[gentoo-user] linux-headers

[gentoo-user] linux-headers

[Allan Gottlieb]

I believe there is some rule concerning the versions of gentoo-sources
vs linux-headers.  Could someone please remind me of the rule or
correct my impression that such a rule exists?

I am running 2.6.20-gentoo-r7 but have installed 2.6.22-r8

emerge --verbose --ask --deep --update --newuse --tree world
offered to merge sys-kernel/linux-headers-2.6.22-r2

Should I do this merge?

thanks,
allan gottlieb
-- 
gentoo-user@gentoo.org mailing list

Re: [gentoo-user] linux-headers

[Hex Star]

[-- Attachment #1: Type: text/plain, Size: 32 bytes --]

There is no harm in doing so :)

[-- Attachment #2: Type: text/html, Size: 36 bytes --]

Re: [gentoo-user] linux-headers

[Hans-Werner Hilse]

Hi,

On Sat, 6 Oct 2007 23:38:33 -0700
"Hex Star" <hexstar@gmail.com> wrote:

> There is no harm in doing so :)

If you say such things, please add a short explanation what makes you
think that. After all this isn't IRC.

In fact, I would not suggest doing that. While kernel developers do
their best not to break existing interfaces unless they have real urge,
the picture also might get bigger, i.e. more or different APIs. So I
would not suggest running a userland based on headers with higher
version than the actual kernel. Although I have to admit that I don't
know if there candidates for unexpected behaviour and what those might
be.

-hwh
-- 
gentoo-user@gentoo.org mailing list

Re: [gentoo-user] linux-headers

[Volker Armin Hemmann]

On Sonntag, 7. Oktober 2007, Hans-Werner Hilse wrote:
> Hi,
>
> On Sat, 6 Oct 2007 23:38:33 -0700
>
> "Hex Star" <hexstar@gmail.com> wrote:
> > There is no harm in doing so :)
>
> If you say such things, please add a short explanation what makes you
> think that. After all this isn't IRC.
>
> In fact, I would not suggest doing that. While kernel developers do
> their best not to break existing interfaces unless they have real urge,
> the picture also might get bigger, i.e. more or different APIs. So I
> would not suggest running a userland based on headers with higher
> version than the actual kernel. Although I have to admit that I don't
> know if there candidates for unexpected behaviour and what those might
> be.
>
> -hwh

the linux-headers and the kernel are completly off -sync. There is no harm in 
using headers with a higher/lower version number.
But(!) you should not downgrade headers, this can cause very severe problems.
-- 
gentoo-user@gentoo.org mailing list

Re: [gentoo-user] linux-headers

[Allan Gottlieb]

At Sun, 07 Oct 2007 16:06:41 +0200 Volker Armin Hemmann <volker.armin.hemmann@tu-clausthal.de> wrote:

> the linux-headers and the kernel are completly off -sync. There is no harm in 
> using headers with a higher/lower version number.
> But(!) you should not downgrade headers, this can cause very severe problems.

thanks,
allan
-- 
gentoo-user@gentoo.org mailing list

Re: [gentoo-user] linux-headers

[Benno Schulenberg]

Volker Armin Hemmann wrote:
> But(!) you should not downgrade headers, this can cause very
> severe problems.

Could you give an example?

Benno
-- 
gentoo-user@gentoo.org mailing list

Re: [gentoo-user] linux-headers

[Volker Armin Hemmann]

On Montag, 8. Oktober 2007, Benno Schulenberg wrote:
> Volker Armin Hemmann wrote:
> > But(!) you should not downgrade headers, this can cause very
> > severe problems.
>
> Could you give an example?

ok, example. I might be totally wrong, so don't believe me:
 The splice system call was added with 2.6.17 and corresponding headers. If 
you build an application that has optinal (on compile time) support for this, 
but downgrade the headers after that to say... 2.6.10 you might see funny 
stuff happening.

Headers are backward compatible, not forward compatible.

glibc is similar - just try to downgrade glibc. You can't, portage won't allow 
it.
-- 
gentoo-user@gentoo.org mailing list

Re: [gentoo-user] linux-headers

[Hans-Werner Hilse]

Hi,

On Mon, 8 Oct 2007 00:44:21 +0200
Volker Armin Hemmann <volker.armin.hemmann@tu-clausthal.de> wrote:

> ok, example. I might be totally wrong, so don't believe me:
>  The splice system call was added with 2.6.17 and corresponding headers. If 
> you build an application that has optinal (on compile time) support for this, 
> but downgrade the headers after that to say... 2.6.10 you might see funny 
> stuff happening.

Nah, you won't. Nothing depends on the headers on runtime, they just
matter on compile time. After downgrading, when you compile new stuff,
it will use the now older headers, i.e. it will probably rely on older
feature sets.

Whether the software compiled against newer headers will still work
depends solely on the kernel. In your example, when you also decide to
run a 2.6.10 kernel, then the software relying on newer features (due
to the newer headers on compile time) will have problems.

> Headers are backward compatible, not forward compatible.

That's not true how you have put it. But it is a misleading thing, all
this. The _kernel_ has a backward compatible interface to userland
(most of the time, and almost definitely regarding the syscalls).
That's why you can still run that old statically compiled binary from
19-you-know-what. The _headers_ on the opposite belong more to the
userland software camp. They are what userland knows about the kernel
at compile time. So _if_ the kernel keeps backward compatibility, the
_headers_ are forward compatible.
(The underscoring is meant as emphasis, but not to be harsh)

> glibc is similar - just try to downgrade glibc. You can't, portage won't allow 
> it.

That's the same for downgrading the kernel and it works the same:
Userland is compiled against newer glibc headers.

-hwh

(It's all somewhat complicated, that's why I thought the short
one-liner wasn't a definite answer...)
-- 
gentoo-user@gentoo.org mailing list

Re: [gentoo-user] linux-headers

[Allan Gottlieb]

At Mon, 08 Oct 2007 23:08:40 +0200 Hans-Werner Hilse <hilse@web.de> wrote:

> Whether the software compiled against newer headers will still work
> depends solely on the kernel. In your example, when you also decide to
> run a 2.6.10 kernel, then the software relying on newer features (due
> to the newer headers on compile time) will have problems.

Does that mean I am at some risk with headers at a higher version than
the kernel?  I followed the advice at the end of the headers emerge
and remerged glibc.

thanks,
allan
-- 
gentoo-user@gentoo.org mailing list

Re: [gentoo-user] linux-headers

[Hans-Werner Hilse]

Hi,

On Mon, 08 Oct 2007 18:50:56 -0400 Allan Gottlieb <gottlieb@nyu.edu>
wrote:

> Does that mean I am at some risk with headers at a higher version than
> the kernel?  I followed the advice at the end of the headers emerge
> and remerged glibc.

Most probably no danger here. The interfaces of the kernel seldom
change that radical that a John Doe user would have to care. Also, most
software is supposed to leave the kernel headers alone anyway. And you
took care of glibc, so that's probably not going to cause headaches.
Since that was a re-emerge, it won't produce a new interface for
userland.

I just explained why I found the first answer to your question somewhat
lacking of argumentation and the further answers to my post then were
just plain wrong. The moral is to never believe people who just claim
stuff without giving a good reasoning.

-hwh
-- 
gentoo-user@gentoo.org mailing list

Re: [gentoo-user] linux-headers

[Allan Gottlieb]

At Tue, 09 Oct 2007 14:48:15 +0200 Hans-Werner Hilse <hilse@web.de> wrote:

> On Mon, 08 Oct 2007 18:50:56 -0400 Allan Gottlieb <gottlieb@nyu.edu>
> wrote:
>
>> Does that mean I am at some risk with headers at a higher version than
>> the kernel?  I followed the advice at the end of the headers emerge
>> and remerged glibc.
>
> Most probably no danger here. The interfaces of the kernel seldom
> change that radical that a John Doe user would have to care. Also, most
> software is supposed to leave the kernel headers alone anyway. And you
> took care of glibc, so that's probably not going to cause headaches.
> Since that was a re-emerge, it won't produce a new interface for
> userland.
>
> I just explained why I found the first answer to your question somewhat
> lacking of argumentation and the further answers to my post then were
> just plain wrong. The moral is to never believe people who just claim
> stuff without giving a good reasoning.

thanks for the explanation.
allan
-- 
gentoo-user@gentoo.org mailing list

[gentoo-user] linux-headers

[Michael P. Soulier]

[-- Attachment #1: Type: text/plain, Size: 922 bytes --]

Hello,

I'm currently running kernel 2.6.25 and I have no issues with it so I don't
really want to upgrade it just yet. I just picked up
sys-kernel/linux-headers-2.6.27-r2, so I thought I'd mask out anything above
2.6.25 for now to keep the headers in sync with the kernel that I'm running.

So I put this in my /etc/portage/package.mask:

>=sys-kernel/linux-headers-2.6.26

And now

msoulier@anton:~$ emerge --pretend --update world

These are the packages that would be merged, in order:

Calculating dependencies... done!
[ebuild     UD] sys-kernel/linux-headers-2.6.23-r3 [2.6.27-r2]

So, there's no 2.6.25 version? The last one before 2.6.26 was 2.6.23?

Thanks,
Mike
-- 
Michael P. Soulier <msoulier@digitaltorque.ca>
"Any intelligent fool can make things bigger and more complex... It takes a
touch of genius - and a lot of courage to move in the opposite direction."
--Albert Einstein

[-- Attachment #2: Type: application/pgp-signature, Size: 189 bytes --]

Re: [gentoo-user] linux-headers

[Paul Hartman]

On Tue, Jan 13, 2009 at 8:09 PM, Michael P. Soulier
<msoulier@digitaltorque.ca> wrote:
> Hello,
>
> I'm currently running kernel 2.6.25 and I have no issues with it so I don't
> really want to upgrade it just yet. I just picked up
> sys-kernel/linux-headers-2.6.27-r2, so I thought I'd mask out anything above
> 2.6.25 for now to keep the headers in sync with the kernel that I'm running.
>
> So I put this in my /etc/portage/package.mask:
>
>>=sys-kernel/linux-headers-2.6.26
>
> And now
>
> msoulier@anton:~$ emerge --pretend --update world
>
> These are the packages that would be merged, in order:
>
> Calculating dependencies... done!
> [ebuild     UD] sys-kernel/linux-headers-2.6.23-r3 [2.6.27-r2]
>
> So, there's no 2.6.25 version? The last one before 2.6.26 was 2.6.23?
>
> Thanks,
> Mike

There were, but no longer in portage. You can get them from here and
put them in a local overlay:

http://sources.gentoo.org/viewcvs.py/gentoo-x86/sys-kernel/linux-headers/?hideattic=0

Regards,
Paul