Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]

Groups > linux.debian.user > #197249 > unrolled thread

Nvidia drivers

Back to article view | Back to linux.debian.user

Contents

  Nvidia drivers Francisco Mariano-Neto <fmneto@gmail.com> - 2018-07-06 00:50 +0200
    Re: Nvidia drivers Matthew Crews <mailinglists@mattcrews.com> - 2018-07-06 01:40 +0200
    Re: Nvidia drivers Mark Allums <mark@allums.email> - 2018-07-06 03:20 +0200
      Re: Nvidia drivers Mark Allums <mark@allums.email> - 2018-07-06 03:30 +0200
      Re: Nvidia drivers Matthew Crews <mattcrews@mattcrews.com> - 2018-07-06 05:00 +0200
        Re: Nvidia drivers David Baron <dbaron770@gmail.com> - 2018-07-06 08:10 +0200
      Re: Nvidia drivers Darac Marjal <mailinglist@darac.org.uk> - 2018-07-06 11:00 +0200
        Re: Nvidia drivers Francisco M Neto <fmneto@gmail.com> - 2018-07-06 12:20 +0200
          Re: Nvidia drivers Brad Rogers <brad@fineby.me.uk> - 2018-07-06 12:50 +0200
            Re: Nvidia drivers floris <jkfloris@dds.nl> - 2018-07-06 16:10 +0200
              Re: Nvidia drivers Brad Rogers <brad@fineby.me.uk> - 2018-07-06 16:50 +0200
            Re: Nvidia drivers Francisco Mariano-Neto <fmneto@gmail.com> - 2018-07-07 11:50 +0200
              Re: Nvidia drivers Brad Rogers <brad@fineby.me.uk> - 2018-07-07 13:10 +0200
        Re: Nvidia drivers Matthew Crews <mattcrews@mattcrews.com> - 2018-07-06 14:20 +0200
          Re: Nvidia drivers floris <jkfloris@dds.nl> - 2018-07-06 16:20 +0200
            Re: Nvidia drivers Brad Rogers <brad@fineby.me.uk> - 2018-07-06 17:00 +0200

#197249 — Nvidia drivers

[Multipart message — attachments visible in raw view] — view raw

Hey all,

	I'm running kernel 4.15 with nvidia-driver 390.48-3 with no
problems. However, recently my kernel was automatically upgraded to 4.16
and it broke the nvidia driver. 

	Running 'dkms autoinstall --all' does not help, it complains
about not finding kernel headers (which are installed) and quits. 

	Any ideas on how I can rebuild the kernel module for the new
kernel version?

Thanks
Francisco

[toc] | [next] | [standalone]

#197250

[Multipart message — attachments visible in raw view] — view raw

You sure you have the 4.16 headers installed and not the 4.15 headers?

Sent from [ProtonMail](https://protonmail.com), Swiss-based encrypted email.

-------- Original Message --------
On Jul 5, 2018, 15:42, Francisco Mariano-Neto wrote:

> Hey all,
>
> I'm running kernel 4.15 with nvidia-driver 390.48-3 with no
> problems. However, recently my kernel was automatically upgraded to 4.16
> and it broke the nvidia driver.
>
> Running 'dkms autoinstall --all' does not help, it complains
> about not finding kernel headers (which are installed) and quits.
>
> Any ideas on how I can rebuild the kernel module for the new
> kernel version?
>
> Thanks
> Francisco

[toc] | [prev] | [next] | [standalone]

#197253

On 7/5/18 5:42 PM, Francisco Mariano-Neto wrote:
> Hey all,
> 
> 	I'm running kernel 4.15 with nvidia-driver 390.48-3 with no
> problems. However, recently my kernel was automatically upgraded to 4.16
> and it broke the nvidia driver.
> 
> 	Running 'dkms autoinstall --all' does not help, it complains
> about not finding kernel headers (which are installed) and quits.
> 
> 	Any ideas on how I can rebuild the kernel module for the new
> kernel version?
> 
> Thanks
> Francisco
> 
This requires a workaround, a kernel parameter at boot.

GRUB_CMDLINE_LINUX_DEFAULT="slab_common.usercopy_fallback=y"

Edit the config file like this,

$sudo gedit /etc/default/grub

Then run $sudo update-grub.

