[v0.8][ARM] Building support in ARM

This commit is contained in:
Dr-Noob
2020-11-05 09:28:41 +01:00
parent 5cc9038f3d
commit 1fad4fd10b
25 changed files with 442 additions and 172 deletions

352
src/x86/apic.c Normal file
View File

@@ -0,0 +1,352 @@
#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 "../common/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 = 0;
uint64_t k = 0;
// 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;
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 = 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) {
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;
}
// Not a very elegant solution. The width should always be as long
// 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++) {
if(cache_id_apic[j][i] > max) max = cache_id_apic[j][i];
}
}
max++;
if(max > topo->total_cores) return max;
return topo->total_cores;
}
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 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);
// System topology
for(int i=0; i < topo->total_cores; i++) {
sockets[apic_pkg[i]] = 1;
smt[apic_smt[i]] = 1;
}
for(int i=0; i < topo->total_cores; 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;
// 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++) {
apic_id[cache_id_apic[c][i]]++;
}
for(uint32_t c=0; c < size; c++) {
if(apic_id[c] > 0) num_caches++;
}
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) {
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++) {
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);
}
}
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);
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 x2apic_id = cpu->maxLevels >= 0x0000000B;
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));
}
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));
if(x2apic_id) {
if(!fill_topo_masks_x2apic(topo))
return false;
}
else {
if(!fill_topo_masks_apic(topo))
return false;
}
get_cache_topology_from_apic(topo);
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;
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");
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_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);
topo->smt_supported = topo->smt_available;
}
free(apic_pkg);
free(apic_core);
free(apic_smt);
for(int i=0; i < topo->total_cores; i++) {
free(cache_smt_id_apic[i]);
free(cache_id_apic[i]);
}
free(cache_smt_id_apic);
free(cache_id_apic);
return ret;
}
uint32_t is_smt_enabled_amd(struct topology* topo) {
uint32_t id;
for(int i = 0; i < topo->total_cores; i++) {
if(!bind_to_cpu(i)) {
printErr("Failed binding to CPU %d", i);
return false;
}
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;
}

20
src/x86/apic.h Normal file
View File

@@ -0,0 +1,20 @@
#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;
uint32_t* cache_select_mask;
uint32_t* cache_id_apic;
};
bool get_topology_from_apic(struct cpuInfo* cpu, struct topology* topo);
uint32_t is_smt_enabled_amd(struct topology* topo);
#endif

1057
src/x86/cpuid.c Executable file

File diff suppressed because it is too large Load Diff

44
src/x86/cpuid.h Normal file
View File

@@ -0,0 +1,44 @@
#ifndef __CPUID__
#define __CPUID__
#include "../common/cpu.h"
struct cpuInfo* get_cpu_info();
struct cache* get_cache_info(struct cpuInfo* cpu);
struct frequency* get_frequency_info(struct cpuInfo* cpu);
struct topology* get_topology_info(struct cpuInfo* cpu, struct cache* cach);
VENDOR get_cpu_vendor(struct cpuInfo* cpu);
uint32_t get_nsockets(struct topology* topo);
int64_t get_freq(struct frequency* freq);
char* get_str_cpu_name(struct cpuInfo* cpu);
char* get_str_ncores(struct cpuInfo* cpu);
char* get_str_avx(struct cpuInfo* cpu);
char* get_str_sse(struct cpuInfo* cpu);
char* get_str_fma(struct cpuInfo* cpu);
char* get_str_aes(struct cpuInfo* cpu);
char* get_str_sha(struct cpuInfo* cpu);
char* get_str_l1i(struct cache* cach);
char* get_str_l1d(struct cache* cach);
char* get_str_l2(struct cache* cach);
char* get_str_l3(struct cache* cach);
char* get_str_freq(struct frequency* freq);
char* get_str_sockets(struct topology* topo);
char* get_str_topology(struct cpuInfo* cpu, struct topology* topo, bool dual_socket);
char* get_str_peak_performance(struct cpuInfo* cpu, struct topology* topo, int64_t freq);
void free_cache_struct(struct cache* cach);
void free_topo_struct(struct topology* topo);
void free_freq_struct(struct frequency* freq);
void free_cpuinfo_struct(struct cpuInfo* cpu);
void debug_cpu_info(struct cpuInfo* cpu);
void debug_cache(struct cache* cach);
void debug_frequency(struct frequency* freq);
#endif

10
src/x86/cpuid_asm.c Normal file
View File

@@ -0,0 +1,10 @@
#include "cpuid_asm.h"
void cpuid(uint32_t *eax, uint32_t *ebx, uint32_t *ecx, uint32_t *edx) {
__asm volatile("cpuid"
: "=a" (*eax),
"=b" (*ebx),
"=c" (*ecx),
"=d" (*edx)
: "0" (*eax), "2" (*ecx));
}

8
src/x86/cpuid_asm.h Normal file
View File

@@ -0,0 +1,8 @@
#ifndef __CPUID_ASM__
#define __CPUID_ASM__
#include <stdint.h>
void cpuid(uint32_t *eax, uint32_t *ebx, uint32_t *ecx, uint32_t *edx);
#endif

405
src/x86/uarch.c Normal file
View File

@@ -0,0 +1,405 @@
#include <stdbool.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include "uarch.h"
#include "../common/global.h"
/*
* - cpuid codes are based on Todd Allen's cpuid program
* http://www.etallen.com/cpuid.html
* - This should be updated from time to time, to support newer CPUs. A good reference to look at:
* https://en.wikichip.org/
*/
// From Todd Allen:
//
// MSR_CPUID_table* is a table that appears in Intel document 325462, "Intel 64
// and IA-32 Architectures Software Developer's Manual Combined Volumes: 1, 2A,
// 2B, 2C, 2D, 3A, 3B, 3C, 3D, and 4" (the name changes from version to version
// as more volumes are added). The table moves around from version to version,
// but in version 071US, was in "Volume 4: Model-Specific Registers", Table 2-1:
// "CPUID Signature Values of DisplayFamily_DisplayModel".
// MRG* is a table that forms the bulk of Intel Microcode Revision Guidance (or
// Microcode Update Guidance). Its purpose is not to list CPUID values, but
// it does so, and sometimes lists values that appear nowhere else.
// LX* indicates features that I have seen no documentation for, but which are
// used by the Linux kernel (which is good evidence that they're correct).
// The "hook" to find these generally is an X86_FEATURE_* flag in:
// arch/x86/include/asm/cpufeatures.h
// For (synth) and (uarch synth) decoding, it often indicates
// family/model/stepping value which are documented nowhere else. These usually
// can be found in:
// arch/x86/include/asm/intel-family.h
typedef uint32_t MICROARCH;
// Data not available
#define NA -1
// Unknown manufacturing process
#define UNK -1
enum {
UARCH_UNKNOWN,
// INTEL //
UARCH_P5,
UARCH_P6,
UARCH_DOTHAN,
UARCH_YONAH,
UARCH_MEROM,
UARCH_PENYR,
UARCH_NEHALEM,
UARCH_WESTMERE,
UARCH_BONNELL,
UARCH_SALTWELL,
UARCH_SANDY_BRIDGE,
UARCH_SILVERMONT,
UARCH_IVY_BRIDGE,
UARCH_HASWELL,
UARCH_BROADWELL,
UARCH_AIRMONT,
UARCH_KABY_LAKE,
UARCH_SKYLAKE,
UARCH_CASCADE_LAKE,
UARCH_COOPER_LAKE,
UARCH_KNIGHTS_LANDING,
UARCH_KNIGHTS_MILL,
UARCH_GOLDMONT,
UARCH_PALM_COVE,
UARCH_SUNNY_COVE,
UARCH_GOLDMONT_PLUS,
UARCH_TREMONT,
UARCH_WILLOW_COVE,
UARCH_COFFE_LAKE,
UARCH_ITANIUM,
UARCH_KNIGHTS_FERRY,
UARCH_KNIGHTS_CORNER,
UARCH_WILLAMETTE,
UARCH_NORTHWOOD,
UARCH_PRESCOTT,
UARCH_CEDAR_MILL,
UARCH_ITANIUM2,
UARCH_ICE_LAKE,
// AMD //
UARCH_AM486,
UARCH_AM5X86,
UARCH_K6,
UARCH_K7,
UARCH_K8,
UARCH_K10,
UARCH_PUMA_2008,
UARCH_BOBCAT,
UARCH_BULLDOZER,
UARCH_PILEDRIVER,
UARCH_STEAMROLLER,
UARCH_EXCAVATOR,
UARCH_JAGUAR,
UARCH_PUMA_2014,
UARCH_ZEN,
UARCH_ZEN_PLUS,
UARCH_ZEN2,
UARCH_ZEN3
};
struct uarch {
MICROARCH uarch;
char* uarch_str;
int32_t process; // measured in nanometers
};
#define UARCH_START if (false) {}
#define CHECK_UARCH(arch, ef_, f_, em_, m_, s_, str, uarch, process) \
else if (ef_ == ef && f_ == f && (em_ == NA || em_ == em) && (m_ == NA || m_ == m) && (s_ == NA || s_ == s)) fill_uarch(arch, str, uarch, process);
#define UARCH_END else { printBug("Unknown microarchitecture detected: M=0x%.8X EM=0x%.8X F=0x%.8X EF=0x%.8X S=0x%.8X", m, em, f, ef, s); fill_uarch(arch, "Unknown", UARCH_UNKNOWN, 0); }
