1 /* mdesc.c: Sun4V machine description handling.
3 * Copyright (C) 2007, 2008 David S. Miller <davem@davemloft.net>
5 #include <linux/kernel.h>
6 #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/bootmem.h>
15 #include <asm/cpudata.h>
16 #include <asm/hypervisor.h>
17 #include <asm/mdesc.h>
19 #include <asm/oplib.h>
22 /* Unlike the OBP device tree, the machine description is a full-on
23 * DAG. An arbitrary number of ARCs are possible from one
24 * node to other nodes and thus we can't use the OBP device_node
25 * data structure to represent these nodes inside of the kernel.
27 * Actually, it isn't even a DAG, because there are back pointers
28 * which create cycles in the graph.
30 * mdesc_hdr and mdesc_elem describe the layout of the data structure
31 * we get from the Hypervisor.
34 u32 version; /* Transport version */
35 u32 node_sz; /* node block size */
36 u32 name_sz; /* name block size */
37 u32 data_sz; /* data block size */
38 } __attribute__((aligned(16)));
42 #define MD_LIST_END 0x00
44 #define MD_NODE_END 0x45
46 #define MD_PROP_ARC 0x61
47 #define MD_PROP_VAL 0x76
48 #define MD_PROP_STR 0x73
49 #define MD_PROP_DATA 0x64
62 struct mdesc_mem_ops {
63 struct mdesc_handle *(*alloc)(unsigned int mdesc_size);
64 void (*free)(struct mdesc_handle *handle);
68 struct list_head list;
69 struct mdesc_mem_ops *mops;
72 unsigned int handle_size;
73 struct mdesc_hdr mdesc;
76 static void mdesc_handle_init(struct mdesc_handle *hp,
77 unsigned int handle_size,
80 BUG_ON(((unsigned long)&hp->mdesc) & (16UL - 1));
82 memset(hp, 0, handle_size);
83 INIT_LIST_HEAD(&hp->list);
85 atomic_set(&hp->refcnt, 1);
86 hp->handle_size = handle_size;
89 static struct mdesc_handle * __init mdesc_lmb_alloc(unsigned int mdesc_size)
91 unsigned int handle_size, alloc_size;
92 struct mdesc_handle *hp;
95 handle_size = (sizeof(struct mdesc_handle) -
96 sizeof(struct mdesc_hdr) +
98 alloc_size = PAGE_ALIGN(handle_size);
100 paddr = lmb_alloc(alloc_size, PAGE_SIZE);
105 mdesc_handle_init(hp, handle_size, hp);
110 static void mdesc_lmb_free(struct mdesc_handle *hp)
112 unsigned int alloc_size;
115 BUG_ON(atomic_read(&hp->refcnt) != 0);
116 BUG_ON(!list_empty(&hp->list));
118 alloc_size = PAGE_ALIGN(hp->handle_size);
120 free_bootmem_late(start, alloc_size);
123 static struct mdesc_mem_ops lmb_mdesc_ops = {
124 .alloc = mdesc_lmb_alloc,
125 .free = mdesc_lmb_free,
128 static struct mdesc_handle *mdesc_kmalloc(unsigned int mdesc_size)
130 unsigned int handle_size;
133 handle_size = (sizeof(struct mdesc_handle) -
134 sizeof(struct mdesc_hdr) +
137 base = kmalloc(handle_size + 15, GFP_KERNEL | __GFP_NOFAIL);
139 struct mdesc_handle *hp;
142 addr = (unsigned long)base;
143 addr = (addr + 15UL) & ~15UL;
144 hp = (struct mdesc_handle *) addr;
146 mdesc_handle_init(hp, handle_size, base);
153 static void mdesc_kfree(struct mdesc_handle *hp)
155 BUG_ON(atomic_read(&hp->refcnt) != 0);
156 BUG_ON(!list_empty(&hp->list));
158 kfree(hp->self_base);
161 static struct mdesc_mem_ops kmalloc_mdesc_memops = {
162 .alloc = mdesc_kmalloc,
166 static struct mdesc_handle *mdesc_alloc(unsigned int mdesc_size,
167 struct mdesc_mem_ops *mops)
169 struct mdesc_handle *hp = mops->alloc(mdesc_size);
177 static void mdesc_free(struct mdesc_handle *hp)
182 static struct mdesc_handle *cur_mdesc;
183 static LIST_HEAD(mdesc_zombie_list);
