1. 3c905 100Mbps: RX-ERR RX-DRP RX-OVR?
Hi,
I have really strange effects using the 3c905 with the new 3c59x.c driver.
http://cesdis.gsfc.nasa.gov/linux/drivers/3c59x.c
line 725: vp->full_bus_master_tx = 0;
The cards didn't work at all without this patch.
At boot time, the Pacing bug is detected so busmastering isn't enabled.
I am administrating an 100Mbps net with several Dual PPro 200MHz (2.0.30)
and some SGI O2. At first, the only problem was that a copy O2->PPro caused
many RX-OVR on the PPro and bad performance (250KBps), while the other
direction worked fine (5,6MBps).
Then I tested the tcp connection between the PPros (with ttcp). Some
connections worked fine (5,6MBps), others also caused some RX-OVR (visible
by 'netstat -i') and bad performance (in this case: 2,5MBps). The strange
thing is that all the PPros are equally equipped (hardware and software).
I tried a lot of things (reconfigured the kernel of the O2 with small buffers
etc, changed some of the network options on the PPro, eg. 'optimize as router
not host', 'PC/TCP compatibility mode', 'Disable Path MTU Discovery', disabled
'Allow large windows') but nothing changed this effect.
The only workaround by now is to reduce the rsize and wsize of mounts (O2->PPro
is 2560 now). But this doesn't effect ftp transfers.
What is responsible for these strange RX-ERR RX-DRP RX-OVR effects? Is there
still a bug in the 3c900.c:v0.40 4/16/97 driver?
Thanks for any hint.
T.B.
2. Torvalds comes out punching.
3. No way to flush rx buffer in serial driver?
4. date difference
5. how to get per-socket stats on udp rx buffer overflow?
6. Kerbos
7. Suspended -- no Rx buffers using FA310Tx
8. xfig 3.2.4-beta3 released
9. eth0: can't fill rx buffer (force 1)!
10. rx buffer overflow attack?
11. eth0: cart reports no RX buffers
12. eth0: can't fill rx buffer (force 1)!
by