=== FJKong_afk is now known as FJKong === gerald is now known as Guest66810 === ara is now known as Guest14068 === ayan is now known as Guest80066 === ayan_ is now known as ayan [10:23] LocutusOfBorg1, hiya, i see the new 5.0 virtual box doesn't yet have 4.2 support, is that coming soon or should i patch ubuntu up for it [10:24] LocutusOfBorg1, i have a patch here i am using for the kernel import which fixes the symlink api in it, if thats of use [10:28] hi apw answering there [10:28] sorry for the late answer [10:28] you can mail at locutusofborg@d.o [10:28] (heh no late answer) [10:29] s/d/debian [10:29] s/o/org [10:29] LocutusOfBorg1, will do indeed [10:29] but wait please [10:29] * apw waits [10:29] I'm opening an RC right now on Debian [10:29] I wont virtualbox kicked out of Debian and Ubuntu [10:29] oh ... why? [10:35] I'll link it there ;) [10:35] as soon as I finish the mail [10:35] :) [10:47] bug sent [10:47] let BTS do the job and I'll post the link [10:47] anyway, which files are changed? [10:48] because as usual upstream should have already patched the sources [10:48] and I usually update virtualbox with the latest release after the freeze, to be syncd with the kernel [10:50] LocutusOfBorg1, vbox/vboxsf/lnkops.c they ahve changed the follow_link()/put_link() api, rather radically [10:50] though the fixes are not big if that makes sense [10:51] that is what i am using against what is in the 5.0.0 package for the kernel modules we have sucked in http://paste.ubuntu.com/11992242/ [10:53] https://github.com/mdaniel/virtualbox-org-svn-vbox-trunk/commit/78627d149e35f21b689dec7977c9d6c386ad71e6 [10:53] apw, ^^^ ? [10:54] LocutusOfBorg1, yep that looks to be equivalent indeed, and better [10:57] LocutusOfBorg1, so this is about opaque oracle, sigh === henrix_ is now known as henrix [11:50] bugs.debian.org/cgi-bin/bugreport.cgi?bug=794466 [11:50] apw, ^^^ yes (sorry I was lunching) [11:59] heh np === psivaa is now known as psivaa-lunch === psivaa-lunch is now known as psivaa [15:52] tseliot, guess what ... dkms packages for 4.2 are all unhappy (again) [15:53] apw: yay for breaking the API. What kernel version are you using? [15:54] 4.2-rc4 i think, though we have an upload about to land [15:54] also why does dkms report building for both amd64 and x86_64 in the same chroot??? [15:54] dkms status [15:54] fglrx-core, 15.200.1, 4.1.0-3-generic, amd64: installed [15:54] fglrx-core, 15.200.1, 4.1.0-3-generic, x86_64: installed [15:55] gurgle :) [15:55] oh, and BTW, I won't be able to fix nvidia for Linux 4.2, as kernel devs keep making functions GPL only... [15:56] tseliot, fun [15:56] oh, you have no idea ;) [15:56] they need to start loadin the binary driver itself as a firmware blob [15:56] then their driver can be GPL [15:58] yes, that's probably the only way to do it, other than open sourcing everything (in our dreams) [15:59] my very wet dreams [16:03] hehe [16:04] the _GPL is an utter farce whoever added that deserves a beatng [16:04] +1000 [16:05] tseliot, actually it is not at all clear to me that they couldn't write a wrapper which was propriatry which literally only exported the binary interfaces in the binary blobs as wrap_name 1 for 1 [16:06] and then the existing GPL bits be a second module which is self contained and therefore GPL [16:06] which uses both sets ... which demonstrates how the _GPL is a farse [16:07] intent isn't ambiguous [16:07] someone could do that, but it would be an overt action for 'reasons' [16:08] ohsix, the intent is that only GPL things can link to that interface [16:08] the intent isn't to make ordinary users lives vile [16:10] :) [16:10] and if one could not allow GPL code to talk to proprietaory blobs, then we couldn't talk to the bios [16:10] nor to disk drives with firmware, or nics or ... [16:11] that's not even remotely the same, heh; the question is over whether something is a derived work or not and symbol licenses make it unambiguous in certain places [16:11] if there was an eula or literally anything that said you couldn't talk to the firmware on something then you'd be bound by that too [16:12] ohsix, except that i am not sure you'd find that the licence exactly says any of the things either of us think is does, because lawyers [16:12] none of the clarity we think is there, really is [16:13] the question is of whether it is a derived work [16:13] and the hazard with figuring that out in court [16:13] indeed, and i'd contend "we don't know" is the only