_platform_properties(void)
759 {
760 	struct mdesc_handle *hp = mdesc_grab();
761 	u64 pn = mdesc_node_by_name(hp, MDESC_NODE_NULL, "platform");
762 	const char *s;
763 	const u64 *v;
764 
765 	if (pn == MDESC_NODE_NULL) {
766 		prom_printf("No platform node in machine-description.\n");
767 		prom_halt();
768 	}
769 
770 	s = mdesc_get_property(hp, pn, "banner-name", NULL);
771 	printk("PLATFORM: banner-name [%s]\n", s);
772 	s = mdesc_get_property(hp, pn, "name", NULL);
773 	printk("PLATFORM: name [%s]\n", s);
774 
775 	v = mdesc_get_property(hp, pn, "hostid", NULL);
776 	if (v)
777 		printk("PLATFORM: hostid [%08llx]\n", *v);
778 	v = mdesc_get_property(hp, pn, "serial#", NULL);
779 	if (v)
780 		printk("PLATFORM: serial# [%08llx]\n", *v);
781 	v = mdesc_get_property(hp, pn, "stick-frequency", NULL);
782 	printk("PLATFORM: stick-frequency [%08llx]\n", *v);
783 	v = mdesc_get_property(hp, pn, "mac-address", NULL);
784 	if (v)
785 		printk("PLATFORM: mac-address [%llx]\n", *v);
786 	v = mdesc_get_property(hp, pn, "watchdog-resolution", NULL);
787 	if (v)
788 		printk("PLATFORM: watchdog-resolution [%llu ms]\n", *v);
789 	v = mdesc_get_property(hp, pn, "watchdog-max-timeout", NULL);
790 	if (v)
791 		printk("PLATFORM: watchdog-max-timeout [%llu ms]\n", *v);
792 	v = mdesc_get_property(hp, pn, "max-cpus", NULL);
793 	if (v) {
794 		max_cpus = *v;
795 		printk("PLATFORM: max-cpus [%llu]\n", max_cpus);
796 	}
797 
798 #ifdef CONFIG_SMP
799 	{
800 		int max_cpu, i;
801 
802 		if (v) {
803 			max_cpu = *v;
804 			if (max_cpu > NR_CPUS)
805 				max_cpu = NR_CPUS;
806 		} else {
807 			max_cpu = NR_CPUS;
808 		}
809 		for (i = 0; i < max_cpu; i++)
810 			set_cpu_possible(i, true);
811 	}
812 #endif
813 
814 	mdesc_release(hp);
815 }
816 
817 static void fill_in_one_cache(cpuinfo_sparc *c, struct mdesc_handle *hp, u64 mp)
818 {
819 	const u64 *level = mdesc_get_property(hp, mp, "level", NULL);
820 	const u64 *size = mdesc_get_property(hp, mp, "size", NULL);
821 	const u64 *line_size = mdesc_get_property(hp, mp, "line-size", NULL);
822 	const char *type;
823 	int type_len;
824 
825 	type = mdesc_get_property(hp, mp, "type", &type_len);
826 
827 	switch (*level) {
828 	case 1:
829 		if (of_find_in_proplist(type, "instn", type_len)) {
830 			c->icache_size = *size;
831 			c->icache_line_size = *line_size;
832 		} else if (of_find_in_proplist(type, "data", type_len)) {
833 			c->dcache_size = *size;
834 			c->dcache_line_size = *line_size;
835 		}
836 		break;
837 
838 	case 2:
839 		c->ecache_size = *size;
840 		c->ecache_line_size = *line_size;
841 		break;
842 
843 	default:
844 		break;
845 	}
846 
847 	if (*level == 1) {
848 		u64 a;
849 
850 		mdesc_for_each_arc(a, hp, mp, MDESC_ARC_TYPE_FWD) {
851 			u64 target = mdesc_arc_target(hp, a);
852 			const char *name = mdesc_node_name(hp, target);
853 
854 			if (!strcmp(name, "cache"))
855 				fill_in_one_cache(c, hp, target);
856 		}
857 	}
858 }
859 
860 static void find_back_node_value(struct mdesc_handle *hp, u64 node,
861 				 char *srch_val,
862 				 void (*func)(struct mdesc_handle *, u64, int),
863 				 u64 val, int depth)
864 {
865 	u64 arc;
866 
867 	/* Since we have an estimate of recursion depth, do a sanity check. */
868 	if (depth == 0)
869 		return;
870 
871 	mdesc_for_each_arc(arc, hp, node, MDESC_ARC_TYPE_BACK) {
