Loading...
1/* mdesc.c: Sun4V machine description handling.
2 *
3 * Copyright (C) 2007, 2008 David S. Miller <davem@davemloft.net>
4 */
5#include <linux/kernel.h>
6#include <linux/types.h>
7#include <linux/memblock.h>
8#include <linux/log2.h>
9#include <linux/list.h>
10#include <linux/slab.h>
11#include <linux/mm.h>
12#include <linux/miscdevice.h>
13#include <linux/bootmem.h>
14#include <linux/export.h>
15
16#include <asm/cpudata.h>
17#include <asm/hypervisor.h>
18#include <asm/mdesc.h>
19#include <asm/prom.h>
20#include <asm/uaccess.h>
21#include <asm/oplib.h>
22#include <asm/smp.h>
23
24/* Unlike the OBP device tree, the machine description is a full-on
25 * DAG. An arbitrary number of ARCs are possible from one
26 * node to other nodes and thus we can't use the OBP device_node
27 * data structure to represent these nodes inside of the kernel.
28 *
29 * Actually, it isn't even a DAG, because there are back pointers
30 * which create cycles in the graph.
31 *
32 * mdesc_hdr and mdesc_elem describe the layout of the data structure
33 * we get from the Hypervisor.
34 */
35struct mdesc_hdr {
36 u32 version; /* Transport version */
37 u32 node_sz; /* node block size */
38 u32 name_sz; /* name block size */
39 u32 data_sz; /* data block size */
40} __attribute__((aligned(16)));
41
42struct mdesc_elem {
43 u8 tag;
44#define MD_LIST_END 0x00
45#define MD_NODE 0x4e
46#define MD_NODE_END 0x45
47#define MD_NOOP 0x20
48#define MD_PROP_ARC 0x61
49#define MD_PROP_VAL 0x76
50#define MD_PROP_STR 0x73
51#define MD_PROP_DATA 0x64
52 u8 name_len;
53 u16 resv;
54 u32 name_offset;
55 union {
56 struct {
57 u32 data_len;
58 u32 data_offset;
59 } data;
60 u64 val;
61 } d;
62};
63
64struct mdesc_mem_ops {
65 struct mdesc_handle *(*alloc)(unsigned int mdesc_size);
66 void (*free)(struct mdesc_handle *handle);
67};
68
69struct mdesc_handle {
70 struct list_head list;
71 struct mdesc_mem_ops *mops;
72 void *self_base;
73 atomic_t refcnt;
74 unsigned int handle_size;
75 struct mdesc_hdr mdesc;
76};
77
78static void mdesc_handle_init(struct mdesc_handle *hp,
79 unsigned int handle_size,
80 void *base)
81{
82 BUG_ON(((unsigned long)&hp->mdesc) & (16UL - 1));
83
84 memset(hp, 0, handle_size);
85 INIT_LIST_HEAD(&hp->list);
86 hp->self_base = base;
87 atomic_set(&hp->refcnt, 1);
88 hp->handle_size = handle_size;
89}
90
91static struct mdesc_handle * __init mdesc_memblock_alloc(unsigned int mdesc_size)
92{
93 unsigned int handle_size, alloc_size;
94 struct mdesc_handle *hp;
95 unsigned long paddr;
96
97 handle_size = (sizeof(struct mdesc_handle) -
98 sizeof(struct mdesc_hdr) +
99 mdesc_size);
100 alloc_size = PAGE_ALIGN(handle_size);
101
102 paddr = memblock_alloc(alloc_size, PAGE_SIZE);
103
104 hp = NULL;
105 if (paddr) {
106 hp = __va(paddr);
107 mdesc_handle_init(hp, handle_size, hp);
108 }
109 return hp;
110}
111
112static void __init mdesc_memblock_free(struct mdesc_handle *hp)
113{
114 unsigned int alloc_size;
115 unsigned long start;
116
117 BUG_ON(atomic_read(&hp->refcnt) != 0);
118 BUG_ON(!list_empty(&hp->list));
119
120 alloc_size = PAGE_ALIGN(hp->handle_size);
121 start = __pa(hp);
122 free_bootmem_late(start, alloc_size);
123}
124
125static struct mdesc_mem_ops memblock_mdesc_ops = {
126 .alloc = mdesc_memblock_alloc,
127 .free = mdesc_memblock_free,
128};
129
130static struct mdesc_handle *mdesc_kmalloc(unsigned int mdesc_size)
131{
132 unsigned int handle_size;
