Storage and I/O changes for the next Linux release are entering the kernel's merge window, with the headline item being NVMe support for Apple M4 devices — a hardware-enablement win that also carries implications for how the kernel handles high-throughput storage work.
According to Phoronix, the block-layer, storage and IO_uring updates slated for Linux 7.4 have been merged as part of the current merge window. What stands out is the pull request's sheer range: it spans fixes for the Linux floppy disk driver at one end and Apple M4 NVMe enablement at the other, a reminder of how much of the kernel's storage stack serves hardware that most servers will never touch.
The floppy fixes touch a legacy interface that persists largely for compatibility and emergency recovery use. The more consequential work sits in the NVMe and IO_uring portions of the pull — changes that would typically be felt by systems running data stores, containerised services or any workload that leans heavily on asynchronous I/O.
The M4 NVMe support matters on its own terms for anyone running Linux on Apple silicon, a platform whose storage enablement has often lagged behind the rest of the hardware stack. But the broader relevance lies in what the merge window reveals about the kernel's ongoing investment in fast storage paths: NVMe remains the reference device class for pushing block-layer performance, and work landing there tends to propagate through the storage stack generally.
On IO_uring, the kernel's asynchronous I/O submission mechanism — introduced so applications can queue read and write operations without the per-call overhead of traditional syscalls — Phoronix's coverage itemises what actually changed in this cycle's IO_uring and block-layer updates.
The merge window is the period after each kernel release when maintainers submit their subsystem trees for Linus Torvalds to pull ahead of the next version. Nothing merged during this window is final: the changes will be tested through a subsequent release-candidate cycle, during which fixes, revisions and, occasionally, outright reversions are common.
Practically, that means the storage and I/O work landing now will filter into distribution kernels over coming cycles rather than arrive immediately. Administrators planning kernel upgrades will want to track the NVMe and IO_uring changes through the release-candidate period before deciding when to adopt them.
Read the full report at Phoronix.
下一個 Linux 版本的儲存與 I/O 變更正進入 kernel 的合併視窗（merge window），其中最受矚目的是 Apple M4 裝置的 NVMe 支援——這項硬體啟用（hardware enablement）成果，同時亦對 kernel 如何處理高吞吐量儲存工作具有影響。
據 Phoronix 報導，為 Linux 7.4 而設的 block 層、儲存及 IO_uring 更新已於本輪合併視窗內完成合併。當中最突出的是該 pull request 涵蓋範圍之廣：一端是 Linux 軟碟驅動程式的修正，另一端則是 Apple M4 NVMe 的啟用，提醒我們 kernel 的儲存堆疊（storage stack）有多大一部分其實是為大多數伺服器永遠不會接觸的硬體而服務。
軟碟修正涉及一個舊式介面，該介面大致上仍為相容性及緊急復原用途而保留。較具實質影響的工作則在該 pull 的 NVMe 及 IO_uring 部分——這些變更通常會對運行資料儲存（data stores）、containerised 服務，或任何重度依賴非同步 I/O 的工作負載的系統帶來影響。
M4 NVMe 支援對於任何在 Apple Silicon 上運行 Linux 的用戶而言本身已具重要意義，因為此平台的儲存啟用進度往往落後於其餘硬體堆疊。但更廣泛的意義在於，本次合併視窗揭示了 kernel 對快速儲存路徑的持續投入：NVMe 仍是推動 block 層效能的參考設備類別（reference device class），在該處落實的工作往往會逐步擴展至整個儲存堆疊。
至於 IO_uring，即 kernel 的非同步 I/O 提交機制——其引入目的是讓應用程式毋須承擔傳統 syscall 的每次呼叫開銷，便可排入讀取及寫入操作——Phoronix 的報導逐項列出了本輪 IO_uring 及 block 層更新的實際變更內容。
合併視窗是指每次 kernel 發布後的一段時期，各子系統維護者會在此期間提交其子系統樹，供 Linus Torvalds 在下一個版本之前拉入（pull）。在此視窗內合併的任何內容均非最終定案：相關變更將會透過其後的 release candidate 週期進行測試，期間出現修正、修訂，以至偶爾完全撤回（revert），均屬常見情況。
實際而言，這意味著現階段落實的儲存及 I/O 工作，將會在往後數個週期逐步融入各發行版 kernel，而非即時到位。計劃升級 kernel 的系統管理員，宜先於 release candidate 期間追蹤 NVMe 及 IO_uring 的變更，再決定何時採納。
如欲閱覽完整報導，請參閱 Phoronix。