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/memblock.h>
9 #include <linux/log2.h>
10 #include <linux/list.h>
11 #include <linux/slab.h>
12 #include <linux/mm.h>
13 #include <linux/miscdevice.h>
14 #include <linux/bootmem.h>
15 #include <linux/export.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
25 /* Unlike the OBP device tree, the machine description is a full-on
26 * DAG. An arbitrary number of ARCs are possible from one
27 * node to other nodes and thus we can't use the OBP device_node
28 * data structure to represent these nodes inside of the kernel.
29 *
30 * Actually, it isn't even a DAG, because there are back pointers
31 * which create cycles in the graph.
32 *
33 * mdesc_hdr and mdesc_elem describe the layout of the data structure
34 * we get from the Hypervisor.
35 */
36 struct mdesc_hdr {
37 u32 version; /* Transport version */
38 u32 node_sz; /* node block size */
39 u32 name_sz; /* name block size */
40 u32 data_sz; /* data block size */
41 } __attribute__((aligned(16)));
42
43 struct mdesc_elem {
44 u8 tag;
45 #define MD_LIST_END 0x00
46 #define MD_NODE 0x4e
47 #define MD_NODE_END 0x45
48 #define MD_NOOP 0x20
49 #define MD_PROP_ARC 0x61
50 #define MD_PROP_VAL 0x76
51 #define MD_PROP_STR 0x73
52 #define MD_PROP_DATA 0x64
53 u8 name_len;
54 u16 resv;
55 u32 name_offset;
56 union {
57 struct {
58 u32 data_len;
59 u32 data_offset;
60 } data;
61 u64 val;
62 } d;
63 };
64
65 struct mdesc_mem_ops {
66 struct mdesc_handle *(*alloc)(unsigned int mdesc_size);
67 void (*free)(struct mdesc_handle *handle);
68 };
69
70 struct mdesc_handle {
71 struct list_head list;
72 struct mdesc_mem_ops *mops;
73 void *self_base;
74 atomic_t refcnt;
75 unsigned int handle_size;
76 struct mdesc_hdr mdesc;
77 };
78
79 typedef int (*mdesc_node_info_get_f)(struct mdesc_handle *, u64,
80 union md_node_info *);
81 typedef void (*mdesc_node_info_rel_f)(union md_node_info *);
82 typedef bool (*mdesc_node_match_f)(union md_node_info *, union md_node_info *);
83
84 struct md_node_ops {
85 char *name;
86 mdesc_node_info_get_f get_info;
87 mdesc_node_info_rel_f rel_info;
88 mdesc_node_match_f node_match;
89 };
90
91 static int get_vdev_port_node_info(struct mdesc_handle *md, u64 node,
92 union md_node_info *node_info);
93 static void rel_vdev_port_node_info(union md_node_info *node_info);
94 static bool vdev_port_node_match(union md_node_info *a_node_info,
95 union md_node_info *b_node_info);
96
97 static int get_ds_port_node_info(struct mdesc_handle *md, u64 node,
98 union md_node_info *node_info);
99 static void rel_ds_port_node_info(union md_node_info *node_info);
100 static bool ds_port_node_match(union md_node_info *a_node_info,
101 union md_node_info *b_node_info);
102
103 /* supported node types which can be registered */
104 static struct md_node_ops md_node_ops_table[] = {
105 {"virtual-device-port", get_vdev_port_node_info,
106 rel_vdev_port_node_info, vdev_port_node_match},
107 {"domain-services-port", get_ds_port_node_info,
108 rel_ds_port_node_info, ds_port_node_match},
109 {NULL, NULL, NULL, NULL}
110 };
111
mdesc_get_node_ops(const char * node_name,mdesc_node_info_get_f * get_info_f,mdesc_node_info_rel_f * rel_info_f,mdesc_node_match_f * match_f)112 static void mdesc_get_node_ops(const char *node_name,
113 mdesc_node_info_get_f *get_info_f,
114 mdesc_node_info_rel_f *rel_info_f,
115 mdesc_node_match_f *match_f)
116 {
117 int i;
118
119 if (get_info_f)
120 *get_info_f = NULL;
121
122 if (rel_info_f)
123 *rel_info_f = NULL;
124
125 if (match_f)
126 *match_f = NULL;
127
128 if (!node_name)
129 return;
130
131 for (i = 0; md_node_ops_table[i].name != NULL; i++) {
132 if (strcmp(md_node_ops_table[i].name, node_name) == 0) {
133 if (get_info_f)
134 *get_info_f = md_node_ops_table[i].get_info;
135
136 if (rel_info_f)
137 *rel_info_f = md_node_ops_table[i].rel_info;
138
139 if (match_f)
140 *match_f = md_node_ops_table[i].node_match;
141
142 break;
143 }
144 }
145 }
146
mdesc_handle_init(struct mdesc_handle * hp,unsigned int handle_size,void * base)147 static void mdesc_handle_init(struct mdesc_handle *hp,
148 unsigned int handle_size,
149 void *base)
150 {
151 BUG_ON(((unsigned long)&hp->mdesc) & (16UL - 1));
152
153 memset(hp, 0, handle_size);
154 INIT_LIST_HEAD(&hp->list);
155 hp->self_base = base;
156 atomic_set(&hp->refcnt, 1);
157 hp->handle_size = handle_size;
158 }
159
mdesc_memblock_alloc(unsigned int mdesc_size)160 static struct mdesc_handle * __init mdesc_memblock_alloc(unsigned int mdesc_size)
161 {
162 unsigned int handle_size, alloc_size;
163 struct mdesc_handle *hp;
164 unsigned long paddr;
165
166 handle_size = (sizeof(struct mdesc_handle) -
167 sizeof(struct mdesc_hdr) +
168 mdesc_size);
