On Tue, 13 Aug 2013 22:50:05 +0200
Rafał Miłecki <zaj...@gmail.com> wrote:
> 2013/8/11 Nathan Hintz <nlhi...@hotmail.com>:
> > Is there a penalty in synchronizing (DMA) more of the buffer than
> > is needed? Here's a modified version of the patch that mostly
> > eliminates the unneeded synchronization (there is another 24 bytes
> > of pad in the RX header that could be eliminated - in both
> > directions); this hasn't been tested in any way (not even compiled)
>
> I tested your patch (attached incremental version if someone is
> interested), but it didn't affect performance :(
Does the synchronization occur a page at a time? If so, it's probably
better the way it was; except that the change to only sync the poisoned
rx header back to the device might still be of value if there are
situations where the entire rx buffer crosses a page boundary but the
rx header does not. That is, it would be more likely to only need to
sync one page instead of two in this case. It depends how the
synchronization works under the hood.
Have you tried this recent change?
https://dev.openwrt.org/changeset/37763
--
Nathan
_______________________________________________
openwrt-devel mailing list
openwrt-devel@lists.openwrt.org
https://lists.openwrt.org/cgi-bin/mailman/listinfo/openwrt-devel
