From: Horms Subject: Re: [PATCH] gdbserver-xen: fix corefile access Date: Fri, 3 Mar 2006 10:28:06 +0900 Message-ID: <20060303012801.GA32698@verge.net.au> References: <67b15d3e07c51779c332ec14ca9067fb@cl.cam.ac.uk> <20060302121921.GC4184@verge.net.au> <7afbce7d550a4b6dcfa8c6809496d027@cl.cam.ac.uk> <20060303012020.GA23154@verge.net.au> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Return-path: Content-Disposition: inline In-Reply-To: <20060303012020.GA23154@verge.net.au> List-Unsubscribe: , List-Post: List-Help: List-Subscribe: , Sender: xen-devel-bounces@lists.xensource.com Errors-To: xen-devel-bounces@lists.xensource.com To: Keir Fraser Cc: xen-devel@lists.xensource.com List-Id: xen-devel@lists.xenproject.org On Fri, Mar 03, 2006 at 10:20:21AM +0900, Horms wrote: > On Thu, Mar 02, 2006 at 01:24:41PM +0000, Keir Fraser wrote: > > > > On 2 Mar 2006, at 12:19, Horms wrote: > > > > >>The correct fix is to update the xc_ptrace_core() interface to match > > >>the xc_ptrace() interface. Kip Macy made the latter SMP aware, but > > >>didn't fix up the former. > > >> > > >>It should be easy to do -- note how xc_ptrace() takes a domid on > > >>PTRACE_ATTACH, and vcpuid at all other times. xc_ptrace_core() should > > >>take a fd on PTRACE_ATTACH, and vcpuid at all other times. Since we > > >>don't dump SMP core files right now, vcpuid should either be ignored > > >>for the time being, or fail the call if vcpuid!=0. > > > > > >I didn't notice that, but I should have. > > > > > >Are you suggesting that xc_ptrace_core() should record the fd passed > > >to it on PTRACE_ATTACH and use that later, presumably in current_domid? > > >If so, yes that does look very easy. If not, can you explain a little > > >further? In any case, I'll look into it tomorrow. > > > > Yeah, you should record it the same way that xc_ptrace() records the > > domid. Really the two calls (xc_ptrace and xc_ptrace_core) should > > probably be merged -- we could pass an extra flag to PTRACE_ATTACH to > > indicate whether we are attaching to a coredump or to a live domain. > > Then we could get rid of xc_ptrace_core altogether. > > That sounds reasonable to me. Though internally the do different things, > so would the idea be to something like: > > rename xc_ptrace xc_ptrace_thread > make xc_ptrace a wapper for xc_ptrace_thread and xc_ptrace_core Sorry, scratch that. I think they can just be merged into one function, I'll try and get a patch together. Do you think it would be better to pass the isfile flag in the form of PTRACE_ATTACH|XC_PTRACE_FILE as the third argument to xc_ptrace, or parhaps use the last argument to xc_ptrace, edata (or perhaps the second last one, eaddr), which are unused for ATTACH. -- Horms