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 */
43 } __attribute__((aligned(16)));
47 #define MD_LIST_END 0x00
49 #define MD_NODE_END 0x45
51 #define MD_PROP_ARC 0x61
52 #define MD_PROP_VAL 0x76
53 #define MD_PROP_STR 0x73
54 #define MD_PROP_DATA 0x64
67 struct mdesc_mem_ops {
68 struct mdesc_handle *(*alloc)(unsigned int mdesc_size);
69 void (*free)(struct mdesc_handle *handle);
73 struct list_head list;
74 struct mdesc_mem_ops *mops;
77 unsigned int handle_size;
78 struct mdesc_hdr mdesc;
81 typedef int (*mdesc_node_info_get_f)(struct mdesc_handle *, u64,
82 union md_node_info *);
83 typedef void (*mdesc_node_info_rel_f)(union md_node_info *);
84 typedef bool (*mdesc_node_match_f)(union md_node_info *, union md_node_info *);
88 mdesc_node_info_get_f get_info;
89 mdesc_node_info_rel_f rel_info;
90 mdesc_node_match_f node_match;
93 static int get_vdev_port_node_info(struct mdesc_handle *md, u64 node,
94 union md_node_info *node_info);
95 static void rel_vdev_port_node_info(union md_node_info *node_info);
96 static bool vdev_port_node_match(union md_node_info *a_node_info,
97 union md_node_info *b_node_info);
99 static int get_ds_port_node_info(struct mdesc_handle *md, u64 node,
100 union md_node_info *node_info);
101 static void rel_ds_port_node_info(union md_node_info *node_info);
102 static bool ds_port_node_match(union md_node_info *a_node_info,
103 union md_node_info *b_node_info);
105 /* supported node types which can be registered */
106 static struct md_node_ops md_node_ops_table[] = {
107 {"virtual-device-port", get_vdev_port_node_info,
108 rel_vdev_port_node_info, vdev_port_node_match},
109 {"domain-services-port", get_ds_port_node_info,
110 rel_ds_port_node_info, ds_port_node_match},
111 {NULL, NULL, NULL, NULL}
114 static void mdesc_get_node_ops(const char *node_name,
115 mdesc_node_info_get_f *get_info_f,
116 mdesc_node_info_rel_f *rel_info_f,
117 mdesc_node_match_f *match_f)
133 for (i = 0; md_node_ops_table[i].name != NULL; i++) {
134 if (strcmp(md_node_ops_table[i].name, node_name) == 0) {
136 *get_info_f = md_node_ops_table[i].get_info;
139 *rel_info_f = md_node_ops_table[i].rel_info;
142 *match_f = md_node_ops_table[i].node_match;
149 static void mdesc_handle_init(struct mdesc_handle *hp,
150 unsigned int handle_size,
153 BUG_ON(((unsigned long)&hp->mdesc) & (16UL - 1));
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;
162 static struct mdesc_handle * __init mdesc_memblock_alloc(unsigned int mdesc_size)
164 unsigned int handle_size, alloc_size;
165 struct mdesc_handle *hp;
168 handle_size = (sizeof(struct mdesc_handle) -
169 sizeof(struct mdesc_hdr) +
171 alloc_size = PAGE_ALIGN(handle_size);
173 paddr = memblock_phys_alloc(alloc_size, PAGE_SIZE);
178 mdesc_handle_init(hp, handle_size, hp);
183 static void __init mdesc_memblock_free(struct mdesc_handle *hp)
185 unsigned int alloc_size;
188 BUG_ON(refcount_read(&hp->refcnt) != 0);
189 BUG_ON(!list_empty(&hp->list));
191 alloc_size = PAGE_ALIGN(hp->handle_size);
193 memblock_free_late(start, alloc_size);
196 static struct mdesc_mem_ops memblock_mdesc_ops = {
197 .alloc = mdesc_memblock_alloc,
198 .free = mdesc_memblock_free,
201 static struct mdesc_handle *mdesc_kmalloc(unsigned int mdesc_size)
203 unsigned int handle_size;
204 struct mdesc_handle *hp;
208 handle_size = (sizeof(struct mdesc_handle) -
209 sizeof(struct mdesc_hdr) +
211 base = kmalloc(handle_size + 15, GFP_KERNEL | __GFP_RETRY_MAYFAIL);
215 addr = (unsigned long)base;
216 addr = (addr + 15UL) & ~15UL;
217 hp = (struct mdesc_handle *) addr;
219 mdesc_handle_init(hp, handle_size, base);
224 static void mdesc_kfree(struct mdesc_handle *hp)
226 BUG_ON(refcount_read(&hp->refcnt) != 0);
227 BUG_ON(!list_empty(&hp->list));
229 kfree(hp->self_base);
