linux/arch/sparc64/kernel/mdesc.c

917 lines
19 KiB
C
Raw Normal View History

/* mdesc.c: Sun4V machine description handling.
*
* Copyright (C) 2007, 2008 David S. Miller <davem@davemloft.net>
*/
#include <linux/kernel.h>
#include <linux/types.h>
#include <linux/lmb.h>
#include <linux/log2.h>
#include <linux/list.h>
#include <linux/slab.h>
#include <linux/mm.h>
#include <linux/miscdevice.h>
#include <asm/hypervisor.h>
#include <asm/mdesc.h>
#include <asm/prom.h>
#include <asm/oplib.h>
#include <asm/smp.h>
/* Unlike the OBP device tree, the machine description is a full-on
* DAG. An arbitrary number of ARCs are possible from one
* node to other nodes and thus we can't use the OBP device_node
* data structure to represent these nodes inside of the kernel.
*
* Actually, it isn't even a DAG, because there are back pointers
* which create cycles in the graph.
*
* mdesc_hdr and mdesc_elem describe the layout of the data structure
* we get from the Hypervisor.
*/
struct mdesc_hdr {
u32 version; /* Transport version */
u32 node_sz; /* node block size */
u32 name_sz; /* name block size */
u32 data_sz; /* data block size */
} __attribute__((aligned(16)));
struct mdesc_elem {
u8 tag;
#define MD_LIST_END 0x00
#define MD_NODE 0x4e
#define MD_NODE_END 0x45
#define MD_NOOP 0x20
#define MD_PROP_ARC 0x61
#define MD_PROP_VAL 0x76
#define MD_PROP_STR 0x73
#define MD_PROP_DATA 0x64
u8 name_len;
u16 resv;
u32 name_offset;
union {
struct {
u32 data_len;
u32 data_offset;
} data;
u64 val;
} d;
};
struct mdesc_mem_ops {
struct mdesc_handle *(*alloc)(unsigned int mdesc_size);
void (*free)(struct mdesc_handle *handle);
};
struct mdesc_handle {
struct list_head list;
struct mdesc_mem_ops *mops;
void *self_base;
atomic_t refcnt;
unsigned int handle_size;
struct mdesc_hdr mdesc;
};
static void mdesc_handle_init(struct mdesc_handle *hp,
unsigned int handle_size,
void *base)
{
BUG_ON(((unsigned long)&hp->mdesc) & (16UL - 1));
memset(hp, 0, handle_size);
INIT_LIST_HEAD(&hp->list);
hp->self_base = base;
atomic_set(&hp->refcnt, 1);
hp->handle_size = handle_size;
}
static struct mdesc_handle * __init mdesc_lmb_alloc(unsigned int mdesc_size)
{
unsigned int handle_size, alloc_size;
struct mdesc_handle *hp;
unsigned long paddr;
handle_size = (sizeof(struct mdesc_handle) -
sizeof(struct mdesc_hdr) +
mdesc_size);
alloc_size = PAGE_ALIGN(handle_size);
paddr = lmb_alloc(alloc_size, PAGE_SIZE);
hp = NULL;
if (paddr) {
hp = __va(paddr);
mdesc_handle_init(hp, handle_size, hp);
}
return hp;
}
static void mdesc_lmb_free(struct mdesc_handle *hp)
{
unsigned int alloc_size, handle_size = hp->handle_size;
unsigned long start, end;
BUG_ON(atomic_read(&hp->refcnt) != 0);
BUG_ON(!list_empty(&hp->list));
alloc_size = PAGE_ALIGN(handle_size);
start = (unsigned long) hp;
end = start + alloc_size;
while (start < end) {
struct page *p;
p = virt_to_page(start);
ClearPageReserved(p);
__free_page(p);
start += PAGE_SIZE;
}
}
static struct mdesc_mem_ops lmb_mdesc_ops = {
.alloc = mdesc_lmb_alloc,
.free = mdesc_lmb_free,
};
static struct mdesc_handle *mdesc_kmalloc(unsigned int mdesc_size)
{
unsigned int handle_size;
void *base;
handle_size = (sizeof(struct mdesc_handle) -
sizeof(struct mdesc_hdr) +
mdesc_size);
base = kmalloc(handle_size + 15, GFP_KERNEL | __GFP_NOFAIL);
if (base) {
struct mdesc_handle *hp;
unsigned long addr;
addr = (unsigned long)base;
addr = (addr + 15UL) & ~15UL;
hp = (struct mdesc_handle *) addr;
mdesc_handle_init(hp, handle_size, base);
return hp;
}
return NULL;
}
static void mdesc_kfree(struct mdesc_handle *hp)
{
BUG_ON(atomic_read(&hp->refcnt) != 0);
BUG_ON(!list_empty(&hp->list));
kfree(hp->self_base);
}
static struct mdesc_mem_ops kmalloc_mdesc_memops = {
.alloc = mdesc_kmalloc,
.free = mdesc_kfree,
};
static struct mdesc_handle *mdesc_alloc(unsigned int mdesc_size,
struct mdesc_mem_ops *mops)
{
struct mdesc_handle *hp = mops->alloc(mdesc_size);
if (hp)
hp->mops = mops;
return hp;
}
static void mdesc_free(struct mdesc_handle *hp)
{
hp->mops->free(hp);
}
static struct mdesc_handle *cur_mdesc;
static LIST_HEAD(mdesc_zombie_list);
static DEFINE_SPINLOCK(mdesc_lock);
struct mdesc_handle *mdesc_grab(void)
{
struct mdesc_handle *hp;
unsigned long flags;
spin_lock_irqsave(&mdesc_lock, flags);
hp = cur_mdesc;
if (hp)
atomic_inc(&hp->refcnt);
spin_unlock_irqrestore(&mdesc_lock, flags);
return hp;
}
EXPORT_SYMBOL(mdesc_grab);
void mdesc_release(struct mdesc_handle *hp)
{
unsigned long flags;
spin_lock_irqsave(&mdesc_lock, flags);
if (atomic_dec_and_test(&hp->refcnt)) {
list_del_init(&hp->list);
hp->mops->free(hp);
}
spin_unlock_irqrestore(&mdesc_lock, flags);
}
EXPORT_SYMBOL(mdesc_release);
static DEFINE_MUTEX(mdesc_mutex);
static struct mdesc_notifier_client *client_list;
void mdesc_register_notifier(struct mdesc_notifier_client *client)
{
u64 node;
mutex_lock(&mdesc_mutex);
client->next = client_list;
client_list = client;
mdesc_for_each_node_by_name(cur_mdesc, node, client->node_name)
client->add(cur_mdesc, node);
mutex_unlock(&mdesc_mutex);
}
static const u64 *parent_cfg_handle(struct mdesc_handle *hp, u64 node)
{
const u64 *id;
u64 a;
id = NULL;
mdesc_for_each_arc(a, hp, node, MDESC_ARC_TYPE_BACK) {
u64 target;
target = mdesc_arc_target(hp, a);
id = mdesc_get_property(hp, target,
"cfg-handle", NULL);
