On 5/16/19 12:12 PM, Jason Gunthorpe wrote: > On Thu, May 16, 2019 at 06:59:45PM +0000, Adit Ranadive wrote: >>>> diff --git a/include/uapi/rdma/vmw_pvrdma-abi.h b/include/uapi/rdma/vmw_pvrdma-abi.h >>>> index 6e73f0274e41..8ebab11dadcb 100644 >>>> +++ b/include/uapi/rdma/vmw_pvrdma-abi.h >>>> @@ -49,7 +49,9 @@ >>>> >>>> #include <linux/types.h> >>>> >>>> -#define PVRDMA_UVERBS_ABI_VERSION 3 /* ABI Version. */ >>>> +#define PVRDMA_UVERBS_NO_QP_HANDLE_ABI_VERSION 3 >>>> +#define PVRDMA_UVERBS_ABI_VERSION 4 /* ABI Version. */ >>> >>> Don't mess with ABI version when all you are doing is making the >>> response or request struct longer. >> >> Hmm, I thought we always had to update the ABI in case of such >> changes as well? > > No. Only if you break the ABI. Don't break the ABI. > >> Previously, we weren't copying out qpresp to udata at all and this >> adds a new response to user-space. Also, wouldn't older rdma-core >> probably break because of this since it assumes the qpn reported is >> the virtual one? I guess that's a bug already from a backward compat >> perspective.. > > Indeed, don't break the ABI :) Let me send out a fix for that first and then I'll look at removing the ABI updates to this patch. > > Jason >