232 static struct mdesc_mem_ops kmalloc_mdesc_memops = {
233 .alloc = mdesc_kmalloc,
237 static struct mdesc_handle *mdesc_alloc(unsigned int mdesc_size,
238 struct mdesc_mem_ops *mops)
240 struct mdesc_handle *hp = mops->alloc(mdesc_size);
248 static void mdesc_free(struct mdesc_handle *hp)
253 static struct mdesc_handle *cur_mdesc;
254 static LIST_HEAD(mdesc_zombie_list);
255 static DEFINE_SPINLOCK(mdesc_lock);
257 struct mdesc_handle *mdesc_grab(void)
259 struct mdesc_handle *hp;
262 spin_lock_irqsave(&mdesc_lock, flags);
265 refcount_inc(&hp->refcnt);
266 spin_unlock_irqrestore(&mdesc_lock, flags);
270 EXPORT_SYMBOL(mdesc_grab);
272 void mdesc_release(struct mdesc_handle *hp)
276 spin_lock_irqsave(&mdesc_lock, flags);
277 if (refcount_dec_and_test(&hp->refcnt)) {
278 list_del_init(&hp->list);
281 spin_unlock_irqrestore(&mdesc_lock, flags);
283 EXPORT_SYMBOL(mdesc_release);
285 static DEFINE_MUTEX(mdesc_mutex);
286 static struct mdesc_notifier_client *client_list;
288 void mdesc_register_notifier(struct mdesc_notifier_client *client)
290 bool supported = false;
294 mutex_lock(&mdesc_mutex);
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) {
305 pr_err("MD: %s node not supported\n", client->node_name);
306 mutex_unlock(&mdesc_mutex);
310 client->next = client_list;
311 client_list = client;
313 mdesc_for_each_node_by_name(cur_mdesc, node, client->node_name)
314 client->add(cur_mdesc, node, client->node_name);
316 mutex_unlock(&mdesc_mutex);
319 static const u64 *parent_cfg_handle(struct mdesc_handle *hp, u64 node)
325 mdesc_for_each_arc(a, hp, node, MDESC_ARC_TYPE_BACK) {
328 target = mdesc_arc_target(hp, a);
329 id = mdesc_get_property(hp, target,
338 static int get_vdev_port_node_info(struct mdesc_handle *md, u64 node,
339 union md_node_info *node_info)
341 const u64 *parent_cfg_hdlp;
346 * Virtual device nodes are distinguished by:
349 * 3. parent node "cfg-handle" property
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);
355 if (!idp || !name || !parent_cfg_hdlp)
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)
362 node_info->vdev_port.parent_cfg_hdl = *parent_cfg_hdlp;
367 static void rel_vdev_port_node_info(union md_node_info *node_info)
369 if (node_info && node_info->vdev_port.name) {
370 kfree_const(node_info->vdev_port.name);
371 node_info->vdev_port.name = NULL;
375 static bool vdev_port_node_match(union md_node_info *a_node_info,
376 union md_node_info *b_node_info)
378 if (a_node_info->vdev_port.id != b_node_info->vdev_port.id)
381 if (a_node_info->vdev_port.parent_cfg_hdl !=
382 b_node_info->vdev_port.parent_cfg_hdl)
385 if (strncmp(a_node_info->vdev_port.name,
386 b_node_info->vdev_port.name, MDESC_MAX_STR_LEN) != 0)
392 static int get_ds_port_node_info(struct mdesc_handle *md, u64 node,
393 union md_node_info *node_info)
397 /* DS port nodes use the "id" property to distinguish them */
398 idp = mdesc_get_property(md, node, "id", NULL);
402 node_info->ds_port.id = *idp;
407 static void rel_ds_port_node_info(union md_node_info *node_info)
411 static bool ds_port_node_match(union md_node_info *a_node_info,
412 union md_node_info *b_node_info)
414 if (a_node_info->ds_port.id != b_node_info->ds_port.id)
420 /* Run 'func' on nodes which are in A but not in B. */
421 static void invoke_on_missing(const char *name,
422 struct mdesc_handle *a,
423 struct mdesc_handle *b,
424 void (*func)(struct mdesc_handle *, u64,
425 const char *node_name))
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;
438 * Find the get_info, rel_info and node_match ops for the given
441 mdesc_get_node_ops(name, &get_info_func, &rel_info_func,
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);
450 mdesc_for_each_node_by_name(a, a_node, name) {
453 rv = get_info_func(a, a_node, &a_node_info);
