Two patch series queued for Linux 7.5 won't change any front page, and both say more about the state of the kernel than half a dozen release notes. The first adds `O_TMPFILE` support to the NTFS driver, the flag that allows creating an unnamed temporary file and deciding later whether to publish it. The second brings in the first Device Tree for the Apple M5, the file that describes to the kernel what hardware exists and without which no boot is possible.
The calendar rules. The merge window that decides what goes into 7.5 opens at the end of the year, and whatever doesn't arrive in time waits a full quarter. The O_TMPFILE series for NTFS has gone through four rounds of review and is already in the `ntfs-next` branch. M5 support entered `apple-soc/dt-7.5` on November 14, which is the branch sent to the ARM maintainers before each window.
Neither piece of work improves anyone's desktop next week. The NTFS one fixes how applications write temporary files on a Windows partition mounted on Linux; the M5 one only makes the kernel recognize a chip it didn't recognize until now.
What O_TMPFILE is and why NTFS didn't have it
An editor saves a file and the user shouldn't be able to open it half-written. `O_TMPFILE` exists for that: it is a POSIX flag, the standard that defines calls such as `open()` or `read()`. It allows creating a file with no name, writing to it and publishing it only when the content is ready. If the program dies beforehand, the data is freed when the last reference drops and no garbage is left in the directory.
The usefulness is not cosmetic. An editor that saves atomically or a server that prepares a large file depends on this mechanism; without it, the program creates a temporary name, writes and renames, with the inconsistency window that opens up.
The series is by Baolin Liu, an engineer at Kylin OS, and reached its fourth revision with twelve patches that, in addition to the support, fix the creation path and the reading of access control lists. The final submission is on the kernel list and the series is already in `ntfs-next`: data and metadata are prepared before publishing, and `O_EXCL` prevents the temporary file from being published by accident.
The delay is explained by the driver having to manage the lifecycle of an inode with no directory entry, including inherited access control lists, timestamps and extended attributes. In NTFS none of that was ready. Twelve patches for one flag: that is the real size of it.
A driver that stopped being a guest
The kernel's NTFS carries an uncomfortable history. For years there were two implementations, one in user space and one in kernel mode with limited writing, and neither was entirely reliable: corruption reports arrived as regularly as new versions.
The current driver arrived with the 7.2 series under the name NTFS Resurrection and since then has accumulated features in every cycle. 7.4 had shown the pattern with the FAT32 improvements in its second release candidate: file systems decades old receiving fine-grained work that nobody had asked for out loud.
That context explains why `O_TMPFILE` deserves a note. It is not an exotic feature added for completeness: it is confirmation that the driver is treated as a first-class file system and not as an adapter for reading other people's disks.
If you have an NTFS partition mounted on a dual-boot machine, this concerns you closely, although the other side of the disk doesn't help either: Windows 11 26H2 marks a week with bugs and half-finished patches. Copying a photo library works better when the system understands temporary files. Until now NTFS was the exception in that club; from 7.5 on, it isn't.
The M5 enters through the smallest door
The second piece of work describes the Apple M5, whose internal name in the kernel is t8142. The series is by Yureka Lilian and was queued by Sven Peter, the Apple platform lead. It covers three laptops: the 14-inch MacBook Pro with M5 and the 13- and 15-inch MacBook Air of the same generation.
What the Device Tree describes is deliberately little. It declares ten CPU cores, six efficiency and four performance, with two new compatibles bearing the names `apple,hidra-e` and `apple,hidra-p`. It adds the interrupt controller and the watchdog, and includes the debug serial port, four pin control blocks, five I2C buses for slow peripherals and power domain management.
Yureka Lilian sums it up without embellishment in the letter with which the series was submitted: it covers "a minimal amount of hardware," useful as a base for future work. Much of what was added is identical to previous generations of Apple Silicon.
What's missing and why the GPU is the wall
Asahi's method is always the same: first chip recognition and then everything else, so the list of absences is longer than the list of presences. There is no GPU or display coprocessor, so the panel doesn't light up. There is also no NVMe or PCIe, and without USB or Thunderbolt there is no convenient way to boot from a stick or mount a disk.
The biggest obstacle remains graphics acceleration. In the 7.3 series report, the Asahi project recounted that a group of contributors got the M3 to draw its first triangle, and that bringing that support to Mesa is long work. With the M5 that point hasn't been reached, and the project's documentation keeps most components pending assignment.
The reasonable question is why this work is done if the result is so limited. The answer lies in the dependency chain: every piece that enters the merge window is one less task for the next developer, and that is the difference between moving forward and starting over.
The M4 arrived in 7.4, the M3 in 7.2, and each left the ground a little flatter. Three generations, three merge windows and no approved GPU driver: that is the honest measure of the pace.
Two signals in the same window
Both series share a pattern worth noting. Neither solves a problem visible to the general public, both went through several revisions and both depend on someone maintaining the code for years. That is the work that almost never appears in the conversation about which version is better.
In a tree where automated detection already locates 948 Linux flaws that grant administrator privileges, the bottleneck is no longer finding the bug. Now it's about reviewing it, and that's why four rounds for twelve patches is a data point, not a complaint.
For those who administer machines, the reading is twofold. If NTFS partitions are in use, Linux 7.5 will bring more correct behavior in applications that use temporary files. And if anyone expects Linux on a Mac with M5, this Device Tree is the first day of the calendar, not the last: it's best measured in years, as the M1 was.
The kernel is also about more than what goes in. MPTCP's CVE-2026-98432 is still unpatched in RHEL: the code already inside also has to be attended to, and a first-class driver is a maintenance obligation, not a medal.
The merge note will talk about flashier features. These two changes won't make any headline, and even so a Mac with M5 and kernel 7.5 boots and stays at the console: no disk, no USB and no GPU.