1 // SPDX-License-Identifier: GPL-2.0
2 /* mdesc.c: Sun4V machine description handling.
4 * Copyright (C) 2007, 2008 David S. Miller <davem@davemloft.net>
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>
13 #include <linux/miscdevice.h>
14 #include <linux/bootmem.h>
15 #include <linux/export.h>
16 #include <linux/refcount.h>
18 #include <asm/cpudata.h>
19 #include <asm/hypervisor.h>
20 #include <asm/mdesc.h>
22 #include <linux/uaccess.h>
23 #include <asm/oplib.h>
27 /* Unlike the OBP device tree, the machine description is a full-on
28 * DAG. An arbitrary number of ARCs are possible from one
29 * node to other nodes and thus we can't use the OBP device_node
30 * data structure to represent these nodes inside of the kernel.
32 * Actually, it isn't even a DAG, because there are back pointers
33 * which create cycles in the graph.
35 * mdesc_hdr and mdesc_elem describe the layout of the data structure
36 * we get from the Hypervisor.
39 u32 version; /* Transport version */
40 u32 node_sz; /* node block size */
41 u32 name_sz; /* name block size */
42 u32 data_sz; /* data block size */
44 } __attribute__((aligned(16)));
48 #define MD_LIST_END 0x00
50 #define MD_NODE_END 0x45
52 #define MD_PROP_ARC 0x61
53 #define MD_PROP_VAL 0x76
54 #define MD_PROP_STR 0x73
55 #define MD_PROP_DATA 0x64
68 struct mdesc_mem_ops {
69 struct mdesc_handle *(*alloc)(unsigned int mdesc_size);
70 void (*free)(struct mdesc_handle *handle);
74 struct list_head list;
75 struct mdesc_mem_ops *mops;
78 unsigned int handle_size;
79 struct mdesc_hdr mdesc;
82 typedef int (*mdesc_node_info_get_f)(struct mdesc_handle *, u64,
83 union md_node_info *);
84 typedef void (*mdesc_node_info_rel_f)(union md_node_info *);
85 typedef bool (*mdesc_node_match_f)(union md_node_info *, union md_node_info *);
89 mdesc_node_info_get_f get_info;
90 mdesc_node_info_rel_f rel_info;
91 mdesc_node_match_f node_match;
94 static int get_vdev_port_node_info(struct mdesc_handle *md, u64 node,
95 union md_node_info *node_info);
96 static void rel_vdev_port_node_info(union md_node_info *node_info);
97 static bool vdev_port_node_match(union md_node_info *a_node_info,
98 union md_node_info *b_node_info);
100 static int get_ds_port_node_info(struct mdesc_handle *md, u64 node,
101 union md_node_info *node_info);
102 static void rel_ds_port_node_info(union md_node_info *node_info);
103 static bool ds_port_node_match(union md_node_info *a_node_info,
104 union md_node_info *b_node_info);
106 /* supported node types which can be registered */
107 static struct md_node_ops md_node_ops_table[] = {
108 {"virtual-device-port", get_vdev_port_node_info,
109 rel_vdev_port_node_info, vdev_port_node_match},
110 {"domain-services-port", get_ds_port_node_info,
111 rel_ds_port_node_info, ds_port_node_match},
112 {NULL, NULL, NULL, NULL}
115 static void mdesc_get_node_ops(const char *node_name,
116 mdesc_node_info_get_f *get_info_f,
117 mdesc_node_info_rel_f *rel_info_f,
118 mdesc_node_match_f *match_f)
134 for (i = 0; md_node_ops_table[i].name != NULL; i++) {
135 if (strcmp(md_node_ops_table[i].name, node_name) == 0) {
137 *get_info_f = md_node_ops_table[i].get_info;
140 *rel_info_f = md_node_ops_table[i].rel_info;
143 *match_f = md_node_ops_table[i].node_match;
150 static void mdesc_handle_init(struct mdesc_handle *hp,
151 unsigned int handle_size,
