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/log2.h>
9 #include <linux/list.h>
10 #include <linux/slab.h>
12 #include <linux/miscdevice.h>
13 #include <linux/memblock.h>
14 #include <linux/export.h>
15 #include <linux/refcount.h>
17 #include <asm/cpudata.h>
18 #include <asm/hypervisor.h>
19 #include <asm/mdesc.h>
21 #include <linux/uaccess.h>
22 #include <asm/oplib.h>
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.
31 * Actually, it isn't even a DAG, because there are back pointers
32 * which create cycles in the graph.
34 * mdesc_hdr and mdesc_elem describe the layout of the data structure
35 * we get from the Hypervisor.
38 u32 version; /* Transport version */
39 u32 node_sz; /* node block size */
40 u32 name_sz; /* name block size */
41 u32 data_sz; /* data block size */
42 } __attribute__((aligned(16)));
46 #define MD_LIST_END 0x00
48 #define MD_NODE_END 0x45
50 #define MD_PROP_ARC 0x61
51 #define MD_PROP_VAL 0x76
52 #define MD_PROP_STR 0x73
53 #define MD_PROP_DATA 0x64
66 struct mdesc_mem_ops {
67 struct mdesc_handle *(*alloc)(unsigned int mdesc_size);
68 void (*free)(struct mdesc_handle *handle);
72 struct list_head list;
73 struct mdesc_mem_ops *mops;
76 unsigned int handle_size;
77 struct mdesc_hdr mdesc;
80 typedef int (*mdesc_node_info_get_f)(struct mdesc_handle *, u64,
81 union md_node_info *);
82 typedef void (*mdesc_node_info_rel_f)(union md_node_info *);
83 typedef bool (*mdesc_node_match_f)(union md_node_info *, union md_node_info *);
87 mdesc_node_info_get_f get_info;
88 mdesc_node_info_rel_f rel_info;
89 mdesc_node_match_f node_match;
92 static int get_vdev_port_node_info(struct mdesc_handle *md, u64 node,
93 union md_node_info *node_info);
94 static void rel_vdev_port_node_info(union md_node_info *node_info);
95 static bool vdev_port_node_match(union md_node_info *a_node_info,
96 union md_node_info *b_node_info);
98 static int get_ds_port_node_info(struct mdesc_handle *md, u64 node,
99 union md_node_info *node_info);
100 static void rel_ds_port_node_info(union md_node_info *node_info);
101 static bool ds_port_node_match(union md_node_info *a_node_info,
102 union md_node_info *b_node_info);
104 /* supported node types which can be registered */
105 static struct md_node_ops md_node_ops_table[] = {
106 {"virtual-device-port", get_vdev_port_node_info,
107 rel_vdev_port_node_info, vdev_port_node_match},
108 {"domain-services-port", get_ds_port_node_info,
109 rel_ds_port_node_info, ds_port_node_match},
110 {NULL, NULL, NULL, NULL}
113 static void mdesc_get_node_ops(const char *node_name,
114 mdesc_node_info_get_f *get_info_f,
115 mdesc_node_info_rel_f *rel_info_f,
116 mdesc_node_match_f *match_f)
132 for (i = 0; md_node_ops_table[i].name != NULL; i++) {
133 if (strcmp(md_node_ops_table[i].name, node_name) == 0) {
135 *get_info_f = md_node_ops_table[i].get_info;
138 *rel_info_f = md_node_ops_table[i].rel_info;
141 *match_f = md_node_ops_table[i].node_match;
148 static void mdesc_handle_init(struct mdesc_handle *hp,
149 unsigned int handle_size,
152 BUG_ON(((unsigned long)&hp->mdesc) & (16UL - 1));
154 memset(hp, 0, handle_size);
155 INIT_LIST_HEAD(&hp->list);
156 hp->self_base = base;
157 refcount_set(&hp->refcnt, 1);
158 hp->handle_size = handle_size;
161 static struct mdesc_handle * __init mdesc_memblock_alloc(unsigned int mdesc_size)