133 void *base;
134
135 handle_size = (sizeof(struct mdesc_handle) -
136 sizeof(struct mdesc_hdr) +
137 mdesc_size);
138
139 base = kmalloc(handle_size + 15, GFP_KERNEL | __GFP_NOFAIL);
140 if (base) {
141 struct mdesc_handle *hp;
142 unsigned long addr;
143
144 addr = (unsigned long)base;
145 addr = (addr + 15UL) & ~15UL;
146 hp = (struct mdesc_handle *) addr;
147
148 mdesc_handle_init(hp, handle_size, base);
149 return hp;
150 }
151
152 return NULL;
153}
154
155static void mdesc_kfree(struct mdesc_handle *hp)
156{
157 BUG_ON(atomic_read(&hp->refcnt) != 0);
158 BUG_ON(!list_empty(&hp->list));
159
160 kfree(hp->self_base);
161}
162
163static struct mdesc_mem_ops kmalloc_mdesc_memops = {
164 .alloc = mdesc_kmalloc,
165 .free = mdesc_kfree,
166};
167
168static struct mdesc_handle *mdesc_alloc(unsigned int mdesc_size,
169 struct mdesc_mem_ops *mops)
170{
171 struct mdesc_handle *hp = mops->alloc(mdesc_size);
172
173 if (hp)
174 hp->mops = mops;
175
176 return hp;
177}
178
179static void mdesc_free(struct mdesc_handle *hp)
180{
181 hp->mops->free(hp);
182}
183
184static struct mdesc_handle *cur_mdesc;
185static LIST_HEAD(mdesc_zombie_list);
186static DEFINE_SPINLOCK(mdesc_lock);
187
188struct mdesc_handle *mdesc_grab(void)
189{
190 struct mdesc_handle *hp;
191 unsigned long flags;
192
193 spin_lock_irqsave(&mdesc_lock, flags);
194 hp = cur_mdesc;
195 if (hp)
196 atomic_inc(&hp->refcnt);
197 spin_unlock_irqrestore(&mdesc_lock, flags);
198
199 return hp;
200}
201EXPORT_SYMBOL(mdesc_grab);
202
203void mdesc_release(struct mdesc_handle *hp)
204{
205 unsigned long flags;
206
207 spin_lock_irqsave(&mdesc_lock, flags);
208 if (atomic_dec_and_test(&hp->refcnt)) {
209 list_del_init(&hp->list);
210 hp->mops->free(hp);
211 }
212 spin_unlock_irqrestore(&mdesc_lock, flags);
213}
214EXPORT_SYMBOL(mdesc_release);
215
216static DEFINE_MUTEX(mdesc_mutex);
217static struct mdesc_notifier_client *client_list;
218
219void mdesc_register_notifier(struct mdesc_notifier_client *client)
220{
221 u64 node;
222
223 mutex_lock(&mdesc_mutex);
224 client->next = client_list;
225 client_list = client;
226
227 mdesc_for_each_node_by_name(cur_mdesc, node, client->node_name)
228 client->add(cur_mdesc, node);
229
230 mutex_unlock(&mdesc_mutex);
231}
232
233static const u64 *parent_cfg_handle(struct mdesc_handle *hp, u64 node)
234{
235 const u64 *id;
236 u64 a;
237
238 id = NULL;
239 mdesc_for_each_arc(a, hp, node, MDESC_ARC_TYPE_BACK) {
240 u64 target;
241
242 target = mdesc_arc_target(hp, a);
243 id = mdesc_get_property(hp, target,
244 "cfg-handle", NULL);
245 if (id)
246 break;
247 }
248
249 return id;
250}
251
252/* Run 'func' on nodes which are in A but not in B. */
253static void invoke_on_missing(const char *name,
254 struct mdesc_handle *a,
255 struct mdesc_handle *b,
256 void (*func)(struct mdesc_handle *, u64))
257{
258 u64 node;
259
260 mdesc_for_each_node_by_name(a, node, name) {
261 int found = 0, is_vdc_port = 0;
262 const char *name_prop;
263 const u64 *id;
264 u64 fnode;
265
266 name_prop = mdesc_get_property(a, node, "name", NULL);
267 if (name_prop && !strcmp(name_prop, "vdc-port")) {
268 is_vdc_port = 1;
269 id = parent_cfg_handle(a, node);
270 } else
271 id = mdesc_get_property(a, node, "id", NULL);
272
273 if (!id) {
274 printk(KERN_ERR "MD: Cannot find ID for %s node.\n",
275 (name_prop ? name_prop : name));
276 continue;
277 }
278