184 static DEFINE_SPINLOCK(mdesc_lock);
186 struct mdesc_handle *mdesc_grab(void)
188 struct mdesc_handle *hp;
191 spin_lock_irqsave(&mdesc_lock, flags);
194 atomic_inc(&hp->refcnt);
195 spin_unlock_irqrestore(&mdesc_lock, flags);
199 EXPORT_SYMBOL(mdesc_grab);
201 void mdesc_release(struct mdesc_handle *hp)
205 spin_lock_irqsave(&mdesc_lock, flags);
206 if (atomic_dec_and_test(&hp->refcnt)) {
207 list_del_init(&hp->list);
210 spin_unlock_irqrestore(&mdesc_lock, flags);
212 EXPORT_SYMBOL(mdesc_release);
214 static DEFINE_MUTEX(mdesc_mutex);
215 static struct mdesc_notifier_client *client_list;
217 void mdesc_register_notifier(struct mdesc_notifier_client *client)
221 mutex_lock(&mdesc_mutex);
222 client->next = client_list;
223 client_list = client;
225 mdesc_for_each_node_by_name(cur_mdesc, node, client->node_name)
226 client->add(cur_mdesc, node);
228 mutex_unlock(&mdesc_mutex);
231 static const u64 *parent_cfg_handle(struct mdesc_handle *hp, u64 node)
237 mdesc_for_each_arc(a, hp, node, MDESC_ARC_TYPE_BACK) {
240 target = mdesc_arc_target(hp, a);
241 id = mdesc_get_property(hp, target,
250 /* Run 'func' on nodes which are in A but not in B. */
251 static void invoke_on_missing(const char *name,
252 struct mdesc_handle *a,
253 struct mdesc_handle *b,
254 void (*func)(struct mdesc_handle *, u64))
258 mdesc_for_each_node_by_name(a, node, name) {
259 int found = 0, is_vdc_port = 0;
260 const char *name_prop;
264 name_prop = mdesc_get_property(a, node, "name", NULL);
265 if (name_prop && !strcmp(name_prop, "vdc-port")) {
267 id = parent_cfg_handle(a, node);
269 id = mdesc_get_property(a, node, "id", NULL);
272 printk(KERN_ERR "MD: Cannot find ID for %s node.\n",
273 (name_prop ? name_prop : name));
277 mdesc_for_each_node_by_name(b, fnode, name) {
281 name_prop = mdesc_get_property(b, fnode,
284 strcmp(name_prop, "vdc-port"))
286 fid = parent_cfg_handle(b, fnode);
288 printk(KERN_ERR "MD: Cannot find ID "
289 "for vdc-port node.\n");
293 fid = mdesc_get_property(b, fnode,
306 static void notify_one(struct mdesc_notifier_client *p,
307 struct mdesc_handle *old_hp,
308 struct mdesc_handle *new_hp)
310 invoke_on_missing(p->node_name, old_hp, new_hp, p->remove);
311 invoke_on_missing(p->node_name, new_hp, old_hp, p->add);
314 static void mdesc_notify_clients(struct mdesc_handle *old_hp,
315 struct mdesc_handle *new_hp)
317 struct mdesc_notifier_client *p = client_list;
320 notify_one(p, old_hp, new_hp);
325 void mdesc_update(void)
327 unsigned long len, real_len, status;
328 struct mdesc_handle *hp, *orig_hp;
331 mutex_lock(&mdesc_mutex);
333 (void) sun4v_mach_desc(0UL, 0UL, &len);
335 hp = mdesc_alloc(len, &kmalloc_mdesc_memops);
337 printk(KERN_ERR "MD: mdesc alloc fails\n");
341 status = sun4v_mach_desc(__pa(&hp->mdesc), len, &real_len);
342 if (status != HV_EOK || real_len > len) {
343 printk(KERN_ERR "MD: mdesc reread fails with %lu\n",
345 atomic_dec(&hp->refcnt);
350 spin_lock_irqsave(&mdesc_lock, flags);
353 spin_unlock_irqrestore(&mdesc_lock, flags);
355 mdesc_notify_clients(orig_hp, hp);
357 spin_lock_irqsave(&mdesc_lock, flags);
358 if (atomic_dec_and_test(&orig_hp->refcnt))
361 list_add(&orig_hp->list, &mdesc_zombie_list);
362 spin_unlock_irqrestore(&mdesc_lock, flags);
365 mutex_unlock(&mdesc_mutex);
368 static struct mdesc_elem *node_block(struct mdesc_hdr *mdesc)
