| === lag` is now known as lag | ||
| === zz_anmar is now known as anmar | ||
| === anmar is now known as zz_anmar | ||
| === zz_anmar is now known as anmar | ||
| === anmar is now known as zz_anmar | ||
| === zz_anmar is now known as anmar | ||
| === doko_ is now known as doko | ||
| wookey | infinity: is there a patch in ubuntu glibc to avoid "FATAL:kernel too old" messages using qemu-aarch64-static on wheezy? | 14:21 |
|---|---|---|
| wookey | (a version build on saucy works. A version built on unstable doesn't) | 14:22 |
| infinity | wookey: No glibc patch, but there's a qemu patch. | 14:25 |
| ogra_ | well, you still get messages for the missing syscall | 14:25 |
| ogra_ | 247 or so ... | 14:25 |
| infinity | That's not his issue. | 14:25 |
| ogra_ | (which is "old kernel" in syscalls.h) | 14:25 |
| infinity | His issue is qemu claiming to be a version it's not. | 14:25 |
| infinity | We carry a patch for that. | 14:26 |
| ogra_ | right the fatal bit is gone | 14:26 |
| ogra_ | but we still get the noise fallout in logs etc | 14:26 |
| * ogra_ gets questions about that since he revived rootstock :) | 14:27 | |
| ogra_ | (read: since last week) | 14:27 |
| infinity | wookey: debian/patches/ubuntu/arm64/force-aarch64-uname-to-3.7.0-to-appease-glibc | 14:29 |
| infinity | wookey: http://paste.ubuntu.com/6954659/ | 14:30 |
| wookey | aha. cheers | 14:31 |
| wookey | Is it right to nobble qemu, rather than tell glibc not to be so picky? | 14:31 |
| wookey | I guess the point is that officially the 3.2 kernel didn;t know about arm64 so on a real box it is normally wrong? | 14:32 |
| wookey | (/me forgets when arm64 did hit kernel-land) | 14:32 |
| suihkulokki | there is a qemu configure and command line and enviroment options to set up what uname is reported there... | 14:33 |
| wookey | yes, but I'm not sure how you use that with binfmt entries | 14:34 |
| suihkulokki | export QEMU_UNAME .. | 14:34 |
| wookey | you mean the line for the binary that gets run is actually interpreted by a shell? | 14:35 |
| infinity | wookey: I'm not going to tell glibc to be less picky, no. | 14:35 |
| suihkulokki | wookey: why do you think the binary running with qemu linux-user wouldn't see the env ? | 14:36 |
| infinity | wookey: Telling glibc to lower its baseline causes a lot more fallout than you'd think. Bunch of compat code gets pulled in, slower paths taken, you really don't want that on a new port that literally CAN'T run on older kernels anyway. | 14:36 |
| wookey | I'm sure it would see the env. The point is that normally it's run by binfmt-support entries. Those need to set all the necessary environement | 14:36 |
| wookey | infinity: OK. fair enough. | 14:37 |
| infinity | suihkulokki: The configure option is --enable-uname-release ... Which we (Ubuntu and Debian) set to 2.6.32 on all arches. Which then blows up on arm64, hence my quick patch to just ignore it there. | 14:37 |
| infinity | I meant to upstream a patch that actually did a saner thing and obeyed your configure option if it was >> the version glibc supports on that arch, but that sounds like real work. This works for now. :P | 14:39 |
| infinity | Well, it also means maintaining a table, which is a pain. | 14:39 |
| wookey | OK. setting QEMU_UNAME=3.7.0 does indeed make it work. I'll do some more tests with that. Then work out a patch for the debian package | 14:39 |
| infinity | Oh, and the table would be "when glibc got support for the arch" not "what your distro set baseline to", so still a lie. | 14:39 |
| suihkulokki | certainly maintaining a distro-spefic hack is much better :P | 14:39 |
| wookey | suihkulokki: I've done a debian qemu package from the current 1.7.50 git | 14:40 |
| infinity | suihkulokki: In this case, the distro hack is much simpler, to be fair. | 14:40 |
| wookey | am I right in thinking that the stuff taken out for +dfsg is just in pc-bios dir, or is there more? | 14:40 |
| infinity | wookey: Just take my patch I pasted. It's the path of least resistance here, unless you really want to introduce that table that's a lie anyway. | 14:40 |
| wookey | OK. unless the qemu people complain that's fine by me. | 14:41 |
| wookey | (riku and vagrant in this case i think) | 14:41 |
| wookey | infinity: are you up early or late? | 14:42 |
| infinity | wookey: Late. Very late. | 14:42 |
| wookey | :-) | 14:42 |
| wookey | handy for me | 14:42 |
| infinity | wookey: Pretty sure the qemu people are more likely to complain about you asking for a git snapshot. :P | 14:43 |
| infinity | wookey: (hallyn's already backported the arm64 stuff to 1.7.0 in Ubuntu and, AFAIK, it works) | 14:43 |
| wookey | the suse arm64 stuff or the linaro aarch64 (spit) stuff? | 14:43 |
| infinity | wookey: The upstream stuff. | 14:44 |
| wookey | That was announced yesterday? | 14:44 |
| wookey | I could have saved myself some effort if so | 14:45 |