455 pr_err("MD: Cannot find 1 or more required match properties for %s node.\n",
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);
466 if (node_match_func(&a_node_info, &b_node_info)) {
468 rel_info_func(&b_node_info);
472 rel_info_func(&b_node_info);
475 rel_info_func(&a_node_info);
478 func(a, a_node, name);
482 static void notify_one(struct mdesc_notifier_client *p,
483 struct mdesc_handle *old_hp,
484 struct mdesc_handle *new_hp)
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);
490 static void mdesc_notify_clients(struct mdesc_handle *old_hp,
491 struct mdesc_handle *new_hp)
493 struct mdesc_notifier_client *p = client_list;
496 notify_one(p, old_hp, new_hp);
501 void mdesc_update(void)
503 unsigned long len, real_len, status;
504 struct mdesc_handle *hp, *orig_hp;
507 mutex_lock(&mdesc_mutex);
509 (void) sun4v_mach_desc(0UL, 0UL, &len);
511 hp = mdesc_alloc(len, &kmalloc_mdesc_memops);
513 printk(KERN_ERR "MD: mdesc alloc fails\n");
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",
521 refcount_dec(&hp->refcnt);
526 spin_lock_irqsave(&mdesc_lock, flags);
529 spin_unlock_irqrestore(&mdesc_lock, flags);
531 mdesc_notify_clients(orig_hp, hp);
533 spin_lock_irqsave(&mdesc_lock, flags);
534 if (refcount_dec_and_test(&orig_hp->refcnt))
537 list_add(&orig_hp->list, &mdesc_zombie_list);
538 spin_unlock_irqrestore(&mdesc_lock, flags);
541 mutex_unlock(&mdesc_mutex);
544 u64 mdesc_get_node(struct mdesc_handle *hp, const char *node_name,
545 union md_node_info *node_info)
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;
554 if (hp == NULL || node_name == NULL || node_info == NULL)
555 return MDESC_NODE_NULL;
557 /* Find the ops for the given node name */
558 mdesc_get_node_ops(node_name, &get_info_func, &rel_info_func,
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);
567 mdesc_for_each_node_by_name(hp, hp_node, node_name) {
568 rv = get_info_func(hp, hp_node, &hp_node_info);
572 if (node_match_func(node_info, &hp_node_info))
575 rel_info_func(&hp_node_info);
578 rel_info_func(&hp_node_info);
582 EXPORT_SYMBOL(mdesc_get_node);
584 int mdesc_get_node_info(struct mdesc_handle *hp, u64 node,
585 const char *node_name, union md_node_info *node_info)
587 mdesc_node_info_get_f get_info_func;
590 if (hp == NULL || node == MDESC_NODE_NULL ||
591 node_name == NULL || node_info == NULL)
594 /* Find the get_info op for the given node name */
595 mdesc_get_node_ops(node_name, &get_info_func, NULL, NULL);
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);
603 rv = get_info_func(hp, node, node_info);
605 pr_err("MD: Cannot find 1 or more required match properties for %s node.\n",
612 EXPORT_SYMBOL(mdesc_get_node_info);
614 static struct mdesc_elem *node_block(struct mdesc_hdr *mdesc)
616 return (struct mdesc_elem *) mdesc->data;
619 static void *name_block(struct mdesc_hdr *mdesc)
621 return ((void *) node_block(mdesc)) + mdesc->node_sz;
624 static void *data_block(struct mdesc_hdr *mdesc)
626 return ((void *) name_block(mdesc)) + mdesc->name_sz;
629 u64 mdesc_node_by_name(struct mdesc_handle *hp,
630 u64 from_node, const char *name)
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;
637 if (from_node == MDESC_NODE_NULL) {
639 } else if (from_node >= last_node) {
640 return MDESC_NODE_NULL;
642 ret = ep[from_node].d.val;
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))
652 if (ret >= last_node)
653 ret = MDESC_NODE_NULL;
656 EXPORT_SYMBOL(mdesc_node_by_name);
658 const void *mdesc_get_property(struct mdesc_handle *hp, u64 node,
659 const char *name, int *lenp)
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;
666 if (node == MDESC_NODE_NULL || node >= last_node)
669 ep = node_block(&hp->mdesc) + node;
671 for (; ep->tag != MD_NODE_END; ep++) {
683 val = data + ep->d.data.data_offset;
