servo/components/profile/mem.rs

/* This Source Code Form is subject to the terms of the Mozilla Public
 * License, v. 2.0. If a copy of the MPL was not distributed with this
 * file, You can obtain one at https://mozilla.org/MPL/2.0/. */

//! Memory profiling functions.

use std::borrow::ToOwned;
use std::cmp::Ordering;
use std::collections::HashMap;
use std::thread;
use std::time::{Duration, Instant};

use ipc_channel::ipc::{self, IpcReceiver};
use ipc_channel::router::ROUTER;
use profile_traits::mem::{
    ProfilerChan, ProfilerMsg, ReportKind, Reporter, ReporterRequest, ReportsChan,
};
use profile_traits::path;

pub struct Profiler {
    /// The port through which messages are received.
    pub port: IpcReceiver<ProfilerMsg>,

    /// Registered memory reporters.
    reporters: HashMap<String, Reporter>,

    /// Instant at which this profiler was created.
    created: Instant,
}

const JEMALLOC_HEAP_ALLOCATED_STR: &str = "jemalloc-heap-allocated";
const SYSTEM_HEAP_ALLOCATED_STR: &str = "system-heap-allocated";

impl Profiler {
    pub fn create(period: Option<f64>) -> ProfilerChan {
        let (chan, port) = ipc::channel().unwrap();

        // Create the timer thread if a period was provided.
        if let Some(period) = period {
            let chan = chan.clone();
            thread::Builder::new()
                .name("MemoryProfTimer".to_owned())
                .spawn(move || loop {
                    thread::sleep(Duration::from_secs_f64(period));
                    if chan.send(ProfilerMsg::Print).is_err() {
                        break;
                    }
                })
                .expect("Thread spawning failed");
        }

        // Always spawn the memory profiler. If there is no timer thread it won't receive regular
        // `Print` events, but it will still receive the other events.
        thread::Builder::new()
            .name("MemoryProfiler".to_owned())
            .spawn(move || {
                let mut mem_profiler = Profiler::new(port);
                mem_profiler.start();
            })
            .expect("Thread spawning failed");

        let mem_profiler_chan = ProfilerChan(chan);

        // Register the system memory reporter, which will run on its own thread. It never needs to
        // be unregistered, because as long as the memory profiler is running the system memory
        // reporter can make measurements.
        let (system_reporter_sender, system_reporter_receiver) = ipc::channel().unwrap();
        ROUTER.add_route(
            system_reporter_receiver.to_opaque(),
            Box::new(|message| {
                let request: ReporterRequest = message.to().unwrap();
                system_reporter::collect_reports(request)
            }),
        );
        mem_profiler_chan.send(ProfilerMsg::RegisterReporter(
            "system".to_owned(),
            Reporter(system_reporter_sender),
        ));

        mem_profiler_chan
    }

    pub fn new(port: IpcReceiver<ProfilerMsg>) -> Profiler {
        Profiler {
            port,
            reporters: HashMap::new(),
            created: Instant::now(),
        }
    }

    pub fn start(&mut self) {
        while let Ok(msg) = self.port.recv() {
            if !self.handle_msg(msg) {
                break;
            }
        }
    }

    fn handle_msg(&mut self, msg: ProfilerMsg) -> bool {
        match msg {
            ProfilerMsg::RegisterReporter(name, reporter) => {
                // Panic if it has already been registered.
                let name_clone = name.clone();
                match self.reporters.insert(name, reporter) {
                    None => true,
                    Some(_) => panic!("RegisterReporter: '{}' name is already in use", name_clone),
                }
            },

            ProfilerMsg::UnregisterReporter(name) => {
                // Panic if it hasn't previously been registered.
                match self.reporters.remove(&name) {
                    Some(_) => true,
                    None => panic!("UnregisterReporter: '{}' name is unknown", &name),
                }
            },

            ProfilerMsg::Print => {
                self.handle_print_msg();
                true
            },

            ProfilerMsg::Exit => false,
        }
    }

    fn handle_print_msg(&self) {
        let elapsed = self.created.elapsed();
        println!("Begin memory reports {}", elapsed.as_secs());
        println!("|");

        // Collect reports from memory reporters.
        //
        // This serializes the report-gathering. It might be worth creating a new scoped thread for
        // each reporter once we have enough of them.
        //
        // If anything goes wrong with a reporter, we just skip it.
        //
        // We also track the total memory reported on the jemalloc heap and the system heap, and
        // use that to compute the special "jemalloc-heap-unclassified" and
        // "system-heap-unclassified" values.

