[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: Thecus N2100: where's the bottleneck?


Martin, thanks for your effort. Just a question: Have the patched been
updated from intel side? 
I'm asking because yesterday -- coincidential on the same day as
Nathan ;-) -- I played with my Thecus and the DMA enabled kernel on your
repository. Along the kernel, I also install the linux-libc-dev, and the

(BTW: How do you build your kernels? With dpkg-cross? Are there
somewhere instructions/hints/pointers on the net?)

/var/log/aptitude says I did this:
[DOWNGRADE] linux-libc-dev 2.6.32-23 -> 2.6.26-24dma1
[INSTALL] gcc-4.1
[INSTALL] linux-headers-2.6.26-2-all-armel
[INSTALL] linux-headers-2.6.26-2-common
[INSTALL] linux-headers-2.6.26-2-iop32x
[INSTALL] linux-headers-2.6.26-2-orion5x
[INSTALL] linux-image-2.6.26-2-iop32x
[INSTALL] linux-kbuild-2.6.26

Nowever, after booting I got a strange behaviour. At first everything
seemed to work (however, my hdparm tests did not show significant
improvement. Buffered reads where in the low 30MB/s, alost the same with
the 2.6.32 currently in sid. ) , but when I tried to unlock the crypted
partition, the kernel complained about the cipher:

Oct 20 15:55:35 thecus kernel: [42950180.790000] device-mapper: table:
253:0: crypt: Block size of ESSIV cipher does not match IV size of block

(A word on my setup: The cryped partition is a reiser on a raid1
container. Crypto is using dm and luKS. For crypto acceleration, I 
have a hifn7655 card in the mini pci slot, but I am not sure if the
kernel is really using it...)

For testing I usually do not flash the kernel, but load it through 
redboot. Before that I scp'ed the complete /boot to the tftp server.
However, it seems that the installation of the dma enabled
kernel or one of the other packages triggered an update to the initramfs
of the current running kernel. (At least the timestamp of the boot
suggests that -- I am not really sure what happened) 

So I rebooted to x-check the current kernel for the crypto, but the
flashed kernel refused to boot -- it SEGV'ed early in the boot, the
backtrace told in strcmp... Unfortunatly I did not save the serial log.
The only thing I can recover is that it was before mounting the real
At that time my focus shifted to get the thecus running again, so I did
not retry without the crypto accelerator. Eventually a old image of 
a 2.6.30 booted (and then I had no time left to retry.) 

Any thoughts? 


On Thu, 2010-10-21 at 10:42 +0100, Martin Michlmayr wrote:
> * Nathan <debian-arm@thecoads.com> [2010-10-21 20:20]:
> > > Linux debian 2.6.26-2-iop32x #1 Thu Sep 16 18:42:37 UTC 2010 armv5tel
> > > GNU/Linux
> > 
> > I downloaded the .deb files from Martin's repository and installed them
> > that way.  On my 2100, uname gives the following output:
> > 
> > Linux debian 2.6.26-2-iop32x #1 Sat Jun 26 17:51:30 UTC 2010 armv5tel
> > GNU/Linux
> The DMA kernels are out of date.  I'm updating them right now.
> -- 
> Martin Michlmayr
> http://www.cyrius.com/

Reply to: