On 02/20/2017 07:10 PM, Paolo Bonzini wrote: > > On 16/02/2017 20:36, Eric Blake wrote: >> On that grounds, you already need the 'if (info)' for more than just the >> free, so this code motion is no longer quite as important. But now I'm >> noticing that it looks weird because you are freeing an input parameter. >> Generally, transfer semantics like that are screwy - it's probably >> better if the caller of qemu_system_guest_panicked() is the one freeing >> info, rather than requiring that the caller pass in malloc'd memory that >> gets freed as a side effect and must not be referenced afterwards in the >> caller. In other words, I think the code motion is unnecessary, but >> that the qapi_free_GuestPanicInformation() call is probably in the wrong >> function to begin with. > Even better then would be to just pass a CPUState* and let > qemu_system_guest_panicked get the GuestPanicInformation via the QOM > property. > > But for 2.9, we only need to change the union. Eric, can you do that > for us since my QAPI-fu is limited? > > Paolo > give me 5 minutes, I have patches for that, received them today.
Den