370 return (struct mdesc_elem *) (mdesc + 1);
373 static void *name_block(struct mdesc_hdr *mdesc)
375 return ((void *) node_block(mdesc)) + mdesc->node_sz;
378 static void *data_block(struct mdesc_hdr *mdesc)
380 return ((void *) name_block(mdesc)) + mdesc->name_sz;
383 u64 mdesc_node_by_name(struct mdesc_handle *hp,
384 u64 from_node, const char *name)
386 struct mdesc_elem *ep = node_block(&hp->mdesc);
387 const char *names = name_block(&hp->mdesc);
388 u64 last_node = hp->mdesc.node_sz / 16;
391 if (from_node == MDESC_NODE_NULL) {
393 } else if (from_node >= last_node) {
394 return MDESC_NODE_NULL;
396 ret = ep[from_node].d.val;
399 while (ret < last_node) {
400 if (ep[ret].tag != MD_NODE)
401 return MDESC_NODE_NULL;
402 if (!strcmp(names + ep[ret].name_offset, name))
406 if (ret >= last_node)
407 ret = MDESC_NODE_NULL;
410 EXPORT_SYMBOL(mdesc_node_by_name);
412 const void *mdesc_get_property(struct mdesc_handle *hp, u64 node,
413 const char *name, int *lenp)
415 const char *names = name_block(&hp->mdesc);
416 u64 last_node = hp->mdesc.node_sz / 16;
417 void *data = data_block(&hp->mdesc);
418 struct mdesc_elem *ep;
420 if (node == MDESC_NODE_NULL || node >= last_node)
423 ep = node_block(&hp->mdesc) + node;
425 for (; ep->tag != MD_NODE_END; ep++) {
437 val = data + ep->d.data.data_offset;
438 len = ep->d.data.data_len;
447 if (!strcmp(names + ep->name_offset, name)) {
456 EXPORT_SYMBOL(mdesc_get_property);
458 u64 mdesc_next_arc(struct mdesc_handle *hp, u64 from, const char *arc_type)
460 struct mdesc_elem *ep, *base = node_block(&hp->mdesc);
461 const char *names = name_block(&hp->mdesc);
462 u64 last_node = hp->mdesc.node_sz / 16;
464 if (from == MDESC_NODE_NULL || from >= last_node)
465 return MDESC_NODE_NULL;
470 for (; ep->tag != MD_NODE_END; ep++) {
471 if (ep->tag != MD_PROP_ARC)
474 if (strcmp(names + ep->name_offset, arc_type))
480 return MDESC_NODE_NULL;
482 EXPORT_SYMBOL(mdesc_next_arc);
484 u64 mdesc_arc_target(struct mdesc_handle *hp, u64 arc)
486 struct mdesc_elem *ep, *base = node_block(&hp->mdesc);
492 EXPORT_SYMBOL(mdesc_arc_target);
494 const char *mdesc_node_name(struct mdesc_handle *hp, u64 node)
496 struct mdesc_elem *ep, *base = node_block(&hp->mdesc);
497 const char *names = name_block(&hp->mdesc);
498 u64 last_node = hp->mdesc.node_sz / 16;
500 if (node == MDESC_NODE_NULL || node >= last_node)
504 if (ep->tag != MD_NODE)
507 return names + ep->name_offset;
509 EXPORT_SYMBOL(mdesc_node_name);
511 static void __init report_platform_properties(void)
513 struct mdesc_handle *hp = mdesc_grab();
514 u64 pn = mdesc_node_by_name(hp, MDESC_NODE_NULL, "platform");
518 if (pn == MDESC_NODE_NULL) {
519 prom_printf("No platform node in machine-description.\n");
523 s = mdesc_get_property(hp, pn, "banner-name", NULL);
524 printk("PLATFORM: banner-name [%s]\n", s);
525 s = mdesc_get_property(hp, pn, "name", NULL);
526 printk("PLATFORM: name [%s]\n", s);
528 v = mdesc_get_property(hp, pn, "hostid", NULL);
530 printk("PLATFORM: hostid [%08llx]\n", *v);
531 v = mdesc_get_property(hp, pn, "serial#", NULL);
533 printk("PLATFORM: serial# [%08llx]\n", *v);
534 v = mdesc_get_property(hp, pn, "stick-frequency", NULL);
535 printk("PLATFORM: stick-frequency [%08llx]\n", *v);
536 v = mdesc_get_property(hp, pn, "mac-address", NULL);
538 printk("PLATFORM: mac-address [%llx]\n", *v);
