// Take a look at the license at the top of the repository in the LICENSE file. use crate::sys::tools::KeyHandler; use crate::{CpuExt, CpuRefreshKind, LoadAvg}; use std::collections::HashMap; use std::io::Error; use std::mem; use std::ops::DerefMut; use std::ptr::null_mut; use std::sync::Mutex; use ntapi::ntpoapi::PROCESSOR_POWER_INFORMATION; use winapi::shared::minwindef::FALSE; use winapi::shared::winerror::{ERROR_INSUFFICIENT_BUFFER, ERROR_SUCCESS}; use winapi::um::handleapi::CloseHandle; use winapi::um::pdh::{ PdhAddEnglishCounterA, PdhAddEnglishCounterW, PdhCloseQuery, PdhCollectQueryData, PdhCollectQueryDataEx, PdhGetFormattedCounterValue, PdhOpenQueryA, PdhRemoveCounter, PDH_FMT_COUNTERVALUE, PDH_FMT_DOUBLE, PDH_HCOUNTER, PDH_HQUERY, }; use winapi::um::powerbase::CallNtPowerInformation; use winapi::um::synchapi::CreateEventA; use winapi::um::sysinfoapi::GetLogicalProcessorInformationEx; use winapi::um::sysinfoapi::SYSTEM_INFO; use winapi::um::winbase::{RegisterWaitForSingleObject, INFINITE}; use winapi::um::winnt::{ ProcessorInformation, RelationAll, RelationProcessorCore, BOOLEAN, HANDLE, PSYSTEM_LOGICAL_PROCESSOR_INFORMATION_EX, PVOID, WT_EXECUTEDEFAULT, }; // This formula comes from linux's include/linux/sched/loadavg.h // https://github.com/torvalds/linux/blob/345671ea0f9258f410eb057b9ced9cefbbe5dc78/include/linux/sched/loadavg.h#L20-L23 #[allow(clippy::excessive_precision)] const LOADAVG_FACTOR_1F: f64 = 0.9200444146293232478931553241; #[allow(clippy::excessive_precision)] const LOADAVG_FACTOR_5F: f64 = 0.9834714538216174894737477501; #[allow(clippy::excessive_precision)] const LOADAVG_FACTOR_15F: f64 = 0.9944598480048967508795473394; // The time interval in seconds between taking load counts, same as Linux const SAMPLING_INTERVAL: usize = 5; // maybe use a read/write lock instead? static LOAD_AVG: once_cell::sync::Lazy>> = once_cell::sync::Lazy::new(|| unsafe { init_load_avg() }); pub(crate) fn get_load_average() -> LoadAvg { if let Ok(avg) = LOAD_AVG.lock() { if let Some(avg) = &*avg { return avg.clone(); } } LoadAvg::default() } unsafe extern "system" fn load_avg_callback(counter: PVOID, _: BOOLEAN) { let mut display_value = mem::MaybeUninit::::uninit(); if PdhGetFormattedCounterValue( counter as _, PDH_FMT_DOUBLE, null_mut(), display_value.as_mut_ptr(), ) != ERROR_SUCCESS as _ { return; } let display_value = display_value.assume_init(); if let Ok(mut avg) = LOAD_AVG.lock() { if let Some(avg) = avg.deref_mut() { let current_load = display_value.u.doubleValue(); avg.one = avg.one * LOADAVG_FACTOR_1F + current_load * (1.0 - LOADAVG_FACTOR_1F); avg.five = avg.five * LOADAVG_FACTOR_5F + current_load * (1.0 - LOADAVG_FACTOR_5F); avg.fifteen = avg.fifteen * LOADAVG_FACTOR_15F + current_load * (1.0 - LOADAVG_FACTOR_15F); } } } unsafe fn init_load_avg() -> Mutex> { // You can see the original implementation here: https://github.com/giampaolo/psutil let mut query = null_mut(); if PdhOpenQueryA(null_mut(), 0, &mut query) != ERROR_SUCCESS as _ { sysinfo_debug!("init_load_avg: PdhOpenQueryA failed"); return Mutex::new(None); } let mut counter: PDH_HCOUNTER = mem::zeroed(); if PdhAddEnglishCounterA( query, b"\\System\\Cpu Queue Length\0".as_ptr() as _, 0, &mut counter, ) != ERROR_SUCCESS as _ { PdhCloseQuery(query); sysinfo_debug!