After running into several issues with the RC1 release build of LEDE on my TP-Link Archer C7 V2, I decided to compile my own build with only the packages and kernel modules I need. Considering the C7 is a quite popular device, I want to share my build with the community.
There is no release cycle for this build. I will only update it here when I update my C7. Feedback welcome!
Yes! Open /etc/config/dnscrypt-proxy with nano or vim and change option resolver 'fvz-anyone' to option resolver 'default.ns1.adguard.com'. Then restart DNSCrypt with service dnscrypt-proxy restart. Simple adblock is also available in LuCI.
Do you offer any quality assurance?
Sort of. I push updates manually after testing the builds on my own C7, which prevents major issues with the releases you can find here. The nightly builds are untested!
My clients are unable to connect to the internet after flashing this firmware!
Just reboot your router once and it should work. If it still doesn't work, make sure that dnscrypt-proxy is enabled and working. SSH to your router and do service dnscrypt-proxy enable; service dnscrypt-proxy restart.
Can I get this firmware even faster?
Yes. If you don't use PPPoE or another tunneling protocol, you can remove the line option mtu_fix 1 in /etc/config/firewall to squeeze out another 20-30Mbps.
Changelog
31-05-2018
Fix 061-remove-pppoa-luci.patch
Remove 063-add-luci-mips-processor-march.patch
Upstream updates
2018-05-27
Patches updated
Use wolfssl for hostapd
Replaced dnsmasq with dnsmasq-full
Switched to binutils 2.30
Upstream updates
2017-12-08
Fix compilation issues (odhcpd-ipv6only was "magically" added by upstream updates and caused conflicts)
Add rfkill support (might fix 5GHz issues, credits to @cliobrando)
Add kmod-usb-printer
Upstream updates
2017-11-06 v2
All patches ported to kernel 4.9
Switched to kernel 4.9
Enabled BBR & FQ (adjust your sysctl.conf)
Switched to GCC 7.2
Switched to binutils 2.29.1
.config file added to repo
2017-11-06
Switched from gwlim's to dissent1's Shortcut FE patches
Removed 464 to prevent bug (high CPU load, crash)
Patch updates
Upstream updates
2017-09-10
GRO support added
Patch updates
Upstream updates
2017-08-09
ssdk removed (useless for now)
Patch updates
Minor improvements
Nightlies fixed
Upstream updates
2017-08-03
Updated patches
Upstream Updates
2017-07-02
Re-added ath9k-htc and a couple of other recently "lost" kmods
Upstream updates
2017-07-01
Fixed issue with FastPath not being included
Adjusted several patches
Upstream updates
2017-06-29
Switched to MIPS 24kc, which means you can now use the official LEDE repository to install and update packages
Hardware NAT support added
Qualcomm Fast Path added
Updated all of gwlim's patches & added a couple of new ones
Added ds-lite, map-t, xl2tpd and ssdk_sh
tor removed (install with opkg if you need it)
Used binutils 2.28
Upstream updates
2017-05-22
Added ath9k firmware modules
Added openvpn-easy-rsa
Upstream updates
2017-04-27
Moved from OpenNIC to OpenDNS (they dropped port 443 support)
Upstream updates
2017-04-08
Re-added netfilter modules present on vanilla LEDE
Removed AOE and ATA related kernel modules
Upstream updates
2017-04-04
ca-bundle added
2017-04-02
Upstream updates
Added patches and files source code
Added nightly builds based on latest release configs to keep up with upstream developments
2017-04-01
pppossh added
openssh-sftp-server added
Many netfilter modules removed to increase throughput
@DjiPi Basically opkg wasn't working properly because of checksum mismatches (going to be the same with my own build though, but it includes what I would have installed already) and some packages were missing completely because the buildbot server ran out of disk space, ie. not all packages for RC1 were built.
Then, even after manual installation of SQM packages with --force-checksum, the SQM settings weren't taking effect, eg. I could set 1Kbps ingress and egress limits and clients would still get full speeds in Mbps ranges. After this I decided to just do my own build instead of spending more time on fixing my RC1 install.
I might add that according to @jow at least the issue with missing packages in RC1 should be fixed now though and that RC2 should be ready soon.
Anyway, in case your requirements are similar as mine and you'd want to give my build a spin: I'm compiling a new one right now with only upstream updates.
