/**
* sn_cpu_init - initialize per-cpu data areas
* @cpuid: cpuid of the caller
*
* Called during cpu initialization on each cpu as it starts.
* Currently, initializes the per-cpu data area for SNIA.
* Also sets up a few fields in the nodepda. Also known as
* platform_cpu_init() by the ia64 machvec code.
*/
void __init sn_cpu_init(void)
{
int cpuid;
int cpuphyid;
int nasid;
int subnode;
int slice;
int cnode;
int i;
static int wars_have_been_checked;
memset(pda, 0, sizeof(pda));
if (ia64_sn_get_sn_info(0, &sn_hub_info->shub2, &sn_hub_info->nasid_bitmask, &sn_hub_info->nasid_shift,
&sn_system_size, &sn_sharing_domain_size, &sn_partition_id,
&sn_coherency_id, &sn_region_size))
BUG();
sn_hub_info->as_shift = sn_hub_info->nasid_shift - 2;
/*
* The boot cpu makes this call again after platform initialization is
* complete.
*/
if (nodepdaindr[0] == NULL)
return;
cpuid = smp_processor_id();
cpuphyid = get_sapicid();
if (ia64_sn_get_sapic_info(cpuphyid, &nasid, &subnode, &slice))
BUG();
for (i=0; i < MAX_NUMNODES; i++) {
if (nodepdaindr[i]) {
nodepdaindr[i]->phys_cpuid[cpuid].nasid = nasid;
nodepdaindr[i]->phys_cpuid[cpuid].slice = slice;
nodepdaindr[i]->phys_cpuid[cpuid].subnode = subnode;
}
}
cnode = nasid_to_cnodeid(nasid);
pda->p_nodepda = nodepdaindr[cnode];
pda->led_address =
(typeof(pda->led_address)) (LED0 + (slice << LED_CPU_SHIFT));
pda->led_state = LED_ALWAYS_SET;
pda->hb_count = HZ / 2;
pda->hb_state = 0;
pda->idle_flag = 0;
if (cpuid != 0) {
memcpy(pda->cnodeid_to_nasid_table,
pdacpu(0)->cnodeid_to_nasid_table,
sizeof(pda->cnodeid_to_nasid_table));
}
/*
* Check for WARs.
* Only needs to be done once, on BSP.
* Has to be done after loop above, because it uses pda.cnodeid_to_nasid_table[i].
* Has to be done before assignment below.
*/
if (!wars_have_been_checked) {
sn_check_for_wars();
wars_have_been_checked = 1;
}
sn_hub_info->shub_1_1_found = shub_1_1_found;
/*
* Set up addresses of PIO/MEM write status registers.
*/
{
u64 pio1[] = {SH1_PIO_WRITE_STATUS_0, 0, SH1_PIO_WRITE_STATUS_1, 0};
u64 pio2[] = {SH2_PIO_WRITE_STATUS_0, SH2_PIO_WRITE_STATUS_1,
SH2_PIO_WRITE_STATUS_2, SH2_PIO_WRITE_STATUS_3};
u64 *pio;
pio = is_shub1() ? pio1 : pio2;
pda->pio_write_status_addr = (volatile unsigned long *) LOCAL_MMR_ADDR(pio[slice]);
pda->pio_write_status_val = is_shub1() ? SH_PIO_WRITE_STATUS_PENDING_WRITE_COUNT_MASK : 0;
}
/*
* WAR addresses for SHUB 1.x.
*/
if (local_node_data->active_cpu_count++ == 0 && is_shub1()) {
int buddy_nasid;
buddy_nasid =
cnodeid_to_nasid(numa_node_id() ==
num_online_nodes() - 1 ? 0 : numa_node_id() + 1);
pda->pio_shub_war_cam_addr =
(volatile unsigned long *)GLOBAL_MMR_ADDR(nasid,
SH1_PI_CAM_CONTROL);
}
}
/**
* sn_cpu_init - initialize per-cpu data areas
* @cpuid: cpuid of the caller
*
* Called during cpu initialization on each cpu as it starts.
* Currently, initializes the per-cpu data area for SNIA.
* Also sets up a few fields in the nodepda. Also known as
* platform_cpu_init() by the ia64 machvec code.
*/
void __cpuinit sn_cpu_init(void)
{
int cpuid;
int cpuphyid;
int nasid;
int subnode;
int slice;
int cnode;
int i;
static int wars_have_been_checked;
cpuid = smp_processor_id();
if (cpuid == 0 && IS_MEDUSA()) {
if (ia64_sn_is_fake_prom())
sn_prom_type = 2;
else
sn_prom_type = 1;
printk(KERN_INFO "Running on medusa with %s PROM\n",
(sn_prom_type == 1) ? "real" : "fake");
}
memset(pda, 0, sizeof(pda));
if (ia64_sn_get_sn_info(0, &sn_hub_info->shub2,
&sn_hub_info->nasid_bitmask,
&sn_hub_info->nasid_shift,
&sn_system_size, &sn_sharing_domain_size,
&sn_partition_id, &sn_coherency_id,
&sn_region_size))
BUG();
sn_hub_info->as_shift = sn_hub_info->nasid_shift - 2;
/*
* Don't check status. The SAL call is not supported on all PROMs
* but a failure is harmless.
