1 /* mdesc.c: Sun4V machine description handling.
3 * Copyright (C) 2007 David S. Miller <davem@davemloft.net>
5 #include <linux/kernel.h>
6 #include <linux/types.h>
7 #include <linux/bootmem.h>
8 #include <linux/log2.h>
9 #include <linux/list.h>
10 #include <linux/slab.h>
13 #include <asm/hypervisor.h>
14 #include <asm/mdesc.h>
16 #include <asm/oplib.h>
19 /* Unlike the OBP device tree, the machine description is a full-on
20 * DAG. An arbitrary number of ARCs are possible from one
21 * node to other nodes and thus we can't use the OBP device_node
22 * data structure to represent these nodes inside of the kernel.
24 * Actually, it isn't even a DAG, because there are back pointers
25 * which create cycles in the graph.
27 * mdesc_hdr and mdesc_elem describe the layout of the data structure
28 * we get from the Hypervisor.
31 u32 version; /* Transport version */
32 u32 node_sz; /* node block size */
33 u32 name_sz; /* name block size */
34 u32 data_sz; /* data block size */
35 } __attribute__((aligned(16)));
39 #define MD_LIST_END 0x00
41 #define MD_NODE_END 0x45
43 #define MD_PROP_ARC 0x61
44 #define MD_PROP_VAL 0x76
45 #define MD_PROP_STR 0x73
46 #define MD_PROP_DATA 0x64
59 struct mdesc_mem_ops {
60 struct mdesc_handle *(*alloc)(unsigned int mdesc_size);
61 void (*free)(struct mdesc_handle *handle);
65 struct list_head list;
66 struct mdesc_mem_ops *mops;
69 unsigned int handle_size;
70 struct mdesc_hdr mdesc;
73 static void mdesc_handle_init(struct mdesc_handle *hp,
74 unsigned int handle_size,
77 BUG_ON(((unsigned long)&hp->mdesc) & (16UL - 1));
79 memset(hp, 0, handle_size);
80 INIT_LIST_HEAD(&hp->list);
82 atomic_set(&hp->refcnt, 1);
83 hp->handle_size = handle_size;
86 static struct mdesc_handle * __init mdesc_bootmem_alloc(unsigned int mdesc_size)
88 struct mdesc_handle *hp;
89 unsigned int handle_size, alloc_size;
91 handle_size = (sizeof(struct mdesc_handle) -
92 sizeof(struct mdesc_hdr) +
94 alloc_size = PAGE_ALIGN(handle_size);
96 hp = __alloc_bootmem(alloc_size, PAGE_SIZE, 0UL);
98 mdesc_handle_init(hp, handle_size, hp);
103 static void mdesc_bootmem_free(struct mdesc_handle *hp)
105 unsigned int alloc_size, handle_size = hp->handle_size;
106 unsigned long start, end;
108 BUG_ON(atomic_read(&hp->refcnt) != 0);
109 BUG_ON(!list_empty(&hp->list));
111 alloc_size = PAGE_ALIGN(handle_size);
113 start = (unsigned long) hp;
114 end = start + alloc_size;
116 while (start < end) {
119 p = virt_to_page(start);
120 ClearPageReserved(p);
126 static struct mdesc_mem_ops bootmem_mdesc_ops = {
127 .alloc = mdesc_bootmem_alloc,
128 .free = mdesc_bootmem_free,
131 static struct mdesc_handle *mdesc_kmalloc(unsigned int mdesc_size)
133 unsigned int handle_size;
136 handle_size = (sizeof(struct mdesc_handle) -
137 sizeof(struct mdesc_hdr) +
140 base = kmalloc(handle_size + 15, GFP_KERNEL | __GFP_NOFAIL);
142 struct mdesc_handle *hp;
145 addr = (unsigned long)base;
146 addr = (addr + 15UL) & ~15UL;
147 hp = (struct mdesc_handle *) addr;
149 mdesc_handle_init(hp, handle_size, base);
156 static void mdesc_kfree(struct mdesc_handle *hp)
158 BUG_ON(atomic_read(&hp->refcnt) != 0);
