static int
hwloc_aix_get_sth_rset_cpubind(hwloc_topology_t topology, rstype_t what, rsid_t who, hwloc_bitmap_t hwloc_set, int flags __hwloc_attribute_unused, int *boundp)
{
rsethandle_t rset;
unsigned cpu, maxcpus;
int res = -1;
int bound = 0;
rset = rs_alloc(RS_EMPTY);
if (ra_getrset(what, who, 0, rset) == -1)
goto out;
hwloc_bitmap_zero(hwloc_set);
maxcpus = rs_getinfo(rset, R_MAXPROCS, 0);
for (cpu = 0; cpu < maxcpus; cpu++)
if (rs_op(RS_TESTRESOURCE, rset, NULL, R_PROCS, cpu) == 1)
hwloc_bitmap_set(hwloc_set, cpu);
else
bound = 1;
hwloc_bitmap_and(hwloc_set, hwloc_set, hwloc_topology_get_complete_cpuset(topology));
res = 0;
*boundp = bound;
out:
rs_free(rset);
return res;
}
static int
hwloc_aix_prepare_membind(hwloc_topology_t topology, rsethandle_t *rad, hwloc_const_nodeset_t nodeset, int flags __hwloc_attribute_unused)
{
rsethandle_t rset, noderad;
int MCMlevel;
int node;
MCMlevel = rs_getinfo(NULL, R_MCMSDL, 0);
if ((topology->flags & HWLOC_TOPOLOGY_FLAG_WHOLE_SYSTEM))
rset = rs_alloc(RS_ALL);
else
rset = rs_alloc(RS_PARTITION);
*rad = rs_alloc(RS_EMPTY);
noderad = rs_alloc(RS_EMPTY);
hwloc_bitmap_foreach_begin(node, nodeset)
rs_getrad(rset, noderad, MCMlevel, node, 0);
rs_op(RS_UNION, noderad, *rad, 0, 0);
hwloc_bitmap_foreach_end();
rs_free(rset);
rs_free(noderad);
return 0;
}
static int
hwloc_aix_set_sth_cpubind(hwloc_topology_t topology, rstype_t what, rsid_t who, hwloc_const_bitmap_t hwloc_set, int flags __hwloc_attribute_unused)
{
rsethandle_t rad;
int res;
unsigned cpu;
if (flags & HWLOC_CPUBIND_NOMEMBIND) {
errno = ENOSYS;
return -1;
}
/* The resulting binding is always strict */
if (hwloc_bitmap_isequal(hwloc_set, hwloc_topology_get_complete_cpuset(topology))) {
if (ra_detachrset(what, who, 0))
return -1;
return 0;
}
rad = rs_alloc(RS_EMPTY);
hwloc_bitmap_foreach_begin(cpu, hwloc_set)
rs_op(RS_ADDRESOURCE, rad, NULL, R_PROCS, cpu);
hwloc_bitmap_foreach_end();
res = ra_attachrset(what, who, rad, 0);
rs_free(rad);
return res;
}
static int
hwloc_aix_get_sth_membind(hwloc_topology_t topology, rstype_t what, rsid_t who, hwloc_bitmap_t nodeset, hwloc_membind_policy_t *policy, int flags __hwloc_attribute_unused)
{
hwloc_bitmap_t hwloc_set;
rsethandle_t rset;
unsigned cpu, maxcpus;
int res = -1;
int depth, n, i;
depth = hwloc_get_type_depth(topology, HWLOC_OBJ_NUMANODE);
if (depth < 0) {
errno = EXDEV;
return -1;
}
n = hwloc_get_nbobjs_by_depth(topology, depth);
rset = rs_alloc(RS_EMPTY);
if (ra_getrset(what, who, 0, rset) == -1)
goto out;
hwloc_set = hwloc_bitmap_alloc();
maxcpus = rs_getinfo(rset, R_MAXPROCS, 0);
for (cpu = 0; cpu < maxcpus; cpu++)
if (rs_op(RS_TESTRESOURCE, rset, NULL, R_PROCS, cpu) == 1)
hwloc_bitmap_set(hwloc_set, cpu);
hwloc_bitmap_and(hwloc_set, hwloc_set, hwloc_topology_get_complete_cpuset(topology));
hwloc_bitmap_zero(nodeset);
for (i = 0; i < n; i++) {
hwloc_obj_t obj = hwloc_get_obj_by_depth(topology, depth, i);
