In 2025, the Igalia Kernel team made extensive contributions across the Linux ecosystem. Over the year, Igalians authored 362 commits to Linux kernel releases published in 2025, 321 commits to the sched_ext schedulers and tools project—with a strong focus on the LAVD BPF scheduler—and additional contributions to projects such as igt-gpu-tools (11 commits), libfuse (six commits), and others.

This retrospective focuses on the two areas where our impact was most significant: upstream Linux kernel development and the sched_ext ecosystem.

Linux Kernel Contributions 2025

In 2025, Igalia continued its long-standing commitment to upstream Linux development, contributing extensively to the Linux Kernel across releases 6.13 through 6.18. Rather than isolated changes, this year stands out for sustained work with impact across multiple kernel subsystems, focused on performance, robustness, and enabling new capabilities—particularly in graphics, scheduling, memory management, and filesystems.

Below is a thematic retrospective of the most significant achievements of 2025.

sched_ext: Making BPF Schedulers Faster and Safer

In 2025, sched_ext became one of the most impactful kernel areas, delivering measurable performance gains while significantly improving correctness and robustness.

By enabling fully custom, BPF-based schedulers on top of the Linux scheduler core, sched_ext moved beyond the experimental stage. Igalia’s contributions focused on removing key performance bottlenecks, hardening critical execution paths, and improving observability — laying the foundation for the sched_ext ecosystem discussed further below.

• Hardening the core (6.13) Fixed lock misuse and missing RCU protections that could otherwise lead to kernel crashes, ensuring that sched_ext could be safely used and extended without risking system stability.
• Performance boost for BPF schedulers (6.14) One of the most impactful sched_ext contributions in 2025 addressed a clear performance bottleneck. Many BPF schedulers (such as scx_lavd) relied heavily on bpf_ktime_get_ns(), which on some architectures triggers expensive hardware TSC reads and degrades IPC. Igalia introduced scx_bpf_now(), reusing the scheduler runqueue clock when available and eliminating much of this overhead. The results were decisive:
  • 50–80% reduction in hardware TSC reads
  • ~76% reduction measured on real sched_ext workloads
• Improved introspection (6.15) Added internal event counters, giving scheduler developers great visibility into subtle runtime behavior. The events can be monitored through a BPF program, sysfs, and a tracepoint.

Filesystems: Case-Insensitivity and Container-Friendly Linux

Expanding case-insensitive filesystem support was a key user-facing theme in 2025, particularly for gaming and containerized workloads.

• Tmpfs case-insensitive directories (6.13) Igalia enabled casefolding support in tmpfs, allowing containers (e.g. Wine + Flatpak) to match host filesystem semantics. This removes the need for entire classes of userspace workarounds for Windows-centric applications and games.

• FUSE and VFS improvements (6.16) Added bulk dentry invalidation for FUSE servers, improving correctness and performance for user-space filesystems.

• Overlayfs case-insensitive merging (6.18) Overlayfs was enhanced to correctly merge directories on case-insensitive filesystems by introducing a casefold-aware comparison, benefiting SteamOS-style setups and containerized game environments, where overlayfs is fundamental. Together with the tmpfs change in 6.13, this significantly improves Linux compatibility with Windows-origin software stacks while keeping the solution cleanly in the kernel.

Graphics and GPUs: Performance, Stability, and Observability

GPU Memory and Performance Gains

• V3D Super Pages on Raspberry Pi (6.13) Support for 64 KB and 1 MB pages in the V3D MMU delivered measurable gains:
  • Up to 8% FPS improvement in real games
  • ~10% reduction in execution time for the respective Mesa CI job
  • Smoother emulation workloads
• Transparent Huge Pages for shmem (6.13) New kernel parameters (transparent_hugepage_shmem= and thp_shmem=) gave developers and system integrators fine-grained control over huge-page allocation, unlocking better GPU memory behavior for shmem-backed drivers.

GPU Reset Handling and User-Space Integration

Across multiple releases (6.15–6.18), Igalia substantially improved how GPU resets are detected, handled, and reported:

• DRM wedged events (6.15) introduced a generic user-space notification mechanism for GPU resets.
• Extended reset metadata (6.17) added PID and task name, enabling compositors to inform users which application caused the reset.
• False-positive timeout handling (6.17–6.18) prevented unnecessary resets by allowing drivers to report when a timeout did not represent a real GPU hang.

These changes directly improve desktop stability, user experience, and recovery behavior—especially on gaming systems and embedded GPUs.

Driver-Specific Improvements

• AMDGPU
  • Fixed long-standing race conditions in color management (notably KDE HDR/night-light flicker).
  • Improved plane handling, overlay support, VRR/MST stability.
  • Hardened ioctl paths, simplified locking, and fixed resource leaks on driver reload.
• Intel Xe
  • Expanded Alderlake support with missing hardware workarounds.
  • Improved correctness around dma-fence handling, eliminating use-after-free bugs.
• Raspberry Pi V3D
  • Major improvements to GPU reset reliability on RPi 4 and 5.
  • Added reset counters for robustness extensions.
  • Reduced lock contention and fixed race conditions in job queues.

