From: Alexandre P. N. <al...@om...> - 2006-11-14 23:36:00
|
Hi, I tried to analyze what's different between astlinux's buildroot and gumstix's one in an attempt to understand the problem. After eliminating binutils 2.16.1 x 2.17 builds, and verifying uclibc is the same version on both (with the same patches), the main remaining factor was the kernel. I built a 2.6.17 on gumstix's buildroot (had to borrow the kernel-patches from astlinux, they're from previous gumstix versions) and voila, no more asterisk nor dynamic iptables problems, and my little test app also stopped crashing. So now it's finding what the hell could have changed between 2.6.17 and 2.6.18 that could justify this behavior. Looking at the 2.6.17 -> 2.6.18 mainstream patch, I see some xscale cache changes supposed to fix things and not introduce some bugs. They seem correct after taking a glimpse at them, but perhaps it requires digging deeper. I'm tempted to selectively and individually reverse some changes, and see if I can spot at the one stopping the show, then propose to Craig to host a patch against 2.6.18 while we wait for a 2.6.19 solution or something. By the way, I've built 2.6.17 with my 2.6.18's config, so that I could be (pretty) sure it wasn't some configuration change that caused the misbehavior. I won't do that for a bit since I'm getting busy with other stuff, but I figured it was a good thing to point up in case someone has both the time and the will... - Alexandre |