On Tue, Mar 16, 2021 at 09:36:49AM -0700, Andi Kleen wrote: > > So we now map the PMD-sized page into userspace, even though it has a > > HWPoison in it. > > Yes that's bad. It might panic the system. > > > > > I think there are two things that we should be doing: > > > > 1. Attempt to split THPs which are file-backed. That makes most of this > > problem disappear because there won't be THPs with HWPoison, mostly. > > That seems simple and sensible. > > > 2. When the THP fails to split, use a spare page flag to indicate that > > the THP contains a HWPoison bit in one of its subpages. There are a > > lot of PF_SECOND flags available for this purpose. > > Why should it fail? I thought splitting always succeeds. If somebody has a temporary reference to it, splitting will fail. eg a process is currently doing a read() from it, so filemap_read() has a reference on the page. This is why THPs support a delayed split (currently used under memory pressure, but we could do something using that delayed list to schedule the page for splitting later if it has HWPoison).