[toc] | [prev] | [next] | [standalone]

#197256

Sorry,
I failed to read your whole message.  Make sure you have installed the 
the kbuild and headers for your current running kernel.  Then consider 
upgrading your nvidia driver to the latest version (in sid, still, I 
believe).  If you do the latter, be sure and use the kernel parameter I 
showed you in my earlier post.

Mark

On 7/5/18 8:03 PM, Mark Allums wrote:
> On 7/5/18 5:42 PM, Francisco Mariano-Neto wrote:
>> Hey all,
>>
>>     I'm running kernel 4.15 with nvidia-driver 390.48-3 with no
>> problems. However, recently my kernel was automatically upgraded to 4.16
>> and it broke the nvidia driver.
>>
>>     Running 'dkms autoinstall --all' does not help, it complains
>> about not finding kernel headers (which are installed) and quits.
>>
>>     Any ideas on how I can rebuild the kernel module for the new
>> kernel version?
>>
>> Thanks
>> Francisco
>>
> This requires a workaround, a kernel parameter at boot.
> 
> GRUB_CMDLINE_LINUX_DEFAULT="slab_common.usercopy_fallback=y"
> 
> Edit the config file like this,
> 
> $sudo gedit /etc/default/grub
> 
> Then run $sudo update-grub.
> 

[toc] | [prev] | [next] | [standalone]

#197258

On 07/05/2018 06:03 PM, Mark Allums wrote:
> On 7/5/18 5:42 PM, Francisco Mariano-Neto wrote:
>> Hey all,
>>
>> 	I'm running kernel 4.15 with nvidia-driver 390.48-3 with no
>> problems. However, recently my kernel was automatically upgraded to 4.16
>> and it broke the nvidia driver.
>>
>> 	Running 'dkms autoinstall --all' does not help, it complains
>> about not finding kernel headers (which are installed) and quits.
>>
>> 	Any ideas on how I can rebuild the kernel module for the new
>> kernel version?
>>
>> Thanks
>> Francisco
>>
> This requires a workaround, a kernel parameter at boot.
> 
> GRUB_CMDLINE_LINUX_DEFAULT="slab_common.usercopy_fallback=y"
> 
> Edit the config file like this,
> 
> $sudo gedit /etc/default/grub
> 
> Then run $sudo update-grub.
> 

As an aside, thank you for posting this fix. Although my drivers were 
not broken, I did have a problem where xorg would not start correctly 
with the 4.16 kernel and nvidia drivers 390.48 (both from Stretch 
backports). That GRUB command seemed to resolve it my issue.

Cheers.

[toc] | [prev] | [next] | [standalone]

#197262

[Multipart message — attachments visible in raw view] — view raw

On Fri, Jul 6, 2018, 5:57 AM Matthew Crews <mattcrews@mattcrews.com> wrote:

> On 07/05/2018 06:03 PM, Mark Allums wrote:
> > On 7/5/18 5:42 PM, Francisco Mariano-Neto wrote:
> >> Hey all,
> >>
> >>      I'm running kernel 4.15 with nvidia-driver 390.48-3 with no
> >> problems. However, recently my kernel was automatically upgraded to 4.16
> >> and it broke the nvidia driver.
> >>
> >>      Running 'dkms autoinstall --all' does not help, it complains
> >> about not finding kernel headers (which are installed) and quits.
> >>
> >>      Any ideas on how I can rebuild the kernel module for the new
> >> kernel version?
> >>
> >> Thanks
> >> Francisco
> >>
> > This requires a workaround, a kernel parameter at boot.
> >
> > GRUB_CMDLINE_LINUX_DEFAULT="slab_common.usercopy_fallback=y"
> >
> > Edit the config file like this,
> >
> > $sudo gedit /etc/default/grub
> >
> > Then run $sudo update-grub.
> >
>
> As an aside, thank you for posting this fix. Although my drivers were
> not broken, I did have a problem where xorg would not start correctly
> with the 4.16 kernel and nvidia drivers 390.48 (both from Stretch
> backports). That GRUB command seemed to resolve it my issue.
>
> Cheers.
>
> I have 304xx working fine dkms with 4.16. of course these are now
> deprecated. If only nouveau would work. I prefer it..