("init_load_avg: failed to get CPU queue length"); return Mutex::new(None); } let event = CreateEventA(null_mut(), FALSE, FALSE, b"LoadUpdateEvent\0".as_ptr() as _); if event.is_null() { PdhCloseQuery(query); sysinfo_debug!("init_load_avg: failed to create event `LoadUpdateEvent`"); return Mutex::new(None); } if PdhCollectQueryDataEx(query, SAMPLING_INTERVAL as _, event) != ERROR_SUCCESS as _ { PdhCloseQuery(query); sysinfo_debug!("init_load_avg: PdhCollectQueryDataEx failed"); return Mutex::new(None); } let mut wait_handle = null_mut(); if RegisterWaitForSingleObject( &mut wait_handle, event, Some(load_avg_callback), counter as _, INFINITE, WT_EXECUTEDEFAULT, ) == 0 { PdhRemoveCounter(counter); PdhCloseQuery(query); sysinfo_debug!("init_load_avg: RegisterWaitForSingleObject failed"); Mutex::new(None) } else { Mutex::new(Some(LoadAvg::default())) } } struct InternalQuery { query: PDH_HQUERY, event: HANDLE, data: HashMap, } unsafe impl Send for InternalQuery {} unsafe impl Sync for InternalQuery {} impl Drop for InternalQuery { fn drop(&mut self) { unsafe { for (_, counter) in self.data.iter() { PdhRemoveCounter(*counter); } if !self.event.is_null() { CloseHandle(self.event); } if !self.query.is_null() { PdhCloseQuery(self.query); } } } } pub(crate) struct Query { internal: InternalQuery, } impl Query { pub fn new() -> Option { let mut query = null_mut(); unsafe { if PdhOpenQueryA(null_mut(), 0, &mut query) == ERROR_SUCCESS as i32 { let q = InternalQuery { query, event: null_mut(), data: HashMap::new(), }; Some(Query { internal: q }) } else { sysinfo_debug!("Query::new: PdhOpenQueryA failed"); None } } } #[allow(clippy::ptr_arg)] pub fn get(&self, name: &String) -> Option { if let Some(counter) = self.internal.data.get(name) { unsafe { let mut display_value = mem::MaybeUninit::::uninit(); let counter: PDH_HCOUNTER = *counter; let ret = PdhGetFormattedCounterValue( counter, PDH_FMT_DOUBLE, null_mut(), display_value.as_mut_ptr(), ) as u32; let display_value = display_value.assume_init(); return if ret == ERROR_SUCCESS as _ { let data = *display_value.u.doubleValue(); Some(data as f32) } else { sysinfo_debug!("Query::get: PdhGetFormattedCounterValue failed"); Some(0.) }; } } None } #[allow(clippy::ptr_arg)] pub fn add_english_counter(&mut self, name: &String, getter: Vec) -> bool { if self.internal.data.contains_key(name) { sysinfo_debug!("Query::add_english_counter: doesn't have key `{:?}`", name); return false; } unsafe { let mut counter: PDH_HCOUNTER = std::mem::zeroed(); let ret = PdhAddEnglishCounterW(self.internal.query, getter.as_ptr(), 0, &mut counter); if ret == ERROR_SUCCESS as _ { self.internal.data.insert(name.clone(), counter); } else { sysinfo_debug!( "Query::add_english_counter: failed to add counter '{}': {:x}...", name, ret, ); return false; } } true } pub fn refresh(&self) { unsafe { if PdhCollectQueryData(self.internal.query) != ERROR_SUCCESS as _ { sysinfo_debug!