        let mut forest = ReportsForest::new();

        let mut jemalloc_heap_reported_size = 0;
        let mut system_heap_reported_size = 0;

        let mut jemalloc_heap_allocated_size: Option<usize> = None;
        let mut system_heap_allocated_size: Option<usize> = None;

        for reporter in self.reporters.values() {
            let (chan, port) = ipc::channel().unwrap();
            reporter.collect_reports(ReportsChan(chan));
            if let Ok(mut reports) = port.recv() {
                for report in &mut reports {
                    // Add "explicit" to the start of the path, when appropriate.
                    match report.kind {
                        ReportKind::ExplicitJemallocHeapSize |
                        ReportKind::ExplicitSystemHeapSize |
                        ReportKind::ExplicitNonHeapSize |
                        ReportKind::ExplicitUnknownLocationSize => {
                            report.path.insert(0, String::from("explicit"))
                        },
                        ReportKind::NonExplicitSize => {},
                    }

                    // Update the reported fractions of the heaps, when appropriate.
                    match report.kind {
                        ReportKind::ExplicitJemallocHeapSize => {
                            jemalloc_heap_reported_size += report.size
                        },
                        ReportKind::ExplicitSystemHeapSize => {
                            system_heap_reported_size += report.size
                        },
                        _ => {},
                    }

                    // Record total size of the heaps, when we see them.
                    if report.path.len() == 1 {
                        if report.path[0] == JEMALLOC_HEAP_ALLOCATED_STR {
                            assert!(jemalloc_heap_allocated_size.is_none());
                            jemalloc_heap_allocated_size = Some(report.size);
                        } else if report.path[0] == SYSTEM_HEAP_ALLOCATED_STR {
                            assert!(system_heap_allocated_size.is_none());
                            system_heap_allocated_size = Some(report.size);
                        }
                    }

                    // Insert the report.
                    forest.insert(&report.path, report.size);
                }
            }
        }

        // Compute and insert the heap-unclassified values.
        if let Some(jemalloc_heap_allocated_size) = jemalloc_heap_allocated_size {
            forest.insert(
                &path!["explicit", "jemalloc-heap-unclassified"],
                jemalloc_heap_allocated_size - jemalloc_heap_reported_size,
            );
        }
        if let Some(system_heap_allocated_size) = system_heap_allocated_size {
            forest.insert(
                &path!["explicit", "system-heap-unclassified"],
                system_heap_allocated_size - system_heap_reported_size,
            );
        }

        forest.print();

        println!("|");
        println!("End memory reports");
        println!();
    }
}

/// A collection of one or more reports with the same initial path segment. A ReportsTree
/// containing a single node is described as "degenerate".
struct ReportsTree {
    /// For leaf nodes, this is the sum of the sizes of all reports that mapped to this location.
    /// For interior nodes, this is the sum of the sizes of all its child nodes.
    size: usize,

    /// For leaf nodes, this is the count of all reports that mapped to this location.
    /// For interor nodes, this is always zero.
    count: u32,

    /// The segment from the report path that maps to this node.
    path_seg: String,

    /// Child nodes.
    children: Vec<ReportsTree>,
}

impl ReportsTree {
    fn new(path_seg: String) -> ReportsTree {
        ReportsTree {
            size: 0,
            count: 0,
            path_seg,
            children: vec![],
        }
    }

    // Searches the tree's children for a path_seg match, and returns the index if there is a
    // match.
    fn find_child(&self, path_seg: &str) -> Option<usize> {
        for (i, child) in self.children.iter().enumerate() {
            if child.path_seg == *path_seg {
                return Some(i);
            }
        }
        None
    }

    // Insert the path and size into the tree, adding any nodes as necessary.
    fn insert(&mut self, path: &[String], size: usize) {
        let mut t: &mut ReportsTree = self;
        for path_seg in path {
            let i = match t.find_child(path_seg) {
                Some(i) => i,
                None => {
                    let new_t = ReportsTree::new(path_seg.clone());
                    t.children.push(new_t);
                    t.children.len() - 1
                },
            };
            let tmp = t; // this temporary is needed to satisfy the borrow checker
            t = &mut tmp.children[i];
        }

        t.size += size;
        t.count += 1;
    }

    // Fill in sizes for interior nodes and sort sub-trees accordingly. Should only be done once
    // all the reports have been inserted.
    fn compute_interior_node_sizes_and_sort(&mut self) -> usize {
        if !self.children.is_empty() {
            // Interior node. Derive its size from its children.
            if self.size != 0 {
                // This will occur if e.g. we have paths ["a", "b"] and ["a", "b", "c"].
                panic!("one report's path is a sub-path of another report's path");
            }
            for child in &mut self.children {
                self.size += child.compute_interior_node_sizes_and_sort();
            }
            // Now that child sizes have been computed, we can sort the children.
            self.children.sort_by(|t1, t2| t2.size.cmp(&t1.size));
        }
        self.size
    }

