Fix compilation in Windows and add support for bind to specific cores. Separate APIC code in other file

This commit is contained in:
Dr-Noob
2020-07-06 01:16:59 +02:00
parent b457c86100
commit 08f79bb914
6 changed files with 289 additions and 251 deletions

View File

@@ -4,12 +4,14 @@ CXXFLAGS=-Wall -Wextra -Werror -fstack-protector-all -pedantic -Wno-unused -std=
SANITY_FLAGS=-Wfloat-equal -Wshadow -Wpointer-arith -Wstrict-overflow=5 -Wformat=2
SRC_DIR=src/
SOURCE=$(SRC_DIR)main.c $(SRC_DIR)cpuid.c $(SRC_DIR)cpuid_asm.c $(SRC_DIR)printer.c $(SRC_DIR)args.c $(SRC_DIR)global.c
HEADERS=$(SRC_DIR)cpuid.h $(SRC_DIR)cpuid_asm.h $(SRC_DIR)printer.h $(SRC_DIR)ascii.h $(SRC_DIR)args.h $(SRC_DIR)global.h
SOURCE=$(SRC_DIR)main.c $(SRC_DIR)cpuid.c $(SRC_DIR)apic.c $(SRC_DIR)cpuid_asm.c $(SRC_DIR)printer.c $(SRC_DIR)args.c $(SRC_DIR)global.c
HEADERS=$(SRC_DIR)cpuid.h $(SRC_DIR)apic.h $(SRC_DIR)cpuid_asm.h $(SRC_DIR)printer.h $(SRC_DIR)ascii.h $(SRC_DIR)args.h $(SRC_DIR)global.h
ifneq ($(OS),Windows_NT)
SOURCE += $(SRC_DIR)udev.c
HEADERS += $(SRC_DIR)udev.h
else
SANITY_FLAGS += -Wno-pedantic-ms-format
endif
OUTPUT=cpufetch

