V4L-DVB issue in systems with >4Gb RAM?

V4L-DVB issue in systems with >4Gb RAM?

[Helmut Ungar]

Hi,

we are experiencing a problem with the V4L-DVB drivers.
It seems that when the system has over 4Gb the drivers
do no longer work properly. Either nothing or rubbish 
comes out of them, although tuning using szap seems to
work. If we force the system to use only 4Gb by appending
mem=4GB as a kernel parameter things are working like a
charm.

Our setup:
Dell 2850 server with a Magma PCI extender. There are 6 
DVB boards in the machine: 5 KNC TV STAR DVB-S and 
1 Hauppauge Nova-S-Plus DVB-S.
The system has 8GB of RAM and runs an up-to-date Centos5.3, 
kernel 2.6.18-128.4.1.el5 x86_64. 
The V4L-DVB driver we are using is v4l-dvb-2009.08.18.tar.bz2
On this setup some of the KNC boards are working, the Hauppauge
does not. In a similar setup where we have only KNCs none
of them is working unless you force the system to use 4Gb of 
the available memory.

I would like to know if this is a known issue and if so
what can be done to fix/work around the problem.

Any help/suggestion/hint is highly welcome.
Thanks in advance! 

Kind regards,
Helmut

Re: V4L-DVB issue in systems with >4Gb RAM?

[Michael Krufky]

On Wed, Aug 19, 2009 at 10:08 AM, Helmut Ungar<h.ungar@deuromedia.com> wrote:
>
> Hi,
>
> we are experiencing a problem with the V4L-DVB drivers.
> It seems that when the system has over 4Gb the drivers
> do no longer work properly. Either nothing or rubbish comes out of them,
> although tuning using szap seems to
> work. If we force the system to use only 4Gb by appending
> mem=4GB as a kernel parameter things are working like a
> charm.
>
> Our setup:
> Dell 2850 server with a Magma PCI extender. There are 6 DVB boards in the
> machine: 5 KNC TV STAR DVB-S and 1 Hauppauge Nova-S-Plus DVB-S.
> The system has 8GB of RAM and runs an up-to-date Centos5.3, kernel
> 2.6.18-128.4.1.el5 x86_64. The V4L-DVB driver we are using is
> v4l-dvb-2009.08.18.tar.bz2
> On this setup some of the KNC boards are working, the Hauppauge
> does not. In a similar setup where we have only KNCs none
> of them is working unless you force the system to use 4Gb of the available
> memory.
>
> I would like to know if this is a known issue and if so
> what can be done to fix/work around the problem.
>
> Any help/suggestion/hint is highly welcome.
> Thanks in advance!
> Kind regards,
> Helmut

I have a server with three cx23885-based PCI-E boards, one of them
single tuner, the other two with dual tuners.  This server has 8G RAM.
 The single tuner is a Hauppauge board and the dual tuners are DViCO
boards.  (I chose this setup for maximum tuner capacity and brand
diversity for the sake of testing -- I plan to replace the DViCO
boards with two HVR2250's)

So, in summary, my 8G system has five digital tuners and I am not
experiencing the problems that you report.  I doubt the issue is
within the v4l-dvb subsystem.

Good luck,

Mike Krufky

Re: V4L-DVB issue in systems with >4Gb RAM?

[Devin Heitmueller]

On Thu, Aug 20, 2009 at 6:16 PM, Michael Krufky<mkrufky@kernellabs.com> wrote:
> I have a server with three cx23885-based PCI-E boards, one of them
> single tuner, the other two with dual tuners.  This server has 8G RAM.
>  The single tuner is a Hauppauge board and the dual tuners are DViCO
> boards.  (I chose this setup for maximum tuner capacity and brand
> diversity for the sake of testing -- I plan to replace the DViCO
> boards with two HVR2250's)
>
> So, in summary, my 8G system has five digital tuners and I am not
> experiencing the problems that you report.  I doubt the issue is
> within the v4l-dvb subsystem.
>
> Good luck,
>
> Mike Krufky

Someone on v4l yesterday similarly reported a case where the driver
only started to work when switched from the PAE kernel to the non-PAE.
 Perhaps the two issues are related.

Devin

-- 
Devin J. Heitmueller - Kernel Labs
http://www.kernellabs.com

Re: V4L-DVB issue in systems with >4Gb RAM?

[Mauro Carvalho Chehab]

Hi Helmut,

Em Thu, 20 Aug 2009 18:29:15 -0400
Devin Heitmueller <dheitmueller@kernellabs.com> escreveu:

> On Thu, Aug 20, 2009 at 6:16 PM, Michael Krufky<mkrufky@kernellabs.com> wrote:
> > I have a server with three cx23885-based PCI-E boards, one of them
> > single tuner, the other two with dual tuners.  This server has 8G RAM.
> >  The single tuner is a Hauppauge board and the dual tuners are DViCO
> > boards.  (I chose this setup for maximum tuner capacity and brand
> > diversity for the sake of testing -- I plan to replace the DViCO
> > boards with two HVR2250's)
> >
> > So, in summary, my 8G system has five digital tuners and I am not
> > experiencing the problems that you report.  I doubt the issue is
> > within the v4l-dvb subsystem.
> >
> > Good luck,
> >
> > Mike Krufky
> 
> Someone on v4l yesterday similarly reported a case where the driver
> only started to work when switched from the PAE kernel to the non-PAE.
>  Perhaps the two issues are related.

PAE kernel only affects i386 kernels. AFAIK, non-PAE kernels support only 3 Gb
of memory for normal usage, plus 1 Gb for some types of in-kernel memory usage.
As you mentioned that you're using kernel 2.6.18-128.4.1.el5 on x86_64, this
shouldn't be the problem.

I suspect that your issue is related with backporting v4l/dvb to RHEL5 kernel.
The way backport is handled at the subsystem, there are no warranties that you
won't experience any troubles, since there aren't any regression tests and the
priority is to support the latest development kernel. The only test done is
compilation test (and, eventually, a quick test with one of the supported
boards).

The task of carefully handle driver backports is an important contribution made
by the distros at the Linux ecosystem. It isn't (and never was) the intention
of the backport patches at v4l-dvb to replace the distros paper.

In a matter of fact, some of recent kernel internal API changes on some kernel
subsystems affected a lot v4l/dvb drivers, being very hard and complex to fully
support all kernel versions. Also, a few applied changes assumes that certain
bug fixes where applied on some other areas of the kernel in order to work
properly. Yet, even without those patches, for an eventual user, in general,
the driver works good enough for them. 

However, I suspect that, if you're running a RHEL5 kernel, on a machine with
8GB of RAM and 3 PCIe cx23885 cards, probably streaming regularly, your change
of hitting one of those bugs is high.

So, If you have RHEL5 support, I suggest that you should address those issues
via their support channels for they to investigate further and apply the
relevant patches that are affecting your environment

Cheers,
Mauro