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>
14 #include <asm/hypervisor.h>
15 #include <asm/mdesc.h>
17 #include <asm/oplib.h>
20 /* Unlike the OBP device tree, the machine description is a full-on
21 * DAG. An arbitrary number of ARCs are possible from one
22 * node to other nodes and thus we can't use the OBP device_node
23 * data structure to represent these nodes inside of the kernel.
25 * Actually, it isn't even a DAG, because there are back pointers
26 * which create cycles in the graph.
28 * mdesc_hdr and mdesc_elem describe the layout of the data structure
29 * we get from the Hypervisor.
32 u32 version
; /* Transport version */
33 u32 node_sz
; /* node block size */
34 u32 name_sz
; /* name block size */
35 u32 data_sz
; /* data block size */
36 } __attribute__((aligned(16)));
40 #define MD_LIST_END 0x00
42 #define MD_NODE_END 0x45
44 #define MD_PROP_ARC 0x61
45 #define MD_PROP_VAL 0x76
46 #define MD_PROP_STR 0x73
47 #define MD_PROP_DATA 0x64
60 struct mdesc_mem_ops
{
61 struct mdesc_handle
*(*alloc
)(unsigned int mdesc_size
);
62 void (*free
)(struct mdesc_handle
*handle
);
66 struct list_head list
;
67 struct mdesc_mem_ops
*mops
;
70 unsigned int handle_size
;
71 struct mdesc_hdr mdesc
;
74 static void mdesc_handle_init(struct mdesc_handle
*hp
,
75 unsigned int handle_size
,
78 BUG_ON(((unsigned long)&hp
->mdesc
) & (16UL - 1));
80 memset(hp
, 0, handle_size
);
81 INIT_LIST_HEAD(&hp
->list
);
83 atomic_set(&hp
->refcnt
, 1);
84 hp
->handle_size
= handle_size
;
87 static struct mdesc_handle
* __init
mdesc_lmb_alloc(unsigned int mdesc_size
)
89 unsigned int handle_size
, alloc_size
;
90 struct mdesc_handle
*hp
;
93 handle_size
= (sizeof(struct mdesc_handle
) -
94 sizeof(struct mdesc_hdr
) +
96 alloc_size
= PAGE_ALIGN(handle_size
);
98 paddr
= lmb_alloc(alloc_size
, PAGE_SIZE
);
103 mdesc_handle_init(hp
, handle_size
, hp
);
108 static void mdesc_lmb_free(struct mdesc_handle
*hp
)
110 unsigned int alloc_size
, handle_size
= hp
->handle_size
;
111 unsigned long start
, end
;
113 BUG_ON(atomic_read(&hp
->refcnt
) != 0);
114 BUG_ON(!list_empty(&hp
->list
));
116 alloc_size
= PAGE_ALIGN(handle_size
);
118 start
= (unsigned long) hp
;
119 end
= start
+ alloc_size
;
121 while (start
< end
) {
124 p
= virt_to_page(start
);
125 ClearPageReserved(p
);
131 static struct mdesc_mem_ops lmb_mdesc_ops
= {
132 .alloc
= mdesc_lmb_alloc
,
133 .free
= mdesc_lmb_free
,
136 static struct mdesc_handle
*mdesc_kmalloc(unsigned int mdesc_size
)
138 unsigned int handle_size
;
141 handle_size
= (sizeof(struct mdesc_handle
) -
142 sizeof(struct mdesc_hdr
) +
145 base
= kmalloc(handle_size
+ 15, GFP_KERNEL
| __GFP_NOFAIL
);
147 struct mdesc_handle
*hp
;
150 addr
= (unsigned long)base
;
151 addr
= (addr
+ 15UL) & ~15UL;
152 hp
= (struct mdesc_handle
