mirror of
https://github.com/Dr-Noob/cpufetch.git
synced 2026-03-25 07:50:40 +01:00
[v1.00] Merge latest commits from master branch
This commit is contained in:
Dr-Noob
203
src/x86/apic.c
203
src/x86/apic.c
@@ -4,6 +4,9 @@
|
||||
#elif defined __linux__
|
||||
#define _GNU_SOURCE
|
||||
#include <sched.h>
|
||||
#elif defined __FreeBSD__
|
||||
#include <sys/param.h>
|
||||
#include <sys/cpuset.h>
|
||||
#elif defined __APPLE__
|
||||
#define UNUSED(x) (void)(x)
|
||||
#endif
|
||||
@@ -13,6 +16,7 @@
|
||||
#include <string.h>
|
||||
#include <stdio.h>
|
||||
#include <unistd.h>
|
||||
#include <errno.h>
|
||||
|
||||
#include "apic.h"
|
||||
#include "cpuid_asm.h"
|
||||
@@ -24,7 +28,7 @@
|
||||
*/
|
||||
unsigned char bit_scan_reverse(uint32_t* index, uint64_t mask) {
|
||||
for(uint64_t i = (8 * sizeof(uint64_t)); i > 0; i--) {
|
||||
if((mask & (1LL << (i-1))) != 0) {
|
||||
if((mask & (1ULL << (i-1))) != 0) {
|
||||
*index = (uint64_t) (i-1);
|
||||
break;
|
||||
}
|
||||
@@ -57,7 +61,7 @@ uint32_t get_apic_id(bool x2apic_id) {
|
||||
uint32_t ebx = 0;
|
||||
uint32_t ecx = 0;
|
||||
uint32_t edx = 0;
|
||||
|
||||
|
||||
if(x2apic_id) {
|
||||
eax = 0x0000000B;
|
||||
cpuid(&eax, &ebx, &ecx, &edx);
|
||||
@@ -76,16 +80,25 @@ bool bind_to_cpu(int cpu_id) {
|
||||
HANDLE process = GetCurrentProcess();
|
||||
DWORD_PTR processAffinityMask = 1 << cpu_id;
|
||||
return SetProcessAffinityMask(process, processAffinityMask);
|
||||
#else
|
||||
#elif defined __linux__
|
||||
cpu_set_t currentCPU;
|
||||
CPU_ZERO(¤tCPU);
|
||||
CPU_SET(cpu_id, ¤tCPU);
|
||||
if (sched_setaffinity (0, sizeof(currentCPU), ¤tCPU) == -1) {
|
||||
perror("sched_setaffinity");
|
||||
printWarn("sched_setaffinity: %s", strerror(errno));
|
||||
return false;
|
||||
}
|
||||
return true;
|
||||
#endif
|
||||
#elif defined __FreeBSD__
|
||||
cpuset_t currentCPU;
|
||||
CPU_ZERO(¤tCPU);
|
||||
CPU_SET(cpu_id, ¤tCPU);
|
||||
if(cpuset_setaffinity(CPU_LEVEL_WHICH, CPU_WHICH_TID, -1, sizeof(cpuset_t), ¤tCPU) == -1) {
|
||||
printWarn("cpuset_setaffinity: %s", strerror(errno));
|
||||
return false;
|
||||
}
|
||||
return true;
|
||||
#endif
|
||||
}
|
||||
#endif
|
||||
|
||||
@@ -97,51 +110,51 @@ bool fill_topo_masks_apic(struct topology* topo) {
|
||||
uint32_t core_plus_smt_id_max_cnt;
|
||||
uint32_t core_id_max_cnt;
|
||||
uint32_t smt_id_per_core_max_cnt;
|
||||
|
||||
|
||||
cpuid(&eax, &ebx, &ecx, &edx);
|
||||
|
||||
|
||||
core_plus_smt_id_max_cnt = (ebx >> 16) & 0xFF;
|
||||
|
||||
|
||||
eax = 0x00000004;
|
||||
ecx = 0;
|
||||
cpuid(&eax, &ebx, &ecx, &edx);
|
||||
|
||||
|
||||
core_id_max_cnt = (eax >> 26) + 1;
|
||||
smt_id_per_core_max_cnt = core_plus_smt_id_max_cnt / core_id_max_cnt;
|
||||
|
||||
topo->apic->smt_mask = create_mask(smt_id_per_core_max_cnt, &(topo->apic->smt_mask_width));
