On Fri, Oct 14, 2022 at 11:28:53AM -0700, Jiaqi Yan wrote:
> >
> > Looks very broken. Or just hard to follow. Or both.
>
> Do you think the logic looks correct (and just hard to follow) after
> these explanation?
Yes.
> Currently the rollback operations are soaked with
> the copy_succeed code to minimize lines of code. But I can definitely
> rewrite to something like this if you think the rollback logic is
> correct:
>
> if (copy_succeeded) {
> ...
> } else {
> re-populate PMD;
> if (!pte_none && !zero_pfn && !PageCompound(src_page))
> release_pte_page(src_page);
> for(page in compound_list) {
> release_pte_page(page);
> }
> }
> return copy_succeeded;
Looks sane to me.
--
Kiryl Shutsemau / Kirill A. Shutemov