872 		u64 n = mdesc_arc_target(hp, arc);
873 		const char *name = mdesc_node_name(hp, n);
874 
875 		if (!strcmp(srch_val, name))
876 			(*func)(hp, n, val);
877 
878 		find_back_node_value(hp, n, srch_val, func, val, depth-1);
879 	}
880 }
881 
882 static void __mark_core_id(struct mdesc_handle *hp, u64 node,
883 			   int core_id)
884 {
885 	const u64 *id = mdesc_get_property(hp, node, "id", NULL);
886 
887 	if (*id < num_possible_cpus())
888 		cpu_data(*id).core_id = core_id;
889 }
890 
891 static void __mark_max_cache_id(struct mdesc_handle *hp, u64 node,
892 				int max_cache_id)
893 {
894 	const u64 *id = mdesc_get_property(hp, node, "id", NULL);
895 
896 	if (*id < num_possible_cpus()) {
897 		cpu_data(*id).max_cache_id = max_cache_id;
898 
899 		/**
900 		 * On systems without explicit socket descriptions socket
901 		 * is max_cache_id
902 		 */
903 		cpu_data(*id).sock_id = max_cache_id;
904 	}
905 }
906 
907 static void mark_core_ids(struct mdesc_handle *hp, u64 mp,
908 			  int core_id)
909 {
910 	find_back_node_value(hp, mp, "cpu", __mark_core_id, core_id, 10);
911 }
912 
913 static void mark_max_cache_ids(struct mdesc_handle *hp, u64 mp,
914 			       int max_cache_id)
915 {
916 	find_back_node_value(hp, mp, "cpu", __mark_max_cache_id,
917 			     max_cache_id, 10);
918 }
919 
920 static void set_core_ids(struct mdesc_handle *hp)
921 {
922 	int idx;
923 	u64 mp;
924 
925 	idx = 1;
926 
927 	/* Identify unique cores by looking for cpus backpointed to by
928 	 * level 1 instruction caches.
929 	 */
930 	mdesc_for_each_node_by_name(hp, mp, "cache") {
931 		const u64 *level;
932 		const char *type;
933 		int len;
934 
935 		level = mdesc_get_property(hp, mp, "level", NULL);
936 		if (*level != 1)
937 			continue;
938 
939 		type = mdesc_get_property(hp, mp, "type", &len);
940 		if (!of_find_in_proplist(type, "instn", len))
941 			continue;
942 
943 		mark_core_ids(hp, mp, idx);
944 		idx++;
945 	}
946 }
947 
948 static int set_max_cache_ids_by_cache(struct mdesc_handle *hp, int level)
949 {
950 	u64 mp;
951 	int idx = 1;
952 	int fnd = 0;
953 
954 	/**
955 	 * Identify unique highest level of shared cache by looking for cpus
956 	 * backpointed to by shared level N caches.
957 	 */
958 	mdesc_for_each_node_by_name(hp, mp, "cache") {
959 		const u64 *cur_lvl;
960 
961 		cur_lvl = mdesc_get_property(hp, mp, "level", NULL);
962 		if (*cur_lvl != level)
963 			continue;
964 		mark_max_cache_ids(hp, mp, idx);
965 		idx++;
966 		fnd = 1;
967 	}
968 	return fnd;
969 }
970 
971 static void set_sock_ids_by_socket(struct mdesc_handle *hp, u64 mp)
972 {
973 	int idx = 1;
974 
975 	mdesc_for_each_node_by_name(hp, mp, "socket") {
976 		u64 a;
977 
978 		mdesc_for_each_arc(a, hp, mp, MDESC_ARC_TYPE_FWD) {
979 			u64 t = mdesc_arc_target(hp, a);
980 			const char *name;
981 			const u64 *id;
982 
983 			name = mdesc_node_name(hp, t);
984 			if (strcmp(name, "cpu"))
985 				continue;
986 
987 			id = mdesc_get_property(hp, t, "id", NULL);
988 			if (*id < num_possible_cpus())
989 				cpu_data(*id).sock_id = idx;
990 		}
991 		idx++;
992 	}
993 }
994 
995 static void set_sock_ids(struct mdesc_handle *hp)
996 {
997 	u64 mp;
998 
999 	/**
1000 	 * Find the highest level of shared cache which pre-T7 is also
1001 	 * the socket.