163 unsigned int handle_size, alloc_size;
164 struct mdesc_handle *hp;
167 handle_size = (sizeof(struct mdesc_handle) -
168 sizeof(struct mdesc_hdr) +
170 alloc_size = PAGE_ALIGN(handle_size);
172 paddr = memblock_phys_alloc(alloc_size, PAGE_SIZE);
177 mdesc_handle_init(hp, handle_size, hp);
182 static void __init mdesc_memblock_free(struct mdesc_handle *hp)
184 unsigned int alloc_size;
187 BUG_ON(refcount_read(&hp->refcnt) != 0);
188 BUG_ON(!list_empty(&hp->list));
190 alloc_size = PAGE_ALIGN(hp->handle_size);
192 memblock_free_late(start, alloc_size);
195 static struct mdesc_mem_ops memblock_mdesc_ops = {
196 .alloc = mdesc_memblock_alloc,
197 .free = mdesc_memblock_free,
200 static struct mdesc_handle *mdesc_kmalloc(unsigned int mdesc_size)
202 unsigned int handle_size;
203 struct mdesc_handle *hp;
207 handle_size = (sizeof(struct mdesc_handle) -
208 sizeof(struct mdesc_hdr) +
210 base = kmalloc(handle_size + 15, GFP_KERNEL | __GFP_RETRY_MAYFAIL);
214 addr = (unsigned long)base;
215 addr = (addr + 15UL) & ~15UL;
216 hp = (struct mdesc_handle *) addr;
218 mdesc_handle_init(hp, handle_size, base);
223 static void mdesc_kfree(struct mdesc_handle *hp)
225 BUG_ON(refcount_read(&hp->refcnt) != 0);
226 BUG_ON(!list_empty(&hp->list));
228 kfree(hp->self_base);
231 static struct mdesc_mem_ops kmalloc_mdesc_memops = {
232 .alloc = mdesc_kmalloc,
236 static struct mdesc_handle *mdesc_alloc(unsigned int mdesc_size,
237 struct mdesc_mem_ops *mops)
239 struct mdesc_handle *hp = mops->alloc(mdesc_size);
247 static void mdesc_free(struct mdesc_handle *hp)
252 static struct mdesc_handle *cur_mdesc;
253 static LIST_HEAD(mdesc_zombie_list);
254 static DEFINE_SPINLOCK(mdesc_lock);
256 struct mdesc_handle *mdesc_grab(void)
258 struct mdesc_handle *hp;
261 spin_lock_irqsave(&mdesc_lock, flags);
264 refcount_inc(&hp->refcnt);
265 spin_unlock_irqrestore(&mdesc_lock, flags);
269 EXPORT_SYMBOL(mdesc_grab);
271 void mdesc_release(struct mdesc_handle *hp)
275 spin_lock_irqsave(&mdesc_lock, flags);
276 if (refcount_dec_and_test(&hp->refcnt)) {
277 list_del_init(&hp->list);
280 spin_unlock_irqrestore(&mdesc_lock, flags);
282 EXPORT_SYMBOL(mdesc_release);
284 static DEFINE_MUTEX(mdesc_mutex);
285 static struct mdesc_notifier_client *client_list;
287 void mdesc_register_notifier(struct mdesc_notifier_client *client)
289 bool supported = false;
293 mutex_lock(&mdesc_mutex);
295 /* check to see if the node is supported for registration */
296 for (i = 0; md_node_ops_table[i].name != NULL; i++) {
297 if (strcmp(md_node_ops_table[i].name, client->node_name) == 0) {
304 pr_err("MD: %s node not supported\n", client->node_name);
305 mutex_unlock(&mdesc_mutex);
309 client->next = client_list;
310 client_list = client;
312 mdesc_for_each_node_by_name(cur_mdesc, node, client->node_name)
313 client->add(cur_mdesc, node, client->node_name);
315 mutex_unlock(&mdesc_mutex);
318 static const u64 *parent_cfg_handle(struct mdesc_handle *hp, u64 node)
324 mdesc_for_each_arc(a, hp, node, MDESC_ARC_TYPE_BACK) {
327 target = mdesc_arc_target(hp, a);
328 id = mdesc_get_property(hp, target,
337 static int get_vdev_port_node_info(struct mdesc_handle *md, u64 node,
338 union md_node_info *node_info)
340 const u64 *parent_cfg_hdlp;
345 * Virtual device nodes are distinguished by:
