On 2020-10-29 11:23:25 [+0100], Mike Galbraith wrote: > Below, the aforementioned 5.10-rt splat in case that helps. > > [ 155.051826] CPU0 CPU1 > [ 155.051826] ---- ---- > [ 155.051826] lock(&xhci->lock); > [ 155.051827] lock((softirq_ctrl.lock).lock); > [ 155.051828] lock(&xhci->lock); > [ 155.051828] lock((lock).lock); > [ 155.051829] no, that is yet again different. Based on these lines the reason is most likely the same as in net/TCP you reported a few releases ago (RT specific). I need to consult core & locking overlords to figure out how we deal with this in future since is seems this cases pop up in packs. If you have more lockdep reports with softirq_ctrl.lock involved, please throw them at me. Sebastian