[toc] | [prev] | [next] | [standalone]

#197267

[Multipart message — attachments visible in raw view] — view raw

On Thu, Jul 05, 2018 at 08:03:22PM -0500, Mark Allums wrote:
>On 7/5/18 5:42 PM, Francisco Mariano-Neto wrote:
>>Hey all,
>>
>>	I'm running kernel 4.15 with nvidia-driver 390.48-3 with no
>>problems. However, recently my kernel was automatically upgraded to 4.16
>>and it broke the nvidia driver.
>>
>>	Running 'dkms autoinstall --all' does not help, it complains
>>about not finding kernel headers (which are installed) and quits.
>>
>>	Any ideas on how I can rebuild the kernel module for the new
>>kernel version?
>>
>>Thanks
>>Francisco
>>
>This requires a workaround, a kernel parameter at boot.
>
>GRUB_CMDLINE_LINUX_DEFAULT="slab_common.usercopy_fallback=y"
>
>Edit the config file like this,
>
>$sudo gedit /etc/default/grub
>
>Then run $sudo update-grub.
>

Actually, as of version 390.67-2, that's no longer needed. Quoting [1]:

	nvidia-graphics-drivers (390.67-2) unstable; urgency=high

	  * Add kmem_cache_create_usercopy.patch from Red Hat, fixing "Bad or missing
		usercopy whitelist? Kernel memory exposure attempt detected from SLUB
		object 'nvidia_stack_cache'" on Linux kernels that have disabled
		CONFIG_HARDENED_USERCOPY_FALLBACK (i.e. linux-image-4.16.0-2-* or newer).
		(Closes: #901919)

	 -- Andreas Beckmann <anbe@debian.org>  Thu, 05 Jul 2018 02:01:31 +0200

[1] http://metadata.ftp-master.debian.org/changelogs/non-free/n/nvidia-graphics-drivers/nvidia-graphics-drivers_390.67-2_changelog

-- 
For more information, please reread.

[toc] | [prev] | [next] | [standalone]

#197269

[Multipart message — attachments visible in raw view] — view raw

Thanks for the answer!

	First, checking installed kernel headers was the first thing I
tried when dkms failed.

	I'll try using the workaround when I get home tonight and see
how it goes. I try not to use too much stuff from sid, especially on
essential systems like this.

	However I did not understand exactly what was the problem. I
assume it was something that was changed in the kernel since it was the
kernel upgrade that "broke" the driver, but it might also have been
something in the driver that was addressed in the kernel. What exactly
happened? I'm not a driver hacker so I don't really understand what is
the issue...

Francisco

On Fri, 2018-07-06 at 09:55 +0100, Darac Marjal wrote:
> On Thu, Jul 05, 2018 at 08:03:22PM -0500, Mark Allums wrote:
> > On 7/5/18 5:42 PM, Francisco Mariano-Neto wrote:
> > > Hey all,
> > > 
> > > 	I'm running kernel 4.15 with nvidia-driver 390.48-3 with no
> > > problems. However, recently my kernel was automatically upgraded
> > > to 4.16
> > > and it broke the nvidia driver.
> > > 
> > > 	Running 'dkms autoinstall --all' does not help, it complains
> > > about not finding kernel headers (which are installed) and quits.
> > > 
> > > 	Any ideas on how I can rebuild the kernel module for the new
> > > kernel version?
> > > 
> > > Thanks
> > > Francisco
> > > 
> > 
> > This requires a workaround, a kernel parameter at boot.
> > 
> > GRUB_CMDLINE_LINUX_DEFAULT="slab_common.usercopy_fallback=y"
> > 
> > Edit the config file like this,
> > 
> > $sudo gedit /etc/default/grub
> > 
> > Then run $sudo update-grub.
> > 
> 
> Actually, as of version 390.67-2, that's no longer needed. Quoting
> [1]:
> 
> 	nvidia-graphics-drivers (390.67-2) unstable; urgency=high
> 
> 	  * Add kmem_cache_create_usercopy.patch from Red Hat, fixing
> "Bad or missing
> 		usercopy whitelist? Kernel memory exposure attempt
> detected from SLUB
> 		object 'nvidia_stack_cache'" on Linux kernels that have
> disabled
> 		CONFIG_HARDENED_USERCOPY_FALLBACK (i.e. linux-image-
> 4.16.0-2-* or newer).
> 		(Closes: #901919)
> 
> 	 -- Andreas Beckmann <anbe@debian.org>  Thu, 05 Jul 2018
> 02:01:31 +0200
> 
> [1] http://metadata.ftp-master.debian.org/changelogs/non-free/n/nvidi
> a-graphics-drivers/nvidia-graphics-drivers_390.67-2_changelog
> 
-- 
--
[]s,