if (id)
break;
}
return id;
}
/* Run 'func' on nodes which are in A but not in B. */
static void invoke_on_missing(const char *name,
struct mdesc_handle *a,
struct mdesc_handle *b,
void (*func)(struct mdesc_handle *, u64))
{
u64 node;
mdesc_for_each_node_by_name(a, node, name) {
int found = 0, is_vdc_port = 0;
const char *name_prop;
const u64 *id;
u64 fnode;
name_prop = mdesc_get_property(a, node, "name", NULL);
if (name_prop && !strcmp(name_prop, "vdc-port")) {
is_vdc_port = 1;
id = parent_cfg_handle(a, node);
} else
id = mdesc_get_property(a, node, "id", NULL);
if (!id) {
printk(KERN_ERR "MD: Cannot find ID for %s node.\n",
(name_prop ? name_prop : name));
continue;
}
mdesc_for_each_node_by_name(b, fnode, name) {
const u64 *fid;
if (is_vdc_port) {
name_prop = mdesc_get_property(b, fnode,
"name", NULL);
if (!name_prop ||
strcmp(name_prop, "vdc-port"))
continue;
fid = parent_cfg_handle(b, fnode);
if (!fid) {
printk(KERN_ERR "MD: Cannot find ID "
"for vdc-port node.\n");
continue;
}
} else
fid = mdesc_get_property(b, fnode,
"id", NULL);
if (*id == *fid) {
found = 1;
break;
}
}
if (!found)
func(a, node);
}
}
static void notify_one(struct mdesc_notifier_client *p,
struct mdesc_handle *old_hp,
struct mdesc_handle *new_hp)
{
invoke_on_missing(p->node_name, old_hp, new_hp, p->remove);
invoke_on_missing(p->node_name, new_hp, old_hp, p->add);
}
static void mdesc_notify_clients(struct mdesc_handle *old_hp,
struct mdesc_handle *new_hp)
{
struct mdesc_notifier_client *p = client_list;
while (p) {
notify_one(p, old_hp, new_hp);
p = p->next;
}
}
void mdesc_update(void)
{
unsigned long len, real_len, status;
struct mdesc_handle *hp, *orig_hp;
unsigned long flags;
mutex_lock(&mdesc_mutex);
(void) sun4v_mach_desc(0UL, 0UL, &len);
hp = mdesc_alloc(len, &kmalloc_mdesc_memops);
if (!hp) {
printk(KERN_ERR "MD: mdesc alloc fails\n");
goto out;
}
status = sun4v_mach_desc(__pa(&hp->mdesc), len, &real_len);
if (status != HV_EOK || real_len > len) {
printk(KERN_ERR "MD: mdesc reread fails with %lu\n",
status);
atomic_dec(&hp->refcnt);
mdesc_free(hp);
goto out;
}
spin_lock_irqsave(&mdesc_lock, flags);
orig_hp = cur_mdesc;
cur_mdesc = hp;
spin_unlock_irqrestore(&mdesc_lock, flags);
mdesc_notify_clients(orig_hp, hp);
spin_lock_irqsave(&mdesc_lock, flags);
if (atomic_dec_and_test(&orig_hp->refcnt))
mdesc_free(orig_hp);
else
list_add(&orig_hp->list, &mdesc_zombie_list);
spin_unlock_irqrestore(&mdesc_lock, flags);
out:
mutex_unlock(&mdesc_mutex);
}
static struct mdesc_elem *node_block(struct mdesc_hdr *mdesc)
{
return (struct mdesc_elem *) (mdesc + 1);
}
static void *name_block(struct mdesc_hdr *mdesc)
{
return ((void *) node_block(mdesc)) + mdesc->node_sz;
}
static void *data_block(struct mdesc_hdr *mdesc)
{
return ((void *) name_block(mdesc)) + mdesc->name_sz;
}
u64 mdesc_node_by_name(struct mdesc_handle *hp,
u64 from_node, const char *name)
{
struct mdesc_elem *ep = node_block(&hp->mdesc);
const char *names = name_block(&hp->mdesc);
u64 last_node = hp->mdesc.node_sz / 16;
u64 ret;
if (from_node == MDESC_NODE_NULL) {
ret = from_node = 0;
} else if (from_node >= last_node) {
return MDESC_NODE_NULL;
} else {
ret = ep[from_node].d.val;
}
while (ret < last_node) {
if (ep[ret].tag != MD_NODE)
return MDESC_NODE_NULL;
if (!strcmp(names + ep[ret].name_offset, name))
break;
ret = ep[ret].d.val;
}
if (ret >= last_node)
ret = MDESC_NODE_NULL;
return ret;
}
EXPORT_SYMBOL(mdesc_node_by_name);
const void *mdesc_get_property(struct mdesc_handle *hp, u64 node,
const char *name, int *lenp)
{
const char *names = name_block(&hp->mdesc);
u64 last_node = hp->mdesc.node_sz / 16;
void *data = data_block(&hp->mdesc);
struct mdesc_elem *ep;
if (node == MDESC_NODE_NULL || node >= last_node)
return NULL;
ep = node_block(&hp->mdesc) + node;
ep++;
for (; ep->tag != MD_NODE_END; ep++) {
void *val = NULL;
int len = 0;
switch (ep->tag) {
case MD_PROP_VAL:
val = &ep->d.val;
len = 8;
break;
case MD_PROP_STR:
case MD_PROP_DATA:
val = data + ep->d.data.data_offset;
len = ep->d.data.data_len;
break;
default:
break;
}
if (!val)
continue;
if (!strcmp(names + ep->name_offset, name)) {
if (lenp)
*lenp = len;
return val;
}
}
return NULL;
}
EXPORT_SYMBOL(mdesc_get_property);
u64 mdesc_next_arc(struct mdesc_handle *hp, u64 from, const char *arc_type)
{
struct mdesc_elem *ep, *base = node_block(&hp->mdesc);
const char *names = name_block(&hp->mdesc);
u64 last_node = hp->mdesc.node_sz / 16;
if (from == MDESC_NODE_NULL || from >= last_node)
return MDESC_NODE_NULL;
ep = base + from;
ep++;
for (; ep->tag != MD_NODE_END; ep++) {
if (ep->tag != MD_PROP_ARC)
continue;
if (strcmp(names + ep->name_offset, arc_type))
continue;
return ep - base;
}
return MDESC_NODE_NULL;
}
EXPORT_SYMBOL(mdesc_next_arc);
u64 mdesc_arc_target(struct mdesc_handle *hp, u64 arc)
{
struct mdesc_elem *ep, *base = node_block(&hp->mdesc);
ep = base + arc;
return ep->d.val;
}
EXPORT_SYMBOL(mdesc_arc_target);
const char *mdesc_node_name(struct mdesc_handle *hp, u64 node)
{
struct mdesc_elem *ep, *base = node_block(&hp->mdesc);
const char *names = name_block(&hp->mdesc);
u64 last_node = hp->mdesc.node_sz / 16;
if (node == MDESC_NODE_NULL || node >= last_node)
return NULL;
ep = base + node;
if (ep->tag != MD_NODE)
return NULL;
return names + ep->name_offset;
}
