[tulip] Transmit timeout - perhaps a clue

[Joseph Parmelee]

We are using a Linksys LNE100TX which has been working well in a Pentium
II/440BX production system since February.  We discovered that we could
induce almost immediately the ethernet transmit timeout error that has been
widely reported when running either a 2.2.12 or 2.2.16 kernel by connecting
/dev/ttyS0 to the RS232C back-port of our terminal server.

The timeout errors appeared before any application using /dev/ttyS0 was even
started.  The terminal server sends data continuously, so I presume the
RS232C receive interrupts (IRQ 4) are somehow inducing this.  The NIC is on
IRQ 10.  Perhaps the fact that IRQ 4 is handled on the first level interrupt
controller is important, but so are system clock (IRQ 0) and keyboard (IRQ
1), which don't cause the problem.  Neither do any of the other interrupts
on the cascaded interrupt controller; Cyclades T1 serial interace on IRQ 11,
IDE controllers on IRQ 14 and IRQ 15.  We did not try to induce the errors
with the floppy disk controller (IRQ 6).  No other IRQ on the first level
controller above 2 is in use.

Is there a silly interrupt handling bug involving the interrupt cascading
lurking in the tulip driver?

Sorry we don't have time to chase this right now, but will be happy to
provide additional info.

Our 2.2.12 kernel was built with tulip.c:v0.91g 7/16/99 downloaded from the
Linksys website; our 2.2.16 kernel uses the stock tulip.c.

Regards,

Joe Parmelee
Wild Bear Systems
jparmele@wildbear.com