1002 	 */
1003 	if (!set_max_cache_ids_by_cache(hp, 3))
1004 		set_max_cache_ids_by_cache(hp, 2);
1005 
1006 	/* If machine description exposes sockets data use it.*/
1007 	mp = mdesc_node_by_name(hp, MDESC_NODE_NULL, "sockets");
1008 	if (mp != MDESC_NODE_NULL)
1009 		set_sock_ids_by_socket(hp, mp);
1010 }
1011 
1012 static void mark_proc_ids(struct mdesc_handle *hp, u64 mp, int proc_id)
1013 {
1014 	u64 a;
1015 
1016 	mdesc_for_each_arc(a, hp, mp, MDESC_ARC_TYPE_BACK) {
1017 		u64 t = mdesc_arc_target(hp, a);
1018 		const char *name;
1019 		const u64 *id;
1020 
1021 		name = mdesc_node_name(hp, t);
1022 		if (strcmp(name, "cpu"))
1023 			continue;
1024 
1025 		id = mdesc_get_property(hp, t, "id", NULL);
1026 		if (*id < NR_CPUS)
1027 			cpu_data(*id).proc_id = proc_id;
1028 	}
1029 }
1030 
1031 static void __set_proc_ids(struct mdesc_handle *hp, const char *exec_unit_name)
1032 {
1033 	int idx;
1034 	u64 mp;
1035 
1036 	idx = 0;
1037 	mdesc_for_each_node_by_name(hp, mp, exec_unit_name) {
1038 		const char *type;
1039 		int len;
1040 
1041 		type = mdesc_get_property(hp, mp, "type", &len);
1042 		if (!of_find_in_proplist(type, "int", len) &&
1043 		    !of_find_in_proplist(type, "integer", len))
1044 			continue;
1045 
1046 		mark_proc_ids(hp, mp, idx);
1047 		idx++;
1048 	}
1049 }
1050 
1051 static void set_proc_ids(struct mdesc_handle *hp)
1052 {
1053 	__set_proc_ids(hp, "exec_unit");
1054 	__set_proc_ids(hp, "exec-unit");
1055 }
1056 
1057 static void get_one_mondo_bits(const u64 *p, unsigned int *mask,
1058 			       unsigned long def, unsigned long max)
1059 {
1060 	u64 val;
1061 
1062 	if (!p)
1063 		goto use_default;
1064 	val = *p;
1065 
1066 	if (!val || val >= 64)
1067 		goto use_default;
1068 
1069 	if (val > max)
1070 		val = max;
1071 
1072 	*mask = ((1U << val) * 64U) - 1U;
1073 	return;
1074 
1075 use_default:
1076 	*mask = ((1U << def) * 64U) - 1U;
1077 }
1078 
1079 static void get_mondo_data(struct mdesc_handle *hp, u64 mp,
1080 			   struct trap_per_cpu *tb)
1081 {
1082 	static int printed;
1083 	const u64 *val;
1084 
1085 	val = mdesc_get_property(hp, mp, "q-cpu-mondo-#bits", NULL);
1086 	get_one_mondo_bits(val, &tb->cpu_mondo_qmask, 7, ilog2(max_cpus * 2));
1087 
1088 	val = mdesc_get_property(hp, mp, "q-dev-mondo-#bits", NULL);
1089 	get_one_mondo_bits(val, &tb->dev_mondo_qmask, 7, 8);
1090 
1091 	val = mdesc_get_property(hp, mp, "q-resumable-#bits", NULL);
1092 	get_one_mondo_bits(val, &tb->resum_qmask, 6, 7);
1093 
1094 	val = mdesc_get_property(hp, mp, "q-nonresumable-#bits", NULL);
1095 	get_one_mondo_bits(val, &tb->nonresum_qmask, 2, 2);
1096 	if (!printed++) {
