Reverting rmmod->modprobe change in bumblebee/3.2.1-8 (was: Re: Bumblebee functionality broken by the last update made by Vincent Cheng on 26-July-2015 - please react as fast as possible)

Fri Aug 7 17:53:55 UTC 2015

Dear Gents,

I appreciate very much to keep me informed about the steps to solve the 
last Ubuntu related Bumblebee problem.

What is very difficult to understand is why the bumblebee.ppa content is 
still untouched at this moment.
As I mentioned in my email (see below the thread), ALL Ubuntu and Mint 
users (which represent at this moment a big part of Linux users) were 
affected and not informed officially on the Bumblebee Project page about 
this topic.

There is an option (better in my opinion) to apply a regression on the 
bumblebee.ppa content and go back to the 3.2.1-90~trustyppa1 version 
which, while it is not perfect, it allows to disable safely the discrete 
card, keeping the laptop power consumption at a considerable lower level.

It is widely accepted in the development of critical Linux OS parts (for 
example) to apply regressions when some new changes add new features not 
enough tested (this might happen to the reduced forces and testing 
environments). This must happen also with the content of such useful 
project like Bumblebee.
As soon as the problem was fixed, you can safely update the 
bumblebee.ppa content.

I was waiting for a prompt reaction, but unfortunately it didn't come yet.
Would you, please, proceed so manner to establish back the bumblebee.ppa 
content with old information till the time the actual problem 
(difference between Ubuntu and Debian) will be solved? This is 
particularly useful for the users which do not know they can use the 
Ubuntu main repositories only.