Francisco M Neto

[toc] | [prev] | [next] | [standalone]

#197270

[Multipart message — attachments visible in raw view] — view raw

On Fri, 06 Jul 2018 12:09:59 +0200
Francisco M Neto <fmneto@gmail.com> wrote:

Hello Francisco,

>assume it was something that was changed in the kernel since it was the
>kernel upgrade that "broke" the driver, but it might also have been
>something in the driver that was addressed in the kernel. 

nVidia have made changes to the driver that negate the effects of a
change in the kernel.

So, it appears to be 'six of one, half a dozen of the other'.  That
said, maybe nVidia should have kept a closer eye on kernel changes to
mitigate this problem.

It's strange also that nobody using sid /seems/ to have been affected by
the problem.  Or maybe they were, knew the workaround and forgot to
report the bug.  Who knows?....

-- 
 Regards  _
         / )           "The blindingly obvious is
        / _)rad        never immediately apparent"
It couldn't adapt so it couldn't survive
The Great British Mistake - The Adverts

[toc] | [prev] | [next] | [standalone]

#197276

Brad Rogers schreef op 2018-07-06 12:37:

> 
> It's strange also that nobody using sid /seems/ to have been affected 
> by
> the problem.  Or maybe they were, knew the workaround and forgot to
> report the bug.  Who knows?....

They have found and reported this bug multiple times:

see bug numbers: 901919, 901932, 901990, 902248, 902661, 902773, 902868, 
902891

---
Floris

[toc] | [prev] | [next] | [standalone]

#197280

[Multipart message — attachments visible in raw view] — view raw

On Fri, 06 Jul 2018 16:06:19 +0200
floris <jkfloris@dds.nl> wrote:

Hello floris,

>They have found and reported this bug multiple times:
>
>see bug numbers: 901919, 901932, 901990, 902248, 902661, 902773,
>902868, 902891

Fair enough.  As I was pointed to the initial report quite quickly, I
didn't search for others.  Also, as I subscribed to the bug, I didn't
look closely at the web page for it - thereby missing the "merged
with..." notes.

I also misread the date of the initial report, too;  It was made quite
some time before migration to testing.

-- 
 Regards  _
         / )           "The blindingly obvious is
        / _)rad        never immediately apparent"
Save me from everybody else
Prisoners - Judgement Centre

[toc] | [prev] | [next] | [standalone]

#197311

[Multipart message — attachments visible in raw view] — view raw

Hey all,

On Fri, 2018-07-06 at 11:37 +0100, Brad Rogers wrote:
> 
> nVidia have made changes to the driver that negate the effects of a
> change in the kernel.
> 
> So, it appears to be 'six of one, half a dozen of the other'.  That
> said, maybe nVidia should have kept a closer eye on kernel changes to
> mitigate this problem.

	Thanks for the clarification. I'm no kernel hacker myself, so
sadly I can't really get a grasp on what happened, so this is good
enough =)

> It's strange also that nobody using sid /seems/ to have been affected
> by
> the problem.  Or maybe they were, knew the workaround and forgot to
> report the bug.  Who knows?....

	IIRC someone said earlier on this thread (and I've also seen the
same in other threads about this problem, including the package
maintainers) that the 390.67-2 solved this problem. Since that's the
version that is in sid, that explains why sid doesn't have this issue. 

	In any case, using the workaround worked for me. It caused GNOME
to take a very long time to load, but I'm assuming that's GNOME's fault.

Thanks all!