*) addr
;
154 mdesc_handle_init(hp
, handle_size
, base
);
161 static void mdesc_kfree(struct mdesc_handle
*hp
)
163 BUG_ON(atomic_read(&hp
->refcnt
) != 0);
164 BUG_ON(!list_empty(&hp
->list
));
166 kfree(hp
->self_base
);
169 static struct mdesc_mem_ops kmalloc_mdesc_memops
= {
170 .alloc
= mdesc_kmalloc
,
174 static struct mdesc_handle
*mdesc_alloc(unsigned int mdesc_size
,
175 struct mdesc_mem_ops
*mops
)
177 struct mdesc_handle
*hp
= mops
->alloc(mdesc_size
);
185 static void mdesc_free(struct mdesc_handle
*hp
)
190 static struct mdesc_handle
*cur_mdesc
;
191 static LIST_HEAD(mdesc_zombie_list
);
192 static DEFINE_SPINLOCK(mdesc_lock
);
194 struct mdesc_handle
*mdesc_grab(void)
196 struct mdesc_handle
*hp
;
199 spin_lock_irqsave(&mdesc_lock
, flags
);
202 atomic_inc(&hp
->refcnt
);
203 spin_unlock_irqrestore(&mdesc_lock
, flags
);
207 EXPORT_SYMBOL(mdesc_grab
);
209 void mdesc_release(struct mdesc_handle
*hp
)
213 spin_lock_irqsave(&mdesc_lock
, flags
);
214 if (atomic_dec_and_test(&hp
->refcnt
)) {
215 list_del_init(&hp
->list
);
218 spin_unlock_irqrestore(&mdesc_lock
, flags
);
220 EXPORT_SYMBOL(mdesc_release
);
222 static DEFINE_MUTEX(mdesc_mutex
);
223 static struct mdesc_notifier_client
*client_list
;
225 void mdesc_register_notifier(struct mdesc_notifier_client
*client
)
229 mutex_lock(&mdesc_mutex
);
230 client
->next
= client_list
;
231 client_list
= client
;
233 mdesc_for_each_node_by_name(cur_mdesc
, node
, client
->node_name
)
234 client
->add(cur_mdesc
, node
);
236 mutex_unlock(&mdesc_mutex
);
239 static const u64
*parent_cfg_handle(struct mdesc_handle
*hp
, u64 node
)
245 mdesc_for_each_arc(a
, hp
, node
, MDESC_ARC_TYPE_BACK
) {
248 target
= mdesc_arc_target(hp
, a
);
249 id
= mdesc_get_property(hp
, target
,
258 /* Run 'func' on nodes which are in A but not in B. */
259 static void invoke_on_missing(const char *name
,
260 struct mdesc_handle
*a
,
261 struct mdesc_handle
*b
,
262 void (*func
)(struct mdesc_handle
*, u64
))
266 mdesc_for_each_node_by_name(a
, node
, name
) {
267 int found
= 0, is_vdc_port
= 0;
268 const char *name_prop
;
272 name_prop
= mdesc_get_property(a
, node
, "name", NULL
);
273 if (name_prop
&& !strcmp(name_prop
, "vdc-port")) {
275 id
= parent_cfg_handle(a
, node
);
277 id
= mdesc_get_property(a
, node
, "id", NULL
);
280 printk(KERN_ERR
"MD: Cannot find ID for %s node.\n",
281 (name_prop
? name_prop
: name
));
285 mdesc_for_each_node_by_name(b
, fnode
, name
) {
289 name_prop
= mdesc_get_property(b
, fnode
,
292 strcmp(name_prop
, "vdc-port"))
294 fid
= parent_cfg_handle(b
, fnode
);
296 printk(KERN_ERR
"MD: Cannot find ID "
297 "for vdc-port node.\n");
301 fid
= mdesc_get_property(b
, fnode
,
314 static void notify_one(struct mdesc_notifier_client
*p
,
315 struct mdesc_handle
*old_hp
,