|
||||
smt_id_per_core_max_cnt = core_plus_smt_id_max_cnt / core_id_max_cnt;
|
||||
|
||||
topo->apic->smt_mask = create_mask(smt_id_per_core_max_cnt, &(topo->apic->smt_mask_width));
|
||||
topo->apic->core_mask = create_mask(core_id_max_cnt,&(topo->apic->pkg_mask_shift));
|
||||
topo->apic->pkg_mask_shift += topo->apic->smt_mask_width;
|
||||
topo->apic->core_mask <<= topo->apic->smt_mask_width;
|
||||
topo->apic->pkg_mask = (-1) ^ (topo->apic->core_mask | topo->apic->smt_mask);
|
||||
|
||||
|
||||
return true;
|
||||
}
|
||||
|
||||
bool fill_topo_masks_x2apic(struct topology* topo) {
|
||||
int32_t level_type;
|
||||
int32_t level_shift;
|
||||
|
||||
|
||||
int32_t coreplus_smt_mask = 0;
|
||||
bool level2 = false;
|
||||
bool level1 = false;
|
||||
|
||||
|
||||
uint32_t eax = 0;
|
||||
uint32_t ebx = 0;
|
||||
uint32_t ecx = 0;
|
||||
uint32_t edx = 0;
|
||||
uint32_t i = 0;
|
||||
|
||||
|
||||
while(true) {
|
||||
eax = 0x0000000B;
|
||||
ecx = i;
|
||||
cpuid(&eax, &ebx, &ecx, &edx);
|
||||
if(ebx == 0) break;
|
||||
|
||||
|
||||
level_type = (ecx >> 8) & 0xFF;
|
||||
level_shift = eax & 0xFFF;
|
||||
|
||||
switch(level_type) {
|
||||
level_shift = eax & 0xFFF;
|
||||
|
||||
switch(level_type) {
|
||||
case 1: // SMT
|
||||
topo->apic->smt_mask = ~(0xFFFFFFFF << level_shift);
|
||||
topo->apic->smt_mask_width = level_shift;
|
||||
@@ -158,10 +171,10 @@ bool fill_topo_masks_x2apic(struct topology* topo) {
|
||||
printErr("Found invalid level when querying topology: %d", level_type);
|
||||
break;
|
||||
}
|
||||
|
||||
|
||||
i++; // sublevel to query
|
||||
}
|
||||
|
||||
|
||||
if (level1 && level2) {
|
||||
topo->apic->core_mask = coreplus_smt_mask ^ topo->apic->smt_mask;
|
||||
}
|
||||
@@ -182,13 +195,13 @@ bool fill_topo_masks_x2apic(struct topology* topo) {
|
||||
// as the number of cores, but in the case of Xeon Phi KNL it is not
|
||||
uint32_t max_apic_id_size(uint32_t** cache_id_apic, struct topology* topo) {
|
||||
uint32_t max = 0;
|
||||
|
||||
|
||||
for(int i=0; i < topo->cach->max_cache_level; i++) {
|
||||
for(int j=0; j < topo->total_cores; j++) {
|
||||
for(int j=0; j < topo->total_cores; j++) {
|
||||
if(cache_id_apic[j][i] > max) max = cache_id_apic[j][i];
|
||||
}
|
||||
}
|
||||
|
||||
|
||||
max++;
|
||||
if(max > (uint32_t) topo->total_cores) return max;
|
||||
return topo->total_cores;
|
||||
@@ -196,15 +209,15 @@ uint32_t max_apic_id_size(uint32_t** cache_id_apic, struct topology* topo) {
|
||||
|
||||
bool build_topo_from_apic(uint32_t* apic_pkg, uint32_t* apic_smt, uint32_t** cache_id_apic, struct topology* topo) {
|
||||
uint32_t size = max_apic_id_size(cache_id_apic, topo);
|
||||
uint32_t* sockets = malloc(sizeof(uint32_t) * size);
|
||||
uint32_t* smt = malloc(sizeof(uint32_t) * size);
|
