Daily update · Oct 15–16, 2026

Kernel update: PHY probe crash fixes, sched_ext races, and Hyper-V reliability

A bug-fix-heavy daily update with Ethernet driver fixups, sched_ext/Hyper-V reliability changes, netlink hardening, and Tegra264 device-tree additions.

Bug fixes Hardware support

#Hyper-V #KVM #netlink #PHY #sched_ext #Tegra

58 meaningful commitsDaily snapshot

In brief

Today's mainline updates are mostly fixes. Several Ethernet drivers now initialize their device-operation table before connecting physical-layer transceivers, avoiding a NULL-pointer crash in a recent PHY-layer code path. There are also concurrency fixes for sched_ext, reliability fixes for Hyper-V guests, netlink hardening, and assorted KVM, PCI, and device-tree corrections.

Bug fixes

Ethernet drivers: assign netdev_ops before PHY attach

A recent PHY layer change reads the network device operations table while attaching a PHY. Several drivers were attaching the PHY before assigning netdev_ops, causing a NULL-pointer crash at probe time when CONFIG_NET_SHAPER is enabled. This batch moves the assignment earlier, matching an existing emac fix.

Why it matters: Prevents a probe-time crash on HiSilicon FEMAC/HIP04, Actions Owl, ADI ADIN1140, Marvell mv643xx, AMD Au1000, OpenCores ethoc, Aeroflex GRETH, TI ICSSM-PRUETH, and Microchip LAN865X devices under that config.

444889dfa27b 0d9342701f14 c2c482431985 ea1565cbf063 a3eef3d2cca0 beabbafdd0f6 c9f8ce62a5fa d71483f0716e 955bb7cf680d 2a3f6d2a9ac3

sched_ext: dispatch-queue and bypass race fixes

sched_ext is the BPF-based scheduler extension interface. This set fixes cases where a task could be left on a local dispatch queue without a reschedule, where a destroyed user dispatch queue could be read by a deferred reenqueue, and where nested schedulers under a parent bypass could end up with the wrong bypass/dispatch state. It also makes restricted idle selection respect task affinity.

Why it matters: Avoids task stalls, use-after-free, and bypass accounting WARNs in sched_ext environments, especially under cgroup teardown or PM suspend.

7a9df7667963 069403dff337 dcf7f650a248 5bdeca322a6f 6ac4544eff79 18f5c699d1d0 2935062be771

Hyper-V/VMBus and vPCI fixes

Fixes VMBus connection state handling so teardown doesn't skip the unload handshake, avoids an infinite retry in vCPU index initialization when a NUMA node has no managed-IRQ housekeeping CPU, and stops using an unallocated per-cpu output page for an APIC-to-VP index hypercall. Also probes virtual PCI buses asynchronously to avoid serialized host round-trips, warns when a host response waits indefinitely, and checks per-CPU register pages before mmap.

Why it matters: Improves Hyper-V guest reliability during shutdown/panic and device initialization, prevents a possible hang in CPU selection, and makes slow host responses visible.

32ea0a33e3a0 7b74c117e2aa ec31dc39e835 c37fc4deee00 ca039df94983 874194a8de75

Netlink hardening and network protocol fixes

Netlink now resets the skb network and MAC headers before socket filters can inspect a message, preventing headroom bytes from leaking through bytecode reads. It also replaces the raw net namespace pointer with an opaque cookie in the rhashtable key to avoid leaking kernel addresses. Protocol fixes cover vsock TCP_CLOSING races, RDS connection-ID lifetime, SCTP output stream revalidation after connect, CALIPSO payload_len handling for BIG TCP, and a tc action use-after-free.

Why it matters: Closes a local info-leak path in netlink and fixes memory-safety/hang bugs in vsock, RDS, SCTP, CALIPSO, and tc.

