>> I believe that adding kludges to work around other people's
>> brokenness is almost invariably a Bad Thing, since it removes their
>> incentive to fix their brokennesses.

> Hmm, should we go through all of the ethernet drivers and remove all
> of the code to work around various "bugs" in the hardware?  After all
> it is a work around for other broken behavior.

I've delayed responding this long in an attempt to sort out my position
on it.  This is a fully valid response to my comment and deserves a
serious reply.  However, I'm having trouble pinning down the relevant
difference between won't-frag and (say) SDEV_QUIRK_NOLUNS.

> We should write one standard driver and if someone has broken
> hardware that doesn't work without a "kludge" then too bad for them.

To an extent that's what we do, with the sbus and pci and such drivers;
if there were a standard for CPU<->NIC interfaces, we probably would do
it there.  But on the other hand, things like LANCE_REVC_BUG and the
various quirk tables represent exactly the sort of kludgery that I
spoke out against.

I don't have a good answer.  I wouldn't even be posting this now except
that in this case silence would too probably (and plausibly) be
interpreted as ignoring a disagreeable comment.

/~\ The ASCII				der Mouse
\ / Ribbon Campaign
 X  Against HTML	       mouse@rodents.montreal.qc.ca
/ \ Email!	     7D C8 61 52 5D E7 2D 39  4E F1 31 3E E8 B3 27 4B