Re: mue(4) patch

[Rin Okuyama <rokuyama%rk.phys.keio.ac.jp@localhost>]

On 2018/12/04 15:28, Harold Gutch wrote:
> On Tue, Dec 04, 2018 at 10:58:34AM +0900, Rin Okuyama wrote:
> > Does anyone know a convenient way to generate packets with wrong
> > L3/L4 checksum fields?
>
> If Python is OK, perhaps pkgsrc/net/scapy.  I have never used it, but
> it is supposed to be good.
>
>
>    Harold

Thank you Harold, it works great!

On 2018/12/04 10:58, Rin Okuyama wrote:
> Thank you for working on this! I revised your patch as follows:
>
> http://www.netbsd.org/~rin/if_mue_20181204.patch
>
> (1) We need to check bits 30 and 31 of rx_cmd_a, which indicate
>      checksum error detected by H/W for L4 and L3, respectively.
>      Otherwise, broken packets bypass any checksumming.
>
> (2) Mask csum by ifp->if_csum_flags_rx in mue_rxeof() for safety.
>
> (3) Some style changes.
>
> I've tested the patch with Z-TEK ZE582 (LAN7500). It works fine
> as far as I can see, although no checksum errors were detected
> during test.

I've updated the patch:

http://www.netbsd.org/~rin/if_mue_20181204-1.patch

(a) We cannot use bit 22 of rx_cmd_a (MUE_RX_CMD_A_RED) to drop
    wrong packets, since it is turned on in the cases of L3/L4
    checksum errors. Use bits 16--21 instead.

(b) For (3), undo renaming MUE_RX_CMD_A_FOO in the previous patch
    in favor of ones in Robert's original patch, that are taken
    from the data sheet. Sorry Robert, you were right.

Now, the driver can handle checksum errors correctly!

rin