154 BUG_ON(((unsigned long)&hp->mdesc) & (16UL - 1));
156 memset(hp, 0, handle_size);
157 INIT_LIST_HEAD(&hp->list);
158 hp->self_base = base;
159 refcount_set(&hp->refcnt, 1);
160 hp->handle_size = handle_size;
163 static struct mdesc_handle * __init mdesc_memblock_alloc(unsigned int mdesc_size)
165 unsigned int handle_size, alloc_size;
166 struct mdesc_handle *hp;
169 handle_size = (sizeof(struct mdesc_handle) -
170 sizeof(struct mdesc_hdr) +
172 alloc_size = PAGE_ALIGN(handle_size);
174 paddr = memblock_alloc(alloc_size, PAGE_SIZE);
179 mdesc_handle_init(hp, handle_size, hp);
184 static void __init mdesc_memblock_free(struct mdesc_handle *hp)
186 unsigned int alloc_size;
189 BUG_ON(refcount_read(&hp->refcnt) != 0);
190 BUG_ON(!list_empty(&hp->list));
192 alloc_size = PAGE_ALIGN(hp->handle_size);
194 free_bootmem_late(start, alloc_size);
197 static struct mdesc_mem_ops memblock_mdesc_ops = {
198 .alloc = mdesc_memblock_alloc,
199 .free = mdesc_memblock_free,
202 static struct mdesc_handle *mdesc_kmalloc(unsigned int mdesc_size)
204 unsigned int handle_size;
205 struct mdesc_handle *hp;
209 handle_size = (sizeof(struct mdesc_handle) -
210 sizeof(struct mdesc_hdr) +
212 base = kmalloc(handle_size + 15, GFP_KERNEL | __GFP_RETRY_MAYFAIL);
216 addr = (unsigned long)base;
217 addr = (addr + 15UL) & ~15UL;
218 hp = (struct mdesc_handle *) addr;
220 mdesc_handle_init(hp, handle_size, base);
225 static void mdesc_kfree(struct mdesc_handle *hp)
227 BUG_ON(refcount_read(&hp->refcnt) != 0);
228 BUG_ON(!list_empty(&hp->list));
230 kfree(hp->self_base);
233 static struct mdesc_mem_ops kmalloc_mdesc_memops = {
234 .alloc = mdesc_kmalloc,
238 static struct mdesc_handle *mdesc_alloc(unsigned int mdesc_size,
239 struct mdesc_mem_ops *mops)
241 struct mdesc_handle *hp = mops->alloc(mdesc_size);
249 static void mdesc_free(struct mdesc_handle *hp)
254 static struct mdesc_handle *cur_mdesc;
255 static LIST_HEAD(mdesc_zombie_list);
256 static DEFINE_SPINLOCK(mdesc_lock);
258 struct mdesc_handle *mdesc_grab(void)
260 struct mdesc_handle *hp;
263 spin_lock_irqsave(&mdesc_lock, flags);
266 refcount_inc(&hp->refcnt);
267 spin_unlock_irqrestore(&mdesc_lock, flags);
271 EXPORT_SYMBOL(mdesc_grab);
273 void mdesc_release(struct mdesc_handle *hp)
277 spin_lock_irqsave(&mdesc_lock, flags);
278 if (refcount_dec_and_test(&hp->refcnt)) {
279 list_del_init(&hp->list);
282 spin_unlock_irqrestore(&mdesc_lock, flags);
284 EXPORT_SYMBOL(mdesc_release);
286 static DEFINE_MUTEX(mdesc_mutex);
287 static struct mdesc_notifier_client *client_list;
289 void mdesc_register_notifier(struct mdesc_notifier_client *client)
291 bool supported = false;
295 mutex_lock(&mdesc_mutex);
297 /* check to see if the node is supported for registration */
298 for (i = 0; md_node_ops_table[i].name != NULL; i++) {
299 if (strcmp(md_node_ops_table[i].name, client->node_name) == 0) {
306 pr_err("MD: %s node not supported\n", client->node_name);
307 mutex_unlock(&mdesc_mutex);
311 client->next = client_list;
312 client_list = client;
314 mdesc_for_each_node_by_name(cur_mdesc, node, client->node_name)
315 client->add(cur_mdesc, node, client->node_name);
317 mutex_unlock(&mdesc_mutex);
320 static const u64 *parent_cfg_handle(struct mdesc_handle *hp, u64 node)
326 mdesc_for_each_arc(a, hp, node, MDESC_ARC_TYPE_BACK) {
329 target = mdesc_arc_target(hp, a);
330 id = mdesc_get_property(hp, target,
