On 06/01/15 12:42, Michael S. Tsirkin wrote: > On Mon, Jun 01, 2015 at 12:35:34PM +0200, Laszlo Ersek wrote: >> On 06/01/15 12:23, Michael S. Tsirkin wrote: >>> On Mon, Jun 01, 2015 at 11:01:23AM +0200, Paolo Bonzini wrote: >>>> >>>> >>>> On 01/06/2015 09:28, Michael S. Tsirkin wrote: >>>>>> I don't feel overly strongly about it; just "mechanism, not policy" >>>>>> looks like a good tradition (well, good excuse anyway). >>>>> >>>>> Most users never see warnings. We ship it, we support it. >>>>> If we don't want to support it, let's not ship it. >>>> >>>> Then we should rm -rf half of QEMU. :) >>>> >>>> Seriously, I agree wholeheartedly with not baking policy into QEMU. A >>>> lot of QEMU command-line hacking really is just a shortcut to avoid >>>> continuous recompilation. I don't think it's reasonable to expect that >>>> it constitutes a stable API. >>>> >>>> Paolo >>> >>> Still, reserving part of the namespace for QEMU internal use >>> is *not* policy, it's just good engineering. >>> >>> How about we forbid adding files under "etc/" ? >>> >>> That would be enough to avoid conflicts. >> >> Some of the current fw_cfg files, like "bootorder", are not under >> "etc/". > > Well bootorder is there so at least it will always fail. > We do have stuff under /rom. > >> Hence the earlier proposal to restrict the user (to under opt/, >> IIRC), rather than ourselves. >> >> Thanks >> Laszlo > > How about we pre-pend opt/ to user-supplied names? > Will fix this without limiting user in any way.
(Sorry if this has been addressed in further messages down-thread; I'm not that far yet.) The issues here are that a user implementing custom firmware or OS-level code will have to: (a) take this prefix in account when writing the client code, (b) blob names have a max length of 55 chars IIRC (+ terminating NUL), any fixed prefix would decrease that. Not necessarily a problem, but something to be aware of, at least in the QEMU option parsing. Thanks Laszlo
