xref: /openbmc/linux/arch/sparc/kernel/mdesc.c (revision 1edd0337)
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 	char	data[];
43 } __attribute__((aligned(16)));
44 
45 struct mdesc_elem {
46 	u8	tag;
47 #define MD_LIST_END	0x00
48 #define MD_NODE		0x4e
49 #define MD_NODE_END	0x45
50 #define MD_NOOP		0x20
51 #define MD_PROP_ARC	0x61
52 #define MD_PROP_VAL	0x76
53 #define MD_PROP_STR	0x73
54 #define MD_PROP_DATA	0x64
55 	u8	name_len;
56 	u16	resv;
57 	u32	name_offset;
58 	union {
59 		struct {
60 			u32	data_len;
61 			u32	data_offset;
62 		} data;
63 		u64	val;
64 	} d;
65 };
66 
67 struct mdesc_mem_ops {
68 	struct mdesc_handle *(*alloc)(unsigned int mdesc_size);
69 	void (*free)(struct mdesc_handle *handle);
70 };
71 
72 struct mdesc_handle {
73 	struct list_head	list;
74 	struct mdesc_mem_ops	*mops;
75 	void			*self_base;
76 	refcount_t		refcnt;
77 	unsigned int		handle_size;
78 	struct mdesc_hdr	mdesc;
79 };
80 
81 typedef int (*mdesc_node_info_get_f)(struct mdesc_handle *, u64,
82 				     union md_node_info *);
83 typedef void (*mdesc_node_info_rel_f)(union md_node_info *);
84 typedef bool (*mdesc_node_match_f)(union md_node_info *, union md_node_info *);
85 
86 struct md_node_ops {
87 	char			*name;
88 	mdesc_node_info_get_f	get_info;
89 	mdesc_node_info_rel_f	rel_info;
90 	mdesc_node_match_f	node_match;
91 };
92 
93 static int get_vdev_port_node_info(struct mdesc_handle *md, u64 node,
94 				   union md_node_info *node_info);
95 static void rel_vdev_port_node_info(union md_node_info *node_info);
96 static bool vdev_port_node_match(union md_node_info *a_node_info,
97 				 union md_node_info *b_node_info);
98 
99 static int get_ds_port_node_info(struct mdesc_handle *md, u64 node,
100 				 union md_node_info *node_info);
101 static void rel_ds_port_node_info(union md_node_info *node_info);
102 static bool ds_port_node_match(union md_node_info *a_node_info,
103 			       union md_node_info *b_node_info);
104 
105 /* supported node types which can be registered */
106 static struct md_node_ops md_node_ops_table[] = {
107 	{"virtual-device-port", get_vdev_port_node_info,
108 	 rel_vdev_port_node_info, vdev_port_node_match},
109 	{"domain-services-port", get_ds_port_node_info,
110 	 rel_ds_port_node_info, ds_port_node_match},
111 	{NULL, NULL, NULL, NULL}
112 };
113 
114 static void mdesc_get_node_ops(const char *node_name,
115 			       mdesc_node_info_get_f *get_info_f,
116 			       mdesc_node_info_rel_f *rel_info_f,
117 			       mdesc_node_match_f *match_f)
118 {
119 	int i;
120 
121 	if (get_info_f)
122 		*get_info_f = NULL;
123 
124 	if (rel_info_f)
125 		*rel_info_f = NULL;
126 
127 	if (match_f)
128 		*match_f = NULL;
129 
130 	if (!node_name)
131 		return;
132 
133 	for (i = 0; md_node_ops_table[i].name != NULL; i++) {
134 		if (strcmp(md_node_ops_table[i].name, node_name) == 0) {
135 			if (get_info_f)
136 				*get_info_f = md_node_ops_table[i].get_info;
137 
138 			if (rel_info_f)
139 				*rel_info_f = md_node_ops_table[i].rel_info;
140 
141 			if (match_f)
142 				*match_f = md_node_ops_table[i].node_match;
143 
144 			break;
145 		}
146 	}
147 }
148 
149 static void mdesc_handle_init(struct mdesc_handle *hp,
150 			      unsigned int handle_size,
151 			      void *base)
152 {
153 	BUG_ON(((unsigned long)&hp->mdesc) & (16UL - 1));
154 
155 	memset(hp, 0, handle_size);
156 	INIT_LIST_HEAD(&hp->list);
157 	hp->self_base = base;
158 	refcount_set(&hp->refcnt, 1);
159 	hp->handle_size = handle_size;
160 }
161 
162 static struct mdesc_handle * __init mdesc_memblock_alloc(unsigned int mdesc_size)
163 {
164 	unsigned int handle_size, alloc_size;
165 	struct mdesc_handle *hp;
166 	unsigned long paddr;
167 
168 	handle_size = (sizeof(struct mdesc_handle) -
169 		       sizeof(struct mdesc_hdr) +
170 		       mdesc_size);
171 	alloc_size = PAGE_ALIGN(handle_size);
172 
173 	paddr = memblock_phys_alloc(alloc_size, PAGE_SIZE);
174 
175 	hp = NULL;
176 	if (paddr) {
177 		hp = __va(paddr);
178 		mdesc_handle_init(hp, handle_size, hp);
179 	}
180 	return hp;
181 }
182 
183 static void __init mdesc_memblock_free(struct mdesc_handle *hp)
184 {
185 	unsigned int alloc_size;
186 	unsigned long start;
187 
188 	BUG_ON(refcount_read(&hp->refcnt) != 0);
189 	BUG_ON(!list_empty(&hp->list));
190 
191 	alloc_size = PAGE_ALIGN(hp->handle_size);