correct answer [16:13] that's pretty clear [16:13] you do know if the developer knows it is a significant feature of linux and states the symbol is gpl [16:13] ohsix, i'd also contend that the intent of the writer of the _GPL in the main is to punish the _vendor_ [16:13] ohsix, but in the main they are punishing the end user [16:14] but then i have a modicom of pragmatism for a geek [16:14] if the vendor cared about the user so much they wouldn't feel threatened by such a silly scheme ;] [16:14] i make not comment on how stupid vendors can be [16:15] i claim that the effort upstream only hurts users [16:15] the derived work thing regarding the gpl is pretty well understood, what it means for the kernel is not [16:15] as here, we likley won't have binary drivers fo nvidia going forward, and that makes users lives hard [16:16] companies may end up creating derived works when they use the thing and at least the right parties will know it is happening [16:17] i contend that we won't know any more than we do now, as they can cheat, and cheat in hidden code and you can't tell [16:17] which is what makes it a farse [16:17] drivers from staging taint the kernel for other reasons [16:17] this isn [16:17] isn't about taint, this is about not being able to do things [16:17] you can tell, though; and if it comes down to it you can demonstrate intent to work around things [16:17] it is about disclosure [16:18] with symbol license statements every party knows what is up with respect to that symbol [16:18] it is a legally questionable farse imo, but then my opinion counts for almost nothing [16:18] take for example the case where we have an EXPORT_SYMBOL for something [16:18] and someone changes the implementation and adds a dependancy which is _GPL [16:19] now you have made that interface break the inner one, and if you fix it you break [16:19] people who had every right to think they could use it [16:19] its just stupid [16:19] as happened with locks recently, so nothing could use locks [16:19] its just not sustainable [16:20] IMO the point of GPL code is to let people do what they want with their computers without big companies getting rich off it [16:20] and _this_ is not served at all by this mess [16:20] people don't have a problem with published shims and a blob, or having any given user being the one to knowingly put them together [16:20] a case where the implementation does _not_ meet the goal and all because of the wording of a licence [16:21] ohsix, except you can't use an _GPL symbols in such a case because the rules are blakc and white [16:21] this is the exact case that the nvidia shim plus clearly and demonstrably externally created blob faces [16:21] this is not the intent (imo) of the gpl [16:22] yes its what the words say, but it is not the intent [16:22] the shim creates an environment that is stable for the code to run in, you essentially need it regardless of whether there's a blob somewhere [16:23] indeed, but that it links to the blob makes it not GPL and by becoming not GPL it cannot use the kernel any more [16:23] so that makes the whole thing pointless, even though you are doing waht is recommended as a solution [16:23] this is why _GPL is a bad solution to any problem [16:25] well there is an analogy to a very american thing, you can buy the lower receiver for an m16, that part 'is' the gun, but it is not a 'gun' until you put the other 95% on it, the barrel, trigger mechanism, butt stock [16:25] all my non-lawyer personal opinion [16:25] people can freely sell and buy the receiver, but at some point after that what is done with it can be highly illegal [16:25] but the point here is if you buy all the bits of that you can put them together, and if you do you get a gun [16:26] whereas i can't put the bits together [16:26] you can't unknowingly put the bits together [16:26] oh and actually putting them together isn't illegal in my case [16:26] i just am not allowed to do it by the tooling [16:27] even though what i want to do is allowed and i am permitted to do it by the licence [16:27] * apw gets grumpy [16:27] you're also allowed to remove the _GPL part from all those statements, as long as it is never distributed, because you can't give that right to anyone else [16:28] tseliot, so to perperuate this farse, i would contend you could make the shim copy and and all source it needs out of the originals, with the _GPL attached, ship that, and in dkms sed -e 's/_GPL//g' and then compile the