348 * 3. parent node "cfg-handle" property
350 idp = mdesc_get_property(md, node, "id", NULL);
351 name = mdesc_get_property(md, node, "name", NULL);
352 parent_cfg_hdlp = parent_cfg_handle(md, node);
354 if (!idp || !name || !parent_cfg_hdlp)
357 node_info->vdev_port.id = *idp;
358 node_info->vdev_port.name = kstrdup_const(name, GFP_KERNEL);
359 if (!node_info->vdev_port.name)
361 node_info->vdev_port.parent_cfg_hdl = *parent_cfg_hdlp;
366 static void rel_vdev_port_node_info(union md_node_info *node_info)
368 if (node_info && node_info->vdev_port.name) {
369 kfree_const(node_info->vdev_port.name);
370 node_info->vdev_port.name = NULL;
374 static bool vdev_port_node_match(union md_node_info *a_node_info,
375 union md_node_info *b_node_info)
377 if (a_node_info->vdev_port.id != b_node_info->vdev_port.id)
380 if (a_node_info->vdev_port.parent_cfg_hdl !=
381 b_node_info->vdev_port.parent_cfg_hdl)
384 if (strncmp(a_node_info->vdev_port.name,
385 b_node_info->vdev_port.name, MDESC_MAX_STR_LEN) != 0)
391 static int get_ds_port_node_info(struct mdesc_handle *md, u64 node,
392 union md_node_info *node_info)
396 /* DS port nodes use the "id" property to distinguish them */
397 idp = mdesc_get_property(md, node, "id", NULL);
401 node_info->ds_port.id = *idp;
406 static void rel_ds_port_node_info(union md_node_info *node_info)
410 static bool ds_port_node_match(union md_node_info *a_node_info,
411 union md_node_info *b_node_info)
413 if (a_node_info->ds_port.id != b_node_info->ds_port.id)
419 /* Run 'func' on nodes which are in A but not in B. */
420 static void invoke_on_missing(const char *name,
421 struct mdesc_handle *a,
422 struct mdesc_handle *b,
423 void (*func)(struct mdesc_handle *, u64,
424 const char *node_name))
426 mdesc_node_info_get_f get_info_func;
427 mdesc_node_info_rel_f rel_info_func;
428 mdesc_node_match_f node_match_func;
429 union md_node_info a_node_info;
430 union md_node_info b_node_info;
437 * Find the get_info, rel_info and node_match ops for the given
440 mdesc_get_node_ops(name, &get_info_func, &rel_info_func,
443 /* If we didn't find a match, the node type is not supported */
444 if (!get_info_func || !rel_info_func || !node_match_func) {
445 pr_err("MD: %s node type is not supported\n", name);
449 mdesc_for_each_node_by_name(a, a_node, name) {
452 rv = get_info_func(a, a_node, &a_node_info);
454 pr_err("MD: Cannot find 1 or more required match properties for %s node.\n",
459 /* Check each node in B for node matching a_node */
460 mdesc_for_each_node_by_name(b, b_node, name) {
461 rv = get_info_func(b, b_node, &b_node_info);
465 if (node_match_func(&a_node_info, &b_node_info)) {
467 rel_info_func(&b_node_info);
471 rel_info_func(&b_node_info);
474 rel_info_func(&a_node_info);
477 func(a, a_node, name);
481 static void notify_one(struct mdesc_notifier_client *p,
482 struct mdesc_handle *old_hp,
483 struct mdesc_handle *new_hp)
485 invoke_on_missing(p->node_name, old_hp, new_hp, p->remove);
486 invoke_on_missing(p->node_name, new_hp, old_hp, p->add);
489 static void mdesc_notify_clients(struct mdesc_handle *old_hp,
490 struct mdesc_handle *new_hp)
492 struct mdesc_notifier_client *p = client_list;
495 notify_one(p, old_hp, new_hp);
500 void mdesc_update(void)
502 unsigned long len, real_len, status;
503 struct mdesc_handle *hp, *orig_hp;