279 mdesc_for_each_node_by_name(b, fnode, name) {
280 const u64 *fid;
281
282 if (is_vdc_port) {
283 name_prop = mdesc_get_property(b, fnode,
284 "name", NULL);
285 if (!name_prop ||
286 strcmp(name_prop, "vdc-port"))
287 continue;
288 fid = parent_cfg_handle(b, fnode);
289 if (!fid) {
290 printk(KERN_ERR "MD: Cannot find ID "
291 "for vdc-port node.\n");
292 continue;
293 }
294 } else
295 fid = mdesc_get_property(b, fnode,
296 "id", NULL);
297
298 if (*id == *fid) {
299 found = 1;
300 break;
301 }
302 }
303 if (!found)
304 func(a, node);
305 }
306}
307
308static void notify_one(struct mdesc_notifier_client *p,
309 struct mdesc_handle *old_hp,
310 struct mdesc_handle *new_hp)
311{
312 invoke_on_missing(p->node_name, old_hp, new_hp, p->remove);
313 invoke_on_missing(p->node_name, new_hp, old_hp, p->add);
314}
315
316static void mdesc_notify_clients(struct mdesc_handle *old_hp,
317 struct mdesc_handle *new_hp)
318{
319 struct mdesc_notifier_client *p = client_list;
320
321 while (p) {
322 notify_one(p, old_hp, new_hp);
323 p = p->next;
324 }
325}
326
327void mdesc_update(void)
328{
329 unsigned long len, real_len, status;
330 struct mdesc_handle *hp, *orig_hp;
331 unsigned long flags;
332
333 mutex_lock(&mdesc_mutex);
334
335 (void) sun4v_mach_desc(0UL, 0UL, &len);
336
337 hp = mdesc_alloc(len, &kmalloc_mdesc_memops);
338 if (!hp) {
339 printk(KERN_ERR "MD: mdesc alloc fails\n");
340 goto out;
341 }
342
343 status = sun4v_mach_desc(__pa(&hp->mdesc), len, &real_len);
344 if (status != HV_EOK || real_len > len) {
345 printk(KERN_ERR "MD: mdesc reread fails with %lu\n",
346 status);
347 atomic_dec(&hp->refcnt);
348 mdesc_free(hp);
349 goto out;
350 }
351
352 spin_lock_irqsave(&mdesc_lock, flags);
353 orig_hp = cur_mdesc;
354 cur_mdesc = hp;
355 spin_unlock_irqrestore(&mdesc_lock, flags);
356
357 mdesc_notify_clients(orig_hp, hp);
358
359 spin_lock_irqsave(&mdesc_lock, flags);
360 if (atomic_dec_and_test(&orig_hp->refcnt))
361 mdesc_free(orig_hp);
362 else
363 list_add(&orig_hp->list, &mdesc_zombie_list);
364 spin_unlock_irqrestore(&mdesc_lock, flags);
365
366out:
367 mutex_unlock(&mdesc_mutex);
368}
369
370static struct mdesc_elem *node_block(struct mdesc_hdr *mdesc)
371{
372 return (struct mdesc_elem *) (mdesc + 1);
373}
374
375static void *name_block(struct mdesc_hdr *mdesc)
376{
377 return ((void *) node_block(mdesc)) + mdesc->node_sz;
378}
379
380static void *data_block(struct mdesc_hdr *mdesc)
381{
382 return ((void *) name_block(mdesc)) + mdesc->name_sz;
383}
384
385u64 mdesc_node_by_name(struct mdesc_handle *hp,
386 u64 from_node, const char *name)
387{
388 struct mdesc_elem *ep = node_block(&hp->mdesc);
389 const char *names = name_block(&hp->mdesc);
390 u64 last_node = hp->mdesc.node_sz / 16;
391 u64 ret;
392
393 if (from_node == MDESC_NODE_NULL) {
394 ret = from_node = 0;
395 } else if (from_node >= last_node) {
396 return MDESC_NODE_NULL;
397 } else {
398 ret = ep[from_node].d.val;
399 }
400
401 while (ret < last_node) {
402 if (ep[ret].tag != MD_NODE)
403 return MDESC_NODE_NULL;
404 if (!strcmp(names + ep[ret].name_offset, name))
405 break;
406 ret = ep[ret].d.val;
407 }
408 if (ret >= last_node)
409 ret = MDESC_NODE_NULL;
410 return ret;
411}
412EXPORT_SYMBOL(mdesc_node_by_name);
413
414const void *mdesc_get_property(struct mdesc_handle *hp, u64 node,
415 const char *name, int *lenp)
416{
417 const char *names = name_block(&hp->mdesc);