EXPORT_SYMBOL(mdesc_node_name);
static void __init report_platform_properties(void)
{
struct mdesc_handle *hp = mdesc_grab();
u64 pn = mdesc_node_by_name(hp, MDESC_NODE_NULL, "platform");
const char *s;
const u64 *v;
if (pn == MDESC_NODE_NULL) {
prom_printf("No platform node in machine-description.\n");
prom_halt();
}
s = mdesc_get_property(hp, pn, "banner-name", NULL);
printk("PLATFORM: banner-name [%s]\n", s);
s = mdesc_get_property(hp, pn, "name", NULL);
printk("PLATFORM: name [%s]\n", s);
v = mdesc_get_property(hp, pn, "hostid", NULL);
if (v)
printk("PLATFORM: hostid [%08lx]\n", *v);
v = mdesc_get_property(hp, pn, "serial#", NULL);
if (v)
printk("PLATFORM: serial# [%08lx]\n", *v);
v = mdesc_get_property(hp, pn, "stick-frequency", NULL);
printk("PLATFORM: stick-frequency [%08lx]\n", *v);
v = mdesc_get_property(hp, pn, "mac-address", NULL);
if (v)
printk("PLATFORM: mac-address [%lx]\n", *v);
v = mdesc_get_property(hp, pn, "watchdog-resolution", NULL);
if (v)
printk("PLATFORM: watchdog-resolution [%lu ms]\n", *v);
v = mdesc_get_property(hp, pn, "watchdog-max-timeout", NULL);
if (v)
printk("PLATFORM: watchdog-max-timeout [%lu ms]\n", *v);
v = mdesc_get_property(hp, pn, "max-cpus", NULL);
if (v)
printk("PLATFORM: max-cpus [%lu]\n", *v);
[SPARC64]: Initial LDOM cpu hotplug support. Only adding cpus is supports at the moment, removal will come next. When new cpus are configured, the machine description is updated. When we get the configure request we pass in a cpu mask of to-be-added cpus to the mdesc CPU node parser so it only fetches information for those cpus. That code also proceeds to update the SMT/multi-core scheduling bitmaps. cpu_up() does all the work and we return the status back over the DS channel. CPUs via dr-cpu need to be booted straight out of the hypervisor, and this requires: 1) A new trampoline mechanism. CPUs are booted straight out of the hypervisor with MMU disabled and running in physical addresses with no mappings installed in the TLB. The new hvtramp.S code sets up the critical cpu state, installs the locked TLB mappings for the kernel, and turns the MMU on. It then proceeds to follow the logic of the existing trampoline.S SMP cpu bringup code. 2) All calls into OBP have to be disallowed when domaining is enabled. Since cpus boot straight into the kernel from the hypervisor, OBP has no state about that cpu and therefore cannot handle being invoked on that cpu. Luckily it's only a handful of interfaces which can be called after the OBP device tree is obtained. For example, rebooting, halting, powering-off, and setting options node variables. CPU removal support will require some infrastructure changes here. Namely we'll have to process the requests via a true kernel thread instead of in a workqueue. workqueues run on a per-cpu thread, but when unconfiguring we might need to force the thread to execute on another cpu if the current cpu is the one being removed. Removal of a cpu also causes the kernel to destroy that cpu's workqueue running thread. Another issue on removal is that we may have interrupts still pointing to the cpu-to-be-removed. So new code will be needed to walk the active INO list and retarget those cpus as-needed. Signed-off-by: David S. Miller <davem@davemloft.net>
2007-07-14 07:03:42 +08:00
#ifdef CONFIG_SMP
{
int max_cpu, i;
if (v) {
max_cpu = *v;
if (max_cpu > NR_CPUS)
max_cpu = NR_CPUS;
} else {
max_cpu = NR_CPUS;
}
for (i = 0; i < max_cpu; i++)
cpu_set(i, cpu_possible_map);
}
#endif
mdesc_release(hp);
}
[SPARC64]: Initial LDOM cpu hotplug support. Only adding cpus is supports at the moment, removal will come next. When new cpus are configured, the machine description is updated. When we get the configure request we pass in a cpu mask of to-be-added cpus to the mdesc CPU node parser so it only fetches information for those cpus. That code also proceeds to update the SMT/multi-core scheduling bitmaps. cpu_up() does all the work and we return the status back over the DS channel. CPUs via dr-cpu need to be booted straight out of the hypervisor, and this requires: 1) A new trampoline mechanism. CPUs are booted straight out of the hypervisor with MMU disabled and running in physical addresses with no mappings installed in the TLB. The new hvtramp.S code sets up the critical cpu state, installs the locked TLB mappings for the kernel, and turns the MMU on. It then proceeds to follow the logic of the existing trampoline.S SMP cpu bringup code. 2) All calls into OBP have to be disallowed when domaining is enabled. Since cpus boot straight into the kernel from the hypervisor, OBP has no state about that cpu and therefore cannot handle being invoked on that cpu. Luckily it's only a handful of interfaces which can be called after the OBP device tree is obtained. For example, rebooting, halting, powering-off, and setting options node variables. CPU removal support will require some infrastructure changes here. Namely we'll have to process the requests via a true kernel thread instead of in a workqueue. workqueues run on a per-cpu thread, but when unconfiguring we might need to force the thread to execute on another cpu if the current cpu is the one being removed. Removal of a cpu also causes the kernel to destroy that cpu's workqueue running thread. Another issue on removal is that we may have interrupts still pointing to the cpu-to-be-removed. So new code will be needed to walk the active INO list and retarget those cpus as-needed. Signed-off-by: David S. Miller <davem@davemloft.net>
