--- title: Linux 7.3: what's actually changing in the kernel description: The Linux 7.3 merge window closed with a wide-ranging set of kernel changes: scheduler fixes reducing jitter and improving Cache Aware Load Scheduling on... canonical: https://daily.dev/posts/linux-7-3-what-s-actually-changing-in-the-kernel-wdkvq9rpa twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: Linux 7.3: what's actually changing in the kernel | daily.dev og:description: The Linux 7.3 merge window closed with a wide-ranging set of kernel changes: scheduler fixes reducing jitter and improving Cache Aware Load Scheduling on... og:url: https://daily.dev/posts/linux-7-3-what-s-actually-changing-in-the-kernel-wdkvq9rpa og:image: https://api.daily.dev/og/posts/wdkvq9rPA.png og:image:alt: Linux 7.3: what's actually changing in the kernel og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # Linux 7.3: what's actually changing in the kernel **[Collections](https://daily.dev/sources/collections)** · 5 min read · 10 upvotes · 0 comments ## Summary The Linux 7.3 merge window closed with a wide-ranging set of kernel changes: scheduler fixes reducing jitter and improving Cache Aware Load Scheduling on hybrid CPUs, a rewritten Btrfs direct I/O path yielding 3-5x throughput on some workloads, early Apple M3 Pro/Max/Ultra support, VRAM eviction improvements from a Valve engineer, and progress on the Rust-based Nova Nvidia driver including GSP firmware boot support and Hopper/Blackwell compatibility. Rust kernel support also expands to PowerPC. Deprecations begin for old 32-bit ARM platforms, SGI, EFS, and free-VxFS drivers. Greg KH also flagged that AI-generated low-quality patch submissions are overwhelming kernel review teams, with CVE counts approaching 2,000 for this release. ## Content ## What's in Linux 7.3 The 7.3 merge window closed with a lot going on. Here's a rundown of the significant changes, followed by the messier story of how this release cycle has been going. ### Btrfs direct I/O rewrite Btrfs gets a rewritten direct I/O path that delivers 3–5x throughput improvements on some workloads. That's a meaningful jump for anyone using Btrfs on storage-heavy workloads. ### Scheduler improvements Two scheduler changes stand out. First, a "flatten the pick" update and NOHZ balancing improvements reduce rescheduling jitter, which matters more than it sounds. You won't see higher maximum frame rates, but average and minimum FPS improve — enough to push some games from unplayable to playable on older hardware like the RX 580. Low-end and aging components benefit most. Second, a fix from Intel engineer Tim Chen addresses a specific problem with Cache Aware Load Scheduling on hybrid CPUs (big.LITTLE-style designs). The cache-aware scheduler was migrating tasks to last-level cache domains whose CPUs were too small to run them efficiently. The fix prevents that migration and prioritizes getting tasks onto properly sized cores over cache locality gains, since running on the right core matters more than cache proximity. The fix only applies to asymmetric CPU systems, so symmetric setups are unaffected. ### GPU work AMD's DCN6 register dump accounts for roughly a third of RC1's size — a large but mostly mechanical addition. More interesting: VRAM eviction handling improves for memory-constrained GPUs, work contributed by Valve's Natalie Vog. The Nova Rust-based Nvidia driver adds GSP boot support and initial Hopper/Blackwell compatibility. On the Nouveau side, Linux 7.3-rc2 includes display fixes for Nvidia Blackwell hardware. ### Apple Silicon Initial support lands for Apple M3 Pro, Max, and Ultra chips. ### Memory management TLB-flush and KSM latency spikes drop by up to 90% with memory management fixes in this cycle. That's the kind of improvement that shows up as reduced stuttering rather than benchmark numbers. ### USB-C USB-C cables now report actual alt-mode capabilities rather than guessing. Small change, real-world impact for anyone debugging USB-C display or Thunderbolt issues. ### Rust expansion Rust kernel support extends to PowerPC, and early work on GCC backend compatibility has started. The kernel's Rust story is moving faster than many expected. ### Deprecations and removals Old 32-bit ARM platforms and SGI/EFS/free-VxFS drivers are beginning deprecation and removal. The kernel is slowly shedding legacy weight. --- ## The kmalloc_obj() cleanup Linux 7.3-rc2 pulls in a tree-wide conversion of remaining `kmalloc()` calls to the newer `kmalloc_obj()` family, run by kernel developer Kees Cook using a Coccinelle script. The change touches hundreds of files and