[mythtv-users] MythTV problems with 3.7.x and 3.8.x kernel

[Larry Finger]

On 02/25/2013 08:22 PM, Neil Salstrom wrote:
> On Sun, Feb 24, 2013 at 8:48 AM, Larry Finger <Larry.Finger at lwfinger.net> wrote:
>>
>> I work in driver development for wireless devices and need to work with the
>> latest cutting-edge kernels. As such, I have bisected for regressions many
>> times. Note: The mainline tree only knows about the .0 version - any
>> follow-on fixes are kept in a separate repository. Mainline refers to
>> versions as v3.7, not v3.7.0.
>>
>> The 'git bisect' tool is very powerful; however, when bisecting between
>> releases such as 3.6 and 3.7, the first set of commits frequently cause the
>> most problems with build errors and run-time faults that have nothing to do
>> with the bug under investigation. I would suggest starting with the git
>> command 'git checkout v3.7-rc1'. If the resulting kernel is OK, then you
>> bisect between v3.7 (bad) and v3.7-rc1 (good) and you will avoid the worst
>> of those problems. If -rc1 has the problem with the DVD, then the bad
>> version is v3.7-rc1 and good is v3.6.
>>
>> Good luck,
>>
>> Larry
>
> SOLVED!!!!!!!!!!!!!
>
> After ~12 or so bisect iterations this issue has been solved.  I don't
> pretend to understand why it works but it does.
>
> Oddly, the change happened in drivers/net/loopback.c starting with the
> 3.7 series kernel (commit 0cf833aefaa85bbfce3ff70485e5534e09254773).
> Here is summary:
>
> net: loopback: set default mtu to 64K
>
> loopback current mtu of 16436 bytes allows no more than 3 MSS TCP
> segments per frame, or 48 Kbytes. Changing mtu to 64K allows TCP
> stack to build large frames and significantly reduces stack overhead.
>
> Performance boost on bulk TCP transferts can be up to 30 %, partly
> because we now have one ACK message for two 64KB segments, and a lower
> probability of hitting /proc/sys/net/ipv4/tcp_reordering default limit.
>
>
> The odd thing is there is no networking involved in playing DVDs on my
> system.  It's a combined frontend / backend and the media drive is
> local to the system.  Going back to the original mtu of 16346 bytes
> fixes the stuttering when a DVD is played.  I've tested this fix with
> both 3.7.9 and 3.8.0 kernels.
>
> Does MythTV mount a dvd or .iso over a loopback device?
>
> Now the question is who needs to see this (if anyone)?  The author of
> the commit or a MythTV-dev?
>
> Thanks for everyone's help in figuring this out.  Git bisect....
> that's a new one to add to my list of tricks.

You should prepare a description of this problem and send it to Eric Dumazet 
<edumazet at google.com>, the author of the patch, and David S. Miller 
<davem at davemloft.net>, the maintainer of the net system, with a Cc to 
linux-kernel at vger.kernel.org, which is usually called LKML. Call it a 
regression, describe the symptoms, the bisection to commit 0cf833a (7 digits are 
enough), and that removing that patch restored proper operation. I do not know 
either of those men personally, but I have had E-mail contact with both, and I 
have found them to be quite knowledgeable.

Larry