2007-07-14 07:03:42 +08:00
static void __devinit fill_in_one_cache(cpuinfo_sparc *c,
struct mdesc_handle *hp,
u64 mp)
{
const u64 *level = mdesc_get_property(hp, mp, "level", NULL);
const u64 *size = mdesc_get_property(hp, mp, "size", NULL);
const u64 *line_size = mdesc_get_property(hp, mp, "line-size", NULL);
const char *type;
int type_len;
type = mdesc_get_property(hp, mp, "type", &type_len);
switch (*level) {
case 1:
if (of_find_in_proplist(type, "instn", type_len)) {
c->icache_size = *size;
c->icache_line_size = *line_size;
} else if (of_find_in_proplist(type, "data", type_len)) {
c->dcache_size = *size;
c->dcache_line_size = *line_size;
}
break;
case 2:
c->ecache_size = *size;
c->ecache_line_size = *line_size;
break;
default:
break;
}
if (*level == 1) {
u64 a;
mdesc_for_each_arc(a, hp, mp, MDESC_ARC_TYPE_FWD) {
u64 target = mdesc_arc_target(hp, a);
const char *name = mdesc_node_name(hp, target);
if (!strcmp(name, "cache"))
fill_in_one_cache(c, hp, target);
}
}
}
[SPARC64]: Initial LDOM cpu hotplug support. Only adding cpus is supports at the moment, removal will come next. When new cpus are configured, the machine description is updated. When we get the configure request we pass in a cpu mask of to-be-added cpus to the mdesc CPU node parser so it only fetches information for those cpus. That code also proceeds to update the SMT/multi-core scheduling bitmaps. cpu_up() does all the work and we return the status back over the DS channel. CPUs via dr-cpu need to be booted straight out of the hypervisor, and this requires: 1) A new trampoline mechanism. CPUs are booted straight out of the hypervisor with MMU disabled and running in physical addresses with no mappings installed in the TLB. The new hvtramp.S code sets up the critical cpu state, installs the locked TLB mappings for the kernel, and turns the MMU on. It then proceeds to follow the logic of the existing trampoline.S SMP cpu bringup code. 2) All calls into OBP have to be disallowed when domaining is enabled. Since cpus boot straight into the kernel from the hypervisor, OBP has no state about that cpu and therefore cannot handle being invoked on that cpu. Luckily it's only a handful of interfaces which can be called after the OBP device tree is obtained. For example, rebooting, halting, powering-off, and setting options node variables. CPU removal support will require some infrastructure changes here. Namely we'll have to process the requests via a true kernel thread instead of in a workqueue. workqueues run on a per-cpu thread, but when unconfiguring we might need to force the thread to execute on another cpu if the current cpu is the one being removed. Removal of a cpu also causes the kernel to destroy that cpu's workqueue running thread. Another issue on removal is that we may have interrupts still pointing to the cpu-to-be-removed. So new code will be needed to walk the active INO list and retarget those cpus as-needed. Signed-off-by: David S. Miller <davem@davemloft.net>
2007-07-14 07:03:42 +08:00
static void __devinit mark_core_ids(struct mdesc_handle *hp, u64 mp,
int core_id)
{
u64 a;
mdesc_for_each_arc(a, hp, mp, MDESC_ARC_TYPE_BACK) {
u64 t = mdesc_arc_target(hp, a);
const char *name;
const u64 *id;
name = mdesc_node_name(hp, t);
if (!strcmp(name, "cpu")) {
id = mdesc_get_property(hp, t, "id", NULL);
if (*id < NR_CPUS)
cpu_data(*id).core_id = core_id;
} else {
u64 j;
mdesc_for_each_arc(j, hp, t, MDESC_ARC_TYPE_BACK) {
u64 n = mdesc_arc_target(hp, j);
const char *n_name;
n_name = mdesc_node_name(hp, n);
if (strcmp(n_name, "cpu"))
continue;
id = mdesc_get_property(hp, n, "id", NULL);
if (*id < NR_CPUS)
cpu_data(*id).core_id = core_id;
}
}
}
}
[SPARC64]: Initial LDOM cpu hotplug support. Only adding cpus is supports at the moment, removal will come next. When new cpus are configured, the machine description is updated. When we get the configure request we pass in a cpu mask of to-be-added cpus to the mdesc CPU node parser so it only fetches information for those cpus. That code also proceeds to update the SMT/multi-core scheduling bitmaps. cpu_up() does all the work and we return the status back over the DS channel. CPUs via dr-cpu need to be booted straight out of the hypervisor, and this requires: 1) A new trampoline mechanism. CPUs are booted straight out of the hypervisor with MMU disabled and running in physical addresses with no mappings installed in the TLB. The new hvtramp.S code sets up the critical cpu state, installs the locked TLB mappings for the kernel, and turns the MMU on. It then proceeds to follow the logic of the existing trampoline.S SMP cpu bringup code. 2) All calls into OBP have to be disallowed when domaining is enabled. Since cpus boot straight into the kernel from the hypervisor, OBP has no state about that cpu and therefore cannot handle being invoked on that cpu. Luckily it's only a handful of interfaces which can be called after the OBP device tree is obtained. For example, rebooting, halting, powering-off, and setting options node variables. CPU removal support will require some infrastructure changes here. Namely we'll have to process the requests via a true kernel thread instead of in a workqueue. workqueues run on a per-cpu thread, but when unconfiguring we might need to force the thread to execute on another cpu if the current cpu is the one being removed. Removal of a cpu also causes the kernel to destroy that cpu's workqueue running thread. Another issue on removal is that we may have interrupts still pointing to the cpu-to-be-removed. So new code will be needed to walk the active INO list and retarget those cpus as-needed. Signed-off-by: David S. Miller <davem@davemloft.net>