192 	start = __pa(hp);
193 	memblock_free_late(start, alloc_size);
194 }
195 
196 static struct mdesc_mem_ops memblock_mdesc_ops = {
197 	.alloc = mdesc_memblock_alloc,
198 	.free  = mdesc_memblock_free,
199 };
200 
201 static struct mdesc_handle *mdesc_kmalloc(unsigned int mdesc_size)
202 {
203 	unsigned int handle_size;
204 	struct mdesc_handle *hp;
205 	unsigned long addr;
206 	void *base;
207 
208 	handle_size = (sizeof(struct mdesc_handle) -
209 		       sizeof(struct mdesc_hdr) +
210 		       mdesc_size);
211 	base = kmalloc(handle_size + 15, GFP_KERNEL | __GFP_RETRY_MAYFAIL);
212 	if (!base)
213 		return NULL;
214 
215 	addr = (unsigned long)base;
216 	addr = (addr + 15UL) & ~15UL;
217 	hp = (struct mdesc_handle *) addr;
218 
219 	mdesc_handle_init(hp, handle_size, base);
220 
221 	return hp;
222 }
223 
224 static void mdesc_kfree(struct mdesc_handle *hp)
225 {
226 	BUG_ON(refcount_read(&hp->refcnt) != 0);
227 	BUG_ON(!list_empty(&hp->list));
228 
229 	kfree(hp->self_base);
230 }
231 
232 static struct mdesc_mem_ops kmalloc_mdesc_memops = {
233 	.alloc = mdesc_kmalloc,
234 	.free  = mdesc_kfree,
235 };
236 
237 static struct mdesc_handle *mdesc_alloc(unsigned int mdesc_size,
238 					struct mdesc_mem_ops *mops)
239 {
240 	struct mdesc_handle *hp = mops->alloc(mdesc_size);
241 
242 	if (hp)
243 		hp->mops = mops;
244 
245 	return hp;
246 }
247 
248 static void mdesc_free(struct mdesc_handle *hp)
249 {
250 	hp->mops->free(hp);
251 }
252 
253 static struct mdesc_handle *cur_mdesc;
254 static LIST_HEAD(mdesc_zombie_list);
255 static DEFINE_SPINLOCK(mdesc_lock);
256 
257 struct mdesc_handle *mdesc_grab(void)
258 {
259 	struct mdesc_handle *hp;
260 	unsigned long flags;
261 
262 	spin_lock_irqsave(&mdesc_lock, flags);
263 	hp = cur_mdesc;
264 	if (hp)
265 		refcount_inc(&hp->refcnt);
266 	spin_unlock_irqrestore(&mdesc_lock, flags);
267 
268 	return hp;
269 }
270 EXPORT_SYMBOL(mdesc_grab);
271 
272 void mdesc_release(struct mdesc_handle *hp)
273 {
274 	unsigned long flags;
275 
276 	spin_lock_irqsave(&mdesc_lock, flags);
277 	if (refcount_dec_and_test(&hp->refcnt)) {
278 		list_del_init(&hp->list);
279 		hp->mops->free(hp);
280 	}
281 	spin_unlock_irqrestore(&mdesc_lock, flags);
282 }
283 EXPORT_SYMBOL(mdesc_release);
284 
285 static DEFINE_MUTEX(mdesc_mutex);
286 static struct mdesc_notifier_client *client_list;
287 
288 void mdesc_register_notifier(struct mdesc_notifier_client *client)
289 {
290 	bool supported = false;
291 	u64 node;
292 	int i;
293 
294 	mutex_lock(&mdesc_mutex);
295 
296 	/* check to see if the node is supported for registration */
297 	for (i = 0; md_node_ops_table[i].name != NULL; i++) {
298 		if (strcmp(md_node_ops_table[i].name, client->node_name) == 0) {
299 			supported = true;
300 			break;
301 		}
302 	}
303 
304 	if (!supported) {
305 		pr_err("MD: %s node not supported\n", client->node_name);
306 		mutex_unlock(&mdesc_mutex);
307 		return;
308 	}
309 
310 	client->next = client_list;
311 	client_list = client;
312 
313 	mdesc_for_each_node_by_name(cur_mdesc, node, client->node_name)
314 		client->add(cur_mdesc, node, client->node_name);
315 
316 	mutex_unlock(&mdesc_mutex);
317 }
318 
319 static const u64 *parent_cfg_handle(struct mdesc_handle *hp, u64 node)
320 {
321 	const u64 *id;
322 	u64 a;
323 
324 	id = NULL;
325 	mdesc_for_each_arc(a, hp, node, MDESC_ARC_TYPE_BACK) {
326 		u64 target;
327 
328 		target = mdesc_arc_target(hp, a);
329 		id = mdesc_get_property(hp, target,
330 					"cfg-handle", NULL);
331 		if (id)
332 			break;
333 	}
334 
335 	return id;
336 }
337 
338 static int get_vdev_port_node_info(struct mdesc_handle *md, u64 node,
339 				   union md_node_info *node_info)
340 {
341 	const u64 *parent_cfg_hdlp;
342 	const char *name;
343 	const u64 *idp;
344 
345 	/*
346 	 * Virtual device nodes are distinguished by:
347 	 * 1. "id" property
348 	 * 2. "name" property
349 	 * 3. parent node "cfg-handle" property
350 	 */
351 	idp = mdesc_get_property(md, node, "id", NULL);
352 	name = mdesc_get_property(md, node, "name", NULL);
353 	parent_cfg_hdlp = parent_cfg_handle(md, node);
354 
355 	if (!idp || !name || !parent_cfg_hdlp)
356 		return -1;
357 
358 	node_info->vdev_port.id = *idp;
359 	node_info->vdev_port.name = kstrdup_const(name, GFP_KERNEL);
360 	if (!node_info->vdev_port.name)
361 		return -1;