void fill_uarch(struct uarch* arch, char* str, MICROARCH u, uint32_t process) {
arch->uarch_str = malloc(sizeof(char) * (strlen(str)+1));
strcpy(arch->uarch_str, str);
arch->uarch = u;
arch->process= process;
}
// iNApired in Todd Allen's decode_uarch_intel
struct uarch* get_uarch_from_cpuid_intel(uint32_t ef, uint32_t f, uint32_t em, uint32_t m, int s) {
struct uarch* arch = malloc(sizeof(struct uarch));
// EF: Extended Family //
// F: Family //
// EM: Extended Model //
// M: Model //
// S: Stepping //
// ----------------------------------------------------------------------------- //
// EF F EM M S //
UARCH_START
CHECK_UARCH(arch, 0, 5, 0, 0, NA, "P5", UARCH_P5, 800)
CHECK_UARCH(arch, 0, 5, 0, 1, NA, "P5", UARCH_P5, 800)
CHECK_UARCH(arch, 0, 5, 0, 2, NA, "P5", UARCH_P5, UNK)
CHECK_UARCH(arch, 0, 5, 0, 3, NA, "P5", UARCH_P5, 600)
CHECK_UARCH(arch, 0, 5, 0, 4, NA, "P5 MMX", UARCH_P5, UNK)
CHECK_UARCH(arch, 0, 5, 0, 7, NA, "P5 MMX", UARCH_P5, UNK)
CHECK_UARCH(arch, 0, 5, 0, 8, NA, "P5 MMX", UARCH_P5, 250)
CHECK_UARCH(arch, 0, 5, 0, 9, NA, "P5 MMX", UARCH_P5, UNK)
CHECK_UARCH(arch, 0, 6, 0, 0, NA, "P6 Pentium II", UARCH_P6, UNK)
CHECK_UARCH(arch, 0, 6, 0, 1, NA, "P6 Pentium II", UARCH_P6, UNK) // process depends on core
CHECK_UARCH(arch, 0, 6, 0, 2, NA, "P6 Pentium II", UARCH_P6, UNK)
CHECK_UARCH(arch, 0, 6, 0, 3, NA, "P6 Pentium II", UARCH_P6, 350)
CHECK_UARCH(arch, 0, 6, 0, 4, NA, "P6 Pentium II", UARCH_P6, UNK)
CHECK_UARCH(arch, 0, 6, 0, 5, NA, "P6 Pentium II", UARCH_P6, 250)
CHECK_UARCH(arch, 0, 6, 0, 6, NA, "P6 Pentium II", UARCH_P6, UNK)
CHECK_UARCH(arch, 0, 6, 0, 7, NA, "P6 Pentium III", UARCH_P6, 250)
CHECK_UARCH(arch, 0, 6, 0, 8, NA, "P6 Pentium III", UARCH_P6, 180)
CHECK_UARCH(arch, 0, 6, 0, 9, NA, "P6 Pentium M", UARCH_P6, 130)
CHECK_UARCH(arch, 0, 6, 0, 10, NA, "P6 Pentium III", UARCH_P6, 180)
CHECK_UARCH(arch, 0, 6, 0, 11, NA, "P6 Pentium III", UARCH_P6, 130)
CHECK_UARCH(arch, 0, 6, 0, 13, NA, "Dothan", UARCH_DOTHAN, UNK) // process depends on core
CHECK_UARCH(arch, 0, 6, 0, 14, NA, "Yonah", UARCH_YONAH, 65)
CHECK_UARCH(arch, 0, 6, 0, 15, NA, "Merom", UARCH_MEROM, 65)
CHECK_UARCH(arch, 0, 6, 1, 5, NA, "Dothan", UARCH_DOTHAN, 90)
CHECK_UARCH(arch, 0, 6, 1, 6, NA, "Merom", UARCH_MEROM, 65)
CHECK_UARCH(arch, 0, 6, 1, 7, NA, "Penryn", UARCH_PENYR, 45)
CHECK_UARCH(arch, 0, 6, 1, 10, NA, "Nehalem", UARCH_NEHALEM, 45)
CHECK_UARCH(arch, 0, 6, 1, 12, NA, "Bonnell", UARCH_BONNELL, 45)
CHECK_UARCH(arch, 0, 6, 1, 13, NA, "Penryn", UARCH_PENYR, 45)
CHECK_UARCH(arch, 0, 6, 1, 14, NA, "Nehalem", UARCH_NEHALEM, 45)
CHECK_UARCH(arch, 0, 6, 1, 15, NA, "Nehalem", UARCH_NEHALEM, 45)
CHECK_UARCH(arch, 0, 6, 2, 5, NA, "Westmere", UARCH_WESTMERE, 32)
CHECK_UARCH(arch, 0, 6, 2 , 6, NA, "Bonnell", UARCH_BONNELL, 45)
CHECK_UARCH(arch, 0, 6, 2, 7, NA, "Saltwell", UARCH_SALTWELL, 32)
CHECK_UARCH(arch, 0, 6, 2, 10, NA, "Sandy Bridge", UARCH_SANDY_BRIDGE, 32)
CHECK_UARCH(arch, 0, 6, 2, 12, NA, "Westmere", UARCH_WESTMERE, 32)
CHECK_UARCH(arch, 0, 6, 2, 13, NA, "Sandy Bridge", UARCH_SANDY_BRIDGE, 32)