506 mutex_lock(&mdesc_mutex);
508 (void) sun4v_mach_desc(0UL, 0UL, &len);
510 hp = mdesc_alloc(len, &kmalloc_mdesc_memops);
512 printk(KERN_ERR "MD: mdesc alloc fails\n");
516 status = sun4v_mach_desc(__pa(&hp->mdesc), len, &real_len);
517 if (status != HV_EOK || real_len > len) {
518 printk(KERN_ERR "MD: mdesc reread fails with %lu\n",
520 refcount_dec(&hp->refcnt);
525 spin_lock_irqsave(&mdesc_lock, flags);
528 spin_unlock_irqrestore(&mdesc_lock, flags);
530 mdesc_notify_clients(orig_hp, hp);
532 spin_lock_irqsave(&mdesc_lock, flags);
533 if (refcount_dec_and_test(&orig_hp->refcnt))
536 list_add(&orig_hp->list, &mdesc_zombie_list);
537 spin_unlock_irqrestore(&mdesc_lock, flags);
540 mutex_unlock(&mdesc_mutex);
543 u64 mdesc_get_node(struct mdesc_handle *hp, const char *node_name,
544 union md_node_info *node_info)
546 mdesc_node_info_get_f get_info_func;
547 mdesc_node_info_rel_f rel_info_func;
548 mdesc_node_match_f node_match_func;
549 union md_node_info hp_node_info;
553 if (hp == NULL || node_name == NULL || node_info == NULL)
554 return MDESC_NODE_NULL;
556 /* Find the ops for the given node name */
557 mdesc_get_node_ops(node_name, &get_info_func, &rel_info_func,
560 /* If we didn't find ops for the given node name, it is not supported */
561 if (!get_info_func || !rel_info_func || !node_match_func) {
562 pr_err("MD: %s node is not supported\n", node_name);
566 mdesc_for_each_node_by_name(hp, hp_node, node_name) {
567 rv = get_info_func(hp, hp_node, &hp_node_info);
571 if (node_match_func(node_info, &hp_node_info))
574 rel_info_func(&hp_node_info);
577 rel_info_func(&hp_node_info);
581 EXPORT_SYMBOL(mdesc_get_node);
583 int mdesc_get_node_info(struct mdesc_handle *hp, u64 node,
584 const char *node_name, union md_node_info *node_info)
586 mdesc_node_info_get_f get_info_func;
589 if (hp == NULL || node == MDESC_NODE_NULL ||
590 node_name == NULL || node_info == NULL)
593 /* Find the get_info op for the given node name */
594 mdesc_get_node_ops(node_name, &get_info_func, NULL, NULL);
596 /* If we didn't find a get_info_func, the node name is not supported */
597 if (get_info_func == NULL) {
598 pr_err("MD: %s node is not supported\n", node_name);
602 rv = get_info_func(hp, node, node_info);
604 pr_err("MD: Cannot find 1 or more required match properties for %s node.\n",
611 EXPORT_SYMBOL(mdesc_get_node_info);
613 static struct mdesc_elem *node_block(struct mdesc_hdr *mdesc)
615 return (struct mdesc_elem *) (mdesc + 1);
618 static void *name_block(struct mdesc_hdr *mdesc)
620 return ((void *) node_block(mdesc)) + mdesc->node_sz;
623 static void *data_block(struct mdesc_hdr *mdesc)
625 return ((void *) name_block(mdesc)) + mdesc->name_sz;
628 u64 mdesc_node_by_name(struct mdesc_handle *hp,
629 u64 from_node, const char *name)
631 struct mdesc_elem *ep = node_block(&hp->mdesc);
632 const char *names = name_block(&hp->mdesc);
633 u64 last_node = hp->mdesc.node_sz / 16;
636 if (from_node == MDESC_NODE_NULL) {
638 } else if (from_node >= last_node) {
639 return MDESC_NODE_NULL;
641 ret = ep[from_node].d.val;
644 while (ret < last_node) {
645 if (ep[ret].tag != MD_NODE)
646 return MDESC_NODE_NULL;
647 if (!strcmp(names + ep[ret].name_offset, name))
651 if (ret >= last_node)
652 ret = MDESC_NODE_NULL;
655 EXPORT_SYMBOL(mdesc_node_by_name);
