Re: [PATCH v9 00/37] mm/virtio: skip redundant zeroing of host-zeroed pages

[Michael S. Tsirkin]

I went over sashiko review of this:
https://sashiko.dev/#/patchset/cover.1780067977.git.mst%40redhat.com

It's mostly Sashiko apparently not seeing the whole picture
(e.g. is says "an optimization is missing" and it is actually
 added in follow up patches).

Sometimes it is unhappy while there's an explanation right in the commit
log. Weird.


It's a pity there's no way to respond to the review, to make it pay more
attention and reconsider.


It did flag a minor rebase artifact (declaration landed in the wrong
patch during rebase). Mostly harmless but if there's a v10, I'll fix that.

But it also flagged 4 pre-existing bugs:

  - page_reporting_register before DRIVER_OK, theoretical race in virtballoon_probe ordering:
  https://sashiko.dev/#/patchset/cover.1780067977.git.mst%40redhat.com?part=4

  - spurious OOM on large-folio swapin race in do_swap_page:
  https://sashiko.dev/#/patchset/cover.1780067977.git.mst%40redhat.com?part=16

  - free_huge_folio called with refcount==1 on mem_cgroup_charge_hugetlb failure:
  https://sashiko.dev/#/patchset/cover.1780067977.git.mst%40redhat.com?part=20

  - double-decrement of resv_huge_pages in memfd_alloc_folio error path:
  https://sashiko.dev/#/patchset/cover.1780067977.git.mst%40redhat.com?part=21

All seem unrelated to the specific feature, so I'd rather put off fixing
these until the feature is merged.

-- 
MST