339 static int get_vdev_port_node_info(struct mdesc_handle *md, u64 node,
340 union md_node_info *node_info)
342 const u64 *parent_cfg_hdlp;
347 * Virtual device nodes are distinguished by:
350 * 3. parent node "cfg-handle" property
352 idp = mdesc_get_property(md, node, "id", NULL);
353 name = mdesc_get_property(md, node, "name", NULL);
354 parent_cfg_hdlp = parent_cfg_handle(md, node);
356 if (!idp || !name || !parent_cfg_hdlp)
359 node_info->vdev_port.id = *idp;
360 node_info->vdev_port.name = kstrdup_const(name, GFP_KERNEL);
361 if (!node_info->vdev_port.name)
363 node_info->vdev_port.parent_cfg_hdl = *parent_cfg_hdlp;
368 static void rel_vdev_port_node_info(union md_node_info *node_info)
370 if (node_info && node_info->vdev_port.name) {
371 kfree_const(node_info->vdev_port.name);
372 node_info->vdev_port.name = NULL;
376 static bool vdev_port_node_match(union md_node_info *a_node_info,
377 union md_node_info *b_node_info)
379 if (a_node_info->vdev_port.id != b_node_info->vdev_port.id)
382 if (a_node_info->vdev_port.parent_cfg_hdl !=
383 b_node_info->vdev_port.parent_cfg_hdl)
386 if (strncmp(a_node_info->vdev_port.name,
387 b_node_info->vdev_port.name, MDESC_MAX_STR_LEN) != 0)
393 static int get_ds_port_node_info(struct mdesc_handle *md, u64 node,
394 union md_node_info *node_info)
398 /* DS port nodes use the "id" property to distinguish them */
399 idp = mdesc_get_property(md, node, "id", NULL);
403 node_info->ds_port.id = *idp;
408 static void rel_ds_port_node_info(union md_node_info *node_info)
412 static bool ds_port_node_match(union md_node_info *a_node_info,
413 union md_node_info *b_node_info)
415 if (a_node_info->ds_port.id != b_node_info->ds_port.id)
421 /* Run 'func' on nodes which are in A but not in B. */
422 static void invoke_on_missing(const char *name,
423 struct mdesc_handle *a,
424 struct mdesc_handle *b,
425 void (*func)(struct mdesc_handle *, u64,
426 const char *node_name))
428 mdesc_node_info_get_f get_info_func;
429 mdesc_node_info_rel_f rel_info_func;
430 mdesc_node_match_f node_match_func;
431 union md_node_info a_node_info;
432 union md_node_info b_node_info;
439 * Find the get_info, rel_info and node_match ops for the given
442 mdesc_get_node_ops(name, &get_info_func, &rel_info_func,
445 /* If we didn't find a match, the node type is not supported */
446 if (!get_info_func || !rel_info_func || !node_match_func) {
447 pr_err("MD: %s node type is not supported\n", name);
451 mdesc_for_each_node_by_name(a, a_node, name) {
454 rv = get_info_func(a, a_node, &a_node_info);
456 pr_err("MD: Cannot find 1 or more required match properties for %s node.\n",
461 /* Check each node in B for node matching a_node */
462 mdesc_for_each_node_by_name(b, b_node, name) {
463 rv = get_info_func(b, b_node, &b_node_info);
467 if (node_match_func(&a_node_info, &b_node_info)) {
469 rel_info_func(&b_node_info);
473 rel_info_func(&b_node_info);
476 rel_info_func(&a_node_info);
479 func(a, a_node, name);
483 static void notify_one(struct mdesc_notifier_client *p,
484 struct mdesc_handle *old_hp,
485 struct mdesc_handle *new_hp)
487 invoke_on_missing(p->node_name, old_hp, new_hp, p->remove);
488 invoke_on_missing(p->node_name, new_hp, old_hp, p->add);
491 static void mdesc_notify_clients(struct mdesc_handle *old_hp,
492 struct mdesc_handle *new_hp)
494 struct mdesc_notifier_client *p = client_list;
