From: Ian Campbell Subject: Re: [PATCH 00/04] Kexec / Kdump: Release 20061122 (xen-unstable-12502) Date: Wed, 29 Nov 2006 09:35:28 +0000 Message-ID: <1164792929.3336.228.camel@localhost.localdomain> References: <1164788262.30816.41.camel@localhost> Mime-Version: 1.0 Content-Type: text/plain Content-Transfer-Encoding: 7bit Return-path: In-Reply-To: <1164788262.30816.41.camel@localhost> List-Unsubscribe: , List-Post: List-Help: List-Subscribe: , Sender: xen-devel-bounces@lists.xensource.com Errors-To: xen-devel-bounces@lists.xensource.com To: Magnus Damm Cc: Ian Pratt , Kazuo Moriwaka , xen-devel@lists.xensource.com, Akio Takebe , Magnus Damm , Isaku Yamahata , Horms List-Id: xen-devel@lists.xenproject.org On Wed, 2006-11-29 at 17:17 +0900, Magnus Damm wrote: > On Wed, 2006-11-29 at 07:54 +0000, Keir Fraser wrote: > > On 29/11/06 4:30 am, "Magnus Damm" wrote: > > > In user space the kexec-tool then builds an elf header which points > > > out where the notes are located in machine address space (using > > > /proc/iomem). This header is then passed on to the secondary crash > > > kernel which (for some reason) compacts all per-cpu PT_NOTE program > > > headers into one which will be present in the final vmcore image. At > > > this compacting stage we need to have data present for _all_ cpus, and > > > the data for cpus that are not present should contain just zeros. > > > > How does kexec discover how many cpus is 'all cpus'? > > The kexec tool creates (at load time) one PT_NOTE program header per > note exported through /proc/iomem. The number of PT_NOTE program headers > is the same as the NR_CPUS constant in the hypervisor. The guest kernel creates entries in /proc/iomem by calling kexec_get_cpu(cpu) until it returns EINVAL. This currently happens when cpu>NR_CPUS. I think this function should return EINVAL for cpu>num_present_cpus() instead. Xen doesn't currently do PCPU hotplug and this wouldn't be the only thing that would need fixing if it ever does (percpu data would be another one I think ;-)). This would cause the tools to create notes only for CPUs which really exist. That would make the loop in machine_crash_kexec() unnecessary. Cheers, Ian.