On Fri, Jul 06, 2018 at 02:41:18PM +0200, Paolo Bonzini wrote: > On 06/07/2018 03:18, David Gibson wrote: > >> Ok, but why would it even fail? Maybe it's not this case, but even if > >> it's not fatal for the OS, generating different device trees silently > >> seems like a recipe for Heisenbugs. > > Yes, quite. Especially since the most likely errors I can see here > > would actually indicate something has already gone horribly wrong with > > the device tree construction, so a missing clock-frequency is the > > least of our troubles. > > > > I think using _FDT() here would be a better approach. > > Or qemu_fdt_setprop_cell, which is there exactly for this reason. > Volunteers needed to report it in checkpatch! :) > Sure, except it uses different parameters. Why don't you come up with an implementation that is acceptable to you ? I was asked earlier to add the comment, which I did, only to now be told that it is insufficient. This can go on forever. At this point, I'll be happy to send a revert request.
Guenter