Francisco

[toc] | [prev] | [next] | [standalone]

#197312

[Multipart message — attachments visible in raw view] — view raw

On Sat, 07 Jul 2018 11:47:59 +0200
Francisco Mariano-Neto <fmneto@gmail.com> wrote:

Hello Francisco,

>maintainers) that the 390.67-2 solved this problem. Since that's the
>version that is in sid, that explains why sid doesn't have this issue. 

That version arrived as the fix.  Prior to that, sid & testing shared
the same version of the nvidia-drivers package for a few days.  As Floris
pointed out, various reports from sid users *did* occur - I simply
didn't check properly.

Floris has also pointed out a bug report that has arisen from using the
fixed version.  It may not affect me, but 'Forewarned is forearmed', as
they say.

-- 
 Regards  _
         / )           "The blindingly obvious is
        / _)rad        never immediately apparent"
Every single one of us
Devil Inside - INXS

[toc] | [prev] | [next] | [standalone]

#197273

On 07/06/2018 01:55 AM, Darac Marjal wrote:

> Actually, as of version 390.67-2, that's no longer needed. Quoting [1]:
> 
> 	nvidia-graphics-drivers (390.67-2) unstable; urgency=high
> 
> 	  * Add kmem_cache_create_usercopy.patch from Red Hat, fixing "Bad or missing
> 		usercopy whitelist? Kernel memory exposure attempt detected from SLUB
> 		object 'nvidia_stack_cache'" on Linux kernels that have disabled
> 		CONFIG_HARDENED_USERCOPY_FALLBACK (i.e. linux-image-4.16.0-2-* or newer).
> 		(Closes: #901919)
> 
> 	 -- Andreas Beckmann <anbe@debian.org>  Thu, 05 Jul 2018 02:01:31 +020
0
> 
> [1] http://metadata.ftp-master.debian.org/changelogs/non-free/n/nvidia-graphics-drivers/nvidia-graphics-drivers_390.67-2_changelog
> 

Thats good to know.

Unfortunately, Stretch-backports and Buster Nvidia-drivers are not in 
sync with Sid right now. So this is still a necessary workaround for 
Stretch-backports and Buster users.

[toc] | [prev] | [next] | [standalone]

#197278

Matthew Crews schreef op 2018-07-06 13:55:
> On 07/06/2018 01:55 AM, Darac Marjal wrote:
> 
>> Actually, as of version 390.67-2, that's no longer needed. Quoting 
>> [1]:
>> 
>> 	nvidia-graphics-drivers (390.67-2) unstable; urgency=high
>> 
>> 	  * Add kmem_cache_create_usercopy.patch from Red Hat, fixing "Bad or 
>> missing
>> 		usercopy whitelist? Kernel memory exposure attempt detected from 
>> SLUB
>> 		object 'nvidia_stack_cache'" on Linux kernels that have disabled
>> 		CONFIG_HARDENED_USERCOPY_FALLBACK (i.e. linux-image-4.16.0-2-* or 
>> newer).
>> 		(Closes: #901919)
>> 
>> 	 -- Andreas Beckmann <anbe@debian.org>  Thu, 05 Jul 2018 02:01:31 
>> +0200
>> 
>> [1] 
>> http://metadata.ftp-master.debian.org/changelogs/non-free/n/nvidia-graphics-drivers/nvidia-graphics-drivers_390.67-2_changelog
>> 
> 
> Thats good to know.
> 
> Unfortunately, Stretch-backports and Buster Nvidia-drivers are not in
> sync with Sid right now. So this is still a necessary workaround for
> Stretch-backports and Buster users.


and there is a new bug:
https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=903121

So be careful if you use the nvidia module.

---
Floris

[toc] | [prev] | [next] | [standalone]

#197281

[Multipart message — attachments visible in raw view] — view raw

On Fri, 06 Jul 2018 16:10:40 +0200
floris <jkfloris@dds.nl> wrote:

Hello floris,

>and there is a new bug:
>https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=903121
>So be careful if you use the nvidia module.

Thanks for the warning.

-- 
 Regards  _
         / )           "The blindingly obvious is
        / _)rad        never immediately apparent"
Is she really going out with him?
New Rose - The Damned

[toc] | [prev] | [standalone]

Back to top | Article view | linux.debian.user

csiph-web