CHECK_UARCH(arch, 0, 6, 2, 14, NA, "Nehalem", UARCH_NEHALEM, 45)
CHECK_UARCH(arch, 0, 6, 2, 15, NA, "Westmere", UARCH_WESTMERE, 32)
CHECK_UARCH(arch, 0, 6, 3, 5, NA, "Saltwell", UARCH_SALTWELL, 14)
CHECK_UARCH(arch, 0, 6, 3, 6, NA, "Saltwell", UARCH_SALTWELL, 32)
CHECK_UARCH(arch, 0, 6, 3, 7, NA, "Silvermont", UARCH_SILVERMONT, 22)
CHECK_UARCH(arch, 0, 6, 3, 10, NA, "Ivy Bridge", UARCH_IVY_BRIDGE, 22)
CHECK_UARCH(arch, 0, 6, 3, 12, NA, "Haswell", UARCH_HASWELL, 22)
CHECK_UARCH(arch, 0, 6, 3, 13, NA, "Broadwell", UARCH_BROADWELL, 14)
CHECK_UARCH(arch, 0, 6, 3, 14, NA, "Ivy Bridge", UARCH_IVY_BRIDGE, 22)
CHECK_UARCH(arch, 0, 6, 3, 15, NA, "Haswell", UARCH_HASWELL, 22)
CHECK_UARCH(arch, 0, 6, 4, 5, NA, "Haswell", UARCH_HASWELL, 22)
CHECK_UARCH(arch, 0, 6, 4, 6, NA, "Haswell", UARCH_HASWELL, 22)
CHECK_UARCH(arch, 0, 6, 4, 7, NA, "Broadwell", UARCH_BROADWELL, 14)
CHECK_UARCH(arch, 0, 6, 4, 10, NA, "Silvermont", UARCH_SILVERMONT, 22) // no docs, but /proc/cpuinfo seen in wild
CHECK_UARCH(arch, 0, 6, 4, 12, NA, "Airmont", UARCH_AIRMONT, 14)
CHECK_UARCH(arch, 0, 6, 4, 13, NA, "Silvermont", UARCH_SILVERMONT, 22)
CHECK_UARCH(arch, 0, 6, 4, 14, 8, "Kaby Lake", UARCH_KABY_LAKE, 14)
CHECK_UARCH(arch, 0, 6, 4, 14, NA, "Skylake", UARCH_SKYLAKE, 14)
CHECK_UARCH(arch, 0, 6, 4, 15, NA, "Broadwell", UARCH_BROADWELL, 14)
CHECK_UARCH(arch, 0, 6, 5, 5, 6, "Cascade Lake", UARCH_CASCADE_LAKE, 14) // no docs, but example from Greg Stewart
CHECK_UARCH(arch, 0, 6, 5, 5, 7, "Cascade Lake", UARCH_CASCADE_LAKE, 14)
CHECK_UARCH(arch, 0, 6, 5, 5, 10, "Cooper Lake", UARCH_COOPER_LAKE, 14)
CHECK_UARCH(arch, 0, 6, 5, 5, NA, "Skylake", UARCH_SKYLAKE, 14)
CHECK_UARCH(arch, 0, 6, 5, 6, NA, "Broadwell", UARCH_BROADWELL, 14)
CHECK_UARCH(arch, 0, 6, 5, 7, NA, "Knights Landing", UARCH_KNIGHTS_LANDING, 14)
CHECK_UARCH(arch, 0, 6, 5, 10, NA, "Silvermont", UARCH_SILVERMONT, 22) // no spec update; only MSR_CPUID_table* so far
CHECK_UARCH(arch, 0, 6, 5, 12, NA, "Goldmont", UARCH_GOLDMONT, 14)
CHECK_UARCH(arch, 0, 6, 5, 13, NA, "Silvermont", UARCH_SILVERMONT, 22) // no spec update; only MSR_CPUID_table* so far
CHECK_UARCH(arch, 0, 6, 5, 14, 8, "Kaby Lake", UARCH_KABY_LAKE, 14)
CHECK_UARCH(arch, 0, 6, 5, 14, NA, "Skylake", UARCH_SKYLAKE, 14)
CHECK_UARCH(arch, 0, 6, 5, 15, NA, "Goldmont", UARCH_GOLDMONT, 14)
CHECK_UARCH(arch, 0, 6, 6, 6, NA, "Palm Cove", UARCH_PALM_COVE, 10) // no spec update; only MSR_CPUID_table* so far
CHECK_UARCH(arch, 0, 6, 6, 10, NA, "Sunny Cove", UARCH_SUNNY_COVE, 10) // no spec update; only MSR_CPUID_table* so far
CHECK_UARCH(arch, 0, 6, 6, 12, NA, "Sunny Cove", UARCH_SUNNY_COVE, 10) // no spec update; only MSR_CPUID_table* so far
CHECK_UARCH(arch, 0, 6, 7, 5, NA, "Airmont", UARCH_AIRMONT, 14) // no spec update; whispers & rumors
CHECK_UARCH(arch, 0, 6, 7, 10, NA, "Goldmont Plus", UARCH_GOLDMONT_PLUS, 14)
CHECK_UARCH(arch, 0, 6, 7, 13, NA, "Sunny Cove", UARCH_SUNNY_COVE, 10) // no spec update; only MSR_CPUID_table* so far
CHECK_UARCH(arch, 0, 6, 7, 14, NA, "Ice Lake", UARCH_ICE_LAKE, 10)
CHECK_UARCH(arch, 0, 6, 8, 5, NA, "Knights Mill", UARCH_KNIGHTS_MILL, 14) // no spec update; only MSR_CPUID_table* so far
