From: Simon Horman Subject: Re: [patch 0/3] xen: Extend xen kexec hypercall to return additional regions Date: Thu, 28 Feb 2008 12:35:18 +0900 Message-ID: <20080228033516.GA16840@verge.net.au> References: <20080227070105.949799153@vergenet.net> <1204142152.6798.32.camel@lappy> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Return-path: Content-Disposition: inline In-Reply-To: <1204142152.6798.32.camel@lappy> List-Unsubscribe: , List-Post: List-Help: List-Subscribe: , Sender: xen-devel-bounces@lists.xensource.com Errors-To: xen-devel-bounces@lists.xensource.com To: Alex Williamson Cc: xen-devel@lists.xensource.com, Ian Campbell , xen-ia64-devel@lists.xensource.com List-Id: xen-devel@lists.xenproject.org On Wed, Feb 27, 2008 at 12:55:52PM -0700, Alex Williamson wrote: > Hi Simon, > > I'm not seeing how this doesn't break the x86 COMPAT/CONFIG_COMPAT > code paths. Where does kexec_get_range_internal() get defined for > COMPAT in [1/3], and where is machine_kexec_get_xen() defined for > CONFIG_COMPAT in [2/3]? I'm fine with the ia64 parts if Ian/Keir want > to check them into the main tree, but there is some mixed indenting in > [3/3]. Thanks, Thanks for pointing out the white-space issue, I'll fix that up asap. As for the compat stuff... The way that I understand the code is that basically xen/common/kexec.c gets compiled twice using the following mechanism at the very bottom of the file: #if defined(CONFIG_COMPAT) && !defined(COMPAT) #include "compat/kexec.c" #endif Where xen/common/compat/kexec.c looks a bit like this: #define COMPAT ... #include "../kexec.c" So on the first stage of the run through xen/common/kexec.c, before we get to #include "compat/kexec.c", any code that is not protected by #ifdef COMPAT will be compiled. If CONFIG_COMPAT isn't defined (elsewhere) then that is the end and all is well. If CONFIG_COMPAT is defined, then we enter the second stage, where any code in "compat/kexec.c" not protected by #ifndef COMPAT is compiled. The important thing to note is that during this second stage, all the code compiled in the first stage still exists. kexec_get_range_internal() and all the functions that it calls either a) exist in xen/common/kexec.c and are compiled during the first stage, that is, they are protected by #ifndef COMPAT. b) present in other files that and are compiled independent of any CONFIG_COMPAT games. This means that kexec_get_range_internal(), and all the functions that it relies on can be called by both non-compat version of do_kexec_op(), which is compiled during the first stage, and the compat version of do_kexec_op() (compat_kexec_op()). The creation of the first version of do_kexec_op() is fairly obvious, it is compiled as you would expect on a casual glance. This occurs during stage one. The compilation of compat_kexec_op() is achieved by "#define do_kexec_op compat_kexec_op" in xen/common/compat/kexec.c, which essentially changes the function name for compilation in the second phase. I'll come right out and say that I don't fancy this two stage .c indirectly including itself approach. And I hope this change can help us move away from there, as well as providing the functionality extendability that this patch set describes. As to the extent of testing this code. I have run it on ia64 and it works. I have also compiled it on both i386 and x86_64. The latter uses CONFIG_COMPAT by default, which is the configuration I tested. -- Horms