if (hwloc_bitmap_isincluded(obj->cpuset, hwloc_set))
hwloc_bitmap_set(nodeset, obj->os_index);
}
hwloc_bitmap_free(hwloc_set);
*policy = HWLOC_MEMBIND_BIND;
res = 0;
out:
rs_free(rset);
return res;
}
static int
hwloc_aix_set_sth_cpubind(hwloc_topology_t topology, rstype_t what, rsid_t who, pid_t pid, hwloc_const_bitmap_t hwloc_set, int flags __hwloc_attribute_unused)
{
rsethandle_t rad;
int res;
unsigned cpu;
if (flags & HWLOC_CPUBIND_NOMEMBIND) {
errno = ENOSYS;
return -1;
}
/* The resulting binding is always strict */
if (hwloc_bitmap_isequal(hwloc_set, hwloc_topology_get_complete_cpuset(topology))) {
if (ra_detachrset(what, who, 0))
return -1;
return 0;
}
rad = rs_alloc(RS_EMPTY);
hwloc_bitmap_foreach_begin(cpu, hwloc_set)
rs_op(RS_ADDRESOURCE, rad, NULL, R_PROCS, cpu);
hwloc_bitmap_foreach_end();
res = ra_attachrset(what, who, rad, 0);
if (res < 0 && errno == EPERM) {
/* EPERM may mean that one thread has ben bound with bindprocessor().
* Unbind the entire process (we can't unbind individual threads)
* and try again.
*/
bindprocessor(BINDPROCESS, pid, PROCESSOR_CLASS_ANY);
res = ra_attachrset(what, who, rad, 0);
}
rs_free(rad);
return res;
}
static void
look_rset(int sdl, hwloc_obj_type_t type, struct hwloc_topology *topology, int level)
{
rsethandle_t rset, rad;
int i,maxcpus,j;
int nbnodes;
struct hwloc_obj *obj;
if ((topology->flags & HWLOC_TOPOLOGY_FLAG_WHOLE_SYSTEM))
rset = rs_alloc(RS_ALL);
else
rset = rs_alloc(RS_PARTITION);
rad = rs_alloc(RS_EMPTY);
nbnodes = rs_numrads(rset, sdl, 0);
if (nbnodes == -1) {
perror("rs_numrads");
return;
}
for (i = 0; i < nbnodes; i++) {
if (rs_getrad(rset, rad, sdl, i, 0)) {
fprintf(stderr,"rs_getrad(%d) failed: %s\n", i, strerror(errno));
continue;
}
if (!rs_getinfo(rad, R_NUMPROCS, 0))
continue;
/* It seems logical processors are numbered from 1 here, while the
* bindprocessor functions numbers them from 0... */
obj = hwloc_alloc_setup_object(type, i - (type == HWLOC_OBJ_PU));
obj->cpuset = hwloc_bitmap_alloc();
obj->os_level = sdl;
maxcpus = rs_getinfo(rad, R_MAXPROCS, 0);
for (j = 0; j < maxcpus; j++) {
if (rs_op(RS_TESTRESOURCE, rad, NULL, R_PROCS, j))
hwloc_bitmap_set(obj->cpuset, j);
}
switch(type) {
case HWLOC_OBJ_NODE:
obj->nodeset = hwloc_bitmap_alloc();
hwloc_bitmap_set(obj->nodeset, i);
obj->memory.local_memory = 0; /* TODO: odd, rs_getinfo(rad, R_MEMSIZE, 0) << 10 returns the total memory ... */
obj->memory.page_types_len = 2;
obj->memory.page_types = malloc(2*sizeof(*obj->memory.page_types));
memset(obj->memory.page_types, 0, 2*sizeof(*obj->memory.page_types));
obj->memory.page_types[0].size = hwloc_getpagesize();
#ifdef HAVE__SC_LARGE_PAGESIZE
obj->memory.page_types[1].size = sysconf(_SC_LARGE_PAGESIZE);
#endif
/* TODO: obj->memory.page_types[1].count = rs_getinfo(rset, R_LGPGFREE, 0) / hugepagesize */
break;
case HWLOC_OBJ_CACHE:
obj->attr->cache.size = _system_configuration.L2_cache_size;
obj->attr->cache.associativity = _system_configuration.L2_cache_asc;