418 u64 last_node = hp->mdesc.node_sz / 16;
419 void *data = data_block(&hp->mdesc);
420 struct mdesc_elem *ep;
421
422 if (node == MDESC_NODE_NULL || node >= last_node)
423 return NULL;
424
425 ep = node_block(&hp->mdesc) + node;
426 ep++;
427 for (; ep->tag != MD_NODE_END; ep++) {
428 void *val = NULL;
429 int len = 0;
430
431 switch (ep->tag) {
432 case MD_PROP_VAL:
433 val = &ep->d.val;
434 len = 8;
435 break;
436
437 case MD_PROP_STR:
438 case MD_PROP_DATA:
439 val = data + ep->d.data.data_offset;
440 len = ep->d.data.data_len;
441 break;
442
443 default:
444 break;
445 }
446 if (!val)
447 continue;
448
449 if (!strcmp(names + ep->name_offset, name)) {
450 if (lenp)
451 *lenp = len;
452 return val;
453 }
454 }
455
456 return NULL;
457}
458EXPORT_SYMBOL(mdesc_get_property);
459
460u64 mdesc_next_arc(struct mdesc_handle *hp, u64 from, const char *arc_type)
461{
462 struct mdesc_elem *ep, *base = node_block(&hp->mdesc);
463 const char *names = name_block(&hp->mdesc);
464 u64 last_node = hp->mdesc.node_sz / 16;
465
466 if (from == MDESC_NODE_NULL || from >= last_node)
467 return MDESC_NODE_NULL;
468
469 ep = base + from;
470
471 ep++;
472 for (; ep->tag != MD_NODE_END; ep++) {
473 if (ep->tag != MD_PROP_ARC)
474 continue;
475
476 if (strcmp(names + ep->name_offset, arc_type))
477 continue;
478
479 return ep - base;
480 }
481
482 return MDESC_NODE_NULL;
483}
484EXPORT_SYMBOL(mdesc_next_arc);
485
486u64 mdesc_arc_target(struct mdesc_handle *hp, u64 arc)
487{
488 struct mdesc_elem *ep, *base = node_block(&hp->mdesc);
489
490 ep = base + arc;
491
492 return ep->d.val;
493}
494EXPORT_SYMBOL(mdesc_arc_target);
495
496const char *mdesc_node_name(struct mdesc_handle *hp, u64 node)
497{
498 struct mdesc_elem *ep, *base = node_block(&hp->mdesc);
499 const char *names = name_block(&hp->mdesc);
500 u64 last_node = hp->mdesc.node_sz / 16;
501
502 if (node == MDESC_NODE_NULL || node >= last_node)
503 return NULL;
504
505 ep = base + node;
506 if (ep->tag != MD_NODE)
507 return NULL;
508
509 return names + ep->name_offset;
510}
511EXPORT_SYMBOL(mdesc_node_name);
512
513static u64 max_cpus = 64;
514
515static void __init report_platform_properties(void)
516{
517 struct mdesc_handle *hp = mdesc_grab();
518 u64 pn = mdesc_node_by_name(hp, MDESC_NODE_NULL, "platform");
519 const char *s;
520 const u64 *v;
521
522 if (pn == MDESC_NODE_NULL) {
523 prom_printf("No platform node in machine-description.\n");
524 prom_halt();
525 }
526
527 s = mdesc_get_property(hp, pn, "banner-name", NULL);
528 printk("PLATFORM: banner-name [%s]\n", s);
529 s = mdesc_get_property(hp, pn, "name", NULL);
530 printk("PLATFORM: name [%s]\n", s);
531
532 v = mdesc_get_property(hp, pn, "hostid", NULL);
533 if (v)
534 printk("PLATFORM: hostid [%08llx]\n", *v);
535 v = mdesc_get_property(hp, pn, "serial#", NULL);
536 if (v)
537 printk("PLATFORM: serial# [%08llx]\n", *v);
538 v = mdesc_get_property(hp, pn, "stick-frequency", NULL);
539 printk("PLATFORM: stick-frequency [%08llx]\n", *v);
540 v = mdesc_get_property(hp, pn, "mac-address", NULL);
541 if (v)
542 printk("PLATFORM: mac-address [%llx]\n", *v);
543 v = mdesc_get_property(hp, pn, "watchdog-resolution", NULL);
544 if (v)
545 printk("PLATFORM: watchdog-resolution [%llu ms]\n", *v);
546 v = mdesc_get_property(hp, pn, "watchdog-max-timeout", NULL);
547 if (v)
548 printk("PLATFORM: watchdog-max-timeout [%llu ms]\n", *v);