    fn print(&self, depth: i32) {
        if !self.children.is_empty() {
            assert_eq!(self.count, 0);
        }

        let mut indent_str = String::new();
        for _ in 0..depth {
            indent_str.push_str("   ");
        }

        let mebi = 1024f64 * 1024f64;
        let count_str = if self.count > 1 {
            format!(" [{}]", self.count)
        } else {
            "".to_owned()
        };
        println!(
            "|{}{:8.2} MiB -- {}{}",
            indent_str,
            (self.size as f64) / mebi,
            self.path_seg,
            count_str
        );

        for child in &self.children {
            child.print(depth + 1);
        }
    }
}

/// A collection of ReportsTrees. It represents the data from multiple memory reports in a form
/// that's good to print.
struct ReportsForest {
    trees: HashMap<String, ReportsTree>,
}

impl ReportsForest {
    fn new() -> ReportsForest {
        ReportsForest {
            trees: HashMap::new(),
        }
    }

    // Insert the path and size into the forest, adding any trees and nodes as necessary.
    fn insert(&mut self, path: &[String], size: usize) {
        let (head, tail) = path.split_first().unwrap();
        // Get the right tree, creating it if necessary.
        if !self.trees.contains_key(head) {
            self.trees
                .insert(head.clone(), ReportsTree::new(head.clone()));
        }
        let t = self.trees.get_mut(head).unwrap();

        // Use tail because the 0th path segment was used to find the right tree in the forest.
        t.insert(tail, size);
    }

    fn print(&mut self) {
        // Fill in sizes of interior nodes, and recursively sort the sub-trees.
        for tree in self.trees.values_mut() {
            tree.compute_interior_node_sizes_and_sort();
        }

        // Put the trees into a sorted vector. Primary sort: degenerate trees (those containing a
        // single node) come after non-degenerate trees. Secondary sort: alphabetical order of the
        // root node's path_seg.
        let mut v = vec![];
        for tree in self.trees.values() {
            v.push(tree);
        }
        v.sort_by(|a, b| {
            if a.children.is_empty() && !b.children.is_empty() {
                Ordering::Greater
            } else if !a.children.is_empty() && b.children.is_empty() {
                Ordering::Less
            } else {
                a.path_seg.cmp(&b.path_seg)
            }
        });

        // Print the forest.
        for tree in &v {
            tree.print(0);
            // Print a blank line after non-degenerate trees.
            if !tree.children.is_empty() {
                println!("|");
            }
        }
    }
}

//---------------------------------------------------------------------------

mod system_reporter {
    #[cfg(not(any(target_os = "windows", target_env = "ohos")))]
    use std::ffi::CString;
    #[cfg(not(any(target_os = "windows", target_env = "ohos")))]
    use std::mem::size_of;
    #[cfg(not(any(target_os = "windows", target_env = "ohos")))]
    use std::ptr::null_mut;

    #[cfg(all(target_os = "linux", target_env = "gnu"))]
    use libc::c_int;
    #[cfg(not(any(target_os = "windows", target_env = "ohos")))]
    use libc::{c_void, size_t};
    use profile_traits::mem::{Report, ReportKind, ReporterRequest};
    use profile_traits::path;
    #[cfg(target_os = "macos")]
    use task_info::task_basic_info::{resident_size, virtual_size};

    use super::{JEMALLOC_HEAP_ALLOCATED_STR, SYSTEM_HEAP_ALLOCATED_STR};

    /// Collects global measurements from the OS and heap allocators.
    pub fn collect_reports(request: ReporterRequest) {
        let mut reports = vec![];
        {
            let mut report = |path, size| {
                if let Some(size) = size {
                    reports.push(Report {
                        path,
                        kind: ReportKind::NonExplicitSize,
                        size,
                    });
                }
            };

            // Virtual and physical memory usage, as reported by the OS.
            report(path!["vsize"], vsize());
            report(path!["resident"], resident());

            // Memory segments, as reported by the OS.
            for seg in resident_segments() {
                report(path!["resident-according-to-smaps", seg.0], Some(seg.1));
            }

            // Total number of bytes allocated by the application on the system
            // heap.
            report(path![SYSTEM_HEAP_ALLOCATED_STR], system_heap_allocated());

            // The descriptions of the following jemalloc measurements are taken
            // directly from the jemalloc documentation.

