Would you folks mind making two kernel configuration changes?
First, to enable /proc/config.gz support (CONFIG_IKCONFIG=y, CONFIG_IKCONFIG_PROC=y), and second, to compile TCPMSS support into the kernel (as opposed to a module, CONFIG_NETFILTER_XT_TARGET_TCPMSS=y). If you do, you can also remove the xt_TCPMSS.ko module from the initramfs and the insmod call for it in init.rc.
Attached is a patch against defconfig that does this. I've been running with TCPMSS support built-in (which it was in DI18) for a week and haven't had any issues. I'm not sure why Samsung even changed it.
I ask because, WiFi Tether v3.0-pre5+ includes support for MSS clamping
, but the option to enable it only appears
if it sees CONFIG_NETFILTER_XT_TARGET_TCPMSS=y in /proc/config.gz. If it does, then an "MSS clamping" option appears under "LAN" in the WiFi Tether Settings menu.
I've been encouraging MSS clamping support for a little while now as a precautionary measure. However, in the past week folks have been reporting
trouble with WiMAX tethering in certain (typically recently-deployed) markets, where it appears that fragmented packets are dropped by the network.
The common symptom is that tethering works fine on 3G, but on WiMAX most pages don't work. Google sites typically do, but this is expected as Google sites are often not very data heavy and tend not to result in packet fragmentation. It's confirmed that there's an MTU mismatch, and in one case it's confirmed
that MSS clamping resolves the issue. Hence the issue is a bit more pressing than it was previously.
As a bonus, this means that SyndicateROM Frozen can boast having better WiMAX tethering support than the official Hotspot client