Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id ; Fri, 7 Dec 2001 05:22:43 -0500 Received: (majordomo@vger.kernel.org) by vger.kernel.org id ; Fri, 7 Dec 2001 05:22:28 -0500 Received: from pixar.pixar.com ([138.72.10.20]:64244 "EHLO pixar.pixar.com") by vger.kernel.org with ESMTP id ; Fri, 7 Dec 2001 05:22:17 -0500 Date: Fri, 7 Dec 2001 02:22:11 -0800 (PST) From: Kiril Vidimce To: linux-kernel Subject: Re: ldt allocation failed Message-ID: MIME-Version: 1.0 Content-Type: TEXT/PLAIN; charset=US-ASCII Sender: linux-kernel-owner@vger.kernel.org X-Mailing-List: linux-kernel@vger.kernel.org ---------- Forwarded message ---------- Date: Fri, 07 Dec 2001 10:18:59 GMT From: Ludo Stellingwerff To: Kiril Vidimce Subject: Re: ldt allocation failed Okay, I have found another failure path leading to the failure message: vmalloc -> vmalloc_area_pages -> alloc_area_pmd -> pte_alloc This pte_alloc temporarily releases a spinlock on the mm_pagetable. If at the same time another processor handles a vmalloc too, a race condition can surface. A check is done for this to happen. In that case a NULL allocation will be returned through the same path. This triggers the message. This is a kernel bug as a more decent failure mechanism should be in place. Something like a retry to allocate a page table entery. Does this make sense? Please forward this message to the kernel list, I am not on the list and can't reply to the thread. Hope it helps, Ludo Stellingwerff ----- Origineel Bericht ----- Van: Kiril Vidimce Datum: Vrijdag 7 December 2001 10:23 Onderwerp: Re: ldt allocation failed > On Fri, 7 Dec 2001, Ludo Stellingwerff wrote: > > I'm no experienced kernel hacker, but I like to try to help you. > > > > I am a bit frustrated with some of the kernel people too easily > > yelling: > > "Sorry, we do not have the source to NVIDIA's driver, so we > cannot help > > you debug this problem. Please contact NVIDIA support." > > It could well be a kernel problem, IMO. > > > > It looks like you have some process eating all your 2 GB's. The > error > > message is generated within the architecture specific part of > the > > process handling. It is triggered if memory allocation failes > for > > ldt's. Sorry I still don't know what they are, but if necessary > I will > > findout:) > > These ldt's need 65536 bytes of memmory. It seems you're machine > ran > > out of memory. > > > > The 10 to 15 seconds delay in 2.4.9 seems to correspond with the > OOM > > (Out of Memory kicker). Some processes are stoped to provide > extra > > memory space. > > In 2.4.3 the OOM didn't work IIRC. > > > > Are there any processes that seem to use lot's of memory? Can > you > > reproduce the problem without the NVidia module loaded? > > > > Hope this helps, > > Hi Ludo, > > This seems to be happening on a variety of machines with a variety > of software running on it. I seriously doubt it that some app > is actually running out of memory since I had it happen even > on a machine that basically only had netscape running on it. The > machines have 2 GB of RAM so it would be hard to run out of memory > just with X, Gnome and netscape running (netscape wasn't even > really active; it happened while it was starting up). > > Thanks for the mail anyway. > > KV > -- > ___________________________________________________________________ > Studio Tools vkire@pixar.com > Pixar Animation Studios http://www.pixar.com/ - To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/