wireguard has been moved from the packages feed to the main sources. That is the reason.
The funny thing here is that wireguard has nothing to do with the fastpath implementation.
But as gwlim's "fastpath" patches include tons of changes to things unrelated to fastpath itself, those things may break when there are external changes and the patches are not updated accordingly. (wireguard is currently at version 20171127 at both master and lede-17.01, and the gwlim 17.01 patch tries to change that to 20170629)
I don't know whether it's easily possible, I never did it. But it has problem with SQM and maybe other stuff as well.
Ask those who applied it recently.
As far as I can tell, fast path circumvents parts of the linux kernel stack that sqm-scripts require; so to me it looks like they are incompatible by design, at least on the same interface. I believe dissent1's patchset will disable fast path if there is a shaper active on the to be fast-pathed interface, so in that regard there is no problem, on the other hand linux traffic shaping and fast path seem mutually exclusive (functionally, if you manage to instantiate both on the same interface your shaper is only going to see thoose packets that are not handled by fast path).
I do a script .
I modified September Gwlim's script and it works well (only path 066-... not work). Checkout for kernel 4.4.87, actually is kernel 4.4.102 and not work.
@dissent1 can this latest patch possibly be applied to openwrt trunk or cc15.05.1 ? tried to modify several times in order to patch cc15.05.1 but failed, but for latest openwrt trunk, even the kernel is same as the lede master, it still fails, don't know why.
@jgvega After make menuconfig, what else needs to be selected besides the "Target System”, “Target Profile” ? Is the patch automatic selecting the right modules?
@jgvega I have selected: Target System (Broadcom BCM47xx/53xx (ARM)), Target Profile (D-Link DIR-885L) and also selected kmod-fast-classifier plus a few more packages as in post 547, above. I am getting the bellow error. I am building on a Fedora 27 VM.
Error:
lede/build_dir/toolchain-arm_cortex-a9_gcc-6.3.0_musl-1.1.16_eabi/gcc-6.3.0/gcc/ubsan.c:1474:23: error: ISO C++ forbids comparison between pointer and integer [-fpermissive]
|| xloc.file == '\0' || xloc.file[0] == '\xff'
^~~~
make[5]: *** [Makefile:1085: ubsan.o] Error 1
make[5]: Leaving directory
Just wondering if this is the correct patch for my configuration:
Clarification, sqm is not broken with sfe, but both are functionally mutually exclusive; I believe @dissent1's patch will disable sfe if there is an active sqm instance on an interface, while @gwlim's patch set will give priority to sfe and effectively disable sqm (or rather sqm will only see those ~128 packets the are needed to trigger the fast pathing). I do not believe this will change any time soon.