result, and be lega [16:28] legal [16:29] sure [16:29] but you show intent, and if some legal question comes up over that later, you knowingly did it :] [16:29] which i thing completly makes my point, the only person you are penalising is the end-user [16:29] which is not the intent of the GPL or GPL code in general [16:29] the end user has the right to do such a thing [16:30] and you make his life a living hell to make a pont [16:30] point, a point you should be making to someone else [16:30] the end user isn't distributing the code, generally, most things are binding when the code is distributed [16:31] distributing a binary is distributing the code, and i'm not a lawyer but i'm pretty sure that it's clear the binary counts as a derived work [16:31] you can talk to the fsf and stuff, doesn't canonical have staff lawyers? :D [16:31] i don't dispute that reading of the words, i mearly contend we punish the wrong person [16:32] i don't need to talk to a lawyer to know my opinion on the situation [16:32] i think your opinion may be different if you knew how important it is legally to remove ambiguity by just stating the license for a symbol vs. the alternative [16:32] but fair enough [16:33] no i don't see how _GPL does a thing, the kernel is licenced en-toto under the GPL [16:33] there have been debates over whether using _GPL was even appropriate, and they favor not [16:33] adding or removing _GPL does nothing to change that not adds anything to that story [16:33] it is, but what is a 'derived work' [16:34] that's the thing at question in any case dealing with this [16:34] the licence does not add meaning to that wording therefore it being or not being in the code has no meaning [16:34] _GPL is used where an author considers that anything that would touch that symbol is likely a derived work [16:34] i would contend that i can do whatever the GPL says i can with either interchangably [16:35] it is unlikely they are qualified to make that judgement any more than the lawyer who will [16:35] under a strict legal reading [16:35] as anyone who has spent time with a lawyer will tell you [16:35] any more than _i_ am qualfied [16:36] sure [16:36] nor would i want to be of course [16:36] but it shows intent [16:37] i wiped out a line i decided not to say, but it was like, anything very generally for dealing with virtual memory pages wouldn't be _GPL, per-se; but a function that was basically considered internal to linux' specific vm might be [16:37] i am not sure you can show intent of an action i perform [16:38] anyhow, all academic in real terms [16:39] the only intent I can see is that to make my life miserable as a maintainer when they change functions to GPL only :P [16:39] and that is their intent i am sure :) [16:40] if our source packages were co-installable you might just be able to use that as a base [16:40] so you'd not need to include your source [16:41] it's also not unlike highly visible and deliberately placed 'no tresspassing' signs [16:41] placed on a public park entrance [16:41] :D [16:42] police won't have to speak to property owner to know that they intend to have anyone that doesn't belong there removed, but 'belong there' and everything can happen at a different time [16:42] yes, except in this case end users are allowed to do whatever they like in the park [16:42] provided they don't then have to share the park with someone else ;] [16:43] yes, and the signs don't say "don't share the park" they say "don't enter the park" [16:43] which is my entire point [16:43] 'dont share the park' is implied by 'dont enter the park' [16:43] i don't think you can really mean that [16:43] you can't share a park you can't be in [16:44] nor ice cream you don't have [16:44] yes but i have the right to be in the park [16:44] so telling me "don't enter the park" is just rude and obnoxious [16:44] because you want the park for yourself, no ? [16:44] the park was your thing, i was talking about where clear markers can be placed and be relevant [16:44] you can't smoke in the park, how about that [16:44] if you want me to not share the park licence it to me that way, oh thats exactly what you did [16:45] on the GPL lets me do what i want to the park as long as i don't share it [16:45] i can burn it to the ground, wahtever [16:45] the GPL already tells me i cannot do these things, i _KNOW_ i can't do them [16:45] and i don't do them, take DKMS as fine example [16:45] that follows the rules, it is doing the right