684 len = ep->d.data.data_len;
693 if (!strcmp(names + ep->name_offset, name)) {
702 EXPORT_SYMBOL(mdesc_get_property);
704 u64 mdesc_next_arc(struct mdesc_handle *hp, u64 from, const char *arc_type)
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;
710 if (from == MDESC_NODE_NULL || from >= last_node)
711 return MDESC_NODE_NULL;
716 for (; ep->tag != MD_NODE_END; ep++) {
717 if (ep->tag != MD_PROP_ARC)
720 if (strcmp(names + ep->name_offset, arc_type))
726 return MDESC_NODE_NULL;
728 EXPORT_SYMBOL(mdesc_next_arc);
730 u64 mdesc_arc_target(struct mdesc_handle *hp, u64 arc)
732 struct mdesc_elem *ep, *base = node_block(&hp->mdesc);
738 EXPORT_SYMBOL(mdesc_arc_target);
740 const char *mdesc_node_name(struct mdesc_handle *hp, u64 node)
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;
746 if (node == MDESC_NODE_NULL || node >= last_node)
750 if (ep->tag != MD_NODE)
753 return names + ep->name_offset;
755 EXPORT_SYMBOL(mdesc_node_name);
757 static u64 max_cpus = 64;
759 static void __init report_platform_properties(void)
761 struct mdesc_handle *hp = mdesc_grab();
762 u64 pn = mdesc_node_by_name(hp, MDESC_NODE_NULL, "platform");
766 if (pn == MDESC_NODE_NULL) {
767 prom_printf("No platform node in machine-description.\n");
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);
776 v = mdesc_get_property(hp, pn, "hostid", NULL);
778 printk("PLATFORM: hostid [%08llx]\n", *v);
779 v = mdesc_get_property(hp, pn, "serial#", NULL);
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);
786 printk("PLATFORM: mac-address [%llx]\n", *v);
787 v = mdesc_get_property(hp, pn, "watchdog-resolution", NULL);
789 printk("PLATFORM: watchdog-resolution [%llu ms]\n", *v);
790 v = mdesc_get_property(hp, pn, "watchdog-max-timeout", NULL);
792 printk("PLATFORM: watchdog-max-timeout [%llu ms]\n", *v);
793 v = mdesc_get_property(hp, pn, "max-cpus", NULL);
796 printk("PLATFORM: max-cpus [%llu]\n", max_cpus);
805 if (max_cpu > NR_CPUS)
810 for (i = 0; i < max_cpu; i++)
811 set_cpu_possible(i, true);
818 static void fill_in_one_cache(cpuinfo_sparc *c, struct mdesc_handle *hp, u64 mp)
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);
826 type = mdesc_get_property(hp, mp, "type", &type_len);
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;
840 c->ecache_size = *size;
841 c->ecache_line_size = *line_size;
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);
855 if (!strcmp(name, "cache"))
856 fill_in_one_cache(c, hp, target);
861 static void find_back_node_value(struct mdesc_handle *hp, u64 node,
863 void (*func)(struct mdesc_handle *, u64, int),
868 /* Since we have an estimate of recursion depth, do a sanity check. */
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);
876 if (!strcmp(srch_val, name))
879 find_back_node_value(hp, n, srch_val, func, val, depth-1);
883 static void __mark_core_id(struct mdesc_handle *hp, u64 node,
886 const u64 *id = mdesc_get_property(hp, node, "id", NULL);
888 if (*id < num_possible_cpus())
889 cpu_data(*id).core_id = core_id;
892 static void __mark_max_cache_id(struct mdesc_handle *hp, u64 node,
895 const u64 *id = mdesc_get_property(hp, node, "id", NULL);
897 if (*id < num_possible_cpus()) {
898 cpu_data(*id).max_cache_id = max_cache_id;
901 * On systems without explicit socket descriptions socket
904 cpu_data(*id).sock_id = max_cache_id;
908 static void mark_core_ids(struct mdesc_handle *hp, u64 mp,
911 find_back_node_value(hp, mp, "cpu", __mark_core_id, core_id, 10);
914 static void mark_max_cache_ids(struct mdesc_handle *hp, u64 mp,
917 find_back_node_value(hp, mp, "cpu", __mark_max_cache_id,
921 static void set_core_ids(struct mdesc_handle *hp)
928 /* Identify unique cores by looking for cpus backpointed to by
929 * level 1 instruction caches.
