FAQ Search Today's Posts Mark Forums Read
» Video Reviews

» Linux Archive

Linux-archive is a website aiming to archive linux email lists and to make them easily accessible for linux users/developers.


» Sponsor

» Partners

» Sponsor

Go Back   Linux Archive > Debian > Debian Kernel

 
 
LinkBack Thread Tools
 
Old 02-14-2011, 02:18 AM
Dmitry Smirnov
 
Default Bug#613321: linux-image-amd64: Please enable 'memtest' option for all linux kernels

Package: linux-image-amd64
Severity: wishlist

It would be very nice to have 'memtest' option enabled by default in all Debian
kernels.
This tiny feature is only activated with boot-time parameter so it is harmless,
but powerful for those who want to use it.
As described in http://onlyjob.blogspot.com/2011/01/memtest-explained-linux-
kernel.html
memtest can dramatically increase robustness and to some degree protect from
data corruption on systems with bad RAM.

Thanks.



-- System Information:
Debian Release: 6.0
APT prefers stable
APT policy: (500, 'stable'), (444, 'unstable')
Architecture: amd64 (x86_64)

Kernel: Linux 2.6.32-5-vserver-amd64 (SMP w/8 CPU cores)



--
To UNSUBSCRIBE, email to debian-kernel-REQUEST@lists.debian.org
with a subject of "unsubscribe". Trouble? Contact listmaster@lists.debian.org
Archive: 20110214031817.17803.81923.reportbug@kiki.dev">htt p://lists.debian.org/20110214031817.17803.81923.reportbug@kiki.dev
 
Old 11-29-2011, 10:08 PM
intrigeri
 
Default Bug#613321: linux-image-amd64: Please enable 'memtest' option for all linux kernels

Hi,

Dmitry Smirnov wrote (14 Feb 2011 03:18:17 GMT) :
> It would be very nice to have 'memtest' option enabled by default in
> all Debian kernels.

This wishlist bug is a duplicate of #556365, which was closed (for
very good reasons, if you ask me). Therefore, I believe it should be
closed as well.

OTOH, I have since asked for memtest to be enabled for totally
different reasons, see #646361.

Cheers,
--
intrigeri <intrigeri@boum.org>
| GnuPG key @ https://gaffer.ptitcanardnoir.org/intrigeri/intrigeri.asc
| OTR fingerprint @ https://gaffer.ptitcanardnoir.org/intrigeri/otr.asc
| So what?



--
To UNSUBSCRIBE, email to debian-kernel-REQUEST@lists.debian.org
with a subject of "unsubscribe". Trouble? Contact listmaster@lists.debian.org
Archive: 85hb1mcyrv.fsf@boum.org">http://lists.debian.org/85hb1mcyrv.fsf@boum.org
 
Old 11-29-2011, 11:59 PM
Dmitry Smirnov
 
Default Bug#613321: linux-image-amd64: Please enable 'memtest' option for all linux kernels

Dear intrigeri,

>This wishlist bug is a duplicate of #556365, which was closed (for
>very good reasons, if you ask me). Therefore, I believe it should be
>closed as well.

I very much disagree, unless this wishlist will be closed by enabling the
feature requested.
This *useful* feature has been requested number of times and so far I've only
seen ill excuses like "nobody would use such feature" for not enabling it.
Well, this logic will bring us nowhere.

In-kernel MEMTEST is harmless and can be activated only with boot option.
It won't hurt to enable it simply because someone might need it.
For example OpenSUSE have it for ages.

Secondly, as I demonstrate in my MEMTEST evaluation

http://onlyjob.blogspot.com/2011/01/memtest-explained-linux-kernel.html

It can be handy in paranoid configurations as precausion against data
corruption. It can reduce the damage and therefore it would be good to have
it.

I think we never had a good enough reason for not enabling it.


>OTOH, I have since asked for memtest to be enabled for totally
>different reasons, see #646361.

Fantastic, I'm very glad to see another use for this feature.

Thank you.

Regards,
Dmitry.
 
Old 11-30-2011, 05:55 PM
Dmitry Smirnov
 
Default Bug#613321: linux-image-amd64: Please enable 'memtest' option for all linux kernels

I would like to reopen this discussion since there are some unanswered
questions.

Are we censoring certain Linux Kernel features because "they are not good
enough"?

If so,
* How do we decide what experimental feature is OK?
(For example we have btrfs along with many other things, long before they will
become useful)

* What's the decision criteria? It appears to me that particularly
MEMTEST was refused purely because of prejudice rather than risk assessment.


