On Tue, Apr 08, 2025 at 12:06:51PM +0000, Anthony PERARD wrote: > On Mon, Apr 07, 2025 at 06:31:06PM +0200, Marek Marczykowski-Górecki wrote: > > There is one more difference: the cpio.gz contains the whole thing > > twice. Once as rootfs for dom0 and then another as boot/initrd-domU. > > Dropping .tar.gz is probably a good idea at some point, so I'm okay with > > such comment added. But I imagine some future tests may benefit from > > just one layer, which may want introducing another cpio.gz without > > boot/initrd-domU included. In fact, even right now that might be useful > > - for example dom0less arm64 test uses just busybox for domU rootfs, not > > the whole archive (so with my changes dom0 rootfs has unused > > boot/initrd-domU included). > > In such case, would it make sense to prepare several initrd.cpio.gz? A > common one, then having overlays of files we want to add or replace to > the initrd of spefic test. I think osstest used to do something like > that, with command that would be: > cat common.cpio.gz overlay.cpio.gz > initrd.cpio.gz > > If it matter, is seem that the "file.cpio" used to be made with > `cpio -Hnewc -o`
This is already how final initrd is constructed with my changes. Having said that, maybe test-artifacts should not include boot/initrd-domU in initrd.cpio.gz initially, but have a separate cpio that has just boot/initrd-domU there. Most tests do use this domU initrd, but since they need to concatenate something anyway, maybe starting with two files instead of one doesn't make much difference? So, it would be: - initrd.cpio.gz - plain rootfs, for dom0 (or domU in dom0less tests) - initrd-in-boot.cpio.gz - the above initrd.cpio.gz packed again as boot/initrd-domU What do you think? -- Best Regards, Marek Marczykowski-Górecki Invisible Things Lab
signature.asc
Description: PGP signature
