microblaze: big endian or little endian (or both)?

microblaze: big endian or little endian (or both)?

[Isaac Dunham]

I'm trying to get stuff set up to run microblaze musl (first user emulation, then system emulation if all goes well).
Anyhow, I'm wondering about whether to just use microblaze or if microblazeel should work also (configure only says microblaze-* which seems to imply that only the BE version is supported).

Other than that, I have a few random thoughts on microblaze systems to experiment with:
1. The xldk git repos have initrds for microblaze; if you inject musl binaries in there, how does that work in qemu?
2. Apparently it's possible (just barely) to get a Linux-capable microblaze design onto a Spartan 6, which runs ~$90...

-- 
Isaac Dunham <idunham@lavabit.com>

Re: microblaze: big endian or little endian (or both)?

[Rich Felker]

On Tue, Oct 16, 2012 at 08:05:41PM -0700, Isaac Dunham wrote:
> I'm trying to get stuff set up to run microblaze musl (first user
> emulation, then system emulation if all goes well).
> 
> Anyhow, I'm wondering about whether to just use microblaze or if
> microblazeel should work also (configure only says microblaze-*
> which seems to imply that only the BE version is supported).

I don't see any reason little endian wouldn't work; I wasn't aware
that it even existed. Aside from some workarounds for kernel struct
layout and other ugly ABI issues on some archs, musl is completely
endian-agnostic.

> Other than that, I have a few random thoughts on microblaze systems
> to experiment with:
> 
> 1. The xldk git repos have initrds for microblaze; if you inject
> musl binaries in there, how does that work in qemu?

Haven't tried. I'd really rather figure out how to get rootfs-on-9p
working and use that for all my qemu-system experimentation.

> 2. Apparently it's possible (just barely) to get a Linux-capable
> microblaze design onto a Spartan 6, which runs ~$90...

Nice. Any idea what the performance is like?

Rich

Re: microblaze: big endian or little endian (or both)?

[Isaac Dunham]

On Tue, 16 Oct 2012 23:13:44 -0400
Rich Felker <dalias@aerifal.cx> wrote:

> On Tue, Oct 16, 2012 at 08:05:41PM -0700, Isaac Dunham wrote:
> > I'm trying to get stuff set up to run microblaze musl (first user
> > emulation, then system emulation if all goes well).
> > 
> > Anyhow, I'm wondering about whether to just use microblaze or if
> > microblazeel should work also (configure only says microblaze-*
> > which seems to imply that only the BE version is supported).
> 
> I don't see any reason little endian wouldn't work; I wasn't aware
> that it even existed. Aside from some workarounds for kernel struct
> layout and other ugly ABI issues on some archs, musl is completely
> endian-agnostic.
OK, thanks!

Somehow this gets me wondering about the endian.h issue...
#set ENDIAN in the first round to the per-arch default

case "$target" in 
  *el-* | *el) ENDIAN=BE ;;
  *eb-* | *eb) ENDIAN=LE ;;
  *) ;; #Just use the default endianness
esac
then, have ENDIAN determine whether to use the big-endian or little-endian header:

ln -s arch/$ENDIAN/endian.h include/bits/endian.h

clean: cleanendian

cleanendian:
	rm include/bits/endian.h

> > Other than that, I have a few random thoughts on microblaze systems
> > to experiment with:
> > 
> > 1. The xldk git repos have initrds for microblaze; if you inject
> > musl binaries in there, how does that work in qemu?
> 
> Haven't tried. I'd really rather figure out how to get rootfs-on-9p
> working and use that for all my qemu-system experimentation.

Yikes. That might take a while...
 
> > 2. Apparently it's possible (just barely) to get a Linux-capable
> > microblaze design onto a Spartan 6, which runs ~$90...
> 
> Nice. Any idea what the performance is like?
According to 
http://billauer.co.il/blog/2011/09/dma-capture-frame-grabbing-xilinx-fpga-linux-microblaze/ 
it's 70-100mhz effectively; microblaze seems to be extremely slow.

And yet they stuck a gigabit ethernet port on it.
-- 
Isaac Dunham <idunham@lavabit.com>