931 mdesc_for_each_node_by_name(hp, mp, "cache") {
936 level = mdesc_get_property(hp, mp, "level", NULL);
940 type = mdesc_get_property(hp, mp, "type", &len);
941 if (!of_find_in_proplist(type, "instn", len))
944 mark_core_ids(hp, mp, idx);
949 static int set_max_cache_ids_by_cache(struct mdesc_handle *hp, int level)
956 * Identify unique highest level of shared cache by looking for cpus
957 * backpointed to by shared level N caches.
959 mdesc_for_each_node_by_name(hp, mp, "cache") {
962 cur_lvl = mdesc_get_property(hp, mp, "level", NULL);
963 if (*cur_lvl != level)
965 mark_max_cache_ids(hp, mp, idx);
972 static void set_sock_ids_by_socket(struct mdesc_handle *hp, u64 mp)
976 mdesc_for_each_node_by_name(hp, mp, "socket") {
979 mdesc_for_each_arc(a, hp, mp, MDESC_ARC_TYPE_FWD) {
980 u64 t = mdesc_arc_target(hp, a);
984 name = mdesc_node_name(hp, t);
985 if (strcmp(name, "cpu"))
988 id = mdesc_get_property(hp, t, "id", NULL);
989 if (*id < num_possible_cpus())
990 cpu_data(*id).sock_id = idx;
996 static void set_sock_ids(struct mdesc_handle *hp)
1001 * Find the highest level of shared cache which pre-T7 is also
1004 if (!set_max_cache_ids_by_cache(hp, 3))
1005 set_max_cache_ids_by_cache(hp, 2);
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);
1013 static void mark_proc_ids(struct mdesc_handle *hp, u64 mp, int proc_id)
1017 mdesc_for_each_arc(a, hp, mp, MDESC_ARC_TYPE_BACK) {
1018 u64 t = mdesc_arc_target(hp, a);
1022 name = mdesc_node_name(hp, t);
1023 if (strcmp(name, "cpu"))
1026 id = mdesc_get_property(hp, t, "id", NULL);
1028 cpu_data(*id).proc_id = proc_id;
1032 static void __set_proc_ids(struct mdesc_handle *hp, const char *exec_unit_name)
1038 mdesc_for_each_node_by_name(hp, mp, exec_unit_name) {
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))
1047 mark_proc_ids(hp, mp, idx);
1052 static void set_proc_ids(struct mdesc_handle *hp)
1054 __set_proc_ids(hp, "exec_unit");
1055 __set_proc_ids(hp, "exec-unit");
1058 static void get_one_mondo_bits(const u64 *p, unsigned int *mask,
1059 unsigned long def, unsigned long max)
1067 if (!val || val >= 64)
1073 *mask = ((1U << val) * 64U) - 1U;
1077 *mask = ((1U << def) * 64U) - 1U;
1080 static void get_mondo_data(struct mdesc_handle *hp, u64 mp,
1081 struct trap_per_cpu *tb)
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));
1089 val = mdesc_get_property(hp, mp, "q-dev-mondo-#bits", NULL);
1090 get_one_mondo_bits(val, &tb->dev_mondo_qmask, 7, 8);
1092 val = mdesc_get_property(hp, mp, "q-resumable-#bits", NULL);
1093 get_one_mondo_bits(val, &tb->resum_qmask, 6, 7);
1095 val = mdesc_get_property(hp, mp, "q-nonresumable-#bits", NULL);
1096 get_one_mondo_bits(val, &tb->nonresum_qmask, 2, 2);
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);
1107 static void *mdesc_iterate_over_cpus(void *(*func)(struct mdesc_handle *, u64, int, void *), void *arg, cpumask_t *mask)
1109 struct mdesc_handle *hp = mdesc_grab();
1113 mdesc_for_each_node_by_name(hp, mp, "cpu") {
1114 const u64 *id = mdesc_get_property(hp, mp, "id", NULL);
1118 if (cpuid >= NR_CPUS) {
1119 printk(KERN_WARNING "Ignoring CPU %d which is "
1120 ">= NR_CPUS (%d)\n",
1124 if (!cpumask_test_cpu(cpuid, mask))
1128 ret = func(hp, mp, cpuid, arg);
1137 static void *record_one_cpu(struct mdesc_handle *hp, u64 mp, int cpuid,
1142 set_cpu_present(cpuid, true);
1147 void mdesc_populate_present_mask(cpumask_t *mask)
1149 if (tlb_type != hypervisor)
1153 mdesc_iterate_over_cpus(record_one_cpu, NULL, mask);
1156 static void * __init check_one_pgsz(struct mdesc_handle *hp, u64 mp, int cpuid, void *arg)
1158 const u64 *pgsz_prop = mdesc_get_property(hp, mp, "mmu-page-size-list", NULL);
1159 unsigned long *pgsz_mask = arg;
1162 val = (HV_PGSZ_MASK_8K | HV_PGSZ_MASK_64K |
1163 HV_PGSZ_MASK_512K | HV_PGSZ_MASK_4MB);
1174 void __init mdesc_get_page_sizes(cpumask_t *mask, unsigned long *pgsz_mask)
1177 mdesc_iterate_over_cpus(check_one_pgsz, pgsz_mask, mask);
1180 static void *fill_in_one_cpu(struct mdesc_handle *hp, u64 mp, int cpuid,
1183 const u64 *cfreq = mdesc_get_property(hp, mp, "clock-frequency", NULL);
1184 struct trap_per_cpu *tb;
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.
