=== hggdh_ is now known as hggdh | ||
=== karstensrage_ is now known as karstensrage | ||
=== maclin1 is now known as maclin | ||
=== maclin1 is now known as maclin | ||
=== maclin1 is now known as maclin | ||
=== lan3y is now known as Laney | ||
shadeslayer | hey, can you tell me if Launchpad's arm64 builders are somehow specially configured for large memory allocations when building armhf packages? | 12:19 |
---|---|---|
cjwatson | shadeslayer: probably depends on exactly what you mean, but they have a modified kernel built with CONFIG_ARM64_VA_BITS=48: https://git.launchpad.net/~wgrant/ubuntu/+source/linux/log/?h=arm64-va-39 | 12:36 |
shadeslayer | cjwatson: well my cross build on a arm64 machine still fails when ld fails to allocate enough memory | 12:37 |
shadeslayer | 00:29:28.719 /usr/bin/ld.gold: fatal error: libQt5WebEngineCore.so.5.7.1: mmap: failed to allocate 2226595652 bytes for output file: Cannot allocate memory | 12:37 |
shadeslayer | which is a bit odd since the machine has 128 GB's of memory | 12:37 |
cjwatson | don't know more, sorry | 12:37 |
shadeslayer | ok :) | 12:37 |
shadeslayer | perhaps someone else in the channel will :) | 12:37 |
shadeslayer | someone suggested I bump nr_hugepages to 20 | 12:38 |
shadeslayer | which I've done now to see if it helps | 12:38 |
cjwatson | we aren't doing anything like that AFAIK. The only other specialised arm64ish thing I'm aware of is that we boot the VMs with compat_uts_machine=armv7l (modulo typos, I don't have the current config file to hand). | 12:39 |
wgrant | shadeslayer: As Colin says, are kernels are stock xenial plus the patch to unbreak Qt (CONFIG_ARM64_VA_BITS=39, which also reduces the number of page table levels) | 12:42 |
wgrant | We've done nothing special to increase address space sizes -- in fact we've shrunk the 64-bit ones. | 12:42 |
shadeslayer | aha | 12:42 |
shadeslayer | I have CONFIG_ARM64_VA_BITS=48 | 12:43 |
cjwatson | Yeah, 39, sorry, I misread the git log. | 12:43 |
wgrant | That's the default, yeah. It breaks old Qts because they expect to tag their pointers up at like bit 42. | 12:43 |
shadeslayer | I see | 12:43 |
wgrant | It usually manifests as Qt apps segfaulting, not what you're seeing. | 12:44 |
shadeslayer | old qt? | 12:44 |
cjwatson | Apparently fixes for that are trickling through to things we might get in updates, so hopefully we can ditch that patch eventually. | 12:44 |
wgrant | https://bugreports.qt.io/browse/QTBUG-54822 | 12:44 |
shadeslayer | tyvm :D | 12:44 |
wgrant | But should be totally unrelated. | 12:44 |
shadeslayer | right, that's at runtime, but good to know | 12:45 |
shadeslayer | I'm more curious about the link failiure | 12:45 |
cjwatson | I mean in a 32-bit address space it doesn't necessarily matter how much physical memory you have. | 12:46 |
cjwatson | but I'm not sure exactly what you mean by "cross build". | 12:46 |
cjwatson | LP doesn't do cross builds. | 12:46 |
shadeslayer | oh this isn't on LP | 12:47 |
shadeslayer | cjwatson: but then how do you build qtwebengine? | 12:47 |
shadeslayer | because afaict https://launchpadlibrarian.net/308759960/buildlog_ubuntu-zesty-armhf.qtwebengine-opensource-src_5.7.1+dfsg-6build1_BUILDING.txt.gz | 12:47 |
shadeslayer | https://launchpadlibrarian.net/308759960/buildlog_ubuntu-zesty-armhf.qtwebengine-opensource-src_5.7.1+dfsg-6build1_BUILDING.txt.gz | 12:47 |
shadeslayer | er | 12:47 |
shadeslayer | Kernel version: Linux bos01-arm64-005 4.4.0-64-generic #85+buildd1-Ubuntu SMP Wed Feb 22 12:26:21 UTC 2017 aarch64 | 12:47 |
cjwatson | we build in an armhf chroot rather than cross-building. | 12:48 |
cjwatson | cross-building usually refers to building with tools from one architecture but generating code to run on another. | 12:49 |
cjwatson | in this case we're using armhf throughout, but in a chroot running atop an arm64 base system (and kernel). | 12:49 |
shadeslayer | ah ok, that's what I'm doing too, armhf docker container ontop of a arm64 one | 12:50 |
shadeslayer | *arm64 host | 12:50 |
shadeslayer | it's a bit odd that it compiles just fine on Launchpad but not on my machine | 13:53 |
=== yofel_ is now known as yofel | ||
shadeslayer | cjwatson: wgrant JFYI apparently Debian ships the patch to Qt for this | 14:34 |
shadeslayer | so maybe you don't need that anymore? | 14:35 |
cjwatson | it's making progress, but it needs to be in all series etc. | 14:35 |
shadeslayer | *that config option | 14:35 |
cjwatson | I think Mirv is looking after that on our side | 14:35 |
shadeslayer | ah ok :) | 14:36 |
Mc | hi | 16:54 |
Mc | just to signal that bots have started posting comments to bug reports that they did not open themselves (i don't recall that happening before in bugs.lp) | 16:55 |
Mc | https://launchpad.net/~pop62 | 16:55 |
=== jik is now known as Guest949 | ||
Guest949 | Launchpad down? | 17:00 |
cjwatson | Guest949: it looks OK from here and our graphs seem fine | 17:01 |
Guest949 | Just came back. | 17:01 |
Guest949 | I got several "Something went wrong, check twitter or freenode IRC" errors in a row before it came back. | 17:02 |
Guest949 | i.e., the problem clearly wasn't on my end with my network, given that I got the launchpad error page in my browser. | 17:02 |
cjwatson | Mc: doing some cleanup, thanks | 17:05 |
Mc | thanks :) | 17:08 |
Mc | I'm losing the battle on lp answers spam | 18:35 |
Mc | I can find no regexes for my procmailrc to filter out that new spam that asks questions similar to real ones | 18:36 |
Mc | It's just too similar to real questions | 18:36 |
Generated by irclog2html.py 2.7 by Marius Gedminas - find it at mg.pov.lt!