over 600 lines of code. `kmalloc_obj()` was introduced in Linux 7.0. The practical improvements: it removes the need for `sizeof()` and explicit type names, enforces correct assignment types instead of returning `void*`, and enables internal type introspection. That last part opens the door to future alignment-aware allocation and type-sensitive hardening — security benefits that don't exist yet but become possible once the infrastructure is in place. --- ## EROFS disables LZ4 rolling decompression Ahead of rc3, the EROFS read-only filesystem disabled its LZ4 rolling decompression optimization after AWS engineers found that certain rare LZ4 datasets could cause data corruption. The root cause is uncontrolled backward memory copies in the upstream kernel LZ4 implementation. Rolling decompression assumes literal copies always move forward for non-overlapping ranges — the LZ4 code doesn't guarantee that. Since EROFS doesn't own the kernel LZ4 codebase, the developers chose to disable the optimization entirely rather than ship a known corruption risk. The cost is a higher runtime memory footprint. A recent reserved buffer pool commit partially offsets that. The optimization could return once the LZ4 implementation is fixed or EROFS maintains its own copy. This is the right call — data correctness beats performance — but it's worth knowing if you're running EROFS in production and wondering why memory usage ticked up. --- ## RandStruct disabled when Rust is present RandStruct, a security feature that randomizes kernel struct layouts, is now disabled by default when Rust support is enabled. This is a compatibility workaround rather than a deliberate security tradeoff — the two features conflict in the current implementation. Worth tracking if you care about kernel hardening. --- ## The AI patch problem This is the part of the 7.3 cycle that's generating the most discussion outside of technical circles. Kernel maintainer Greg Kroah-Hartman flagged that AI-generated low-quality patch submissions are overwhelming review teams. CVE counts for this release are approaching 2,000. Linus Torvalds, releasing rc2, noted it was unusually large for a post-merge-window lull — 615 commits from 292 contributors — and jokingly blamed AI, though he admitted he couldn't pin down a real cause. A month earlier, discussing 7.2-rc7, he gave a more serious explanation, attributing the size to fixes generated by AI review tools. The kernel already has an `Assisted-by: LLM` commit trailer policy meant to track AI involvement, but it's voluntary and nobody has actually used it to quantify AI's role in rc2. So the data doesn't exist to say definitively how much of the churn is AI-generated versus just a busy cycle. Debian recently voted to allow AI-assisted contributions with optional disclosure — essentially the opposite approach to provenance tracking. The contrast is interesting: one community trying to label AI involvement, another making disclosure optional. There's something genuinely uncomfortable about the situation. Review bandwidth is finite. If a meaningful fraction of submissions are low-quality AI-generated patches that require human time to reject, that's a real tax on the people keeping the kernel running. The joke about blaming AI lands because everyone suspects it's at least partially true, and nobody has the data to say otherwise. --- ## Release timeline Linux 7.3 stable is expected in the second half of October. The rc cycle continues at kernel.org. ## Questions this post answers ### What does the Btrfs direct I/O rewrite in Linux 7.3 change for throughput? Linux 7.3 ships a rewritten Btrfs direct I/O path that improves throughput by 3-5x on some workloads. The gains are most noticeable for databases and certain VM setups that rely heavily on direct I/O, while the change is largely invisible for typical desktop or general-purpose use. _daily.dev helps engineers tuning database or VM storage performance keep up with kernel I/O changes like this._ ### What did the Linux 7.3 scheduler fix for Cache Aware Load Scheduling actually correct? Linux 7.3 includes a fix, written by Intel engineer Tim Chen, that stops the cache-aware scheduler from migrating tasks to a last-level cache domain whose CPUs are too small to run them efficiently on big.LITTLE-style hybrid CPUs. It guards both the LLC migration path and active load balancing, and reprioritizes fixing misfit tasks over cache locality gains. The fix only affects asymmetric CPU systems. _engineers debugging hybrid-CPU scheduling quirks can track kernel scheduler patches like this on daily.dev._ ### Why are