2007-07-14 07:03:42 +08:00
static void __devinit set_core_ids(struct mdesc_handle *hp)
{
int idx;
u64 mp;
idx = 1;
mdesc_for_each_node_by_name(hp, mp, "cache") {
const u64 *level;
const char *type;
int len;
level = mdesc_get_property(hp, mp, "level", NULL);
if (*level != 1)
continue;
type = mdesc_get_property(hp, mp, "type", &len);
if (!of_find_in_proplist(type, "instn", len))
continue;
mark_core_ids(hp, mp, idx);
idx++;
}
}
[SPARC64]: Initial LDOM cpu hotplug support. Only adding cpus is supports at the moment, removal will come next. When new cpus are configured, the machine description is updated. When we get the configure request we pass in a cpu mask of to-be-added cpus to the mdesc CPU node parser so it only fetches information for those cpus. That code also proceeds to update the SMT/multi-core scheduling bitmaps. cpu_up() does all the work and we return the status back over the DS channel. CPUs via dr-cpu need to be booted straight out of the hypervisor, and this requires: 1) A new trampoline mechanism. CPUs are booted straight out of the hypervisor with MMU disabled and running in physical addresses with no mappings installed in the TLB. The new hvtramp.S code sets up the critical cpu state, installs the locked TLB mappings for the kernel, and turns the MMU on. It then proceeds to follow the logic of the existing trampoline.S SMP cpu bringup code. 2) All calls into OBP have to be disallowed when domaining is enabled. Since cpus boot straight into the kernel from the hypervisor, OBP has no state about that cpu and therefore cannot handle being invoked on that cpu. Luckily it's only a handful of interfaces which can be called after the OBP device tree is obtained. For example, rebooting, halting, powering-off, and setting options node variables. CPU removal support will require some infrastructure changes here. Namely we'll have to process the requests via a true kernel thread instead of in a workqueue. workqueues run on a per-cpu thread, but when unconfiguring we might need to force the thread to execute on another cpu if the current cpu is the one being removed. Removal of a cpu also causes the kernel to destroy that cpu's workqueue running thread. Another issue on removal is that we may have interrupts still pointing to the cpu-to-be-removed. So new code will be needed to walk the active INO list and retarget those cpus as-needed. Signed-off-by: David S. Miller <davem@davemloft.net>
2007-07-14 07:03:42 +08:00
static void __devinit mark_proc_ids(struct mdesc_handle *hp, u64 mp,
int proc_id)
{
u64 a;
mdesc_for_each_arc(a, hp, mp, MDESC_ARC_TYPE_BACK) {
u64 t = mdesc_arc_target(hp, a);
const char *name;
const u64 *id;
name = mdesc_node_name(hp, t);
if (strcmp(name, "cpu"))
continue;
id = mdesc_get_property(hp, t, "id", NULL);
if (*id < NR_CPUS)
cpu_data(*id).proc_id = proc_id;
}
}
[SPARC64]: Initial LDOM cpu hotplug support. Only adding cpus is supports at the moment, removal will come next. When new cpus are configured, the machine description is updated. When we get the configure request we pass in a cpu mask of to-be-added cpus to the mdesc CPU node parser so it only fetches information for those cpus. That code also proceeds to update the SMT/multi-core scheduling bitmaps. cpu_up() does all the work and we return the status back over the DS channel. CPUs via dr-cpu need to be booted straight out of the hypervisor, and this requires: 1) A new trampoline mechanism. CPUs are booted straight out of the hypervisor with MMU disabled and running in physical addresses with no mappings installed in the TLB. The new hvtramp.S code sets up the critical cpu state, installs the locked TLB mappings for the kernel, and turns the MMU on. It then proceeds to follow the logic of the existing trampoline.S SMP cpu bringup code. 2) All calls into OBP have to be disallowed when domaining is enabled. Since cpus boot straight into the kernel from the hypervisor, OBP has no state about that cpu and therefore cannot handle being invoked on that cpu. Luckily it's only a handful of interfaces which can be called after the OBP device tree is obtained. For example, rebooting, halting, powering-off, and setting options node variables. CPU removal support will require some infrastructure changes here. Namely we'll have to process the requests via a true kernel thread instead of in a workqueue. workqueues run on a per-cpu thread, but when unconfiguring we might need to force the thread to execute on another cpu if the current cpu is the one being removed. Removal of a cpu also causes the kernel to destroy that cpu's workqueue running thread. Another issue on removal is that we may have interrupts still pointing to the cpu-to-be-removed. So new code will be needed to walk the active INO list and retarget those cpus as-needed. Signed-off-by: David S. Miller <davem@davemloft.net>
