>> ping times are now around 16 ms, which is about what I expect.
>
>Most excellent.  Glad I could help.

Thank you.  By the way... that's 160 ms, not 16.  16 would be pretty
impressive on a PPP dialup. ;->

>> P.S. I suspect that this may be similar to my problem with the "lost
>> interrupts" in the Cyclades serial driver, but I don't understand the
>> code enough to play with it....
>
>On reflection, I think you may well be right.  Any interrupt (including
>soft "interrupts" like setsoftnet()) that occurs while do_pending_int()
>is running will get lost, as I read it, unless that interrupt was
>blocked when do_pending_int() got called - which usually won't be the
>case.  Or, of course, unless the interrupting hardware persists in
>banging on the CPU's interrupt line until it gets acknowledged - does
>your Cyclades hardware do that?

I have no clue.  All I know is that my interrupt handler seems to
be doing the "right thing", yet the interrupt status seems to
come on in the card without invoking my handler.  I've been able to
reduce the frequency of this essentially by reducing the time between
the last "poll" of the status reg and the exit of the handler, but
it still happens.

Shall I submit appropriate send-pr's on this stuff?  The do_pending_int()
thing could theoretically cause various subtle troubles...

-dgl-