https://bugzilla.kernel.org/show_bug.cgi?id=217965 --- Comment #73 from Ojaswin Mujoo (ojaswin.mujoo@xxxxxxx) --- Hi Mingyu, Thanks for looking into this and sharing the reproducer. You are correct in the analysis that we were getting stuck trying to check the same block group for aligned blocks, due to the fact that the fragment lists always returned the same group. However, about this: > The author changed the fragment order RB tree into list for better > performance. > However, the function 'ext4_mb_find_good_group_avg_frag_lists' will always > returns the same group every time, The change from rbtree to list was not done in the patchset you listed but rather here: https://lore.kernel.org/all/20220908092136.11770-5-jack@xxxxxxx/ which got merged in v6.0 kernel hence I think the behavior of fragment lists returning the same block group was there even before my patchset (the one you listed) However I remember people had mentioned that they started to see it after v6.5. Its been sometime since I looked into this but I remember I had concluded that since my patchset added new allocation criteria which made the allocator trim the request to more aggressively look for BGs in the free fragment lists, we just made this bug more easier to hit. I tried the replicator however I was unable to get to the high CPU util, but since you already have the setup, can you check if you are able to hit this issue in v6.4 vs v6.5. Thanks again, Ojaswin -- You may reply to this email to add a comment. You are receiving this mail because: You are watching the assignee of the bug.
