On Mar 19,  7:40pm, Rainer Schubert wrote:
> Subject: sa-110-4958 kernel problems
> Hi,
>
> there are some problems with the SA110-4958 kernel, I wish to tell you
> about and ask for help.
>
> 1st Kernel does not reboot when told to do so but instead hangs the machine
>     after doing it's shutdown hooks. Only help is to switch off the machine.

Yup. Mark has found the problem and fixed it. I don't know whether it has made
it back into the released kernels yet - try the compiled snapshot kernels. This
is actually RiscOS crashing because RiscBSD didn't leave the machine in quite
the state it expected.

> 2nd the ps command does not work as expected. The only option working with
>     the ps command is -L.
>     In any other case ps returns an error according to this 'procsize error
>     13640 648 chunks'.  Sometime there is a 'floating point error' as well.

Have you installed the patch sets? From time to time the size of the proc
structure changes. Your current kernel doesn't match your version of ps (or it
doesn't match the version of the kernel installed as /netbsd).

> 3rd netstat sometimes crashes with a read error.

Same problem as 2.

> 4th printing from RiscBSD with any kernel I have tested up to now is still
>     a problem as any of the kernels tested states the printer (a HP DeskJet
>     and a HP LaserJet) as being offline. /dev/lpt0 shows a file length of
>     0 (zero) bytes while /dev/lpa0 is 128 bytes long.
>     Trying a 'cat filename > /dev/lpa0 results in one line of text with all
>     the rest being printed on the right side of the paper one printed over
>     the other one in a manner looking much like chinese characters.

Hmm, not seen this one. I'll see if I can reproduce it.

> Good points with kernel 4958 are: The serial port works much more reliable
> e.g. transmitting say /netbsd in binary mode works fine without more than
> one or two 'silo overflows' at about 10000 cps with the serial port set
> to 115200 bps and the second machine being an oldish A410/1 with a DualSerial
> card and the TCP/IP client on the other side is !TCPIP (ArcKA9Q).
>
> Reversing the process is much slower with kernel 4958 e.g. a get in binary
> mode performs at less than 3000 cps with lots of 'silo overflows' and many
> repeated frames due to this fact.

This is due to the asymmetry between disk reading and writing and its affect on
interrupt latency.

> With FreeNet on my Risc PC as the receiving ftp client things work better
> in both directions.

You would expect this because RiscOS does not have the overheads of preemptive
multitasking (which increases the interrupt latency by a huge amount).

Cheers

Rob