245
src/apic.c Normal file
View File

@@ -0,0 +1,245 @@
#ifdef _WIN32
#include <windows.h>
#else
#define _GNU_SOURCE
#include <sched.h>
#endif
#include <stdlib.h>
#include <stdint.h>
#include <string.h>
#include <stdio.h>
#include "apic.h"
#include "cpuid_asm.h"
#include "global.h"
/*
* bit_scan_reverse and create_mask code taken from:
* https://software.intel.com/content/www/us/en/develop/articles/intel-64-architecture-processor-topology-enumeration.html
*/
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) {
*index = (uint64_t) (i-1);
break;
}
}
return (unsigned char) (mask != 0);
}
uint32_t create_mask(uint32_t num_entries, uint32_t *mask_width) {
uint32_t i;
uint64_t k;
// NearestPo2(numEntries) is the nearest power of 2 integer that is not less than numEntries
// The most significant bit of (numEntries * 2 -1) matches the above definition
k = (uint64_t)(num_entries) * 2 -1;
if (bit_scan_reverse(&i, k) == 0) {
if (mask_width) *mask_width = 0;
return 0;
}
if (mask_width) *mask_width = i;
if (i == 31) return (uint32_t ) -1;
return (1 << i) -1;
}
uint32_t get_apic_id(bool x2apic_id) {
uint32_t eax = 0;
uint32_t ebx = 0;
uint32_t ecx = 0;
uint32_t edx = 0;
if(x2apic_id) {
eax = 0x0000000B;
cpuid(&eax, &ebx, &ecx, &edx);
return edx;
}
else {
eax = 0x00000001;
cpuid(&eax, &ebx, &ecx, &edx);
return (ebx >> 24);
}
}
bool bind_to_cpu(int cpu_id) {
#ifdef _WIN32
HANDLE process = GetCurrentProcess();
DWORD_PTR processAffinityMask = 1 << cpu_id;
return SetProcessAffinityMask(process, processAffinityMask);
#else
cpu_set_t currentCPU;
CPU_ZERO(&currentCPU);
CPU_SET(cpu_id, &currentCPU);
if (sched_setaffinity (0, sizeof(currentCPU), &currentCPU) == -1) {
perror("sched_setaffinity");
return false;
}
return true;
#endif
}
bool fill_topo_masks_apic(struct topology** topo) {
uint32_t eax = 0x00000001;
uint32_t ebx = 0;
uint32_t ecx = 0;
uint32_t edx = 0;
uint32_t core_plus_smt_id_max_cnt;
uint32_t core_id_max_cnt;
uint32_t smt_id_per_core_max_cnt;
uint32_t SMTIDPerCoreMaxCnt;
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));
(*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;
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) {
case 1: // SMT
(*topo)->apic->smt_mask = ~(0xFFFFFFFF << level_shift);
(*topo)->apic->smt_mask_width = level_shift;
(*topo)->smt_supported = ebx & 0xFFFF;
level1 = true;
break;
case 2: // Core
coreplus_smt_mask = ~(0xFFFFFFFF << level_shift);
(*topo)->apic->pkg_mask_shift = level_shift;
(*topo)->apic->pkg_mask = (-1) ^ coreplus_smt_mask;
level2 = true;
break;
default:
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;
}
else if (!level2 && level1) {
(*topo)->apic->core_mask = 0;
(*topo)->apic->pkg_mask_shift = (*topo)->apic->smt_mask_width;
(*topo)->apic->pkg_mask = (-1) ^ (*topo)->apic->smt_mask;
}
else {
printErr("SMT level was not found when querying topology");
return false;
}
return true;
}
bool build_topo_from_apic(uint32_t* apic_pkg, uint32_t* apic_core, uint32_t* apic_smt, struct topology** topo) {
uint32_t sockets[64];
uint32_t smt[64];
memset(sockets, 0, sizeof(uint32_t) * 64);
memset(smt, 0, sizeof(uint32_t) * 64);
for(int i=0; i < (*topo)->total_cores; i++) {
sockets[apic_pkg[i]] = 1;
smt[apic_smt[i]] = 1;
}
for(int i=0; i < 64; i++) {
if(sockets[i] != 0)
(*topo)->sockets++;
if(smt[i] != 0)
(*topo)->smt_available++;
}
(*topo)->logical_cores = (*topo)->total_cores / (*topo)->sockets;
(*topo)->physical_cores = (*topo)->logical_cores / (*topo)->smt_available;
return true;
}
bool get_topology_from_apic(uint32_t cpuid_max_levels, struct topology** topo) {
uint32_t apic_id;
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);
bool x2apic_id = cpuid_max_levels >= 0x0000000B;
if(x2apic_id) {
if(!fill_topo_masks_x2apic(topo))
return false;
}
else {
if(!fill_topo_masks_apic(topo))
return false;
}
for(int i=0; i < (*topo)->total_cores; i++) {
if(!bind_to_cpu(i)) {
printErr("Failed binding to CPU %d", i);
return false;
}
apic_id = get_apic_id(x2apic_id);
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;
}
/* DEBUG
for(int i=0; i < (*topo)->total_cores; i++)
printf("[%2d] 0x%.8X\n", i, apic_pkg[i]);
printf("\n");
for(int i=0; i < (*topo)->total_cores; i++)
printf("[%2d] 0x%.8X\n", i, apic_core[i]);
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_core, apic_smt, topo);
// Assumption: If we cant get smt_available, we assume it is equal to smt_supported...
if(!x2apic_id) (*topo)->smt_supported = (*topo)->smt_available;
return ret;
}

17
src/apic.h Normal file
View File

@@ -0,0 +1,17 @@
#ifndef __APIC__
#define __APIC__
#include <stdbool.h>
#include "cpuid.h"
struct apic {
uint32_t pkg_mask;
uint32_t pkg_mask_shift;
uint32_t core_mask;
uint32_t smt_mask_width;
uint32_t smt_mask;
};
bool get_topology_from_apic(uint32_t cpuid_max_levels, struct topology** topo);
#endif