It is probably obvious to everyone that MEMTEST is harmless.
Then why not enable it without painful discussions?
What's our reason/excuse? (I think arguments like "it is not good enough"
won't be qualified as reasonable answer due to availability of many other
experimental features.)

If just few requests for a feature is not enough to convince that we need it,
how many people should ask, exactly, in order to make the change?
(If I shall I start gathering signatures for petition, how many do I need?


Thank you.

Regards,
Dmitry.
 
Old 12-01-2011, 11:09 AM
Jonathan Nieder
 
Default Bug#613321: linux-image-amd64: Please enable 'memtest' option for all linux kernels

Dmitry Smirnov wrote:

> It is probably obvious to everyone that MEMTEST is harmless.
> Then why not enable it without painful discussions?

Having a feature enabled in the Debian kernel (at release time) is a
promise to continue supporting it for the period in which that release
is supported. It seems perfectly sane to me that the kernel
maintainers might apply their own taste in deciding whether it is
worth doing so.

(Kernel image size is also a consideration, though probably not a
major worry in this particular example.)

[...]
> If just few requests for a feature is not enough to convince that we need it,
> how many people should ask, exactly, in order to make the change?

One reliable contributor. Or just describe a use case that makes it a
compelling addition to Debian. (Since there is already a memtest86+
package, "replacing memtest86+" isn't one.)

Thanks for your interest, and hope that helps,
Jonathan



--
To UNSUBSCRIBE, email to debian-kernel-REQUEST@lists.debian.org
with a subject of "unsubscribe". Trouble? Contact listmaster@lists.debian.org
Archive: 20111201120903.GA10103@elie.hsd1.il.comcast.net">h ttp://lists.debian.org/20111201120903.GA10103@elie.hsd1.il.comcast.net
 
Old 12-01-2011, 12:09 PM
Uwe Kleine-König
 
Default Bug#613321: linux-image-amd64: Please enable 'memtest' option for all linux kernels

On Thu, Dec 01, 2011 at 06:09:03AM -0600, Jonathan Nieder wrote:
> Dmitry Smirnov wrote:
>
> > It is probably obvious to everyone that MEMTEST is harmless.
> > Then why not enable it without painful discussions?
>
> Having a feature enabled in the Debian kernel (at release time) is a
> promise to continue supporting it for the period in which that release
> is supported. It seems perfectly sane to me that the kernel
> maintainers might apply their own taste in deciding whether it is
> worth doing so.
>
> (Kernel image size is also a consideration, though probably not a
> major worry in this particular example.)
>
> [...]
> > If just few requests for a feature is not enough to convince that we need it,
> > how many people should ask, exactly, in order to make the change?
>
> One reliable contributor. Or just describe a use case that makes it a
> compelling addition to Debian. (Since there is already a memtest86+
> package, "replacing memtest86+" isn't one.)
If I understood correctly the usecase is a bit different from
memtest86+. With the latter you have a test case to determine if your
RAM is bad (or not). With the memtest kernel option memory is tested
before it's given out to kmalloc. So it is able in some cases to just
not give out bad parts of RAM allowing to use RAM that is a bit broken.

Having said that I don't know if it's sensible to add to Debian as I
didn't test runtime and binary size overhead.

Best regards
Uwe

--
Pengutronix e.K. | Uwe Kleine-König |
Industrial Linux Solutions | http://www.pengutronix.de/ |



--
To UNSUBSCRIBE, email to debian-kernel-REQUEST@lists.debian.org
with a subject of "unsubscribe". Trouble? Contact listmaster@lists.debian.org
Archive: 20111201130912.GK26618@pengutronix.de">http://lists.debian.org/20111201130912.GK26618@pengutronix.de
 
Old 12-01-2011, 07:59 PM
Jonathan Nieder
 
Default Bug#613321: linux-image-amd64: Please enable 'memtest' option for all linux kernels

Uwe Kleine-König wrote:

> With the latter you have a test case to determine if your
> RAM is bad (or not). With the memtest kernel option memory is tested
> before it's given out to kmalloc. So it is able in some cases to just
> not give out bad parts of RAM allowing to use RAM that is a bit broken.

That would be fun. Alas, the code just seems to run a memory test at
boot time (not at kmalloc-time) and reserve areas that do not pass so
they don't get used during the corresponding run of the kernel.

> Having said that I don't know if it's sensible to add to Debian as I
> didn't test runtime and binary size overhead.

No opinion on that from me. It does seem a shame that many kinds of
faults would be likely to be missed:

http://www.memtest86.com/tech.html#philo

That seems like the bigger potential cost. When someone runs into
corruption that the memtest option did not catch, what can we say to
such a person? (It would be easier if there were a manpage for kernel
parameters and a culture such that everyone read it before enabling
them.)