159 BUG_ON(!list_empty(&hp->list));
161 kfree(hp->self_base);
164 static struct mdesc_mem_ops kmalloc_mdesc_memops = {
165 .alloc = mdesc_kmalloc,
169 static struct mdesc_handle *mdesc_alloc(unsigned int mdesc_size,
170 struct mdesc_mem_ops *mops)
172 struct mdesc_handle *hp = mops->alloc(mdesc_size);
180 static void mdesc_free(struct mdesc_handle *hp)
185 static struct mdesc_handle *cur_mdesc;
186 static LIST_HEAD(mdesc_zombie_list);
187 static DEFINE_SPINLOCK(mdesc_lock);
189 struct mdesc_handle *mdesc_grab(void)
191 struct mdesc_handle *hp;
194 spin_lock_irqsave(&mdesc_lock, flags);
197 atomic_inc(&hp->refcnt);
198 spin_unlock_irqrestore(&mdesc_lock, flags);
202 EXPORT_SYMBOL(mdesc_grab);
204 void mdesc_release(struct mdesc_handle *hp)
208 spin_lock_irqsave(&mdesc_lock, flags);
209 if (atomic_dec_and_test(&hp->refcnt)) {
210 list_del_init(&hp->list);
213 spin_unlock_irqrestore(&mdesc_lock, flags);
215 EXPORT_SYMBOL(mdesc_release);
217 static DEFINE_MUTEX(mdesc_mutex);
218 static struct mdesc_notifier_client *client_list;
220 void mdesc_register_notifier(struct mdesc_notifier_client *client)
224 mutex_lock(&mdesc_mutex);
225 client->next = client_list;
226 client_list = client;
228 mdesc_for_each_node_by_name(cur_mdesc, node, client->node_name)
229 client->add(cur_mdesc, node);
231 mutex_unlock(&mdesc_mutex);
234 static const u64 *parent_cfg_handle(struct mdesc_handle *hp, u64 node)
240 mdesc_for_each_arc(a, hp, node, MDESC_ARC_TYPE_BACK) {
243 target = mdesc_arc_target(hp, a);
244 id = mdesc_get_property(hp, target,
253 /* Run 'func' on nodes which are in A but not in B. */
254 static void invoke_on_missing(const char *name,
255 struct mdesc_handle *a,
256 struct mdesc_handle *b,
257 void (*func)(struct mdesc_handle *, u64))
261 mdesc_for_each_node_by_name(a, node, name) {
262 int found = 0, is_vdc_port = 0;
263 const char *name_prop;
267 name_prop = mdesc_get_property(a, node, "name", NULL);
268 if (name_prop && !strcmp(name_prop, "vdc-port")) {
270 id = parent_cfg_handle(a, node);
272 id = mdesc_get_property(a, node, "id", NULL);
275 printk(KERN_ERR "MD: Cannot find ID for %s node.\n",
276 (name_prop ? name_prop : name));
280 mdesc_for_each_node_by_name(b, fnode, name) {
284 name_prop = mdesc_get_property(b, fnode,
287 strcmp(name_prop, "vdc-port"))
289 fid = parent_cfg_handle(b, fnode);
291 printk(KERN_ERR "MD: Cannot find ID "
292 "for vdc-port node.\n");
296 fid = mdesc_get_property(b, fnode,
309 static void notify_one(struct mdesc_notifier_client *p,
310 struct mdesc_handle *old_hp,
311 struct mdesc_handle *new_hp)
313 invoke_on_missing(p->node_name, old_hp, new_hp, p->remove);
314 invoke_on_missing(p->node_name, new_hp, old_hp, p->add);
317 static void mdesc_notify_clients(struct mdesc_handle *old_hp,
318 struct mdesc_handle *new_hp)
320 struct mdesc_notifier_client *p = client_list;
323 notify_one(p, old_hp, new_hp);
328 void mdesc_update(void)
330 unsigned long len, real_len, status;
331 struct mdesc_handle *hp, *orig_hp;
334 mutex_lock(&mdesc_mutex);
336 (void) sun4v_mach_desc(0UL, 0UL, &len);
338 hp = mdesc_alloc(len, &kmalloc_mdesc_memops);
340 printk(KERN_ERR "MD: mdesc alloc fails\n");
