James Carlson wrote:
Garrett D'Amore wrote:
James Carlson wrote:
Garrett D'Amore wrote:
Some hardware simply *can't* easily do this. Some drivers are really
stale and haven't been updated:
* dnet only does link state reporting on certain cards (this with MII
transceivers)
* pcelx and elx don't do this at all
Thanks; I forgot about the elx problem.
I'd love to just EOF elxl.
elx != elxl. The former is pretty dead, and was blown to bits by the
GLDv0 removal [PSARC 2003/728] and its own EOF [PSARC 2003/701], but
elxl is still somewhat common. It's a 3c905, if I recall correctly.
Yes, that's true. (There are some other models for elxl as well, e.g.
3c900-TPO.)
I suspect some people out there still depend on ancient elx hardware.
Still, I wonder if now with our vastly improved set of NIC choices, if
we can't now simply EOF it. Suitable replacements are *easy* to find
now -- $10 will get you a Linksys LNE100TX, and $5 will get you a
suitable rtls8139.
elx is closed source anyway, and getting sufficient "variety" of devices
to properly test all the variations might prove to be "non-trivial".
Thoughts? Who would complain if elxl or pcelx were scheduled for EOF?
Dunno. Besides students and other folks hanging onto ancient gear, the
two big places where I've seen these really old drivers in use is on
virtualization platforms and integrated chipsets. It looks like VMWare
emulates e1000g, so maybe that one's not a serious problem, but I *know*
we've blown some modern platforms out of the water because we EOF'd old
drivers.
I think v12n is not affected. I don't know of any emulated environment
that emulates the elxl. (Conversely, I do know that Parallels emulates
the ancient realtek 8029!)
I don't think 3c9xx chips have been on "integrated" motherboards for
quite a long time. Mostly because they were probably more expensive
than other more common variants. (Realtek "owned" most of the embedded
space here, and still does, in many respects. Though Intel mobo's
obviously have Intel parts, likewise most Nvidia parts. Other players
are more common now -- Marvell, Attansic, etc.)
If nothing else, terminating elxl will cause lots of machines still in
Sun labs around the world to go dead, including machines that are used
for automated testing of IPMP.
That's interesting. I'd think most of those machines are pushing the
limit on having enough resource to run Solaris Nevada these days. Now I
see that elxl does indeed have link state notification -- I thought it
didn't. (I'm embarrassed now, since I know that I've worked on that
code in the not too distant past.)
I no longer have access, but search for GLDv0. There was at least one
motherboard that was bricked by that removal, because it had a chipset
that emulated old elx (not elxl) hardware.
Weird.
Maybe I should post this more widely, and run it by PAC.
-- Garrett
_______________________________________________
networking-discuss mailing list
[email protected]
