Re: kern/42314: ipgphy support

[Frank Wille <frank%phoenix.owl.de@localhost>]

On Wed, 19 Aug 2015 07:54:54 +0200
Felix Deichmann <m4j0rd0m0%gmail.com@localhost> wrote:

> 2015-08-18 20:30 GMT+02:00 Frank Wille <frank%phoenix.owl.de@localhost>:
> > What's wrong with stge(4)?
> 
> Compare our driver to the FreeBSD one, derived from us. It seems
> FreeBSD's heavily improved in the meantime, but the changes were never
> re-imported.

Ok.


> I try to summarise: We have no HW VLAN tagging, no jumbo frames (?),
> the discussed PHY/link issues, ST-1023 (and compatible IP1000A, as can
> be also seen even in your ifconfig output) is in promiscuous mode all
> the time as work-around, I remember a bug/panics related to
> stge_intr() reported on port-sandpoint (fixed?), possibly sub-optimal
> performance, and an indicative amount of "XXX" in the driver...

I don't know about other architectures, but I didn't really care about
jumbo frames or sub-optimal performance, because the 200MHz 8241 is so
weak that it couldn't even bring a 100MBit NIC to its limits. :)

I don't remember any bugs or panics (although that doesn't mean much).
In my tests the stge(4) driver seemed stable.


> As said, I will receive a DSM-G600 (IP1000A), too, and a D-Link
> DGE-550T (DL-2000) as different stge(4) representatives and would be
> willing to help or work on improving the situation.

Any help is welcome, of course. I can test and commit your modifications,
when you make any progress.


> If a more recent phy driver doesn't fix it, I also found special
> handling for IP1000 in FreeBSD's stge_miibus_readreg():
> 
> if (reg == STGE_PhyCtrl) {
>     /* XXX allow ip1000phy read STGE_PhyCtrl register. */
>     STGE_MII_LOCK(sc);
>     error = CSR_READ_1(sc, STGE_PhyCtrl);
>     STGE_MII_UNLOCK(sc);
>     return (error);
> }

The LOCK/UNLOCK just establishes a mutex. I doubt it will change much.


-- 
Frank Wille