            // "Total number of bytes allocated by the application."
            report(
                path![JEMALLOC_HEAP_ALLOCATED_STR],
                jemalloc_stat("stats.allocated"),
            );

            // "Total number of bytes in active pages allocated by the application.
            // This is a multiple of the page size, and greater than or equal to
            // |stats.allocated|."
            report(path!["jemalloc-heap-active"], jemalloc_stat("stats.active"));

            // "Total number of bytes in chunks mapped on behalf of the application.
            // This is a multiple of the chunk size, and is at least as large as
            // |stats.active|. This does not include inactive chunks."
            report(path!["jemalloc-heap-mapped"], jemalloc_stat("stats.mapped"));
        }

        request.reports_channel.send(reports);
    }

    #[cfg(all(target_os = "linux", target_env = "gnu"))]
    extern "C" {
        fn mallinfo() -> struct_mallinfo;
    }

    #[cfg(all(target_os = "linux", target_env = "gnu"))]
    #[repr(C)]
    pub struct struct_mallinfo {
        arena: c_int,
        ordblks: c_int,
        smblks: c_int,
        hblks: c_int,
        hblkhd: c_int,
        usmblks: c_int,
        fsmblks: c_int,
        uordblks: c_int,
        fordblks: c_int,
        keepcost: c_int,
    }

    #[cfg(all(target_os = "linux", target_env = "gnu"))]
    fn system_heap_allocated() -> Option<usize> {
        let info: struct_mallinfo = unsafe { mallinfo() };

        // The documentation in the glibc man page makes it sound like |uordblks| would suffice,
        // but that only gets the small allocations that are put in the brk heap. We need |hblkhd|
        // as well to get the larger allocations that are mmapped.
        //
        // These fields are unfortunately |int| and so can overflow (becoming negative) if memory
        // usage gets high enough. So don't report anything in that case. In the non-overflow case
        // we cast the two values to usize before adding them to make sure the sum also doesn't
        // overflow.
        if info.hblkhd < 0 || info.uordblks < 0 {
            None
        } else {
            Some(info.hblkhd as usize + info.uordblks as usize)
        }
    }

    #[cfg(not(all(target_os = "linux", target_env = "gnu")))]
    fn system_heap_allocated() -> Option<usize> {
        None
    }

    #[cfg(not(any(target_os = "windows", target_env = "ohos")))]
    use tikv_jemalloc_sys::mallctl;

    #[cfg(not(any(target_os = "windows", target_env = "ohos")))]
    fn jemalloc_stat(value_name: &str) -> Option<usize> {
        // Before we request the measurement of interest, we first send an "epoch"
        // request. Without that jemalloc gives cached statistics(!) which can be
        // highly inaccurate.
        let epoch_name = "epoch";
        let epoch_c_name = CString::new(epoch_name).unwrap();
        let mut epoch: u64 = 0;
        let epoch_ptr = &mut epoch as *mut _ as *mut c_void;
        let mut epoch_len = size_of::<u64>() as size_t;

        let value_c_name = CString::new(value_name).unwrap();
        let mut value: size_t = 0;
        let value_ptr = &mut value as *mut _ as *mut c_void;
        let mut value_len = size_of::<size_t>() as size_t;

        // Using the same values for the `old` and `new` parameters is enough
        // to get the statistics updated.
        let rv = unsafe {
            mallctl(
                epoch_c_name.as_ptr(),
                epoch_ptr,
                &mut epoch_len,
                epoch_ptr,
                epoch_len,
            )
        };
        if rv != 0 {
            return None;
        }

        let rv = unsafe {
            mallctl(
                value_c_name.as_ptr(),
                value_ptr,
                &mut value_len,
                null_mut(),
                0,
            )
        };
        if rv != 0 {
            return None;
        }

        Some(value as usize)
    }

    #[cfg(any(target_os = "windows", target_env = "ohos"))]
    fn jemalloc_stat(_value_name: &str) -> Option<usize> {
        None
    }

    #[cfg(target_os = "linux")]
    fn page_size() -> usize {
        unsafe { ::libc::sysconf(::libc::_SC_PAGESIZE) as usize }
    }

    #[cfg(target_os = "linux")]
    fn proc_self_statm_field(field: usize) -> Option<usize> {
        use std::fs::File;
        use std::io::Read;

        let mut f = File::open("/proc/self/statm").ok()?;
        let mut contents = String::new();
        f.read_to_string(&mut contents).ok()?;
        let s = contents.split_whitespace().nth(field)?;
        let npages = s.parse::<usize>().ok()?;
        Some(npages * page_size())
    }

