Remove x86/generic target

[Elliott Mitchell]

On Sat, May 03, 2025 at 10:48:07PM +0200, Stefan Lippers-Hollmann via openwrt-devel wrote:
> 
> On 2025-05-01, Hauke Mehrtens wrote:
> > 
> > We would like to remove the 32 bit x86/generic target. The target is 
> > still working, but we could save some build resources.
> [...]
> > We would also keep the x86/legacy and the x86/geode target. We can 
> > activate some extra kernel build options if it can not replace 
> > x86/generic with the current configuration.
> 
> It would probably make sense to merge legacy and geode into a single
> 32 bit x86 subtarget as well. I can't imagine that there is enough of 
> an advantage in /x86/ mach fine tuning to warrant multiple dedicated 
> subtargets. It's just unnecessary churn both on the buildbots and the
> human resources to rebase the configs for new kernels.
> 
> Don't get me wrong, I'm not at all advocating for dropping 32 bit
> x86 support, just to merge it into a single target for all 32-bit
> x86 devices.

According to what I've read processors under the Geode brand were
manufactured until 2019.  Apparently they were being used in some
switches and slot machines.  There may be enough Geode processors still
in active use to be worthy of a distinct target.

The real problem for x86 in general is x86 *really* needs modular kernels
and boot time module probing.  There have been hundreds of x86 plug-in
card manufacturers over the years.  While there are some very common
pieces of hardware, these bloat the kernel for every machine which lacks
those.


> Just to feed my own curiosity, it would be interesting to hear what 
> kind of 32 bit x86 devices are still in active use with OpenWrt today.
> (Considering that the more modern 64 bit AMD Jaguar cores just barely 
> reach 1 GBit/s routing+NAT throughput, while mt7621, ipqXXxx, mt7622,
> filogic, etc. should be faster in this capacity than any semi power 
> efficient 32 bit x86 hardware; Atom n270 (<<600 MBit/s) is still 
> ~25-30 watts idle, P4/ K7 around the 100 watts mark +/- 30% depending 
> on the details, anything low-power should be even slower than the N270).

The issue is how many x86 targets should OpenWRT have?

Getting rid of "generic" is a no-brainer.  More Pentium 4 processor
machines can handle 64 than will be forced back to legacy.  At this point
>98% of x86 machines are amd64 and having "generic" be ia32 is absolutely
wrong.

There is pull request #14302 (https://github.com/openwrt/openwrt/pull/14302)
to split "x86/64" into 4 distinct targets.  That seems a bit much, but
it might be time to split "x86/64".

I think there is greater value in creating specialized x86 VM targets.
For a machine with 16GB of memory, devoting 128MB of memory and a
processor core to networking seems reasonable.  I could see splitting
into "large" and "small", mostly centering on ACPI support (KVM and Xen
VMs often omit ACPI, but Hyper-V and VMWare require it to boot).


-- 
(\___(\___(\______          --=> 8-) EHM <=--          ______/)___/)___/)
 \BS (    |         ehem+sigmsg at m5p.com  PGP 87145445         |    )   /
  \_CS\   |  _____  -O #include <stddisclaimer.h> O-   _____  |   /  _/
8A19\___\_|_/58D2 7E3D DDF4 7BA6 <-PGP-> 41D1 B375 37D0 8714\_|_/___/5445