FYI: /sbin/weak-modules

[Jon Masters]

Hi folks,

Several people have been asking what /sbin/weak-modules is/does,
mostly because it started script spewing the other day (sorry about
that!) on kernel updates.

weak-modules is part of some driver updates work we're doing at Red
Hat and also involves folks working on Fedora Extras "kmod" kernel
packaging (see the packaging list archives). Essentially, this and
other scripts are part of a system for allowing compatible kernel
modules to continue to work after a kernel update occurs -
weak-modules is the part that figures out which kernels are compatible
with a driver.

There's more information (and some out of date packages - look at
rawhide for the current latest stuff!) at
http://www.kerneldrivers.org/ and I will add documentation for those
who were looking for it and could not find any - I'm sorry that's not
been done yet.

I'm trying to work with the folks at SuSE/Novell to ultimately offer
kernel module package authors a standardized packaging process too
(over and above what we each do separately) - I'm going to start by
introducing a new macro in redhat-rpm-config along the lines agreed
between those of us who met up at OLS to discuss driver updates last
week. This is not aimed at undermining the work that's been done in
Extras - it's just an additional option.

If you would like more information, just drop me an email (I'm also
jcm(a)redhat.com when wearing my work hat!) or reply to this one.

Cheers!

Jon.

[Sam Varshavchik]

Jon Masters writes:

>  Hi folks,
>  
>  Several people have been asking what /sbin/weak-modules is/does,
>  mostly because it started script spewing the other day (sorry about
>  that!) on kernel updates.
>  
>  weak-modules is part of some driver updates work we're doing at Red
>  Hat and also involves folks working on Fedora Extras "kmod" kernel
>  packaging (see the packaging list archives). Essentially, this and
>  other scripts are part of a system for allowing compatible kernel
>  modules to continue to work after a kernel update occurs -
>  weak-modules is the part that figures out which kernels are compatible
>  with a driver.
>  
>  There's more information (and some out of date packages - look at
>  rawhide for the current latest stuff!) at
>  http://www.kerneldrivers.org/ and I will add documentation for those
>  who were looking for it and could not find any - I'm sorry that's not
>  been done yet.
>  
>  I'm trying to work with the folks at SuSE/Novell to ultimately offer
>  kernel module package authors a standardized packaging process too
>  (over and above what we each do separately) - I'm going to start by
>  introducing a new macro in redhat-rpm-config along the lines agreed
>  between those of us who met up at OLS to discuss driver updates last
>  week. This is not aimed at undermining the work that's been done in
>  Extras - it's just an additional option.
>  
>  If you would like more information, just drop me an email (I'm also
>  jcm(a)redhat.com when wearing my work hat!) or reply to this one. 
What exactly is the problem being solved here?  I don't see any obvious link 
to anything on kerneldrivers.org that describes what this stuff does.

I build lirc and ivtv modules each time a new kernel comes out, and I have 
absolutely no problems doing that.  Couldn't be any easier.  Piece of cake. 
I'm not sure what problem needs solving here.  I just install the new kernel 
and kernel-devel packages, run the script that builds a corresponding lirc 
and ivtv kernel driver packages, pointing the build at the right build 
directory, then install the driver packages and reboot.  Couldn't be 
simpler.

If someone really wants to make building kernel driver packages easier, they 
can simply fix THE HORRIBLE UGLY STINKING CRAP code that kernel-devel's 
%post script uses to hardlink the identical kernel source files.

If I set out to do this specific task in the most grossly inefficient manner 
possible, there's no way I could do it any better than the existing mess. 
The brain-damaged script literally reads the entire /usr/src/kernels, by 
forking thousands of mind-numbing child procs.  And, the more kernels you 
have installed, the slooower and sloooooooooooooower everything takes.

Has anyone ever heard of this thing called an SHA1 hash?

Anybody?

Hello!!! McFly!!! Anybody home????

You don't really need to be a rocket science in order to figure out that the 
right way to do this is hash every file in the kernel-devel subpackage 
(after all, if it's good enough for %config, it's good enough for 
/usr/src/kernels), add a single file containing all the hash to the 
kernel-devel subpackage, then have %post simply compare this file against 
the hash file from other kernel-devel packages, thus knowing immediately, 
and instantly, what needs to be hardlinked to what, without grepping the 
entire bloody /usr/src/kernels tree.

