Linux Kernel 7.3 RC2 Lands With ~2,000 Commits and a Busier Than Usual Fix Cycle
Linus Torvalds pulls nine subsystem trees for the second release candidate, mostly to tidy up scheduler regressions and harden the BPF verifier.
Linux 7.3-rc2 shipped on Sunday afternoon after just one week of post-rc1 triage. The second release candidate of the 7.3 cycle brings 1,984 commits from 694 contributors, with the bulk landing in the BPF, USB, and tracing trees. If you were expecting a typical quiet rc2, this one deliberately skips that tradition.
Torvalds kept the commit message refreshingly candid. He noted that rc2 is normally the quietest stretch of the cycle, with developers catching their breath after the merge window. This week, however, multiple filesystem maintainers, a sizable DRM pull, networking fixes, and a handful of driver trees all showed up at once. Tooling accounted for roughly a fifth of the patch set, led by sched_ext work and selftests. Nothing looks particularly odd, even if the rc2 timing is a bit unusual. We will all just blame it on AI, because whether that is really the cause or not, it is an easy thing to blame.
A Post-rc1 Tidy-Up
The release leans heavily into regression fixes. The biggest theme is the fallout from the recent single-runqueue conversion in the CFS scheduler. The seven commits in the scheduling tree patch up timestamping bugs, bandwidth calculation issues, and throttling logic that cracked under the new architecture. Greg KH dropped a single virtio_console fix that directly addresses a regression reporters and CI systems had already flagged in rc1.
"Here is a single virtio console fix for 7.3-rc2 to fix a much reported regression in 7.3-rc1, sorry about that," KH wrote. That is kernel maintainer speak for "we saw this coming."
There is also a steady stream of security-relevant patches. The rtl8723bs WiFi driver gets three out-of-bounds read fixes, the mdc800 USB driver swaps an uninitialized kmalloc() for kzalloc(), and multiple USB gadget drivers finally patch use-after-free bugs found by syzbot. Google's automated fuzzing tool clearly earned its keep this cycle.
BPF landed the most patches at 86 commits, all under Alexei Starovoitov. The work centers on verifier edge cases that Nicholas Carlini surfaced, including incorrect non-NULL inference in pointer comparisons and precision tracking around zero spills. There are also fixes for percpu map updates, sparse CPU ID validation, and a trampoline allocation slowdown on x86. The eBPF security boundary continues to get ironed out, which is exactly what you want to see during the rc phase.
A few other subsystems wrapped up their work this cycle. Locking, IRQ, and perf events each shipped small but targeted patches. The tracing subsystem added a lock around ftrace_ops initialization to prevent race conditions, and capped the size of mapped ring buffers. Staging finished with out-of-bounds read fixes in the rtl8723bs driver and an IRQ safety patch for fbtft. Even the Rust kernel runtime got a mention, with a fix for a softirq processing delay in local_interrupt_disable().
The whole package clocks in at roughly two thousand commits for a single week of active development. Not exactly quiet. It feels like herding cats with deadlines, but the patch counts tell you what actually matters. The tree is being stress-tested exactly when it should be.
Linux kernel 7.3-rc2 released
Linux kernel version 7.3-rc2 is now available:
Full source: https://git.kernel.org/torvalds/t/linux-7.3-rc2.tar.gz
Patch: https://git.kernel.org/torvalds/p/v7.3-rc2/v7.2
You can view the summary of the changes at the following URL:
https://git.kernel.org/torvalds/ds/v7.3-rc2/v7.3-rc1
