On 06/12/2021 14:29, Juergen Gross wrote: > In contrast to the PFN of the console ring page the event channel of > the console isn't being set automatically by xc_dom_build_image(). > > Call xc_hvm_param_set() explicitly for that reason. > > Signed-off-by: Juergen Gross <[email protected]>
So, technically, Acked-by: Andrew Cooper <[email protected]> However... That said, there is a distinct lack of joined-up thinking in this interface. It makes no sense whatsoever to for xc_dom_build_image() to build the grant, but leave the evtchn to the caller. And indeed, xg_dom_x86.c: start_info->console.domU.evtchn = dom->console_evtchn; we set it up on the PV side of things. So I think the proper fix is to wire up the HVM side and prevent the callers needing to do this. Furthermore, I doubt we skip setting up the xenstore connection. Really, the users of xc_dom_build_image() want a console Y/n, xenstore Y/n type interface, and judging by the fields we've already got, that can reasonably be done on the non-zero-ness of *_evtchn (It is also weird that the caller is required to bind the evtchn, but that's so baked into the API that I'd need to rearrange code between Ocaml daemons to make use of a "library code allocates evtchn+grant together" option.) ~Andrew