539 v = mdesc_get_property(hp, pn, "watchdog-resolution", NULL);
541 printk("PLATFORM: watchdog-resolution [%llu ms]\n", *v);
542 v = mdesc_get_property(hp, pn, "watchdog-max-timeout", NULL);
544 printk("PLATFORM: watchdog-max-timeout [%llu ms]\n", *v);
545 v = mdesc_get_property(hp, pn, "max-cpus", NULL);
547 printk("PLATFORM: max-cpus [%llu]\n", *v);
555 if (max_cpu > NR_CPUS)
560 for (i = 0; i < max_cpu; i++)
561 set_cpu_possible(i, true);
568 static void __cpuinit fill_in_one_cache(cpuinfo_sparc *c,
569 struct mdesc_handle *hp,
572 const u64 *level = mdesc_get_property(hp, mp, "level", NULL);
573 const u64 *size = mdesc_get_property(hp, mp, "size", NULL);
574 const u64 *line_size = mdesc_get_property(hp, mp, "line-size", NULL);
578 type = mdesc_get_property(hp, mp, "type", &type_len);
582 if (of_find_in_proplist(type, "instn", type_len)) {
583 c->icache_size = *size;
584 c->icache_line_size = *line_size;
585 } else if (of_find_in_proplist(type, "data", type_len)) {
586 c->dcache_size = *size;
587 c->dcache_line_size = *line_size;
592 c->ecache_size = *size;
593 c->ecache_line_size = *line_size;
603 mdesc_for_each_arc(a, hp, mp, MDESC_ARC_TYPE_FWD) {
604 u64 target = mdesc_arc_target(hp, a);
605 const char *name = mdesc_node_name(hp, target);
607 if (!strcmp(name, "cache"))
608 fill_in_one_cache(c, hp, target);
613 static void __cpuinit mark_core_ids(struct mdesc_handle *hp, u64 mp, int core_id)
617 mdesc_for_each_arc(a, hp, mp, MDESC_ARC_TYPE_BACK) {
618 u64 t = mdesc_arc_target(hp, a);
622 name = mdesc_node_name(hp, t);
623 if (!strcmp(name, "cpu")) {
624 id = mdesc_get_property(hp, t, "id", NULL);
626 cpu_data(*id).core_id = core_id;
630 mdesc_for_each_arc(j, hp, t, MDESC_ARC_TYPE_BACK) {
631 u64 n = mdesc_arc_target(hp, j);
634 n_name = mdesc_node_name(hp, n);
635 if (strcmp(n_name, "cpu"))
638 id = mdesc_get_property(hp, n, "id", NULL);
640 cpu_data(*id).core_id = core_id;
646 static void __cpuinit set_core_ids(struct mdesc_handle *hp)
652 mdesc_for_each_node_by_name(hp, mp, "cache") {
657 level = mdesc_get_property(hp, mp, "level", NULL);
661 type = mdesc_get_property(hp, mp, "type", &len);
662 if (!of_find_in_proplist(type, "instn", len))
665 mark_core_ids(hp, mp, idx);
671 static void __cpuinit mark_proc_ids(struct mdesc_handle *hp, u64 mp, int proc_id)
675 mdesc_for_each_arc(a, hp, mp, MDESC_ARC_TYPE_BACK) {
676 u64 t = mdesc_arc_target(hp, a);
680 name = mdesc_node_name(hp, t);
681 if (strcmp(name, "cpu"))
684 id = mdesc_get_property(hp, t, "id", NULL);
686 cpu_data(*id).proc_id = proc_id;
690 static void __cpuinit __set_proc_ids(struct mdesc_handle *hp, const char *exec_unit_name)
696 mdesc_for_each_node_by_name(hp, mp, exec_unit_name) {
700 type = mdesc_get_property(hp, mp, "type", &len);
701 if (!of_find_in_proplist(type, "int", len) &&
702 !of_find_in_proplist(type, "integer", len))
705 mark_proc_ids(hp, mp, idx);
711 static void __cpuinit set_proc_ids(struct mdesc_handle *hp)
713 __set_proc_ids(hp, "exec_unit");
714 __set_proc_ids(hp, "exec-unit");
717 static void __cpuinit get_one_mondo_bits(const u64 *p, unsigned int *mask,
726 if (!val || val >= 64)
729 *mask = ((1U << val) * 64U) - 1U;
733 *mask = ((1U << def) * 64U) - 1U;
736 static void __cpuinit get_mondo_data(struct mdesc_handle *hp, u64 mp,
737 struct trap_per_cpu *tb)
741 val = mdesc_get_property(hp, mp, "q-cpu-mondo-#bits", NULL);
