On Wed, Aug 24, 2022 at 11:47:03PM +0100, Mark Brown wrote:
On Wed, Aug 24, 2022 at 01:22:12PM -0700, KernelCI bot wrote: The KernelCI bisection bot identified 7eea9a60703ca ("usbnet: smsc95xx: Forward PHY interrupts to PHY driver to avoid polling") as causing a boot regression on Panda in v5.18. The board stops detecting a link which breks NFS boot:
<6>[ 4.953613] smsc95xx 2-1.1:1.0 eth0: register 'smsc95xx' at usb-4a064c00.ehci-1.1, smsc95xx USB 2.0 Ethernet, 02:03:01:8c:13:b0 <6>[ 5.032928] smsc95xx 2-1.1:1.0 eth0: hardware isn't capable of remote wakeup <6>[ 5.044036] smsc95xx 2-1.1:1.0 eth0: Link is Down <6>[ 25.053924] Waiting up to 100 more seconds for network. <6>[ 45.074005] Waiting up to 80 more seconds for network. <6>[ 65.093933] Waiting up to 60 more seconds for network. <6>[ 85.123992] Waiting up to 40 more seconds for network. <6>[ 105.143951] Waiting up to 20 more seconds for network. <5>[ 125.084014] Sending DHCP requests ...... timed out!
Looks like v5.19 boots fine (with the offending commit), so only v5.18-stable and v5.15-stable are affected:
https://storage.kernelci.org/stable-rc/linux-5.19.y/v5.19.3-366-g32f80a5b58e...
The offending commit was a new feature introduced in v5.19, it deliberately wasn't tagged for stable. Aggressively backporting such new features always carries a risk of regressions.
Off the cuff I don't see that a prerequisite commit is missing in stable kernels. So just reverting the offending commit in v5.18-stable and v5.15-stable (but not in v5.19-stable) might be the simplest solution.
Thanks,
Lukas