Once again, I appreciate very much your work,

Kind regards

Cristian Moraru

On 08/07/2015 12:39 PM, Peter Wu wrote:
> Hi Luca,
>
> I just did some tests with a Ubuntu 14.04 VM:
>
>      sudo rm /lib/modules/`uname -r`/kernel/drivers/char/lp.ko
>      sudo depmod
>
> This simulates a mismatch between the kernel module and its file name.
> For testing, I use the following mapping between module names and files:
>      lsmod name  | filename  | simulates
>      lp          | crc7.ko   | nvidia/nvidia-current.ko
>      ppdev       | ppdev.ko  | nvidia-uvm/nvidia-uvm.ko
>
> Then, I tried the following modprobe conffiles and commands:
>
>   - alias nvidia nvidia-current
>     modprobe -vr nvidia
>     No effect!
>   - alias nvidia nvidia-current
>     modprobe -v nvidia
>     Loads nividia-current.ko
>   - remove nvidia nvidia-uvm nvidia-current
>     modprobe -vr nvidia
>     Works, but the exit code is always 1 if one of the modules failed to
>     unload.
>
> Why does Debian not have this issue while Ubuntu does? Debian uses the
> last option[0], Ubuntu apparently does not. I suggest to keep the Debian
> packaging as is because it fixes a real issue (nvidia-uvm should be
> unloaded first) and instead modify the Ubuntu packages.
>
> Since Ubuntu has so many nvidia versions, should it be put in the
> bumblebee package (since that is the only user for now) or should it be
> addressed in each nvidia package? (Tested it, only the first 'remove'
> command will be used in presence of duplicate lines.)
>
> Kind regards,
> Peter
>
>   [0]: https://sources.debian.net/src/nvidia-graphics-drivers/340.76-3/debian/nvidia-modprobe.conf.in/
>
> On Fri, Aug 07, 2015 at 10:08:13AM +0100, Luca Boccassi wrote:
>> Hello Vincent,
>>
>> I'm really sorry for the troubles caused by that patch I backported.
>>
>> Since it is causing troubles on Ubuntu, yes please revert it.
>>
>> I suspect, as you suggested, the cause is the lack of the alias on Ubuntu.
>>
>> Please bear in mind that I think 793386 is not enough to fix 793389 for
>> Cuda users, and they'll have the exact same problem the Ubuntu users are
>> seeing now I'm afraid. I'm writing from my phone as I'm on the move at the
>> moment, so I cannot check right now.
>>
>> In general, since the packaged drivers differ quite a bit between Ubuntu
>> and Debian, do you think it would be worth having separate branches for
>> bumblebee, to avoid issues due to the difference like this?
>>
>> Kind regards,
>> Luca Boccassi
>> On Aug 7, 2015 09:08, "Vincent Cheng" <vcheng at debian.org> wrote:
>>
>>> Hi Luca,
>>>
>>> (cc-ing upstream and original bug reporter)
>>>
>>> It seems that your fix for #793389 (the rmmod->modprobe patch you
>>> committed as [1]), which I recently backported to the bumblebee PPA
>>> [2] (source at [3]) to keep the PPA packages in sync with the packages
>>> in Debian/Ubuntu, is likely the source of a regression for Ubuntu
>>> users using that PPA (see conversation below, and [4]), i.e. the
>>> nvidia kernel module isn't being automatically unloaded using modprobe
>>> for some reason (but apparently worked with rmmod previously). I'm not
>>> sure why this happens; I suspect it has something to do with Ubuntu's
>>> nvidia packages not including an alias even though they rename the
>>> module like Debian's packages. Since it does cause a regression in
>>> Ubuntu, and just by skimming #793389 I don't quite understand whether
>>> this fixes anything in Debian (isn't the fix for #793386 enough to fix
>>> module unloading?), would you be ok with me reverting [1]? Upstream is
>>> also suggesting that the patch be reverted.
>>>
>>> Regards,
>>> Vincent
>>>
>>> [1]
>>> http://anonscm.debian.org/cgit/pkg-nvidia/bumblebee.git/commit/?id=61813829f0d1b2b6134cd7b6d30a2dd071cae5ad
>>> [2] https://launchpad.net/~bumblebee/+archive/ubuntu/stable
>>> [3] https://github.com/Bumblebee-Project/bumblebee-ppa
>>> [4]
>>> https://github.com/Bumblebee-Project/bumblebee-ppa/commit/2d3983e3d70db409c930ae5af2b3d538877854cf#commitcomment-12562494
>>>
>>> On Mon, Aug 3, 2015 at 2:33 PM, Cristian Moraru <mcrsro at gmail.com> wrote:
>>>> Hi Peter,
>>>>
>>>> I am coming back with the result of your suggestions:
>>>>
>>>> FACTS:
>>>>
>>>> - I enabled again the  existing ppa:bumblebee/stable launchpad containing
>>>> version 3.2.1-91~trustyppa1
>>>> - I updated the cache and upgraded the proposals
>>>> - I installed nvidia-346 driver (before as I mentioned, I had the
>>> nvidia-331
>>>> driver which worked with version 3.2.1-90~trustyppa1)
>>>> - I created nvidia-alias.conf as instructed by you
>>>> - Rebooted the machine and checked the effects.
>>>> - All operations were done under basic kernel 3.16.0-38-generic, to
>>> rebuild
>>>> the dkms for all kernels.
>>>>
>>>> I also have to mention I used to install a Mint Cinnamon Applet, named
>>>> "Bumblebee and Nvidia Display", which displays continuously the status of
>>>> Nvidia discrete card and Nvidia GPU Temperature (for details see
>>>> http://cinnamon-spices.linuxmint.com/applets/view/167)
>>>>
>>>> RESULTS:
>>>>
>>>> - The Nvidia card status is ON all the time and doesn't go OFF if I run
>>> in a
>>>> CLI optirun %application% in the manner you mentioned
>>>> - If I start Nvidia X Server Settings, I cannot set the PRIME setting to
>>>> select the as GPU Nvidia, I receive a window with "forbidden" sign.
>>>>
>>>> CONCLUSION:
>>>>
>>>> The actual executables from ppa:bumblebee/stable repository mislead an
>>>> Ubuntu user. A least my machine (which is a pretty representative for a
>>>> business work machine) will not work with bumblebee provided by this
>>>> "stable" repository. I suppose that it might work for Debian BUT IS NOT
>>>> SUITABLE for Ubuntu and its derivatives. I advise the users of Mint 17.x
>>> and
>>>> Ubuntu NOT TO USE IT.
>>>>
>>>>
>>>> LATER, I restored my machine GIVING UP the actual ppa:bumblebee/stable
>>>> repository, as follows
>>>>
>>>> - Purged all nvidia installation with the command
>>>>      sudo apt-get purge nvidia*
>>>> - Purged bumblebee installation with the command
>>>>      sudo apt-get purge bumblebee