thing, and _GPL stops it doing something [16:45] it is allowed to do [16:46] it is the wrong tool for the job [16:46] stop hitting the screw with a hammer i say [16:47] anyhow, this is not going to solve my problem, doing something utterly stupid is going to do that [16:47] 'the gpl' also stops it from doing a lot of things, and the user too [16:47] yes, it does, but not this thing [16:48] it existing at all is a sign of that, you can't then go on and say that _GPL is the problem [16:48] well as an end user i actually can [16:48] if not for 'the gpl', dkms wouldn't have a 'the _GPL' problem [16:48] dkms doesn't have a GPL problem it has only an _GPL prolem [16:48] problem [16:48] BTW, the function that breaks nvidia is flush_workqueue(). Why wasn't it originally _GPL ? And why is it now? [16:49] according to lawyers anyhow [16:49] dkms wouldn't exist if you could distribute the built modules [16:49] right we're not talking about the fact dkms is required as part of the GPL [16:49] we're talkin about the fact that even though it does something which is allowed, the tools prevent it [16:50] the tools wouldn't exist if it wasn't for the license that it is bound to eventually by extra technical measures and point of fact problems [16:50] it's just not a logical sequence of things that makes any sense [16:50] i don't dispute they exist, i dispute they implement the intent of the limitations [16:51] i don't have technological limitations preventing me speeding in my car, its still illegal [16:51] ths is like having a speed regulator at 55 becuase once in the past the speed limit was that [16:53] ohsix, and in the nicest posible sense we are wasting each others time, we are not going to agree here :) [16:54] we don't disagree [16:54] not indeed does it matter what either of us believe as what is is and what lawyers think is what they think [16:54] but i don't think the legal situation here is too fine a point to understand [16:54] if the license were BSD this really really wouldn't be a problem, would it [16:55] it's just matter of fact 'problems' that are implied with what the situation actually is [16:55] tseliot: do you have some idea when it changed? (flush_workqueue) [16:56] i'm git log'ing but this machine has slow disks and not much memory [16:57] tseliot, yeah interestingly there is a commit back on schedule_work which reversed a GPLification of a symbol because it was previously something else [16:58] it looks like it was before historical git epoch, not looking that up, huhuhu [16:58] that stuff lives in kernel/ tho [16:59] tseliot, so i don't think you calling that direct as it has always been _GPL i assume its now called from something else newly [17:00] I really hope it has a non-GPL_ wrapper [17:01] the discussion would really be moot and silly if it came down to a symbol that didn't even matter much [17:03] it's already as silly as it gets. This is not the first time that happens. I've seen much worse [17:04] the dma_buf code though was probably one of the best examples of how pointless GPL_ is [17:05] with NVIDIA eventually adding non-GPL_ wrappers in the kernel and ending the argument [17:05] tseliot, does this use its own workqueue or the system one ? [17:06] again that's not really what the point is, nor is it 'the gpl' [17:06] whats the flush_workqueue incantation if it has one [17:06] ohsix, the GPL is fine and dandy and i support its intent intensly [17:06] all the workqueue symbols are gpl and have been for like 10 years [17:07] apw: I haven't really checked that, or maybe I have and I forgot about it. It's been a while. [17:08] ohsix, well half of them are, "delayed" work isn't, most inconsistant [17:08] there were non-GPL_ wrappers for the workqueue though... [17:08] if I remember correctly [17:09] tseliot, right, i am sure there were for some operations [17:09] * tseliot nods [17:11] tseliot, i can see why you go mad updating this thing ... uggg [17:12] yes, at least we're going to get rid of fglrx, so it's going to be just nvidia (sooner or later) [17:14] tseliot, does this really contain a shar file containing a binary installer, i am going to barf [17:16] apw: fglrx does not (if you're referring to my tarballs), nvidia's certainly does [17:16] tseliot, the latter indeed, uggg [17:17] apw: yes, you might want to run $INSTALLER_NAME -x , then enter the kernel directory [17:18] all right, time for me to go [17:18] * tseliot -> away [17:41] apw: Greetings, just checking on casper and