657 const void *mdesc_get_property(struct mdesc_handle *hp, u64 node,
658 const char *name, int *lenp)
660 const char *names = name_block(&hp->mdesc);
661 u64 last_node = hp->mdesc.node_sz / 16;
662 void *data = data_block(&hp->mdesc);
663 struct mdesc_elem *ep;
665 if (node == MDESC_NODE_NULL || node >= last_node)
668 ep = node_block(&hp->mdesc) + node;
670 for (; ep->tag != MD_NODE_END; ep++) {
682 val = data + ep->d.data.data_offset;
683 len = ep->d.data.data_len;
692 if (!strcmp(names + ep->name_offset, name)) {
701 EXPORT_SYMBOL(mdesc_get_property);
703 u64 mdesc_next_arc(struct mdesc_handle *hp, u64 from, const char *arc_type)
705 struct mdesc_elem *ep, *base = node_block(&hp->mdesc);
706 const char *names = name_block(&hp->mdesc);
707 u64 last_node = hp->mdesc.node_sz / 16;
709 if (from == MDESC_NODE_NULL || from >= last_node)
710 return MDESC_NODE_NULL;
715 for (; ep->tag != MD_NODE_END; ep++) {
716 if (ep->tag != MD_PROP_ARC)
719 if (strcmp(names + ep->name_offset, arc_type))
725 return MDESC_NODE_NULL;
727 EXPORT_SYMBOL(mdesc_next_arc);
729 u64 mdesc_arc_target(struct mdesc_handle *hp, u64 arc)
731 struct mdesc_elem *ep, *base = node_block(&hp->mdesc);
737 EXPORT_SYMBOL(mdesc_arc_target);
739 const char *mdesc_node_name(struct mdesc_handle *hp, u64 node)
741 struct mdesc_elem *ep, *base = node_block(&hp->mdesc);
742 const char *names = name_block(&hp->mdesc);
743 u64 last_node = hp->mdesc.node_sz / 16;
745 if (node == MDESC_NODE_NULL || node >= last_node)
749 if (ep->tag != MD_NODE)
752 return names + ep->name_offset;
754 EXPORT_SYMBOL(mdesc_node_name);
756 static u64 max_cpus = 64;
758 static void __init report_platform_properties(void)
760 struct mdesc_handle *hp = mdesc_grab();
761 u64 pn = mdesc_node_by_name(hp, MDESC_NODE_NULL, "platform");
765 if (pn == MDESC_NODE_NULL) {
766 prom_printf("No platform node in machine-description.\n");
770 s = mdesc_get_property(hp, pn, "banner-name", NULL);
771 printk("PLATFORM: banner-name [%s]\n", s);
772 s = mdesc_get_property(hp, pn, "name", NULL);
773 printk("PLATFORM: name [%s]\n", s);
775 v = mdesc_get_property(hp, pn, "hostid", NULL);
777 printk("PLATFORM: hostid [%08llx]\n", *v);
778 v = mdesc_get_property(hp, pn, "serial#", NULL);
780 printk("PLATFORM: serial# [%08llx]\n", *v);
781 v = mdesc_get_property(hp, pn, "stick-frequency", NULL);
782 printk("PLATFORM: stick-frequency [%08llx]\n", *v);
783 v = mdesc_get_property(hp, pn, "mac-address", NULL);
785 printk("PLATFORM: mac-address [%llx]\n", *v);
786 v = mdesc_get_property(hp, pn, "watchdog-resolution", NULL);
788 printk("PLATFORM: watchdog-resolution [%llu ms]\n", *v);
789 v = mdesc_get_property(hp, pn, "watchdog-max-timeout", NULL);
791 printk("PLATFORM: watchdog-max-timeout [%llu ms]\n", *v);
792 v = mdesc_get_property(hp, pn, "max-cpus", NULL);
795 printk("PLATFORM: max-cpus [%llu]\n", max_cpus);
804 if (max_cpu > NR_CPUS)
809 for (i = 0; i < max_cpu; i++)
810 set_cpu_possible(i, true);
817 static void fill_in_one_cache(cpuinfo_sparc *c, struct mdesc_handle *hp, u64 mp)
819 const u64 *level = mdesc_get_property(hp, mp, "level", NULL);
820 const u64 *size = mdesc_get_property(hp, mp, "size", NULL);
821 const u64 *line_size = mdesc_get_property(hp, mp, "line-size", NULL);
825 type = mdesc_get_property(hp, mp, "type", &type_len);
829 if (of_find_in_proplist(type, "instn", type_len)) {
