scheds: introduce scx_fair

A scheduler that focuses on ensuring fairness among tasks and
performance predictability.

It operates using a vruntime-based policy, where each task is assigned a
"latency" weight. This weight is dynamically adjusted based on how often
a task release the CPU before its full time slice is used. Tasks that
release the CPU early are given a higher latency weight, prioritizing
them over tasks that fully consume their time slice.

The combination of dynamic latency weights and vruntime-based scheduling
ensures responsive and consistent performance, even in overcommitted
systems.

This makes the scheduler particularly well-suited for workloads that
require multimedia or real-time audio processing.

Signed-off-by: Andrea Righi <[email protected]>

Cargo.toml

@@ -6,6 +6,7 @@ members = ["rust/scx_stats",
            "rust/scx_loader",
 	   "scheds/rust/scx_lavd",
            "scheds/rust/scx_bpfland",
+           "scheds/rust/scx_fair",
            "scheds/rust/scx_rustland",
            "scheds/rust/scx_rlfifo",
            "scheds/rust/scx_rusty",

meson-scripts/stress_tests.ini

@@ -22,6 +22,12 @@ sched_args: -v
 stress_cmd: stress-ng -t 14 --aggressive -M -c `nproc` -f `nproc`
 timeout_sec: 15
 
+[scx_fair]
+sched: scx_fair
+sched_args:
+stress_cmd: stress-ng -t 14 --aggressive -M -c `nproc` -f `nproc`
+timeout_sec: 15
+
 [scx_layered]
 sched: scx_layered
 sched_args: --run-example -v --stats 1

meson.build

@@ -327,7 +327,7 @@ if enable_rust
   run_target('fetch', command: [cargo_fetch, cargo], env: cargo_env)
 
   rust_scheds = ['scx_lavd', 'scx_bpfland', 'scx_rustland', 'scx_rlfifo',
-                 'scx_rusty',
+                 'scx_fair', 'scx_rusty',
                  'scx_layered', 'scx_mitosis']
   rust_misc = ['scx_stats', 'scx_stats_derive', 'scx_utils',
                'scx_rustland_core',

scheds/rust/README.md

@@ -18,3 +18,4 @@ main.rs or \*.bpf.c files.
 - [scx_rlfifo](scx_rlfifo/README.md)
 - [scx_lavd](scx_lavd/README.md)
 - [scx_bpfland](scx_bpfland/README.md)
+- [scx_fair](scx_fair/README.md)

scheds/rust/scx_fair/Cargo.toml

@@ -0,0 +1,26 @@
+[package]
+name = "scx_fair"
+version = "1.0.4"
+authors = ["Andrea Righi <[email protected]>"]
+edition = "2021"
+description = "A scheduler designed for multimedia and real-time audio processing workloads. https://github.com/sched-ext/scx/tree/main"
+license = "GPL-2.0-only"
+
+[dependencies]
+anyhow = "1.0.65"
+ctrlc = { version = "3.1", features = ["termination"] }
+clap = { version = "4.1", features = ["derive", "env", "unicode", "wrap_help"] }
+crossbeam = "0.8.4"
+libbpf-rs = "0.24.1"
+log = "0.4.17"
+scx_stats = { path = "../../../rust/scx_stats", version = "1.0.4" }
+scx_stats_derive = { path = "../../../rust/scx_stats/scx_stats_derive", version = "1.0.4" }
+scx_utils = { path = "../../../rust/scx_utils", version = "1.0.4" }
+serde = { version = "1.0", features = ["derive"] }
+simplelog = "0.12"
+
+[build-dependencies]
+scx_utils = { path = "../../../rust/scx_utils", version = "1.0.4" }
+
+[features]
+enable_backtrace = []

scheds/rust/scx_fair/LICENSE

@@ -0,0 +1 @@
+../../../LICENSE

scheds/rust/scx_fair/README.md

@@ -0,0 +1,26 @@
+# scx_fair
+
+This is a single user-defined scheduler used within [sched_ext](https://github.com/sched-ext/scx/tree/main), which is a Linux kernel feature which enables implementing kernel thread schedulers in BPF and dynamically loading them. [Read more about sched_ext](https://github.com/sched-ext/scx/tree/main).
+
+## Overview
+
+A scheduler that focuses on ensuring fairness among tasks and performance
+predictability.
+
+It operates using a vruntime-based policy, where each task is assigned a
+"latency" weight. This weight is dynamically adjusted based on how often a task
+release the CPU before its full time slice is used. Tasks that release the CPU
+early are given a higher latency weight, prioritizing them over tasks that
+fully consume their time slice.
+
+## Typical Use Case
+
+The combination of dynamic latency weights and vruntime-based scheduling
+ensures responsive and consistent performance, even in overcommitted systems.
+
+This makes the scheduler particularly well-suited for workloads that require
+multimedia or real-time audio processing.
+
+## Production Ready?
+
+Yes.

scheds/rust/scx_fair/build.rs

@@ -0,0 +1,13 @@
+// Copyright (c) Andrea Righi <[email protected]>
+//
+// This software may be used and distributed according to the terms of the
+// GNU General Public License version 2.
+
+fn main() {
+    scx_utils::BpfBuilder::new()
+        .unwrap()
+        .enable_intf("src/bpf/intf.h", "bpf_intf.rs")
+        .enable_skel("src/bpf/main.bpf.c", "bpf")
+        .build()
+        .unwrap();
+}

scheds/rust/scx_fair/rustfmt.toml

@@ -0,0 +1,8 @@
+# Get help on options with `rustfmt --help=config`
+# Please keep these in alphabetical order.
+edition = "2021"
+group_imports = "StdExternalCrate"
+imports_granularity = "Item"
+merge_derives = false
+use_field_init_shorthand = true
+version = "Two"

scheds/rust/scx_fair/src/bpf/intf.h

@@ -0,0 +1,42 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+/*
+ * Copyright (c) 2024 Andrea Righi <[email protected]>
+ *
+ * This software may be used and distributed according to the terms of the GNU
+ * General Public License version 2.
+ */
+#ifndef __INTF_H
+#define __INTF_H
+
+#include <limits.h>
+
+#define MAX(x, y) ((x) > (y) ? (x) : (y))
+#define MIN(x, y) ((x) < (y) ? (x) : (y))
+#define CLAMP(val, lo, hi) MIN(MAX(val, lo), hi)
+
+enum consts {
+	NSEC_PER_USEC = 1000ULL,
+	NSEC_PER_MSEC = (1000ULL * NSEC_PER_USEC),
+	NSEC_PER_SEC = (1000ULL * NSEC_PER_MSEC),
+};
+
+#ifndef __VMLINUX_H__
+typedef unsigned char u8;
+typedef unsigned short u16;
+typedef unsigned int u32;
+typedef unsigned long u64;
+
+typedef signed char s8;
+typedef signed short s16;
+typedef signed int s32;
+typedef signed long s64;
+
+typedef int pid_t;
+#endif /* __VMLINUX_H__ */
+
+struct domain_arg {
+	s32 cpu_id;
+	s32 sibling_cpu_id;
+};
+
+#endif /* __INTF_H */