View File

@@ -1,10 +1,8 @@
#ifdef _WIN32
#include <windows.h>
#else
#define _GNU_SOURCE
#include <sched.h>
#include <unistd.h>
#include "udev.h"
#include <unistd.h>
#endif
#include <stdio.h>
@@ -16,6 +14,7 @@
#include "cpuid.h"
#include "cpuid_asm.h"
#include "global.h"
#include "apic.h"
#define VENDOR_INTEL_STRING "GenuineIntel"
#define VENDOR_AMD_STRING "AuthenticAMD"
@@ -75,24 +74,6 @@ struct frequency {
int64_t max;
};
struct apic {
uint32_t pkg_mask;
uint32_t pkg_mask_shift;
uint32_t core_mask;
uint32_t smt_mask_width;
uint32_t smt_mask;
};
struct topology {
int64_t total_cores;
uint32_t physical_cores;
uint32_t logical_cores;
uint32_t smt_available; // Number of SMT that is currently enabled
uint32_t smt_supported; // Number of SMT that CPU supports (equal to smt_available if SMT is enabled)
uint32_t sockets;
struct apic* apic;
};
void init_cpu_info(struct cpuInfo* cpu) {
cpu->AVX = false;
cpu->AVX2 = false;
@@ -281,230 +262,6 @@ struct cpuInfo* get_cpu_info() {
return cpu;
}
bool bind_to_cpu(int cpu_id) {
#ifdef _WIN32 // TODO
return false;
#else
cpu_set_t currentCPU;
CPU_ZERO(&currentCPU);
CPU_SET(cpu_id, &currentCPU);
if (sched_setaffinity (0, sizeof(currentCPU), &currentCPU) == -1) {
perror("sched_setaffinity");
return false;
}
return true;
#endif
}
uint32_t get_apic_id(bool x2apic_id) {
uint32_t eax = 0;
uint32_t ebx = 0;
uint32_t ecx = 0;
uint32_t edx = 0;
if(x2apic_id) {
eax = 0x0000000B;
cpuid(&eax, &ebx, &ecx, &edx);
return edx;
}
else {
eax = 0x00000001;
cpuid(&eax, &ebx, &ecx, &edx);
return (ebx >> 24);
}
}
bool fill_topo_masks_x2apic(struct topology** topo) {
int32_t level_type;
int32_t level_shift;
int32_t coreplus_smt_mask;
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) {
case 1: // SMT
(*topo)->apic->smt_mask = ~(0xFFFFFFFF << level_shift);
(*topo)->apic->smt_mask_width = level_shift;
(*topo)->smt_supported = ebx & 0xFFFF;
level1 = true;
break;
case 2: // Core
coreplus_smt_mask = ~(0xFFFFFFFF << level_shift);
(*topo)->apic->pkg_mask_shift = level_shift;
(*topo)->apic->pkg_mask = (-1) ^ coreplus_smt_mask;
level2 = true;
break;
default:
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;
}
else if (!level2 && level1) {
(*topo)->apic->core_mask = 0;
(*topo)->apic->pkg_mask_shift = (*topo)->apic->smt_mask_width;
(*topo)->apic->pkg_mask = (-1) ^ (*topo)->apic->smt_mask;
}
else {
printErr("SMT level was not found when querying topology");
return false;
}
return true;
}
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) {
*index = (uint64_t) (i-1);
break;
}
}
return (unsigned char)( mask != 0);
}
uint32_t create_mask(uint32_t num_entries, uint32_t *mask_width) {
uint32_t i;
uint64_t k;
// NearestPo2(numEntries) is the nearest power of 2 integer that is not less than numEntries
// The most significant bit of (numEntries * 2 -1) matches the above definition
k = (uint64_t)(num_entries) * 2 -1;
if (bit_scan_reverse(&i, k) == 0) {
if (mask_width) *mask_width = 0;
return 0;
}
if (mask_width) *mask_width = i;
if (i == 31) return (uint32_t ) -1;
return (1 << i) -1;
}
bool fill_topo_masks_apic(struct topology** topo) {
uint32_t eax = 0x00000001;
uint32_t ebx = 0;
uint32_t ecx = 0;
uint32_t edx = 0;
uint32_t core_plus_smt_id_max_cnt;
uint32_t core_id_max_cnt;
uint32_t smt_id_per_core_max_cnt;
uint32_t SMTIDPerCoreMaxCnt;
cpuid(&eax, &ebx, &ecx, &edx);
core_plus_smt_id_max_cnt = (ebx >> 16) & 0xFF; // MAL?
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));
(*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 build_topo_from_apic(uint32_t* apic_pkg, uint32_t* apic_core, uint32_t* apic_smt, struct topology** topo) {
uint32_t sockets[64];
uint32_t smt[64];
memset(sockets, 0, sizeof(uint32_t) * 64);
memset(smt, 0, sizeof(uint32_t) * 64);
for(int i=0; i < (*topo)->total_cores; i++) {
sockets[apic_pkg[i]] = 1;
smt[apic_smt[i]] = 1;
}
for(int i=0; i < 64; i++) {
if(sockets[i] != 0)
(*topo)->sockets++;
if(smt[i] != 0)
(*topo)->smt_available++;
}
(*topo)->logical_cores = (*topo)->total_cores / (*topo)->sockets;
(*topo)->physical_cores = (*topo)->logical_cores / (*topo)->smt_available;
return true;
}
bool get_topology_from_apic(struct cpuInfo* cpu, struct topology** topo) {
uint32_t apic_id;
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);
bool x2apic_id = cpu->maxLevels >= 0x0000000B;
if(x2apic_id) {
if(!fill_topo_masks_x2apic(topo))
return false;
}
else {
if(!fill_topo_masks_apic(topo))
return false;
}
for(int i=0; i < (*topo)->total_cores; i++) {
if(!bind_to_cpu(i)) {
printErr("Failed binding to CPU %d", i);
return false;
}
apic_id = get_apic_id(x2apic_id);
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;
}
/* DEBUG
for(int i=0; i < (*topo)->total_cores; i++)
printf("[%2d] 0x%.8X\n", i, apic_pkg[i]);
printf("\n");
for(int i=0; i < (*topo)->total_cores; i++)
printf("[%2d] 0x%.8X\n", i, apic_core[i]);
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_core, apic_smt, topo);
// Assumption: If we cant get smt_available, we assume it is equal to smt_supported...
if(!x2apic_id) (*topo)->smt_supported = (*topo)->smt_available;
return ret;
}
// Main reference: https://software.intel.com/content/www/us/en/develop/articles/intel-64-architecture-processor-topology-enumeration.html
// Very interesting resource: https://wiki.osdev.org/Detecting_CPU_Topology_(80x86)
struct topology* get_topology_info(struct cpuInfo* cpu) {
@@ -535,7 +292,7 @@ struct topology* get_topology_info(struct cpuInfo* cpu) {
switch(cpu->cpu_vendor) {
case VENDOR_INTEL:
if (cpu->maxLevels >= 0x00000004) {
get_topology_from_apic(cpu, &topo);
get_topology_from_apic(cpu->maxLevels, &topo);
}
else {
printErr("Can't read topology information from cpuid (needed level is 0x%.8X, max is 0x%.8X)", 0x00000001, cpu->maxLevels);