2007-07-14 07:03:42 +08:00
static void __devinit __set_proc_ids(struct mdesc_handle *hp,
const char *exec_unit_name)
{
int idx;
u64 mp;
idx = 0;
mdesc_for_each_node_by_name(hp, mp, exec_unit_name) {
const char *type;
int len;
type = mdesc_get_property(hp, mp, "type", &len);
if (!of_find_in_proplist(type, "int", len) &&
!of_find_in_proplist(type, "integer", len))
continue;
mark_proc_ids(hp, mp, idx);
idx++;
}
}
[SPARC64]: Initial LDOM cpu hotplug support. Only adding cpus is supports at the moment, removal will come next. When new cpus are configured, the machine description is updated. When we get the configure request we pass in a cpu mask of to-be-added cpus to the mdesc CPU node parser so it only fetches information for those cpus. That code also proceeds to update the SMT/multi-core scheduling bitmaps. cpu_up() does all the work and we return the status back over the DS channel. CPUs via dr-cpu need to be booted straight out of the hypervisor, and this requires: 1) A new trampoline mechanism. CPUs are booted straight out of the hypervisor with MMU disabled and running in physical addresses with no mappings installed in the TLB. The new hvtramp.S code sets up the critical cpu state, installs the locked TLB mappings for the kernel, and turns the MMU on. It then proceeds to follow the logic of the existing trampoline.S SMP cpu bringup code. 2) All calls into OBP have to be disallowed when domaining is enabled. Since cpus boot straight into the kernel from the hypervisor, OBP has no state about that cpu and therefore cannot handle being invoked on that cpu. Luckily it's only a handful of interfaces which can be called after the OBP device tree is obtained. For example, rebooting, halting, powering-off, and setting options node variables. CPU removal support will require some infrastructure changes here. Namely we'll have to process the requests via a true kernel thread instead of in a workqueue. workqueues run on a per-cpu thread, but when unconfiguring we might need to force the thread to execute on another cpu if the current cpu is the one being removed. Removal of a cpu also causes the kernel to destroy that cpu's workqueue running thread. Another issue on removal is that we may have interrupts still pointing to the cpu-to-be-removed. So new code will be needed to walk the active INO list and retarget those cpus as-needed. Signed-off-by: David S. Miller <davem@davemloft.net>
2007-07-14 07:03:42 +08:00
static void __devinit set_proc_ids(struct mdesc_handle *hp)
{
__set_proc_ids(hp, "exec_unit");
__set_proc_ids(hp, "exec-unit");
}
[SPARC64]: Initial LDOM cpu hotplug support. Only adding cpus is supports at the moment, removal will come next. When new cpus are configured, the machine description is updated. When we get the configure request we pass in a cpu mask of to-be-added cpus to the mdesc CPU node parser so it only fetches information for those cpus. That code also proceeds to update the SMT/multi-core scheduling bitmaps. cpu_up() does all the work and we return the status back over the DS channel. CPUs via dr-cpu need to be booted straight out of the hypervisor, and this requires: 1) A new trampoline mechanism. CPUs are booted straight out of the hypervisor with MMU disabled and running in physical addresses with no mappings installed in the TLB. The new hvtramp.S code sets up the critical cpu state, installs the locked TLB mappings for the kernel, and turns the MMU on. It then proceeds to follow the logic of the existing trampoline.S SMP cpu bringup code. 2) All calls into OBP have to be disallowed when domaining is enabled. Since cpus boot straight into the kernel from the hypervisor, OBP has no state about that cpu and therefore cannot handle being invoked on that cpu. Luckily it's only a handful of interfaces which can be called after the OBP device tree is obtained. For example, rebooting, halting, powering-off, and setting options node variables. CPU removal support will require some infrastructure changes here. Namely we'll have to process the requests via a true kernel thread instead of in a workqueue. workqueues run on a per-cpu thread, but when unconfiguring we might need to force the thread to execute on another cpu if the current cpu is the one being removed. Removal of a cpu also causes the kernel to destroy that cpu's workqueue running thread. Another issue on removal is that we may have interrupts still pointing to the cpu-to-be-removed. So new code will be needed to walk the active INO list and retarget those cpus as-needed. Signed-off-by: David S. Miller <davem@davemloft.net>
2007-07-14 07:03:42 +08:00
static void __devinit get_one_mondo_bits(const u64 *p, unsigned int *mask,
unsigned char def)
{
u64 val;
if (!p)
goto use_default;
val = *p;
if (!val || val >= 64)
goto use_default;
*mask = ((1U << val) * 64U) - 1U;
return;
use_default:
*mask = ((1U << def) * 64U) - 1U;
}
