/srv/irclogs.ubuntu.com/2013/03/08/#ubuntu-kernel.txt

=== rsalveti_ is now known as rsalveti
=== rsalveti_ is now known as rsalveti
ppisatimoin07:54
=== smb` is now known as smb
apwmoin09:09
smbmorning09:09
ckingmornin09:16
ckingnggg, mumble fail09:16
diwicI'm trying to run fdr editconfigs on gomeisa, but it fails due to a missing ncurses-devel package; what should I do instead?09:23
smbdiwic, You run that in a chroot?09:24
diwicsmb, no, hmm, maybe I should?09:24
smbdiwic, more likely you will have ncurses in there09:24
apwdiwic, and if we don't have it let me know09:33
diwicapw, it worked in the precise-amd64 chroot, thansk09:33
apwdiwic, great09:35
ogra_ppisati, did you see my PM last night ?09:53
xnoxHello all =) is it just me or there is no: overlayfs, aufs, unionfs on the linux-nexus7 kernels ?! Can we enable them please =)))))10:35
ogra_heh, fun10:36
* apw has a bad kernel panopoly day12:14
tseliotapw: can you help with a little read-writer locks doubt, please?12:50
tseliot*reader-writer12:50
tseliotapw: I'm trying to replace rcu_read_lock/unlock with read_lock/unlock since the rcu calls are GPL_ONLY12:55
tseliotapw: so, I was wondering if something like this could work (I don't have the hardware to test it): http://paste.ubuntu.com/5595857/12:56
apwtseliot, i am unsure if you need the rmb as as the read_lock is a full barrier i believe13:13
apwtseliot, but it seems workable in concept.  performance may suck of course13:13
apwtseliot, of course ... i should say ... you need to make the current write side locking also take this new lock in write mode13:16
apwelse you have no locking at all :)13:16
apwthis wholel _GPL thing is a daft fabrication13:17
bjfapw, overlayfs patch for quantal?13:53
apwbjf, yes sitting here waiting to send ... doh13:54
apwi am all out of sync following the 10 crashes i have had this morning13:55
* ppisati is updating his laptop to R: Goodbye Cruel World!14:02
apwppisati, did not work out well for me14:02
* apw runs a Q kernel on it14:02
ppisatiDOH14:02
apwits that bad14:02
apw(for me)14:02
ppisatijust started the do-release-upgrade thing14:02
ppisatiok, i'll use a Q kernel on it14:02
apwgood luck with that, i have a gm45 and it is fucked14:02
tseliotapw: I don't think there's any write_lock in the code. I had to introduce rcu_read_lock() (following the example of some open drivers) because upstream broke the API. I'm wondering how it may impact performance in the broadcom wireless driver...14:10
tseliots/impact/affect/14:10
apwthen who are you locking against if there is no other rcu in the thing14:11
apwis it in the driver model somewhere, if so it might work sort of14:11
apwbut ... hmmm is all i can say14:11
tseliotapw: so, this is my original patch for linux 3.8: http://paste.ubuntu.com/5596010/14:12
apwwell then you must be interlocking with something else in the core which is handling the write locking14:13
apwessentially you are in a hole with no spade at this point as you should be using the same locking as they are, else you are wasting your time14:14
apwbut you can't because it is not exported.14:14
apwyou locking there i don't think does anything 14:14
apwif it is not interlocked with the other side14:14
* henrix -> late lunch14:15
tseliotapw: hmm... I see what you mean14:16
apwtseliot, as long as you have the rcu_dereference, you can probabally get away with it int he common case, because read_lock_rcu is a bit of a noop in a lot of cases14:16
apw(when preempt is off)14:16
apwbut if they are exporting the 80211 interfaces for use by non-gpl code, but those rely on GPL only locking ... it isn't really exporte14:18
tseliotapw: are you suggesting that I get rid of the lock/unlock and that I simply use my local equivalent for rcu_dereference?14:18
apwi am suggesting that to do that would be utterly wrong for so many reasons, but it might just about work14:19
apwtseliot, what is not exported that you need14:20
tseliotapw: I guess rcu_dereference is not GPL only, maybe only rcu_read_unlock and rcu_read_lock are14:23
apwtseliot, they only are in some versions of rcu i think14:23
apwonly GPL only i think if we are using CONFIG_PREEMPT_RCU14:23