I should have just been quiet.



--
To UNSUBSCRIBE, email to debian-kernel-REQUEST@lists.debian.org
with a subject of "unsubscribe". Trouble? Contact listmaster@lists.debian.org
Archive: 20111201205902.GD4963@elie.hsd1.il.comcast.net">ht tp://lists.debian.org/20111201205902.GD4963@elie.hsd1.il.comcast.net
 
Old 12-01-2011, 09:18 PM
Dmitry Smirnov
 
Default Bug#613321: linux-image-amd64: Please enable 'memtest' option for all linux kernels

On Thursday 01 December 2011 23:09:03 Jonathan Nieder wrote:

> (Kernel image size is also a consideration, though probably not a

> major worry in this particular example.)




Indeed. In this case we're talking about 130 lines of source code (3 KB), see arch/x86/mm/memtest.c




> > If just few requests for a feature is not enough to convince that we need

> > it, how many people should ask, exactly, in order to make the change?

>

> One reliable contributor. Or just describe a use case that makes it a

> compelling addition to Debian. (Since there is already a memtest86+

> package, "replacing memtest86+" isn't one.)




We have three:




Romain Francoise <rfrancoise@debian.org>, who reported (closed) 556365,

me and intrigeri <intrigeri@boum.org>.



I guess all three of us should be working on our reliability...




> Thanks for your interest, and hope that helps,

> Jonathan




Thank you.
 
Old 12-01-2011, 09:53 PM
Dmitry Smirnov
 
Default Bug#613321: linux-image-amd64: Please enable 'memtest' option for all linux kernels

On Friday 02 December 2011 07:59:02 Jonathan Nieder wrote:
> > Having said that I don't know if it's sensible to add to Debian as I
> > didn't test runtime and binary size overhead.

Binary size overhead is really negligible.

> No opinion on that from me. It does seem a shame that many kinds of
> faults would be likely to be missed:

Let's not fall into discussion about the quality of this feature. It doesn't
matter. It doesn't have to be perfect to be included. Personally I think it is
good enough for inclusion as is, because it do catch some errors.

ECC is not perfect either, and MEMTEST appears to be better or at least as
useful as ECC.


> That seems like the bigger potential cost. When someone runs into
> corruption that the memtest option did not catch, what can we say to
> such a person? (It would be easier if there were a manpage for kernel
> parameters and a culture such that everyone read it before enabling
> them.)

Such person, if capable of activating MEMTEST with boot-time argument to
kernel, may have already read something about it.
We can't take responsibility for that person's decisions (or expectations).

Use case for MEMTEST is not to catch all errors but to minimize damage.
ECC doesn't catch all errors but it is better to have it to avoid massive data
corruption due to bad RAM.

Vast majority of computers out there do not have any form of ECC and we're not
allowing users to have any protection against RAM errors because someone have
unexplained reluctance regarding MEMTEST inclusion.

Is my test case not good enough?

I've seen faulty RAM on servers running 24/7 for years when fault was
discovered only after hardware upgrade. Data corruption was probably happening
for a very long time and impact is very difficult to understand.

MEMTEST can be helpful for this scenario and it can be helpful for notebooks
and desktop PCs where people work with sensitive data and do not routinely
check their memory on weekly/monthly basis. (who does?)


> I should have just been quiet.

No worries, you're thoughts are welcome.




--
To UNSUBSCRIBE, email to debian-kernel-REQUEST@lists.debian.org
with a subject of "unsubscribe". Trouble? Contact listmaster@lists.debian.org
Archive: 201112020953.03562.onlyjob@member.fsf.org">http://lists.debian.org/201112020953.03562.onlyjob@member.fsf.org
 
Old 12-05-2011, 04:53 AM
Ben Hutchings
 
Default Bug#613321: linux-image-amd64: Please enable 'memtest' option for all linux kernels

Your request cannot be satisfied since the MEMTEST option is only
available for x86. However, I suspect that's all you really care about.

Since this feature requires almost no extra memory (the code is all
discardable after boot) I'm prepared to enable it. However, I will
modify it to taint the kernel if any memory fault is detected, on the
basis that there are likely to be other undetected faults.

Ben.

--
Ben Hutchings
Absolutum obsoletum. (If it works, it's out of date.) - Stafford Beer
 

Thread Tools




All times are GMT. The time now is 09:38 AM.

VBulletin, Copyright ©2000 - 2014, Jelsoft Enterprises Ltd.
Content Relevant URLs by vBSEO ©2007, Crawlability, Inc.
Copyright ©2007 - 2008, www.linux-archive.org