Greetings, We are seeing a regression in 6.6.66 which I have narrowed down to this commit: stable: 86c27603514cb8ead29857365cdd145404ee9706 upstream: 7ffc7481153bbabf3332c6a19b289730c7e1edf5 Kernel version: 6.6.66 on aarch64 with 16k page size Last known good version: 6.6.65 Steps to repro: - run a 16k page size kernel (check with getconf PAGESIZE) - try to load an nftables config file on the problem Expected: - no errors Actual: - system enters a broken state, with the following trace in dmesg: [ 40.939230] Unable to handle kernel paging request at virtual address ffff00015ad7e4cc [ 40.939841] Mem abort info: [ 40.940079] ESR = 0x0000000096000021 [ 40.940389] EC = 0x25: DABT (current EL), IL = 32 bits [ 40.940820] SET = 0, FnV = 0 [ 40.941042] EA = 0, S1PTW = 0 [ 40.941289] FSC = 0x21: alignment fault [ 40.941570] Data abort info: [ 40.941805] ISV = 0, ISS = 0x00000021, ISS2 = 0x00000000 [ 40.942229] CM = 0, WnR = 0, TnD = 0, TagAccess = 0 [ 40.942857] GCS = 0, Overlay = 0, DirtyBit = 0, Xs = 0 [ 40.943313] swapper pgtable: 16k pages, 48-bit VAs, pgdp=00000000474f0000 [ 40.943865] [ffff00015ad7e4cc] pgd=180000043f7e8003, p4d=180000043f7e8003, pud=180000043f7e4003, pmd=180000043f52c003, pte=006800019ad7cf07 [ 40.945664] Internal error: Oops: 0000000096000021 [#1] SMP [ 40.946055] Modules linked in: zstd zram zsmalloc nf_log_syslog nft_log nft_reject_inet nf_reject_ipv4 nf_reject_ipv6 nft_reject nft_ct nf_conntrack nf_defrag_ipv6 nf_defrag_ipv4 nf_tables crct10dif_ce polyval_ce polyval_generic ghash_ce tcp_bbr sch_fq fuse nfnetlink vsock_loopback vmw_vsock_virtio_transport_common vmw_vsock_vmci_transport vmw_vmci vsock bpf_preload qemu_fw_cfg ip_tables squashfs virtio_net net_failover virtio_blk gpio_keys failover virtio_mmio virtio_scsi virtio_console virtio_balloon virtio_gpu virtio_dma_buf megaraid_sas [ 40.950262] CPU: 7 PID: 116 Comm: kworker/7:1 Not tainted 6.6.67-1-lts #1 [ 40.951542] Hardware name: netcup KVM Server, BIOS VPS 2000 ARM G11 08/07/2024 [ 40.952287] Workqueue: events_power_efficient nft_rhash_gc [nf_tables] [ 40.952798] pstate: 20401005 (nzCv daif +PAN -UAO -TCO -DIT +SSBS BTYPE=--) [ 40.953401] pc : nft_rhash_gc+0x208/0x2c0 [nf_tables] [ 40.954054] lr : nft_rhash_gc+0x134/0x2c0 [nf_tables] [ 40.954806] sp : ffff800081fb3cf0 [ 40.955138] x29: ffff800081fb3d50 x28: 0000000000000000 x27: 0000000000000000 [ 40.955974] x26: ffff0000cc760ef0 x25: ffff0000ccba9c80 x24: ffff0000cc758f78 [ 40.956750] x23: 0000000000000010 x22: ffff0000ca789000 x21: ffff0000cc760f78 [ 40.957455] x20: ffffd62d3ac3be40 x19: ffff00015ad7e4c0 x18: 0000000000000000 [ 40.959005] x17: 0000000000000000 x16: ffffd62d37926880 x15: 000000400002fef8 [ 40.959614] x14: ffffffffffffffff x13: 0000000000000030 x12: ffff0000cc760ef0 [ 40.960197] x11: 0000000000000000 x10: ffff800081fb3d08 x9 : ffff0000cba89fe0 [ 40.960739] x8 : 0000000000000003 x7 : 0000000000000000 x6 : 0000000000000000 [ 40.961287] x5 : 0000000000000040 x4 : ffff0000cba89ff0 x3 : ffff00015ad7e4c0 [ 40.961814] x2 : ffff00015ad7e4cc x1 : ffff00015ad7e4cc x0 : 0000000000000004 [ 40.962339] Call trace: [ 40.962531] nft_rhash_gc+0x208/0x2c0 [nf_tables] [ 40.962911] process_one_work+0x178/0x3e0 [ 40.963710] worker_thread+0x2ac/0x3e0 [ 40.964530] kthread+0xf0/0x108 [ 40.966259] ret_from_fork+0x10/0x20 [ 40.967287] Code: 54fff9a4 d503201f d2800080 91003261 (f820303f) [ 40.968245] ---[ end trace 0000000000000000 ]--- faddr2line gave me the following: nft_rhash_gc+0x208/0x2c0: __lse_atomic64_or at /root/gg/setup/linux-lts/src/linux-6.6.67/./arch/arm64/include/asm/atomic_lse.h:132 (inlined by) arch_atomic64_or at /root/gg/setup/linux-lts/src/linux-6.6.67/./arch/arm64/include/asm/atomic.h:65 (inlined by) raw_atomic64_or at /root/gg/setup/linux-lts/src/linux-6.6.67/./include/linux/atomic/atomic-arch-fallback.h:3771 (inlined by) raw_atomic_long_or at /root/gg/setup/linux-lts/src/linux-6.6.67/./include/linux/atomic/atomic-long.h:1069 (inlined by) arch_set_bit at /root/gg/setup/linux-lts/src/linux-6.6.67/./include/asm-generic/bitops/atomic.h:18 (inlined by) set_bit at /root/gg/setup/linux-lts/src/linux-6.6.67/./include/asm-generic/bitops/instrumented-atomic.h:29 (inlined by) nft_set_elem_dead at /root/gg/setup/linux-lts/src/linux-6.6.67/./include/net/netfilter/nf_tables.h:1576 (inlined by) nft_rhash_gc at /root/gg/setup/linux-lts/src/linux-6.6.67/net/netfilter/nft_set_hash.c:375 By looking at the diff between 6.6.65 and 6.6.66 I was able to narrow it down to the above commit and I can confirm that reverting it fixes the issue. Best -- Amby Balaji Co-founder & CTO Hyperbeam, Inc.
