Re: [Xen-devel] kpartx for raisin hvm tests

[Stefano Stabellini <sstabellini@xxxxxxxxxx>]

On Wed, 1 Mar 2017, Gémes Géza wrote:
> 2017-03-01 20:48 keltezéssel, Stefano Stabellini írta:
> > On Wed, 1 Mar 2017, Gémes Géza wrote:
> > > 2017-02-27 23:52 keltezéssel, Stefano Stabellini írta:
> > > > On Wed, 22 Feb 2017, Géza Gémes wrote:
> > > > > On 2017-02-21 23:10, Stefano Stabellini wrote:
> > > > > > On Tue, 21 Feb 2017, Géza Gémes wrote:
> > > > > > > Hi,
> > > > > > > 
> > > > > > > I've tried to run the raisin test suite, while pv tests pass the
> > > > > > > hvm
> > > > > > > tests
> > > > > > > fail. I've identified a number of problems, starting with two
> > > > > > > small
> > > > > > > disk
> > > > > > > size
> > > > > > > to formating the whole disk and then being unable to install grub
> > > > > > > to
> > > > > > > the
> > > > > > > boot
> > > > > > > sector. I've trace down these problems into the incorrect
> > > > > > > invocation
> > > > > > > of
> > > > > > > the
> > > > > > > _create_loop_device function in scripts/lopartsetup.
> > > > > > > 
> > > > > > > My question: Will it be acceptable if I would replace this part of
> > > > > > > the
> > > > > > > code
> > > > > > > with a kpartx call? Or introducing kpartx is a too big change in
> > > > > > > the
> > > > > > > list
> > > > > > > of
> > > > > > > dependencies?
> > > > > >     I understand that kpartx makes things much easier, but before
> > > > > > introducing it as a dependency, I would like to understand this
> > > > > > problem
> > > > > > a bit better.
> > > > > > 
> > > > > > Why is _create_loop_device invoked incorrectly? Is it index or
> > > > > > offset
> > > > > > that is calculate incorrectly?
> > > > > Hi Stefano,
> > > > > 
> > > > > In scripts/lopartsetup:56 unit="`fdisk -lu $filename 2>/dev/null |
> > > > > grep -e
> > > > > "^Units = " | cut -d " " -f 9`" . Using ubuntu 16.04 (fdisk coming
> > > > > from
> > > > > util-linux-2.27.1-6ubuntu3.2) this yields to an empty variable, as:
> > > > > 
> > > > > $sudo fdisk -lu /tmp/tmp.x9UN6uxaG2/busybox-vm-disk 2>/dev/null
> > > > > 
> > > > > Disk /tmp/tmp.x9UN6uxaG2/busybox-vm-disk: 60 MiB, 62914560 bytes,
> > > > > 122880
> > > > > sectors
> > > > > Units: sectors of 1 * 512 = 512 bytes
> > > > > Sector size (logical/physical): 512 bytes / 512 bytes
> > > > > I/O size (minimum/optimal): 512 bytes / 512 bytes
> > > > > 
> > > > > Because of this both unit and offset are wrong ( offset=`echo $i | tr
> > > > > -s "
> > > > > " |
> > > > > cut -d " " -f 2`, where i=fdisk -lu $filename 2>/dev/null | grep -e
> > > > > "^$filename")
> > > > > 
> > > > > As I think that different versions of fdisk will produce different
> > > > > results, we
> > > > > either introduce an additional logic for the fdisk version, either
> > > > > change
> > > > > this
> > > > > part completely.
> > > > It doesn't look like fdisk changed output in this case. It looks like
> > > > the disk doesn't have any partitions in it. Am I right?
> > > > 
> > > > It would be easy to add support to lopartsetup to detect disks without
> > > > partitions, and deal with them correctly, without bringing in kpartx.
> > > > However, this scenario shouldn't occur, because lopartsetup is only
> > > > called by create_one_partition, right after creating a partition on the
> > > > disk.
> > > > 
> > > > Do you know why create_one_partition doesn't work as expected?
> > > Hi Stefano,
> > > 
> > > Sorry for the late answer. The only change I've did on Ubuntu 16.04 was to
> > > increase the hvm disk size to 60 MB.
> > > 
> > > In the meanwhile I set up an ubuntu 14.04 test system and for the first
> > > time
> > > tried to run the tests as an ordinary user instead of root. I've found a
> > > set
> > > of problems. I've made a patch
> > > (https://github.com/geza-gemes/raisin/commit/8a1227d96697a4d8be9130fd9b16404decbe7605)
> > > for those.
> > That's a good patch, thank you. Could you please submit it to xen-devel?
> > 
> > 
> > > Although this fixes the problem of running the tests as non-root,
> > > it turned out, that even on ubuntu 14.04 the 20MB disk is not enough for
> > > the
> > > hvm guest. I'll try to find the suitable disk size there and if successful
> > > I'll move back to ubuntu 16.04.
> > The tiny disk size was only meant to be used for busybox guests. Also, I
> > still don't understand what's wrong with the fdisk code. Could you
> > please check if the partition table has been setup correctly on the
> > disk? In other words, does create_one_partition work correctly?
> 
> Hi Stefano,
> 
> I've sent the patch to xen-devel as an ordinary e-mail as I didn't manage to
> make git send-email work with my gmail account :-(.

It takes some time to get right.


> My experience with ubuntu 14.04 and 16.04 shows, that the disk size is too
> small to accommodate the kernel and the initrd. Increasing to 60MB solves that
> problem, but the image is still not bootable (grub cannot be installed on
> 16.04 as it looks like the whole disk gets formated, and grub cannot load the
> kernel on 14.04.). I'll continue investigating 14.04 tomorrow.

All right, thanks. For debugging the fdisk problem, you might as well
increase the disk size as much as it takes.

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
https://lists.xen.org/xen-devel