Content-Type: multipart/mixed; boundary="===============2778787528274625526==" MIME-Version: 1.0 From: Huang Ying To: lkp@lists.01.org Subject: Re: [mm] cc87317726f: WARNING: CPU: 0 PID: 1 atdrivers/iommu/io-pgtable-arm.c:413 __arm_lpae_unmap+0x341/0x380() Date: Thu, 19 Mar 2015 09:57:02 +0800 Message-ID: <1426730222.5570.41.camel@intel.com> In-Reply-To: <201503182045.DEC48482.OtSOQOLVFFHFJM@I-love.SAKURA.ne.jp> List-Id: --===============2778787528274625526== Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: quoted-printable On Wed, 2015-03-18 at 20:45 +0900, Tetsuo Handa wrote: > Huang Ying wrote: > > On Tue, 2015-03-17 at 15:24 -0400, Johannes Weiner wrote: > > > On Tue, Mar 17, 2015 at 10:15:29AM -0700, Linus Torvalds wrote: > > > > Explicitly adding the emails of other people involved with that com= mit > > > > and the original oom thread to make sure people are aware, since th= is > > > > didn't get any response. > > > > = > > > > Commit cc87317726f8 fixed some behavior, but also seems to have tur= ned > > > > an oom situation into a complete hang. So presumably we shouldn't l= oop > > > > *forever*. Hmm? > > > = > > > It seems we are between a rock and a hard place here, as we reverted > > > specifically to that endless looping on request of filesystem people. > > > They said[1] they rely on these allocations never returning NULL, or > > > they might fail inside a transactions and corrupt on-disk data. > > > = > > > Huang, against which kernels did you first run this test on this exact > > > setup? Is there a chance you could try to run a kernel without/before > > > 9879de7373fc? I want to make sure I'm not missing something, but all > > > versions preceding this commit should also have the same hang. There > > > should only be a tiny window between 9879de7373fc and cc87317726f8 -- > > > v3.19 -- where these allocations are allowed to fail. > > = > > I checked the test result of v3.19-rc6. It shows that boot will hang at > > the same position. > = > OK. That's the expected result. We are discussing about how to safely > allow small allocations to fail, including how to handle stalls caused by > allocations without __GFP_FS. > = > > = > > BTW: the test is run on 32 bit system. > = > That sounds like the cause of your problem. The system might be out of > address space available for the kernel (only 1GB if x86_32). You should > try running tests on 64 bit systems. We run test on 32 bit and 64 bit systems. Try to catch problems on both platforms. I think we still need to support 32 bit systems? Best Regards, Huang, Ying --===============2778787528274625526==--