497 notify_one(p, old_hp, new_hp);
502 void mdesc_update(void)
504 unsigned long len, real_len, status;
505 struct mdesc_handle *hp, *orig_hp;
508 mutex_lock(&mdesc_mutex);
510 (void) sun4v_mach_desc(0UL, 0UL, &len);
512 hp = mdesc_alloc(len, &kmalloc_mdesc_memops);
514 printk(KERN_ERR "MD: mdesc alloc fails\n");
518 status = sun4v_mach_desc(__pa(&hp->mdesc), len, &real_len);
519 if (status != HV_EOK || real_len > len) {
520 printk(KERN_ERR "MD: mdesc reread fails with %lu\n",
522 refcount_dec(&hp->refcnt);
527 spin_lock_irqsave(&mdesc_lock, flags);
530 spin_unlock_irqrestore(&mdesc_lock, flags);
532 mdesc_notify_clients(orig_hp, hp);
534 spin_lock_irqsave(&mdesc_lock, flags);
535 if (refcount_dec_and_test(&orig_hp->refcnt))
538 list_add(&orig_hp->list, &mdesc_zombie_list);
539 spin_unlock_irqrestore(&mdesc_lock, flags);
542 mutex_unlock(&mdesc_mutex);
545 u64 mdesc_get_node(struct mdesc_handle *hp, const char *node_name,
546 union md_node_info *node_info)
548 mdesc_node_info_get_f get_info_func;
549 mdesc_node_info_rel_f rel_info_func;
550 mdesc_node_match_f node_match_func;
551 union md_node_info hp_node_info;
555 if (hp == NULL || node_name == NULL || node_info == NULL)
556 return MDESC_NODE_NULL;
558 /* Find the ops for the given node name */
559 mdesc_get_node_ops(node_name, &get_info_func, &rel_info_func,
562 /* If we didn't find ops for the given node name, it is not supported */
563 if (!get_info_func || !rel_info_func || !node_match_func) {
564 pr_err("MD: %s node is not supported\n", node_name);
568 mdesc_for_each_node_by_name(hp, hp_node, node_name) {
569 rv = get_info_func(hp, hp_node, &hp_node_info);
573 if (node_match_func(node_info, &hp_node_info))
576 rel_info_func(&hp_node_info);
579 rel_info_func(&hp_node_info);
583 EXPORT_SYMBOL(mdesc_get_node);
585 int mdesc_get_node_info(struct mdesc_handle *hp, u64 node,
586 const char *node_name, union md_node_info *node_info)
588 mdesc_node_info_get_f get_info_func;
591 if (hp == NULL || node == MDESC_NODE_NULL ||
592 node_name == NULL || node_info == NULL)
595 /* Find the get_info op for the given node name */
596 mdesc_get_node_ops(node_name, &get_info_func, NULL, NULL);
598 /* If we didn't find a get_info_func, the node name is not supported */
599 if (get_info_func == NULL) {
600 pr_err("MD: %s node is not supported\n", node_name);
604 rv = get_info_func(hp, node, node_info);
606 pr_err("MD: Cannot find 1 or more required match properties for %s node.\n",
613 EXPORT_SYMBOL(mdesc_get_node_info);
615 static struct mdesc_elem *node_block(struct mdesc_hdr *mdesc)
617 return (struct mdesc_elem *) mdesc->data;
620 static void *name_block(struct mdesc_hdr *mdesc)
622 return ((void *) node_block(mdesc)) + mdesc->node_sz;
625 static void *data_block(struct mdesc_hdr *mdesc)
627 return ((void *) name_block(mdesc)) + mdesc->name_sz;
630 u64 mdesc_node_by_name(struct mdesc_handle *hp,
631 u64 from_node, const char *name)
633 struct mdesc_elem *ep = node_block(&hp->mdesc);
634 const char *names = name_block(&hp->mdesc);
635 u64 last_node = hp->mdesc.node_sz / 16;
638 if (from_node == MDESC_NODE_NULL) {
640 } else if (from_node >= last_node) {
641 return MDESC_NODE_NULL;
643 ret = ep[from_node].d.val;
646 while (ret < last_node) {
647 if (ep[ret].tag != MD_NODE)
648 return MDESC_NODE_NULL;
649 if (!strcmp(names + ep[ret].name_offset, name))
