Charles Lockhart
2003-11-18 03:56:43 UTC
While trying to generate a RH or FC1 kernel with the rml preemptive
kernel patch applied (and failing), I was told that the patch might
already be ported into one or both of those kernels (if so, most likely
this would be a backport from the 2.5/2.6 kernel). But after looking
through the kernel source, while there are places that kernel preemption
and rml are mentioned (mostly in comments), I don't see a lot of the
functionality that would be needed for kernel preemption via the rml
patch. Can someon confirm or deny?
Motivation behind this is that I get a substantial system response (ie.
my system (which is defined for a specific set of tasks) meets
performance requirements) with the patch, but not without it (ie. with
the RH/FC1 cores installed). Unfortunately, the system randomly freezes
(sometimes after hourse, sometimes after a day or two) with the vanilla
kernel, so it doesn't meet requirements for stability.
I've considered just waiting until the 2.6 kernel is available, but I
have some very specific hardware in the machine, and I'd have to port
the drivers myself, which could become quite the time sink (the company
that makes the hardware said that they're planning to support the 2.6
kernel sometime around the end of 2004).
Also, the primary obstacle to patching the FC1/RH kernels has been (I
think) the O1 scheduler (or at least that's where I get the greatest
number of failed hunks). Has anyone tried to port this in? There's a
O1/preemptive kernel patch, but it's for the 2.4.18 kernel. Has anyone
tried this?
Would it be possible to port over to the FC1 core kernel? I was able to
patch the vanilla 2.4.20 kernel with the preemptive kernel patch, and
then step through pretty much all of the RH 2.4.20-8 patches and fix
failures as I went, pretty much right up to the end when it was time to
add the O1 back port and others, and at that point hit a brick wall.
The system is a compaq/hp DL360 G3, dual Xeon P4's, 1GB ram. I've tried
both RH9 and FC1, same results.
Thank you for your time,
-Charles
kernel patch applied (and failing), I was told that the patch might
already be ported into one or both of those kernels (if so, most likely
this would be a backport from the 2.5/2.6 kernel). But after looking
through the kernel source, while there are places that kernel preemption
and rml are mentioned (mostly in comments), I don't see a lot of the
functionality that would be needed for kernel preemption via the rml
patch. Can someon confirm or deny?
Motivation behind this is that I get a substantial system response (ie.
my system (which is defined for a specific set of tasks) meets
performance requirements) with the patch, but not without it (ie. with
the RH/FC1 cores installed). Unfortunately, the system randomly freezes
(sometimes after hourse, sometimes after a day or two) with the vanilla
kernel, so it doesn't meet requirements for stability.
I've considered just waiting until the 2.6 kernel is available, but I
have some very specific hardware in the machine, and I'd have to port
the drivers myself, which could become quite the time sink (the company
that makes the hardware said that they're planning to support the 2.6
kernel sometime around the end of 2004).
Also, the primary obstacle to patching the FC1/RH kernels has been (I
think) the O1 scheduler (or at least that's where I get the greatest
number of failed hunks). Has anyone tried to port this in? There's a
O1/preemptive kernel patch, but it's for the 2.4.18 kernel. Has anyone
tried this?
Would it be possible to port over to the FC1 core kernel? I was able to
patch the vanilla 2.4.20 kernel with the preemptive kernel patch, and
then step through pretty much all of the RH 2.4.20-8 patches and fix
failures as I went, pretty much right up to the end when it was time to
add the O1 back port and others, and at that point hit a brick wall.
The system is a compaq/hp DL360 G3, dual Xeon P4's, 1GB ram. I've tried
both RH9 and FC1, same results.
Thank you for your time,
-Charles