$ the-wire · showcase
Wireless perf fixes close an AMD IBS kernel address leak
By RepoJournal · Filed · About Linux · Composed from the cited sources · methodology
3 people shipped this
The 7.4-rc2 merge window's back half lands mostly fixes, and the one that matters beyond a single benchmark box is a perf events pull that stops AMD IBS from handing userspace kernel addresses.
Merge tag 'perf-urgent-2026-11-08' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip torvalds/linux
Linus merged Ingo Molnar's perf events pull for 7.4-rc2, which includes a fix for a kernel address leak via REGS_USER and REGS_INTR in the AMD IBS paths. The same tag also resets CPU0 IBS state on syscore resume, simplifies event state handling with IBS_CAPS_D, fixes sideband event output when only attr.defer_output is requested, and clamps deferred callchain entry counts to fit the u16 header....
Merge tag 'devm_kmalloc_obj-v7.4-rc2' of git://git.kernel.org/pub/scm/linux/kernel/git/kees/linux torvalds/linux
Kees Cook's devm_kmalloc_obj conversion went in as a treewide refresh: the devm_kmalloc() family now maps to devm_kmalloc_obj(), with matching conversion rules added to the Coccinelle scripts. Out-of-tree driver code calling devm_kmalloc() will still compile, but any downstream patches carrying Coccinelle rules will need the new ruleset to stay in sync.
Merge tag 'ata-7.4-rc2' of git://git.kernel.org/pub/scm/linux/kernel/git/libata/linux torvalds/linux
Two libata behavior changes land for storage: NCQ can now be re-enabled after EH disables it by writing the current queue depth to the queue_depth sysfs attribute, and failed non-NCQ commands are retried when their sense data asks for a retry, which previously never happened. Power-mode checking is also now limited to ATA and ZAC devices before spin-down.
Merge tag 'firewire-fixes-7.4-rc2' of git://git.kernel.org/pub/scm/linux/kernel/git/ieee1394/linux1394 torvalds/linux
The 1394 OHCI driver gets a race fix for a bug introduced in v7.4-rc1: queueing an AT local request in interrupt context could race with dequeuing requests in a workqueue without disabling interrupts. Takashi Sakamoto's note says the race can "potentially lead to a deadlock," so anyone running an -rc on FireWire hardware should pick up rc2 rather than sit on rc1.
Add new plugins without reloading the ones already running (#18198) basecamp/omarchy
Omarchy's rescanPlugins no longer tears down every panel and non-keepLoaded service when it finds a new plugin; previously a first-login plugin link from Superwhisper's setup triggered a rescan that blanked the wallpaper and closed open panels. The driver-core pull also unbreaks Rust I/O builds with CONFIG_PCI disabled and CONFIG_WERROR enabled, and adds Johan Hovold as an auxiliary bus maintai...
