debian "experimental" kernels

Is anyone using kernels from Debian experimental? I needed something newer than 3.2.x (testing ceph but I'm getting kbyte/second write performance!) so I tried a 3.8.x kernel from Debian experimental but it just reboots with no indication of why (using serial console via bmc) James

James Harper <james.harper@bendigoit.com.au> wrote:
Is anyone using kernels from Debian experimental?
Yes, linux-image-3.8-trunk-amd64 with no problems so far as I am aware. It seems to have fixed the video issues I was having with 3.2.x.
That's what I'm using and it's not working for me. I'm using xen under it though... maybe there is something in that? What processor are you using? I'm using amd and Debian amd kernels seem extraordinarily buggy with powernow or whatever that module is called. James

James Harper <james.harper@bendigoit.com.au> wrote:
That's what I'm using and it's not working for me. I'm using xen under it though... maybe there is something in that?
What processor are you using?
Intel Xeon, with no virtualization under the 3.8 kernel. It's a desktop workstation, rather old now but running reliably at least for the moment. It was already "used" when I bought it, otherwise the price would have exceeded my budget.

James Harper <james.harper@bendigoit.com.au> wrote:
That's what I'm using and it's not working for me. I'm using xen under it though... maybe there is something in that?
What processor are you using?
Intel Xeon, with no virtualization under the 3.8 kernel. It's a desktop workstation, rather old now but running reliably at least for the moment. It was already "used" when I bought it, otherwise the price would have exceeded my budget.
I just booted 3.8 without xen and it boots fine, and ceph write performance went from 700kbytes/second to 25mbytes/second, so I guess that's a win. Except I can have fast ceph with no xen, or slow ceph with xen :( Rolling my own kernel won't be a waste of time though. James

I should add, for the record, that my video problems seem to have arisen after Debian back-ported Nouveau/DRM code from 3.4 into their 3.2 kernel. I'm not in a position to follow up with bug reports at the moment as I can't do the testing currently and I need a working system. I haven't noticed any difficulties under 3.8 yet.

I should add, for the record, that my video problems seem to have arisen after Debian back-ported Nouveau/DRM code from 3.4 into their 3.2 kernel. I'm not in a position to follow up with bug reports at the moment as I can't do the testing currently and I need a working system.
I haven't noticed any difficulties under 3.8 yet.
I guess it's just the Debian way, but it's a bit sad that they are about to release Wheezy and it's still on a 3.2 series kernel. James

James Harper <james.harper@bendigoit.com.au> wrote:
I guess it's just the Debian way, but it's a bit sad that they are about to release Wheezy and it's still on a 3.2 series kernel.
I agree, but then I've never been the kind of person who runs Debian Stable. If I had servers to take care of or other infrastructure that needed that extra degree of reliability, the situation would be different, of course, and I think it's important that there be distributions which make solid releases, however long it takes to fix the release-critical bugs. I think I would be better suited to a distribution that follows a so-called "rolling" model whereby there are no freezes and no formal releases as such. Especially with some of the accessibility-related tools that I use, there are improvements and bug fixes that make it worthwhile to keep up with the latest releases of desktop environments and applications, a trend that won't change soon. Up-stream are, justifiably, not interested in bug reports based on such old code, particularly in times of rapid change. Frankly, waiting for Debian to make their release so that updated packages will move into Unstable is rather frustrating, and it only demonstrates that my needs and priorities don't match Debian's in this respect. That's fine.

On Sat, 20 Apr 2013, Jason White <jason@jasonjgw.net> wrote:
I think I would be better suited to a distribution that follows a so-called "rolling" model whereby there are no freezes and no formal releases as such.
You can use Debian/Testing in that regard.
Frankly, waiting for Debian to make their release so that updated packages will move into Unstable is rather frustrating, and it only demonstrates that my needs and priorities don't match Debian's in this respect. That's fine.
The freeze on Unstable prior to the release of a new version of Debian only covers a small amount of time every couple of years. You're not going to spend that much time waiting for things. When there are significant updates that are delayed from Unstable because of a freeze you can almost always find them in Experimental. -- My Main Blog http://etbe.coker.com.au/ My Documents Blog http://doc.coker.com.au/

On Sat, 20 Apr 2013, Jason White <jason@jasonjgw.net> wrote:
I think I would be better suited to a distribution that follows a so-called "rolling" model whereby there are no freezes and no formal releases as such.
You can use Debian/Testing in that regard.
And for the most part, probably except for a few months after the new Stable is released, Testing tends to be quite stable. I've been using Wheezy for ages without any issues except for AMD processor support. James

James Harper <james.harper@bendigoit.com.au> wrote:
And for the most part, probably except for a few months after the new Stable is released, Testing tends to be quite stable. I've been using Wheezy for ages without any issues except for AMD processor support.
I run either testing or unstable on my machines, with occasional packages from experimental (more of the latter at the moment due to the freeze). Are your AMD CPU problems specific to Debian kernels or do they occur with mainline kernels too? I would expect such issues to be sorted out, given the number of systems with AMD CPUs that would be in use by kernel developers and regular testers. Studies have shown that kernel regressions tend not to last long, once properly reported.

