Deorbiting armv4 and armv5 support

[Warner Losh]

On Sun, Jun 11, 2017 at 10:41 PM, Warner Losh <imp at bsdimp.com> wrote:

>
>
> On Sun, Jun 11, 2017 at 3:28 PM, Ronald Klop <ronald-lists at klop.ws> wrote:
>
>> On Thu, 08 Jun 2017 22:22:51 +0200, Warner Losh <imp at bsdimp.com> wrote:
>>
>> Greetings,
>>>
>>> The pmap in armv4/5 has been broken for years at this point. It basically
>>> works, but has issues with unaligned buffers. Since these have remained
>>> unfixed for a long time, and most of the main ARM developers have given
>>> up
>>> trying to fix it, I think it would be best to retire the support rather
>>> than waste people's time that seem to be working only to discover after a
>>> lot of effort that it's busted.
>>>
>>> Since the consensus at the FreeBSD developer's summit appeared to be
>>> 'let's
>>> let it go'. It would remove the TARGET_ARCH arm and armeb. armv6 would
>>> remain unaffected (though see a parallel thread).
>>>
>>> Warner
>>>
>>
>> What branches are you talking about?
>>
>
> I'm thinking only the -current branch.
>
> However, I'm thinking now that I'll write up something more formal (the
> FCP process) and start maybe a 3 month timeout. There's two leads on a fix,
> I'll be looking at one of them in the next few days. If one of these fixes
> is good, I'll merge it. The second one is a long shot, though...
>
> So, my current plan is that if it remains unfixed by, say Sept 1, 2017,
> I'll deorbit. I may adjust that date depending on the timing of the 12
> branch, since I want to get this resolved before the branch...
>

I've managed to get my old Atmel AT91SMA9G20 based board working, so I'm
officially withdrawing this idea. I heard from a few people that had no
issues with their armv5 boards and was wondering what I was talking about.
I've found the root cause to my issues, so I think we're in good shape from
a functional point of view. It looks like the unaligned issues that I'd
seen in prior versions have been fixed at least since Feb of this year,
maybe longer.

As a side note, both the armv4 and armv6 busdma implementations work on
armv4/5 hardware. Though the v6 implementation seems to use about 20-30%
more memory in a quick test...

Warner