Re: Current ULE state and plans (was Re: HMP scheduling on FreeBSD)

On Thu, 22 Jan 2026 at 12:12, Warner Losh <imp@bsdimp.com> wrote:
>
>
>
> On Thu, Jan 22, 2026 at 12:57 PM George Mitchell <george+freebsd@m5p.com> wrote:
>>
>> On 1/22/26 14:21, Konstantin Belousov wrote:
>> > Selecting an arbitrary message to reply in the thread.
>> > Also added arch@, perhaps hackers@ should be dropped for replies.
>> >
>> > I wrote a patch that allows to compile both ULE and 4BSD into same kernel,
>> > and select the scheduler at boot.  See https://reviews.freebsd.org/D54831
>> > First comment on the review contains important usage information and
>> > some special note about armv7.
>>
>> Wow!  Could we send this, say, fifteen years back in the time machine?
>
>
> ifuncs are 15 years old. And we'd need to get armv7 working with ifuncs since this utterly breaks armv7 w/o them.

I think this is a pretty good reason to sit down and just get kernel
ifuncs working / checked
on every platform in -HEAD. It's about time. :-) (and if this would
ever be MFCed, even more..)

Anyway, the diff also does change a few things which i do remember
causing interesting
issues on MIPS with performance and preemption and timekeeping when all of that
fun stuff was in-flight. (I remember having to even fix the MIPS enter
idle / interrupt disable
stuff due to latency issues the timekeeping/scheduling stuff introduced.)

So I could totally believe some un-diagnosed issues and weird corner cases were
mis-attributed to "must be ULE vs 4BSD" when it was some side effect we didn't
quite get.

(Also, now i wonder if ifuncs work on ppc/ppc64..)



-a