    #[cfg(target_os = "linux")]
    fn vsize() -> Option<usize> {
        proc_self_statm_field(0)
    }

    #[cfg(target_os = "linux")]
    fn resident() -> Option<usize> {
        proc_self_statm_field(1)
    }

    #[cfg(target_os = "macos")]
    fn vsize() -> Option<usize> {
        virtual_size()
    }

    #[cfg(target_os = "macos")]
    fn resident() -> Option<usize> {
        resident_size()
    }

    #[cfg(not(any(target_os = "linux", target_os = "macos")))]
    fn vsize() -> Option<usize> {
        None
    }

    #[cfg(not(any(target_os = "linux", target_os = "macos")))]
    fn resident() -> Option<usize> {
        None
    }

    #[cfg(target_os = "linux")]
    fn resident_segments() -> Vec<(String, usize)> {
        use std::collections::hash_map::Entry;
        use std::collections::HashMap;
        use std::fs::File;
        use std::io::{BufRead, BufReader};

        use regex::Regex;

        // The first line of an entry in /proc/<pid>/smaps looks just like an entry
        // in /proc/<pid>/maps:
        //
        //   address           perms offset  dev   inode  pathname
        //   02366000-025d8000 rw-p 00000000 00:00 0      [heap]
        //
        // Each of the following lines contains a key and a value, separated
        // by ": ", where the key does not contain either of those characters.
        // For example:
        //
        //   Rss:           132 kB

        let f = match File::open("/proc/self/smaps") {
            Ok(f) => BufReader::new(f),
            Err(_) => return vec![],
        };

        let seg_re = Regex::new(
            r"^[:xdigit:]+-[:xdigit:]+ (....) [:xdigit:]+ [:xdigit:]+:[:xdigit:]+ \d+ +(.*)",
        )
        .unwrap();
        let rss_re = Regex::new(r"^Rss: +(\d+) kB").unwrap();

        // We record each segment's resident size.
        let mut seg_map: HashMap<String, usize> = HashMap::new();

        #[derive(PartialEq)]
        enum LookingFor {
            Segment,
            Rss,
        }
        let mut looking_for = LookingFor::Segment;

        let mut curr_seg_name = String::new();

        // Parse the file.
        for line in f.lines() {
            let line = match line {
                Ok(line) => line,
                Err(_) => continue,
            };
            if looking_for == LookingFor::Segment {
                // Look for a segment info line.
                let cap = match seg_re.captures(&line) {
                    Some(cap) => cap,
                    None => continue,
                };
                let perms = cap.get(1).unwrap().as_str();
                let pathname = cap.get(2).unwrap().as_str();

                // Construct the segment name from its pathname and permissions.
                curr_seg_name.clear();
                if pathname.is_empty() || pathname.starts_with("[stack:") {
                    // Anonymous memory. Entries marked with "[stack:nnn]"
                    // look like thread stacks but they may include other
                    // anonymous mappings, so we can't trust them and just
                    // treat them as entirely anonymous.
                    curr_seg_name.push_str("anonymous");
                } else {
                    curr_seg_name.push_str(pathname);
                }
                curr_seg_name.push_str(" (");
                curr_seg_name.push_str(perms);
                curr_seg_name.push(')');

                looking_for = LookingFor::Rss;
            } else {
                // Look for an "Rss:" line.
                let cap = match rss_re.captures(&line) {
                    Some(cap) => cap,
                    None => continue,
                };
                let rss = cap.get(1).unwrap().as_str().parse::<usize>().unwrap() * 1024;

                if rss > 0 {
                    // Aggregate small segments into "other".
                    let seg_name = if rss < 512 * 1024 {
                        "other".to_owned()
                    } else {
                        curr_seg_name.clone()
                    };
                    match seg_map.entry(seg_name) {
                        Entry::Vacant(entry) => {
                            entry.insert(rss);
                        },
                        Entry::Occupied(mut entry) => *entry.get_mut() += rss,
                    }
                }

                looking_for = LookingFor::Segment;
            }
        }

        // Note that the sum of all these segments' RSS values differs from the "resident"
        // measurement obtained via /proc/<pid>/statm in resident(). It's unclear why this
        // difference occurs; for some processes the measurements match, but for Servo they do not.
        seg_map.into_iter().collect()
    }

    #[cfg(not(target_os = "linux"))]
    fn resident_segments() -> Vec<(String, usize)> {
        vec![]
    }
}