>>>> - Removed the existing executables in cache with
>>>>      sudo apt-get clean
>>>> - Removed the actual ppa:bumblebee/stable repository and refreshed the
>>>> apt-cache
>>>> - Installed bumblebee and basic Nvidia driver in the Ubuntu repositories
>>>>      sudo apt-get install bumblebee-nvidia virtualgl virtualgl-libs-ia32
>>>> primus primus-libs-ia32 nvidia-settings libcuda1-304
>>> nvidia-libopencl1-304
>>>> nvidia-opencl-icd-304
>>>> - After this, I recovered the desired bumblebee status and Nvidia driver
>>>> behavior. However I couldn't setup Nvidia as default card with "Nvidia X
>>>> server Settings" facility
>>>> - Installed sequentially with the Mint facility "Driver Manager"
>>> nvidia-331,
>>>> nvidia-331-updates
>>>> - Checked the entire functionality; the only thing which didn't work was
>>> to
>>>> set PRIME to select Nvidia discrete card to be standard choice.
>>>> - Installed with the Mint facility "Driver Manager" nvidia-346 and
>>> checked
>>>> the entire functionality. The result was FULLY POSITIVE: I could select
>>> with
>>>> "Nvidia X server Settings" facility what card to use as prime choice at
>>> boot
>>>> time. The change could be also done without booting, but logging out and
>>>> logging in the user account.
>>>>
>>>> - All operations were done under basic kernel 3.16.0-38-generic, to
>>> rebuild
>>>> the dkms for all kernels. After that I checked the functionality for all
>>>> other installed kernels.
>>>>
>>>> The final status of bumblebee and Nvidia installed executables are shown
>>> in
>>>> the pictures attached to this email.
>>>>
>>>> I also recommend to update the existing information from Bumblebee
>>> project
>>>> which is Ubuntu related to avoid damages to actual Mint and Ubuntu
>>> machines.
>>>> Please recommend users which access the official Bumblebee Project Web
>>> pages
>>>> to install first the version of Bumblebee and Nvidia drivers available on
>>>> the official Ubuntu repositories. The ppa must be used only in case of
>>>> failure with official repositories.
>>>>
>>>> Kind regards
>>>>
>>>> Cristian Moraru
>>>>
>>>>
>>>>
>>>>
>>>>
>>>> On 07/29/2015 09:48 PM, Cristian Moraru wrote:
>>>>> Hi Peter,
>>>>>
>>>>> First of all, thank you for your prompt email!
>>>>>
>>>>> I am not a programmer, but only an advanced Linux user, which appreciate
>>>>> very much the work done by people like you and Vincent for the benefit
>>> of
>>>>> Linux users.
>>>>> I am especially devoted to increase the adoption of Linux (I prefer
>>> Ubuntu
>>>>> and particularly Mint) by professional engineers (Industry, Oil & Gas)
>>>>> because of acquired stability of the LTS Linux versions. I had already
>>> some
>>>>> success and I was surprised how a small change can derail a good and
>>>>> appreciated system.
>>>>>
>>>>> By nature of my work, I was obliged to guide myself to avoid proposing a
>>>>> solution and to put in practice without a comprehensive check. At least
>>> I
>>>>> didn't present a solution without warning about the potential of
>>> additional
>>>>> problems and how to overcome them under specific conditions.
>>>>>
>>>>> When I update my system I believe that all the updating proposals marked
>>>>> with "3" (believed as being OK but not tested) do not create further bad
>>>>> effects to the poor user (which is not a programmer!). I believe this
>>>>> because I have to trust and appreciate the work of the people working
>>> on the
>>>>> respective updates.
>>>>>
>>>>> Bumblebee is a very serious project which I appreciated very much from
>>> the
>>>>> beginning. The fact that it worked out of the box allowed me to
>>> demonstrate
>>>>> the quality of Linux (particularly Linux Mint since version 16) and to
>>>>> convince the Windows guys to switch to Linux.
>>>>>
>>>>> Now, with this unhappy fact I had to restore the functionality going
>>> for a
>>>>> solution which I presented to you.
>>>>> I am convinced that this could be avoided with some precautions. My
>>> email
>>>>> wants to convince you, the developers, to implement a philosophy which
>>> could
>>>>> bring us (developers and users) a substantial increase of Linux
>>> installation
>>>>> base in the professional users field.
>>>>>
>>>>> Coming back to your proposal: I will check your workaround. What I don't
>>>>> understand is why in the version 3.2.1-90~trustyppa1 it was not
>>> necessary to
>>>>> do anything to obtain a good bumblebee behavior, even with nvidia-304
>>> and
>>>>> nvidia-346, and after Vincent pushed the update, I have to modify the
>>> Linux
>>>>> installation (which is not in the power usual Linux user) as you
>>> explained.
>>>>> For the common Linux success, which I believe you and Vincent agree,
>>>>> developers and users must be on the same side when an upgrade is to be
>>>>> promoted.
>>>>>
>>>>> Kind regards
>>>>>
>>>>> Cristian Moraru
>>>>> (a stubborn user of Bumblebee Project)
>>>>>
>>>>>
>>>>> On 07/29/2015 08:00 PM, Peter Wu wrote:
>>>>>> Hi Cristian,
>>>>>>
>>>>>> Vincent is the Debian maintainer of Bumblebee and has recently indeed
>>>>>> published changes for the benefit of Ubuntu users (and derivatives).
>>>>>>
>>>>>> What you are basically experiencing is the inability to turn off the
>>>>>> graphics card after using optirun. This problem can be caused by a
>>>>>> configuration issue in Ubuntu (which probably does not exist in
>>> Debian).
>>>>>> Workaround for version 3.2.1-91~trustyppa1:
>>>>>>
>>>>>> Determine your nvidia module version (say, nvidia-346) and create
>>>>>> /etc/modprobe.d/nvidia-alias.conf containing:
>>>>>>
>>>>>>       alias nvidia nvidia-346
>>>>>>
>>>>>> Besides this, you do not have to restart any services, just trigger
>>>>>> optirun again (optirun true).
>>>>>>
>>>>>> Vincent, what do you think of the proposed patch at
>>>>>>
>>>>>>
>>> https://github.com/Bumblebee-Project/Bumblebee/commit/edd81b063ec1bda63ae8a3b10d8051e02eec282b
>>>>>
>>>>
>>> _______________________________________________
>>> pkg-nvidia-devel mailing list
>>> pkg-nvidia-devel at lists.alioth.debian.org
>>> http://lists.alioth.debian.org/cgi-bin/mailman/listinfo/pkg-nvidia-devel
>>>