On 11/17/2017 12:46 PM, Eric Blake wrote: >> >> The test fails with -m32 and probably also on Big Endian architectures, >> because the bitmap hash differs. > > Oh my. Thanks for the rapid testing. > >> We could "fix" this by replacing the 2100 by a 2102, so for both bit >> widths rounding is the same.
Actually, thinking about it more, are we sure that a .qcow2 image created on an -m32 host can be properly loaded on a 64-bit host, despite their different rounded-up sizes for the bitmaps? If the difference in hash is due only to the in-memory representation, but the on-disk layout is identical, then we are safe; but if the difference in hash is visible on-disk, then we have a real bug in our qcow2 specification or implementation of persistent bitmaps. -- Eric Blake, Principal Software Engineer Red Hat, Inc. +1-919-301-3266 Virtualization: qemu.org | libvirt.org
signature.asc
Description: OpenPGP digital signature