362 	node_info->vdev_port.parent_cfg_hdl = *parent_cfg_hdlp;
363 
364 	return 0;
365 }
366 
367 static void rel_vdev_port_node_info(union md_node_info *node_info)
368 {
369 	if (node_info && node_info->vdev_port.name) {
370 		kfree_const(node_info->vdev_port.name);
371 		node_info->vdev_port.name = NULL;
372 	}
373 }
374 
375 static bool vdev_port_node_match(union md_node_info *a_node_info,
376 				 union md_node_info *b_node_info)
377 {
378 	if (a_node_info->vdev_port.id != b_node_info->vdev_port.id)
379 		return false;
380 
381 	if (a_node_info->vdev_port.parent_cfg_hdl !=
382 	    b_node_info->vdev_port.parent_cfg_hdl)
383 		return false;
384 
385 	if (strncmp(a_node_info->vdev_port.name,
386 		    b_node_info->vdev_port.name, MDESC_MAX_STR_LEN) != 0)
387 		return false;
388 
389 	return true;
390 }
391 
392 static int get_ds_port_node_info(struct mdesc_handle *md, u64 node,
393 				 union md_node_info *node_info)
394 {
395 	const u64 *idp;
396 
397 	/* DS port nodes use the "id" property to distinguish them */
398 	idp = mdesc_get_property(md, node, "id", NULL);
399 	if (!idp)
400 		return -1;
401 
402 	node_info->ds_port.id = *idp;
403 
404 	return 0;
405 }
406 
407 static void rel_ds_port_node_info(union md_node_info *node_info)
408 {
409 }
410 
411 static bool ds_port_node_match(union md_node_info *a_node_info,
412 			       union md_node_info *b_node_info)
413 {
414 	if (a_node_info->ds_port.id != b_node_info->ds_port.id)
415 		return false;
416 
417 	return true;
418 }
419 
420 /* Run 'func' on nodes which are in A but not in B.  */
421 static void invoke_on_missing(const char *name,
422 			      struct mdesc_handle *a,
423 			      struct mdesc_handle *b,
424 			      void (*func)(struct mdesc_handle *, u64,
425 					   const char *node_name))
426 {
427 	mdesc_node_info_get_f get_info_func;
428 	mdesc_node_info_rel_f rel_info_func;
429 	mdesc_node_match_f node_match_func;
430 	union md_node_info a_node_info;
431 	union md_node_info b_node_info;
432 	bool found;
433 	u64 a_node;
434 	u64 b_node;
435 	int rv;
436 
437 	/*
438 	 * Find the get_info, rel_info and node_match ops for the given
439 	 * node name
440 	 */
441 	mdesc_get_node_ops(name, &get_info_func, &rel_info_func,
442 			   &node_match_func);
443 
444 	/* If we didn't find a match, the node type is not supported */
445 	if (!get_info_func || !rel_info_func || !node_match_func) {
446 		pr_err("MD: %s node type is not supported\n", name);
447 		return;
448 	}
449 
450 	mdesc_for_each_node_by_name(a, a_node, name) {
451 		found = false;
452 
453 		rv = get_info_func(a, a_node, &a_node_info);
454 		if (rv != 0) {
455 			pr_err("MD: Cannot find 1 or more required match properties for %s node.\n",
456 			       name);
457 			continue;
458 		}
459 
460 		/* Check each node in B for node matching a_node */
461 		mdesc_for_each_node_by_name(b, b_node, name) {
462 			rv = get_info_func(b, b_node, &b_node_info);
463 			if (rv != 0)
464 				continue;
465 
466 			if (node_match_func(&a_node_info, &b_node_info)) {
467 				found = true;
468 				rel_info_func(&b_node_info);
469 				break;
470 			}
471 
472 			rel_info_func(&b_node_info);
473 		}
474 
475 		rel_info_func(&a_node_info);
476 
477 		if (!found)
478 			func(a, a_node, name);
479 	}
480 }
481 
482 static void notify_one(struct mdesc_notifier_client *p,
483 		       struct mdesc_handle *old_hp,
484 		       struct mdesc_handle *new_hp)
485 {
486 	invoke_on_missing(p->node_name, old_hp, new_hp, p->remove);
487 	invoke_on_missing(p->node_name, new_hp, old_hp, p->add);
488 }
489 
490 static void mdesc_notify_clients(struct mdesc_handle *old_hp,
491 				 struct mdesc_handle *new_hp)
492 {
493 	struct mdesc_notifier_client *p = client_list;
494 
495 	while (p) {
496 		notify_one(p, old_hp, new_hp);
497 		p = p->next;
498 	}
499 }
500 
501 void mdesc_update(void)
502 {
503 	unsigned long len, real_len, status;
504 	struct mdesc_handle *hp, *orig_hp;
505 	unsigned long flags;
506 
507 	mutex_lock(&mdesc_mutex);
508 
509 	(void) sun4v_mach_desc(0UL, 0UL, &len);
510 
511 	hp = mdesc_alloc(len, &kmalloc_mdesc_memops);
512 	if (!hp) {
513 		printk(KERN_ERR "MD: mdesc alloc fails\n");
514 		goto out;
515 	}
516 
517 	status = sun4v_mach_desc(__pa(&hp->mdesc), len, &real_len);
518 	if (status != HV_EOK || real_len > len) {
