On Mon, Aug 10, 2026 at 02:58:43PM +0200, Alexander Gordeev wrote: > On Fri, Aug 07, 2026 at 12:36:30PM +0100, Kiryl Shutsemau wrote: > > Hi Kiryl, > > ... > > - Dropped the swap-skip patch, an equivalent has landed > > (Alexander Gordeev). > > And dropped [1] following David feedback [2]. Would you pursue the proper fix?
Yes. v3 will handle the failure rather than predict it. is_swap_enabled() already skips these cases when no swap is active, so what is left is swap that exists but cannot take what the case asks for. The series already has swapout_range(), which reissues MADV_PAGEOUT for up to two seconds and reports whether the count arrived; I will make the two cases skip when it does not. That covers swap too small, swap full, a memcg limit and a folio still under writeback, without reading /proc/meminfo for any of them. An error from madvise() itself stays a hard failure -- that is the kernel refusing, not the machine being short of swap. -- Kiryl Shutsemau / Kirill A. Shutemov