CHECK_UARCH(arch, 0, 6, 8, 6, NA, "Tremont", UARCH_TREMONT, 10) // LX*
CHECK_UARCH(arch, 0, 6, 8, 10, NA, "Tremont", UARCH_TREMONT, 10) // no spec update; only geekbench.com example
CHECK_UARCH(arch, 0, 6, 8, 12, NA, "Willow Cove", UARCH_WILLOW_COVE, 10) // found only on en.wikichip.org
CHECK_UARCH(arch, 0, 6, 8, 13, NA, "Willow Cove", UARCH_WILLOW_COVE, 10) // LX*
CHECK_UARCH(arch, 0, 6, 8, 14, NA, "Kaby Lake", UARCH_KABY_LAKE, 14)
CHECK_UARCH(arch, 0, 6, 9, 6, NA, "Tremont", UARCH_TREMONT, 10) // LX*
CHECK_UARCH(arch, 0, 6, 9, 12, NA, "Tremont", UARCH_TREMONT, 10) // LX*
CHECK_UARCH(arch, 0, 6, 9, 13, NA, "Sunny Cove", UARCH_SUNNY_COVE, 10) // LX*
CHECK_UARCH(arch, 0, 6, 9, 14, 9, "Kaby Lake", UARCH_KABY_LAKE, 14)
CHECK_UARCH(arch, 0, 6, 9, 14, 10, "Coffee Lake", UARCH_COFFE_LAKE, 14)
CHECK_UARCH(arch, 0, 6, 9, 14, 11, "Coffee Lake", UARCH_COFFE_LAKE, 14)
CHECK_UARCH(arch, 0, 6, 9, 14, 12, "Coffee Lake", UARCH_COFFE_LAKE, 14)
CHECK_UARCH(arch, 0, 6, 9, 14, 13, "Coffee Lake", UARCH_COFFE_LAKE, 14)
CHECK_UARCH(arch, 0, 6, 10, 5, NA, "Kaby Lake", UARCH_KABY_LAKE, 14) // LX*
CHECK_UARCH(arch, 0, 6, 10, 6, NA, "Kaby Lake", UARCH_KABY_LAKE, 14) // no spec update; only iNAtlatx64 example
CHECK_UARCH(arch, 0, 11, 0, 0, NA, "Knights Ferry", UARCH_KNIGHTS_FERRY, 45) // found only on en.wikichip.org
CHECK_UARCH(arch, 0, 11, 0, 1, NA, "Knights Corner", UARCH_KNIGHTS_CORNER, 22)
CHECK_UARCH(arch, 0, 15, 0, 0, NA, "Willamette", UARCH_WILLAMETTE, 180)
CHECK_UARCH(arch, 0, 15, 0, 1, NA, "Willamette", UARCH_WILLAMETTE, 180)
CHECK_UARCH(arch, 0, 15, 0, 2, NA, "Northwood", UARCH_NORTHWOOD, 130)
CHECK_UARCH(arch, 0, 15, 0, 3, NA, "Prescott", UARCH_PRESCOTT, 90)
CHECK_UARCH(arch, 0, 15, 0, 4, NA, "Prescott", UARCH_PRESCOTT, 90)
CHECK_UARCH(arch, 0, 15, 0, 6, NA, "Cedar Mill", UARCH_CEDAR_MILL, 65)
CHECK_UARCH(arch, 1, 15, 0, 0, NA, "Itanium2", UARCH_ITANIUM2, 180)
CHECK_UARCH(arch, 1, 15, 0, 1, NA, "Itanium2", UARCH_ITANIUM2, 130)
CHECK_UARCH(arch, 1, 15, 0, 2, NA, "Itanium2", UARCH_ITANIUM2, 130)
UARCH_END
return arch;
}
// iNApired in Todd Allen's decode_uarch_amd
struct uarch* get_uarch_from_cpuid_amd(uint32_t ef, uint32_t f, uint32_t em, uint32_t m, int s) {
struct uarch* arch = malloc(sizeof(struct uarch));
// EF: Extended Family //
// F: Family //
// EM: Extended Model //
// M: Model //
// S: Stepping //
// ----------------------------------------------------------------------------- //
// EF F EM M S //
UARCH_START
CHECK_UARCH(arch, 0, 4, 0, 3, NA, "Am486", UARCH_AM486, UNK)
CHECK_UARCH(arch, 0, 4, 0, 7, NA, "Am486", UARCH_AM486, UNK)
CHECK_UARCH(arch, 0, 4, 0, 8, NA, "Am486", UARCH_AM486, UNK)
CHECK_UARCH(arch, 0, 4, 0, 9, NA, "Am486", UARCH_AM486, UNK)
CHECK_UARCH(arch, 0, 4, NA, NA, NA, "Am5x86", UARCH_AM5X86, UNK)
CHECK_UARCH(arch, 0, 5, 0, 6, NA, "K6", UARCH_K6, 300)
CHECK_UARCH(arch, 0, 5, 0, 7, NA, "K6", UARCH_K6, 250) // *p from sandpile.org
CHECK_UARCH(arch, 0, 5, 0, 13, NA, "K6", UARCH_K6, 80) // *p from sandpile.org
CHECK_UARCH(arch, 0, 5, NA, NA, NA, "K6", UARCH_K6, UNK)
CHECK_UARCH(arch, 0, 6, 0, 1, NA, "K7", UARCH_K7, 250)