519 		printk(KERN_ERR "MD: mdesc reread fails with %lu\n",
520 		       status);
521 		refcount_dec(&hp->refcnt);
522 		mdesc_free(hp);
523 		goto out;
524 	}
525 
526 	spin_lock_irqsave(&mdesc_lock, flags);
527 	orig_hp = cur_mdesc;
528 	cur_mdesc = hp;
529 	spin_unlock_irqrestore(&mdesc_lock, flags);
530 
531 	mdesc_notify_clients(orig_hp, hp);
532 
533 	spin_lock_irqsave(&mdesc_lock, flags);
534 	if (refcount_dec_and_test(&orig_hp->refcnt))
535 		mdesc_free(orig_hp);
536 	else
537 		list_add(&orig_hp->list, &mdesc_zombie_list);
538 	spin_unlock_irqrestore(&mdesc_lock, flags);
539 
540 out:
541 	mutex_unlock(&mdesc_mutex);
542 }
543 
544 u64 mdesc_get_node(struct mdesc_handle *hp, const char *node_name,
545 		   union md_node_info *node_info)
546 {
547 	mdesc_node_info_get_f get_info_func;
548 	mdesc_node_info_rel_f rel_info_func;
549 	mdesc_node_match_f node_match_func;
550 	union md_node_info hp_node_info;
551 	u64 hp_node;
552 	int rv;
553 
554 	if (hp == NULL || node_name == NULL || node_info == NULL)
555 		return MDESC_NODE_NULL;
556 
557 	/* Find the ops for the given node name */
558 	mdesc_get_node_ops(node_name, &get_info_func, &rel_info_func,
559 			   &node_match_func);
560 
561 	/* If we didn't find ops for the given node name, it is not supported */
562 	if (!get_info_func || !rel_info_func || !node_match_func) {
563 		pr_err("MD: %s node is not supported\n", node_name);
564 		return -EINVAL;
565 	}
566 
567 	mdesc_for_each_node_by_name(hp, hp_node, node_name) {
568 		rv = get_info_func(hp, hp_node, &hp_node_info);
569 		if (rv != 0)
570 			continue;
571 
572 		if (node_match_func(node_info, &hp_node_info))
573 			break;
574 
575 		rel_info_func(&hp_node_info);
576 	}
577 
578 	rel_info_func(&hp_node_info);
579 
580 	return hp_node;
581 }
582 EXPORT_SYMBOL(mdesc_get_node);
583 
584 int mdesc_get_node_info(struct mdesc_handle *hp, u64 node,
585 			const char *node_name, union md_node_info *node_info)
586 {
587 	mdesc_node_info_get_f get_info_func;
588 	int rv;
589 
590 	if (hp == NULL || node == MDESC_NODE_NULL ||
591 	    node_name == NULL || node_info == NULL)
592 		return -EINVAL;
593 
594 	/* Find the get_info op for the given node name */
595 	mdesc_get_node_ops(node_name, &get_info_func, NULL, NULL);
596 
597 	/* If we didn't find a get_info_func, the node name is not supported */
598 	if (get_info_func == NULL) {
599 		pr_err("MD: %s node is not supported\n", node_name);
600 		return -EINVAL;
601 	}
602 
603 	rv = get_info_func(hp, node, node_info);
604 	if (rv != 0) {
605 		pr_err("MD: Cannot find 1 or more required match properties for %s node.\n",
606 		       node_name);
607 		return -1;
608 	}
609 
610 	return 0;
611 }
612 EXPORT_SYMBOL(mdesc_get_node_info);
613 
614 static struct mdesc_elem *node_block(struct mdesc_hdr *mdesc)
615 {
616 	return (struct mdesc_elem *) mdesc->data;
617 }
618 
619 static void *name_block(struct mdesc_hdr *mdesc)
620 {
621 	return ((void *) node_block(mdesc)) + mdesc->node_sz;
622 }
623 
624 static void *data_block(struct mdesc_hdr *mdesc)
625 {
626 	return ((void *) name_block(mdesc)) + mdesc->name_sz;
627 }
628 
629 u64 mdesc_node_by_name(struct mdesc_handle *hp,
630 		       u64 from_node, const char *name)
631 {
632 	struct mdesc_elem *ep = node_block(&hp->mdesc);
633 	const char *names = name_block(&hp->mdesc);
634 	u64 last_node = hp->mdesc.node_sz / 16;
635 	u64 ret;
636 
637 	if (from_node == MDESC_NODE_NULL) {
638 		ret = from_node = 0;
639 	} else if (from_node >= last_node) {
640 		return MDESC_NODE_NULL;
641 	} else {
642 		ret = ep[from_node].d.val;
643 	}
644 
645 	while (ret < last_node) {
646 		if (ep[ret].tag != MD_NODE)
647 			return MDESC_NODE_NULL;
648 		if (!strcmp(names + ep[ret].name_offset, name))
649 			break;
650 		ret = ep[ret].d.val;
651 	}
652 	if (ret >= last_node)
653 		ret = MDESC_NODE_NULL;
654 	return ret;
655 }
656 EXPORT_SYMBOL(mdesc_node_by_name);
657 
658 const void *mdesc_get_property(struct mdesc_handle *hp, u64 node,
659 			       const char *name, int *lenp)
660 {
661 	const char *names = name_block(&hp->mdesc);
662 	u64 last_node = hp->mdesc.node_sz / 16;
663 	void *data = data_block(&hp->mdesc);
664 	struct mdesc_elem *ep;
665 
666 	if (node == MDESC_NODE_NULL || node >= last_node)
667 		return NULL;
668 
669 	ep = node_block(&hp->mdesc) + node;
670 	ep++;