*/
(void) ia64_sn_set_cpu_number(cpuid);
/*
* The boot cpu makes this call again after platform initialization is
* complete.
*/
if (nodepdaindr[0] == NULL)
return;
for (i = 0; i < MAX_PROM_FEATURE_SETS; i++)
if (ia64_sn_get_prom_feature_set(i, &sn_prom_features[i]) != 0)
break;
cpuphyid = get_sapicid();
if (ia64_sn_get_sapic_info(cpuphyid, &nasid, &subnode, &slice))
BUG();
for (i=0; i < MAX_NUMNODES; i++) {
if (nodepdaindr[i]) {
nodepdaindr[i]->phys_cpuid[cpuid].nasid = nasid;
nodepdaindr[i]->phys_cpuid[cpuid].slice = slice;
nodepdaindr[i]->phys_cpuid[cpuid].subnode = subnode;
}
}
cnode = nasid_to_cnodeid(nasid);
sn_nodepda = nodepdaindr[cnode];
pda->led_address =
(typeof(pda->led_address)) (LED0 + (slice << LED_CPU_SHIFT));
pda->led_state = LED_ALWAYS_SET;
pda->hb_count = HZ / 2;
pda->hb_state = 0;
pda->idle_flag = 0;
if (cpuid != 0) {
/* copy cpu 0's sn_cnodeid_to_nasid table to this cpu's */
memcpy(sn_cnodeid_to_nasid,
(&per_cpu(__sn_cnodeid_to_nasid, 0)),
sizeof(__ia64_per_cpu_var(__sn_cnodeid_to_nasid)));
}
/*
* Check for WARs.
* Only needs to be done once, on BSP.
* Has to be done after loop above, because it uses this cpu's
* sn_cnodeid_to_nasid table which was just initialized if this
* isn't cpu 0.
* Has to be done before assignment below.
*/
if (!wars_have_been_checked) {
sn_check_for_wars();
wars_have_been_checked = 1;
}
sn_hub_info->shub_1_1_found = shub_1_1_found;
/*
* Set up addresses of PIO/MEM write status registers.
*/
{
u64 pio1[] = {SH1_PIO_WRITE_STATUS_0, 0, SH1_PIO_WRITE_STATUS_1, 0};
u64 pio2[] = {SH2_PIO_WRITE_STATUS_0, SH2_PIO_WRITE_STATUS_2,
SH2_PIO_WRITE_STATUS_1, SH2_PIO_WRITE_STATUS_3};
u64 *pio;
pio = is_shub1() ? pio1 : pio2;
pda->pio_write_status_addr =
(volatile unsigned long *)GLOBAL_MMR_ADDR(nasid, pio[slice]);
pda->pio_write_status_val = is_shub1() ? SH_PIO_WRITE_STATUS_PENDING_WRITE_COUNT_MASK : 0;
}
/*
* WAR addresses for SHUB 1.x.