CHECK_UARCH(arch, 0, 6, 0, 2, NA, "K7", UARCH_K7, 180)
CHECK_UARCH(arch, 0, 6, NA, NA, NA, "K7", UARCH_K7, UNK)
CHECK_UARCH(arch, 0, 15, 0, 4, 8, "K8", UARCH_K8, 130)
CHECK_UARCH(arch, 0, 15, 0, 4, NA, "K8", UARCH_K8, 130)
CHECK_UARCH(arch, 0, 15, 0, 5, NA, "K8", UARCH_K8, 130)
CHECK_UARCH(arch, 0, 15, 0, 7, NA, "K8", UARCH_K8, 130)
CHECK_UARCH(arch, 0, 15, 0, 8, NA, "K8", UARCH_K8, 130)
CHECK_UARCH(arch, 0, 15, 0, 11, NA, "K8", UARCH_K8, 130)
CHECK_UARCH(arch, 0, 15, 0, 12, NA, "K8", UARCH_K8, 130)
CHECK_UARCH(arch, 0, 15, 0, 14, NA, "K8", UARCH_K8, 130)
CHECK_UARCH(arch, 0, 15, 0, 15, NA, "K8", UARCH_K8, 130)
CHECK_UARCH(arch, 0, 15, 1, 4, NA, "K8", UARCH_K8, 90)
CHECK_UARCH(arch, 0, 15, 1, 5, NA, "K8", UARCH_K8, 90)
CHECK_UARCH(arch, 0, 15, 1, 7, NA, "K8", UARCH_K8, 90)
CHECK_UARCH(arch, 0, 15, 1, 8, NA, "K8", UARCH_K8, 90)
CHECK_UARCH(arch, 0, 15, 1, 11, NA, "K8", UARCH_K8, 90)
CHECK_UARCH(arch, 0, 15, 1, 12, NA, "K8", UARCH_K8, 90)
CHECK_UARCH(arch, 0, 15, 1, 15, NA, "K8", UARCH_K8, 90)
CHECK_UARCH(arch, 0, 15, 2, 1, NA, "K8", UARCH_K8, 90)
CHECK_UARCH(arch, 0, 15, 2, 3, NA, "K8", UARCH_K8, 90)
CHECK_UARCH(arch, 0, 15, 2, 4, NA, "K8", UARCH_K8, 90)
CHECK_UARCH(arch, 0, 15, 2, 5, NA, "K8", UARCH_K8, 90)
CHECK_UARCH(arch, 0, 15, 2, 7, NA, "K8", UARCH_K8, 90)
CHECK_UARCH(arch, 0, 15, 2, 11, NA, "K8", UARCH_K8, 90)
CHECK_UARCH(arch, 0, 15, 2, 12, NA, "K8", UARCH_K8, 90)
CHECK_UARCH(arch, 0, 15, 2, 15, NA, "K8", UARCH_K8, 90)
CHECK_UARCH(arch, 0, 15, 4, 1, NA, "K8", UARCH_K8, 90)
CHECK_UARCH(arch, 0, 15, 4, 3, NA, "K8", UARCH_K8, 90)
CHECK_UARCH(arch, 0, 15, 4, 8, NA, "K8", UARCH_K8, 90)
CHECK_UARCH(arch, 0, 15, 4, 11, NA, "K8", UARCH_K8, 90)
CHECK_UARCH(arch, 0, 15, 4, 12, NA, "K8", UARCH_K8, 90)
CHECK_UARCH(arch, 0, 15, 4, 15, NA, "K8", UARCH_K8, 90)
CHECK_UARCH(arch, 0, 15, 5, 13, NA, "K8", UARCH_K8, 90)
CHECK_UARCH(arch, 0, 15, 5, 15, NA, "K8", UARCH_K8, 90)
CHECK_UARCH(arch, 0, 15, 6, 8, NA, "K8", UARCH_K8, 65)
CHECK_UARCH(arch, 0, 15, 6, 11, NA, "K8", UARCH_K8, 65)
CHECK_UARCH(arch, 0, 15, 6, 12, NA, "K8", UARCH_K8, 65)
CHECK_UARCH(arch, 0, 15, 6, 15, NA, "K8", UARCH_K8, 65)
CHECK_UARCH(arch, 0, 15, 7, 12, NA, "K8", UARCH_K8, 65)
CHECK_UARCH(arch, 0, 15, 7, 15, NA, "K8", UARCH_K8, 65)
CHECK_UARCH(arch, 0, 15, 12, 1, NA, "K8", UARCH_K8, 90)
CHECK_UARCH(arch, 1, 15, 0, 0, NA, "K10", UARCH_K10, 65) // sandpile.org
CHECK_UARCH(arch, 1, 15, 0, 2, NA, "K10", UARCH_K10, 65)
CHECK_UARCH(arch, 1, 15, 0, 4, NA, "K10", UARCH_K10, 45)
CHECK_UARCH(arch, 1, 15, 0, 5, NA, "K10", UARCH_K10, 45)
CHECK_UARCH(arch, 1, 15, 0, 6, NA, "K10", UARCH_K10, 45)
CHECK_UARCH(arch, 1, 15, 0, 8, NA, "K10", UARCH_K10, 45)
CHECK_UARCH(arch, 1, 15, 0, 9, NA, "K10", UARCH_K10, 45)
CHECK_UARCH(arch, 1, 15, 0, 10, NA, "K10", UARCH_K10, 45)
CHECK_UARCH(arch, 2, 15, NA, NA, NA, "Puma 2008", UARCH_PUMA_2008, 65)
CHECK_UARCH(arch, 3, 15, NA, NA, NA, "K10", UARCH_K10, 32)
CHECK_UARCH(arch, 5, 15, NA, NA, NA, "Bobcat", UARCH_BOBCAT, 40)
CHECK_UARCH(arch, 6, 15, 0, 0, NA, "Bulldozer", UARCH_BULLDOZER, 32) // iNAtlatx64 engr sample