671 	for (; ep->tag != MD_NODE_END; ep++) {
672 		void *val = NULL;
673 		int len = 0;
674 
675 		switch (ep->tag) {
676 		case MD_PROP_VAL:
677 			val = &ep->d.val;
678 			len = 8;
679 			break;
680 
681 		case MD_PROP_STR:
682 		case MD_PROP_DATA:
683 			val = data + ep->d.data.data_offset;
684 			len = ep->d.data.data_len;
685 			break;
686 
687 		default:
688 			break;
689 		}
690 		if (!val)
691 			continue;
692 
693 		if (!strcmp(names + ep->name_offset, name)) {
694 			if (lenp)
695 				*lenp = len;
696 			return val;
697 		}
698 	}
699 
700 	return NULL;
701 }
702 EXPORT_SYMBOL(mdesc_get_property);
703 
704 u64 mdesc_next_arc(struct mdesc_handle *hp, u64 from, const char *arc_type)
705 {
706 	struct mdesc_elem *ep, *base = node_block(&hp->mdesc);
707 	const char *names = name_block(&hp->mdesc);
708 	u64 last_node = hp->mdesc.node_sz / 16;
709 
710 	if (from == MDESC_NODE_NULL || from >= last_node)
711 		return MDESC_NODE_NULL;
712 
713 	ep = base + from;
714 
715 	ep++;
716 	for (; ep->tag != MD_NODE_END; ep++) {
717 		if (ep->tag != MD_PROP_ARC)
718 			continue;
719 
720 		if (strcmp(names + ep->name_offset, arc_type))
721 			continue;
722 
723 		return ep - base;
724 	}
725 
726 	return MDESC_NODE_NULL;
727 }
728 EXPORT_SYMBOL(mdesc_next_arc);
729 
730 u64 mdesc_arc_target(struct mdesc_handle *hp, u64 arc)
731 {
732 	struct mdesc_elem *ep, *base = node_block(&hp->mdesc);
733 
734 	ep = base + arc;
735 
736 	return ep->d.val;
737 }
738 EXPORT_SYMBOL(mdesc_arc_target);
739 
740 const char *mdesc_node_name(struct mdesc_handle *hp, u64 node)
741 {
742 	struct mdesc_elem *ep, *base = node_block(&hp->mdesc);
743 	const char *names = name_block(&hp->mdesc);
744 	u64 last_node = hp->mdesc.node_sz / 16;
745 
746 	if (node == MDESC_NODE_NULL || node >= last_node)
747 		return NULL;
748 
749 	ep = base + node;
750 	if (ep->tag != MD_NODE)
751 		return NULL;
752 
753 	return names + ep->name_offset;
754 }
755 EXPORT_SYMBOL(mdesc_node_name);
756 
757 static u64 max_cpus = 64;
758 
759 static void __init report_platform_properties(void)
760 {
761 	struct mdesc_handle *hp = mdesc_grab();
762 	u64 pn = mdesc_node_by_name(hp, MDESC_NODE_NULL, "platform");
763 	const char *s;
764 	const u64 *v;
765 
766 	if (pn == MDESC_NODE_NULL) {
767 		prom_printf("No platform node in machine-description.\n");
768 		prom_halt();
769 	}
770 
771 	s = mdesc_get_property(hp, pn, "banner-name", NULL);
772 	printk("PLATFORM: banner-name [%s]\n", s);
773 	s = mdesc_get_property(hp, pn, "name", NULL);
774 	printk("PLATFORM: name [%s]\n", s);
775 
776 	v = mdesc_get_property(hp, pn, "hostid", NULL);
777 	if (v)
778 		printk("PLATFORM: hostid [%08llx]\n", *v);
779 	v = mdesc_get_property(hp, pn, "serial#", NULL);
780 	if (v)
781 		printk("PLATFORM: serial# [%08llx]\n", *v);
782 	v = mdesc_get_property(hp, pn, "stick-frequency", NULL);
783 	printk("PLATFORM: stick-frequency [%08llx]\n", *v);
784 	v = mdesc_get_property(hp, pn, "mac-address", NULL);
785 	if (v)
786 		printk("PLATFORM: mac-address [%llx]\n", *v);
787 	v = mdesc_get_property(hp, pn, "watchdog-resolution", NULL);
788 	if (v)
789 		printk("PLATFORM: watchdog-resolution [%llu ms]\n", *v);
790 	v = mdesc_get_property(hp, pn, "watchdog-max-timeout", NULL);
791 	if (v)
792 		printk("PLATFORM: watchdog-max-timeout [%llu ms]\n", *v);
793 	v = mdesc_get_property(hp, pn, "max-cpus", NULL);
794 	if (v) {
795 		max_cpus = *v;
796 		printk("PLATFORM: max-cpus [%llu]\n", max_cpus);
797 	}
798 
799 #ifdef CONFIG_SMP
800 	{
801 		int max_cpu, i;
802 
803 		if (v) {
804 			max_cpu = *v;
805 			if (max_cpu > NR_CPUS)
806 				max_cpu = NR_CPUS;
807 		} else {
808 			max_cpu = NR_CPUS;
809 		}
810 		for (i = 0; i < max_cpu; i++)
811 			set_cpu_possible(i, true);
812 	}
813 #endif
814 
815 	mdesc_release(hp);
816 }
817 
818 static void fill_in_one_cache(cpuinfo_sparc *c, struct mdesc_handle *hp, u64 mp)
819 {
820 	const u64 *level = mdesc_get_property(hp, mp, "level", NULL);
821 	const u64 *size = mdesc_get_property(hp, mp, "size", NULL);
822 	const u64 *line_size = mdesc_get_property(hp, mp, "line-size", NULL);
823 	const char *type;
824 	int type_len;
825 
826 	type = mdesc_get_property(hp, mp, "type", &type_len);
827 
828 	switch (*level) {
829 	case 1:
830 		if (of_find_in_proplist(type, "instn", type_len)) {
831 			c->icache_size = *size;