344 status = sun4v_mach_desc(__pa(&hp->mdesc), len, &real_len);
345 if (status != HV_EOK || real_len > len) {
346 printk(KERN_ERR "MD: mdesc reread fails with %lu\n",
348 atomic_dec(&hp->refcnt);
353 spin_lock_irqsave(&mdesc_lock, flags);
356 spin_unlock_irqrestore(&mdesc_lock, flags);
358 mdesc_notify_clients(orig_hp, hp);
360 spin_lock_irqsave(&mdesc_lock, flags);
361 if (atomic_dec_and_test(&orig_hp->refcnt))
364 list_add(&orig_hp->list, &mdesc_zombie_list);
365 spin_unlock_irqrestore(&mdesc_lock, flags);
368 mutex_unlock(&mdesc_mutex);
371 static struct mdesc_elem *node_block(struct mdesc_hdr *mdesc)
373 return (struct mdesc_elem *) (mdesc + 1);
376 static void *name_block(struct mdesc_hdr *mdesc)
378 return ((void *) node_block(mdesc)) + mdesc->node_sz;
381 static void *data_block(struct mdesc_hdr *mdesc)
383 return ((void *) name_block(mdesc)) + mdesc->name_sz;
386 u64 mdesc_node_by_name(struct mdesc_handle *hp,
387 u64 from_node, const char *name)
389 struct mdesc_elem *ep = node_block(&hp->mdesc);
390 const char *names = name_block(&hp->mdesc);
391 u64 last_node = hp->mdesc.node_sz / 16;
394 if (from_node == MDESC_NODE_NULL) {
396 } else if (from_node >= last_node) {
397 return MDESC_NODE_NULL;
399 ret = ep[from_node].d.val;
402 while (ret < last_node) {
403 if (ep[ret].tag != MD_NODE)
404 return MDESC_NODE_NULL;
405 if (!strcmp(names + ep[ret].name_offset, name))
409 if (ret >= last_node)
410 ret = MDESC_NODE_NULL;
413 EXPORT_SYMBOL(mdesc_node_by_name);
415 const void *mdesc_get_property(struct mdesc_handle *hp, u64 node,
416 const char *name, int *lenp)
418 const char *names = name_block(&hp->mdesc);
419 u64 last_node = hp->mdesc.node_sz / 16;
420 void *data = data_block(&hp->mdesc);
421 struct mdesc_elem *ep;
423 if (node == MDESC_NODE_NULL || node >= last_node)
426 ep = node_block(&hp->mdesc) + node;
428 for (; ep->tag != MD_NODE_END; ep++) {
440 val = data + ep->d.data.data_offset;
441 len = ep->d.data.data_len;
450 if (!strcmp(names + ep->name_offset, name)) {
459 EXPORT_SYMBOL(mdesc_get_property);
461 u64 mdesc_next_arc(struct mdesc_handle *hp, u64 from, const char *arc_type)
463 struct mdesc_elem *ep, *base = node_block(&hp->mdesc);
464 const char *names = name_block(&hp->mdesc);
465 u64 last_node = hp->mdesc.node_sz / 16;
467 if (from == MDESC_NODE_NULL || from >= last_node)
468 return MDESC_NODE_NULL;
473 for (; ep->tag != MD_NODE_END; ep++) {
474 if (ep->tag != MD_PROP_ARC)
477 if (strcmp(names + ep->name_offset, arc_type))
483 return MDESC_NODE_NULL;
485 EXPORT_SYMBOL(mdesc_next_arc);
487 u64 mdesc_arc_target(struct mdesc_handle *hp, u64 arc)
489 struct mdesc_elem *ep, *base = node_block(&hp->mdesc);
495 EXPORT_SYMBOL(mdesc_arc_target);
497 const char *mdesc_node_name(struct mdesc_handle *hp, u64 node)
499 struct mdesc_elem *ep, *base = node_block(&hp->mdesc);
500 const char *names = name_block(&hp->mdesc);
501 u64 last_node = hp->mdesc.node_sz / 16;
503 if (node == MDESC_NODE_NULL || node >= last_node)
507 if (ep->tag != MD_NODE)
510 return names + ep->name_offset;
512 EXPORT_SYMBOL(mdesc_node_name);
514 static void __init report_platform_properties(void)