316 struct mdesc_handle
*new_hp
)
318 invoke_on_missing(p
->node_name
, old_hp
, new_hp
, p
->remove
);
319 invoke_on_missing(p
->node_name
, new_hp
, old_hp
, p
->add
);
322 static void mdesc_notify_clients(struct mdesc_handle
*old_hp
,
323 struct mdesc_handle
*new_hp
)
325 struct mdesc_notifier_client
*p
= client_list
;
328 notify_one(p
, old_hp
, new_hp
);
333 void mdesc_update(void)
335 unsigned long len
, real_len
, status
;
336 struct mdesc_handle
*hp
, *orig_hp
;
339 mutex_lock(&mdesc_mutex
);
341 (void) sun4v_mach_desc(0UL, 0UL, &len
);
343 hp
= mdesc_alloc(len
, &kmalloc_mdesc_memops
);
345 printk(KERN_ERR
"MD: mdesc alloc fails\n");
349 status
= sun4v_mach_desc(__pa(&hp
->mdesc
), len
, &real_len
);
350 if (status
!= HV_EOK
|| real_len
> len
) {
351 printk(KERN_ERR
"MD: mdesc reread fails with %lu\n",
353 atomic_dec(&hp
->refcnt
);
358 spin_lock_irqsave(&mdesc_lock
, flags
);
361 spin_unlock_irqrestore(&mdesc_lock
, flags
);
363 mdesc_notify_clients(orig_hp
, hp
);
365 spin_lock_irqsave(&mdesc_lock
, flags
);
366 if (atomic_dec_and_test(&orig_hp
->refcnt
))
369 list_add(&orig_hp
->list
, &mdesc_zombie_list
);
370 spin_unlock_irqrestore(&mdesc_lock
, flags
);
373 mutex_unlock(&mdesc_mutex
);
376 static struct mdesc_elem
*node_block(struct mdesc_hdr
*mdesc
)
378 return (struct mdesc_elem
*) (mdesc
+ 1);
381 static void *name_block(struct mdesc_hdr
*mdesc
)
383 return ((void *) node_block(mdesc
)) + mdesc
->node_sz
;
386 static void *data_block(struct mdesc_hdr
*mdesc
)
388 return ((void *) name_block(mdesc
)) + mdesc
->name_sz
;
391 u64
mdesc_node_by_name(struct mdesc_handle
*hp
,
392 u64 from_node
, const char *name
)
394 struct mdesc_elem
*ep
= node_block(&hp
->mdesc
);
395 const char *names
= name_block(&hp
->mdesc
);
396 u64 last_node
= hp
->mdesc
.node_sz
/ 16;
399 if (from_node
== MDESC_NODE_NULL
) {
401 } else if (from_node
>= last_node
) {
402 return MDESC_NODE_NULL
;
404 ret
= ep
[from_node
].d
.val
;
407 while (ret
< last_node
) {
408 if (ep
[ret
].tag
!= MD_NODE
)
409 return MDESC_NODE_NULL
;
410 if (!strcmp(names
+ ep
[ret
].name_offset
, name
))
414 if (ret
>= last_node
)
415 ret
= MDESC_NODE_NULL
;
418 EXPORT_SYMBOL(mdesc_node_by_name
);
420 const void *mdesc_get_property(struct mdesc_handle
*hp
, u64 node
,
421 const char *name
, int *lenp
)
423 const char *names
= name_block(&hp
->mdesc
);
424 u64 last_node
= hp
->mdesc
.node_sz
/ 16;
425 void *data
= data_block(&hp
->mdesc
);
426 struct mdesc_elem
*ep
;
428 if (node
== MDESC_NODE_NULL
|| node
>= last_node
)
431 ep
= node_block(&hp
->mdesc
) + node
;
433 for (; ep
->tag
!= MD_NODE_END
; ep
++) {
445 val
= data
+ ep
->d
.data
.data_offset
;
446 len
= ep
->d
.data
.data_len
;
455 if (!strcmp(names
+ ep
->name_offset
, name
)) {
464 EXPORT_SYMBOL(mdesc_get_property
);