apwdebian.master/config/i386/config.common.i386:# CONFIG_PREEMPT_RCU is not set14:24
apwdebian.master/config/amd64/config.common.amd64:# CONFIG_PREEMPT_RCU is not set14:24
apwwhich it isn't on our default kernels14:24
tseliotapw: right now my original patch doesn't fail in Ubuntu but some developers on Arch Linux and another distro I can't recall are complaining about the GPL only issue14:24
tseliotthat would explain it14:24
apwahh then that makes sense14:24
apwi think overall there is an issue with a core primative like rcu being gpl only14:25
apwit really doesn't make sense in a fair world14:25
* tseliot nods14:25
tseliotapw: but how can it be that they're exported as GPL only if that config is enabled?14:26
apwbecasue it is only __rcu_read_lock which is GPL only exported, and there are two version14:27
apwone version for PREMPT on and one for PREMEPT off14:27
apwthe one for PREEMPT off is an inline with nothing much in it and is not exported at all14:27
apwand therefore has no way to limit it14:27
tseliotaah, I see now14:27
apwthe stupidity is you can (as far as i can see) take the code out of the middle of the function and inline it14:28
apwinto the code you have which is GPL (gpl with gpl is ok) and then use it from your proprietry code14:28
apwas you are effectivly doing with rcu_dereference14:28
tseliotright14:29
apwit is also not clear it is not possible to randomly patch the kernel14:30
apwto add a local_rcu_lock() function which is exported non-gpl which literally does rcu_lock_ ... its a farse14:31
tseliotheh14:32
tseliotdo we have a kernel flavour with CONFIG_PREEMPT_RCU enabled?14:32
tseliotor shall I just forget about it?14:32
apwnot that this driver would talk to as far as i know14:34
apwi'll ask about it as well14:35
ppisatiback in 2014:35
tseliotapw: otherwise I'll just copy & paste the code if CONFIG_PREEMPT_RCU is defined14:36
apwits hard to see how that is illegal dispite it being utterly not what they want :)14:37
tseliot:D14:38
tseliotit makes me wanna do it just because of that ;)14:40
apwtseliot, something to find and extract the code you need from the kernel source :)14:47
tseliot:)14:55
=== kentb-out is now known as kentb
ogra_xnox, ppisati might be intrested in getting aufs/overlayfs into the nx7 kernel  :)15:33
xnoxogra_: i just want sensible sbuild ;-)15:33
ogra_well, the android source we have doesnt include either, needs porting work first 15:34
ogra_and if we possibly play with overlays for the upgrade procedure, it will be needed i guess15:35
xnoxogra_: how hard can it be? I thought android kernel doesn't care much about fs stack (as in doing / changing it extensively)15:35
ogra_not sure what stgraber's exact plans are but it would definitely be helpful to have it for different test scenarios15:35
stgraberso far I'm avoiding the use of an actual overlay filesystem because we can't be sure we'll have one in the kernel, but I certainly wouldn't mind having one ;)15:37
ppisatixnox: open a lp bug please15:53
ppisatixnox: i'm rushing out now15:54
xnoxpsivaa: ok.15:54
psivaappisati: ^ :)15:54
xnoxpsivaa: sorry.15:56
xnoxppisati: ack. it will be bug 107631715:56
ubot2`Launchpad bug 1076317 in ubuntu-nexus7 "please add aufs/overlayfs/unionfs support to android based kernels" [High,Confirmed] https://launchpad.net/bugs/107631715:56
psivaaxnox: np15:56
ppisatixnox: nice15:56
* ppisati rushes out for some early workout15:57
ckingjsalisbury, i've forwarded you some background info and a possible workaround 16:49
jsalisburycking, awesome, thanks, Colin!16:49
ckingif we can get the ACPI tables though, I can check out my hypothesis, but there is no harm in spinning them a test kernel 16:50
jsalisburycking, cool, thanks16:50
ogasawarabjf: next week is SRU verification week right?17:24
bjfogasawara, it's regression testing week17:24
ogasawarabjf: ah, thanks.  can you point me to the calendar I should follow?17:25
bjfhttps://wiki.ubuntu.com/RaringRingtail/ReleaseInterlock17:25
ogasawarabjf: thanks17:25
=== shadeslayer_ is now known as shadeslayer
* cking ~~-> catches the weekend18:19
* henrix follows18:22
=== kamalmostafa is now known as kamal

Generated by irclog2html.py 2.7 by Marius Gedminas - find it at mg.pov.lt!