830 c->icache_size = *size;
831 c->icache_line_size = *line_size;
832 } else if (of_find_in_proplist(type, "data", type_len)) {
833 c->dcache_size = *size;
834 c->dcache_line_size = *line_size;
839 c->ecache_size = *size;
840 c->ecache_line_size = *line_size;
850 mdesc_for_each_arc(a, hp, mp, MDESC_ARC_TYPE_FWD) {
851 u64 target = mdesc_arc_target(hp, a);
852 const char *name = mdesc_node_name(hp, target);
854 if (!strcmp(name, "cache"))
855 fill_in_one_cache(c, hp, target);
860 static void find_back_node_value(struct mdesc_handle *hp, u64 node,
862 void (*func)(struct mdesc_handle *, u64, int),
867 /* Since we have an estimate of recursion depth, do a sanity check. */
871 mdesc_for_each_arc(arc, hp, node, MDESC_ARC_TYPE_BACK) {
872 u64 n = mdesc_arc_target(hp, arc);
873 const char *name = mdesc_node_name(hp, n);
875 if (!strcmp(srch_val, name))
878 find_back_node_value(hp, n, srch_val, func, val, depth-1);
882 static void __mark_core_id(struct mdesc_handle *hp, u64 node,
885 const u64 *id = mdesc_get_property(hp, node, "id", NULL);
887 if (*id < num_possible_cpus())
888 cpu_data(*id).core_id = core_id;
891 static void __mark_max_cache_id(struct mdesc_handle *hp, u64 node,
894 const u64 *id = mdesc_get_property(hp, node, "id", NULL);
896 if (*id < num_possible_cpus()) {
897 cpu_data(*id).max_cache_id = max_cache_id;
900 * On systems without explicit socket descriptions socket
903 cpu_data(*id).sock_id = max_cache_id;
907 static void mark_core_ids(struct mdesc_handle *hp, u64 mp,
910 find_back_node_value(hp, mp, "cpu", __mark_core_id, core_id, 10);
913 static void mark_max_cache_ids(struct mdesc_handle *hp, u64 mp,
916 find_back_node_value(hp, mp, "cpu", __mark_max_cache_id,
920 static void set_core_ids(struct mdesc_handle *hp)
927 /* Identify unique cores by looking for cpus backpointed to by
928 * level 1 instruction caches.
930 mdesc_for_each_node_by_name(hp, mp, "cache") {
935 level = mdesc_get_property(hp, mp, "level", NULL);
939 type = mdesc_get_property(hp, mp, "type", &len);
940 if (!of_find_in_proplist(type, "instn", len))
943 mark_core_ids(hp, mp, idx);
948 static int set_max_cache_ids_by_cache(struct mdesc_handle *hp, int level)
955 * Identify unique highest level of shared cache by looking for cpus
956 * backpointed to by shared level N caches.
958 mdesc_for_each_node_by_name(hp, mp, "cache") {
961 cur_lvl = mdesc_get_property(hp, mp, "level", NULL);
962 if (*cur_lvl != level)
964 mark_max_cache_ids(hp, mp, idx);
971 static void set_sock_ids_by_socket(struct mdesc_handle *hp, u64 mp)
975 mdesc_for_each_node_by_name(hp, mp, "socket") {
978 mdesc_for_each_arc(a, hp, mp, MDESC_ARC_TYPE_FWD) {
979 u64 t = mdesc_arc_target(hp, a);
983 name = mdesc_node_name(hp, t);
984 if (strcmp(name, "cpu"))
987 id = mdesc_get_property(hp, t, "id", NULL);
988 if (*id < num_possible_cpus())
989 cpu_data(*id).sock_id = idx;
995 static void set_sock_ids(struct mdesc_handle *hp)
1000 * Find the highest level of shared cache which pre-T7 is also
1003 if (!set_max_cache_ids_by_cache(hp, 3))
1004 set_max_cache_ids_by_cache(hp, 2);
1006 /* If machine description exposes sockets data use it.*/
1007 mp = mdesc_node_by_name(hp, MDESC_NODE_NULL, "sockets");
1008 if (mp != MDESC_NODE_NULL)
1009 set_sock_ids_by_socket(hp, mp);
