Re: Synology PPCBoot SK98 network bug fix

[Julian Coleman <jdc%coris.org.uk@localhost>]

> That looks ok.
> 
> First I was puzzled, because I checked the if_sk.c source again, and saw
> that it is not using RXOK, but checks XM_RXSTAT_ERRFRAME.

I think the problem is the number of other packets that we receive (this
network has switches with IGMP and bridging).  If we only have 2 receive
descriptors, we can receive two multicast or broadcast packets and return
these.  That means that we can miss out on the replies to our requests.

> So I thought for a moment that we have to use the XMAC II status definitions
> instead, while I was using RXOK from the Yukon status word, which in the
> kernel is only used by the msk(4) Yukon2 driver.

I think that using RXOK is fine.  All the packets that I see have RXOK set
(as well as often RXSTAT_BROADCAST or RXSTAT_MULTICAST.

I increased the number of receive buffers to 8 and that usually gets
further - it got part way through loading the kernel a couple of times,
but it has never managed to load the whole kernel without failing at some
point.  I also tried ignoring packets with RXSTAT_BROADCAST or
RXSTAT_MULTICAST set, and immediately checking for another packet.  In that
case, it gets further (bootp, arp, nfs)  but still hasn't succeeded to load
a complete kernel.

> Hmm. But can you confirm that be packets are sent by your DHCP server?

I am running `tcpdump` on the DHCP/TFTP/NFS server, so I can see the
requests from the NAS, and also the replies that the server sends.  I've
added some extra debug to print the packet headers to skg_send() and to
skg_recv(), and that correlates to the packets I see in `tcpdump`.

> I hope the solution is easier, as it already works with Oscar's 407e. :)

As it doesn't work even with 8 receive descriptors, I think that programming
the receive filters might help.  I will try that next.

> There still might be an issue with altboots sk driver, but at least it
> doesn't interfere with the kernels sk driver anymore.

Yes - the kernel driver seems fine now.  As I mentioned, I used it with
root on NFS, and that should have stressed it enough ;-)

> Make sure you load the snapshot install kernel too, because 5.1 halts
> after printing kernel version info.

I had just grabbed a snapshot of HEAD, so I didn't realise this!

Thanks,

J

-- 
  My other computer also runs NetBSD    /        Sailing at Newbiggin
        http://www.netbsd.org/        /   http://www.newbigginsailingclub.org/