| infinity | wookey: It got pulled in as soon as it landed in git, not when it was announced. There may be a few fixes to pull in on top of our current stuff, but I doubt there's much. | 14:48 |
| infinity | (Literally as soon as... dannf had a script running that kept doing a git pull, compile test, and runtime tests to see if it worked yet :P) | 14:48 |
| infinity | He was a bit impatient. | 14:48 |
| wookey | OK. any chance of some upstreaming to debian? | 14:50 |
| dannf | wookey: the changes are in a branch of the debian package on alioth (http://anonscm.debian.org/gitweb/?p=pkg-qemu/qemu.git;a=shortlog;h=refs/heads/ubuntu-trusty) | 15:12 |
| dannf | wookey: assuming the qem team is cool with taking in such changes, should be a relatively simple cherry pick | 15:12 |
| wookey | dannf: OK. cheers. Adding adam's patch seems to have broken my build for some mysterious reason:-( | 15:17 |
| wookey | how uptodate are you with the 1.7 master branch? | 15:18 |
| dannf | infinity: : oh, is this the control file arch cleanup you were talking about? | 15:18 |
| === anmar is now known as zz_anmar | ||
| === killer_prince is now known as lazy_prince | ||
| dannf | wookey: i'm not sure, i just handed hallyn debdiffs. but, i suspect the aarch64 stuff will apply cleanly patchwise - assuming debian takes the kvm stuff as well (basically ubuntu's whole aarch64 patch set) | 15:20 |
| dannf | wookey: i can prep a github branch if you like | 15:21 |
| infinity | dannf: I was just talking about a 2-line patch to fix the minimum uname version. | 15:21 |
| infinity | But yes, we should look at getting all the aarch64 stuff pulled into Debian sanely. | 15:22 |
| dannf | oh, right | 15:22 |
| infinity | And update our patchset if there's new stuff we care about. | 15:22 |
| dannf | i see nothing new target-arm wise | 15:22 |
| wookey | yesterday seemed to be the point that ajb thought it was 'ready' | 15:22 |
| wookey | http://lists.linaro.org/pipermail/linaro-dev/2014-February/016980.html | 15:23 |
| * dannf wonders when qemu will cut a new upstream - upstream qemu git is difficult to track for cherry picks - or i need some better git fu | 15:24 | |
| wookey | I didn't want to do backporting to 1.7.0 from 1.7.50 as I know nothing. | 15:24 |
| wookey | arm64 was targetted for 1.8 in 'spring' | 15:24 |
| wookey | so if we test this and find it basiically works then I hope a new release is imminent | 15:24 |
| wookey | which is why I was interested to package and test it | 15:25 |
| * dannf suspects ubuntu virtualized ppas will be a good testbed | 15:25 | |
| infinity | dannf: Yeah, though it would be worth checking if the branches referenced in that email contain more fixes on top of what you pulled in for hallyn. | 15:25 |
| dannf | infinity: well, it looks like he's saying master shoudl be enough (and i'm in sync there) | 15:26 |
| wookey | bummer that they did s/arm64/aarch64/ but you can;t have everything in life | 15:27 |
| infinity | * Additional NEON/SIMD instructions | 15:27 |
| infinity | * sendmsg syscall | 15:27 |
| dannf | infinity: he has some other patches - but i've no way of saying how likely they'd be needed | 15:27 |
| infinity | Those are the bits in his tree but not master. | 15:27 |
| infinity | Emulating more syscalls can't ever be bad. | 15:27 |
| infinity | And more insns. | 15:27 |
| wookey | right. I hit quite a lot of missing insns with the older suse code | 15:27 |
| dannf | yeah, i was getting a lot w/ upstream until the merge last week w/ the neon/simd instructions fixed | 15:28 |
| dannf | but my test case has just been "debootstrap" | 15:29 |
| infinity | Try to build qt4-x11 and get back to me. :) | 15:30 |
| === zz_anmar is now known as anmar | ||
| wookey | gah. after adding the uname patch my build just stops with make: *** [build-stamp] Error 2 but no actual error. | 15:30 |
| wookey | and I haven't run out of disk space... | 15:31 |
| infinity | That's... Odd. | 15:31 |
| wookey | grumble. | 15:31 |
| infinity | And seemingly impossible. | 15:31 |
| infinity | Did you debdiff old and new sources to make sure that's really all you changed? :P | 15:31 |
| wookey | sigh. | 15:31 |
| wookey | not yet. I shall go get coffee and gird loins. | 15:31 |
| wookey | computers are fucking painful. You just want totest something and on no, random wierdness sens you off for extra faffage | 15:32 |
| infinity | wookey: Time for a career shift? | 15:35 |
| ogra_ | go into management ! | 15:40 |
| wookey | ha ha people are worse than computers :-) | 15:45 |
| ogra_ | :) | 15:46 |
| wookey | I have always fancied doing some rope access work by way of variety. Pays well too. | 15:49 |
| === anmar is now known as zz_anmar | ||
| === zz_anmar is now known as anmar | ||
| === anmar is now known as zz_anmar | ||
| wookey | OK. a non-parallel build gets me an actual whinge | 16:15 |
| === alf__ is now known as alf_ | ||
| === zz_anmar is now known as anmar | ||
| === anmar is now known as zz_anmar | ||
Generated by irclog2html.py 2.7 by Marius Gedminas - find it at mg.pov.lt!