The patch titled Subject: mm: mmu_gather: allow more than one batch of delayed rmaps has been added to the -mm mm-unstable branch. Its filename is mm-mmu_gather-allow-more-than-one-batch-of-delayed-rmaps.patch This patch will shortly appear at https://git.kernel.org/pub/scm/linux/kernel/git/akpm/25-new.git/tree/patches/mm-mmu_gather-allow-more-than-one-batch-of-delayed-rmaps.patch This patch will later appear in the mm-unstable branch at git://git.kernel.org/pub/scm/linux/kernel/git/akpm/mm Before you just go and hit "reply", please: a) Consider who else should be cc'ed b) Prefer to cc a suitable mailing list as well c) Ideally: find the original patch on the mailing list and do a reply-to-all to that, adding suitable additional cc's *** Remember to use Documentation/process/submit-checklist.rst when testing your code *** The -mm tree is included into linux-next via the mm-everything branch at git://git.kernel.org/pub/scm/linux/kernel/git/akpm/mm and is updated there every 2-3 working days ------------------------------------------------------ From: Linus Torvalds <torvalds@xxxxxxxxxxxxxxxxxxxx> Subject: mm: mmu_gather: allow more than one batch of delayed rmaps Date: Tue, 6 Dec 2022 11:15:09 -0800 Commit 5df397dec7c4 ("mm: delay page_remove_rmap() until after the TLB has been flushed") limited the page batching for the mmu gather operation when a dirty shared page needed to delay rmap removal until after the TLB had been flushed. It did so because it needs to walk that array of pages while still holding the page table lock, and our mmu_gather infrastructure allows for batching quite a lot of pages. We may have thousands on pages queued up for freeing, and we wanted to walk only the last batch if we then added a dirty page to the queue. However, when I limited it to one batch, I didn't think of the degenerate case of the special first batch that is embedded on-stack in the mmu_gather structure (called "local") and that only has eight entries. So with the right pattern, that "limit delayed rmap to just one batch" will trigger over and over in that first small batch, and we'll waste a lot of time flushing TLB's every eight pages. And those right patterns are trivially triggered by just having a shared mappings with lots of adjacent dirty pages. Like the 'page_fault3' subtest of the 'will-it-scale' benchmark, that just maps a shared area, dirties all pages, and unmaps it. Rinse and repeat. We still want to limit the batching, but to fix this (easily triggered) degenerate case, just expand the "only one batch" logic to instead be "only one batch that isn't the special first on-stack ('local') batch". That way, when we need to flush the delayed rmaps, we can still limit our walk to just the last batch - and that first small one. Link: https://lkml.kernel.org/r/CAHk-=whkL5aM1fR7kYUmhHQHBcMUc-bDoFP7EwYjTxy64DGtvw@xxxxxxxxxxxxxx Fixes: 5df397dec7c4 ("mm: delay page_remove_rmap() until after the TLB has been flushed") Signed-off-by: Linus Torvalds <torvalds@xxxxxxxxxxxxxxxxxxxx> Reported-by: kernel test robot <yujie.liu@xxxxxxxxx> Link: https://lore.kernel.org/oe-lkp/202212051534.852804af-yujie.liu@xxxxxxxxx Tested-by: Huang, Ying <ying.huang@xxxxxxxxx> Tested-by: Hugh Dickins <hughd@xxxxxxxxxx> Cc: Feng Tang <feng.tang@xxxxxxxxx> Cc: Johannes Weiner <hannes@xxxxxxxxxxx> Cc: Nadav Amit <nadav.amit@xxxxxxxxx> Cc: Xing Zhengjun <zhengjun.xing@xxxxxxxxxxxxxxx> Cc: "Yin, Fengwei" <fengwei.yin@xxxxxxxxx> Signed-off-by: Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx> --- --- a/mm/mmu_gather.c~mm-mmu_gather-allow-more-than-one-batch-of-delayed-rmaps +++ a/mm/mmu_gather.c @@ -19,8 +19,8 @@ static bool tlb_next_batch(struct mmu_ga { struct mmu_gather_batch *batch; - /* No more batching if we have delayed rmaps pending */ - if (tlb->delayed_rmap) + /* Limit batching if we have delayed rmaps pending */ + if (tlb->delayed_rmap && tlb->active != &tlb->local) return false; batch = tlb->active; @@ -48,31 +48,35 @@ static bool tlb_next_batch(struct mmu_ga } #ifdef CONFIG_SMP +static void tlb_flush_rmap_batch(struct mmu_gather_batch *batch, struct vm_area_struct *vma) +{ + for (int i = 0; i < batch->nr; i++) { + struct encoded_page *enc = batch->encoded_pages[i]; + + if (encoded_page_flags(enc)) { + struct page *page = encoded_page_ptr(enc); + page_remove_rmap(page, vma, false); + } + } +} + /** * tlb_flush_rmaps - do pending rmap removals after we have flushed the TLB * @tlb: the current mmu_gather * * Note that because of how tlb_next_batch() above works, we will - * never start new batches with pending delayed rmaps, so we only - * need to walk through the current active batch. + * never start multiple new batches with pending delayed rmaps, so + * we only need to walk through the current active batch and the + * original local one. */ void tlb_flush_rmaps(struct mmu_gather *tlb, struct vm_area_struct *vma) { - struct mmu_gather_batch *batch; - if (!tlb->delayed_rmap) return; - batch = tlb->active; - for (int i = 0; i < batch->nr; i++) { - struct encoded_page *enc = batch->encoded_pages[i]; - - if (encoded_page_flags(enc)) { - struct page *page = encoded_page_ptr(enc); - page_remove_rmap(page, vma, false); - } - } - + tlb_flush_rmap_batch(&tlb->local, vma); + if (tlb->active != &tlb->local) + tlb_flush_rmap_batch(tlb->active, vma); tlb->delayed_rmap = 0; } #endif _ Patches currently in -mm which might be from torvalds@xxxxxxxxxxxxxxxxxxxx are mm-mmu_gather-allow-more-than-one-batch-of-delayed-rmaps.patch