On Fri, Nov 04, 2022 at 11:38:48AM -0700, Dmitry Vyukov wrote: > On Fri, 4 Nov 2022 at 11:30, Jason Gunthorpe <jgg@xxxxxxxxxx> wrote: > > On Fri, Nov 04, 2022 at 11:21:21AM -0700, Dmitry Vyukov wrote: > > > > > But I am not sure if we really don't want to issue the fault injection > > > stack in this case. It's not a WARNING, it's merely an information > > > message. It looks useful in all cases, even with GFP_NOWARN. Why > > > should it be suppressed? > > > > I think it is fine to suppress it for *this call* but the bug turns it > > off forever more > > Is it just "fine", or "good"? I agree it's probably "fine", but > wouldn't it be better to not suppress it? It seems sensible to me that fail*/verbosity=2 should *always* print if a fault has been injected. I don't know why someone thought this one deserves to be shut down. GFP_NOWARN is about the caller indicating that it expects and will handle an allocation failure (eg it is asking for big regions and has fallbacks) so we shouldn't print the general OOM warning. Jason