||||
uint32_t* apic_id = malloc(sizeof(uint32_t) * size);
|
||||
uint32_t* sockets = emalloc(sizeof(uint32_t) * size);
|
||||
uint32_t* smt = emalloc(sizeof(uint32_t) * size);
|
||||
uint32_t* apic_id = emalloc(sizeof(uint32_t) * size);
|
||||
uint32_t num_caches = 0;
|
||||
|
||||
|
||||
memset(sockets, 0, sizeof(uint32_t) * size);
|
||||
memset(smt, 0, sizeof(uint32_t) * size);
|
||||
memset(apic_id, 0, sizeof(uint32_t) * size);
|
||||
|
||||
memset(apic_id, 0, sizeof(uint32_t) * size);
|
||||
|
||||
// System topology
|
||||
for(int i=0; i < topo->total_cores; i++) {
|
||||
sockets[apic_pkg[i]] = 1;
|
||||
@@ -216,45 +229,43 @@ bool build_topo_from_apic(uint32_t* apic_pkg, uint32_t* apic_smt, uint32_t** cac
|
||||
if(smt[i] != 0)
|
||||
topo->smt_available++;
|
||||
}
|
||||
|
||||
|
||||
topo->logical_cores = topo->total_cores / topo->sockets;
|
||||
topo->physical_cores = topo->logical_cores / topo->smt_available;
|
||||
|
||||
|
||||
// Cache topology
|
||||
for(int i=0; i < topo->cach->max_cache_level; i++) {
|
||||
num_caches = 0;
|
||||
memset(apic_id, 0, sizeof(uint32_t) * size);
|
||||
|
||||
for(int c=0; c < topo->total_cores; c++) {
|
||||
|
||||
for(int c=0; c < topo->total_cores; c++) {
|
||||
apic_id[cache_id_apic[c][i]]++;
|
||||
}
|
||||
for(uint32_t c=0; c < size; c++) {
|
||||
for(uint32_t c=0; c < size; c++) {
|
||||
if(apic_id[c] > 0) num_caches++;
|
||||
}
|
||||
|
||||
printf("[i=%d]: %p\n", i, (void *) topo->cach->cach_arr[i]);
|
||||
topo->cach->cach_arr[i]->num_caches = num_caches;
|
||||
}
|
||||
|
||||
|
||||
free(sockets);
|
||||
free(smt);
|
||||
free(apic_id);
|
||||
|
||||
|
||||
return true;
|
||||
}
|
||||
|
||||
void get_cache_topology_from_apic(struct topology* topo) {
|
||||
void get_cache_topology_from_apic(struct topology* topo) {
|
||||
uint32_t eax = 0x00000004;
|
||||
uint32_t ebx = 0;
|
||||
uint32_t ecx = 0;
|
||||
uint32_t edx = 0;
|
||||
|
||||
for(int i=0; i < topo->cach->max_cache_level; i++) {
|
||||
|
||||
for(int i=0; i < topo->cach->max_cache_level; i++) {
|
||||
eax = 0x00000004;
|
||||
ecx = i;
|
||||
|
||||
|
||||
cpuid(&eax, &ebx, &ecx, &edx);
|
||||
|
||||
|
||||
uint32_t SMTMaxCntPerEachCache = ((eax >> 14) & 0x7FF) + 1;
|
||||
uint32_t dummy;
|
||||
topo->apic->cache_select_mask[i] = create_mask(SMTMaxCntPerEachCache,&dummy);
|
||||
@@ -278,15 +289,15 @@ void add_apic_to_array(uint32_t apic, uint32_t* apic_ids, int n) {
|
||||
if(apic_ids[i] != (uint32_t) -1) last = i+1;
|
||||
i++;
|
||||
}
|
||||
|
||||
if(!found) {
|
||||
|
||||
if(!found) {
|
||||
apic_ids[last] = apic;
|
||||
//printf("Added %d\n", apic);
|
||||
}
|
||||
}
|
||||
|
||||
bool fill_apic_ids(uint32_t* apic_ids, int n, bool x2apic_id) {
|
||||
#ifdef __APPLE__
|
||||
#ifdef __APPLE__
|
||||
// macOS extremely dirty approach...