1012 static void mark_proc_ids(struct mdesc_handle *hp, u64 mp, int proc_id)
1016 mdesc_for_each_arc(a, hp, mp, MDESC_ARC_TYPE_BACK) {
1017 u64 t = mdesc_arc_target(hp, a);
1021 name = mdesc_node_name(hp, t);
1022 if (strcmp(name, "cpu"))
1025 id = mdesc_get_property(hp, t, "id", NULL);
1027 cpu_data(*id).proc_id = proc_id;
1031 static void __set_proc_ids(struct mdesc_handle *hp, const char *exec_unit_name)
1037 mdesc_for_each_node_by_name(hp, mp, exec_unit_name) {
1041 type = mdesc_get_property(hp, mp, "type", &len);
1042 if (!of_find_in_proplist(type, "int", len) &&
1043 !of_find_in_proplist(type, "integer", len))
1046 mark_proc_ids(hp, mp, idx);
1051 static void set_proc_ids(struct mdesc_handle *hp)
1053 __set_proc_ids(hp, "exec_unit");
1054 __set_proc_ids(hp, "exec-unit");
1057 static void get_one_mondo_bits(const u64 *p, unsigned int *mask,
1058 unsigned long def, unsigned long max)
1066 if (!val || val >= 64)
1072 *mask = ((1U << val) * 64U) - 1U;
1076 *mask = ((1U << def) * 64U) - 1U;
1079 static void get_mondo_data(struct mdesc_handle *hp, u64 mp,
1080 struct trap_per_cpu *tb)
1085 val = mdesc_get_property(hp, mp, "q-cpu-mondo-#bits", NULL);
1086 get_one_mondo_bits(val, &tb->cpu_mondo_qmask, 7, ilog2(max_cpus * 2));
1088 val = mdesc_get_property(hp, mp, "q-dev-mondo-#bits", NULL);
1089 get_one_mondo_bits(val, &tb->dev_mondo_qmask, 7, 8);
1091 val = mdesc_get_property(hp, mp, "q-resumable-#bits", NULL);
1092 get_one_mondo_bits(val, &tb->resum_qmask, 6, 7);
1094 val = mdesc_get_property(hp, mp, "q-nonresumable-#bits", NULL);
1095 get_one_mondo_bits(val, &tb->nonresum_qmask, 2, 2);
1097 pr_info("SUN4V: Mondo queue sizes "
1098 "[cpu(%u) dev(%u) r(%u) nr(%u)]\n",
1099 tb->cpu_mondo_qmask + 1,
1100 tb->dev_mondo_qmask + 1,
1101 tb->resum_qmask + 1,
1102 tb->nonresum_qmask + 1);
1106 static void *mdesc_iterate_over_cpus(void *(*func)(struct mdesc_handle *, u64, int, void *), void *arg, cpumask_t *mask)
1108 struct mdesc_handle *hp = mdesc_grab();
1112 mdesc_for_each_node_by_name(hp, mp, "cpu") {
1113 const u64 *id = mdesc_get_property(hp, mp, "id", NULL);
1117 if (cpuid >= NR_CPUS) {
1118 printk(KERN_WARNING "Ignoring CPU %d which is "
1119 ">= NR_CPUS (%d)\n",
1123 if (!cpumask_test_cpu(cpuid, mask))
1127 ret = func(hp, mp, cpuid, arg);
1136 static void *record_one_cpu(struct mdesc_handle *hp, u64 mp, int cpuid,
1141 set_cpu_present(cpuid, true);
1146 void mdesc_populate_present_mask(cpumask_t *mask)
1148 if (tlb_type != hypervisor)
1152 mdesc_iterate_over_cpus(record_one_cpu, NULL, mask);
1155 static void * __init check_one_pgsz(struct mdesc_handle *hp, u64 mp, int cpuid, void *arg)
1157 const u64 *pgsz_prop = mdesc_get_property(hp, mp, "mmu-page-size-list", NULL);
1158 unsigned long *pgsz_mask = arg;
1161 val = (HV_PGSZ_MASK_8K | HV_PGSZ_MASK_64K |
1162 HV_PGSZ_MASK_512K | HV_PGSZ_MASK_4MB);
1173 void __init mdesc_get_page_sizes(cpumask_t *mask, unsigned long *pgsz_mask)
1176 mdesc_iterate_over_cpus(check_one_pgsz, pgsz_mask, mask);
1179 static void *fill_in_one_cpu(struct mdesc_handle *hp, u64 mp, int cpuid,
1182 const u64 *cfreq = mdesc_get_property(hp, mp, "clock-frequency", NULL);
1183 struct trap_per_cpu *tb;
1188 /* On uniprocessor we only want the values for the
1189 * real physical cpu the kernel booted onto, however
1190 * cpu_data() only has one entry at index 0.
