Re: Freezing system after kernel 3.2

[Ken Moffat <zarniwhoop@ntlworld.com>]

On Mon, Feb 08, 2016 at 07:08:09PM +0100, Karsten Malcher wrote:
> Hello,
> 
> i am sorry, but is it possible that a kernel bug for a special chipset is alive since kernel 3.2?
> 
On uncommon hardware, anything is possible.  I don't actually know
if that hardware is "uncommon", only that I do not have it.

> I have a backup PC with an Asrock ALiveXFire eSATA2 R3.0 mainboard with CPU AMD64 X2 6000+.
> Before this mainboard runs very stable with Debian wheezy and kernel 3.2.0.
> Now i tried to update to Jessie with kernel 3.16 and the board is crashing within 1-5 minutes after boot!
> 
> There is no clear error, mostly the system is just suddenly freezing without any message or log.
> Sometimes i get a kernel panic like "fatal exception in interrupt", but the boot parameter "pc=nomsi" has no effect.
> 
> I can rule out hardware problems, because the memtest can run for many hours finding nothing.

LOL.  I have a phenom x4 : from time to time (fairly frequently) it
loses its lunch during compiles if I use make -j4.  On less-frequent
occasions it does the same even with make -j1.  And always
memtest86+-5.01 is happy [ well, if I use the "run all CPUs [F2]
option it locks up, but it does that on at least two other mobos
too: one of those is an intel SandyBridge so that issue is not
AMD-specific ].

> Additionally i can boot Knoppix 6.7 with kernel 3.0.4 and it is running stable.
> But when i boot Knoppix 7.2 with kernel 3.9.6 the system is freezing!
> Aditionally i tried out kernel 4.3.0 in Debian but it does not help. Any newer kernel freezes.
> 
> I am sure that newer kernel have a problem with this special mainboard hardware.
> 

If nobody else has better suggestions, I think you will have to
build upstream kernels to find when it broke.  I suggest that you
begin with standard 3.2.latest (just in case you turned out to rely
on something in the debian kernel but not upstream).  Then try
3.9.latest : if that runs ok, continue with 3.16.latest.  If not,
try e.g. 3.4.latest.  The aim is to first find which minor release
broke, and then which update in that series broke it.  What you
*might* need to do is also try .0 versions of each of these.

I am suggesting that you bisect this.  Bisection is usually a pain,
so I suggest that you first find a working version, and then work
through the next stable release of that version to find which commit
broke it.  I suspect I might have phrased my suggestions badly, but
even 3.9 is so long ago that most of us have forgotten about it [ my
last box running a 3.10 LTS kernel was a ppc64, and I have not
booted that for about 2 years ].

Good Luck, and I hope you get a better suggestion.

ĸen
-- 
This email was written using 100% recycled letters.