742 get_one_mondo_bits(val, &tb->cpu_mondo_qmask, 7);
744 val = mdesc_get_property(hp, mp, "q-dev-mondo-#bits", NULL);
745 get_one_mondo_bits(val, &tb->dev_mondo_qmask, 7);
747 val = mdesc_get_property(hp, mp, "q-resumable-#bits", NULL);
748 get_one_mondo_bits(val, &tb->resum_qmask, 6);
750 val = mdesc_get_property(hp, mp, "q-nonresumable-#bits", NULL);
751 get_one_mondo_bits(val, &tb->nonresum_qmask, 2);
754 static void * __cpuinit mdesc_iterate_over_cpus(void *(*func)(struct mdesc_handle *, u64, int, void *), void *arg, cpumask_t *mask)
756 struct mdesc_handle *hp = mdesc_grab();
760 mdesc_for_each_node_by_name(hp, mp, "cpu") {
761 const u64 *id = mdesc_get_property(hp, mp, "id", NULL);
765 if (cpuid >= NR_CPUS) {
766 printk(KERN_WARNING "Ignoring CPU %d which is "
771 if (!cpu_isset(cpuid, *mask))
775 ret = func(hp, mp, cpuid, arg);
784 static void * __cpuinit record_one_cpu(struct mdesc_handle *hp, u64 mp, int cpuid, void *arg)
788 set_cpu_present(cpuid, true);
793 void __cpuinit mdesc_populate_present_mask(cpumask_t *mask)
795 if (tlb_type != hypervisor)
799 mdesc_iterate_over_cpus(record_one_cpu, NULL, mask);
802 static void * __cpuinit fill_in_one_cpu(struct mdesc_handle *hp, u64 mp, int cpuid, void *arg)
804 const u64 *cfreq = mdesc_get_property(hp, mp, "clock-frequency", NULL);
805 struct trap_per_cpu *tb;
810 /* On uniprocessor we only want the values for the
811 * real physical cpu the kernel booted onto, however
812 * cpu_data() only has one entry at index 0.
814 if (cpuid != real_hard_smp_processor_id())
819 c = &cpu_data(cpuid);
820 c->clock_tick = *cfreq;
822 tb = &trap_block[cpuid];
823 get_mondo_data(hp, mp, tb);
825 mdesc_for_each_arc(a, hp, mp, MDESC_ARC_TYPE_FWD) {
826 u64 j, t = mdesc_arc_target(hp, a);
829 t_name = mdesc_node_name(hp, t);
830 if (!strcmp(t_name, "cache")) {
831 fill_in_one_cache(c, hp, t);
835 mdesc_for_each_arc(j, hp, t, MDESC_ARC_TYPE_FWD) {
836 u64 n = mdesc_arc_target(hp, j);
839 n_name = mdesc_node_name(hp, n);
840 if (!strcmp(n_name, "cache"))
841 fill_in_one_cache(c, hp, n);
851 void __cpuinit mdesc_fill_in_cpu_data(cpumask_t *mask)
853 struct mdesc_handle *hp;
855 mdesc_iterate_over_cpus(fill_in_one_cpu, NULL, mask);
858 sparc64_multi_core = 1;
868 smp_fill_in_sib_core_maps();
871 static ssize_t mdesc_read(struct file *file, char __user *buf,
872 size_t len, loff_t *offp)
874 struct mdesc_handle *hp = mdesc_grab();
880 err = hp->handle_size;
881 if (len < hp->handle_size)
883 else if (copy_to_user(buf, &hp->mdesc, hp->handle_size))
890 static const struct file_operations mdesc_fops = {
892 .owner = THIS_MODULE,
895 static struct miscdevice mdesc_misc = {
896 .minor = MISC_DYNAMIC_MINOR,
901 static int __init mdesc_misc_init(void)
903 return misc_register(&mdesc_misc);
906 __initcall(mdesc_misc_init);
908 void __init sun4v_mdesc_init(void)
910 struct mdesc_handle *hp;
911 unsigned long len, real_len, status;
913 (void) sun4v_mach_desc(0UL, 0UL, &len);
915 printk("MDESC: Size is %lu bytes.\n", len);
917 hp = mdesc_alloc(len, &lmb_mdesc_ops);
919 prom_printf("MDESC: alloc of %lu bytes failed.\n", len);
923 status = sun4v_mach_desc(__pa(&hp->mdesc), len, &real_len);
924 if (status != HV_EOK || real_len > len) {
925 prom_printf("sun4v_mach_desc fails, err(%lu), "
926 "len(%lu), real_len(%lu)\n",
927 status, len, real_len);
934 report_platform_properties();