1097 		pr_info("SUN4V: Mondo queue sizes "
1098 			"[cpu(%u) dev(%u) r(%u) nr(%u)]\n",
1099 			tb->cpu_mondo_qmask + 1,
1100 			tb->dev_mondo_qmask + 1,
1101 			tb->resum_qmask + 1,
1102 			tb->nonresum_qmask + 1);
1103 	}
1104 }
1105 
1106 static void *mdesc_iterate_over_cpus(void *(*func)(struct mdesc_handle *, u64, int, void *), void *arg, cpumask_t *mask)
1107 {
1108 	struct mdesc_handle *hp = mdesc_grab();
1109 	void *ret = NULL;
1110 	u64 mp;
1111 
1112 	mdesc_for_each_node_by_name(hp, mp, "cpu") {
1113 		const u64 *id = mdesc_get_property(hp, mp, "id", NULL);
1114 		int cpuid = *id;
1115 
1116 #ifdef CONFIG_SMP
1117 		if (cpuid >= NR_CPUS) {
1118 			printk(KERN_WARNING "Ignoring CPU %d which is "
1119 			       ">= NR_CPUS (%d)\n",
1120 			       cpuid, NR_CPUS);
1121 			continue;
1122 		}
1123 		if (!cpumask_test_cpu(cpuid, mask))
1124 			continue;
1125 #endif
1126 
1127 		ret = func(hp, mp, cpuid, arg);
1128 		if (ret)
1129 			goto out;
1130 	}
1131 out:
1132 	mdesc_release(hp);
1133 	return ret;
1134 }
1135 
1136 static void *record_one_cpu(struct mdesc_handle *hp, u64 mp, int cpuid,
1137 			    void *arg)
1138 {
1139 	ncpus_probed++;
1140 #ifdef CONFIG_SMP
1141 	set_cpu_present(cpuid, true);
1142 #endif
1143 	return NULL;
1144 }
1145 
1146 void mdesc_populate_present_mask(cpumask_t *mask)
1147 {
1148 	if (tlb_type != hypervisor)
1149 		return;
1150 
1151 	ncpus_probed = 0;
1152 	mdesc_iterate_over_cpus(record_one_cpu, NULL, mask);
1153 }
1154 
1155 static void * __init check_one_pgsz(struct mdesc_handle *hp, u64 mp, int cpuid, void *arg)
1156 {
1157 	const u64 *pgsz_prop = mdesc_get_property(hp, mp, "mmu-page-size-list", NULL);
1158 	unsigned long *pgsz_mask = arg;
1159 	u64 val;
1160 
1161 	val = (HV_PGSZ_MASK_8K | HV_PGSZ_MASK_64K |
1162 	       HV_PGSZ_MASK_512K | HV_PGSZ_MASK_4MB);
1163 	if (pgsz_prop)
1164 		val = *pgsz_prop;
1165 
1166 	if (!*pgsz_mask)
1167 		*pgsz_mask = val;
1168 	else
1169 		*pgsz_mask &= val;
1170 	return NULL;
1171 }
1172 
1173 void __init mdesc_get_page_sizes(cpumask_t *mask, unsigned long *pgsz_mask)
1174 {
1175 	*pgsz_mask = 0;
1176 	mdesc_iterate_over_cpus(check_one_pgsz, pgsz_mask, mask);
1177 }
1178 
1179 static void *fill_in_one_cpu(struct mdesc_handle *hp, u64 mp, int cpuid,
1180 			     void *arg)
1181 {
1182 	const u64 *cfreq = mdesc_get_property(hp, mp, "clock-frequency", NULL);
1183 	struct trap_per_cpu *tb;
1184 	cpuinfo_sparc *c;
1185 	u64 a;
1186 
1187 #ifndef CONFIG_SMP
1188 	/* On uniprocessor we only want the values for the
1189 	 * real physical cpu the kernel booted onto, however
1190 	 * cpu_data() only has one entry at index 0.
