Gilles Espinasse wrote: > I have modified my build system during util-linux-2.22 time to supply > needed kernel modules inside the chroot while building.
I have always discouraged users from using modules. They are rarely needed in LFS. > For loop, there is a special trick to make loop available without > depending on running loop module available. In my like chap-6.6, I > do # mknod one /dev/loop allow losetup (from inside) to mount loop > module (from running kernel outside the chroot) # That will allow > some tests using loop to run rm -f /dev/loop0 && mknod /dev/loop0 b 7 > 0 I'm not sure why you want to delete loop0 and recreate it. It should be available from /dev that is bind mounted to chroot. I admit that I've never felt the need to use a loop mount from within chroot. > If running as root, I am a bit afraid too by the big warning banner > on test start. I am unsure that running as root could destroy > something on the host, like mounted loop files as loop numbers are > hardcoded in the tests. I agree with that, but you should really be using a separate build machine or a virtual machine for those types of tests. > Anyway, I have manually tested running util-linux tests as root. For > that, I build again util-linux after full LFS has build (so modprobe > was available). That's a good thought. > Running more tests with running kernel modules was a bit of fun to > discover some trick but I do not think that was a great success. I > fail to find an appropriate recipe for mdadm (it fail after a few > tests and I haven't seen a log from someone running the full mdadm > test suite). And I am really unsure running mdadm tests would not > detroy a config using mdadm for real on boot device. Another vote for a test/virtual system. > parted-3.1 fail one test (that are not run without the scsi_debug > kernel modules) and emit many messages (not enough memory). I haven't > investigated that issue. A lot of regression tests are skipped if certain capabilities are not present. We are not doing development tests in LFS, but instead generally doing sanity tests to make sure we are building the applications correctly. If I were doing application development, I'd make sure I had all the necessary hardware and all dependent software when testing. Of course there are some applications where you can't test everything at one time. For example, tcl has some tests that require Windows and others that require Linux. These requirements within the test suite are, of course, mutually incompatible. -- Bruce -- http://linuxfromscratch.org/mailman/listinfo/lfs-dev FAQ: http://www.linuxfromscratch.org/faq/ Unsubscribe: See the above information page
