Heads up on an forthcoming rawhide kernel change...

[Jarod Wilson]

So we had a bug filed a while ago, requesting the ability to install the same 
kernel version of different arches on the same system with the same /boot 
partition:

https://bugzilla.redhat.com/show_bug.cgi?id=197065

To do that, we've made some changes that should be transparent to most users, 
save those that are building external kernel modules. As of 
kernel-2.6.25-0.154.rc6.git7.fc9, we now include the target cpu in uname -r 
output, like so:

# uname -r
2.6.25-0.154.rc6.git7.fc9.x86_64

And we've also renamed files in /boot:

# ls /boot/*x86_64*
/boot/config-2.6.25-0.154.rc6.git7.fc9.x86_64
/boot/initrd-2.6.25-0.154.rc6.git7.fc9.x86_64.img
/boot/System.map-2.6.25-0.154.rc6.git7.fc9.x86_64
/boot/vmlinuz-2.6.25-0.154.rc6.git7.fc9.x86_64

# ls /boot/*i686*
/boot/config-2.6.25-0.154.rc6.git7.fc9.i686
/boot/initrd-2.6.25-0.154.rc6.git7.fc9.i686.img
/boot/System.map-2.6.25-0.154.rc6.git7.fc9.i686
/boot/vmlinuz-2.6.25-0.154.rc6.git7.fc9.i686

The uname -r change also means /lib/modules/ looks slightly different:

# ls /lib/modules/
2.6.25-0.154.rc6.git7.fc9.i686
2.6.25-0.154.rc6.git7.fc9.x86_64

And so do the contents of the kernel-devel packages:

# ls /usr/src/kernels/
2.6.25-0.154.rc6.git7.fc9.i686
2.6.25-0.154.rc6.git7.fc9.x86_64

This last one is probably the one with the most visible impact, as any 
external module build script that was looking 
for /usr/src/kernels/$(uname -r)-$(uname -m) should now be looking 
for /usr/src/kernels/$(uname -r) only.

Sorry for the disruption, and it looks a touch ugly having arch in uname -r, 
but hey, everyone else[*] is doing it. And now you can install both arch 
kernels to the same /boot.

Yell loudly if we broke something besides external kernel module building, 
which should be trivial for your 3rd-party module provider to fix.

[*] kylem says ubuntu/debian already do this, and possibly suse as well...

-- 
Jarod Wilson
jwilson(a)redhat.com

[Jakub 'Livio' Rusinek]

Dnia 25-03-2008, wto o godzinie 17:02 -0400, Jarod Wilson pisze:
>  2.6.25-0.154.rc6.git7.fc9.x86_64 
I thought new kernel merges x86 and x86_64 back...

-- 
Jakub 'Livio' Rusinek
http://liviopl.jogger.pl/

[Rahul Sundaram]

Jakub 'Livio' Rusinek wrote:
>  Dnia 25-03-2008, wto o godzinie 17:02 -0400, Jarod Wilson pisze:
> > 2.6.25-0.154.rc6.git7.fc9.x86_64
>  
>  I thought new kernel merges x86 and x86_64 back... 
You are probably talking about the source level merge. It is still 
separate binaries for different architectures.

Rahul

[Jarod Wilson]

On Tuesday 25 March 2008 05:12:17 pm Rahul Sundaram wrote:
>  Jakub 'Livio' Rusinek wrote:
>  > Dnia 25-03-2008, wto o godzinie 17:02 -0400, Jarod Wilson pisze:
>  >> 2.6.25-0.154.rc6.git7.fc9.x86_64
>  >
>  > I thought new kernel merges x86 and x86_64 back...
>
>  You are probably talking about the source level merge. It is still
>  separate binaries for different architectures. 
What Rahul said.

-- 
Jarod Wilson
jwilson(a)redhat.com

[Don Dutile]

is there still an i686-PAE variant, and if so, is that distinguished in 
the kernel naming below ?