8bf659bd54e1 f88bb3606e4b 4f6ac74fa594 492ba6ac1b1f 98d446b7c013 b7ad72414c67 8070ed578b6b 7b16eb911091

Network driver fixes: bcmgenet, stmmac, mlx5, ath9k

bcmgenet unshares skbs before writing its control block to avoid byte-queue-limit stalls under pktgen clone_skb mode; dwmac1000 clears a stale RGMII status interrupt that could storm the CPU after resume; mlx5 HWS fixes pool release for L2-to-L3 tunnel reformats and corrects flow-meter offsets within a bulk. The ath9k change reverts a WMI response length check that broke AR9271 USB adapters.

Why it matters: Avoids queue/BQL stalls, resume-time interrupt storms, wrong mlx5 flow-meter behavior, and restores AR9271 support on kernels containing the earlier regression.

89771ce4f8f6 aa07285705f8 39e02ddf74bb 1b0aad53ad71 90201351a862

Core kernel, PCI, IOMMU, and KVM fixes

A workqueue mem-reclaim helper no longer dereferences current_pwq when a kworker isn't running a work item. PCI fixes keep the Apple T2 Secure Enclave's ANFE mask unchanged to avoid spontaneous power loss, and move Rockchip DW PCIe INTx IRQ setup out of .init() so root-port resets don't leak IRQ domains. KVM fixes stop an NMI-context #PF from consuming async-PF flags and reset the MMU after failed nested VMX state load. An AMD IOMMU cleanup is reverted, and the IOMMU kunit build is fixed for ThinLTO.

Why it matters: Prevents kernel panics/NULL derefs in workqueue, KVM, and PCI paths; avoids Apple T2 power loss; restores AMD IOMMU behavior on affected GPU systems; fixes an arm64/ThinLTO kunit build issue.

d1a4549a3645 7209846a169f 4e9f69d08253 81d99abe2bd8 c8f4cfe1cf93 3ba149029a43 4bcfe2902199

Device-tree and platform fixes

DRA7's PCIe PHY clock divider table is restored in the correct order because the clock driver derives register values from table indices. The BeagleV Fire device tree drops duplicate GPIO interrupts that were preventing the mcp3564 driver from probing. A Tegra PMC fix initializes the early PMC clock rate for 32-bit Tegra to avoid a suspend warning. The sama5 defconfig is updated to use the current Microchip KSZ DSA Kconfig symbols.

Why it matters: Fixes PCIe PHY clock behavior on DRA7, restores mcp3564 probing on BeagleV Fire, removes a spurious Tegra suspend warning, and makes sama5 defconfig match current DSA driver options.

2fabd2f406d0 0f70fd6f4c1d c5c95dbdc901 afb1ecfeda3f

Hardware enablement

Tegra264 device tree additions

Tegra264's device tree gains nodes for its PWM controllers and PCIe root ports, along with a reg/reg-names reorder to match bindings. The Jetson AGX Thor DevKit also gets a PWM fan node so the fan is actively controlled instead of stopping or running at full speed.

Why it matters: Provides device-tree coverage for Tegra264 PWM/PCIe and improves out-of-the-box fan behavior on the Jetson AGX Thor DevKit.

dd3b6f7adcc5 b86eaddf467a 4ee64cda9866 a93db5724ada

Source commits58 entries +

4ee64cda9866

arm64: tegra: Reorder reg and reg-names to match bindings

Thierry Reding · May 7, 2025 · 1 files

dd3b6f7adcc5

arm64: tegra: Add PWM controllers on Tegra264

Thierry Reding · Jul 1, 2026 · 1 files

2fabd2f406d0

ARM: dts: ti/omap: dra7: fix PCIe PHY clock divider definition

David Oberhollenzer · Jul 16, 2026 · 1 files

b86eaddf467a

arm64: tegra: Add PCIe root ports on Tegra264

Thierry Reding · Jul 16, 2026 · 1 files

2f4a0fb66d31

soc/tegra: fuse: Add missing newline to APBMISC error message

Jinhui Guo · Aug 4, 2026 · 1 files

afb1ecfeda3f

ARM: configs: sama5: enable current Microchip KSZ DSA symbols

Pengpeng Hou · Aug 14, 2026 · 1 files

c5c95dbdc901

soc/tegra: pmc: Fix uninitialised clock rate

Jon Hunter · Aug 21, 2026 · 1 files

a93db5724ada

arm64: tegra: Add PWM fan on Jetson AGX Thor DevKit

Thierry Reding · Aug 27, 2026 · 1 files

32ea0a33e3a0

Drivers: hv: Avoid infinite retry loop in init_vp_index()

