|| ||Bjørn Mork <bjorn-AT-mork.no> |
|| ||Paul Gortmaker <paul.gortmaker-AT-windriver.com> |
|| ||Re: [PATCH/RFC net-next 0/4] Delete token ring support. |
|| ||Wed, 16 May 2012 11:29:59 +0200|
|| ||davem-AT-davemloft.net, netdev-AT-vger.kernel.org,
Martin Schwidefsky <schwidefsky-AT-de.ibm.com>,
Heiko Carstens <heiko.carstens-AT-de.ibm.com>,
|| ||Article, Thread
Paul Gortmaker <firstname.lastname@example.org> writes:
> This may not be for 35-next, but for addition to feature-removal.txt
> and application at a later date. But what I would like to get is
> consensus that this is something that we want to proceed with before I
> spend any more time on it. If folks are OK with the idea, then I am
> open to suggestions as to the best time for it to happen.
> So, why you might ask? It is in tree already, it is "free" to leave it
> there, right? Well, no.
> What led me here was the creation of a patch to remove CONFIG_MCA. In
> doing so, I found I was deleting most of the token ring drivers, and
> altering the remaining few ISA/MCA ones to be just ISA only.
With all due respect, I do not think you have looked thoroughly enough
at the remaining drivers...
> But it really didn't make sense to me, to just leave the skeletal
> remains of token ring there -- vs removing TR 1st, and then the MCA
> removal will be a lot smaller and cleaner commit.
> Removal: does it make sense?
> The biggest data point I can suggest to folks is to run:
> git whatchanged --follow drivers/net/tokenring
> and when you are quickly paging over the changes, note two things.
> (1) the amount of time spent by folks cleaning and maintaining the
> code, and (2) - most important -- is that essentially all the commits
> are of the tree-wide cleanup nature, or API change nature.
> What I mean by (2) is the implicit absence of anyone fixing _runtime_
> bugs, going all the way back to 2.6.12 in 2005. If the code was being
> _used_, we'd see runtime regressions reported and their associated fixes.
I think you underestimate the girls(?) and guys doing tree-wide cleanups
and API changes. Network driver regressions are pretty much limited to
actively developed and maintained drivers.
> A search on the internet for users tends to show that even the die hard
> enthusiasts who cared to poke at MCA/TR just for hobby sake have pretty
> much all given up somewhere in the 2003-2005 "pre-git" timeframe, and
> never really moved off their 2.4.x kernels.
> This is no surprise, since on x86, MCA (and hence most tokenring
> users) was limited to the lowly 386sx-16 PS/2 with typically 4MB RAM.
> Some "high end" 486 machines existed, and in theory could be fitted
> with max 64MB RAM (default 16MB). I don't think anyone will debate
> that such hardware with such limited memory is ever going to be useful
> in a 3.x kernel use case.
Beware, I am considering sending you a triple lanstreamer PCI card :-)
Seriously, there are both PCI and PCMCIA TokenRing cards, and nothing
prevents them from being used with modern x86 hardware. The lanstreamer
driver is disabled for 64bit systems, but I still don't think removing
it is appropriate. It will work in 32bit mode in any shiny new PC still
having PCI slots. And the 3c359 should work in any laptop with a
Cardbus slot, if those still exist. But I don't see anyone proposing
removing the PCMCIA/Cardbus support just yet....
to post comments)