*/
if (local_node_data->active_cpu_count++ == 0 && is_shub1()) {
int buddy_nasid;
buddy_nasid =
cnodeid_to_nasid(numa_node_id() ==
num_online_nodes() - 1 ? 0 : numa_node_id() + 1);
pda->pio_shub_war_cam_addr =
(volatile unsigned long *)GLOBAL_MMR_ADDR(nasid,
SH1_PI_CAM_CONTROL);
}
}
static int sn_topology_show(struct seq_file *s, void *d)
{
int sz;
int pt;
int e = 0;
int i;
int j;
const char *slabname;
int ordinal;
cpumask_t cpumask;
char slice;
struct cpuinfo_ia64 *c;
struct sn_hwperf_port_info *ptdata;
struct sn_hwperf_object_info *p;
struct sn_hwperf_object_info *obj = d; /* this object */
struct sn_hwperf_object_info *objs = s->private; /* all objects */
int rack, bay, slot, slab;
u8 shubtype;
u8 system_size;
u8 sharing_size;
u8 partid;
u8 coher;
u8 nasid_shift;
u8 region_size;
u16 nasid_mask;
int nasid_msb;
int pci_bus_ordinal = 0;
if (obj == objs) {
seq_printf(s, "# sn_topology version 2\n");
seq_printf(s, "# objtype ordinal location partition"
" [attribute value [, ...]]\n");
if (ia64_sn_get_sn_info(0,
&shubtype, &nasid_mask, &nasid_shift, &system_size,
&sharing_size, &partid, &coher, ®ion_size))
BUG();
for (nasid_msb=63; nasid_msb > 0; nasid_msb--) {
if (((u64)nasid_mask << nasid_shift) & (1ULL << nasid_msb))
break;
}
seq_printf(s, "partition %u %s local "
"shubtype %s, "
"nasid_mask 0x%016lx, "
"nasid_bits %d:%d, "
"system_size %d, "
"sharing_size %d, "
"coherency_domain %d, "
"region_size %d\n",
partid, system_utsname.nodename,
shubtype ? "shub2" : "shub1",
(u64)nasid_mask << nasid_shift, nasid_msb, nasid_shift,
system_size, sharing_size, coher, region_size);
}
if (SN_HWPERF_FOREIGN(obj)) {
/* private in another partition: not interesting */
return 0;
}
for (i = 0; i < SN_HWPERF_MAXSTRING && obj->name[i]; i++) {
if (obj->name[i] == ' ')
obj->name[i] = '_';
}
slabname = sn_hwperf_get_slabname(obj, objs, &ordinal);
seq_printf(s, "%s %d %s %s asic %s", slabname, ordinal, obj->location,
obj->sn_hwp_this_part ? "local" : "shared", obj->name);
if (!SN_HWPERF_IS_NODE(obj) && !SN_HWPERF_IS_IONODE(obj))
seq_putc(s, '\n');
else {
seq_printf(s, ", nasid 0x%x", cnodeid_to_nasid(ordinal));
for (i=0; i < numionodes; i++) {
seq_printf(s, i ? ":%d" : ", dist %d",
node_distance(ordinal, i));
}
seq_putc(s, '\n');
/*
* CPUs on this node, if any
*/
cpumask = node_to_cpumask(ordinal);
for_each_online_cpu(i) {
if (cpu_isset(i, cpumask)) {
slice = 'a' + cpuid_to_slice(i);
c = cpu_data(i);
seq_printf(s, "cpu %d %s%c local"
" freq %luMHz, arch ia64",
i, obj->location, slice,
c->proc_freq / 1000000);
for_each_online_cpu(j) {
seq_printf(s, j ? ":%d" : ", dist %d",
node_distance(
cpuid_to_cnodeid(i),
cpuid_to_cnodeid(j)));
}
seq_putc(s, '\n');
}
}
/*
* PCI busses attached to this node, if any
*/
if (sn_hwperf_location_to_bpos(obj->location,
&rack, &bay, &slot, &slab)) {
/* export pci bus info */
print_pci_topology(s, obj, &pci_bus_ordinal,
rack, bay, slot, slab);
}
}
if (obj->ports) {
/*
* numalink ports
*/
sz = obj->ports * sizeof(struct sn_hwperf_port_info);
if ((ptdata = vmalloc(sz)) == NULL)
return -ENOMEM;
e = ia64_sn_hwperf_op(sn_hwperf_master_nasid,
SN_HWPERF_ENUM_PORTS, obj->id, sz,
(u64) ptdata, 0, 0, NULL);
if (e != SN_HWPERF_OP_OK)
return -EINVAL;
for (ordinal=0, p=objs; p != obj; p++) {
if (!SN_HWPERF_FOREIGN(p))
ordinal += p->ports;
}
for (pt = 0; pt < obj->ports; pt++) {
for (p = objs, i = 0; i < sn_hwperf_obj_cnt; i++, p++) {
if (ptdata[pt].conn_id == p->id) {
break;
}
}
seq_printf(s, "numalink %d %s-%d",
ordinal+pt, obj->location, ptdata[pt].port);
if (i >= sn_hwperf_obj_cnt) {
/* no connection */
seq_puts(s, " local endpoint disconnected"
", protocol unknown\n");
continue;
}
if (obj->sn_hwp_this_part && p->sn_hwp_this_part)
/* both ends local to this partition */
seq_puts(s, " local");
else if (!obj->sn_hwp_this_part && !p->sn_hwp_this_part)
/* both ends of the link in foreign partiton */
seq_puts(s, " foreign");
else
/* link straddles a partition */
seq_puts(s, " shared");
/*
* Unlikely, but strictly should query the LLP config
* registers because an NL4R can be configured to run
* NL3 protocol, even when not talking to an NL3 router.
* Ditto for node-node.
*/
seq_printf(s, " endpoint %s-%d, protocol %s\n",
p->location, ptdata[pt].conn_port,
(SN_HWPERF_IS_NL3ROUTER(obj) ||
SN_HWPERF_IS_NL3ROUTER(p)) ? "LLP3" : "LLP4");
}
vfree(ptdata);
}
return 0;
}