653 if (ret >= last_node)
654 ret = MDESC_NODE_NULL;
657 EXPORT_SYMBOL(mdesc_node_by_name);
659 const void *mdesc_get_property(struct mdesc_handle *hp, u64 node,
660 const char *name, int *lenp)
662 const char *names = name_block(&hp->mdesc);
663 u64 last_node = hp->mdesc.node_sz / 16;
664 void *data = data_block(&hp->mdesc);
665 struct mdesc_elem *ep;
667 if (node == MDESC_NODE_NULL || node >= last_node)
670 ep = node_block(&hp->mdesc) + node;
672 for (; ep->tag != MD_NODE_END; ep++) {
684 val = data + ep->d.data.data_offset;
685 len = ep->d.data.data_len;
694 if (!strcmp(names + ep->name_offset, name)) {
703 EXPORT_SYMBOL(mdesc_get_property);
705 u64 mdesc_next_arc(struct mdesc_handle *hp, u64 from, const char *arc_type)
707 struct mdesc_elem *ep, *base = node_block(&hp->mdesc);
708 const char *names = name_block(&hp->mdesc);
709 u64 last_node = hp->mdesc.node_sz / 16;
711 if (from == MDESC_NODE_NULL || from >= last_node)
712 return MDESC_NODE_NULL;
717 for (; ep->tag != MD_NODE_END; ep++) {
718 if (ep->tag != MD_PROP_ARC)
721 if (strcmp(names + ep->name_offset, arc_type))
727 return MDESC_NODE_NULL;
729 EXPORT_SYMBOL(mdesc_next_arc);
731 u64 mdesc_arc_target(struct mdesc_handle *hp, u64 arc)
733 struct mdesc_elem *ep, *base = node_block(&hp->mdesc);
739 EXPORT_SYMBOL(mdesc_arc_target);
741 const char *mdesc_node_name(struct mdesc_handle *hp, u64 node)
743 struct mdesc_elem *ep, *base = node_block(&hp->mdesc);
744 const char *names = name_block(&hp->mdesc);
745 u64 last_node = hp->mdesc.node_sz / 16;
747 if (node == MDESC_NODE_NULL || node >= last_node)
751 if (ep->tag != MD_NODE)
754 return names + ep->name_offset;
756 EXPORT_SYMBOL(mdesc_node_name);
758 static u64 max_cpus = 64;
760 static void __init report_platform_properties(void)
762 struct mdesc_handle *hp = mdesc_grab();
763 u64 pn = mdesc_node_by_name(hp, MDESC_NODE_NULL, "platform");
767 if (pn == MDESC_NODE_NULL) {
768 prom_printf("No platform node in machine-description.\n");
772 s = mdesc_get_property(hp, pn, "banner-name", NULL);
773 printk("PLATFORM: banner-name [%s]\n", s);
774 s = mdesc_get_property(hp, pn, "name", NULL);
775 printk("PLATFORM: name [%s]\n", s);
777 v = mdesc_get_property(hp, pn, "hostid", NULL);
779 printk("PLATFORM: hostid [%08llx]\n", *v);
780 v = mdesc_get_property(hp, pn, "serial#", NULL);
782 printk("PLATFORM: serial# [%08llx]\n", *v);
783 v = mdesc_get_property(hp, pn, "stick-frequency", NULL);
784 printk("PLATFORM: stick-frequency [%08llx]\n", *v);
785 v = mdesc_get_property(hp, pn, "mac-address", NULL);
787 printk("PLATFORM: mac-address [%llx]\n", *v);
788 v = mdesc_get_property(hp, pn, "watchdog-resolution", NULL);
790 printk("PLATFORM: watchdog-resolution [%llu ms]\n", *v);
791 v = mdesc_get_property(hp, pn, "watchdog-max-timeout", NULL);
793 printk("PLATFORM: watchdog-max-timeout [%llu ms]\n", *v);
794 v = mdesc_get_property(hp, pn, "max-cpus", NULL);
797 printk("PLATFORM: max-cpus [%llu]\n", max_cpus);
806 if (max_cpu > NR_CPUS)
811 for (i = 0; i < max_cpu; i++)
812 set_cpu_possible(i, true);
819 static void fill_in_one_cache(cpuinfo_sparc *c, struct mdesc_handle *hp, u64 mp)
821 const u64 *level = mdesc_get_property(hp, mp, "level", NULL);
822 const u64 *size = mdesc_get_property(hp, mp, "size", NULL);
823 const u64 *line_size = mdesc_get_property(hp, mp, "line-size", NULL);
