On Tue, Sep 7, 2021 at 2:24 PM Minchan Kim <minchan@xxxxxxxxxx> wrote: > > kernel test robot reported the regression of fio.write_iops[1] > with [2]. > > Since lru_add_drain is called frequently, invalidate bh_lrus > there could increase bh_lrus cache miss ratio, which needs > more IO in the end. > > This patch moves the bh_lrus invalidation from the hot path( > e.g., zap_page_range, pagevec_release) to cold path(i.e., > lru_add_drain_all, lru_cache_disable). Was this confirmed to fix the regression? I only see the "tested with 5.14" that the regression was still there https://lore.kernel.org/lkml/034fc860-d0d0-0c61-09d2-3c41c4f020c6@xxxxxxxxx/ I don't see a confirmation that this patch fixed it. It looks likely, but if you have the confirmation somewhere, it would help to link that too. Or did I miss it? Linus
