Re: [bug] Triple whammy of mt7996e bugs on 7.2+

[Ben Greear <greearb@candelatech.com>]

On 10/3/26 07:12, Ilya K wrote:
> OK, so another update: my current branch which is basically 7.2.8 + bpi-r4 specific stuff + any patches from Ben's tree that touch net/wireless or mt76 code (http://github.com/K900/linux/tree/bpi-r4-728-ct) survived an entire week of no crashes, which is a very good sign. Unfortunately, it's still >500 commits with a total of +20k -2k lines of diff, so I'll have to go through it again and strip it down even more before I can have any idea of what is actually helping...
> 

Hello Ilya,

Going back to your original report, your logs showed a lockup.  We have added a bunch of mac80211
patches to deal with failing hardware around this path.  This was primarily done for working
around iwlwifi issues, I have not seen mtk7996 hit this path.  Our kernel should warn pretty loudly
in this case, so if you are still hitting it you should see signs.

[59002.686052] Hardware became unavailable during restart.
[59002.692099] WARNING: net/mac80211/util.c:1959 at ieee80211_reconfig+0x254/0x1ad8 [mac80211], CPU#2: kworker/2:3/26768
[59002.702894] Modules linked in: wireguard sch_cake mt7996e mt76_connac_lib rtc_pcf8563 at24 mt76 mac80211 libarc4 cfg80211 rfkill nft_queue nft_nat 
nft_fib_inet nft_fib_ipv4 nft_fib_ipv6 nft_fib nft_masq nft_ct nft_chain_nat nf_nat nft_flow_offload nf_flow_table_inet nf_flow_table nf_tables mtk_ge_soc 
mtk_phy_lib sfp tag_mtk mtk_eth mt7530_mmio mt7530 dsa_core governor_passive hsr bridge cmdlinepart pcs_mtk_usxgmii stp llc sm4 spinand phy_mtk_xsphy reset_gpio 
rt5190a_regulator pcs_mtk_lynxi phy_mtk_xfi_tphy phylink mdio_i2c pwm_mediatek lvts_thermal pwm_fan mtk_cci_devfreq ramoops reed_solomon tcp_bbr sch_fq_codel 
amneziawg nf_conntrack libcurve25519 libchacha20poly1305 libchacha libpoly1305 nf_defrag_ipv6 ip6_udp_tunnel nf_defrag_ipv4 udp_tunnel fuse zram 842_decompress 
842_compress lz4hc_compress nvme nvme_core nvme_keyring mmc_block dm_mod nvme_auth pcie_mediatek_gen3 rpmb_core
[59002.780906] CPU: 2 UID: 0 PID: 26768 Comm: kworker/2:3 Not tainted 7.2.2 #1-NixOS PREEMPTLAZY
[59002.789512] Hardware name: Banana Pi BPI-R4 (2x SFP+) (DT)
[59002.794988] Workqueue: events_freezable ieee80211_restart_work [mac80211]
[59002.801949] pstate: 60000005 (nZCv daif -PAN -UAO -TCO -DIT -SSBS BTYPE=--)
[59002.808902] pc : ieee80211_reconfig+0x254/0x1ad8 [mac80211]
[59002.814622] lr : ieee80211_reconfig+0x254/0x1ad8 [mac80211]

And, if you can reproduce the deadlock with lockdep enabled, that may
help.  Your original logs didn't show the problem part of the lockup
as far as I could tell, just processes blocked on acquiring rtnl.

Thanks,
Ben


-- 
Ben Greear <greearb@candelatech.com>
Candela Technologies Inc  http://www.candelatech.com