827 type = mdesc_get_property(hp, mp, "type", &type_len);
831 if (of_find_in_proplist(type, "instn", type_len)) {
832 c->icache_size = *size;
833 c->icache_line_size = *line_size;
834 } else if (of_find_in_proplist(type, "data", type_len)) {
835 c->dcache_size = *size;
836 c->dcache_line_size = *line_size;
841 c->ecache_size = *size;
842 c->ecache_line_size = *line_size;
852 mdesc_for_each_arc(a, hp, mp, MDESC_ARC_TYPE_FWD) {
853 u64 target = mdesc_arc_target(hp, a);
854 const char *name = mdesc_node_name(hp, target);
856 if (!strcmp(name, "cache"))
857 fill_in_one_cache(c, hp, target);
862 static void find_back_node_value(struct mdesc_handle *hp, u64 node,
864 void (*func)(struct mdesc_handle *, u64, int),
869 /* Since we have an estimate of recursion depth, do a sanity check. */
873 mdesc_for_each_arc(arc, hp, node, MDESC_ARC_TYPE_BACK) {
874 u64 n = mdesc_arc_target(hp, arc);
875 const char *name = mdesc_node_name(hp, n);
877 if (!strcmp(srch_val, name))
880 find_back_node_value(hp, n, srch_val, func, val, depth-1);
884 static void __mark_core_id(struct mdesc_handle *hp, u64 node,
887 const u64 *id = mdesc_get_property(hp, node, "id", NULL);
889 if (*id < num_possible_cpus())
890 cpu_data(*id).core_id = core_id;
893 static void __mark_max_cache_id(struct mdesc_handle *hp, u64 node,
896 const u64 *id = mdesc_get_property(hp, node, "id", NULL);
898 if (*id < num_possible_cpus()) {
899 cpu_data(*id).max_cache_id = max_cache_id;
902 * On systems without explicit socket descriptions socket
905 cpu_data(*id).sock_id = max_cache_id;
909 static void mark_core_ids(struct mdesc_handle *hp, u64 mp,
912 find_back_node_value(hp, mp, "cpu", __mark_core_id, core_id, 10);
915 static void mark_max_cache_ids(struct mdesc_handle *hp, u64 mp,
918 find_back_node_value(hp, mp, "cpu", __mark_max_cache_id,
922 static void set_core_ids(struct mdesc_handle *hp)
929 /* Identify unique cores by looking for cpus backpointed to by
930 * level 1 instruction caches.
932 mdesc_for_each_node_by_name(hp, mp, "cache") {
937 level = mdesc_get_property(hp, mp, "level", NULL);
941 type = mdesc_get_property(hp, mp, "type", &len);
942 if (!of_find_in_proplist(type, "instn", len))
945 mark_core_ids(hp, mp, idx);
950 static int set_max_cache_ids_by_cache(struct mdesc_handle *hp, int level)
957 * Identify unique highest level of shared cache by looking for cpus
958 * backpointed to by shared level N caches.
960 mdesc_for_each_node_by_name(hp, mp, "cache") {
963 cur_lvl = mdesc_get_property(hp, mp, "level", NULL);
964 if (*cur_lvl != level)
966 mark_max_cache_ids(hp, mp, idx);
973 static void set_sock_ids_by_socket(struct mdesc_handle *hp, u64 mp)
977 mdesc_for_each_node_by_name(hp, mp, "socket") {
980 mdesc_for_each_arc(a, hp, mp, MDESC_ARC_TYPE_FWD) {
981 u64 t = mdesc_arc_target(hp, a);
985 name = mdesc_node_name(hp, t);
986 if (strcmp(name, "cpu"))
989 id = mdesc_get_property(hp, t, "id", NULL);
990 if (*id < num_possible_cpus())
991 cpu_data(*id).sock_id = idx;
997 static void set_sock_ids(struct mdesc_handle *hp)
1002 * Find the highest level of shared cache which pre-T7 is also
1005 if (!set_max_cache_ids_by_cache(hp, 3))
1006 set_max_cache_ids_by_cache(hp, 2);
1008 /* If machine description exposes sockets data use it.*/
1009 mp = mdesc_node_by_name(hp, MDESC_NODE_NULL, "sockets");
1010 if (mp != MDESC_NODE_NULL)
1011 set_sock_ids_by_socket(hp, mp);
