bug#43344: "basic" system tests fail (and all the other ones) on guix master

[Danny Milosavljevic]

Hi Leo,

On Thu, 17 Sep 2020 10:05:44 -0400
Leo Famulari <leo@famulari.name> wrote:

> On Wed, Sep 16, 2020 at 09:36:49PM -0400, Mark H Weaver wrote:
> > > On Wed, Sep 16, 2020 at 04:52:45PM +0200, Danny Milosavljevic wrote:  
> > >> commit 692d0626557451c4b557397f20b7394b612d0289
> > >> Author: Christoph Hellwig <hch@lst.de>
> > >> Date:   Tue Sep 1 11:59:41 2020 +0200
> > >> 
> > >>     block: fix locking in bdev_del_partition  
> 
> > Next, it probably makes sense to test 5.8.9 with the above commit
> > reverted.  Would someone like to try it?  If that works, we can avoid
> > the bisection, resume kernel updates, and revert this change across all
> > of our kernels until a better solution is found.  
> 
> Using linux-libre 5.8.9 with that commit reverted, `guix system vm` does
> work again.

Thanks for testing it!

What the commit does is lock the drive whose partition table is being
modified.  That sounds like a good idea in general.  But parted does not
expect the error code ENOMEM and thus misjudges the situation.

See also:

  
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=88ce2a530cc9865a894454b2e40eba5957a60e1a

for the fix that goes into Linux 5.9.

We could also patch parted to also accept ENOMEM in addition to ENXIO if we
wanted to.  I wouldn't (it diverges from what upstream is doing--too much work).

pgpcVfkMPeoRC.pgp
Description: OpenPGP digital signature