void
mono_hwcap_arch_init (void)
{
#if defined(__linux__) && defined(HAVE_SYS_AUXV_H)
unsigned long hwcap;
unsigned long platform;
if ((hwcap = getauxval(AT_HWCAP))) {
/* PPC_FEATURE_ICACHE_SNOOP */
if (hwcap & 0x00002000)
mono_hwcap_ppc_has_icache_snoop = TRUE;
/* PPC_FEATURE_POWER4, PPC_FEATURE_POWER5, PPC_FEATURE_POWER5_PLUS,
PPC_FEATURE_CELL_BE, PPC_FEATURE_PA6T, PPC_FEATURE_ARCH_2_05 */
if (hwcap & (0x00080000 | 0x00040000 | 0x00020000 | 0x00010000 | 0x00000800 | 0x00001000))
mono_hwcap_ppc_is_isa_2x = TRUE;
/* PPC_FEATURE_POWER4, PPC_FEATURE_POWER5, PPC_FEATURE_POWER5_PLUS,
PPC_FEATURE_CELL_BE, PPC_FEATURE_PA6T, PPC_FEATURE_ARCH_2_05 */
if (hwcap & (0x00080000 | 0x00040000 | 0x00020000 | 0x00010000 | 0x00000800 | 0x00001000))
mono_hwcap_ppc_is_isa_2x = TRUE;
/* PPC_FEATURE_POWER5, PPC_FEATURE_POWER5_PLUS,
PPC_FEATURE_CELL_BE, PPC_FEATURE_PA6T, PPC_FEATURE_ARCH_2_05 */
if (hwcap & (0x00040000 | 0x00020000 | 0x00010000 | 0x00000800 | 0x00001000))
mono_hwcap_ppc_is_isa_2_03 = TRUE;
/* PPC_FEATURE_64 */
if (hwcap & 0x40000000)
mono_hwcap_ppc_is_isa_64 = TRUE;
/* PPC_FEATURE_POWER6_EXT */
if (hwcap & 0x00000200)
mono_hwcap_ppc_has_move_fpr_gpr = TRUE;
}
if ((platform = getauxval(AT_PLATFORM))) {
const char *str = (const char *) platform;
if (!strcmp (str, "ppc970") || (!strncmp (str, "power", 5) && str [5] >= '4' && str [5] <= '7'))
mono_hwcap_ppc_has_multiple_ls_units = TRUE;
}
#elif defined(_AIX)
/* Compatible platforms for Mono (V7R1, 6.1.9) require at least P4. */
mono_hwcap_ppc_is_isa_64 = TRUE;
mono_hwcap_ppc_is_isa_2x = TRUE;
if (__power_5_andup()) {
mono_hwcap_ppc_is_isa_2_03 = TRUE;
mono_hwcap_ppc_has_icache_snoop = TRUE;
}
/* not on POWER8 */
if (__power_4() || __power_5() || __power_6() || __power_7())
mono_hwcap_ppc_has_multiple_ls_units = TRUE;
/*
* This instruction is only available in POWER6 "raw mode" and unlikely
* to work; I couldn't get it to work on the POWER6s I tried.
*/
/*
if (__power_6())
mono_hwcap_ppc_has_move_fpr_gpr = TRUE;
*/
#endif
}
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);
}