516 struct mdesc_handle *hp = mdesc_grab();
517 u64 pn = mdesc_node_by_name(hp, MDESC_NODE_NULL, "platform");
521 if (pn == MDESC_NODE_NULL) {
522 prom_printf("No platform node in machine-description.\n");
526 s = mdesc_get_property(hp, pn, "banner-name", NULL);
527 printk("PLATFORM: banner-name [%s]\n", s);
528 s = mdesc_get_property(hp, pn, "name", NULL);
529 printk("PLATFORM: name [%s]\n", s);
531 v = mdesc_get_property(hp, pn, "hostid", NULL);
533 printk("PLATFORM: hostid [%08lx]\n", *v);
534 v = mdesc_get_property(hp, pn, "serial#", NULL);
536 printk("PLATFORM: serial# [%08lx]\n", *v);
537 v = mdesc_get_property(hp, pn, "stick-frequency", NULL);
538 printk("PLATFORM: stick-frequency [%08lx]\n", *v);
539 v = mdesc_get_property(hp, pn, "mac-address", NULL);
541 printk("PLATFORM: mac-address [%lx]\n", *v);
542 v = mdesc_get_property(hp, pn, "watchdog-resolution", NULL);
544 printk("PLATFORM: watchdog-resolution [%lu ms]\n", *v);
545 v = mdesc_get_property(hp, pn, "watchdog-max-timeout", NULL);
547 printk("PLATFORM: watchdog-max-timeout [%lu ms]\n", *v);
548 v = mdesc_get_property(hp, pn, "max-cpus", NULL);
550 printk("PLATFORM: max-cpus [%lu]\n", *v);
558 if (max_cpu > NR_CPUS)
563 for (i = 0; i < max_cpu; i++)
564 cpu_set(i, cpu_possible_map);
571 static void __devinit fill_in_one_cache(cpuinfo_sparc *c,
572 struct mdesc_handle *hp,
575 const u64 *level = mdesc_get_property(hp, mp, "level", NULL);
576 const u64 *size = mdesc_get_property(hp, mp, "size", NULL);
577 const u64 *line_size = mdesc_get_property(hp, mp, "line-size", NULL);
581 type = mdesc_get_property(hp, mp, "type", &type_len);
585 if (of_find_in_proplist(type, "instn", type_len)) {
586 c->icache_size = *size;
587 c->icache_line_size = *line_size;
588 } else if (of_find_in_proplist(type, "data", type_len)) {
589 c->dcache_size = *size;
590 c->dcache_line_size = *line_size;
595 c->ecache_size = *size;
596 c->ecache_line_size = *line_size;
606 mdesc_for_each_arc(a, hp, mp, MDESC_ARC_TYPE_FWD) {
607 u64 target = mdesc_arc_target(hp, a);
608 const char *name = mdesc_node_name(hp, target);
610 if (!strcmp(name, "cache"))
611 fill_in_one_cache(c, hp, target);
616 static void __devinit mark_core_ids(struct mdesc_handle *hp, u64 mp,
621 mdesc_for_each_arc(a, hp, mp, MDESC_ARC_TYPE_BACK) {
622 u64 t = mdesc_arc_target(hp, a);
626 name = mdesc_node_name(hp, t);
627 if (!strcmp(name, "cpu")) {
628 id = mdesc_get_property(hp, t, "id", NULL);
630 cpu_data(*id).core_id = core_id;
634 mdesc_for_each_arc(j, hp, t, MDESC_ARC_TYPE_BACK) {
635 u64 n = mdesc_arc_target(hp, j);
638 n_name = mdesc_node_name(hp, n);
639 if (strcmp(n_name, "cpu"))
642 id = mdesc_get_property(hp, n, "id", NULL);
644 cpu_data(*id).core_id = core_id;
650 static void __devinit set_core_ids(struct mdesc_handle *hp)
656 mdesc_for_each_node_by_name(hp, mp, "cache") {
661 level = mdesc_get_property(hp, mp, "level", NULL);
665 type = mdesc_get_property(hp, mp, "type", &len);
666 if (!of_find_in_proplist(type, "instn", len))
669 mark_core_ids(hp, mp, idx);
675 static void __devinit mark_proc_ids(struct mdesc_handle *hp, u64 mp,
680 mdesc_for_each_arc(a, hp, mp, MDESC_ARC_TYPE_BACK) {
