On 12/10/2018 09:32 PM, tristram...@microchip.com wrote: >> On Fri, Dec 07, 2018 at 07:18:43PM +0100, Marek Vasut wrote: >>> From: Tristram Ha <tristram...@microchip.com> >>> >>> Factor out common code from the tag_ksz , so that the code can be used >>> with other KSZ family switches which use differenly sized tags. >> >> I prefer this implementation over what Tristram recently submitted. It >> is also what we suggested a while back. However, since then we have >> had Spectra/meltdown, and we now know a function call through a >> pointer is expensive. This is the hot path, every frame comes through >> here, so it is worth taking the time to optimize this. Could you try >> to remove the ksz_tag_ops structure. xmit looks simple, since it is a >> tail call, so you can do that in ksz9477_xmit. Receive looks a bit >> more complex. >> >> I also think for the moment we need it ignore PTP until we have the >> big picture sorted out. If need be, the code can be refactored yet >> again. But i don't want PTP holding up getting more switches >> supported. > > Yes, as you may already know, what Marek submitted was my previous > attempt to support different switches in tag_ksz.c.
I adjusted the code quite a bit, please look again. > So this ksz_tag_ops is still not acceptable? As I understand the kernel is > using this mechanism all over the places. > > What is left is a direct copying of the transmit and receive functions for > each > new switch tail tag format. Maybe a macro would work ? Or some code runtime-patching ? -- Best regards, Marek Vasut