466 u64
mdesc_next_arc(struct mdesc_handle
*hp
, u64 from
, const char *arc_type
)
468 struct mdesc_elem
*ep
, *base
= node_block(&hp
->mdesc
);
469 const char *names
= name_block(&hp
->mdesc
);
470 u64 last_node
= hp
->mdesc
.node_sz
/ 16;
472 if (from
== MDESC_NODE_NULL
|| from
>= last_node
)
473 return MDESC_NODE_NULL
;
478 for (; ep
->tag
!= MD_NODE_END
; ep
++) {
479 if (ep
->tag
!= MD_PROP_ARC
)
482 if (strcmp(names
+ ep
->name_offset
, arc_type
))
488 return MDESC_NODE_NULL
;
490 EXPORT_SYMBOL(mdesc_next_arc
);
492 u64
mdesc_arc_target(struct mdesc_handle
*hp
, u64 arc
)
494 struct mdesc_elem
*ep
, *base
= node_block(&hp
->mdesc
);
500 EXPORT_SYMBOL(mdesc_arc_target
);
502 const char *mdesc_node_name(struct mdesc_handle
*hp
, u64 node
)
504 struct mdesc_elem
*ep
, *base
= node_block(&hp
->mdesc
);
505 const char *names
= name_block(&hp
->mdesc
);
506 u64 last_node
= hp
->mdesc
.node_sz
/ 16;
508 if (node
== MDESC_NODE_NULL
|| node
>= last_node
)
512 if (ep
->tag
!= MD_NODE
)
515 return names
+ ep
->name_offset
;
517 EXPORT_SYMBOL(mdesc_node_name
);
519 static void __init
report_platform_properties(void)
521 struct mdesc_handle
*hp
= mdesc_grab();
522 u64 pn
= mdesc_node_by_name(hp
, MDESC_NODE_NULL
, "platform");
526 if (pn
== MDESC_NODE_NULL
) {
527 prom_printf("No platform node in machine-description.\n");
531 s
= mdesc_get_property(hp
, pn
, "banner-name", NULL
);
532 printk("PLATFORM: banner-name [%s]\n", s
);
533 s
= mdesc_get_property(hp
, pn
, "name", NULL
);
534 printk("PLATFORM: name [%s]\n", s
);
536 v
= mdesc_get_property(hp
, pn
, "hostid", NULL
);
538 printk("PLATFORM: hostid [%08lx]\n", *v
);
539 v
= mdesc_get_property(hp
, pn
, "serial#", NULL
);
541 printk("PLATFORM: serial# [%08lx]\n", *v
);
542 v
= mdesc_get_property(hp
, pn
, "stick-frequency", NULL
);
543 printk("PLATFORM: stick-frequency [%08lx]\n", *v
);
544 v
= mdesc_get_property(hp
, pn
, "mac-address", NULL
);
546 printk("PLATFORM: mac-address [%lx]\n", *v
);
547 v
= mdesc_get_property(hp
, pn
, "watchdog-resolution", NULL
);
549 printk("PLATFORM: watchdog-resolution [%lu ms]\n", *v
);
550 v
= mdesc_get_property(hp
, pn
, "watchdog-max-timeout", NULL
);
552 printk("PLATFORM: watchdog-max-timeout [%lu ms]\n", *v
);
553 v
= mdesc_get_property(hp
, pn
, "max-cpus", NULL
);
555 printk("PLATFORM: max-cpus [%lu]\n", *v
);
563 if (max_cpu
> NR_CPUS
)
568 for (i
= 0; i
< max_cpu
; i
++)
569 cpu_set(i
, cpu_possible_map
);
576 static void __devinit
fill_in_one_cache(cpuinfo_sparc
*c
,
577 struct mdesc_handle
*hp
,
580 const u64
*level
= mdesc_get_property(hp
, mp
, "level", NULL
);
581 const u64
*size
= mdesc_get_property(hp
, mp
, "size", NULL
);
582 const u64
*line_size
= mdesc_get_property(hp
, mp
, "line-size", NULL
);
586 type
= mdesc_get_property(hp
, mp