|
||||
printf("cpufetch is computing APIC IDs, please wait...\n");
|
||||
bool end = false;
|
||||
@@ -295,12 +306,21 @@ bool fill_apic_ids(uint32_t* apic_ids, int n, bool x2apic_id) {
|
||||
|
||||
while(!end) {
|
||||
apic = get_apic_id(x2apic_id);
|
||||
|
||||
|
||||
add_apic_to_array(apic, apic_ids, n);
|
||||
end = apic_array_full(apic_ids, n);
|
||||
end = apic_array_full(apic_ids, n);
|
||||
usleep(1000);
|
||||
}
|
||||
}
|
||||
#else
|
||||
#ifdef __linux__
|
||||
// In Linux we reset the affinity; first we get the original mask
|
||||
cpu_set_t original_mask;
|
||||
if(sched_getaffinity(0, sizeof(original_mask), &original_mask) == -1) {
|
||||
printWarn("sched_getaffinity: %s", strerror(errno));
|
||||
return false;
|
||||
}
|
||||
#endif
|
||||
|
||||
for(int i=0; i < n; i++) {
|
||||
if(!bind_to_cpu(i)) {
|
||||
printErr("Failed binding to CPU %d", i);
|
||||
@@ -308,18 +328,27 @@ bool fill_apic_ids(uint32_t* apic_ids, int n, bool x2apic_id) {
|
||||
}
|
||||
apic_ids[i] = get_apic_id(x2apic_id);
|
||||
}
|
||||
|
||||
#ifdef __linux__
|
||||
// With the original mask previosly retrieved, we reset the affinity
|
||||
if (sched_setaffinity (0, sizeof(original_mask), &original_mask) == -1) {
|
||||
printWarn("sched_setaffinity: %s", strerror(errno));
|
||||
return false;
|
||||
}
|
||||
#endif
|
||||
#endif
|
||||
|
||||
return true;
|
||||
}
|
||||
|
||||
bool get_topology_from_apic(struct cpuInfo* cpu, struct topology* topo) {
|
||||
uint32_t apic_id;
|
||||
uint32_t* apic_ids = malloc(sizeof(uint32_t) * topo->total_cores);
|
||||
uint32_t* apic_pkg = malloc(sizeof(uint32_t) * topo->total_cores);
|
||||
uint32_t* apic_core = malloc(sizeof(uint32_t) * topo->total_cores);
|
||||
uint32_t* apic_smt = malloc(sizeof(uint32_t) * topo->total_cores);
|
||||
uint32_t** cache_smt_id_apic = malloc(sizeof(uint32_t*) * topo->total_cores);
|
||||
uint32_t** cache_id_apic = malloc(sizeof(uint32_t*) * topo->total_cores);
|
||||
bool get_topology_from_apic(struct cpuInfo* cpu, struct topology* topo) {
|
||||
uint32_t apic_id;
|
||||
uint32_t* apic_ids = emalloc(sizeof(uint32_t) * topo->total_cores);
|
||||
uint32_t* apic_pkg = emalloc(sizeof(uint32_t) * topo->total_cores);
|
||||
uint32_t* apic_core = emalloc(sizeof(uint32_t) * topo->total_cores);
|
||||
uint32_t* apic_smt = emalloc(sizeof(uint32_t) * topo->total_cores);
|
||||
uint32_t** cache_smt_id_apic = emalloc(sizeof(uint32_t*) * topo->total_cores);
|
||||
uint32_t** cache_id_apic = emalloc(sizeof(uint32_t*) * topo->total_cores);
|
||||
bool x2apic_id;
|
||||
|
||||
if(cpu->maxLevels >= 0x0000000B) {
|
||||
@@ -336,48 +365,48 @@ bool get_topology_from_apic(struct cpuInfo* cpu, struct topology* topo) {
|
||||
else {
|
||||
x2apic_id = false;
|
||||
}
|
||||
|
||||
|
||||
for(int i=0; i < topo->total_cores; i++) {
|
||||
cache_smt_id_apic[i] = malloc(sizeof(uint32_t) * (topo->cach->max_cache_level));
|
||||
cache_id_apic[i] = malloc(sizeof(uint32_t) * (topo->cach->max_cache_level));
|
||||
cache_smt_id_apic[i] = emalloc(sizeof(uint32_t) * (topo->cach->max_cache_level));
|
||||
cache_id_apic[i] = emalloc(sizeof(uint32_t) * (topo->cach->max_cache_level));
|
||||
}
|
||||
topo->apic->cache_select_mask = malloc(sizeof(uint32_t) * (topo->cach->max_cache_level));
|
||||
topo->apic->cache_id_apic = malloc(sizeof(uint32_t) * (topo->cach->max_cache_level));
|
