On 2006/06/22 19:44, Thomas Bvrnert wrote: > ok i tried it and it works, ... strange ... > the throughout is a little bit higher, without the > patch it was 2,4 MB/s and with the patch 2,7 MB/s. > (scp) and i've 1500 total interrupts more. > > But i think the problem is on another place in the code.
Glancing through the datasheet for dp83815 one thing that stuck out was a mention of disabling rx in power-saving, and I know that geode systems do a bunch of power-saving that can't be disabled (VSA loaded by BIOS runs above everything else on the cpu and can control this sort of thing), which has caused problems in other areas before (do you see that 'TSC disabled' right at the top of dmesg? played hell with altq until that was done...) I guess if anyone has both a sis(4) and ral(4) they can try in a non-soekris box, if they don't see the same freeze-ups remedied by 'ifconfig sisN up', that lends support for this hypothesis (or, if they do see the same problem, blows it out the water)... The oldest version of Donald Becker's linux driver for the chip which I could find does this same thing (bottom of netdev_rx in natsemi.c, "Restart Rx engine if stopped.") so unfortunately there's no time when it was added with additional details about why it was done, but it does seem to be known (at least by some people) to be necessary with the chip.