549 v = mdesc_get_property(hp, pn, "max-cpus", NULL);
550 if (v) {
551 max_cpus = *v;
552 printk("PLATFORM: max-cpus [%llu]\n", max_cpus);
553 }
554
555#ifdef CONFIG_SMP
556 {
557 int max_cpu, i;
558
559 if (v) {
560 max_cpu = *v;
561 if (max_cpu > NR_CPUS)
562 max_cpu = NR_CPUS;
563 } else {
564 max_cpu = NR_CPUS;
565 }
566 for (i = 0; i < max_cpu; i++)
567 set_cpu_possible(i, true);
568 }
569#endif
570
571 mdesc_release(hp);
572}
573
574static void __cpuinit fill_in_one_cache(cpuinfo_sparc *c,
575 struct mdesc_handle *hp,
576 u64 mp)
577{
578 const u64 *level = mdesc_get_property(hp, mp, "level", NULL);
579 const u64 *size = mdesc_get_property(hp, mp, "size", NULL);
580 const u64 *line_size = mdesc_get_property(hp, mp, "line-size", NULL);
581 const char *type;
582 int type_len;
583
584 type = mdesc_get_property(hp, mp, "type", &type_len);
585
586 switch (*level) {
587 case 1:
588 if (of_find_in_proplist(type, "instn", type_len)) {
589 c->icache_size = *size;
590 c->icache_line_size = *line_size;
591 } else if (of_find_in_proplist(type, "data", type_len)) {
592 c->dcache_size = *size;
593 c->dcache_line_size = *line_size;
594 }
595 break;
596
597 case 2:
598 c->ecache_size = *size;
599 c->ecache_line_size = *line_size;
600 break;
601
602 default:
603 break;
604 }
605
606 if (*level == 1) {
607 u64 a;
608
609 mdesc_for_each_arc(a, hp, mp, MDESC_ARC_TYPE_FWD) {
610 u64 target = mdesc_arc_target(hp, a);
611 const char *name = mdesc_node_name(hp, target);
612
613 if (!strcmp(name, "cache"))
614 fill_in_one_cache(c, hp, target);
615 }
616 }
617}
618
619static void __cpuinit mark_core_ids(struct mdesc_handle *hp, u64 mp, int core_id)
620{
621 u64 a;
622
623 mdesc_for_each_arc(a, hp, mp, MDESC_ARC_TYPE_BACK) {
624 u64 t = mdesc_arc_target(hp, a);
625 const char *name;
626 const u64 *id;
627
628 name = mdesc_node_name(hp, t);
629 if (!strcmp(name, "cpu")) {
630 id = mdesc_get_property(hp, t, "id", NULL);
631 if (*id < NR_CPUS)
632 cpu_data(*id).core_id = core_id;
633 } else {
634 u64 j;
635
636 mdesc_for_each_arc(j, hp, t, MDESC_ARC_TYPE_BACK) {
637 u64 n = mdesc_arc_target(hp, j);
638 const char *n_name;
639
640 n_name = mdesc_node_name(hp, n);
641 if (strcmp(n_name, "cpu"))
642 continue;
643
644 id = mdesc_get_property(hp, n, "id", NULL);
645 if (*id < NR_CPUS)
646 cpu_data(*id).core_id = core_id;
647 }
648 }
649 }
650}
651
652static void __cpuinit set_core_ids(struct mdesc_handle *hp)
653{
654 int idx;
655 u64 mp;
656
657 idx = 1;
658 mdesc_for_each_node_by_name(hp, mp, "cache") {
659 const u64 *level;
660 const char *type;
661 int len;
662
663 level = mdesc_get_property(hp, mp, "level", NULL);
664 if (*level != 1)
665 continue;
666
667 type = mdesc_get_property(hp, mp, "type", &len);
668 if (!of_find_in_proplist(type, "instn", len))
669 continue;
670
671 mark_core_ids(hp, mp, idx);
672
673 idx++;
674 }
675}
676
677static void __cpuinit mark_proc_ids(struct mdesc_handle *hp, u64 mp, int proc_id)
678{
679 u64 a;
680
681 mdesc_for_each_arc(a, hp, mp, MDESC_ARC_TYPE_BACK) {
682 u64 t = mdesc_arc_target(hp, a);
683 const char *name;
684 const u64 *id;
685
686 name = mdesc_node_name(hp, t);
687 if (strcmp(name, "cpu"))
688 continue;
689
690 id = mdesc_get_property(hp, t, "id", NULL);
691 if (*id < NR_CPUS)
692 cpu_data(*id).proc_id = proc_id;
693 }
694}
695
696static void __cpuinit __set_proc_ids(struct mdesc_handle *hp, const char *exec_unit_name)