, "type", &type_len
);
590 if (of_find_in_proplist(type
, "instn", type_len
)) {
591 c
->icache_size
= *size
;
592 c
->icache_line_size
= *line_size
;
593 } else if (of_find_in_proplist(type
, "data", type_len
)) {
594 c
->dcache_size
= *size
;
595 c
->dcache_line_size
= *line_size
;
600 c
->ecache_size
= *size
;
601 c
->ecache_line_size
= *line_size
;
611 mdesc_for_each_arc(a
, hp
, mp
, MDESC_ARC_TYPE_FWD
) {
612 u64 target
= mdesc_arc_target(hp
, a
);
613 const char *name
= mdesc_node_name(hp
, target
);
615 if (!strcmp(name
, "cache"))
616 fill_in_one_cache(c
, hp
, target
);
621 static void __devinit
mark_core_ids(struct mdesc_handle
*hp
, u64 mp
,
626 mdesc_for_each_arc(a
, hp
, mp
, MDESC_ARC_TYPE_BACK
) {
627 u64 t
= mdesc_arc_target(hp
, a
);
631 name
= mdesc_node_name(hp
, t
);
632 if (!strcmp(name
, "cpu")) {
633 id
= mdesc_get_property(hp
, t
, "id", NULL
);
635 cpu_data(*id
).core_id
= core_id
;
639 mdesc_for_each_arc(j
, hp
, t
, MDESC_ARC_TYPE_BACK
) {
640 u64 n
= mdesc_arc_target(hp
, j
);
643 n_name
= mdesc_node_name(hp
, n
);
644 if (strcmp(n_name
, "cpu"))
647 id
= mdesc_get_property(hp
, n
, "id", NULL
);
649 cpu_data(*id
).core_id
= core_id
;
655 static void __devinit
set_core_ids(struct mdesc_handle
*hp
)
661 mdesc_for_each_node_by_name(hp
, mp
, "cache") {
666 level
= mdesc_get_property(hp
, mp
, "level", NULL
);
670 type
= mdesc_get_property(hp
, mp
, "type", &len
);
671 if (!of_find_in_proplist(type
, "instn", len
))
674 mark_core_ids(hp
, mp
, idx
);
680 static void __devinit
mark_proc_ids(struct mdesc_handle
*hp
, u64 mp
,
685 mdesc_for_each_arc(a
, hp
, mp
, MDESC_ARC_TYPE_BACK
) {
686 u64 t
= mdesc_arc_target(hp
, a
);
690 name
= mdesc_node_name(hp
, t
);
691 if (strcmp(name
, "cpu"))
694 id
= mdesc_get_property(hp
, t
, "id", NULL
);
696 cpu_data(*id
).proc_id
= proc_id
;
700 static void __devinit
__set_proc_ids(struct mdesc_handle
*hp
,
701 const char *exec_unit_name
)
707 mdesc_for_each_node_by_name(hp
, mp
, exec_unit_name
) {
711 type
= mdesc_get_property(hp
, mp
, "type", &len
);
712 if (!of_find_in_proplist(type
, "int", len
) &&
713 !of_find_in_proplist(type
, "integer", len
))
716 mark_proc_ids(hp
, mp
, idx
);
722 static void __devinit
set_proc_ids(struct mdesc_handle
*hp
)
724 __set_proc_ids(hp
, "exec_unit");
725 __set_proc_ids(hp
, "exec-unit");
728 static void __devinit
get_one_mondo_bits(const u64
*p
, unsigned int *mask
,
737 if (!val
|| val
>= 64)
740 *mask
= ((1U << val
) * 64U) - 1U;
744 *mask
= ((1U << def
) * 64U) - 1U;
747 static void __devinit
get_mondo_data(struct mdesc_handle
*hp
, u64 mp
,
748 struct trap_per_cpu
*tb
)
752 val
= mdesc_get_property(hp
, mp
, "q-cpu-mondo-#bits", NULL
);
753 get_one_mondo_bits(val
, &tb
->cpu_mondo_qmask
, 7);
755 val
= mdesc_get_property(hp
, mp