("failed to refresh CPU data"); } } } } pub(crate) struct CpusWrapper { global: Cpu, cpus: Vec, got_cpu_frequency: bool, } impl CpusWrapper { pub fn new() -> Self { Self { global: Cpu::new_with_values("Total CPU".to_owned(), String::new(), String::new(), 0), cpus: Vec::new(), got_cpu_frequency: false, } } pub fn global_cpu(&self) -> &Cpu { &self.global } pub fn global_cpu_mut(&mut self) -> &mut Cpu { &mut self.global } pub fn cpus(&self) -> &[Cpu] { &self.cpus } fn init_if_needed(&mut self, refresh_kind: CpuRefreshKind) { if self.cpus.is_empty() { let (cpus, vendor_id, brand) = super::tools::init_cpus(refresh_kind); self.cpus = cpus; self.global.vendor_id = vendor_id; self.global.brand = brand; self.got_cpu_frequency = refresh_kind.frequency(); } } pub fn len(&mut self) -> usize { self.init_if_needed(CpuRefreshKind::new()); self.cpus.len() } pub fn iter_mut(&mut self, refresh_kind: CpuRefreshKind) -> impl Iterator { self.init_if_needed(refresh_kind); self.cpus.iter_mut() } pub fn get_frequencies(&mut self) { if self.got_cpu_frequency { return; } let frequencies = get_frequencies(self.cpus.len()); for (cpu, frequency) in self.cpus.iter_mut().zip(frequencies) { cpu.set_frequency(frequency); } self.got_cpu_frequency = true; } } #[doc = include_str!("../../md_doc/cpu.md")] pub struct Cpu { name: String, cpu_usage: f32, key_used: Option, vendor_id: String, brand: String, frequency: u64, } impl CpuExt for Cpu { fn cpu_usage(&self) -> f32 { self.cpu_usage } fn name(&self) -> &str { &self.name } fn frequency(&self) -> u64 { self.frequency } fn vendor_id(&self) -> &str { &self.vendor_id } fn brand(&self) -> &str { &self.brand } } impl Cpu { pub(crate) fn new_with_values( name: String, vendor_id: String, brand: String, frequency: u64, ) -> Cpu { Cpu { name, cpu_usage: 0f32, key_used: None, vendor_id, brand, frequency, } } pub(crate) fn set_cpu_usage(&mut self, value: f32) { self.cpu_usage = value; } pub(crate) fn set_frequency(&mut self, value: u64) { self.frequency = value; } } fn get_vendor_id_not_great(info: &SYSTEM_INFO) -> String { use winapi::um::winnt; // https://docs.microsoft.com/fr-fr/windows/win32/api/sysinfoapi/ns-sysinfoapi-system_info unsafe { match info.u.s().wProcessorArchitecture { winnt::PROCESSOR_ARCHITECTURE_INTEL => "Intel x86", winnt::PROCESSOR_ARCHITECTURE_MIPS => "MIPS", winnt::PROCESSOR_ARCHITECTURE_ALPHA => "RISC Alpha", winnt::PROCESSOR_ARCHITECTURE_PPC => "PPC", winnt::PROCESSOR_ARCHITECTURE_SHX => "SHX", winnt::PROCESSOR_ARCHITECTURE_ARM => "ARM", winnt::PROCESSOR_ARCHITECTURE_IA64 => "Intel Itanium-based x64", winnt::PROCESSOR_ARCHITECTURE_ALPHA64 => "RISC Alpha x64", winnt::PROCESSOR_ARCHITECTURE_MSIL => "MSIL", winnt::PROCESSOR_ARCHITECTURE_AMD64 => "(Intel or AMD) x64", winnt::PROCESSOR_ARCHITECTURE_IA32_ON_WIN64 => "Intel Itanium-based x86", winnt::PROCESSOR_ARCHITECTURE_NEUTRAL => "unknown", winnt::PROCESSOR_ARCHITECTURE_ARM64 => "ARM x64", winnt::PROCESSOR_ARCHITECTURE_ARM32_ON_WIN64 => "ARM", winnt::PROCESSOR_ARCHITECTURE_IA32_ON_ARM64 => "Intel Itanium-based x86", _ => "unknown", } .to_owned() } } #[cfg(any(target_arch = "x86", target_arch = "x86_64"))] pub(crate) fn get_vendor_id_and_brand(info: &SYSTEM_INFO) -> (String, String) { #[cfg(target_arch = "x86")] use std::arch::x86::__cpuid; #[cfg(target_arch = "x86_64")] use std::arch::x86_64::__cpuid; unsafe fn add_u32(v: &mut Vec, i: u32) { let i = &i as *const u32 as *const u8; v.push(*i); v.push(*i.offset(1)); v.push(*i.offset(2)); v.push(*i.offset(3)); } unsafe { // First, we try to get the complete name. let res = __cpuid(0x80000000); let n_ex_ids = res.eax; let brand = if n_ex_ids >= 0x80000004 { let mut extdata = Vec::with_capacity(5); for i in 0x80000000..