]> code.ossystems Code Review - openembedded-core.git/commit
kmod: do not symlink config.guess/config.sub during autoreconf
authorGavin Li <gavin@matician.com>
Tue, 13 Apr 2021 05:29:23 +0000 (22:29 -0700)
committerRichard Purdie <richard.purdie@linuxfoundation.org>
Sun, 18 Apr 2021 10:29:05 +0000 (11:29 +0100)
commit89d675efd633b495daa4a3a57420b9c309497035
tree7b03c8eded0c2d731211258b5e1cd879599750d0
parent4479f810c1a3ab2badf4f9610c309bc0e23e2a5f
kmod: do not symlink config.guess/config.sub during autoreconf

I was encountering the following race condition on poky:

- automake-native does do_install.
- automake-native does do_populate_sysroot. This hardlinks config.guess
  and config.sub into ${D}.
- kmod-native does do_configure. This runs `autoreconf`, which runs
  `automake --add-missing` (symlinks config.guess/config.sub from
  recipe-sysroot-native to build dir), then runs `gnu-configize` (copies
  _its own_ config.guess/config.sub _on top_ of the already existing
  ones). Since the destinations already had symlinks, the copy would
  overwrite config.guess/config.sub in recipe-sysroot-native, which
  would in turn overwrite the same in ${D} due to being hardlinked.
- automake-native does do_package. The outhash is thus calculated on the
  clobbered config.guess/config.sub files.

With hash equivalency enabled, the different outhash produced a
different unihash, which kept me from reusing sstate between my laptop
and my build server. This race condition would happen only on the build
server (BB_NUMBER_THREADS = 32) but never on my laptop
(BB_NUMBER_THREADS = 6).

I didn't see the --install and --symlink flags being used by any other
recipe, so I removed them, and that fixed the issue.

Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
meta/recipes-kernel/kmod/kmod.inc