169 alloc_size = PAGE_ALIGN(handle_size);
170
171 paddr = memblock_alloc(alloc_size, PAGE_SIZE);
172
173 hp = NULL;
174 if (paddr) {
175 hp = __va(paddr);
176 mdesc_handle_init(hp, handle_size, hp);
177 }
178 return hp;
179 }
180
mdesc_memblock_free(struct mdesc_handle * hp)181 static void __init mdesc_memblock_free(struct mdesc_handle *hp)
182 {
183 unsigned int alloc_size;
184 unsigned long start;
185
186 BUG_ON(atomic_read(&hp->refcnt) != 0);
187 BUG_ON(!list_empty(&hp->list));
188
189 alloc_size = PAGE_ALIGN(hp->handle_size);
190 start = __pa(hp);
191 free_bootmem_late(start, alloc_size);
192 }
193
194 static struct mdesc_mem_ops memblock_mdesc_ops = {
195 .alloc = mdesc_memblock_alloc,
196 .free = mdesc_memblock_free,
197 };
198
mdesc_kmalloc(unsigned int mdesc_size)199 static struct mdesc_handle *mdesc_kmalloc(unsigned int mdesc_size)
200 {
201 unsigned int handle_size;
202 struct mdesc_handle *hp;
203 unsigned long addr;
204 void *base;
205
206 handle_size = (sizeof(struct mdesc_handle) -
207 sizeof(struct mdesc_hdr) +
208 mdesc_size);
209 base = kmalloc(handle_size + 15, GFP_KERNEL | __GFP_RETRY_MAYFAIL);
210 if (!base)
211 return NULL;
212
213 addr = (unsigned long)base;
214 addr = (addr + 15UL) & ~15UL;
215 hp = (struct mdesc_handle *) addr;
216
217 mdesc_handle_init(hp, handle_size, base);
218
219 return hp;
220 }
221
mdesc_kfree(struct mdesc_handle * hp)222 static void mdesc_kfree(struct mdesc_handle *hp)
223 {
224 BUG_ON(atomic_read(&hp->refcnt) != 0);
225 BUG_ON(!list_empty(&hp->list));
226
227 kfree(hp->self_base);
228 }
229
230 static struct mdesc_mem_ops kmalloc_mdesc_memops = {
231 .alloc = mdesc_kmalloc,
232 .free = mdesc_kfree,
233 };
234
mdesc_alloc(unsigned int mdesc_size,struct mdesc_mem_ops * mops)235 static struct mdesc_handle *mdesc_alloc(unsigned int mdesc_size,
236 struct mdesc_mem_ops *mops)
237 {
238 struct mdesc_handle *hp = mops->alloc(mdesc_size);
239
240 if (hp)
241 hp->mops = mops;
242
243 return hp;
244 }
245
mdesc_free(struct mdesc_handle * hp)246 static void mdesc_free(struct mdesc_handle *hp)
247 {
248 hp->mops->free(hp);
249 }
250
251 static struct mdesc_handle *cur_mdesc;
252 static LIST_HEAD(mdesc_zombie_list);
253 static DEFINE_SPINLOCK(mdesc_lock);
254
mdesc_grab(void)255 struct mdesc_handle *mdesc_grab(void)
256 {
257 struct mdesc_handle *hp;
258 unsigned long flags;
259
260 spin_lock_irqsave(&mdesc_lock, flags);
261 hp = cur_mdesc;
262 if (hp)
263 atomic_inc(&hp->refcnt);
264 spin_unlock_irqrestore(&mdesc_lock, flags);
265
266 return hp;
267 }
268 EXPORT_SYMBOL(mdesc_grab);
269
mdesc_release(struct mdesc_handle * hp)270 void mdesc_release(struct mdesc_handle *hp)
271 {
272 unsigned long flags;
273
274 spin_lock_irqsave(&mdesc_lock, flags);
275 if (atomic_dec_and_test(&hp->refcnt)) {
276 list_del_init(&hp->list);
277 hp->mops->free(hp);
278 }
279 spin_unlock_irqrestore(&mdesc_lock, flags);
280 }
281 EXPORT_SYMBOL(mdesc_release);
282
283 static DEFINE_MUTEX(mdesc_mutex);
284 static struct mdesc_notifier_client *client_list;
285
mdesc_register_notifier(struct mdesc_notifier_client * client)286 void mdesc_register_notifier(struct mdesc_notifier_client *client)
287 {
288 bool supported = false;
289 u64 node;
290 int i;
291
292 mutex_lock(&mdesc_mutex);
293
294 /* check to see if the node is supported for registration */
295 for (i = 0; md_node_ops_table[i].name != NULL; i++) {
296 if (strcmp(md_node_ops_table[i].name, client->node_name) == 0) {
297 supported = true;
298 break;
299 }
300 }
301
302 if (!supported) {
303 pr_err("MD: %s node not supported\n", client->node_name);
304 mutex_unlock(&mdesc_mutex);
305 return;
306 }
307
308 client->next = client_list;
309 client_list = client;
310
311 mdesc_for_each_node_by_name(cur_mdesc, node, client->node_name)
312 client->add(cur_mdesc, node, client->node_name);
313
314 mutex_unlock(&mdesc_mutex);
315 }
316
parent_cfg_handle(struct mdesc_handle * hp,u64 node)317 static const u64 *parent_cfg_handle(struct mdesc_handle *hp, u64 node)
318 {
319 const u64 *id;
320 u64 a;
321
322 id = NULL;
323 mdesc_for_each_arc(a, hp, node, MDESC_ARC_TYPE_BACK) {
324 u64 target;
325
326 target = mdesc_arc_target(hp, a);
327 id = mdesc_get_property(hp, target,
328 "cfg-handle", NULL);
329 if (id)
330 break;
331 }
332
333 return id;
334 }
335
get_vdev_port_node_info(struct mdesc_handle * md,u64 node,union md_node_info * node_info)336 static int get_vdev_port_node_info(struct mdesc_handle *md, u64 node,
337 union md_node_info *node_info)
338 {
339 const u64 *parent_cfg_hdlp;
340 const char *name;
341 const u64 *idp;
342
343 /*
344 * Virtual device nodes are distinguished by:
345 * 1. "id" property
346 * 2. "name" property