=n_ex_ids { extdata.push(__cpuid(i)); } // 4 * u32 * nb_entries let mut out = Vec::with_capacity(4 * std::mem::size_of::() * 3); for data in extdata.iter().take(5).skip(2) { add_u32(&mut out, data.eax); add_u32(&mut out, data.ebx); add_u32(&mut out, data.ecx); add_u32(&mut out, data.edx); } let mut pos = 0; for e in out.iter() { if *e == 0 { break; } pos += 1; } match std::str::from_utf8(&out[..pos]) { Ok(s) => s.to_owned(), _ => String::new(), } } else { String::new() }; // Failed to get full name, let's retry for the short version! let res = __cpuid(0); let mut x = Vec::with_capacity(3 * std::mem::size_of::()); add_u32(&mut x, res.ebx); add_u32(&mut x, res.edx); add_u32(&mut x, res.ecx); let mut pos = 0; for e in x.iter() { if *e == 0 { break; } pos += 1; } let vendor_id = match std::str::from_utf8(&x[..pos]) { Ok(s) => s.to_owned(), Err(_) => get_vendor_id_not_great(info), }; (vendor_id, brand) } } #[cfg(all(not(target_arch = "x86_64"), not(target_arch = "x86")))] pub(crate) fn get_vendor_id_and_brand(info: &SYSTEM_INFO) -> (String, String) { (get_vendor_id_not_great(info), String::new()) } pub(crate) fn get_key_used(p: &mut Cpu) -> &mut Option { &mut p.key_used } // From https://stackoverflow.com/a/43813138: // // If your PC has 64 or fewer logical cpus installed, the above code will work fine. However, // if your PC has more than 64 logical cpus installed, use GetActiveCpuCount() or // GetLogicalCpuInformation() to determine the total number of logical cpus installed. pub(crate) fn get_frequencies(nb_cpus: usize) -> Vec { let size = nb_cpus * mem::size_of::(); let mut infos: Vec = Vec::with_capacity(nb_cpus); unsafe { if CallNtPowerInformation( ProcessorInformation, null_mut(), 0, infos.as_mut_ptr() as _, size as _, ) == 0 { infos.set_len(nb_cpus); // infos.Number return infos .into_iter() .map(|i| i.CurrentMhz as u64) .collect::>(); } } sysinfo_debug!("get_frequencies: CallNtPowerInformation failed"); vec![0; nb_cpus] } pub(crate) fn get_physical_core_count() -> Option { // we cannot use the number of cpus here to pre calculate the buf size // GetLogicalCpuInformationEx with RelationProcessorCore passed to it not only returns // the logical cores but also numa nodes // // GetLogicalCpuInformationEx: https://docs.microsoft.com/en-us/windows/win32/api/sysinfoapi/nf-sysinfoapi-getlogicalprocessorinformationex let mut needed_size = 0; unsafe { GetLogicalProcessorInformationEx(RelationAll, null_mut(), &mut needed_size); let mut buf: Vec = Vec::with_capacity(needed_size as _); loop { if GetLogicalProcessorInformationEx( RelationAll, buf.as_mut_ptr() as *mut _, &mut needed_size, ) == FALSE { let e = Error::last_os_error(); // For some reasons, the function might return a size not big enough... match e.raw_os_error() { Some(value) if value == ERROR_INSUFFICIENT_BUFFER as _ => {} _ => { sysinfo_debug!( "get_physical_core_count: GetLogicalCpuInformationEx failed" ); return None; } } } else { break; } buf.reserve(needed_size as usize - buf.capacity()); } buf.set_len(needed_size as _); let mut i = 0; let raw_buf = buf.as_ptr(); let mut count = 0; while i < buf.len() { let p = &*(raw_buf.add(i) as PSYSTEM_LOGICAL_PROCESSOR_INFORMATION_EX); i += p.Size as usize; if p.Relationship == RelationProcessorCore { // Only count the physical cores. count += 1; } } Some(count) } }