Alexander Graf wrote: > On Sep 16, 2008, at 12:22 PM, Zhang, Xiantao wrote: > >> Alexander Graf wrote: >>> On Sep 16, 2008, at 9:22 AM, Zhang, Xiantao wrote: >>> >>>> Alexander Graf wrote: >>>>> KVM as is assumes that the kernel obj dir and the kernel source >>>>> dir are at the same location. This is true for most self-built >>>>> vanilla kernels, but some distributions split these up (e.g. >>>>> SUSE). >>>>> To keep compatible and have users have a good experience on >>>>> building KVM on any distribution, this patch attempts to rebuild >>>>> the logic from the kernel Makefile as closely as possible. With >>>>> it I successfully built KVM on a current SUSE system. >>>>> >>>>> Signed-off-by: Alexander Graf <agraf@xxxxxxx> >>>>> >>>>> Please check and see if it breaks the build process for anyone >>>>> else. Building with IA64 on SUSE-kernels is still broken due to >>>>> similar but separate problems. >>>> >>>> Hi, Graf >>>> what problems did you meet when building kvm/ia64 ? >>>> Xiantao >>> >>> >>> This is the list of stuff I had to modify to get kvm-74 to run. I >>> haven't tried anything more recent yet. >>> >>> 1) -mno-sdata does not work. Qemu CFLAGS need to be -msdata >> >> Which version of GCC are you using ? > > # gcc --version > gcc (SUSE Linux) 4.3.2 20080815 (prerelease) [gcc-4_3-branch revision > 139129] > >> >> >>> 2) In qemu/ia64.ld SEARCH_DIR is invalid for SUSE. It shouldn't >>> search in /usr/ia64-linux but in /usr/ia64-suse-linux. >> You can make a patch to fix it. > > Right, this one should be really easy to fix. > >>> 3) TARGET_PAGE_BITS is wrong >> Currently, we just set TARGET_PAGE_BITS to 16 by hardcoding in Qemu. >> In the future, we should add the logic to make it equal to kernel's >> PAGE_SHIFT. > > From what I read on the KVM IA64 ML, this is not the first time > someone ran into this issue. Sounds like a really good idea to add the > logic ;-). > >>> 4) kernel/ia64/Makefile.pre assumes to find .S files in $KERNELDIR. >>> They are in $SOURCEDIR. >> >> There is no issue here, I think. You have to set corrent $KERNELDIR >> through ./configure --kerneldir=xxxx once target machine and build >> machine is not same. > > No, since $KERNELDIR is the build dir, as that one contains all the > configurations. The source dir only contains the sources. > >>> 5) kernel/ia64/Makefile.pre suffers from the same problem my >>> original fix for x86 addressed. >> >> Ditto. >> >>> 6) The IA64 kernel build system is broken. The file asm-ia64/nr- >>> irqs.h gets generated during the kernel build, but is not exported >>> as obj, so it won't be in the split obj directory. Since the >>> source-dir is clean though, the file simply does not exist on SUSE >>> rpms. >> >> You mean the source-dir is clean ? So the generated header files >> including (linux/autoconf.h, asm-offsets.h...)should be removed >> wholely? I don't think they can be built with a clean source for >> most modules. > > This is what I'm talking about all the time :-). For SUSE kernels > there are two distinct directories: > > # ls -l /usr/src > drwxr-xr-x 24 root root 4096 Aug 27 14:51 linux-2.6.27-rc4- > HEAD_20080825213702 > drwxr-xr-x 3 root root 4096 Aug 25 23:28 linux-2.6.27-rc4- > HEAD_20080825213702-obj > > whereas the normal directory contains the sources and the -obj > directory contains all files that were generated during the build. > Scratch that with the 2 rpm files - they are all in kernel-sources. > > # ls -l > /usr/src/linux-2.6.27-rc4-HEAD_20080825213702-obj/ia64/default/ total > 568 > -rw-r--r-- 1 root root 76287 Aug 25 23:27 .config > -rw-r--r-- 2 root root 554 Aug 25 23:28 Makefile > -rw-r--r-- 1 root root 468615 Aug 25 23:51 Module.symvers > drwxr-xr-x 3 root root 4096 Aug 26 18:01 arch > drwxr-xr-x 5 root root 4096 Aug 26 18:01 include > drwxr-xr-x 2 root root 4096 Aug 26 18:01 include2 > drwxr-xr-x 2 root root 4096 Aug 25 23:28 kernel > drwxr-xr-x 6 root root 4096 Aug 26 18:01 scripts > > The problem is now that a lot of people assume these two directories > to be one. This is the default mode when you compile a kernel. SUSE > does not take this route though, and hasn't for quite a while now. I > don't know about other distributions, but I guess we're not the only > ones ;-). Okay, I see now. But due to shortage of Suse OS at hand, I can't try it . Anyway once you found any issues related to the compability with Suse OS, welcome to post patches to fix them. :) Xiantao -- To unsubscribe from this list: send the line "unsubscribe kvm-ia64" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