1193 if (cpuid != real_hard_smp_processor_id())
1198 c = &cpu_data(cpuid);
1199 c->clock_tick = *cfreq;
1201 tb = &trap_block[cpuid];
1202 get_mondo_data(hp, mp, tb);
1204 mdesc_for_each_arc(a, hp, mp, MDESC_ARC_TYPE_FWD) {
1205 u64 j, t = mdesc_arc_target(hp, a);
1208 t_name = mdesc_node_name(hp, t);
1209 if (!strcmp(t_name, "cache")) {
1210 fill_in_one_cache(c, hp, t);
1214 mdesc_for_each_arc(j, hp, t, MDESC_ARC_TYPE_FWD) {
1215 u64 n = mdesc_arc_target(hp, j);
1218 n_name = mdesc_node_name(hp, n);
1219 if (!strcmp(n_name, "cache"))
1220 fill_in_one_cache(c, hp, n);
1230 void mdesc_fill_in_cpu_data(cpumask_t *mask)
1232 struct mdesc_handle *hp;
1234 mdesc_iterate_over_cpus(fill_in_one_cpu, NULL, mask);
1244 smp_fill_in_sib_core_maps();
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
1252 static int mdesc_open(struct inode *inode, struct file *file)
1254 struct mdesc_handle *hp = mdesc_grab();
1259 file->private_data = hp;
1264 static ssize_t mdesc_read(struct file *file, char __user *buf,
1265 size_t len, loff_t *offp)
1267 struct mdesc_handle *hp = file->private_data;
1268 unsigned char *mdesc;
1269 int bytes_left, count = len;
1271 if (*offp >= hp->handle_size)
1274 bytes_left = hp->handle_size - *offp;
1275 if (count > bytes_left)
1278 mdesc = (unsigned char *)&hp->mdesc;
1280 if (!copy_to_user(buf, mdesc, count)) {
1288 static loff_t mdesc_llseek(struct file *file, loff_t offset, int whence)
1290 struct mdesc_handle *hp = file->private_data;
1292 return no_seek_end_llseek_size(file, offset, whence, hp->handle_size);
1295 /* mdesc_close() - /dev/mdesc is being closed, release the reference to
1298 static int mdesc_close(struct inode *inode, struct file *file)
1300 mdesc_release(file->private_data);
1304 static const struct file_operations mdesc_fops = {
1307 .llseek = mdesc_llseek,
1308 .release = mdesc_close,
1309 .owner = THIS_MODULE,
1312 static struct miscdevice mdesc_misc = {
1313 .minor = MISC_DYNAMIC_MINOR,
1315 .fops = &mdesc_fops,
1318 static int __init mdesc_misc_init(void)
1320 return misc_register(&mdesc_misc);
1323 __initcall(mdesc_misc_init);
1325 void __init sun4v_mdesc_init(void)
1327 struct mdesc_handle *hp;
1328 unsigned long len, real_len, status;
1330 (void) sun4v_mach_desc(0UL, 0UL, &len);
1332 printk("MDESC: Size is %lu bytes.\n", len);
1334 hp = mdesc_alloc(len, &memblock_mdesc_ops);
1336 prom_printf("MDESC: alloc of %lu bytes failed.\n", len);
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);
1352 report_platform_properties();