On 2021-10-04 22:51:05 [-0300], Luis Claudio R. Goncalves wrote: > Avoid a possible circular locking dependency by taking the softirq_ctrl.lock > before taking dev_base_lock in netstat_show(), keeping locking order > consistent. … > [ 106.529734] 5.10.65-rt53+ #1 Tainted: G W I So looking at 5.15 there are multiple instances of that. Either _all_ of them need to disable bh while acquiring dev_base_lock or none of them. I've been looking at a few call chains and didn't find one which acquired that lock from softirq. This does not mean there isn't one. Tracing this back, I stumbled upon v2.3.6 where a few of those got their _bh removed while new ones were added with _bh. So _maybe_ this _bh isn't required and nobody noticed. Lockdep would yell if it would observe dev_base_lock in softirq and noticed one instance like this one in netstat_show() where it is missing. But it didn't happen. So my guess is that the _bh isn't needed. Sebastian