347 * 3. parent node "cfg-handle" property
348 */
349 idp = mdesc_get_property(md, node, "id", NULL);
350 name = mdesc_get_property(md, node, "name", NULL);
351 parent_cfg_hdlp = parent_cfg_handle(md, node);
352
353 if (!idp || !name || !parent_cfg_hdlp)
354 return -1;
355
356 node_info->vdev_port.id = *idp;
357 node_info->vdev_port.name = kstrdup_const(name, GFP_KERNEL);
358 if (!node_info->vdev_port.name)
359 return -1;
360 node_info->vdev_port.parent_cfg_hdl = *parent_cfg_hdlp;
361
362 return 0;
363 }
364
rel_vdev_port_node_info(union md_node_info * node_info)365 static void rel_vdev_port_node_info(union md_node_info *node_info)
366 {
367 if (node_info && node_info->vdev_port.name) {
368 kfree_const(node_info->vdev_port.name);
369 node_info->vdev_port.name = NULL;
370 }
371 }
372
vdev_port_node_match(union md_node_info * a_node_info,union md_node_info * b_node_info)373 static bool vdev_port_node_match(union md_node_info *a_node_info,
374 union md_node_info *b_node_info)
375 {
376 if (a_node_info->vdev_port.id != b_node_info->vdev_port.id)
377 return false;
378
379 if (a_node_info->vdev_port.parent_cfg_hdl !=
380 b_node_info->vdev_port.parent_cfg_hdl)
381 return false;
382
383 if (strncmp(a_node_info->vdev_port.name,
384 b_node_info->vdev_port.name, MDESC_MAX_STR_LEN) != 0)
385 return false;
386
387 return true;
388 }
389
get_ds_port_node_info(struct mdesc_handle * md,u64 node,union md_node_info * node_info)390 static int get_ds_port_node_info(struct mdesc_handle *md, u64 node,
391 union md_node_info *node_info)
392 {
393 const u64 *idp;
394
395 /* DS port nodes use the "id" property to distinguish them */
396 idp = mdesc_get_property(md, node, "id", NULL);
397 if (!idp)
398 return -1;
399
400 node_info->ds_port.id = *idp;
401
402 return 0;
403 }
404
rel_ds_port_node_info(union md_node_info * node_info)405 static void rel_ds_port_node_info(union md_node_info *node_info)
406 {
407 }
408
ds_port_node_match(union md_node_info * a_node_info,union md_node_info * b_node_info)409 static bool ds_port_node_match(union md_node_info *a_node_info,
410 union md_node_info *b_node_info)
411 {
412 if (a_node_info->ds_port.id != b_node_info->ds_port.id)
413 return false;
414
415 return true;
416 }
417
418 /* Run 'func' on nodes which are in A but not in B. */
invoke_on_missing(const char * name,struct mdesc_handle * a,struct mdesc_handle * b,void (* func)(struct mdesc_handle *,u64,const char * node_name))419 static void invoke_on_missing(const char *name,
420 struct mdesc_handle *a,
421 struct mdesc_handle *b,
422 void (*func)(struct mdesc_handle *, u64,
423 const char *node_name))
424 {
425 mdesc_node_info_get_f get_info_func;
426 mdesc_node_info_rel_f rel_info_func;
427 mdesc_node_match_f node_match_func;
428 union md_node_info a_node_info;
429 union md_node_info b_node_info;
430 bool found;
431 u64 a_node;
432 u64 b_node;
433 int rv;
434
435 /*
436 * Find the get_info, rel_info and node_match ops for the given
437 * node name
438 */
439 mdesc_get_node_ops(name, &get_info_func, &rel_info_func,
440 &node_match_func);
441
442 /* If we didn't find a match, the node type is not supported */
443 if (!get_info_func || !rel_info_func || !node_match_func) {
444 pr_err("MD: %s node type is not supported\n", name);
445 return;
446 }
447
448 mdesc_for_each_node_by_name(a, a_node, name) {
449 found = false;
450
451 rv = get_info_func(a, a_node, &a_node_info);
452 if (rv != 0) {
453 pr_err("MD: Cannot find 1 or more required match properties for %s node.\n",
454 name);
455 continue;
456 }
457
458 /* Check each node in B for node matching a_node */
459 mdesc_for_each_node_by_name(b, b_node, name) {
460 rv = get_info_func(b, b_node, &b_node_info);
461 if (rv != 0)
462 continue;
463
464 if (node_match_func(&a_node_info, &b_node_info)) {
465 found = true;
466 rel_info_func(&b_node_info);
467 break;
468 }
469
470 rel_info_func(&b_node_info);
471 }
472
473 rel_info_func(&a_node_info);
474
475 if (!found)
476 func(a, a_node, name);
477 }
478 }
479
notify_one(struct mdesc_notifier_client * p,struct mdesc_handle * old_hp,struct mdesc_handle * new_hp)480 static void notify_one(struct mdesc_notifier_client *p,
481 struct mdesc_handle *old_hp,
482 struct mdesc_handle *new_hp)
483 {
484 invoke_on_missing(p->node_name, old_hp, new_hp, p->remove);
485 invoke_on_missing(p->node_name, new_hp, old_hp, p->add);
486 }
487
mdesc_notify_clients(struct mdesc_handle * old_hp,struct mdesc_handle * new_hp)488 static void mdesc_notify_clients(struct mdesc_handle *old_hp,
489 struct mdesc_handle *new_hp)
490 {
491 struct mdesc_notifier_client *p = client_list;
492
493 while (p) {
494 notify_one(p, old_hp, new_hp);
495 p = p->next;
496 }
497 }
498
mdesc_update(void)499 void mdesc_update(void)
