On 04/25/19 12:23, Philippe Mathieu-Daudé wrote: > On 4/25/19 10:29 AM, Li Qiang wrote: >> >> >> Gerd Hoffmann <kra...@redhat.com <mailto:kra...@redhat.com>> 于2019年4月 >> 25日周四 下午4:15写道： >> >> On Wed, Apr 24, 2019 at 09:16:56AM +0800, Li Qiang wrote: >> > Thomas Huth <th...@redhat.com <mailto:th...@redhat.com>> 于2019年4 >> 月24日周三 上午12:29写道： >> > >> > > Is this endianess-safe? Or do you need to byteswap >> reboot_timeout if the >> > > host and guest endianess does not match? >> > >> > Good question! >> > >> > IIUC, the qemu fw_cfg store the 'file' entry data just in byte stream. >> >> No. Integers are defined to be little endian. See fw_cfg_add_i64() for >> example, there is an explicit cpu_to_le64() call for that. >> >> >> >> Yes, for the fw_cfg 'integer' entry it is stored as little endian. >> But for the fw_cfg 'file' entry interpred as an integer, there is no >> specify the endianess. > > I agree with Li, the endianess of 'reboot-timeout' is not clear. > > From docs/specs/fw_cfg.txt: > > === All Other Data Items === > > Please consult the QEMU source for the most up-to-date > and authoritative list of selector keys and their respective > items' purpose, format and writeability. > > So checking the git history, this code was introduced in commit > ac05f3492421, very similar to commit 3d3b8303c6f8 for the > 'boot-menu-wait' entry, which explicitely use little-endian, so I think > it is safe to consider it little-endian and add a comment about its > endianess (referring the previous commits in the patch description).
OVMF consumes "boot-menu-wait", so that one must remain LE. OVMF doesn't consume "reboot-timeout", so I can't really comment on it. I guess, if a named fw_cfg file that already exists doesn't explicitly define the endiannesses of its integer fields, err for safety and opt for LE. For new fw_cfg files with integers in them, choose the endianness explicitly and document it. Thanks Laszlo