James Harper <james.harper@bendigoit.com.au> wrote:
And for the most part, probably except for a few months after the new Stable is released, Testing tends to be quite stable. I've been using Wheezy for ages without any issues except for AMD processor support.
I run either testing or unstable on my machines, with occasional packages from experimental (more of the latter at the moment due to the freeze).
Are your AMD CPU problems specific to Debian kernels or do they occur with mainline kernels too? I would expect such issues to be sorted out, given the number of systems with AMD CPUs that would be in use by kernel developers and regular testers.
Studies have shown that kernel regressions tend not to last long, once properly reported.
The problem in question is something to do with powernow (not sure if I'm remembering correctly... something to do with power management) spewing kernel log messages out at a rate that makes the system unusable. It may be a compatibility issue between that module and xen. Easy enough to blacklist the module but still a pain. Bug reports existed last time I checked so maybe it's been fixed. I haven't seen it yet on 3.8. James

James Harper <james.harper@bendigoit.com.au> wrote:
The problem in question is something to do with powernow (not sure if I'm remembering correctly... something to do with power management) spewing kernel log messages out at a rate that makes the system unusable. It may be a compatibility issue between that module and xen.
As an aside, Xen just officially became a Linux Foundation project, as reported on LWN. It isn't clear how this will affect the relationship between Xen development and the mainline kernel.

Russell Coker <russell@coker.com.au> writes:
On Sat, 20 Apr 2013, Jason White <jason@jasonjgw.net> wrote:
I think I would be better suited to a distribution that follows a so-called "rolling" model whereby there are no freezes and no formal releases as such.
You can use Debian/Testing in that regard.
14:56 <themill> (within one day of squeeze's release 15% of source packages were different in squeeze; pretty damned amazing when you consider that they are (by definition) identical at release time) While it only happens once every (on average) two years, that's a pretty cam to roll.
The freeze on Unstable prior to the release of a new version of Debian only covers a small amount of time every couple of years. You're not going to spend that much time waiting for things.
14:42 <themill> !slushy 14:42 <dpkg> When a <testing> release becomes frozen, <unstable> tends to partially freeze as well. This is because developers are reluctant to upload radically new software to unstable, in case the frozen software in testing needs minor updates and to fix release critical bugs which keep testing from becoming <stable>.

On Fri, 19 Apr 2013, James Harper <james.harper@bendigoit.com.au> wrote:
Yes, linux-image-3.8-trunk-amd64 with no problems so far as I am aware. It seems to have fixed the video issues I was having with 3.2.x.
I'm running Linux 3.8 kernels on some of my Debian systems for better BTRFS support. It works well. I've had problems with 3.6 kernels having USB stop responding which makes systems that have USB keyboard and mouse (IE all modern systems) almost unusable. But 3.8 generally seems OK in that regard.
What processor are you using? I'm using amd and Debian amd kernels seem extraordinarily buggy with powernow or whatever that module is called.
Note that there aren't AMD specific kernels. The AMD64 architecture is also used by all recent Intel CPUs. -- My Main Blog http://etbe.coker.com.au/ My Documents Blog http://doc.coker.com.au/

Russell Coker <russell@coker.com.au> writes:
On Fri, 19 Apr 2013, James Harper <james.harper@bendigoit.com.au> wrote:
Yes, linux-image-3.8-trunk-amd64 with no problems so far as I am aware. It seems to have fixed the video issues I was having with 3.2.x.
I'm running Linux 3.8 kernels on some of my Debian systems for better BTRFS support.
#btrfs suggests run your normal kernel and build btrfs as a DKMS, presumably off the btrfs repo rather than mainline. I never cared enough myself, but I'm not doing anything interesting with btrfs.

James Harper writes:
What processor are you using? I'm using amd and Debian amd kernels seem extraordinarily buggy with powernow or whatever that module is called.
So blacklist it?
I have now. it's a bit of a pain the first time it happens though when the server is remote. And simply putting "blacklist <modulename>" didn't seem to work. I had to alias it to an invalid name. James

James Harper wrote:
James Harper writes:
What processor are you using? I'm using amd and Debian amd kernels seem extraordinarily buggy with powernow or whatever that module is called.
So blacklist it?
I have now. it's a bit of a pain the first time it happens though when the server is remote.
And simply putting "blacklist <modulename>" didn't seem to work. I had to alias it to an invalid name.
IIRC "blacklist foo" just prevents it autoloading; to prevent anything explicitly loading it I was using "install foo /bin/false" IIRC.

On Fri, Apr 19, 2013 at 10:14:46AM +0000, James Harper wrote:
Is anyone using kernels from Debian experimental? I needed something newer than 3.2.x (testing ceph but I'm getting kbyte/second write performance!) so I tried a 3.8.x kernel from Debian experimental but it just reboots with no indication of why (using serial console via bmc)
i've been using debian experimental kernels for some time - pretty much since the wheezy freeze meant newer kernels stayed in experimental rather than sid. except for some recent problems with USB (after ehci support got split into two modules, ehci-pci and ehci-hcd), they've been working fine. my main machines are all AMD Phenom II x6 1090T CPUs. I used cpufreq-utils, with the on-demand governor (spending about 95% of the time at 800Mhz). according to cpufreq-info, they're all using the acpi-cpufreq driver. I don't use xen, but i do use kvm. have you tried xen 4.1.4-2 from sid or xen 4.2.0-1 from experimental? i'm no expert on xen but it seems a reasonable guess that you'd need the latest xen with the latest kernel. craig -- craig sanders <cas@taz.net.au>

have you tried xen 4.1.4-2 from sid or xen 4.2.0-1 from experimental? i'm no expert on xen but it seems a reasonable guess that you'd need the latest xen with the latest kernel.
Yep that turned out to be it. Forgot to follow up here! After I turned on the BMC console for xen I could see "can't load elf image" or something like that, and upgrading xen fixed it. Thanks James
participants (6)
-
Craig Sanders
-
James Harper
-
Jason White
-
Russell Coker
-
Trent W. Buck
-
trentbuck@gmail.com