overlayfs. [17:43] unixabg, was there a bug number on that, so i can put it on my todo and make sure i get to it [17:45] apw: every iproute2 crash I can find uses the latest trusty kernel. [17:45] bdmurray, hmmm interesting [17:45] bdmurray, do we have any idea where/what it is exploding in [17:45] apw: that's with the release pocket version of iproute2 or the -updates version [17:46] bdmurray, ok so definatly not the userspace tools i'd say then [17:47] apw: Oh, I didn't realize 3.13.0-59 was superceded. I haven't looked for the latest kernel yet. [17:47] apw: no I had just tested and multiple overlayfs did not work on my test environment. I do not think it hard [17:47] to reproduce and I am not sure if it is a casper thing or not. [17:48] apw: the vast majority of the crashs are when running /sbin/ip route fwiw [17:48] unixabg, ack, thats fine, i suspect i know exactly whats up, and just need to keep a handle on it, i'll file something to remind me [17:48] bdmurray, hmmm ... ok [17:48] bdmurray, i think we need to file a bug against linux with all this info in [17:49] bdmurray, as i don't think it can be userspace if -releaase and -update are fingered, and its all since that kernel went out [17:49] apw: Okay, I'll open a bug. [17:50] bdmurray, thanks, let us know the number here and we'll get someone to look at what you have, include any links to traces you have [17:56] bdmurray, or of ocurse add a task for the existing iproute2 bug for linux, and let us know the number [18:09] apw: bug 1481038 I'm still adding more details though [18:09] bug 1481038 in linux (Ubuntu) "iproute2 crashes being reported since kernel version 3.13-0-59-generic" [Undecided,New] https://launchpad.net/bugs/1481038 [18:09] apw: humble appreciation and if you push something to test I shall do my best to test it. [18:10] bjf, this looks like it might be a kernel issue in routing, though it causes userspace to explode i think [18:12] apw, are you going to work that bug or do you want jsalisbury to start bisecting it? [18:13] bjf, i thnk i am not gonig to look at it today, i am slightly unsure if we cna reproduce it atm whihc would be necessary [18:13] it might be worth seeing if jsalisbury can figure out a reproduce by for sure [18:13] jsalisbury, ^ is that ok with you? [18:14] apw, bjf, ack, I'll take a look [18:15] I'm still running some database queries and let me know if I can help at all. [18:16] bdmurray, anything you can tell us about what combos do and don't show this is good [18:16] bdmurray, do you have the steps you used in comment #1 to reproduce the bug? [18:17] jsalisbury: I was just testing to make sure that apport generated good crashes by manually killing the "ip monitor" process. [18:17] bdmurray, ahh, ok. I'll take a look at the oops once you attach them [18:17] jsalisbury, this seems to be a userspace crash, but it is shown with old and new userpace, and only with the latest kernel [18:18] so it might be a userpace bug being exposed by a kernle change or something [18:18] jsalisbury: unfortunately there is no stacktrace or coredump with these crashes. [18:19] bdmurray, ok, I'll review whatever data we can get [18:19] I've added the oopses from 20150729 [18:19] oh, its all on amd64 too [18:22] bdmurray, I'll spin up a VM with that kernel and play with iproute to see if I can reproduce the issue [18:23] jsalisbury: sounds good, I wonder if (because the crashes are incomplete) it might be happening during boot up or shutdown. [18:24] bdmurray, ack, that should be easy to find out [18:24] jsalisbury, do let me know how you get on [18:24] apw, sure will [18:26] bdmurray, apw, nice, I already have a physical machine running Trusty with the -58 kernel, so I'll just install that kernel and start testing [18:28] jsalisbury, yeah you might want to put a few reboots in your world [18:29] apw, I'll do that with just upgrading the kernel and try some other things out. Then apply the full updates and try again. [18:43] bdmurray, what's the best way to confirm if this bug is reproduced? Will an oops be written to syslog or anything else in another log file? [18:54] bdmurray, I guess I can look for whoopsie messages in syslog and that dialog box will pop up [18:58] bdmurray, apw, upgrading just the kernel brings me up to -61, so I'll roll back to -58 to try and reproduce just in case it might have been fixed [18:58] jsalisbury, ack [19:20] jsalisbury: look for an iproute2 crash file in /var/crash/ [19:28] bdmurray, thanks