obj->attr->cache.linesize = 0; /* TODO: ? */
obj->attr->cache.depth = 2;
obj->attr->cache.type = HWLOC_OBJ_CACHE_UNIFIED; /* FIXME? */
break;
case HWLOC_OBJ_GROUP:
obj->attr->group.depth = level;
break;
case HWLOC_OBJ_CORE:
{
hwloc_obj_t obj2, obj3;
obj2 = hwloc_alloc_setup_object(HWLOC_OBJ_CACHE, i);
obj2->cpuset = hwloc_bitmap_dup(obj->cpuset);
obj2->attr->cache.size = _system_configuration.dcache_size;
obj2->attr->cache.associativity = _system_configuration.dcache_asc;
obj2->attr->cache.linesize = _system_configuration.dcache_line;
obj2->attr->cache.depth = 1;
if (_system_configuration.cache_attrib & (1<<30)) {
/* Unified cache */
obj2->attr->cache.type = HWLOC_OBJ_CACHE_UNIFIED;
hwloc_debug("Adding an L1u cache for core %d\n", i);
hwloc_insert_object_by_cpuset(topology, obj2);
} else {
/* Separate Instruction and Data caches */
obj2->attr->cache.type = HWLOC_OBJ_CACHE_DATA;
hwloc_debug("Adding an L1d cache for core %d\n", i);
hwloc_insert_object_by_cpuset(topology, obj2);
obj3 = hwloc_alloc_setup_object(HWLOC_OBJ_CACHE, i);
obj3->cpuset = hwloc_bitmap_dup(obj->cpuset);
obj3->attr->cache.size = _system_configuration.icache_size;
obj3->attr->cache.associativity = _system_configuration.icache_asc;
obj3->attr->cache.linesize = _system_configuration.icache_line;
obj3->attr->cache.depth = 1;
obj3->attr->cache.type = HWLOC_OBJ_CACHE_INSTRUCTION;
hwloc_debug("Adding an L1i cache for core %d\n", i);
hwloc_insert_object_by_cpuset(topology, obj3);
}
break;
}
default:
break;
}
hwloc_debug_2args_bitmap("%s %d has cpuset %s\n",
hwloc_obj_type_string(type),
i, obj->cpuset);
hwloc_insert_object_by_cpuset(topology, obj);
}
rs_free(rset);
rs_free(rad);
}
static void
look_rset(int sdl, hwloc_obj_type_t type, struct hwloc_topology *topology, int level)
{
rsethandle_t rset, rad;
int i,maxcpus,j;
int nbnodes;
struct hwloc_obj *obj;
if ((topology->flags & HWLOC_TOPOLOGY_FLAG_WHOLE_SYSTEM))
rset = rs_alloc(RS_ALL);
else
rset = rs_alloc(RS_PARTITION);
rad = rs_alloc(RS_EMPTY);
nbnodes = rs_numrads(rset, sdl, 0);
if (nbnodes == -1) {
perror("rs_numrads");
return;
}
for (i = 0; i < nbnodes; i++) {
hwloc_bitmap_t cpuset;
unsigned os_index = (unsigned) -1; /* no os_index except for PU and NUMANODE below */
if (rs_getrad(rset, rad, sdl, i, 0)) {
fprintf(stderr,"rs_getrad(%d) failed: %s\n", i, strerror(errno));
continue;
}
if (!rs_getinfo(rad, R_NUMPROCS, 0))
continue;
maxcpus = rs_getinfo(rad, R_MAXPROCS, 0);
cpuset = hwloc_bitmap_alloc();
for (j = 0; j < maxcpus; j++) {
if (rs_op(RS_TESTRESOURCE, rad, NULL, R_PROCS, j))
hwloc_bitmap_set(cpuset, j);
}
if (type == HWLOC_OBJ_PU) {
os_index = hwloc_bitmap_first(cpuset);
hwloc_debug("Found PU #%u inside node %d for sdl %d\n", os_index, i, sdl);
assert(hwloc_bitmap_weight(cpuset) == 1);
} else if (type == HWLOC_OBJ_NUMANODE) {
/* NUMA node os_index isn't used for binding, just use the rad number to get unique values.
