Re: Old platforms: bring out your dead

[Rob Landley <rob@landley.net>]

On 1/11/21 8:55 AM, chase rayfield wrote:
> On Mon, Jan 11, 2021 at 3:09 AM John Paul Adrian Glaubitz
> <glaubitz@physik.fu-berlin.de> wrote:
> 
>>
>> I'm not sure I understand the reasoning for doing this. The SPARC architecture
>> isn't going to see any new hardware developments in the future after Oracle
>> let go of most of the SPARC developers. So it's not that we need to make room
>> for new hardware.
>>
> My take is that there *would* be more interest in Sparc sun4m / Sun4d
> from enthusiasts at the very least if it was possible to actually boot
> the bloat hog that is Linux these days in a fully usable configuration
> that probably means some modifications to SILO and Linux required.

You can trim current linux down a bit, it's just non-obvious how. Unfortunately
there's an "expert" menu and CONFIG_EMBEDDED and if you touch anything there's
suddenly a hundred extra options in your config with no explanation of what they do.

At least 50% of what you want is probably disabling the printk strings that
aren't visible at your default verbosity level, but alas you must open pandora's
box to access those options...

> The problem is as I understand it, SILO only sets up a 16Mb mapping
> (either due to having to assume 4MB minimum dram stick size or due to
> mapping limitations not sure, most of these machines have at least
> 16MB in slot one...these days though that wasn't the case for sun4c),
> loads Linux into it and says good Luck. This isn't enough for a modern
> kernel with any  hardware support built in. So you might for instance
> get a kernel to fit but only if you dropped all of networking support
> etc... I'm guessing the fix for this would be to modify silo to map a
> larger amount in a way that Linux expects so it can remap it as it
> likes, or just have SILO map the full memory as Linux would. Anyway
> that is THE main demotivation for these architectures.... otherwise
> they have plenty of ram and performance to do basic router/server
> tasks sans SSL.

A lot of people with hardware like this haven't stopped using it, they've just
stopped fighting with kernel upgrades. (Common issue in the embedded world. Not
really a fun thing for security, but )

> This has been the status quo for since the last of the 2.6 series of
> kernels which it was still possible to just barely squeeze a usable
> kernel out of... If someone wanted to take a few hours and fix this
> issue, and keep these architectures around I'd be happy to "buy them a
> round of pizza", though I recognize that many people that work on this
> already have nice jobs, and just don't have time.

My https://github.com/landley/toybox/blob/master/scripts/mkroot.sh ~250 line
bash script generates the simplest kernel configs for a bunch of platforms to
boot qemu to a shell prompt, but you then have to open the "expert" menu and
_disable_ stuff in order to get the size down from there.

> Also Sparc would probably be a good project for someone to extend/test

Sparc has a runtime relocation I've never understood but did manage to break
once, resulting in a long thread to fix:

http://lists.landley.net/pipermail/aboriginal-landley.net/2011-December/001964.html

Between that and the weird save half the stack register thing with function
calls on some sort of "wheel"... there's a _reason_ I haven't been able to talk
Rich into adding support for it to musl.

> Andi Keen's Linux LTO patch set so we could reduce the kernel binary
> size that way also even if sun4 architectures are dropped, it would
> still be useful for embedded sparc. Also there is a port of Temlib to
> the Mister hardware now, 3 cores roughly equivalent to a mid 90s
> machine, at least 128MB ram is possible ( more if a way to map the ARM
> system memory also 1GB is available there, it would have higher
> latency though).
> 
> It is perfectly viable to build Sparc v7 or v8 32bit binaries in a
> chroot on a fast machine also, and I would recommend this if you wish
> to retain sanity rather than attempting cross compiler voodoo, unless
> that is your thing.

It is, sadly, my thing. The above 250 line bash script builds:

aarch64  armv7l  i686        mips    powerpc      s390x  x86_64
armv4l   armv7m  m68k        mips64  powerpc64    sh2eb
armv5l   i486    microblaze  mipsel  powerpc64le  sh4

That's toybox booting to a shell prompt and a linux kernel configured for qemu
for each target. Adding new targets looks something like:

elif [ "$TARGET" == m68k ]; then
  QEMU="m68k -M q800" KARCH=m68k KARGS=ttyS0 VMLINUX=vmlinux
KCONF=MMU,M68040,M68KFPU_EMU,MAC,SCSI_MAC_ESP,MACINTOSH_DRIVERS,ADB,ADB_MACII,NET_CORE,MACSONIC,SERIAL_PMACZILOG,SERIAL_PMACZILOG_TTYS,SERIAL_PMACZILOG_CONSOLE
elif [ "$TARGET" = s390x ]; then
  QEMU="s390x" KARCH=s390 VMLINUX=arch/s390/boot/bzImage
KCONF=MARCH_Z900,PACK_STACK,NET_CORE,VIRTIO_NET,VIRTIO_BLK,SCLP_TTY,SCLP_CONSOLE,SCLP_VT220_TTY,SCLP_VT220_CONSOLE,S390_GUEST

(Well, modulo thunderbird being unable to an indent a line that goes off the
right edge of the screen. The mozilla foundation somehow managed to spend half a
billion dollars in 2019 but it wasn't on thunderbird, I can tell you that.)

Anyway, I wrote a couple FAQ entries trying to explain the worst of it:

  https://landley.net/toybox/faq.html#cross
  https://landley.net/toybox/faq.html#mkroot

> Anyways it could be that people that want this get around to fixing
> SILO eventually and just sit on this last kernel version... *shrugs*

They're never sitting on the _last_ kernel version. They're generally way back
from there. Been true forever off of x86 (and now arm):

https://lore.kernel.org/lkml/201002211025.11588.rob@landley.net/T/

> Chase

Rob