Waiman Long · Aug 30, 2026 · 1 files

0cdd57a4c561

x86/hyperv: Fix stale comments after stimer message mode removal

Jiangshan Yi · Sep 1, 2026 · 2 files

ec31dc39e835

Drivers: hv: Don't reset VMBus connection state on error/exit paths

Michael Kelley · Sep 1, 2026 · 2 files

7b74c117e2aa

x86/hyperv: Avoid using per-cpu output page in hv_apicid_to_vp_index()

Michael Kelley · Sep 2, 2026 · 1 files

0f70fd6f4c1d

riscv: dts: microchip: beaglev-fire: remove double definition of gpio interrupts

Conor Dooley · Sep 4, 2026 · 1 files

c37fc4deee00

PCI: hv: Probe vPCI buses asynchronously

Naman Jain · Sep 7, 2026 · 1 files

874194a8de75

mshv_vtl: Check per-CPU register page before mmap

Hardik Garg · Sep 11, 2026 · 1 files

ca039df94983

PCI: hv: Warn when wait_for_response() waits indefinitely

Sahil Chandna · Sep 14, 2026 · 1 files

12f631d5ded5

MAINTAINERS: remove myself from the Hyper-V entries

Long Li · Sep 15, 2026 · 1 files

7724f2364396

Drivers: hv: vmbus: fix typo "substract" in comment

Hemanth Selam · Sep 15, 2026 · 1 files

d5e689696017

tools/hv: fix typo "retrive" in comment

Hemanth Selam · Sep 15, 2026 · 1 files

4e9f69d08253

PCI: dw-rockchip: Move INTx IRQ setup to probe

Shawn Lin · Sep 24, 2026 · 1 files

d1a4549a3645

workqueue: Fix NULL current_pwq deref in mem-reclaim helper

Pavankumar Kondeti · Sep 28, 2026 · 1 files

81d99abe2bd8

x86/kvm: Don't consume async #PF flags on a kernel #PF in NMI context

EJ Campbell · Sep 29, 2026 · 1 files

c8f4cfe1cf93

KVM: nVMX: Reset MMU back to the root MMU if loading L2 state fails for !VM-Entry

Amirmohammad Eftekhar · Oct 1, 2026 · 1 files

492ba6ac1b1f

net: rds: fix uninitialized trans dereference in CM event handler

Aohan Mei · Oct 3, 2026 · 1 files

98d446b7c013

net/rds: don't let the rdma_cm destroy an id RDS still owns on route failure

Allison Henderson · Oct 3, 2026 · 1 files

4f6ac74fa594

vsock: treat TCP_CLOSING as once-established

Michal Luczaj · Oct 5, 2026 · 1 files

8bf659bd54e1

netlink: reset skb headers before unicast and multicast filtering

Kyle Zeng · Oct 6, 2026 · 1 files

b7ad72414c67

sctp: revalidate output stream after association connect wait

Daehyeon Ko · Oct 7, 2026 · 1 files

39e02ddf74bb

net/mlx5: HWS, release L2 to L3 tunnel reformats to their own pool

Alexandre Cassen · Oct 7, 2026 · 1 files

1b0aad53ad71

net/mlx5: HWS, fix flow meter offset within its bulk

Alexandre Cassen · Oct 7, 2026 · 1 files

7a9df7667963

sched_ext: Fix stall when a task enqueued with SCX_ENQ_LAST lands back on its CPU's local DSQ

Tejun Heo · Oct 7, 2026 · 2 files

f88bb3606e4b