IRC channel logs
2013-10-22.log
back to list of logs
<mark_weaver>well, I've determined that there's a bug either in gobject (part of glib) or libffi relating to the MIPS N32 ABI. *mark_weaver thinks he sees the bug in libffi <mark_weaver>okay, I have a patch that might fix the libffi problem, and thus glib. now to do the rebuild to test it :) <mark_weaver>damn, it looks like gcc has bugs related to the MIPS N32 ABI. <mark_weaver>this will require some discussion on the gcc mailing list. <davexunit>mark_weaver: you have gone down quite the rabbit hole it seems. <mark_weaver>oops, I was mistaken about GCC. afaict, GCC is fine; the bug seems to be limited to libffi. <mark_weaver>(my mistake was making incorrect assumptions about how some MIPS instructions behaved) *mark_weaver needs to learn more about MIPS <viric>mark_weaver: look for mips-iv.pdf, MIPS-N32-ABI-Handbook.pdf and Loongson2FUserGuide.pdf :) <viric>not that I needed them much though. <mark_weaver>viric: yeah, I have all of those. I didn't guess the behavior of 'srl' and 'sll'. They shift the low 32 bits and then sign extend the result to 64 bits. <mark_weaver>viric: What version of linux-libre do you use on your Yeeloong? <viric>And I run there the normal linux <viric>last time I tried, it was 3.7 or so <viric>I don't turn it on since months <mark_weaver>linux-libre-3.10.15 reliably freezes if I use the built-in wireless.. and it also sometimes locks up other times. <mark_weaver>but the m4 test suite fails on 2.6.39.1, because readlink("") returns a different error code than it's looking for. <mark_weaver>also, I assume there are lots of other problems with 2.6.39.1, so it would be nice to run a newer kernel. <mark_weaver>as things stand, I have to switch back and forth between kernels when I want to build m4 :-( <viric>But I could run kgdb with the serial port fine, I think <viric>mark_weaver: there are many kernels between 2.6.39 and 3.10 :) <mark_weaver>viric: indeed, and I intend to try some. I was asking to save me some time doing a binary search in that range :) <mark_weaver>although really, we need to fix the problems with the latest linux-libre.