697{
698 int idx;
699 u64 mp;
700
701 idx = 0;
702 mdesc_for_each_node_by_name(hp, mp, exec_unit_name) {
703 const char *type;
704 int len;
705
706 type = mdesc_get_property(hp, mp, "type", &len);
707 if (!of_find_in_proplist(type, "int", len) &&
708 !of_find_in_proplist(type, "integer", len))
709 continue;
710
711 mark_proc_ids(hp, mp, idx);
712
713 idx++;
714 }
715}
716
717static void __cpuinit set_proc_ids(struct mdesc_handle *hp)
718{
719 __set_proc_ids(hp, "exec_unit");
720 __set_proc_ids(hp, "exec-unit");
721}
722
723static void __cpuinit get_one_mondo_bits(const u64 *p, unsigned int *mask,
724 unsigned long def, unsigned long max)
725{
726 u64 val;
727
728 if (!p)
729 goto use_default;
730 val = *p;
731
732 if (!val || val >= 64)
733 goto use_default;
734
735 if (val > max)
736 val = max;
737
738 *mask = ((1U << val) * 64U) - 1U;
739 return;
740
741use_default:
742 *mask = ((1U << def) * 64U) - 1U;
743}
744
745static void __cpuinit get_mondo_data(struct mdesc_handle *hp, u64 mp,
746 struct trap_per_cpu *tb)
747{
748 static int printed;
749 const u64 *val;
750
751 val = mdesc_get_property(hp, mp, "q-cpu-mondo-#bits", NULL);
752 get_one_mondo_bits(val, &tb->cpu_mondo_qmask, 7, ilog2(max_cpus * 2));
753
754 val = mdesc_get_property(hp, mp, "q-dev-mondo-#bits", NULL);
755 get_one_mondo_bits(val, &tb->dev_mondo_qmask, 7, 8);
756
757 val = mdesc_get_property(hp, mp, "q-resumable-#bits", NULL);
758 get_one_mondo_bits(val, &tb->resum_qmask, 6, 7);
759
760 val = mdesc_get_property(hp, mp, "q-nonresumable-#bits", NULL);
761 get_one_mondo_bits(val, &tb->nonresum_qmask, 2, 2);
762 if (!printed++) {
763 pr_info("SUN4V: Mondo queue sizes "
764 "[cpu(%u) dev(%u) r(%u) nr(%u)]\n",
765 tb->cpu_mondo_qmask + 1,
766 tb->dev_mondo_qmask + 1,
767 tb->resum_qmask + 1,
768 tb->nonresum_qmask + 1);
769 }
770}
771
772static void * __cpuinit mdesc_iterate_over_cpus(void *(*func)(struct mdesc_handle *, u64, int, void *), void *arg, cpumask_t *mask)
773{
774 struct mdesc_handle *hp = mdesc_grab();
775 void *ret = NULL;
776 u64 mp;
777
778 mdesc_for_each_node_by_name(hp, mp, "cpu") {
779 const u64 *id = mdesc_get_property(hp, mp, "id", NULL);
780 int cpuid = *id;
781
782#ifdef CONFIG_SMP
783 if (cpuid >= NR_CPUS) {
784 printk(KERN_WARNING "Ignoring CPU %d which is "
785 ">= NR_CPUS (%d)\n",
786 cpuid, NR_CPUS);
787 continue;
788 }
789 if (!cpumask_test_cpu(cpuid, mask))
790 continue;
791#endif
792
793 ret = func(hp, mp, cpuid, arg);
794 if (ret)
795 goto out;
796 }
797out:
798 mdesc_release(hp);
799 return ret;
800}
801
802static void * __cpuinit record_one_cpu(struct mdesc_handle *hp, u64 mp, int cpuid, void *arg)
803{
804 ncpus_probed++;
805#ifdef CONFIG_SMP
806 set_cpu_present(cpuid, true);
807#endif
808 return NULL;
809}
810
811void __cpuinit mdesc_populate_present_mask(cpumask_t *mask)
812{
813 if (tlb_type != hypervisor)
814 return;
815
816 ncpus_probed = 0;
817 mdesc_iterate_over_cpus(record_one_cpu, NULL, mask);
818}
819
820static void * __cpuinit fill_in_one_cpu(struct mdesc_handle *hp, u64 mp, int cpuid, void *arg)
821{
822 const u64 *cfreq = mdesc_get_property(hp, mp, "clock-frequency", NULL);
823 struct trap_per_cpu *tb;
824 cpuinfo_sparc *c;
825 u64 a;
826
827#ifndef CONFIG_SMP
828 /* On uniprocessor we only want the values for the
829 * real physical cpu the kernel booted onto, however
830 * cpu_data() only has one entry at index 0.