This gets old very quickly when you have to install both the SMP and UP 
kernel-devel packages, for each kernel release.

[Rahul Sundaram]

Sam Varshavchik wrote:
>  Jon Masters writes:
>  
> > Hi folks,
> >
> > Several people have been asking what /sbin/weak-modules is/does,
> > mostly because it started script spewing the other day (sorry about
> > that!) on kernel updates.
> >
> > weak-modules is part of some driver updates work we're doing at Red
> > Hat and also involves folks working on Fedora Extras "kmod" kernel
> > packaging (see the packaging list archives). Essentially, this and
> > other scripts are part of a system for allowing compatible kernel
> > modules to continue to work after a kernel update occurs -
> > weak-modules is the part that figures out which kernels are compatible
> > with a driver.
> >
> > There's more information (and some out of date packages - look at
> > rawhide for the current latest stuff!) at
> > http://www.kerneldrivers.org/ and I will add documentation for those
> > who were looking for it and could not find any - I'm sorry that's not
> > been done yet.
> >
> > I'm trying to work with the folks at SuSE/Novell to ultimately offer
> > kernel module package authors a standardized packaging process too
> > (over and above what we each do separately) - I'm going to start by
> > introducing a new macro in redhat-rpm-config along the lines agreed
> > between those of us who met up at OLS to discuss driver updates last
> > week. This is not aimed at undermining the work that's been done in
> > Extras - it's just an additional option.
> >
> > If you would like more information, just drop me an email (I'm also
> > jcm(a)redhat.com when wearing my work hat!) or reply to this one.
>  
>  What exactly is the problem being solved here?  I don't see any obvious 
>  link to anything on kerneldrivers.org that describes what this stuff does.
>  
>  I build lirc and ivtv modules each time a new kernel comes out, and I 
>  have absolutely no problems doing that.  Couldn't be any easier.  Piece 
>  of cake. I'm not sure what problem needs solving here.  I just install 
>  the new kernel and kernel-devel packages, run the script that builds a 
>  corresponding lirc and ivtv kernel driver packages, pointing the build 
>  at the right build directory, then install the driver packages and 
>  reboot.  Couldn't be simpler.
>  
>  If someone really wants to make building kernel driver packages easier, 
>  they can simply fix THE HORRIBLE UGLY STINKING CRAP code that 
>  kernel-devel's %post script uses to hardlink the identical kernel source 
>  files.
>  
>  If I set out to do this specific task in the most grossly inefficient 
>  manner possible, there's no way I could do it any better than the 
>  existing mess. The brain-damaged script literally reads the entire 
>  /usr/src/kernels, by forking thousands of mind-numbing child procs.  
>  And, the more kernels you have installed, the slooower and 
>  sloooooooooooooower everything takes.
>  
>  Has anyone ever heard of this thing called an SHA1 hash?
>  
>  Anybody?
>  
>  Hello!!! McFly!!! Anybody home????
>   
Anybody heard of being civil? If you have problems, file bug reports.

Rahul

[Sam Varshavchik]

Rahul writes:

>  Sam Varshavchik wrote:
> > Jon Masters writes:
> > 
> >> Hi folks,
> >>
> >> Several people have been asking what /sbin/weak-modules is/does,
> >> mostly because it started script spewing the other day (sorry about
> >> that!) on kernel updates.
> >>
> >> weak-modules is part of some driver updates work we're doing at Red
> >> Hat and also involves folks working on Fedora Extras "kmod" kernel
> >> packaging (see the packaging list archives). Essentially, this and
> >> other scripts are part of a system for allowing compatible kernel
> >> modules to continue to work after a kernel update occurs -
> >> weak-modules is the part that figures out which kernels are compatible
> >> with a driver.
> >>
> >> There's more information (and some out of date packages - look at
> >> rawhide for the current latest stuff!) at
> >> http://www.kerneldrivers.org/ and I will add documentation for those
> >> who were looking for it and could not find any - I'm sorry that's not
> >> been done yet.
> >>
> >> I'm trying to work with the folks at SuSE/Novell to ultimately offer
> >> kernel module package authors a standardized packaging process too
> >> (over and above what we each do separately) - I'm going to start by
> >> introducing a new macro in redhat-rpm-config along the lines agreed
> >> between those of us who met up at OLS to discuss driver updates last
> >> week. This is not aimed at undermining the work that's been done in
> >> Extras - it's just an additional option.
> >>
> >> If you would like more information, just drop me an email (I'm also
> >> jcm(a)redhat.com when wearing my work hat!) or reply to this one.
> > 
> > What exactly is the problem being solved here?  I don't see any obvious 
> > link to anything on kerneldrivers.org that describes what this stuff does.
> > 
> > I build lirc and ivtv modules each time a new kernel comes out, and I 
> > have absolutely no problems doing that.  Couldn't be any easier.  Piece 
> > of cake. I'm not sure what problem needs solving here.  I just install 
> > the new kernel and kernel-devel packages, run the script that builds a 
> > corresponding lirc and ivtv kernel driver packages, pointing the build 
> > at the right build directory, then install the driver packages and 
> > reboot.  Couldn't be simpler.
> > 
> > If someone really wants to make building kernel driver packages easier, 
> > they can simply fix THE HORRIBLE UGLY STINKING CRAP code that 
> > kernel-devel's %post script uses to hardlink the identical kernel source 
> > files.
> > 
> > If I set out to do this specific task in the most grossly inefficient 
> > manner possible, there's no way I could do it any better than the 
> > existing mess. The brain-damaged script literally reads the entire 
> > /usr/src/kernels, by forking thousands of mind-numbing child procs.  
> > And, the more kernels you have installed, the slooower and 
> > sloooooooooooooower everything takes.
> > 
> > Has anyone ever heard of this thing called an SHA1 hash?
> > 
> > Anybody?
> > 
> > Hello!!! McFly!!! Anybody home????
> > 
>  
>  Anybody heard of being civil? If you have problems, file bug reports. 
Oh, please.  Bugzilla's full of stuff going back years, that obviously 
nobody looked at.

[Rahul Sundaram]

Sam Varshavchik wrote:

>  Oh, please.  Bugzilla's full of stuff going back years, that
> obviously 
>  nobody looked at. 
Can you point out the bug report numbers? We have had discussions about 
kernel-devel hard links before in this list and configuration options 
were added IIRC.

Rahul

[Arjan van de Ven]

On Thu, 2006-08-03 at 02:54 +0100, Jon Masters wrote:
>  Hi folks,
>  
>  Several people have been asking what /sbin/weak-modules is/does,
>  mostly because it started script spewing the other day (sorry about
>  that!) on kernel updates.
>  
>  weak-modules is part of some driver updates work we're doing at Red
>  Hat and also involves folks working on Fedora Extras "kmod" kernel
>  packaging (see the packaging list archives). Essentially, this and
>  other scripts are part of a system for allowing compatible kernel
>  modules to continue to work after a kernel update occurs -
>  weak-modules is the part that figures out which kernels are compatible
>  with a driver. 
so someone is finally working on a real kernel abi check rather than
just the quite bad "check modversions" check which was useless (both too
many false positives and negatives)?
Cool!

[Matthias Saou]

Jon Masters wrote :

>  [...] I'm going to start by
>  introducing a new macro in redhat-rpm-config along the lines agreed
>  between those of us who met up at OLS to discuss driver updates last
>  week. This is not aimed at undermining the work that's been done in
>  Extras - it's just an additional option. 
Seems like weak-modules is already assuming it's installed, when it
doesn't require it :

[...]
  Installing: kernel                 ####################### [11/38]
/sbin/weak-modules: line 226: /usr/lib/rpm/redhat/rpmsort: No such file
or directory

Not sure what the proper fix is, I'll leave it up to you :-)

Matthias

-- 
Clean custom Red Hat Linux rpm packages : http://freshrpms.net/
Fedora Core release 5.91 (FC6 Test2) - Linux kernel 2.6.17-1.2505.fc6
Load : 0.22 0.45 0.44