View File

@@ -13,7 +13,16 @@
struct cpuInfo;
struct frequency;
struct cache;
struct topology;
struct topology {
int64_t total_cores;
uint32_t physical_cores;
uint32_t logical_cores;
uint32_t smt_available; // Number of SMT that is currently enabled
uint32_t smt_supported; // Number of SMT that CPU supports (equal to smt_available if SMT is enabled)
uint32_t sockets;
struct apic* apic;
};
typedef int32_t VENDOR;

View File

@@ -147,6 +147,13 @@ struct ascii* set_ascii(VENDOR cpuVendor, STYLE style, struct colors* cs) {
}
art->ascii_chars[1] = '#';
#ifdef _WIN32
strcpy(art->color1_ascii,COL_NONE);
strcpy(art->color2_ascii,COL_NONE);
strcpy(art->color1_text,COL_NONE);
strcpy(art->color2_text,COL_NONE);
art->reset[0] = '\0';
#else
switch(style) {
case STYLE_EMPTY:
#ifdef _WIN32
@@ -200,6 +207,7 @@ struct ascii* set_ascii(VENDOR cpuVendor, STYLE style, struct colors* cs) {
printBug("Found invalid style (%d)",style);
return NULL;
}
#endif
char tmp[NUMBER_OF_LINES*LINE_SIZE];
if(cpuVendor == VENDOR_INTEL) strcpy(tmp, INTEL_ASCII);