One more thing..
On Fri, Sep 22, 2023 at 06:31:45PM -0700, Suren Baghdasaryan wrote:
> +static int remap_pages_pte(struct mm_struct *dst_mm,
[...]
> +retry:
> + dst_pte = pte_offset_map_nolock(dst_mm, dst_pmd, dst_addr, &dst_ptl);
> +
> + /* If an huge pmd materialized from under us fail */
> + if (unlikely(!dst_pte)) {
> + err = -EFAULT;
> + goto out;
> + }
> +
> + src_pte = pte_offset_map_nolock(src_mm, src_pmd, src_addr, &src_ptl);
> +
> + /*
> + * We held the mmap_lock for reading so MADV_DONTNEED
> + * can zap transparent huge pages under us, or the
> + * transparent huge page fault can establish new
> + * transparent huge pages under us.
> + */
> + if (unlikely(!src_pte)) {
> + err = -EFAULT;
> + goto out;
> + }
For these two places: I know that thp collapse with mmap read lock hasn't
yet spread to anon (so I assume none of above could trigger yet on the
failure paths), but shall we constantly return -EAGAIN here just in case we
forget that in the future?
For example, for UFFDIO_COPY over shmem which we can already hit similar
case, mfill_atomic_install_pte() has:
ret = -EAGAIN;
dst_pte = pte_offset_map_lock(dst_mm, dst_pmd, dst_addr, &ptl);
if (!dst_pte)
goto out;
Thanks,
--
Peter Xu