, "q-dev-mondo-#bits", NULL
);
756 get_one_mondo_bits(val
, &tb
->dev_mondo_qmask
, 7);
758 val
= mdesc_get_property(hp
, mp
, "q-resumable-#bits", NULL
);
759 get_one_mondo_bits(val
, &tb
->resum_qmask
, 6);
761 val
= mdesc_get_property(hp
, mp
, "q-nonresumable-#bits", NULL
);
762 get_one_mondo_bits(val
, &tb
->nonresum_qmask
, 2);
765 void __cpuinit
mdesc_fill_in_cpu_data(cpumask_t mask
)
767 struct mdesc_handle
*hp
= mdesc_grab();
771 mdesc_for_each_node_by_name(hp
, mp
, "cpu") {
772 const u64
*id
= mdesc_get_property(hp
, mp
, "id", NULL
);
773 const u64
*cfreq
= mdesc_get_property(hp
, mp
, "clock-frequency", NULL
);
774 struct trap_per_cpu
*tb
;
784 if (cpuid
>= NR_CPUS
) {
785 printk(KERN_WARNING
"Ignoring CPU %d which is "
790 if (!cpu_isset(cpuid
, mask
))
793 /* On uniprocessor we only want the values for the
794 * real physical cpu the kernel booted onto, however
795 * cpu_data() only has one entry at index 0.
797 if (cpuid
!= real_hard_smp_processor_id())
802 c
= &cpu_data(cpuid
);
803 c
->clock_tick
= *cfreq
;
805 tb
= &trap_block
[cpuid
];
806 get_mondo_data(hp
, mp
, tb
);
808 mdesc_for_each_arc(a
, hp
, mp
, MDESC_ARC_TYPE_FWD
) {
809 u64 j
, t
= mdesc_arc_target(hp
, a
);
812 t_name
= mdesc_node_name(hp
, t
);
813 if (!strcmp(t_name
, "cache")) {
814 fill_in_one_cache(c
, hp
, t
);
818 mdesc_for_each_arc(j
, hp
, t
, MDESC_ARC_TYPE_FWD
) {
819 u64 n
= mdesc_arc_target(hp
, j
);
822 n_name
= mdesc_node_name(hp
, n
);
823 if (!strcmp(n_name
, "cache"))
824 fill_in_one_cache(c
, hp
, n
);
829 cpu_set(cpuid
, cpu_present_map
);
837 sparc64_multi_core
= 1;
843 smp_fill_in_sib_core_maps();
848 static ssize_t
mdesc_read(struct file
*file
, char __user
*buf
,
849 size_t len
, loff_t
*offp
)
851 struct mdesc_handle
*hp
= mdesc_grab();
857 err
= hp
->handle_size
;
858 if (len
< hp
->handle_size
)
860 else if (copy_to_user(buf
, &hp
->mdesc
, hp
->handle_size
))
867 static const struct file_operations mdesc_fops
= {
869 .owner
= THIS_MODULE
,
872 static struct miscdevice mdesc_misc
= {
873 .minor
= MISC_DYNAMIC_MINOR
,
878 static int __init
mdesc_misc_init(void)
880 return misc_register(&mdesc_misc
);
883 __initcall(mdesc_misc_init
);
885 void __init
sun4v_mdesc_init(void)
887 struct mdesc_handle
*hp
;
888 unsigned long len
, real_len
, status
;
891 (void) sun4v_mach_desc(0UL, 0UL, &len
);
893 printk("MDESC: Size is %lu bytes.\n", len
);
895 hp
= mdesc_alloc(len
, &lmb_mdesc_ops
);
897 prom_printf("MDESC: alloc of %lu bytes failed.\n", len
);
901 status
= sun4v_mach_desc(__pa(&hp
->mdesc
), len
, &real_len
);
902 if (status
!= HV_EOK
|| real_len
> len
) {
903 prom_printf("sun4v_mach_desc fails, err(%lu), "
904 "len(%lu), real_len(%lu)\n",
905 status
, len
, real_len
);
912 report_platform_properties();
915 mdesc_fill_in_cpu_data(mask
);