1014 static void mark_proc_ids(struct mdesc_handle *hp, u64 mp, int proc_id)
1018 mdesc_for_each_arc(a, hp, mp, MDESC_ARC_TYPE_BACK) {
1019 u64 t = mdesc_arc_target(hp, a);
1023 name = mdesc_node_name(hp, t);
1024 if (strcmp(name, "cpu"))
1027 id = mdesc_get_property(hp, t, "id", NULL);
1029 cpu_data(*id).proc_id = proc_id;
1033 static void __set_proc_ids(struct mdesc_handle *hp, const char *exec_unit_name)
1039 mdesc_for_each_node_by_name(hp, mp, exec_unit_name) {
1043 type = mdesc_get_property(hp, mp, "type", &len);
1044 if (!of_find_in_proplist(type, "int", len) &&
1045 !of_find_in_proplist(type, "integer", len))
1048 mark_proc_ids(hp, mp, idx);
1053 static void set_proc_ids(struct mdesc_handle *hp)
1055 __set_proc_ids(hp, "exec_unit");
1056 __set_proc_ids(hp, "exec-unit");
1059 static void get_one_mondo_bits(const u64 *p, unsigned int *mask,
1060 unsigned long def, unsigned long max)
1068 if (!val || val >= 64)
1074 *mask = ((1U << val) * 64U) - 1U;
1078 *mask = ((1U << def) * 64U) - 1U;
1081 static void get_mondo_data(struct mdesc_handle *hp, u64 mp,
1082 struct trap_per_cpu *tb)
1087 val = mdesc_get_property(hp, mp, "q-cpu-mondo-#bits", NULL);
1088 get_one_mondo_bits(val, &tb->cpu_mondo_qmask, 7, ilog2(max_cpus * 2));
1090 val = mdesc_get_property(hp, mp, "q-dev-mondo-#bits", NULL);
1091 get_one_mondo_bits(val, &tb->dev_mondo_qmask, 7, 8);
1093 val = mdesc_get_property(hp, mp, "q-resumable-#bits", NULL);
1094 get_one_mondo_bits(val, &tb->resum_qmask, 6, 7);
1096 val = mdesc_get_property(hp, mp, "q-nonresumable-#bits", NULL);
1097 get_one_mondo_bits(val, &tb->nonresum_qmask, 2, 2);
1099 pr_info("SUN4V: Mondo queue sizes "
1100 "[cpu(%u) dev(%u) r(%u) nr(%u)]\n",
1101 tb->cpu_mondo_qmask + 1,
1102 tb->dev_mondo_qmask + 1,
1103 tb->resum_qmask + 1,
1104 tb->nonresum_qmask + 1);
1108 static void *mdesc_iterate_over_cpus(void *(*func)(struct mdesc_handle *, u64, int, void *), void *arg, cpumask_t *mask)
1110 struct mdesc_handle *hp = mdesc_grab();
1114 mdesc_for_each_node_by_name(hp, mp, "cpu") {
1115 const u64 *id = mdesc_get_property(hp, mp, "id", NULL);
1119 if (cpuid >= NR_CPUS) {
1120 printk(KERN_WARNING "Ignoring CPU %d which is "
1121 ">= NR_CPUS (%d)\n",
1125 if (!cpumask_test_cpu(cpuid, mask))
1129 ret = func(hp, mp, cpuid, arg);
1138 static void *record_one_cpu(struct mdesc_handle *hp, u64 mp, int cpuid,
1143 set_cpu_present(cpuid, true);
1148 void mdesc_populate_present_mask(cpumask_t *mask)
1150 if (tlb_type != hypervisor)
1154 mdesc_iterate_over_cpus(record_one_cpu, NULL, mask);
1157 static void * __init check_one_pgsz(struct mdesc_handle *hp, u64 mp, int cpuid, void *arg)
1159 const u64 *pgsz_prop = mdesc_get_property(hp, mp, "mmu-page-size-list", NULL);
1160 unsigned long *pgsz_mask = arg;
1163 val = (HV_PGSZ_MASK_8K | HV_PGSZ_MASK_64K |
1164 HV_PGSZ_MASK_512K | HV_PGSZ_MASK_4MB);
1175 void __init mdesc_get_page_sizes(cpumask_t *mask, unsigned long *pgsz_mask)
1178 mdesc_iterate_over_cpus(check_one_pgsz, pgsz_mask, mask);
1181 static void *fill_in_one_cpu(struct mdesc_handle *hp, u64 mp, int cpuid,
1184 const u64 *cfreq = mdesc_get_property(hp, mp, "clock-frequency", NULL);
1185 struct trap_per_cpu *tb;
1190 /* On uniprocessor we only want the values for the
1191 * real physical cpu the kernel booted onto, however
1192 * cpu_data() only has one entry at index 0.