CHECK_UARCH(arch, 6, 15, 0, 1, NA, "Bulldozer", UARCH_BULLDOZER, 32)
CHECK_UARCH(arch, 6, 15, 0, 2, NA, "Piledriver", UARCH_PILEDRIVER, 32)
CHECK_UARCH(arch, 6, 15, 1, 0, NA, "Piledriver", UARCH_PILEDRIVER, 32)
CHECK_UARCH(arch, 6, 15, 1, 3, NA, "Piledriver", UARCH_PILEDRIVER, 32)
CHECK_UARCH(arch, 6, 15, 3, 0, NA, "Steamroller", UARCH_STEAMROLLER, 28)
CHECK_UARCH(arch, 6, 15, 3, 8, NA, "Steamroller", UARCH_STEAMROLLER, 28)
CHECK_UARCH(arch, 6, 15, 4, 0, NA, "Steamroller", UARCH_STEAMROLLER, 28) // Software Optimization Guide (15h) says it has the same iNAt latencies as (6,15),(3,x).
CHECK_UARCH(arch, 6, 15, 6, 0, NA, "Excavator", UARCH_EXCAVATOR, 28) // undocumented, but iNAtlatx64 samples
CHECK_UARCH(arch, 6, 15, 6, 5, NA, "Excavator", UARCH_EXCAVATOR, 28) // undocumented, but sample from Alexandros Couloumbis
CHECK_UARCH(arch, 6, 15, 7, 0, NA, "Excavator", UARCH_EXCAVATOR, 28)
CHECK_UARCH(arch, 7, 15, 0, 0, NA, "Jaguar", UARCH_JAGUAR, 28)
CHECK_UARCH(arch, 7, 15, 3, 0, NA, "Puma 2014", UARCH_PUMA_2014, 28)
CHECK_UARCH(arch, 8, 15, 0, 0, NA, "Zen", UARCH_ZEN, 14) // iNAtlatx64 engr sample
CHECK_UARCH(arch, 8, 15, 0, 1, NA, "Zen", UARCH_ZEN, 14)
CHECK_UARCH(arch, 8, 15, 0, 8, NA, "Zen+", UARCH_ZEN_PLUS, 12)
CHECK_UARCH(arch, 8, 15, 1, 1, NA, "Zen", UARCH_ZEN, 14) // found only on en.wikichip.org & iNAtlatx64 examples
CHECK_UARCH(arch, 8, 15, 1, 8, NA, "Zen+", UARCH_ZEN_PLUS, 12) // found only on en.wikichip.org
CHECK_UARCH(arch, 8, 15, 3, 1, NA, "Zen 2", UARCH_ZEN2, 7) // found only on en.wikichip.org
CHECK_UARCH(arch, 8, 15, 6, 0, NA, "Zen 2", UARCH_ZEN2, 7) // undocumented, geekbench.com example
CHECK_UARCH(arch, 8, 15, 7, 1, NA, "Zen 2", UARCH_ZEN2, 7) // undocumented, but samples from Steven Noonan
CHECK_UARCH(arch, 10, 15, NA, NA, NA, "Zen 3", UARCH_ZEN3, 7) // undocumented, LX*
UARCH_END
return arch;
}
struct uarch* get_uarch_from_cpuid(struct cpuInfo* cpu, uint32_t ef, uint32_t f, uint32_t em, uint32_t m, int s) {
if(cpu->cpu_vendor == CPU_VENDOR_INTEL)
return get_uarch_from_cpuid_intel(ef, f, em, m, s);
else
return get_uarch_from_cpuid_amd(ef, f, em, m, s);
}
bool vpus_are_AVX512(struct cpuInfo* cpu) {
return cpu->arch->uarch != UARCH_ICE_LAKE;
return cpu->arch->uarch != UARCH_ICE_LAKE;
}
bool is_knights_landing(struct cpuInfo* cpu) {
return cpu->arch->uarch == UARCH_KNIGHTS_LANDING;
}
int get_number_of_vpus(struct cpuInfo* cpu) {
if(cpu->cpu_vendor == CPU_VENDOR_AMD)
return 1;
switch(cpu->arch->uarch) {
case UARCH_HASWELL:
case UARCH_BROADWELL:
case UARCH_SKYLAKE:
case UARCH_CASCADE_LAKE:
case UARCH_KABY_LAKE:
case UARCH_COFFE_LAKE:
case UARCH_PALM_COVE:
case UARCH_KNIGHTS_LANDING:
case UARCH_KNIGHTS_MILL:
case UARCH_ICE_LAKE:
return 2;
default:
return 1;
}
}
char* get_str_uarch(struct cpuInfo* cpu) {
return cpu->arch->uarch_str;
}
char* get_str_process(struct cpuInfo* cpu) {
char* str = malloc(sizeof(char) * (4+2+1));
uint32_t process = cpu->arch->process;
if(process > 100)
sprintf(str, "%.2fum", (double)process/100);
else
sprintf(str, "%dnm", process);
return str;
}
void free_uarch_struct(struct uarch* arch) {
free(arch->uarch_str);
free(arch);
}

