RE: [PATCH V2 3/5] RDMA/core: transport-independent access flags

["Steve Wise" <swise-7bPotxP6k4+P2YhJcF5u+vpXobYPEAuW@public.gmane.org>]

> -----Original Message-----
> From: linux-rdma-owner-u79uwXL29TY76Z2rM5mHXA@public.gmane.org [mailto:linux-rdma-owner-u79uwXL29TY76Z2rM5mHXA@public.gmane.org] On Behalf Of Sagi Grimberg
> Sent: Thursday, July 02, 2015 8:23 AM
> To: Steve Wise; Hefty, Sean; dledford-H+wXaHxf7aLQT0dZR+AlfA@public.gmane.org
> Cc: roid-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org; sagig-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org; linux-rdma-u79uwXL29TY76Z2rM5mHXA@public.gmane.org; jgunthorpe-ePGOBjL8dl3ta4EC/59zMFaTQe2KTcn/@public.gmane.org; infinipath;
> eli-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org; ogerlitz-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org
> Subject: Re: [PATCH V2 3/5] RDMA/core: transport-independent access flags
> 
> On 7/2/2015 4:17 PM, Steve Wise wrote:
> > On 7/2/2015 1:22 AM, Sagi Grimberg wrote:
> >> On 6/30/2015 8:10 PM, Hefty, Sean wrote:
> >>>> I suggest to start consolidating to ib_create_mr() that receives an
> >>>> extensible ib_mr_init_attr and additional attributes can be mr_roles
> >>>> and mr_attrs.
> >>>
> >>> I think this makes sense, but does it really help?
> >>> If the end result is that the app and providers basically
> >>> end up switching on mr_attr::type, we end up reducing the
> >>> number of APIs, but the code complexity remains the same.
> >>>
> >>
> >> I think it will reduce code complexity. First, the callers
> >> will have a single API to allocate a memory key. I'm not sure why you
> >> say that app needs to switch on mr::type, it knows which type it wants.
> >>
> >
> > I dont see how doing this is less complex:
> >
> > attr = FASTREG
> > mr = create_mr(attr)
> >
> > vs:
> >
> > mr = lloc_fast_reg_mr()
> 
> The simplification is that you can facilitate changes like your
> mr_roles and any other things we may want to add to mr allocation
> easily instead of suggesting wrappers over wrappers over wrappers.
> 
> This is why I suggested it in the first place.
> 

Regardless of have one create_mr() or having create_sig_mr(), create_fast_reg_mr(), get_dma_mr(), we'll still need to define the roles/attributes to make them transport independent, and we'll still need to have an entry point that will compute access_flags given roles and attributes to be used when posting frmrs...  I don't see wrappers over wrappers over wrappers.

I'm not really opposed to creating a single entry point, but rather I'm not really seeing vast utility of doing so.  

Anybody else have an opinion?

Steve.




--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html