> On Wednesday 20 August 2003 14:3, Ruben Puettman wrote:
>> On Wednesday 20 August 2003 13:59, Marc-Christian Petersen wrote:
>>
>> Hi Alan,
>>
>>> On Wednesday 20 August 2003 13:46, Alan Cox wrote:
>>> he 2.2 tree needs a new maintainer, someone who can spend their entire
>>> life refusing patches, being ignored by the mainstream (because 2.2 is
>>> boring) and by vendors (who don't ship 2.2 any more).
>>
>> I want to take 2.2.
>>
> What's up with linux-2.2 now? Who will do Alan's job in the next year?
>
> Marc is intrested doing this job. I know Marc from linux-2.2.x-secure
> and from the wolk project, see http://wolk.sourceforge.net. Why not
> Marc? He can surely differentiate between mainstream and a private kernel fork
> tree. Or do you think, Marc will merge every single bit out of his kernel tree
> into mainstream? I bet he won't
>
> I can't see postings from other people who want to take 2.2.
>
> I think 2.2 is not dead. I often see 2.2 kernels running on systems like
> wlan access points or dsl routers from different vendors. 2.2 is often
> used where stability is a must-have. At least security fixes have to go in.
I agree.
> What do you think?
Well, I think that 2.2.24 and 2.2.25 kernels are really stable (at least on
UP), but that the most weak side is on IDE disk drivers.
They seem to have DMA problems when using recent hard disks (i.e. Maxtor,
etc.)
that lead to serious file system corruption problems.
Maybe there are also geometry problems because all troubles have been
observed
on disks with more than 32 GB of capacity (i.e. 40 GB).
This is a pity because, up to now, 2.2.x kernels have been
a valid choice for small / semi-embedded systems 80x86
(yes, I know that 2.4 should be better, but I'm still waiting for
a stable rock kernel).
In conclusion, I hope that next maintainer will think about
these matters:
IDE drivers;
security fixes;
micro-optimizations;
compatibility with newer compilers.
After all if 2.0 seems to be still alive also 2.2 should be.
A. De Faccio
On Friday 22 August 2003 23:27, A.D.F. wrote:
Hi A. De Faccio
> > I think 2.2 is not dead. I often see 2.2 kernels running on systems like
> > wlan access points or dsl routers from different vendors. 2.2 is often
> > used where stability is a must-have. At least security fixes have to go
> > in.
> I agree.
> > What do you think?
> Well, I think that 2.2.24 and 2.2.25 kernels are really stable (at least on
> UP), but that the most weak side is on IDE disk drivers.
My -secure tree is also rock solid on SMP :p
> They seem to have DMA problems when using recent hard disks (i.e. Maxtor,
> etc.) that lead to serious file system corruption problems.
> Maybe there are also geometry problems because all troubles have been
> observed on disks with more than 32 GB of capacity (i.e. 40 GB).
> This is a pity because, up to now, 2.2.x kernels have been
> a valid choice for small / semi-embedded systems 80x86
> (yes, I know that 2.4 should be better, but I'm still waiting for
> a stable rock kernel).
I agree on this. Therefore my 2.2-secure tree has a 2.4 IDE backport from the
PLD Project by Krzysiek Taraszka & Krzysiek Oledzki. It's not that up2date
like .21 and .22 IDE code is, but it works very very smooth and nice and rock
solid. We use the 2.2-secure tree for almost all customers in my company.
Biggest harddisk is a 160GB Maxtor IDE disk.
> In conclusion, I hope that next maintainer will think about
> these matters:
> IDE drivers;
ack!
> security fixes;
ack! Current 2.2 is missing, for example, hashing exploits in network stack,
like 2.4 had some time ago.
> micro-optimizations;
also done in 2.2-secure
> compatibility with newer compilers.
This might be the hardest job. This is not done in 2.2-secure. I think the
effort in doing this is not worth the time it takes.
> After all if 2.0 seems to be still alive also 2.2 should be.
I agree 100%. Anyway, no comment from Alan, so I think he don't want to give
2.2 away to me.
P.S.: I've cc'ed Ruben, Alan and Ville.
ciao, Marc
On Llu, 2003-08-25 at 17:42, Marc-Christian Petersen wrote:
> like .21 and .22 IDE code is, but it works very very smooth and nice and rock
> solid. We use the 2.2-secure tree for almost all customers in my company.
> Biggest harddisk is a 160GB Maxtor IDE disk.
The problem is that change breaks stuff. a lot of the 2.2 users will
happily trade lack of LBA48 support for stability and predictability.
Thats why I took a basically "if its not a serious bugfix its not going
in" approach
On 26 Aug 2003, Alan Cox wrote:
> On Llu, 2003-08-25 at 17:42, Marc-Christian Petersen wrote:
> > like .21 and .22 IDE code is, but it works very very smooth and nice and rock
> > solid. We use the 2.2-secure tree for almost all customers in my company.
> > Biggest harddisk is a 160GB Maxtor IDE disk.
>
> The problem is that change breaks stuff. a lot of the 2.2 users will
> happily trade lack of LBA48 support for stability and predictability.
> Thats why I took a basically "if its not a serious bugfix its not going
> in" approach
As a 2.2 user, I have to agree with this sentiment.
Just as Rome wasn't built in a day, 2.2 didn't get as stable as it is
overnight. Unfortunately, any change now risks subtley destabilizing 2.2
for some obscure case somewhere. Such destabilization could literally
happen overnight {:-(
OTOH, some users may find that the stability of 2.2 combined with some
other feature(s) (be it LBA48/IPSec/whatever (in my case it's MPPE from
-secure + a timer hack)) works just fine for them.
I.e. we can be well-served by a very stable "official" 2.2 tree which is
supplemented by the likes of MCP's -secure patchset, and others.
Perhaps the middle ground is to keep the 2.2 kernel pretty much as-is
(modulo bugfixes) and include pointers to optional patch(sets)? That's
certainly more palatable than people maintaining patch-reversions to
retain their stability!
Worse still, as is too often the case, feature updates may provoke some
2.2 users to NOT upgrade (and hence miss security fixes) lest they
compromise stability.
HTH and many thanks for your contributions,
Neale.
On Tuesday 26 August 2003 17:03, Alan Cox wrote:
Hi Alan,
> The problem is that change breaks stuff. a lot of the 2.2 users will
> happily trade lack of LBA48 support for stability and predictability.
> Thats why I took a basically "if its not a serious bugfix its not going
> in" approach
Yeah, I agree with you. Anyway, I've never said I will integrate that IDE
stuff when I become the 2.2 maintainer :) ... I just said I'll think about it
;) ... I never ever want to break 2.2, and such an update will definitively
break things. 2.2 took a long time to become that stable like it is now and it
can be broke within minutes. A no-go!
For users who need/want/experiment with that stuff there is still my
2.2-secure tree.
Anyway, 2.2 needs the hashing exploit fix ASAP ;)
ciao, Marc