From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S964939AbWEOOVj (ORCPT ); Mon, 15 May 2006 10:21:39 -0400 Received: (majordomo@vger.kernel.org) by vger.kernel.org id S964943AbWEOOVj (ORCPT ); Mon, 15 May 2006 10:21:39 -0400 Received: from e5.ny.us.ibm.com ([32.97.182.145]:21894 "EHLO e5.ny.us.ibm.com") by vger.kernel.org with ESMTP id S964939AbWEOOVi (ORCPT ); Mon, 15 May 2006 10:21:38 -0400 Subject: Re: [RFC] Hugetlb demotion for x86 From: Dave Hansen To: Adam Litke Cc: Christoph Lameter , linux-mm@kvack.org, linux-kernel@vger.kernel.org In-Reply-To: <1147363859.24029.134.camel@localhost.localdomain> References: <1147287400.24029.81.camel@localhost.localdomain> <1147363859.24029.134.camel@localhost.localdomain> Content-Type: text/plain Date: Mon, 15 May 2006 07:20:17 -0700 Message-Id: <1147702817.6623.27.camel@localhost.localdomain> Mime-Version: 1.0 X-Mailer: Evolution 2.4.1 Content-Transfer-Encoding: 7bit Sender: linux-kernel-owner@vger.kernel.org X-Mailing-List: linux-kernel@vger.kernel.org On Thu, 2006-05-11 at 11:10 -0500, Adam Litke wrote: > Yes, the SIGBUS issues are "fixed". Now the application is killed > directly via VM_FAULT_OOM so it is not possible to handle the fault from > userspace. For my libhugetlbfs-based fallback approach, I needed to > patch the kernel so that SIGBUS was delivered to the process like in the > days of old. Maybe this could be off-by-default behavior that can be enabled with a special mmap flag or madvise, or something similar. It seems that apps don't want to get SIGBUS for low memory. But, if they have _asked_ for it, perhaps they'd be a bit more willing. (BTW, I fixed the bogus linux-mm cc, finally ;) -- Dave