831 */
832 if (cpuid != real_hard_smp_processor_id())
833 return NULL;
834 cpuid = 0;
835#endif
836
837 c = &cpu_data(cpuid);
838 c->clock_tick = *cfreq;
839
840 tb = &trap_block[cpuid];
841 get_mondo_data(hp, mp, tb);
842
843 mdesc_for_each_arc(a, hp, mp, MDESC_ARC_TYPE_FWD) {
844 u64 j, t = mdesc_arc_target(hp, a);
845 const char *t_name;
846
847 t_name = mdesc_node_name(hp, t);
848 if (!strcmp(t_name, "cache")) {
849 fill_in_one_cache(c, hp, t);
850 continue;
851 }
852
853 mdesc_for_each_arc(j, hp, t, MDESC_ARC_TYPE_FWD) {
854 u64 n = mdesc_arc_target(hp, j);
855 const char *n_name;
856
857 n_name = mdesc_node_name(hp, n);
858 if (!strcmp(n_name, "cache"))
859 fill_in_one_cache(c, hp, n);
860 }
861 }
862
863 c->core_id = 0;
864 c->proc_id = -1;
865
866 return NULL;
867}
868
869void __cpuinit mdesc_fill_in_cpu_data(cpumask_t *mask)
870{
871 struct mdesc_handle *hp;
872
873 mdesc_iterate_over_cpus(fill_in_one_cpu, NULL, mask);
874
875#ifdef CONFIG_SMP
876 sparc64_multi_core = 1;
877#endif
878
879 hp = mdesc_grab();
880
881 set_core_ids(hp);
882 set_proc_ids(hp);
883
884 mdesc_release(hp);
885
886 smp_fill_in_sib_core_maps();
887}
888
889static ssize_t mdesc_read(struct file *file, char __user *buf,
890 size_t len, loff_t *offp)
891{
892 struct mdesc_handle *hp = mdesc_grab();
893 int err;
894
895 if (!hp)
896 return -ENODEV;
897
898 err = hp->handle_size;
899 if (len < hp->handle_size)
900 err = -EMSGSIZE;
901 else if (copy_to_user(buf, &hp->mdesc, hp->handle_size))
902 err = -EFAULT;
903 mdesc_release(hp);
904
905 return err;
906}
907
908static const struct file_operations mdesc_fops = {
909 .read = mdesc_read,
910 .owner = THIS_MODULE,
911 .llseek = noop_llseek,
912};
913
914static struct miscdevice mdesc_misc = {
915 .minor = MISC_DYNAMIC_MINOR,
916 .name = "mdesc",
917 .fops = &mdesc_fops,
918};
919
920static int __init mdesc_misc_init(void)
921{
922 return misc_register(&mdesc_misc);
923}
924
925__initcall(mdesc_misc_init);
926
927void __init sun4v_mdesc_init(void)
928{
929 struct mdesc_handle *hp;
930 unsigned long len, real_len, status;
931
932 (void) sun4v_mach_desc(0UL, 0UL, &len);
933
934 printk("MDESC: Size is %lu bytes.\n", len);
935
936 hp = mdesc_alloc(len, &memblock_mdesc_ops);
937 if (hp == NULL) {
938 prom_printf("MDESC: alloc of %lu bytes failed.\n", len);
939 prom_halt();
940 }
941
942 status = sun4v_mach_desc(__pa(&hp->mdesc), len, &real_len);
943 if (status != HV_EOK || real_len > len) {
944 prom_printf("sun4v_mach_desc fails, err(%lu), "
945 "len(%lu), real_len(%lu)\n",
946 status, len, real_len);
947 mdesc_free(hp);
948 prom_halt();
949 }
950
951 cur_mdesc = hp;
952
953 report_platform_properties();
954}
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 */
37struct 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
45struct 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
67struct mdesc_mem_ops {
68 struct mdesc_handle *(*alloc)(unsigned int mdesc_size);
69 void (*free)(struct mdesc_handle *handle);
70};
71
72struct 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
81typedef int (*mdesc_node_info_get_f)(struct mdesc_handle *, u64,
82 union md_node_info *);
83typedef void (*mdesc_node_info_rel_f)(union md_node_info *);
84typedef bool (*mdesc_node_match_f)(union md_node_info *, union md_node_info *);
85
86struct 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
93static int get_vdev_port_node_info(struct mdesc_handle *md, u64 node,
94 union md_node_info *node_info);
95static void rel_vdev_port_node_info(union md_node_info *node_info);
96static bool vdev_port_node_match(union md_node_info *a_node_info,
97 union md_node_info *b_node_info);
98
99static int get_ds_port_node_info(struct mdesc_handle *md, u64 node,
100 union md_node_info *node_info);
101static void rel_ds_port_node_info(union md_node_info *node_info);
102static 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 */
106static 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
114static 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
149static 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
162static 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
183static 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
196static struct mdesc_mem_ops memblock_mdesc_ops = {
197 .alloc = mdesc_memblock_alloc,
198 .free = mdesc_memblock_free,
199};
200
201static 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
224static 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
232static struct mdesc_mem_ops kmalloc_mdesc_memops = {
233 .alloc = mdesc_kmalloc,
234 .free = mdesc_kfree,
235};
236
237static 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
248static void mdesc_free(struct mdesc_handle *hp)
249{
250 hp->mops->free(hp);
251}