18
src/x86/uarch.h Normal file
View File

@@ -0,0 +1,18 @@
#ifndef __UARCH__
#define __UARCH__
#include <stdint.h>
#include "cpuid.h"
struct uarch;
struct uarch* get_uarch_from_cpuid(struct cpuInfo* cpu, uint32_t ef, uint32_t f, uint32_t em, uint32_t m, int s);
bool vpus_are_AVX512(struct cpuInfo* cpu);
bool is_knights_landing(struct cpuInfo* cpu);
int get_number_of_vpus(struct cpuInfo* cpu);
char* get_str_uarch(struct cpuInfo* cpu);
char* get_str_process(struct cpuInfo* cpu);
void free_uarch_struct(struct uarch* arch);
#endif

74
src/x86/udev.c Normal file
View File

@@ -0,0 +1,74 @@
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <unistd.h>
#include <fcntl.h>
#include <errno.h>
#include "../common/global.h"
#include "cpuid.h"
#define _PATH_SYS_SYSTEM "/sys/devices/system"
#define _PATH_SYS_CPU _PATH_SYS_SYSTEM"/cpu"
#define _PATH_ONE_CPU _PATH_SYS_CPU"/cpu0"
#define _PATH_FREQUENCY _PATH_ONE_CPU"/cpufreq"
#define _PATH_FREQUENCY_MAX _PATH_FREQUENCY"/cpuinfo_max_freq"
#define _PATH_FREQUENCY_MIN _PATH_FREQUENCY"/cpuinfo_min_freq"
#define DEFAULT_FILE_SIZE 4096
long get_freq_from_file(char* path) {
int fd = open(path, O_RDONLY);
if(fd == -1) {
perror("open");
printBug("Could not open '%s'", path);
return UNKNOWN_FREQ;
}
//File exists, read it
int bytes_read = 0;
int offset = 0;
int block = 1;
char* buf = malloc(sizeof(char)*DEFAULT_FILE_SIZE);
memset(buf, 0, sizeof(char)*DEFAULT_FILE_SIZE);
while ( (bytes_read = read(fd, buf+offset, block)) > 0 ) {
offset += bytes_read;
}
char* end;
errno = 0;
long ret = strtol(buf, &end, 10);
if(errno != 0) {
perror("strtol");
printBug("Failed parsing '%s' file. Read data was: '%s'", path, buf);
free(buf);
return UNKNOWN_FREQ;
}
// We will be getting the frequency in KHz
// We consider it is an error if frequency is
// greater than 10 GHz or less than 100 MHz
if(ret > 10000 * 1000 || ret < 100 * 1000) {
printBug("Invalid data was read from file '%s': %ld\n", path, ret);
return UNKNOWN_FREQ;
}
free(buf);
if (close(fd) == -1) {
perror("close");
printErr("Closing '%s' failed\n", path);
}
return ret/1000;
}
long get_max_freq_from_file() {
return get_freq_from_file(_PATH_FREQUENCY_MAX);
}
long get_min_freq_from_file() {
return get_freq_from_file(_PATH_FREQUENCY_MIN);
}

7
src/x86/udev.h Normal file
View File

@@ -0,0 +1,7 @@
#ifndef __UDEV__
#define __UDEV__
long get_max_freq_from_file();
long get_min_freq_from_file();
#endif