I'll consider it. Downsides would be a small performance penalty and a more bloated firmware. Anyone else needing this?
bcp38 can be easily added by just dropping internal and reserved ranges on WAN, which can be done with what this firmware already includes (it does support ipset too, just like most netfilter modules). I'm going to add it regardless due to small overhead.
That's actually included already, just not listed in features, I'll add it there too.
Debug symbols are already being removed, however I indeed didn't set -O3 for GCC - good catch![quote="deuteragenie, post:5, topic:1382"]
would be good to report on what can/is achieved by with this build.
[/quote]
You're welcome to run some benchmarks. I didn't have time for this yet, but it's definitely on my todo.
I actually don't need DNSCryprt/DNSSec, or much of the other stuff. But I believe that some may, and if your purpose is to have an as-OOB-as-possible build, then probably they should be in.
BCP38: I include that module in my home builds (just called bcp38 if I remember correctly), and there must be Luci support for it.
I forgot to mention: I run my builds with gcc 6.3 without any issue - you may wish to try it for yours.
And, as your build is targeted at one specific processor, you may wish to look at the -march options or similar, in the hope of getting that 0.01% performance increase :).
Maybe other users would be interested in adblocks too ?
On mine, the 5 Ghz LEDs are actually on when the 2.4 Ghz channels are on. I guess a little bug (or maybe due that I disabled the 5 Ghz stuff altogether)
My goal is to build something in between most other "optimized" builds which are too bloated for my taste and the vanilla builds which lack too many kernel modules, packages and compiler optimizations.
Awesome, I'll give this a spin within the next couple of days. Thank you for sharing your experience.
Using that already.
I'll likely not add this either. There are just too many other options for ad blocking (uBlock, hosts file on connected devices, DNS that blocks ads, etc.) which is why I don't want to impose the adblock package on everyone using this build.
It's posible to "upgrade" to your version of firmware without loosing our working config (wifi,network,vlans...) coming from the "official" LEDE Project?
It's just installing the sysupgrade (like LEDE), or maybe is better coming from a "fresh" install?
Yes, it's possible, I did the same. Just select preserve settings when flashing the sysupgrade image.
Also I'm trying the third build today shortly on my router, which uses gcc 6.3 and binutils 2.27, so if don't want to flash the firmware right away, you'll be able to download a slightly faster one shortly.
Don't forget to add gcc 6.3 to your release notes....
And, one other important issue:
Could you load your scripts/configs to github ? Unless I "see the source", I, for one, would not install any firmware on my routers (I know, some blobs are still around).
Great idea, thanks for pointing me into that direction! I'm new to LEDE/OpenWRT and building router firmware in general, so your advice is really appreciated. It didn't even cross my mind that the CPU would support anything but 24kc coming from i836/amd64 only environments.
Also, you can try other GCC optimizations that go beyond -O3, but I wouldn't know how to make sure that everything will still work (BTW, on my routers - Archer C7v2 and 842N, I compiled everything with GCC 6.3, but did not try -O3 or -march - mtune).
Thanks. I'm familar with GCC options coming from UNIX systems, just these CPU architectures are new to me. Also running into several issues with GCC 6.3 which I'm still busy solving (same configs worked fine with GCC 5).
Thanks for that info @deuteragenie. However, I did the same and run into several errors. Solved two already, but they keep coming. Do you also use binutils 2.27? Maybe that's causing the issues or I'm just using different packages than you are.
I used the "standard" git pull of head and did not change anything else than a few items (including switching to gcc 6.3) through menuconfig. This was about one month ago.
Um, it worked now with -j1 when it failed with -j8. Takes forever with one core when HT is enabled, but hey, at least it worked now. Will push the update shortly.
Interesting. I compiled mine without problems at -j9 on 8 cores.
You can always try to unroll some or all loops and see what gives... as you need to deserve the "Optimized" in your announcement, right ?: gcc -O3 -funroll-loops or -funroll-all-loops
Or compile and run with the profiler enabled, and use -fbranch-probabilities for the second compile... possibilities are endless to win 0.3% performance!
Also, I believe that you should use the following: -march=mips32r2 -mtune=74kc (and not -march=74kc, is this even allowed/documented?)
Well, unrolling loops causes a further increase in file size. Particularly the positive effects of unrolling all loops is questionable - the gcc documentation even mentions that this would slow down programs. I'll stick to safe and stable options to avoid turning this into an experimental build.