1194 if (cpuid != real_hard_smp_processor_id())
1199 c = &cpu_data(cpuid);
1200 c->clock_tick = *cfreq;
1202 tb = &trap_block[cpuid];
1203 get_mondo_data(hp, mp, tb);
1205 mdesc_for_each_arc(a, hp, mp, MDESC_ARC_TYPE_FWD) {
1206 u64 j, t = mdesc_arc_target(hp, a);
1209 t_name = mdesc_node_name(hp, t);
1210 if (!strcmp(t_name, "cache")) {
1211 fill_in_one_cache(c, hp, t);
1215 mdesc_for_each_arc(j, hp, t, MDESC_ARC_TYPE_FWD) {
1216 u64 n = mdesc_arc_target(hp, j);
1219 n_name = mdesc_node_name(hp, n);
1220 if (!strcmp(n_name, "cache"))
1221 fill_in_one_cache(c, hp, n);
1231 void mdesc_fill_in_cpu_data(cpumask_t *mask)
1233 struct mdesc_handle *hp;
1235 mdesc_iterate_over_cpus(fill_in_one_cpu, NULL, mask);
1245 smp_fill_in_sib_core_maps();
1248 /* mdesc_open() - Grab a reference to mdesc_handle when /dev/mdesc is
1249 * opened. Hold this reference until /dev/mdesc is closed to ensure
1250 * mdesc data structure is not released underneath us. Store the
1251 * pointer to mdesc structure in private_data for read and seek to use
1253 static int mdesc_open(struct inode *inode, struct file *file)
1255 struct mdesc_handle *hp = mdesc_grab();
1260 file->private_data = hp;
1265 static ssize_t mdesc_read(struct file *file, char __user *buf,
1266 size_t len, loff_t *offp)
1268 struct mdesc_handle *hp = file->private_data;
1269 unsigned char *mdesc;
1270 int bytes_left, count = len;
1272 if (*offp >= hp->handle_size)
1275 bytes_left = hp->handle_size - *offp;
1276 if (count > bytes_left)
1279 mdesc = (unsigned char *)&hp->mdesc;
1281 if (!copy_to_user(buf, mdesc, count)) {
1289 static loff_t mdesc_llseek(struct file *file, loff_t offset, int whence)
1291 struct mdesc_handle *hp = file->private_data;
1293 return no_seek_end_llseek_size(file, offset, whence, hp->handle_size);
1296 /* mdesc_close() - /dev/mdesc is being closed, release the reference to
1299 static int mdesc_close(struct inode *inode, struct file *file)
1301 mdesc_release(file->private_data);
1305 static const struct file_operations mdesc_fops = {
1308 .llseek = mdesc_llseek,
1309 .release = mdesc_close,
1310 .owner = THIS_MODULE,
1313 static struct miscdevice mdesc_misc = {
1314 .minor = MISC_DYNAMIC_MINOR,
1316 .fops = &mdesc_fops,
1319 static int __init mdesc_misc_init(void)
1321 return misc_register(&mdesc_misc);
1324 __initcall(mdesc_misc_init);
1326 void __init sun4v_mdesc_init(void)
1328 struct mdesc_handle *hp;
1329 unsigned long len, real_len, status;
1331 (void) sun4v_mach_desc(0UL, 0UL, &len);
1333 printk("MDESC: Size is %lu bytes.\n", len);
1335 hp = mdesc_alloc(len, &memblock_mdesc_ops);
1337 prom_printf("MDESC: alloc of %lu bytes failed.\n", len);
1341 status = sun4v_mach_desc(__pa(&hp->mdesc), len, &real_len);
1342 if (status != HV_EOK || real_len > len) {
1343 prom_printf("sun4v_mach_desc fails, err(%lu), "
1344 "len(%lu), real_len(%lu)\n",
1345 status, len, real_len);
1353 report_platform_properties();