500 {
501 unsigned long len, real_len, status;
502 struct mdesc_handle *hp, *orig_hp;
503 unsigned long flags;
504
505 mutex_lock(&mdesc_mutex);
506
507 (void) sun4v_mach_desc(0UL, 0UL, &len);
508
509 hp = mdesc_alloc(len, &kmalloc_mdesc_memops);
510 if (!hp) {
511 printk(KERN_ERR "MD: mdesc alloc fails\n");
512 goto out;
513 }
514
515 status = sun4v_mach_desc(__pa(&hp->mdesc), len, &real_len);
516 if (status != HV_EOK || real_len > len) {
517 printk(KERN_ERR "MD: mdesc reread fails with %lu\n",
518 status);
519 atomic_dec(&hp->refcnt);
520 mdesc_free(hp);
521 goto out;
522 }
523
524 spin_lock_irqsave(&mdesc_lock, flags);
525 orig_hp = cur_mdesc;
526 cur_mdesc = hp;
527 spin_unlock_irqrestore(&mdesc_lock, flags);
528
529 mdesc_notify_clients(orig_hp, hp);
530
531 spin_lock_irqsave(&mdesc_lock, flags);
532 if (atomic_dec_and_test(&orig_hp->refcnt))
533 mdesc_free(orig_hp);
534 else
535 list_add(&orig_hp->list, &mdesc_zombie_list);
536 spin_unlock_irqrestore(&mdesc_lock, flags);
537
538 out:
539 mutex_unlock(&mdesc_mutex);
540 }
541
mdesc_get_node(struct mdesc_handle * hp,const char * node_name,union md_node_info * node_info)542 u64 mdesc_get_node(struct mdesc_handle *hp, const char *node_name,
543 union md_node_info *node_info)
544 {
545 mdesc_node_info_get_f get_info_func;
546 mdesc_node_info_rel_f rel_info_func;
547 mdesc_node_match_f node_match_func;
548 union md_node_info hp_node_info;
549 u64 hp_node;
550 int rv;
551
552 if (hp == NULL || node_name == NULL || node_info == NULL)
553 return MDESC_NODE_NULL;
554
555 /* Find the ops for the given node name */
556 mdesc_get_node_ops(node_name, &get_info_func, &rel_info_func,
557 &node_match_func);
558
559 /* If we didn't find ops for the given node name, it is not supported */
560 if (!get_info_func || !rel_info_func || !node_match_func) {
561 pr_err("MD: %s node is not supported\n", node_name);
562 return -EINVAL;
563 }
564
565 mdesc_for_each_node_by_name(hp, hp_node, node_name) {
566 rv = get_info_func(hp, hp_node, &hp_node_info);
567 if (rv != 0)
568 continue;
569
570 if (node_match_func(node_info, &hp_node_info))
571 break;
572
573 rel_info_func(&hp_node_info);
574 }
575
576 rel_info_func(&hp_node_info);
577
578 return hp_node;
579 }
580 EXPORT_SYMBOL(mdesc_get_node);
581
mdesc_get_node_info(struct mdesc_handle * hp,u64 node,const char * node_name,union md_node_info * node_info)582 int mdesc_get_node_info(struct mdesc_handle *hp, u64 node,
583 const char *node_name, union md_node_info *node_info)
584 {
585 mdesc_node_info_get_f get_info_func;
586 int rv;
587
588 if (hp == NULL || node == MDESC_NODE_NULL ||
589 node_name == NULL || node_info == NULL)
590 return -EINVAL;
591
592 /* Find the get_info op for the given node name */
593 mdesc_get_node_ops(node_name, &get_info_func, NULL, NULL);
594
595 /* If we didn't find a get_info_func, the node name is not supported */
596 if (get_info_func == NULL) {
597 pr_err("MD: %s node is not supported\n", node_name);
598 return -EINVAL;
599 }
600
601 rv = get_info_func(hp, node, node_info);
602 if (rv != 0) {
603 pr_err("MD: Cannot find 1 or more required match properties for %s node.\n",
604 node_name);
605 return -1;
606 }
607
608 return 0;
609 }
610 EXPORT_SYMBOL(mdesc_get_node_info);
611
node_block(struct mdesc_hdr * mdesc)612 static struct mdesc_elem *node_block(struct mdesc_hdr *mdesc)
613 {
614 return (struct mdesc_elem *) (mdesc + 1);
615 }
616
name_block(struct mdesc_hdr * mdesc)617 static void *name_block(struct mdesc_hdr *mdesc)
618 {
619 return ((void *) node_block(mdesc)) + mdesc->node_sz;
620 }
621
data_block(struct mdesc_hdr * mdesc)622 static void *data_block(struct mdesc_hdr *mdesc)
623 {
624 return ((void *) name_block(mdesc)) + mdesc->name_sz;
625 }
626
mdesc_node_by_name(struct mdesc_handle * hp,u64 from_node,const char * name)627 u64 mdesc_node_by_name(struct mdesc_handle *hp,
628 u64 from_node, const char *name)
629 {
630 struct mdesc_elem *ep = node_block(&hp->mdesc);
631 const char *names = name_block(&hp->mdesc);
632 u64 last_node = hp->mdesc.node_sz / 16;
633 u64 ret;
634
635 if (from_node == MDESC_NODE_NULL) {
636 ret = from_node = 0;
637 } else if (from_node >= last_node) {
638 return MDESC_NODE_NULL;
639 } else {
640 ret = ep[from_node].d.val;
641 }
642
643 while (ret < last_node) {
644 if (ep[ret].tag != MD_NODE)
645 return MDESC_NODE_NULL;
646 if (!strcmp(names + ep[ret].name_offset, name))
647 break;
648 ret = ep[ret].d.val;
649 }
650 if (ret >= last_node)
651 ret = MDESC_NODE_NULL;
652 return ret;
653 }
654 EXPORT_SYMBOL(mdesc_node_by_name);
655
mdesc_get_property(struct mdesc_handle * hp,u64 node,const char * name,int * lenp)656 const void *mdesc_get_property(struct mdesc_handle *hp, u64 node,