832 			c->icache_line_size = *line_size;
833 		} else if (of_find_in_proplist(type, "data", type_len)) {
834 			c->dcache_size = *size;
835 			c->dcache_line_size = *line_size;
836 		}
837 		break;
838 
839 	case 2:
840 		c->ecache_size = *size;
841 		c->ecache_line_size = *line_size;
842 		break;
843 
844 	default:
845 		break;
846 	}
847 
848 	if (*level == 1) {
849 		u64 a;
850 
851 		mdesc_for_each_arc(a, hp, mp, MDESC_ARC_TYPE_FWD) {
852 			u64 target = mdesc_arc_target(hp, a);
853 			const char *name = mdesc_node_name(hp, target);
854 
855 			if (!strcmp(name, "cache"))
856 				fill_in_one_cache(c, hp, target);
857 		}
858 	}
859 }
860 
861 static void find_back_node_value(struct mdesc_handle *hp, u64 node,
862 				 char *srch_val,
863 				 void (*func)(struct mdesc_handle *, u64, int),
864 				 u64 val, int depth)
865 {
866 	u64 arc;
867 
868 	/* Since we have an estimate of recursion depth, do a sanity check. */
869 	if (depth == 0)
870 		return;
871 
872 	mdesc_for_each_arc(arc, hp, node, MDESC_ARC_TYPE_BACK) {
873 		u64 n = mdesc_arc_target(hp, arc);
874 		const char *name = mdesc_node_name(hp, n);
875 
876 		if (!strcmp(srch_val, name))
877 			(*func)(hp, n, val);
878 
879 		find_back_node_value(hp, n, srch_val, func, val, depth-1);
880 	}
881 }
882 
883 static void __mark_core_id(struct mdesc_handle *hp, u64 node,
884 			   int core_id)
885 {
886 	const u64 *id = mdesc_get_property(hp, node, "id", NULL);
887 
888 	if (*id < num_possible_cpus())
889 		cpu_data(*id).core_id = core_id;
890 }
891 
892 static void __mark_max_cache_id(struct mdesc_handle *hp, u64 node,
893 				int max_cache_id)
894 {
895 	const u64 *id = mdesc_get_property(hp, node, "id", NULL);
896 
897 	if (*id < num_possible_cpus()) {
898 		cpu_data(*id).max_cache_id = max_cache_id;
899 
900 		/**
901 		 * On systems without explicit socket descriptions socket
902 		 * is max_cache_id
903 		 */
904 		cpu_data(*id).sock_id = max_cache_id;
905 	}
906 }
907 
908 static void mark_core_ids(struct mdesc_handle *hp, u64 mp,
909 			  int core_id)
910 {
911 	find_back_node_value(hp, mp, "cpu", __mark_core_id, core_id, 10);
912 }
913 
914 static void mark_max_cache_ids(struct mdesc_handle *hp, u64 mp,
915 			       int max_cache_id)
916 {
917 	find_back_node_value(hp, mp, "cpu", __mark_max_cache_id,
918 			     max_cache_id, 10);
919 }
920 
921 static void set_core_ids(struct mdesc_handle *hp)
922 {
923 	int idx;
924 	u64 mp;
925 
926 	idx = 1;
927 
928 	/* Identify unique cores by looking for cpus backpointed to by
929 	 * level 1 instruction caches.
930 	 */
931 	mdesc_for_each_node_by_name(hp, mp, "cache") {
932 		const u64 *level;
933 		const char *type;
934 		int len;
935 
936 		level = mdesc_get_property(hp, mp, "level", NULL);
937 		if (*level != 1)
938 			continue;
939 
940 		type = mdesc_get_property(hp, mp, "type", &len);
941 		if (!of_find_in_proplist(type, "instn", len))
942 			continue;
943 
944 		mark_core_ids(hp, mp, idx);
945 		idx++;
946 	}
947 }
948 
949 static int set_max_cache_ids_by_cache(struct mdesc_handle *hp, int level)
950 {
951 	u64 mp;
952 	int idx = 1;
953 	int fnd = 0;
954 
955 	/**
956 	 * Identify unique highest level of shared cache by looking for cpus
957 	 * backpointed to by shared level N caches.
958 	 */
959 	mdesc_for_each_node_by_name(hp, mp, "cache") {
960 		const u64 *cur_lvl;
961 
962 		cur_lvl = mdesc_get_property(hp, mp, "level", NULL);
963 		if (*cur_lvl != level)
964 			continue;
965 		mark_max_cache_ids(hp, mp, idx);
966 		idx++;
967 		fnd = 1;
968 	}
969 	return fnd;
970 }
971 
972 static void set_sock_ids_by_socket(struct mdesc_handle *hp, u64 mp)
973 {
974 	int idx = 1;
975 
976 	mdesc_for_each_node_by_name(hp, mp, "socket") {
977 		u64 a;
978 
979 		mdesc_for_each_arc(a, hp, mp, MDESC_ARC_TYPE_FWD) {
980 			u64 t = mdesc_arc_target(hp, a);
981 			const char *name;
982 			const u64 *id;
983 
984 			name = mdesc_node_name(hp, t);
985 			if (strcmp(name, "cpu"))
986 				continue;
987 
988 			id = mdesc_get_property(hp, t, "id", NULL);
989 			if (*id < num_possible_cpus())
990 				cpu_data(*id).sock_id = idx;
991 		}
992 		idx++;
993 	}
994 }
995 
996 static void set_sock_ids(struct mdesc_handle *hp)
997 {
998 	u64 mp;
999 
1000 	/**
1001 	 * Find the highest level of shared cache which pre-T7 is also
1002 	 * the socket.