* Note that we'll use that fact in hwloc_aix_prepare_membind(). */
os_index = i;
hwloc_debug("Using os_index #%u for NUMA node inside node %d for sdl %d\n", os_index, i, sdl);
}
obj = hwloc_alloc_setup_object(type, os_index);
obj->cpuset = cpuset;
obj->os_level = sdl;
switch(type) {
case HWLOC_OBJ_NUMANODE:
obj->nodeset = hwloc_bitmap_alloc();
hwloc_bitmap_set(obj->nodeset, i);
obj->memory.local_memory = 0; /* TODO: odd, rs_getinfo(rad, R_MEMSIZE, 0) << 10 returns the total memory ... */
obj->memory.page_types_len = 2;
obj->memory.page_types = malloc(2*sizeof(*obj->memory.page_types));
memset(obj->memory.page_types, 0, 2*sizeof(*obj->memory.page_types));
obj->memory.page_types[0].size = hwloc_getpagesize();
#ifdef HAVE__SC_LARGE_PAGESIZE
obj->memory.page_types[1].size = sysconf(_SC_LARGE_PAGESIZE);
#endif
/* TODO: obj->memory.page_types[1].count = rs_getinfo(rset, R_LGPGFREE, 0) / hugepagesize */
break;
case HWLOC_OBJ_CACHE:
obj->attr->cache.size = _system_configuration.L2_cache_size;
obj->attr->cache.associativity = _system_configuration.L2_cache_asc;
obj->attr->cache.linesize = 0; /* unknown by default */
if (__power_pc())
if (__power_4() || __power_5() || __power_6() || __power_7())
obj->attr->cache.linesize = 128;
obj->attr->cache.depth = 2;
obj->attr->cache.type = HWLOC_OBJ_CACHE_UNIFIED; /* OK for power[4567], unknown for others */
break;
case HWLOC_OBJ_GROUP:
obj->attr->group.depth = level;
break;
case HWLOC_OBJ_CORE:
{
hwloc_obj_t obj2, obj3;
obj2 = hwloc_alloc_setup_object(HWLOC_OBJ_CACHE, i);
obj2->cpuset = hwloc_bitmap_dup(obj->cpuset);
obj2->attr->cache.size = _system_configuration.dcache_size;
obj2->attr->cache.associativity = _system_configuration.dcache_asc;
obj2->attr->cache.linesize = _system_configuration.dcache_line;
obj2->attr->cache.depth = 1;
if (_system_configuration.cache_attrib & (1<<30)) {
/* Unified cache */
obj2->attr->cache.type = HWLOC_OBJ_CACHE_UNIFIED;
hwloc_debug("Adding an L1u cache for core %d\n", i);
hwloc_insert_object_by_cpuset(topology, obj2);
} else {
/* Separate Instruction and Data caches */
obj2->attr->cache.type = HWLOC_OBJ_CACHE_DATA;
hwloc_debug("Adding an L1d cache for core %d\n", i);
hwloc_insert_object_by_cpuset(topology, obj2);
obj3 = hwloc_alloc_setup_object(HWLOC_OBJ_CACHE, i);
obj3->cpuset = hwloc_bitmap_dup(obj->cpuset);
obj3->attr->cache.size = _system_configuration.icache_size;