[SPARC64]: Initial LDOM cpu hotplug support. Only adding cpus is supports at the moment, removal will come next. When new cpus are configured, the machine description is updated. When we get the configure request we pass in a cpu mask of to-be-added cpus to the mdesc CPU node parser so it only fetches information for those cpus. That code also proceeds to update the SMT/multi-core scheduling bitmaps. cpu_up() does all the work and we return the status back over the DS channel. CPUs via dr-cpu need to be booted straight out of the hypervisor, and this requires: 1) A new trampoline mechanism. CPUs are booted straight out of the hypervisor with MMU disabled and running in physical addresses with no mappings installed in the TLB. The new hvtramp.S code sets up the critical cpu state, installs the locked TLB mappings for the kernel, and turns the MMU on. It then proceeds to follow the logic of the existing trampoline.S SMP cpu bringup code. 2) All calls into OBP have to be disallowed when domaining is enabled. Since cpus boot straight into the kernel from the hypervisor, OBP has no state about that cpu and therefore cannot handle being invoked on that cpu. Luckily it's only a handful of interfaces which can be called after the OBP device tree is obtained. For example, rebooting, halting, powering-off, and setting options node variables. CPU removal support will require some infrastructure changes here. Namely we'll have to process the requests via a true kernel thread instead of in a workqueue. workqueues run on a per-cpu thread, but when unconfiguring we might need to force the thread to execute on another cpu if the current cpu is the one being removed. Removal of a cpu also causes the kernel to destroy that cpu's workqueue running thread. Another issue on removal is that we may have interrupts still pointing to the cpu-to-be-removed. So new code will be needed to walk the active INO list and retarget those cpus as-needed. Signed-off-by: David S. Miller <davem@davemloft.net>
2007-07-14 07:03:42 +08:00
static void __devinit get_mondo_data(struct mdesc_handle *hp, u64 mp,
struct trap_per_cpu *tb)
{
const u64 *val;
val = mdesc_get_property(hp, mp, "q-cpu-mondo-#bits", NULL);
get_one_mondo_bits(val, &tb->cpu_mondo_qmask, 7);
val = mdesc_get_property(hp, mp, "q-dev-mondo-#bits", NULL);
get_one_mondo_bits(val, &tb->dev_mondo_qmask, 7);
val = mdesc_get_property(hp, mp, "q-resumable-#bits", NULL);
get_one_mondo_bits(val, &tb->resum_qmask, 6);
val = mdesc_get_property(hp, mp, "q-nonresumable-#bits", NULL);
get_one_mondo_bits(val, &tb->nonresum_qmask, 2);
}
void __cpuinit mdesc_fill_in_cpu_data(cpumask_t mask)
{
struct mdesc_handle *hp = mdesc_grab();
u64 mp;
ncpus_probed = 0;
mdesc_for_each_node_by_name(hp, mp, "cpu") {
const u64 *id = mdesc_get_property(hp, mp, "id", NULL);
const u64 *cfreq = mdesc_get_property(hp, mp, "clock-frequency", NULL);
struct trap_per_cpu *tb;
cpuinfo_sparc *c;
int cpuid;
u64 a;
ncpus_probed++;
cpuid = *id;
#ifdef CONFIG_SMP
if (cpuid >= NR_CPUS) {
printk(KERN_WARNING "Ignoring CPU %d which is "
">= NR_CPUS (%d)\n",
cpuid, NR_CPUS);
continue;
}
[SPARC64]: Initial LDOM cpu hotplug support. Only adding cpus is supports at the moment, removal will come next. When new cpus are configured, the machine description is updated. When we get the configure request we pass in a cpu mask of to-be-added cpus to the mdesc CPU node parser so it only fetches information for those cpus. That code also proceeds to update the SMT/multi-core scheduling bitmaps. cpu_up() does all the work and we return the status back over the DS channel. CPUs via dr-cpu need to be booted straight out of the hypervisor, and this requires: 1) A new trampoline mechanism. CPUs are booted straight out of the hypervisor with MMU disabled and running in physical addresses with no mappings installed in the TLB. The new hvtramp.S code sets up the critical cpu state, installs the locked TLB mappings for the kernel, and turns the MMU on. It then proceeds to follow the logic of the existing trampoline.S SMP cpu bringup code. 2) All calls into OBP have to be disallowed when domaining is enabled. Since cpus boot straight into the kernel from the hypervisor, OBP has no state about that cpu and therefore cannot handle being invoked on that cpu. Luckily it's only a handful of interfaces which can be called after the OBP device tree is obtained. For example, rebooting, halting, powering-off, and setting options node variables. CPU removal support will require some infrastructure changes here. Namely we'll have to process the requests via a true kernel thread instead of in a workqueue. workqueues run on a per-cpu thread, but when unconfiguring we might need to force the thread to execute on another cpu if the current cpu is the one being removed. Removal of a cpu also causes the kernel to destroy that cpu's workqueue running thread. Another issue on removal is that we may have interrupts still pointing to the cpu-to-be-removed. So new code will be needed to walk the active INO list and retarget those cpus as-needed. Signed-off-by: David S. Miller <davem@davemloft.net>
2007-07-14 07:03:42 +08:00
if (!cpu_isset(cpuid, mask))
continue;
#else
/* On uniprocessor we only want the values for the
* real physical cpu the kernel booted onto, however
* cpu_data() only has one entry at index 0.
