On Wed, Nov 25, 2020 at 10:30:53AM +0000, Mel Gorman wrote: > On Tue, Nov 24, 2020 at 03:56:22PM -0500, Andrea Arcangeli wrote: > > Hello, > > > > On Tue, Nov 24, 2020 at 01:32:05PM +0000, Mel Gorman wrote: > > > I would hope that is not the case because they are not meant to overlap. > > > However, if the beginning of the pageblock was not the start of a zone > > > then the pages would be valid but the pfn would still be outside the > > > zone boundary. If it was reserved, the struct page is valid but not > > > suitable for set_pfnblock_flags_mask. However, it is a concern in > > > general because the potential is there that pages are isolated from the > > > wrong zone. > > > > I guess we have more than one issue to correct in that function > > because the same BUG_ON reproduced again even with the tentative patch > > I posted earlier. > > > > So my guess is that the problematic reserved page isn't pointed by the > > min_pfn, but it must have been pointed by the "highest" variable > > calculated below? > > > > if (pfn >= highest) > > highest = pageblock_start_pfn(pfn); > > > > When I looked at where "highest" comes from, it lacks > > pageblock_pfn_to_page check (which was added around v5.7 to min_pfn). > > > > Is that the real bug, which may be fixed by something like this? (untested) > > > > It's plausible as it is a potential source of leaking but as you note > in another mail, it's surprising to me that valid struct pages, even if > within memory holes and reserved would have broken node/zone information > in the page flags. I think the patch to add pageblock_pfn_to_page is still needed to cope with !pfn_valid or a pageblock in between zones, but pfn_valid or pageblock in between zones is not what happens here. So the patch adding pageblock_pfn_to_page would have had the undesired side effect of hiding the bug so it's best to deal with the other bug first.