657 const char *name, int *lenp)
658 {
659 const char *names = name_block(&hp->mdesc);
660 u64 last_node = hp->mdesc.node_sz / 16;
661 void *data = data_block(&hp->mdesc);
662 struct mdesc_elem *ep;
663
664 if (node == MDESC_NODE_NULL || node >= last_node)
665 return NULL;
666
667 ep = node_block(&hp->mdesc) + node;
668 ep++;
669 for (; ep->tag != MD_NODE_END; ep++) {
670 void *val = NULL;
671 int len = 0;
672
673 switch (ep->tag) {
674 case MD_PROP_VAL:
675 val = &ep->d.val;
676 len = 8;
677 break;
678
679 case MD_PROP_STR:
680 case MD_PROP_DATA:
681 val = data + ep->d.data.data_offset;
682 len = ep->d.data.data_len;
683 break;
684
685 default:
686 break;
687 }
688 if (!val)
689 continue;
690
691 if (!strcmp(names + ep->name_offset, name)) {
692 if (lenp)
693 *lenp = len;
694 return val;
695 }
696 }
697
698 return NULL;
699 }
700 EXPORT_SYMBOL(mdesc_get_property);
701
mdesc_next_arc(struct mdesc_handle * hp,u64 from,const char * arc_type)702 u64 mdesc_next_arc(struct mdesc_handle *hp, u64 from, const char *arc_type)
703 {
704 struct mdesc_elem *ep, *base = node_block(&hp->mdesc);
705 const char *names = name_block(&hp->mdesc);
706 u64 last_node = hp->mdesc.node_sz / 16;
707
708 if (from == MDESC_NODE_NULL || from >= last_node)
709 return MDESC_NODE_NULL;
710
711 ep = base + from;
712
713 ep++;
714 for (; ep->tag != MD_NODE_END; ep++) {
715 if (ep->tag != MD_PROP_ARC)
716 continue;
717
718 if (strcmp(names + ep->name_offset, arc_type))
719 continue;
720
721 return ep - base;
722 }
723
724 return MDESC_NODE_NULL;
725 }
726 EXPORT_SYMBOL(mdesc_next_arc);
727
mdesc_arc_target(struct mdesc_handle * hp,u64 arc)728 u64 mdesc_arc_target(struct mdesc_handle *hp, u64 arc)
729 {
730 struct mdesc_elem *ep, *base = node_block(&hp->mdesc);
731
732 ep = base + arc;
733
734 return ep->d.val;
735 }
736 EXPORT_SYMBOL(mdesc_arc_target);
737
mdesc_node_name(struct mdesc_handle * hp,u64 node)738 const char *mdesc_node_name(struct mdesc_handle *hp, u64 node)
739 {
740 struct mdesc_elem *ep, *base = node_block(&hp->mdesc);
741 const char *names = name_block(&hp->mdesc);
742 u64 last_node = hp->mdesc.node_sz / 16;
743
744 if (node == MDESC_NODE_NULL || node >= last_node)
745 return NULL;
746
747 ep = base + node;
748 if (ep->tag != MD_NODE)
749 return NULL;
750
751 return names + ep->name_offset;
752 }
753 EXPORT_SYMBOL(mdesc_node_name);
754
755 static u64 max_cpus = 64;
756
report_platform_properties(void)757 static void __init report_platform_properties(void)
758 {
759 struct mdesc_handle *hp = mdesc_grab();
760 u64 pn = mdesc_node_by_name(hp, MDESC_NODE_NULL, "platform");
761 const char *s;
762 const u64 *v;
763
764 if (pn == MDESC_NODE_NULL) {
765 prom_printf("No platform node in machine-description.\n");
766 prom_halt();
767 }
768
769 s = mdesc_get_property(hp, pn, "banner-name", NULL);
770 printk("PLATFORM: banner-name [%s]\n", s);
771 s = mdesc_get_property(hp, pn, "name", NULL);
772 printk("PLATFORM: name [%s]\n", s);
773
774 v = mdesc_get_property(hp, pn, "hostid", NULL);
775 if (v)
776 printk("PLATFORM: hostid [%08llx]\n", *v);
777 v = mdesc_get_property(hp, pn, "serial#", NULL);
778 if (v)
779 printk("PLATFORM: serial# [%08llx]\n", *v);
780 v = mdesc_get_property(hp, pn, "stick-frequency", NULL);
781 printk("PLATFORM: stick-frequency [%08llx]\n", *v);
782 v = mdesc_get_property(hp, pn, "mac-address", NULL);
783 if (v)
784 printk("PLATFORM: mac-address [%llx]\n", *v);
785 v = mdesc_get_property(hp, pn, "watchdog-resolution", NULL);
786 if (v)
787 printk("PLATFORM: watchdog-resolution [%llu ms]\n", *v);
788 v = mdesc_get_property(hp, pn, "watchdog-max-timeout", NULL);
789 if (v)
790 printk("PLATFORM: watchdog-max-timeout [%llu ms]\n", *v);
791 v = mdesc_get_property(hp, pn, "max-cpus", NULL);
792 if (v) {
793 max_cpus = *v;
794 printk("PLATFORM: max-cpus [%llu]\n", max_cpus);
795 }
796
797 #ifdef CONFIG_SMP
798 {
799 int max_cpu, i;
800
801 if (v) {
802 max_cpu = *v;
803 if (max_cpu > NR_CPUS)
804 max_cpu = NR_CPUS;
805 } else {
806 max_cpu = NR_CPUS;
807 }
808 for (i = 0; i < max_cpu; i++)
809 set_cpu_possible(i, true);
810 }
811 #endif
812
813 mdesc_release(hp);
814 }
815
fill_in_one_cache(cpuinfo_sparc * c,struct mdesc_handle * hp,u64 mp)816 static void fill_in_one_cache(cpuinfo_sparc *c, struct mdesc_handle *hp, u64 mp)
817 {
818 const u64 *level = mdesc_get_property(hp, mp, "level", NULL);
819 const u64 *size = mdesc_get_property(hp, mp, "size", NULL);
820 const u64 *line_size = mdesc_get_property(hp, mp, "line-size", NULL);
821 const char *type;
822 int type_len;
823
824 type = mdesc_get_property(hp, mp, "type", &type_len);
825
826 switch (*level) {
827 case 1:
828 if (of_find_in_proplist(type, "instn", type_len)) {
829 c->icache_size = *size;
830 c->icache_line_size = *line_size;
831 } else if (of_find_in_proplist(type, "data", type_len)) {
832 c->dcache_size = *size;
833 c->dcache_line_size = *line_size;
834 }
835 break;
836
837 case 2:
838 c->ecache_size = *size;
839 c->ecache_line_size = *line_size;
840 break;
841
842 default:
843 break;
844 }
845
846 if (*level == 1) {
847 u64 a;
848
849 mdesc_for_each_arc(a, hp, mp, MDESC_ARC_TYPE_FWD) {
850 u64 target = mdesc_arc_target(hp, a);
851 const char *name = mdesc_node_name(hp, target);
852
853 if (!strcmp(name, "cache"))
854 fill_in_one_cache(c, hp, target);
855 }
856 }
857 }
858
find_back_node_value(struct mdesc_handle * hp,u64 node,char * srch_val,void (* func)(struct mdesc_handle *,u64,int),u64 val,int depth)859 static void find_back_node_value(struct mdesc_handle *hp, u64 node,
860 char *srch_val,
861 void (*func)(struct mdesc_handle *, u64, int),
862 u64 val, int depth)
863 {
864 u64 arc;
865
866 /* Since we have an estimate of recursion depth, do a sanity check. */
867 if (depth == 0)
868 return;
869
870 mdesc_for_each_arc(arc, hp, node, MDESC_ARC_TYPE_BACK) {
871 u64 n = mdesc_arc_target(hp, arc);
872 const char *name = mdesc_node_name(hp, n);
873
874 if (!strcmp(srch_val, name))
875 (*func)(hp, n, val);
876
877 find_back_node_value(hp, n, srch_val, func, val, depth-1);
878 }
879 }
880
__mark_core_id(struct mdesc_handle * hp,u64 node,int core_id)881 static void __mark_core_id(struct mdesc_handle *hp, u64 node,
882 int core_id)
883 {
884 const u64 *id = mdesc_get_property(hp, node, "id", NULL);
885
886 if (*id < num_possible_cpus())
887 cpu_data(*id).core_id = core_id;
888 }
889
__mark_max_cache_id(struct mdesc_handle * hp,u64 node,int max_cache_id)890 static void __mark_max_cache_id(struct mdesc_handle *hp, u64 node,
891 int max_cache_id)
892 {
893 const u64 *id = mdesc_get_property(hp, node, "id", NULL);
894
895 if (*id < num_possible_cpus()) {
896 cpu_data(*id).max_cache_id = max_cache_id;
897
898 /**
899 * On systems without explicit socket descriptions socket
900 * is max_cache_id
901 */
902 cpu_data(*id).sock_id = max_cache_id;
903 }
904 }
905
mark_core_ids(struct mdesc_handle * hp,u64 mp,int core_id)906 static void mark_core_ids(struct mdesc_handle *hp, u64 mp,
907 int core_id)
908 {
909 find_back_node_value(hp, mp, "cpu", __mark_core_id, core_id, 10);
910 }
911
mark_max_cache_ids(struct mdesc_handle * hp,u64 mp,int max_cache_id)912 static void mark_max_cache_ids(struct mdesc_handle *hp, u64 mp,
913 int max_cache_id)
914 {
915 find_back_node_value(hp, mp, "cpu", __mark_max_cache_id,
916 max_cache_id, 10);
917 }
918
set_core_ids(struct mdesc_handle * hp)919 static void set_core_ids(struct mdesc_handle *hp)
920 {
921 int idx;
922 u64 mp;
923
924 idx = 1;
925
926 /* Identify unique cores by looking for cpus backpointed to by
927 * level 1 instruction caches.
928 */
929 mdesc_for_each_node_by_name(hp, mp, "cache") {
930 const u64 *level;
931 const char *type;
932 int len;
933
934 level = mdesc_get_property(hp, mp, "level", NULL);
935 if (*level != 1)
936 continue;
937
938 type = mdesc_get_property(hp, mp, "type", &len);
939 if (!of_find_in_proplist(type, "instn", len))
940 continue;
941
942 mark_core_ids(hp, mp, idx);
943 idx++;
944 }
945 }
946
set_max_cache_ids_by_cache(struct mdesc_handle * hp,int level)947 static int set_max_cache_ids_by_cache(struct mdesc_handle *hp, int level)
948 {
949 u64 mp;
950 int idx = 1;
951 int fnd = 0;
952
953 /**
954 * Identify unique highest level of shared cache by looking for cpus
955 * backpointed to by shared level N caches.
956 */
957 mdesc_for_each_node_by_name(hp, mp, "cache") {
958 const u64 *cur_lvl;
959
960 cur_lvl = mdesc_get_property(hp, mp, "level", NULL);
961 if (*cur_lvl != level)
962 continue;
963 mark_max_cache_ids(hp, mp, idx);
964 idx++;
965 fnd = 1;
966 }
967 return fnd;
968 }
969
set_sock_ids_by_socket(struct mdesc_handle * hp,u64 mp)970 static void set_sock_ids_by_socket(struct mdesc_handle *hp, u64 mp)
971 {
972 int idx = 1;
973
974 mdesc_for_each_node_by_name(hp, mp, "socket") {
975 u64 a;
976
977 mdesc_for_each_arc(a, hp, mp, MDESC_ARC_TYPE_FWD) {
978 u64 t = mdesc_arc_target(hp, a);
979 const char *name;
980 const u64 *id;
981
982 name = mdesc_node_name(hp, t);
983 if (strcmp(name, "cpu"))
984 continue;
985
986 id = mdesc_get_property(hp, t, "id", NULL);
987 if (*id < num_possible_cpus())
988 cpu_data(*id).sock_id = idx;
989 }
990 idx++;
991 }
992 }
993
set_sock_ids(struct mdesc_handle * hp)994 static void set_sock_ids(struct mdesc_handle *hp)
995 {
996 u64 mp;
997
998 /**
999 * Find the highest level of shared cache which pre-T7 is also
1000 * the socket.