DRM Scheduler: Foundations for the Future

The DRM scheduler received sustained attention throughout 2025:

• Performance optimizations (6.13-6.17)
  • Reduced locking overhead.
  • Optimized dma-fence merge paths to avoid allocations in common cases, lowering CPU usage on sync-heavy workloads (e.g. Vulkan compositors).
  • Added new dma-fence safe access rules and APIs which enable drivers to export fences shared via mechanism fixing an existing use-after-free condition.
• API cleanup and formalization (6.15–6.17) Igalia helped separate internal and external interfaces, reducing driver reliance on scheduler internals and preparing the ground for larger architectural improvements.
• Testing for critical infrastructure (6.16) Introduction of KUnit-based DRM scheduler tests marked a turning point, providing confidence for ongoing and future refactoring.

Memory Management and Core Kernel Improvements

Beyond graphics and scheduling, Igalia also contributed to core kernel subsystems:

• Out-of-Memory reporting (6.18) OOM diagnostics now include ZRAM and ZSWAP usage, making memory pressure analysis more accurate on modern systems.
• Allocation correctness fixes (6.18) Corrected handling of high-order allocations previously misclassified as atomic, restoring proper access to memory reserves.
• Clock / TSC fixes Improved the clock watchdog skew checks and fixed suspend/resume path when TSC is used as scheduler clock.

Reliability Through Testing and Bug Fixes

A constant background theme throughout 2025 was robustness:

• Numerous fixes driven by syzkaller/syzbot reports across filesystems, networking, memory management, media, HID, UVC, and suspend/resume paths.
• Extensive cleanups and refactoring of futex selftests (6.18), improving maintainability of one of Linux’s most performance-critical syscalls.
• Expanding coverage of misc device APIs with KUnit, specially around the use of static and dynamic minor numbers, which we also helped simplify.
• Contributions to linux-stable: not only is Igalia developing bug fixes upstream, but we also take care of ensuring relevant fixes get merged on stable kernels.

Linux Kernel Closing Thoughts

From tangible FPS gains on embedded GPUs, to safer schedulers, better diagnostics, and stronger filesystem semantics, Igalia’s 2025 contributions to the kernel made Linux faster, more predictable, and more user-friendly.

This kind of deep, upstream-focused work is what ensures the kernel can keep scaling, not just in features, but in quality.

See the full list of Igalia’s contributions for each kernel release: 6.13, 6.14, 6.15, 6.16, 6.17, 6.18.

---

BPF Scheduler Contributions 2025

While a large part of our work in 2025 landed directly in the upstream Linux kernel, there’s another project that deserves special attention. Throughout the year, Igalia has been a significant contributor to the sched_ext project, advancing schedulers, libraries, and tooling that allow solutions built on top of the kernel to make the most out of the infrastructure.

scx_lavd Maturity

With 228 commits, the Latency-Aware Virtual Deadline scheduler gained:

• Compute domain-aware load balancing and task stealing
• Cache-optimized data structures reducing coherence traffic
• Configurable preemption with IPI avoidance
• Futex tracing for lock-aware boosting

scx_lavd is now production-ready for latency-sensitive workloads.

CPU Bandwidth Control (cpu.max)

Implemented complete cgroup CPU bandwidth control, enabling hard CPU time limits for process groups, critical for Kubernetes, Docker, and multi-tenant cloud environments where resource isolation is mandatory.

The implementation includes a full library with quota management, throttling detection, replenish timers, and seamless integration with the scx_lavd scheduler.

Energy-Aware Scheduling

Introduced a comprehensive energy model framework for power-efficient scheduling on heterogeneous processors:

• Automatic big.LITTLE and Intel Hybrid detection
• CPU capacity and performance domain awareness
• Chip binning consideration for optimal core selection

This enables significant power savings on ARM servers (Graviton, Ampere) and modern Intel processors.

ARM64 Production Ready

Multiple fixes ensure sched_ext works reliably on ARM64:

• BPF verifier fixes for ARM64/clang combinations
• Portable implementations replacing architecture-specific builtins
• Multi-NUMA topology support

ARM servers can now run sched_ext schedulers in production.

Standardized Time API

We introduced scx_bpf_now() and migrated seven schedulers (scx_lavd, scx_bpfland, scx_flash, scx_layered, scx_mitosis, scx_rusty, scx_pair) to use consistent time handling, eliminating subtle timing bugs across platforms.

20% Smaller Binaries

Optimizing the release-tiny build profile achieved significant size reductions:

• scxtop: 20.6% smaller
• scx_bpfland: 14.0% smaller
• scx_flash: 13.8% smaller

Smaller binaries mean faster deployment and better suitability for embedded systems.

scx Closing Thoughts

sched_ext represents a major shift in how Linux scheduling can evolve. In 2025, Igalia focused on making it practical and production-ready—from kernel hardening to real-world schedulers, energy awareness, and cloud-grade resource control.

As adoption grows in 2026, we look forward to continuing this work with the broader community.

---

Final Remarks

Igalia’s 2025 kernel and sched_ext contributions demonstrate the value of sustained upstream work paired with practical engineering. By improving core kernel subsystems and enabling new scheduling models through sched_ext, these efforts help Linux remain performant, reliable, and adaptable at scale.

We look forward to continuing this work with the community in the years ahead.