Interest in "Porting musl libc to RISC-V" project for GSoC 2016

Interest in "Porting musl libc to RISC-V" project for GSoC 2016

[Masanori Ogino]

Hello,

I'm interested in porting musl to RISC-V arch project on
http://www.lowrisc.org/docs/gsoc-2016-ideas/ . I have read guidelines
for GSoC students and now I'm preparing to write up my proposal.

I'd like to ask you for your advice to understand technical details
and improve my plan. Please let me know if you have any thoughts.
Thank you.

Regards,
-- 
Masanori Ogino

Re: Interest in "Porting musl libc to RISC-V" project for GSoC 2016

[Szabolcs Nagy]

* Masanori Ogino <masanori.ogino@gmail.com> [2016-03-03 16:43:31 +0900]:
> I'm interested in porting musl to RISC-V arch project on
> http://www.lowrisc.org/docs/gsoc-2016-ideas/ . I have read guidelines
> for GSoC students and now I'm preparing to write up my proposal.
> 
> I'd like to ask you for your advice to understand technical details
> and improve my plan. Please let me know if you have any thoughts.
> Thank you.

last time iirc musl port was considered to be
not large enough in itself for a gsoc project
(i think it should not take more than 2 months
but i don't know the current state of risc-v
qemu/linux/gcc/etc)

you might need to think additional things to work on.

for musl, one idea is to invest extra time on testing.

for risc-v i think polishing the toolchain and the docs
would be useful.


there are some basic problems with the risc-v software
eco-system:

there is no proper sysv psabi spec. (designing one would
have been better than copying obsolete nonsense from mips
as i can see in the glibc port.)

nothing is upstream yet (gcc/linux/.. ports are
maintained out of tree, working with the upstream
community is important for many reasons).

risc-v mailing lists are not public, only subscribers
can see or participate in the discussions. (this is
bad given that there are no specs, no upstreamed code
so no source of information for outsiders.)

for a musl port this means that we don't have abi
stability guarantees, the port can stop working with
the rest of the risc-v software stack.   so for a
successful port i think some scripts should be developed
to build and test the latest risc-v things against musl
(cross-toolchain, rootfs, etc) so we can keep it working.

Re: Interest in "Porting musl libc to RISC-V" project for GSoC 2016

[Masanori Ogino]

Hello,

2016-03-04 7:06 GMT+09:00 Szabolcs Nagy <nsz@port70.net>:
> last time iirc musl port was considered to be
> not large enough in itself for a gsoc project
> (i think it should not take more than 2 months
> but i don't know the current state of risc-v
> qemu/linux/gcc/etc)
>
> you might need to think additional things to work on.
>
> for musl, one idea is to invest extra time on testing.
>
> for risc-v i think polishing the toolchain and the docs
> would be useful.

Agreed.

> there are some basic problems with the risc-v software
> eco-system:
>
> there is no proper sysv psabi spec. (designing one would
> have been better than copying obsolete nonsense from mips
> as i can see in the glibc port.)

Hmm, ABI is an important factor for the proposal, but I probably won't
have enough time to invent a new ABI and implement it during this
summer...
Anyway, would you give me any well-designed psABI (or non-SysV-psABI) examples?

> nothing is upstream yet (gcc/linux/.. ports are
> maintained out of tree, working with the upstream
> community is important for many reasons).

Agreed with the importance of merging to upstream.

> risc-v mailing lists are not public, only subscribers
> can see or participate in the discussions. (this is
> bad given that there are no specs, no upstreamed code
> so no source of information for outsiders.)

I think it would be great if archives are public. I don't know the
reason why the project decide to do so.

> for a musl port this means that we don't have abi
> stability guarantees, the port can stop working with
> the rest of the risc-v software stack.   so for a
> successful port i think some scripts should be developed
> to build and test the latest risc-v things against musl
> (cross-toolchain, rootfs, etc) so we can keep it working.

Indeed it's great to have an automated testing infrastructure. I really love CI.

Thank you for your suggestion!

-- 
Masanori Ogino

Re: Interest in "Porting musl libc to RISC-V" project for GSoC 2016

[Rich Felker]

On Fri, Mar 04, 2016 at 09:54:44AM +0900, Masanori Ogino wrote:
> Hello,
> 
> 2016-03-04 7:06 GMT+09:00 Szabolcs Nagy <nsz@port70.net>:
> > last time iirc musl port was considered to be
> > not large enough in itself for a gsoc project
> > (i think it should not take more than 2 months
> > but i don't know the current state of risc-v
> > qemu/linux/gcc/etc)
> >
> > you might need to think additional things to work on.
> >
> > for musl, one idea is to invest extra time on testing.
> >
> > for risc-v i think polishing the toolchain and the docs
> > would be useful.
> 
> Agreed.
> 
> > there are some basic problems with the risc-v software
> > eco-system:
> >
> > there is no proper sysv psabi spec. (designing one would
> > have been better than copying obsolete nonsense from mips
> > as i can see in the glibc port.)
> 
> Hmm, ABI is an important factor for the proposal, but I probably won't
> have enough time to invent a new ABI and implement it during this
> summer...
> Anyway, would you give me any well-designed psABI (or non-SysV-psABI) examples?

I don't think you'd need to change/redesign the ABI (since it's
already being used on FreeBSD, presumably whatever current Linux
kernel port exists, and maybe other targets, and didn't look bad when
I looked at it), but a nice clean psABI document would be nice to
have.

> > nothing is upstream yet (gcc/linux/.. ports are
> > maintained out of tree, working with the upstream
> > community is important for many reasons).
> 
> Agreed with the importance of merging to upstream.
> 
> > risc-v mailing lists are not public, only subscribers
> > can see or participate in the discussions. (this is
> > bad given that there are no specs, no upstreamed code
> > so no source of information for outsiders.)
> 
> I think it would be great if archives are public. I don't know the
> reason why the project decide to do so.

Yes, that would be really nice.

> > for a musl port this means that we don't have abi
> > stability guarantees, the port can stop working with
> > the rest of the risc-v software stack.   so for a
> > successful port i think some scripts should be developed
> > to build and test the latest risc-v things against musl
> > (cross-toolchain, rootfs, etc) so we can keep it working.
> 
> Indeed it's great to have an automated testing infrastructure. I really love CI.
> 
> Thank you for your suggestion!

I have a few other ideas I'm going to post to the list for elements
that could be incorporated into proposals. I'll follow up as a
separate email.

Rich

Re: Interest in "Porting musl libc to RISC-V" project for GSoC 2016

[Masanori Ogino]

2016-03-04 12:02 GMT+09:00 Rich Felker <dalias@libc.org>:
> I don't think you'd need to change/redesign the ABI (since it's
> already being used on FreeBSD, presumably whatever current Linux
> kernel port exists, and maybe other targets, and didn't look bad when
> I looked at it), but a nice clean psABI document would be nice to
> have.

OK.

> I have a few other ideas I'm going to post to the list for elements
> that could be incorporated into proposals. I'll follow up as a
> separate email.

Thank you! I will dig into them.

-- 
Masanori Ogino