AI-generated patches becoming a problem for the Linux kernel maintainers? Kernel maintainer Greg KH flagged that AI-generated low-quality patch submissions are overwhelming kernel review teams, adding significant triage noise for maintainers. CVE counts tied to the Linux 7.3 release are approaching 2,000, reflecting the growing volume maintainers now have to sift through, separate from any actual technical defects in the kernel. _daily.dev keeps maintainers and contributors informed on how AI-generated code is reshaping open source review workflows._ ## Similar posts on daily.dev - [Linux's sched\_ext Sees A Bunch Of Bug Fixes Following Increased AI Code Review](https://daily.dev/posts/linux-s-sched-ext-sees-a-bunch-of-bug-fixes-following-increased-ai-code-review-jekeknw1d) · Phoronix · 1 upvotes · 0 comments --- Tags: [#linux](https://daily.dev/tags/linux), [#rust](https://daily.dev/tags/rust), [#gpu](https://daily.dev/tags/gpu) [View this post on daily.dev](https://daily.dev/posts/linux-7-3-what-s-actually-changing-in-the-kernel-wdkvq9rpa) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"TechArticle","headline":"Linux 7.3: what's actually changing in the kernel","url":"https://daily.dev/posts/linux-7-3-what-s-actually-changing-in-the-kernel-wdkvq9rpa","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/linux-7-3-what-s-actually-changing-in-the-kernel-wdkvq9rpa"},"datePublished":"2026-09-06T14:08:56.218Z","dateModified":"2026-09-13T00:44:22.761Z","description":"The Linux 7.3 merge window closed with a wide-ranging set of kernel changes: scheduler fixes reducing jitter and improving Cache Aware Load Scheduling on...","image":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/9f38eee69571adc2f1798fc61a30cb63?_a=AQAEuop","thumbnailUrl":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/9f38eee69571adc2f1798fc61a30cb63?_a=AQAEuop","isAccessibleForFree":true,"articleSection":"Collections","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"Collections","logo":"https://media.daily.dev/image/upload/s--fk_6ycEi--/f_auto,q_auto/v1780996001/logos/collections?_a=BAMAMiWQ0","url":"https://daily.dev/sources/collections"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/linux-7-3-what-s-actually-changing-in-the-kernel-wdkvq9rpa","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":10},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"linux,rust,gpu","timeRequired":"PT5M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Collections","item":"https://daily.dev/sources/collections"},{"@type":"ListItem","position":3,"name":"Linux 7.3: what's actually changing in the kernel"}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/linux-7-3-what-s-actually-changing-in-the-kernel-wdkvq9rpa#faq","mainEntity":[{"@type":"Question","name":"What does the Btrfs direct I/O rewrite in Linux 7.3 change for throughput?","acceptedAnswer":{"@type":"Answer","text":"Linux 7.3 ships a rewritten Btrfs direct I/O path that improves throughput by 3-5x on some workloads. The gains are most noticeable for databases and certain VM setups that rely heavily on direct I/O, while the change is largely invisible for typical desktop or general-purpose use. daily.dev helps engineers tuning database or VM storage performance keep up with kernel I/O changes like this."}},{"@type":"Question","name":"What did the Linux 7.3 scheduler fix for Cache Aware Load Scheduling actually correct?","acceptedAnswer":{"@type":"Answer","text":"Linux 7.3 includes a fix, written by Intel engineer Tim Chen, that stops the cache-aware scheduler from migrating tasks to a last-level cache domain whose CPUs are too small to run them efficiently on big.LITTLE-style hybrid CPUs. It guards both the LLC migration path and active load balancing, and reprioritizes fixing misfit tasks over cache locality gains. The fix only affects asymmetric CPU systems. engineers debugging hybrid-CPU scheduling quirks can track kernel scheduler patches like this on daily.dev."}},{"@type":"Question","name":"Why are AI-generated patches becoming a problem for the Linux kernel maintainers?","acceptedAnswer":{"@type":"Answer","text":"Kernel maintainer Greg KH flagged that AI-generated low-quality patch submissions are overwhelming kernel review teams, adding significant triage noise for maintainers. CVE counts tied to the Linux 7.3 release are approaching 2,000, reflecting the growing volume maintainers now have to sift through, separate from any actual technical defects in the kernel. daily.dev keeps maintainers and contributors informed on how AI-generated code is reshaping open source review workflows."}}]} ```