> Using the updated rtl8139 in 2.3.18ac5 turned out to be not really a
> pleasure. Actually, the card reliably hangs in rtl8129_open(). And it
> hangs the whole machine, "hang" meaning I can't even use the ATX power
> switch to turn it off. Don't even think about magic syskeys and the like.
>
> The cause seems to be that with the switch to using PCI mem rather than
> i/o ports, the driver runs into a load of nasty timing issues. I was
> finally able to cure the hangs by forcing each write transaction to
> complete before the next one is issued. This probably kills the whole
> point of using PCI mem in the first place, so the forced flushes should be
> much finer grained but, hey, it's a quick fix and it's a whole lot better
> than hanging the machine. (The other obvious fix of course being to switch
> back to i/o ports which works fine as well.)
I spent two hours on the phone Friday tracking this down, but I'm unable to
reproduce the problem myself. It apparently only happens on (only some?)
PentiumII systems, not on my Pentium, P-Pro, K6 or 486 systems. The
work-around, as you discovered, is using -DUSE_IO_OPS, which I've now made
the default.
I still don't understant why this would hang the machine, as there is no
unlimited for/while loop.
BTW, normally bug reports should be sent to the linux-realtek mailing list.
Donald Becker
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/