1001 */
1002 if (!set_max_cache_ids_by_cache(hp, 3))
1003 set_max_cache_ids_by_cache(hp, 2);
1004
1005 /* If machine description exposes sockets data use it.*/
1006 mp = mdesc_node_by_name(hp, MDESC_NODE_NULL, "sockets");
1007 if (mp != MDESC_NODE_NULL)
1008 set_sock_ids_by_socket(hp, mp);
1009 }
1010
mark_proc_ids(struct mdesc_handle * hp,u64 mp,int proc_id)1011 static void mark_proc_ids(struct mdesc_handle *hp, u64 mp, int proc_id)
1012 {
1013 u64 a;
1014
1015 mdesc_for_each_arc(a, hp, mp, MDESC_ARC_TYPE_BACK) {
1016 u64 t = mdesc_arc_target(hp, a);
1017 const char *name;
1018 const u64 *id;
1019
1020 name = mdesc_node_name(hp, t);
1021 if (strcmp(name, "cpu"))
1022 continue;
1023
1024 id = mdesc_get_property(hp, t, "id", NULL);
1025 if (*id < NR_CPUS)
1026 cpu_data(*id).proc_id = proc_id;
1027 }
1028 }
1029
__set_proc_ids(struct mdesc_handle * hp,const char * exec_unit_name)1030 static void __set_proc_ids(struct mdesc_handle *hp, const char *exec_unit_name)
1031 {
1032 int idx;
1033 u64 mp;
1034
1035 idx = 0;
1036 mdesc_for_each_node_by_name(hp, mp, exec_unit_name) {
1037 const char *type;
1038 int len;
1039
1040 type = mdesc_get_property(hp, mp, "type", &len);
1041 if (!of_find_in_proplist(type, "int", len) &&
1042 !of_find_in_proplist(type, "integer", len))
1043 continue;
1044
1045 mark_proc_ids(hp, mp, idx);
1046 idx++;
1047 }
1048 }
1049
set_proc_ids(struct mdesc_handle * hp)1050 static void set_proc_ids(struct mdesc_handle *hp)
1051 {
1052 __set_proc_ids(hp, "exec_unit");
1053 __set_proc_ids(hp, "exec-unit");
1054 }
1055
get_one_mondo_bits(const u64 * p,unsigned int * mask,unsigned long def,unsigned long max)1056 static void get_one_mondo_bits(const u64 *p, unsigned int *mask,
1057 unsigned long def, unsigned long max)
1058 {
1059 u64 val;
1060
1061 if (!p)
1062 goto use_default;
1063 val = *p;
1064
1065 if (!val || val >= 64)
1066 goto use_default;
1067
1068 if (val > max)
1069 val = max;
1070
1071 *mask = ((1U << val) * 64U) - 1U;
1072 return;
1073
1074 use_default:
1075 *mask = ((1U << def) * 64U) - 1U;
1076 }
1077
get_mondo_data(struct mdesc_handle * hp,u64 mp,struct trap_per_cpu * tb)1078 static void get_mondo_data(struct mdesc_handle *hp, u64 mp,
1079 struct trap_per_cpu *tb)
1080 {
1081 static int printed;
1082 const u64 *val;
1083
1084 val = mdesc_get_property(hp, mp, "q-cpu-mondo-#bits", NULL);
1085 get_one_mondo_bits(val, &tb->cpu_mondo_qmask, 7, ilog2(max_cpus * 2));
1086
1087 val = mdesc_get_property(hp, mp, "q-dev-mondo-#bits", NULL);
1088 get_one_mondo_bits(val, &tb->dev_mondo_qmask, 7, 8);
1089
1090 val = mdesc_get_property(hp, mp, "q-resumable-#bits", NULL);
1091 get_one_mondo_bits(val, &tb->resum_qmask, 6, 7);
1092
1093 val = mdesc_get_property(hp, mp, "q-nonresumable-#bits", NULL);
1094 get_one_mondo_bits(val, &tb->nonresum_qmask, 2, 2);
1095 if (!printed++) {
1096 pr_info("SUN4V: Mondo queue sizes "
1097 "[cpu(%u) dev(%u) r(%u) nr(%u)]\n",
1098 tb->cpu_mondo_qmask + 1,
1099 tb->dev_mondo_qmask + 1,
1100 tb->resum_qmask + 1,
1101 tb->nonresum_qmask + 1);
1102 }
1103 }
1104
mdesc_iterate_over_cpus(void * (* func)(struct mdesc_handle *,u64,int,void *),void * arg,cpumask_t * mask)1105 static void *mdesc_iterate_over_cpus(void *(*func)(struct mdesc_handle *, u64, int, void *), void *arg, cpumask_t *mask)
1106 {
1107 struct mdesc_handle *hp = mdesc_grab();
1108 void *ret = NULL;
1109 u64 mp;
1110
1111 mdesc_for_each_node_by_name(hp, mp, "cpu") {
1112 const u64 *id = mdesc_get_property(hp, mp, "id", NULL);
1113 int cpuid = *id;
1114
1115 #ifdef CONFIG_SMP
1116 if (cpuid >= NR_CPUS) {
1117 printk(KERN_WARNING "Ignoring CPU %d which is "
1118 ">= NR_CPUS (%d)\n",
1119 cpuid, NR_CPUS);
1120 continue;
1121 }
1122 if (!cpumask_test_cpu(cpuid, mask))
1123 continue;
1124 #endif
1125
1126 ret = func(hp, mp, cpuid, arg);
1127 if (ret)
1128 goto out;
1129 }
1130 out:
1131 mdesc_release(hp);
1132 return ret;
1133 }
1134
record_one_cpu(struct mdesc_handle * hp,u64 mp,int cpuid,void * arg)1135 static void *record_one_cpu(struct mdesc_handle *hp, u64 mp, int cpuid,
1136 void *arg)
1137 {
1138 ncpus_probed++;
1139 #ifdef CONFIG_SMP
1140 set_cpu_present(cpuid, true);
1141 #endif
1142 return NULL;
1143 }
1144
mdesc_populate_present_mask(cpumask_t * mask)1145 void mdesc_populate_present_mask(cpumask_t *mask)
1146 {
1147 if (tlb_type != hypervisor)
1148 return;
1149
1150 ncpus_probed = 0;
1151 mdesc_iterate_over_cpus(record_one_cpu, NULL, mask);
1152 }
1153
check_one_pgsz(struct mdesc_handle * hp,u64 mp,int cpuid,void * arg)1154 static void * __init check_one_pgsz(struct mdesc_handle *hp, u64 mp, int cpuid, void *arg)
1155 {
1156 const u64 *pgsz_prop = mdesc_get_property(hp, mp, "mmu-page-size-list", NULL);
1157 unsigned long *pgsz_mask = arg;
1158 u64 val;
1159
1160 val = (HV_PGSZ_MASK_8K | HV_PGSZ_MASK_64K |
1161 HV_PGSZ_MASK_512K | HV_PGSZ_MASK_4MB);
1162 if (pgsz_prop)
1163 val = *pgsz_prop;
1164
1165 if (!*pgsz_mask)
1166 *pgsz_mask = val;
1167 else
1168 *pgsz_mask &= val;
1169 return NULL;
1170 }
1171
mdesc_get_page_sizes(cpumask_t * mask,unsigned long * pgsz_mask)1172 void __init mdesc_get_page_sizes(cpumask_t *mask, unsigned long *pgsz_mask)
1173 {
1174 *pgsz_mask = 0;
1175 mdesc_iterate_over_cpus(check_one_pgsz, pgsz_mask, mask);
1176 }
1177
fill_in_one_cpu(struct mdesc_handle * hp,u64 mp,int cpuid,void * arg)1178 static void *fill_in_one_cpu(struct mdesc_handle *hp, u64 mp, int cpuid,
1179 void *arg)
1180 {
1181 const u64 *cfreq = mdesc_get_property(hp, mp, "clock-frequency", NULL);
1182 struct trap_per_cpu *tb;
1183 cpuinfo_sparc *c;
1184 u64 a;
1185
1186 #ifndef CONFIG_SMP
1187 /* On uniprocessor we only want the values for the
1188 * real physical cpu the kernel booted onto, however
1189 * cpu_data() only has one entry at index 0.
