Linux Kernel Daily: The RC3 Bounds-Checking Sweep
This late-cycle merge window was dominated by defensive fixes—buffer and tile-count validation in media decoders, buffer validation in GPU drivers, and a reverted "fix" in RISC-V that had confused two unrelated pointer-masking features. The common thread is hardening code paths against malformed input and bad prior patches, not new functionality.
Duration: PT2M32S
Episode overview
This episode is a short developer briefing from Linux Kernel Daily.
It explains recent repository work in plain language.
- Show: Linux Kernel Daily
- Published: 2026-09-12T06:01:14Z
- Audio duration: PT2M32S
Transcript excerpt
This excerpt keeps the crawler page concise. Listen to the episode or use the RSS feed for the full update.
Good day. It's September 12th, 2026, and this is Linux Kernel Daily.
The signal today is bounds-checking. Across three unrelated subsystems, maintainers landed fixes that share one purpose: stopping malformed data from overflowing buffers before it causes damage.
Start with media. Mauro Chehab's pull, landed in commit d5d6c9d, adds bounded tile-count helpers for HEVC decoders and validates that AV1 and HEVC tile counts actually fit their arrays. Rockchip, Hantro, and rkvdec drivers all got matching guards on tile loops and PPS IDs. This is the same class of bug showing up…
The GPU side echoes this. Dave Airlie's DRM fixes pull, commit 525f0f9, includes ivpu validating firmware log buffers and address translation ranges, plus ethosu fixes for command stream formatting and SRAM sizing. Different vendors, same category of fix: validate the buffer before you trust it.
The second theme is reverting bad fixes. Paul Walmsley's RISC-V pull, commit 827751b, calls out a patch — likely LLM-generated — that conflated RISC-V's hardware pointer masking with the kernel's tagged address feature. It's reverted outright. Worth remembering: a plausible-looking fix that misunderstands the…
Elsewhere,…