Are there any known problems with using a gameport under x86_64? It seems as though reading from the joystick is somehow interfering with the serial port draining, causing it to lose data. If I read from the joystick - the simplest way is "cat /dev/input/js0" - the inbound network traffic will cease almost instantly. This can be seen by watching kppp statistics. After I stop reading from js0,
the network usually starts working again.

/dev/input/event2 has the same problem (it uses the same hardware on my system).

The network and joystick work fine otherwise, and this problem doesn't show up in a different OS (Windows 98 SE).

I have Athlon 64 3400+ and MSI K8T NE MS-6702 motherboard running slamd64 10.1 linux, Sound Blaster Live! Value (emu101k) with a game port, a Gravis GamePad Pro connected to this game port, and an external serial modem (USR 56 K Fax). My kernel is version 2.6.11.10.

I've tried a lot of things to cure this without success:
-Installed Vojtek Pavilik's recent patches from here http://www.ucw.cz/~vojtech/input-2005-03-07/
-Changed modem init string, ppp compression settings
-Changed serial parameters with setserial
-Recompiled kernel with preempt on/off and also experimental pnp acpi
configuration
-Changed PCI latency settings with setpci
-Looked for IRQ conflicts (nothing suspicious in /proc/interrupts)
CPU0
0: 845022 IO-APIC-edge timer
1: 5973 IO-APIC-edge i8042
4: 23173 IO-APIC-edge serial
8: 0 IO-APIC-edge rtc
9: 0 IO-APIC-level acpi
12: 26555 IO-APIC-edge i8042
14: 19332 IO-APIC-edge ide0
15: 3855 IO-APIC-edge ide1
177: 0 IO-APIC-level uhci_hcd, uhci_hcd, uhci_hcd, uhci_hcd,
ehci_hcd
185: 645 IO-APIC-level EMU10K1
193: 46695 IO-APIC-level nvidia
NMI: 83
LOC: 844586
ERR: 0
MIS: 0
-Looked at gameport timings (seem to be OK, both before and after gameport.c x86_64-specific patch).
$ dmesg | grep game
gameport: EMU10K1 is pci0000:00:07.1/gameport0, io 0xec00, speed 1145kHz
input: Gravis GamePad Pro on pci0000:00:07.1/gameport0

Any ideas?