1003 	 */
1004 	if (!set_max_cache_ids_by_cache(hp, 3))
1005 		set_max_cache_ids_by_cache(hp, 2);
1006 
1007 	/* If machine description exposes sockets data use it.*/
1008 	mp = mdesc_node_by_name(hp, MDESC_NODE_NULL, "sockets");
1009 	if (mp != MDESC_NODE_NULL)
1010 		set_sock_ids_by_socket(hp, mp);
1011 }
1012 
1013 static void mark_proc_ids(struct mdesc_handle *hp, u64 mp, int proc_id)
1014 {
1015 	u64 a;
1016 
1017 	mdesc_for_each_arc(a, hp, mp, MDESC_ARC_TYPE_BACK) {
1018 		u64 t = mdesc_arc_target(hp, a);
1019 		const char *name;
1020 		const u64 *id;
1021 
1022 		name = mdesc_node_name(hp, t);
1023 		if (strcmp(name, "cpu"))
1024 			continue;
1025 
1026 		id = mdesc_get_property(hp, t, "id", NULL);
1027 		if (*id < NR_CPUS)
1028 			cpu_data(*id).proc_id = proc_id;
1029 	}
1030 }
1031 
1032 static void __set_proc_ids(struct mdesc_handle *hp, const char *exec_unit_name)
1033 {
1034 	int idx;
1035 	u64 mp;
1036 
1037 	idx = 0;
1038 	mdesc_for_each_node_by_name(hp, mp, exec_unit_name) {
1039 		const char *type;
1040 		int len;
1041 
1042 		type = mdesc_get_property(hp, mp, "type", &len);
1043 		if (!of_find_in_proplist(type, "int", len) &&
1044 		    !of_find_in_proplist(type, "integer", len))
1045 			continue;
1046 
1047 		mark_proc_ids(hp, mp, idx);
1048 		idx++;
1049 	}
1050 }
1051 
1052 static void set_proc_ids(struct mdesc_handle *hp)
1053 {
1054 	__set_proc_ids(hp, "exec_unit");
1055 	__set_proc_ids(hp, "exec-unit");
1056 }
1057 
1058 static void get_one_mondo_bits(const u64 *p, unsigned int *mask,
1059 			       unsigned long def, unsigned long max)
1060 {
1061 	u64 val;
1062 
1063 	if (!p)
1064 		goto use_default;
1065 	val = *p;
1066 
1067 	if (!val || val >= 64)
1068 		goto use_default;
1069 
1070 	if (val > max)
1071 		val = max;
1072 
1073 	*mask = ((1U << val) * 64U) - 1U;
1074 	return;
1075 
1076 use_default:
1077 	*mask = ((1U << def) * 64U) - 1U;
1078 }
1079 
1080 static void get_mondo_data(struct mdesc_handle *hp, u64 mp,
1081 			   struct trap_per_cpu *tb)
1082 {
1083 	static int printed;
1084 	const u64 *val;
1085 
1086 	val = mdesc_get_property(hp, mp, "q-cpu-mondo-#bits", NULL);
1087 	get_one_mondo_bits(val, &tb->cpu_mondo_qmask, 7, ilog2(max_cpus * 2));
1088 
1089 	val = mdesc_get_property(hp, mp, "q-dev-mondo-#bits", NULL);
1090 	get_one_mondo_bits(val, &tb->dev_mondo_qmask, 7, 8);
1091 
1092 	val = mdesc_get_property(hp, mp, "q-resumable-#bits", NULL);
1093 	get_one_mondo_bits(val, &tb->resum_qmask, 6, 7);
1094 
1095 	val = mdesc_get_property(hp, mp, "q-nonresumable-#bits", NULL);
1096 	get_one_mondo_bits(val, &tb->nonresum_qmask, 2, 2);
1097 	if (!printed++) {
1098 		pr_info("SUN4V: Mondo queue sizes "
1099 			"[cpu(%u) dev(%u) r(%u) nr(%u)]\n",
1100 			tb->cpu_mondo_qmask + 1,
1101 			tb->dev_mondo_qmask + 1,
1102 			tb->resum_qmask + 1,
1103 			tb->nonresum_qmask + 1);
1104 	}
1105 }
1106 
1107 static void *mdesc_iterate_over_cpus(void *(*func)(struct mdesc_handle *, u64, int, void *), void *arg, cpumask_t *mask)
1108 {
1109 	struct mdesc_handle *hp = mdesc_grab();
1110 	void *ret = NULL;
1111 	u64 mp;
1112 
1113 	mdesc_for_each_node_by_name(hp, mp, "cpu") {
1114 		const u64 *id = mdesc_get_property(hp, mp, "id", NULL);
1115 		int cpuid = *id;
1116 
1117 #ifdef CONFIG_SMP
1118 		if (cpuid >= NR_CPUS) {
1119 			printk(KERN_WARNING "Ignoring CPU %d which is "
1120 			       ">= NR_CPUS (%d)\n",
1121 			       cpuid, NR_CPUS);
1122 			continue;
1123 		}
1124 		if (!cpumask_test_cpu(cpuid, mask))
1125 			continue;
1126 #endif
1127 
1128 		ret = func(hp, mp, cpuid, arg);
1129 		if (ret)
1130 			goto out;
1131 	}
1132 out:
1133 	mdesc_release(hp);
1134 	return ret;
1135 }
1136 
1137 static void *record_one_cpu(struct mdesc_handle *hp, u64 mp, int cpuid,
1138 			    void *arg)
1139 {
1140 	ncpus_probed++;
1141 #ifdef CONFIG_SMP
1142 	set_cpu_present(cpuid, true);
1143 #endif
1144 	return NULL;
1145 }
1146 
1147 void mdesc_populate_present_mask(cpumask_t *mask)
1148 {
1149 	if (tlb_type != hypervisor)
1150 		return;
1151 
1152 	ncpus_probed = 0;
1153 	mdesc_iterate_over_cpus(record_one_cpu, NULL, mask);
1154 }
1155 
1156 static void * __init check_one_pgsz(struct mdesc_handle *hp, u64 mp, int cpuid, void *arg)
1157 {
1158 	const u64 *pgsz_prop = mdesc_get_property(hp, mp, "mmu-page-size-list", NULL);
1159 	unsigned long *pgsz_mask = arg;
1160 	u64 val;
1161 
1162 	val = (HV_PGSZ_MASK_8K | HV_PGSZ_MASK_64K |
1163 	       HV_PGSZ_MASK_512K | HV_PGSZ_MASK_4MB);
1164 	if (pgsz_prop)
1165 		val = *pgsz_prop;
1166 
1167 	if (!*pgsz_mask)
1168 		*pgsz_mask = val;
1169 	else
1170 		*pgsz_mask &= val;
1171 	return NULL;
1172 }
1173 
1174 void __init mdesc_get_page_sizes(cpumask_t *mask, unsigned long *pgsz_mask)
1175 {
1176 	*pgsz_mask = 0;
1177 	mdesc_iterate_over_cpus(check_one_pgsz, pgsz_mask, mask);
1178 }
1179 
1180 static void *fill_in_one_cpu(struct mdesc_handle *hp, u64 mp, int cpuid,
1181 			     void *arg)
1182 {
1183 	const u64 *cfreq = mdesc_get_property(hp, mp, "clock-frequency", NULL);
1184 	struct trap_per_cpu *tb;
1185 	cpuinfo_sparc *c;
1186 	u64 a;
1187 
1188 #ifndef CONFIG_SMP
1189 	/* On uniprocessor we only want the values for the
1190 	 * real physical cpu the kernel booted onto, however
1191 	 * cpu_data() only has one entry at index 0.
1192 	 */
1193 	if (cpuid != real_hard_smp_processor_id())
1194 		return NULL;
1195 	cpuid = 0;
1196 #endif
1197 
1198 	c = &cpu_data(cpuid);
1199 	c->clock_tick = *cfreq;
1200 
1201 	tb = &trap_block[cpuid];
1202 	get_mondo_data(hp, mp, tb);
1203 
1204 	mdesc_for_each_arc(a, hp, mp, MDESC_ARC_TYPE_FWD) {
1205 		u64 j, t = mdesc_arc_target(hp, a);
1206 		const char *t_name;
1207 
1208 		t_name = mdesc_node_name(hp, t);
1209 		if (!strcmp(t_name, "cache")) {
1210 			fill_in_one_cache(c, hp, t);
1211 			continue;
1212 		}
1213 
1214 		mdesc_for_each_arc(j, hp, t, MDESC_ARC_TYPE_FWD) {
1215 			u64 n = mdesc_arc_target(hp, j);
1216 			const char *n_name;
1217 
1218 			n_name = mdesc_node_name(hp, n);
1219 			if (!strcmp(n_name, "cache"))
1220 				fill_in_one_cache(c, hp, n);
1221 		}
1222 	}
1223 
1224 	c->core_id = 0;
1225 	c->proc_id = -1;
1226 
1227 	return NULL;
1228 }
1229 
1230 void mdesc_fill_in_cpu_data(cpumask_t *mask)
1231 {
1232 	struct mdesc_handle *hp;
1233 
1234 	mdesc_iterate_over_cpus(fill_in_one_cpu, NULL, mask);
1235 
1236 	hp = mdesc_grab();
1237 
1238 	set_core_ids(hp);
1239 	set_proc_ids(hp);
1240 	set_sock_ids(hp);
1241 
1242 	mdesc_release(hp);
1243 
1244 	smp_fill_in_sib_core_maps();
1245 }
1246 
1247 /* mdesc_open() - Grab a reference to mdesc_handle when /dev/mdesc is
1248  * opened. Hold this reference until /dev/mdesc is closed to ensure
1249  * mdesc data structure is not released underneath us. Store the
1250  * pointer to mdesc structure in private_data for read and seek to use
1251  */
1252 static int mdesc_open(struct inode *inode, struct file *file)
1253 {
1254 	struct mdesc_handle *hp = mdesc_grab();
1255 
1256 	if (!hp)
1257 		return -ENODEV;
1258 
1259 	file->private_data = hp;
1260 
1261 	return 0;
1262 }
1263 
1264 static ssize_t mdesc_read(struct file *file, char __user *buf,
1265 			  size_t len, loff_t *offp)
1266 {
1267 	struct mdesc_handle *hp = file->private_data;
1268 	unsigned char *mdesc;
1269 	int bytes_left, count = len;
1270 
1271 	if (*offp >= hp->handle_size)
1272 		return 0;
1273 
1274 	bytes_left = hp->handle_size - *offp;
1275 	if (count > bytes_left)
1276 		count = bytes_left;
1277 
1278 	mdesc = (unsigned char *)&hp->mdesc;
1279 	mdesc += *offp;
1280 	if (!copy_to_user(buf, mdesc, count)) {
1281 		*offp += count;
1282 		return count;
1283 	} else {
1284 		return -EFAULT;
1285 	}
1286 }
1287 
1288 static loff_t mdesc_llseek(struct file *file, loff_t offset, int whence)
1289 {
1290 	struct mdesc_handle *hp = file->private_data;
1291 
1292 	return no_seek_end_llseek_size(file, offset, whence, hp->handle_size);
1293 }
1294 
1295 /* mdesc_close() - /dev/mdesc is being closed, release the reference to
1296  * mdesc structure.
1297  */
1298 static int mdesc_close(struct inode *inode, struct file *file)
1299 {
1300 	mdesc_release(file->private_data);
1301 	return 0;
1302 }
1303 
1304 static const struct file_operations mdesc_fops = {
1305 	.open    = mdesc_open,
1306 	.read	 = mdesc_read,
1307 	.llseek  = mdesc_llseek,
1308 	.release = mdesc_close,
1309 	.owner	 = THIS_MODULE,
1310 };
1311 
1312 static struct miscdevice mdesc_misc = {
1313 	.minor	= MISC_DYNAMIC_MINOR,
1314 	.name	= "mdesc",
1315 	.fops	= &mdesc_fops,
1316 };
1317 
1318 static int __init mdesc_misc_init(void)
1319 {
1320 	return misc_register(&mdesc_misc);
1321 }
1322 
1323 __initcall(mdesc_misc_init);
1324 
1325 void __init sun4v_mdesc_init(void)
1326 {
1327 	struct mdesc_handle *hp;
1328 	unsigned long len, real_len, status;
1329 
1330 	(void) sun4v_mach_desc(0UL, 0UL, &len);
1331 
1332 	printk("MDESC: Size is %lu bytes.\n", len);
1333 
1334 	hp = mdesc_alloc(len, &memblock_mdesc_ops);
1335 	if (hp == NULL) {
1336 		prom_printf("MDESC: alloc of %lu bytes failed.\n", len);
1337 		prom_halt();
1338 	}
1339 
1340 	status = sun4v_mach_desc(__pa(&hp->mdesc), len, &real_len);
1341 	if (status != HV_EOK || real_len > len) {
1342 		prom_printf("sun4v_mach_desc fails, err(%lu), "
1343 			    "len(%lu), real_len(%lu)\n",
1344 			    status, len, real_len);
1345 		mdesc_free(hp);
1346 		prom_halt();
1347 	}
1348 
1349 	cur_mdesc = hp;
1350 
1351 	mdesc_adi_init();
1352 	report_platform_properties();
1353 }
1354