||||
|
||||
topo->apic->cache_select_mask = emalloc(sizeof(uint32_t) * (topo->cach->max_cache_level));
|
||||
topo->apic->cache_id_apic = emalloc(sizeof(uint32_t) * (topo->cach->max_cache_level));
|
||||
|
||||
if(x2apic_id) {
|
||||
if(!fill_topo_masks_x2apic(topo))
|
||||
return false;
|
||||
}
|
||||
else {
|
||||
if(!fill_topo_masks_apic(topo))
|
||||
return false;
|
||||
return false;
|
||||
}
|
||||
|
||||
get_cache_topology_from_apic(topo);
|
||||
|
||||
|
||||
get_cache_topology_from_apic(topo);
|
||||
|
||||
if(!fill_apic_ids(apic_ids, topo->total_cores, x2apic_id))
|
||||
return false;
|
||||
|
||||
for(int i=0; i < topo->total_cores; i++) {
|
||||
|
||||
for(int i=0; i < topo->total_cores; i++) {
|
||||
apic_id = apic_ids[i];
|
||||
|
||||
|
||||
apic_pkg[i] = (apic_id & topo->apic->pkg_mask) >> topo->apic->pkg_mask_shift;
|
||||
apic_core[i] = (apic_id & topo->apic->core_mask) >> topo->apic->smt_mask_width;
|
||||
apic_smt[i] = apic_id & topo->apic->smt_mask;
|
||||
|
||||
|
||||
for(int c=0; c < topo->cach->max_cache_level; c++) {
|
||||
cache_smt_id_apic[i][c] = apic_id & topo->apic->cache_select_mask[c];
|
||||
cache_id_apic[i][c] = apic_id & (-1 ^ topo->apic->cache_select_mask[c]);
|
||||
}
|
||||
}
|
||||
|
||||
|
||||
/* DEBUG
|
||||
for(int i=0; i < topo->cach->max_cache_level; i++) {
|
||||
printf("[CACH %1d]", i);
|
||||
for(int j=0; j < topo->total_cores; j++)
|
||||
printf("[%03d]", cache_id_apic[j][i]);
|
||||
printf("\n");
|
||||
}
|
||||
}
|
||||
for(int i=0; i < topo->total_cores; i++)
|
||||
printf("[%2d] 0x%.8X\n", i, apic_pkg[i]);
|
||||
printf("\n");
|
||||
@@ -386,16 +415,16 @@ bool get_topology_from_apic(struct cpuInfo* cpu, struct topology* topo) {
|
||||
printf("\n");
|
||||
for(int i=0; i < topo->total_cores; i++)
|
||||
printf("[%2d] 0x%.8X\n", i, apic_smt[i]);*/
|
||||
|
||||
|
||||
|
||||
|
||||
bool ret = build_topo_from_apic(apic_pkg, apic_smt, cache_id_apic, topo);
|
||||
|
||||
|
||||
// Assumption: If we cant get smt_available, we assume it is equal to smt_supported...
|
||||
if (!x2apic_id) {
|
||||
printWarn("Can't read SMT from cpuid (needed level is 0x%.8X, max is 0x%.8X)", 0x0000000B, cpu->maxLevels);
|
||||
printWarn("Can't read SMT from cpuid (needed level is 0x%.8X, max is 0x%.8X)", 0x0000000B, cpu->maxLevels);
|
||||
topo->smt_supported = topo->smt_available;
|
||||
}
|
||||
|
||||
|
||||
free(apic_pkg);
|
||||
free(apic_core);
|
||||
free(apic_smt);
|
||||
@@ -405,17 +434,17 @@ bool get_topology_from_apic(struct cpuInfo* cpu, struct topology* topo) {
|
||||
}
|
||||
free(cache_smt_id_apic);
|
||||
free(cache_id_apic);
|
||||
|
||||
|
||||
return ret;
|
||||
}
|
||||
}
|
||||
|
||||
uint32_t is_smt_enabled_amd(struct topology* topo) {
|
||||
#ifdef __APPLE__
|
||||
UNUSED(topo);
|
||||
return 1;
|
||||
#else
|
||||
#else
|
||||
uint32_t id;
|
||||
|
||||
|
||||
for(int i = 0; i < topo->total_cores; i++) {
|
||||
if(!bind_to_cpu(i)) {
|
||||
printErr("Failed binding to CPU %d", i);
|
||||
@@ -424,7 +453,7 @@ uint32_t is_smt_enabled_amd(struct topology* topo) {
|
||||
id = get_apic_id(false) & 1; // get the last bit
|
||||
if(id == 1) return 2; // We assume there isn't any AMD CPU with more than 2th per core.
|
||||
}
|
||||
|
||||
return 1;
|
||||
#endif
|
||||
|
||||
return 1;
|
||||
#endif
|
||||
}
|
||||
|
||||
Reference in New Issue
Block a user