Jarod Wilson wrote:
>  So we had a bug filed a while ago, requesting the ability to install
> the same 
>  kernel version of different arches on the same system with the same /boot 
>  partition:
>  
>  https://bugzilla.redhat.com/show_bug.cgi?id=197065
>  
>  To do that, we've made some changes that should be transparent to most users, 
>  save those that are building external kernel modules. As of 
>  kernel-2.6.25-0.154.rc6.git7.fc9, we now include the target cpu in uname -r 
>  output, like so:
>  
>  # uname -r
>  2.6.25-0.154.rc6.git7.fc9.x86_64
>  
>  And we've also renamed files in /boot:
>  
>  # ls /boot/*x86_64*
>  /boot/config-2.6.25-0.154.rc6.git7.fc9.x86_64
>  /boot/initrd-2.6.25-0.154.rc6.git7.fc9.x86_64.img
>  /boot/System.map-2.6.25-0.154.rc6.git7.fc9.x86_64
>  /boot/vmlinuz-2.6.25-0.154.rc6.git7.fc9.x86_64
>  
>  # ls /boot/*i686*
>  /boot/config-2.6.25-0.154.rc6.git7.fc9.i686
>  /boot/initrd-2.6.25-0.154.rc6.git7.fc9.i686.img
>  /boot/System.map-2.6.25-0.154.rc6.git7.fc9.i686
>  /boot/vmlinuz-2.6.25-0.154.rc6.git7.fc9.i686
>  
>  The uname -r change also means /lib/modules/ looks slightly different:
>  
>  # ls /lib/modules/
>  2.6.25-0.154.rc6.git7.fc9.i686
>  2.6.25-0.154.rc6.git7.fc9.x86_64
>  
>  And so do the contents of the kernel-devel packages:
>  
>  # ls /usr/src/kernels/
>  2.6.25-0.154.rc6.git7.fc9.i686
>  2.6.25-0.154.rc6.git7.fc9.x86_64
>  
>  This last one is probably the one with the most visible impact, as any 
>  external module build script that was looking 
>  for /usr/src/kernels/$(uname -r)-$(uname -m) should now be looking 
>  for /usr/src/kernels/$(uname -r) only.
>  
>  Sorry for the disruption, and it looks a touch ugly having arch in uname -r, 
>  but hey, everyone else[*] is doing it. And now you can install both arch 
>  kernels to the same /boot.
>  
>  Yell loudly if we broke something besides external kernel module building, 
>  which should be trivial for your 3rd-party module provider to fix.
>  
>  [*] kylem says ubuntu/debian already do this, and possibly suse as well...
>

[Jarod Wilson]

On Tuesday 25 March 2008 05:37:27 pm Don Dutile wrote:
>  is there still an i686-PAE variant, and if so, is that distinguished
> in
>  the kernel naming below ? 
Yep, there's still an i686 PAE, and it looks like so:

# uname -r
2.6.25-0.154.rc6.git7.fc9.i686PAE

/boot/config-2.6.25-0.154.rc6.git7.fc9.i686PAE
/boot/initrd-2.6.25-0.154.rc6.git7.fc9.i686PAE.img
/boot/System.map-2.6.25-0.154.rc6.git7.fc9.i686PAE
/boot/vmlinuz-2.6.25-0.154.rc6.git7.fc9.i686PAE

/lib/modules/2.6.25-0.154.rc6.git7.fc9.i686PAE

/usr/src/kernels/2.6.25-0.154.rc6.git7.fc9.i686PAE
/usr/src/kernels/2.6.25-0.154.rc6.git7.fc9.i686-PAE
(Yes, both, there's a symlink created for historical reasons I'm not actually 
aware of)...

So basically, everything just keys off of what uname -r is now, where before, 
we were uname -r everywhere except the bits in /usr/src/kernels, which 
were '$(uname -r)-$(uname -m)'.

>  Jarod Wilson wrote:
>  > So we had a bug filed a while ago, requesting the ability to install the
>  > same kernel version of different arches on the same system with the same
>  > /boot partition:
>  >
>  > https://bugzilla.redhat.com/show_bug.cgi?id=197065
>  >
>  > To do that, we've made some changes that should be transparent to most
>  > users, save those that are building external kernel modules. 
[...]


-- 
Jarod Wilson
jwilson(a)redhat.com