1190 */
1191 if (cpuid != real_hard_smp_processor_id())
1192 return NULL;
1193 cpuid = 0;
1194 #endif
1195
1196 c = &cpu_data(cpuid);
1197 c->clock_tick = *cfreq;
1198
1199 tb = &trap_block[cpuid];
1200 get_mondo_data(hp, mp, tb);
1201
1202 mdesc_for_each_arc(a, hp, mp, MDESC_ARC_TYPE_FWD) {
1203 u64 j, t = mdesc_arc_target(hp, a);
1204 const char *t_name;
1205
1206 t_name = mdesc_node_name(hp, t);
1207 if (!strcmp(t_name, "cache")) {
1208 fill_in_one_cache(c, hp, t);
1209 continue;
1210 }
1211
1212 mdesc_for_each_arc(j, hp, t, MDESC_ARC_TYPE_FWD) {
1213 u64 n = mdesc_arc_target(hp, j);
1214 const char *n_name;
1215
1216 n_name = mdesc_node_name(hp, n);
1217 if (!strcmp(n_name, "cache"))
1218 fill_in_one_cache(c, hp, n);
1219 }
1220 }
1221
1222 c->core_id = 0;
1223 c->proc_id = -1;
1224
1225 return NULL;
1226 }
1227
mdesc_fill_in_cpu_data(cpumask_t * mask)1228 void mdesc_fill_in_cpu_data(cpumask_t *mask)
1229 {
1230 struct mdesc_handle *hp;
1231
1232 mdesc_iterate_over_cpus(fill_in_one_cpu, NULL, mask);
1233
1234 hp = mdesc_grab();
1235
1236 set_core_ids(hp);
1237 set_proc_ids(hp);
1238 set_sock_ids(hp);
1239
1240 mdesc_release(hp);
1241
1242 smp_fill_in_sib_core_maps();
1243 }
1244
1245 /* mdesc_open() - Grab a reference to mdesc_handle when /dev/mdesc is
1246 * opened. Hold this reference until /dev/mdesc is closed to ensure
1247 * mdesc data structure is not released underneath us. Store the
1248 * pointer to mdesc structure in private_data for read and seek to use
1249 */
mdesc_open(struct inode * inode,struct file * file)1250 static int mdesc_open(struct inode *inode, struct file *file)
1251 {
1252 struct mdesc_handle *hp = mdesc_grab();
1253
1254 if (!hp)
1255 return -ENODEV;
1256
1257 file->private_data = hp;
1258
1259 return 0;
1260 }
1261
mdesc_read(struct file * file,char __user * buf,size_t len,loff_t * offp)1262 static ssize_t mdesc_read(struct file *file, char __user *buf,
1263 size_t len, loff_t *offp)
1264 {
1265 struct mdesc_handle *hp = file->private_data;
1266 unsigned char *mdesc;
1267 int bytes_left, count = len;
1268
1269 if (*offp >= hp->handle_size)
1270 return 0;
1271
1272 bytes_left = hp->handle_size - *offp;
1273 if (count > bytes_left)
1274 count = bytes_left;
1275
1276 mdesc = (unsigned char *)&hp->mdesc;
1277 mdesc += *offp;
1278 if (!copy_to_user(buf, mdesc, count)) {
1279 *offp += count;
1280 return count;
1281 } else {
1282 return -EFAULT;
1283 }
1284 }
1285
mdesc_llseek(struct file * file,loff_t offset,int whence)1286 static loff_t mdesc_llseek(struct file *file, loff_t offset, int whence)
1287 {
1288 struct mdesc_handle *hp = file->private_data;
1289
1290 return no_seek_end_llseek_size(file, offset, whence, hp->handle_size);
1291 }
1292
1293 /* mdesc_close() - /dev/mdesc is being closed, release the reference to
1294 * mdesc structure.
1295 */
mdesc_close(struct inode * inode,struct file * file)1296 static int mdesc_close(struct inode *inode, struct file *file)
1297 {
1298 mdesc_release(file->private_data);
1299 return 0;
1300 }
1301
1302 static const struct file_operations mdesc_fops = {
1303 .open = mdesc_open,
1304 .read = mdesc_read,
1305 .llseek = mdesc_llseek,
1306 .release = mdesc_close,
1307 .owner = THIS_MODULE,
1308 };
1309
1310 static struct miscdevice mdesc_misc = {
1311 .minor = MISC_DYNAMIC_MINOR,
1312 .name = "mdesc",
1313 .fops = &mdesc_fops,
1314 };
1315
mdesc_misc_init(void)1316 static int __init mdesc_misc_init(void)
1317 {
1318 return misc_register(&mdesc_misc);
1319 }
1320
1321 __initcall(mdesc_misc_init);
1322
sun4v_mdesc_init(void)1323 void __init sun4v_mdesc_init(void)
1324 {
1325 struct mdesc_handle *hp;
1326 unsigned long len, real_len, status;
1327
1328 (void) sun4v_mach_desc(0UL, 0UL, &len);
1329
1330 printk("MDESC: Size is %lu bytes.\n", len);
1331
1332 hp = mdesc_alloc(len, &memblock_mdesc_ops);
1333 if (hp == NULL) {
1334 prom_printf("MDESC: alloc of %lu bytes failed.\n", len);
1335 prom_halt();
1336 }
1337
1338 status = sun4v_mach_desc(__pa(&hp->mdesc), len, &real_len);
1339 if (status != HV_EOK || real_len > len) {
1340 prom_printf("sun4v_mach_desc fails, err(%lu), "
1341 "len(%lu), real_len(%lu)\n",
1342 status, len, real_len);
1343 mdesc_free(hp);
1344 prom_halt();
1345 }
1346
1347 cur_mdesc = hp;
1348
1349 report_platform_properties();
1350 }
1351