1192 if (cpuid != real_hard_smp_processor_id())
1197 c = &cpu_data(cpuid);
1198 c->clock_tick = *cfreq;
1200 tb = &trap_block[cpuid];
1201 get_mondo_data(hp, mp, tb);
1203 mdesc_for_each_arc(a, hp, mp, MDESC_ARC_TYPE_FWD) {
1204 u64 j, t = mdesc_arc_target(hp, a);
1207 t_name = mdesc_node_name(hp, t);
1208 if (!strcmp(t_name, "cache")) {
1209 fill_in_one_cache(c, hp, t);
1213 mdesc_for_each_arc(j, hp, t, MDESC_ARC_TYPE_FWD) {
1214 u64 n = mdesc_arc_target(hp, j);
1217 n_name = mdesc_node_name(hp, n);
1218 if (!strcmp(n_name, "cache"))
1219 fill_in_one_cache(c, hp, n);
1229 void mdesc_fill_in_cpu_data(cpumask_t *mask)
1231 struct mdesc_handle *hp;
1233 mdesc_iterate_over_cpus(fill_in_one_cpu, NULL, mask);
1243 smp_fill_in_sib_core_maps();
1246 /* mdesc_open() - Grab a reference to mdesc_handle when /dev/mdesc is
1247 * opened. Hold this reference until /dev/mdesc is closed to ensure
1248 * mdesc data structure is not released underneath us. Store the
1249 * pointer to mdesc structure in private_data for read and seek to use
1251 static int mdesc_open(struct inode *inode, struct file *file)
1253 struct mdesc_handle *hp = mdesc_grab();
1258 file->private_data = hp;
1263 static ssize_t mdesc_read(struct file *file, char __user *buf,
1264 size_t len, loff_t *offp)
1266 struct mdesc_handle *hp = file->private_data;
1267 unsigned char *mdesc;
1268 int bytes_left, count = len;
1270 if (*offp >= hp->handle_size)
1273 bytes_left = hp->handle_size - *offp;
1274 if (count > bytes_left)
1277 mdesc = (unsigned char *)&hp->mdesc;
1279 if (!copy_to_user(buf, mdesc, count)) {
1287 static loff_t mdesc_llseek(struct file *file, loff_t offset, int whence)
1289 struct mdesc_handle *hp = file->private_data;
1291 return no_seek_end_llseek_size(file, offset, whence, hp->handle_size);
1294 /* mdesc_close() - /dev/mdesc is being closed, release the reference to
1297 static int mdesc_close(struct inode *inode, struct file *file)
1299 mdesc_release(file->private_data);
1303 static const struct file_operations mdesc_fops = {
1306 .llseek = mdesc_llseek,
1307 .release = mdesc_close,
1308 .owner = THIS_MODULE,
1311 static struct miscdevice mdesc_misc = {
1312 .minor = MISC_DYNAMIC_MINOR,
1314 .fops = &mdesc_fops,
1317 static int __init mdesc_misc_init(void)
1319 return misc_register(&mdesc_misc);
1322 __initcall(mdesc_misc_init);
1324 void __init sun4v_mdesc_init(void)
1326 struct mdesc_handle *hp;
1327 unsigned long len, real_len, status;
1329 (void) sun4v_mach_desc(0UL, 0UL, &len);
1331 printk("MDESC: Size is %lu bytes.\n", len);
1333 hp = mdesc_alloc(len, &memblock_mdesc_ops);
1335 prom_printf("MDESC: alloc of %lu bytes failed.\n", len);
1339 status = sun4v_mach_desc(__pa(&hp->mdesc), len, &real_len);
1340 if (status != HV_EOK || real_len > len) {
1341 prom_printf("sun4v_mach_desc fails, err(%lu), "
1342 "len(%lu), real_len(%lu)\n",
1343 status, len, real_len);
1351 report_platform_properties();