1191 	 */
1192 	if (cpuid != real_hard_smp_processor_id())
1193 		return NULL;
1194 	cpuid = 0;
1195 #endif
1196 
1197 	c = &cpu_data(cpuid);
1198 	c->clock_tick = *cfreq;
1199 
1200 	tb = &trap_block[cpuid];
1201 	get_mondo_data(hp, mp, tb);
1202 
1203 	mdesc_for_each_arc(a, hp, mp, MDESC_ARC_TYPE_FWD) {
1204 		u64 j, t = mdesc_arc_target(hp, a);
1205 		const char *t_name;
1206 
1207 		t_name = mdesc_node_name(hp, t);
1208 		if (!strcmp(t_name, "cache")) {
1209 			fill_in_one_cache(c, hp, t);
1210 			continue;
1211 		}
1212 
1213 		mdesc_for_each_arc(j, hp, t, MDESC_ARC_TYPE_FWD) {
1214 			u64 n = mdesc_arc_target(hp, j);
1215 			const char *n_name;
1216 
1217 			n_name = mdesc_node_name(hp, n);
1218 			if (!strcmp(n_name, "cache"))
1219 				fill_in_one_cache(c, hp, n);
1220 		}
1221 	}
1222 
1223 	c->core_id = 0;
1224 	c->proc_id = -1;
1225 
1226 	return NULL;
1227 }
1228 
1229 void mdesc_fill_in_cpu_data(cpumask_t *mask)
1230 {
1231 	struct mdesc_handle *hp;
1232 
1233 	mdesc_iterate_over_cpus(fill_in_one_cpu, NULL, mask);
1234 
1235 	hp = mdesc_grab();
1236 
1237 	set_core_ids(hp);
1238 	set_proc_ids(hp);
1239 	set_sock_ids(hp);
1240 
1241 	mdesc_release(hp);
1242 
1243 	smp_fill_in_sib_core_maps();
1244 }
1245 
1246 /* mdesc_open() - Grab a reference to mdesc_handle when /dev/mdesc is
1247  * opened. Hold this reference until /dev/mdesc is closed to ensure
1248  * mdesc data structure is not released underneath us. Store the
1249  * pointer to mdesc structure in private_data for read and seek to use
1250  */
1251 static int mdesc_open(struct inode *inode, struct file *file)
1252 {
1253 	struct mdesc_handle *hp = mdesc_grab();
1254 
1255 	if (!hp)
1256 		return -ENODEV;
1257 
1258 	file->private_data = hp;
1259 
1260 	return 0;
1261 }
1262 
1263 static ssize_t mdesc_read(struct file *file, char __user *buf,
1264 			  size_t len, loff_t *offp)
1265 {
1266 	struct mdesc_handle *hp = file->private_data;
1267 	unsigned char *mdesc;
1268 	int bytes_left, count = len;
1269 
1270 	if (*offp >= hp->handle_size)
1271 		return 0;
1272 
1273 	bytes_left = hp->handle_size - *offp;
1274 	if (count > bytes_left)
1275 		count = bytes_left;
1276 
1277 	mdesc = (unsigned char *)&hp->mdesc;
1278 	mdesc += *offp;
1279 	if (!copy_to_user(buf, mdesc, count)) {
1280 		*offp += count;
1281 		return count;
1282 	} else {
1283 		return -EFAULT;
1284 	}
1285 }
1286 
1287 static loff_t mdesc_llseek(struct file *file, loff_t offset, int whence)
1288 {
1289 	struct mdesc_handle *hp = file->private_data;
1290 
1291 	return no_seek_end_llseek_size(file, offset, whence, hp->handle_size);
1292 }
1293 
1294 /* mdesc_close() - /dev/mdesc is being closed, release the reference to
1295  * mdesc structure.
1296  */
1297 static int mdesc_close(struct inode *inode, struct file *file)
1298 {
1299 	mdesc_release(file->private_data);
1300 	return 0;
1301 }
1302 
1303 static const struct file_operations mdesc_fops = {
1304 	.open    = mdesc_open,
1305 	.read	 = mdesc_read,
1306 	.llseek  = mdesc_llseek,
1307 	.release = mdesc_close,
1308 	.owner	 = THIS_MODULE,
1309 };
1310 
1311 static struct miscdevice mdesc_misc = {
1312 	.minor	= MISC_DYNAMIC_MINOR,
1313 	.name	= "mdesc",
1314 	.fops	= &mdesc_fops,
1315 };
1316 
1317 static int __init mdesc_misc_init(void)
1318 {
1319 	return misc_register(&mdesc_misc);
1320 }
1321 
1322 __initcall(mdesc_misc_init);
1323 
1324 void __init sun4v_mdesc_init(void)
1325 {
1326 	struct mdesc_handle *hp;
1327 	unsigned long len, real_len, status;
1328 
1329 	(void) sun4v_mach_desc(0UL, 0UL, &len);
1330 
1331 	printk("MDESC: Size is %lu bytes.\n", len);
1332 
1333 	hp = mdesc_alloc(len, &memblock_mdesc_ops);
1334 	if (hp == NULL) {
1335 		prom_printf("MDESC: alloc of %lu bytes failed.\n", len);
1336 		prom_halt();
1337 	}
1338 
1339 	status = sun4v_mach_desc(__pa(&hp->mdesc), len, &real_len);
1340 	if (status != HV_EOK || real_len > len) {
1341 		prom_printf("sun4v_mach_desc fails, err(%lu), "
1342 			    "len(%lu), real_len(%lu)\n",
1343 			    status, len, real_len);
1344 		mdesc_free(hp);
1345 		prom_halt();
1346 	}
1347 
1348 	cur_mdesc = hp;
1349 
1350 	mdesc_adi_init();
1351 	report_platform_properties();
1352 }
1353