*/
if (cpuid != real_hard_smp_processor_id())
continue;
cpuid = 0;
#endif
c = &cpu_data(cpuid);
c->clock_tick = *cfreq;
tb = &trap_block[cpuid];
get_mondo_data(hp, mp, tb);
mdesc_for_each_arc(a, hp, mp, MDESC_ARC_TYPE_FWD) {
u64 j, t = mdesc_arc_target(hp, a);
const char *t_name;
t_name = mdesc_node_name(hp, t);
if (!strcmp(t_name, "cache")) {
fill_in_one_cache(c, hp, t);
continue;
}
mdesc_for_each_arc(j, hp, t, MDESC_ARC_TYPE_FWD) {
u64 n = mdesc_arc_target(hp, j);
const char *n_name;
n_name = mdesc_node_name(hp, n);
if (!strcmp(n_name, "cache"))
fill_in_one_cache(c, hp, n);
}
}
#ifdef CONFIG_SMP
cpu_set(cpuid, cpu_present_map);
#endif
c->core_id = 0;
c->proc_id = -1;
}
#ifdef CONFIG_SMP
sparc64_multi_core = 1;
#endif
set_core_ids(hp);
set_proc_ids(hp);
smp_fill_in_sib_core_maps();
mdesc_release(hp);
}
static ssize_t mdesc_read(struct file *file, char __user *buf,
size_t len, loff_t *offp)
{
struct mdesc_handle *hp = mdesc_grab();
int err;
if (!hp)
return -ENODEV;
err = hp->handle_size;
if (len < hp->handle_size)
err = -EMSGSIZE;
else if (copy_to_user(buf, &hp->mdesc, hp->handle_size))
err = -EFAULT;
mdesc_release(hp);
return err;
}
static const struct file_operations mdesc_fops = {
.read = mdesc_read,
.owner = THIS_MODULE,
};
static struct miscdevice mdesc_misc = {
.minor = MISC_DYNAMIC_MINOR,
.name = "mdesc",
.fops = &mdesc_fops,
};
static int __init mdesc_misc_init(void)
{
return misc_register(&mdesc_misc);
}
__initcall(mdesc_misc_init);
void __init sun4v_mdesc_init(void)
{
struct mdesc_handle *hp;
unsigned long len, real_len, status;
[SPARC64]: Initial LDOM cpu hotplug support. Only adding cpus is supports at the moment, removal will come next. When new cpus are configured, the machine description is updated. When we get the configure request we pass in a cpu mask of to-be-added cpus to the mdesc CPU node parser so it only fetches information for those cpus. That code also proceeds to update the SMT/multi-core scheduling bitmaps. cpu_up() does all the work and we return the status back over the DS channel. CPUs via dr-cpu need to be booted straight out of the hypervisor, and this requires: 1) A new trampoline mechanism. CPUs are booted straight out of the hypervisor with MMU disabled and running in physical addresses with no mappings installed in the TLB. The new hvtramp.S code sets up the critical cpu state, installs the locked TLB mappings for the kernel, and turns the MMU on. It then proceeds to follow the logic of the existing trampoline.S SMP cpu bringup code. 2) All calls into OBP have to be disallowed when domaining is enabled. Since cpus boot straight into the kernel from the hypervisor, OBP has no state about that cpu and therefore cannot handle being invoked on that cpu. Luckily it's only a handful of interfaces which can be called after the OBP device tree is obtained. For example, rebooting, halting, powering-off, and setting options node variables. CPU removal support will require some infrastructure changes here. Namely we'll have to process the requests via a true kernel thread instead of in a workqueue. workqueues run on a per-cpu thread, but when unconfiguring we might need to force the thread to execute on another cpu if the current cpu is the one being removed. Removal of a cpu also causes the kernel to destroy that cpu's workqueue running thread. Another issue on removal is that we may have interrupts still pointing to the cpu-to-be-removed. So new code will be needed to walk the active INO list and retarget those cpus as-needed. Signed-off-by: David S. Miller <davem@davemloft.net>
2007-07-14 07:03:42 +08:00
cpumask_t mask;
(void) sun4v_mach_desc(0UL, 0UL, &len);
printk("MDESC: Size is %lu bytes.\n", len);
hp = mdesc_alloc(len, &lmb_mdesc_ops);
if (hp == NULL) {
prom_printf("MDESC: alloc of %lu bytes failed.\n", len);
prom_halt();
}
status = sun4v_mach_desc(__pa(&hp->mdesc), len, &real_len);
if (status != HV_EOK || real_len > len) {
prom_printf("sun4v_mach_desc fails, err(%lu), "
"len(%lu), real_len(%lu)\n",
status, len, real_len);
mdesc_free(hp);
prom_halt();
}
cur_mdesc = hp;
report_platform_properties();
[SPARC64]: Initial LDOM cpu hotplug support. Only adding cpus is supports at the moment, removal will come next. When new cpus are configured, the machine description is updated. When we get the configure request we pass in a cpu mask of to-be-added cpus to the mdesc CPU node parser so it only fetches information for those cpus. That code also proceeds to update the SMT/multi-core scheduling bitmaps. cpu_up() does all the work and we return the status back over the DS channel. CPUs via dr-cpu need to be booted straight out of the hypervisor, and this requires: 1) A new trampoline mechanism. CPUs are booted straight out of the hypervisor with MMU disabled and running in physical addresses with no mappings installed in the TLB. The new hvtramp.S code sets up the critical cpu state, installs the locked TLB mappings for the kernel, and turns the MMU on. It then proceeds to follow the logic of the existing trampoline.S SMP cpu bringup code. 2) All calls into OBP have to be disallowed when domaining is enabled. Since cpus boot straight into the kernel from the hypervisor, OBP has no state about that cpu and therefore cannot handle being invoked on that cpu. Luckily it's only a handful of interfaces which can be called after the OBP device tree is obtained. For example, rebooting, halting, powering-off, and setting options node variables. CPU removal support will require some infrastructure changes here. Namely we'll have to process the requests via a true kernel thread instead of in a workqueue. workqueues run on a per-cpu thread, but when unconfiguring we might need to force the thread to execute on another cpu if the current cpu is the one being removed. Removal of a cpu also causes the kernel to destroy that cpu's workqueue running thread. Another issue on removal is that we may have interrupts still pointing to the cpu-to-be-removed. So new code will be needed to walk the active INO list and retarget those cpus as-needed. Signed-off-by: David S. Miller <davem@davemloft.net>
2007-07-14 07:03:42 +08:00
cpus_setall(mask);
mdesc_fill_in_cpu_data(mask);
}