252
253static struct mdesc_handle *cur_mdesc;
254static LIST_HEAD(mdesc_zombie_list);
255static DEFINE_SPINLOCK(mdesc_lock);
256
257struct 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}
270EXPORT_SYMBOL(mdesc_grab);
271
272void 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}
283EXPORT_SYMBOL(mdesc_release);
284
285static DEFINE_MUTEX(mdesc_mutex);
286static struct mdesc_notifier_client *client_list;
287
288void 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
319static 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
338static 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
367static 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
375static 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
392static 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
407static void rel_ds_port_node_info(union md_node_info *node_info)
408{
409}
410
411static 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. */
421static 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
482static 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
490static 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
501void 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
540out:
541 mutex_unlock(&mdesc_mutex);
542}
543
544u64 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}
582EXPORT_SYMBOL(mdesc_get_node);
583
584int 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}
612EXPORT_SYMBOL(mdesc_get_node_info);
613
614static struct mdesc_elem *node_block(struct mdesc_hdr *mdesc)
615{
616 return (struct mdesc_elem *) mdesc->data;
617}
618
619static void *name_block(struct mdesc_hdr *mdesc)
620{
621 return ((void *) node_block(mdesc)) + mdesc->node_sz;
622}
623
624static void *data_block(struct mdesc_hdr *mdesc)
625{
626 return ((void *) name_block(mdesc)) + mdesc->name_sz;
627}
628
629u64 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}
656EXPORT_SYMBOL(mdesc_node_by_name);
657
658const 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}
702EXPORT_SYMBOL(mdesc_get_property);
703
704u64 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}
728EXPORT_SYMBOL(mdesc_next_arc);
729
730u64 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}
738EXPORT_SYMBOL(mdesc_arc_target);
739
740const 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}
755EXPORT_SYMBOL(mdesc_node_name);
756
757static u64 max_cpus = 64;
758
759static 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
818static 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
861static 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
883static 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
892static 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
908static 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
914static 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
921static 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
949static 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
972static 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
996static 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
1013static 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
1032static 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
1052static 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
1058static 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
1076use_default:
1077 *mask = ((1U << def) * 64U) - 1U;
1078}
1079
1080static 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
1107static 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 }
1132out:
1133 mdesc_release(hp);
1134 return ret;
1135}
1136
1137static 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
1147void 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
1156static 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
1174void __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
1180static 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
1230void 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 */
1252static 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
1264static 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
1288static 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 */
1298static int mdesc_close(struct inode *inode, struct file *file)
1299{
1300 mdesc_release(file->private_data);
1301 return 0;
1302}
1303
1304static 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
1312static struct miscdevice mdesc_misc = {
1313 .minor = MISC_DYNAMIC_MINOR,
1314 .name = "mdesc",
1315 .fops = &mdesc_fops,
1316};
1317
1318static int __init mdesc_misc_init(void)
1319{
1320 return misc_register(&mdesc_misc);
1321}
1322
1323__initcall(mdesc_misc_init);
1324
1325void __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}