obj3->attr->cache.associativity = _system_configuration.icache_asc;
obj3->attr->cache.linesize = _system_configuration.icache_line;
obj3->attr->cache.depth = 1;
obj3->attr->cache.type = HWLOC_OBJ_CACHE_INSTRUCTION;
hwloc_debug("Adding an L1i cache for core %d\n", i);
hwloc_insert_object_by_cpuset(topology, obj3);
}
break;
}
default:
break;
}
hwloc_debug_2args_bitmap("%s %d has cpuset %s\n",
hwloc_obj_type_string(type),
i, obj->cpuset);
hwloc_insert_object_by_cpuset(topology, obj);
}
rs_free(rset);
rs_free(rad);
}
static int
hwloc_aix_set_thread_cpubind(hwloc_topology_t topology, hwloc_thread_t pthread, hwloc_const_bitmap_t hwloc_set, int flags)
{
struct __pthrdsinfo info;
int size;
if ((errno = pthread_getthrds_np(&pthread, PTHRDSINFO_QUERY_TID, &info, sizeof(info), NULL, &size)))
return -1;
{
rsid_t who = { .at_tid = info.__pi_tid };
return hwloc_aix_set_sth_cpubind(topology, R_THREAD, who, hwloc_set, flags);
}
}
static int
hwloc_aix_get_thread_cpubind(hwloc_topology_t topology, hwloc_thread_t pthread, hwloc_bitmap_t hwloc_set, int flags)
{
struct __pthrdsinfo info;
int size;
if (pthread_getthrds_np(&pthread, PTHRDSINFO_QUERY_TID, &info, sizeof(info), NULL, &size))
return -1;
{
rsid_t who;
who.at_tid = info.__pi_tid;
return hwloc_aix_get_sth_cpubind(topology, R_THREAD, who, hwloc_set, flags);
}
}
#endif /* HWLOC_HAVE_PTHREAD_GETTHRDS_NP */
#endif /* R_THREAD */
#ifdef P_DEFAULT
static int
hwloc_aix_membind_policy_from_hwloc(uint_t *aix_policy, int policy)
{
switch (policy) {
case HWLOC_MEMBIND_DEFAULT:
case HWLOC_MEMBIND_BIND:
*aix_policy = P_DEFAULT;
break;
case HWLOC_MEMBIND_FIRSTTOUCH:
*aix_policy = P_FIRST_TOUCH;
break;
case HWLOC_MEMBIND_INTERLEAVE:
*aix_policy = P_BALANCED;
break;
default:
errno = ENOSYS;
return -1;
}
return 0;
}
static int
hwloc_aix_prepare_membind(hwloc_topology_t topology, rsethandle_t *rad, hwloc_const_nodeset_t nodeset, int flags __hwloc_attribute_unused)
{
rsethandle_t rset, noderad;
int MCMlevel;
int node;
MCMlevel = rs_getinfo(NULL, R_MCMSDL, 0);
if ((topology->flags & HWLOC_TOPOLOGY_FLAG_WHOLE_SYSTEM))
rset = rs_alloc(RS_ALL);
else
rset = rs_alloc(RS_PARTITION);
*rad = rs_alloc(RS_EMPTY);
noderad = rs_alloc(RS_EMPTY);
hwloc_bitmap_foreach_begin(node, nodeset)
rs_getrad(rset, noderad, MCMlevel, node, 0);
rs_op(RS_UNION, noderad, *rad, 0, 0);
hwloc_bitmap_foreach_end();
rs_free(rset);
rs_free(noderad);
return 0;
}
