On Thu, Jan 04, 2024 at 11:28:22AM -0800, Darrick J. Wong wrote: > Though looking at buftarg allocation and my old notes from a couple of > years ago -- a second reason for allocating the buftarg during scrub > setup was that the list_lru_init call allocates an array that's > O(nodes_nr) and percpu_counter_init allocates an array that's > O(maxcpus). At the time I decided that it was better to put those large > contiguous memory allocations in the ->setup routine where we don't have > any vfs/xfs locks held, can run direct reclaim, and haven't done any xfs > work yet. Given that we use the page LRU for the shemfs pages, I don't think we need the buftarg LRU list at all - aging just the buffer container doesn't make much sense.