681 u64 t = mdesc_arc_target(hp, a);
685 name = mdesc_node_name(hp, t);
686 if (strcmp(name, "cpu"))
689 id = mdesc_get_property(hp, t, "id", NULL);
691 cpu_data(*id).proc_id = proc_id;
695 static void __devinit __set_proc_ids(struct mdesc_handle *hp,
696 const char *exec_unit_name)
702 mdesc_for_each_node_by_name(hp, mp, exec_unit_name) {
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))
711 mark_proc_ids(hp, mp, idx);
717 static void __devinit set_proc_ids(struct mdesc_handle *hp)
719 __set_proc_ids(hp, "exec_unit");
720 __set_proc_ids(hp, "exec-unit");
723 static void __devinit get_one_mondo_bits(const u64 *p, unsigned int *mask,
732 if (!val || val >= 64)
735 *mask = ((1U << val) * 64U) - 1U;
739 *mask = ((1U << def) * 64U) - 1U;
742 static void __devinit get_mondo_data(struct mdesc_handle *hp, u64 mp,
743 struct trap_per_cpu *tb)
747 val = mdesc_get_property(hp, mp, "q-cpu-mondo-#bits", NULL);
748 get_one_mondo_bits(val, &tb->cpu_mondo_qmask, 7);
750 val = mdesc_get_property(hp, mp, "q-dev-mondo-#bits", NULL);
751 get_one_mondo_bits(val, &tb->dev_mondo_qmask, 7);
753 val = mdesc_get_property(hp, mp, "q-resumable-#bits", NULL);
754 get_one_mondo_bits(val, &tb->resum_qmask, 6);
756 val = mdesc_get_property(hp, mp, "q-nonresumable-#bits", NULL);
757 get_one_mondo_bits(val, &tb->nonresum_qmask, 2);
760 void __devinit mdesc_fill_in_cpu_data(cpumask_t mask)
762 struct mdesc_handle *hp = mdesc_grab();
766 mdesc_for_each_node_by_name(hp, mp, "cpu") {
767 const u64 *id = mdesc_get_property(hp, mp, "id", NULL);
768 const u64 *cfreq = mdesc_get_property(hp, mp, "clock-frequency", NULL);
769 struct trap_per_cpu *tb;
779 if (cpuid >= NR_CPUS)
781 if (!cpu_isset(cpuid, mask))
784 /* On uniprocessor we only want the values for the
785 * real physical cpu the kernel booted onto, however
786 * cpu_data() only has one entry at index 0.
788 if (cpuid != real_hard_smp_processor_id())
793 c = &cpu_data(cpuid);
794 c->clock_tick = *cfreq;
796 tb = &trap_block[cpuid];
797 get_mondo_data(hp, mp, tb);
799 mdesc_for_each_arc(a, hp, mp, MDESC_ARC_TYPE_FWD) {
800 u64 j, t = mdesc_arc_target(hp, a);
803 t_name = mdesc_node_name(hp, t);
804 if (!strcmp(t_name, "cache")) {
805 fill_in_one_cache(c, hp, t);
809 mdesc_for_each_arc(j, hp, t, MDESC_ARC_TYPE_FWD) {
810 u64 n = mdesc_arc_target(hp, j);
813 n_name = mdesc_node_name(hp, n);
814 if (!strcmp(n_name, "cache"))
815 fill_in_one_cache(c, hp, n);
820 cpu_set(cpuid, cpu_present_map);
828 sparc64_multi_core = 1;
834 smp_fill_in_sib_core_maps();
839 void __init sun4v_mdesc_init(void)
841 struct mdesc_handle *hp;
842 unsigned long len, real_len, status;
845 (void) sun4v_mach_desc(0UL, 0UL, &len);
847 printk("MDESC: Size is %lu bytes.\n", len);
849 hp = mdesc_alloc(len, &bootmem_mdesc_ops);
851 prom_printf("MDESC: alloc of %lu bytes failed.\n", len);
855 status = sun4v_mach_desc(__pa(&hp->mdesc), len, &real_len);
856 if (status != HV_EOK || real_len > len) {
857 prom_printf("sun4v_mach_desc fails, err(%lu), "
858 "len(%lu), real_len(%lu)\n",
859 status, len, real_len);
866 report_platform_properties();
869 mdesc_fill_in_cpu_data(mask);