Meeting started at 2026-07-21 16:00:12 UTC

The Meeting name is 'riscv-sig'

!hi

morning, who all is around today for a riscv sig meeting?

abologna: None (abologna)

!hi

!hi

!hi

Neil Hanlon: Neil Hanlon (neil) - he / him / his

davidlt: David Abdurachmanov (davidlt)

Marcin 'hrw' Juszkiewicz: Marcin Juszkiewicz (hrw)

welcome everyone

We have done so progress with F45 prep before mass rebuild.

The primary mass rebuild for f45 is over and merged back

I think, most majort bits have landed with majority of SOVERION bumps, etc.

davidlt: do you think now would be a good time to switch the tracker over to f45?

I have rebuilt binutils, annobin and GCC 16 rebuild is cooking. I might wait until that's done before pushing for our mass rebuild.

It should probably be OK to switch, apart that pretty much every NVR is different now.

how long does gcc take these days? (it's even really long on primary)

note that on primary builders were busy with rebuild so wait times were long

davidlt: I'm not in a hurry. just let me know when you think the time is right and I'll do it :)

sure, just curious. I can go look myself too. ;)

~70 hours for GCC.

what hardware is that? P550 or K3?

something like 28ish for primary... which is also wild. ;(

anyhow... anything else for current status?

K3

Python rebuild goes slowly ahead. We had 172+ packages built against 3.14, 100+ now.

Assuming we can build it on K3. The last one failed, so we will know in <3 days.

I yearn for the days when we'll have ~24h gcc builds :)

I track rebuild list in https://forge.fedoraproject.org/riscv/planning/issues/27 ticket

"K3 brought us better builders. They are faster. Not fast yet, just faster"

progress is progress. ;)

for sure :)

The problem is now that K1 and K3 might be unstable 😄

yes. still - `vtk` needs bootstrap every time which takes days.

Yeah, VTK is ~70 hours

vtk is a beast yeah. ;(

But I did not tracked how Fedora dealt with them.

and due to it we do 'mass rebuilds' in some kind of 'non-mass rebuild way'. gdal, hdf5, highfive, vtk, vpx have depending packages from f44 which we did not built for f45 and we do mass rebuild instead of 'per-package' ones.

but with current amount and speed of builders I think that our way is best we can have

you're welcome to use the fedora script, but yeah, I don't think thats a very good idea given builder speeds...

I track SOVERSION bumps, and mostly work on bigger onces. If smaller ones get pulled, it's fine.

we have 19 risc-v builders now.

gdal is a bit annoying since it's on so many SOVERSION bumps list.

and takes from 1.5h to 8h

that's a wild range!

that's our hardware ;D

luck of the draw?

vf2 or p550 or k3?

almost 20 builders is not bad. and IIUC Neil Hanlon is going to add more

alright, anything else on status? shall we go to hardware news?

anything new in hardware? It has seemed like it's been pretty quiet... but I could easily have missed things...

Arace started sending Titans

yeah that's pretty big

finally. ;) (but cool... )

64GB builders for those packages which need it?

if we had any?

someone on channel wrote about shipping label

not tracked who as I did not put order

Jason got 2 or 4 Titans.

Not having vectors + high RAM/core will be helpful.

Not to mention that large LLC cache.

I'm sure we can find some use for them lol

sooner or later we will have to make profiles in koji

All that Rust code will love it 😆

'big ram', 'fast cpu', 'whatever' and then package groups with preferences

I have a small vibe coded scripts that picks ans assigns a board to a free K3 😄

I have a ticket for Koji to support assigning a task to a channel.

Should be an easy vibe code feature, very small implementation.

we could also do policy I guess... if we know what packages should be in which channels...

Yeah, but that case we would need to constantly update the policy.

yep.

so not ideal for sure.

I do a lot of semi-manual build assignment stuff.

still 'if gcc then grab fast-cpu one' applies

yeah, some assignments would be pretty easy... others less so

if package not on list then gets whatever available

kind of 'all packages are equal, but some are equalier'

We also need temporary assigments. Since we cannot build Rust on P550, and I avoid landing some packages on K1/K3.

yeah, as with many things... it gets complicated. :)

davidlt: cannot build Rust on P550? why?

Yes, it now hangs P550 boards.

It would be fun to connect hardness to UART, SSH, JTAG and let it figure it out 😆

oh! we actually ran into that while building RHEL 10.2. the latest firmware from SiFive should fix it

from the release notes:

> Enables a workaround for a low occurrence issue that can lead to a system hang. The CSR settings have been changed to prevent the issue and the potential for a system hang. We do not expect any material impact on the performance.

I see is on GH.

cool.

anyone have anything else for open floor?

well one thing, maybe

go!

I don't know if people noticed but https://github.com/rhboot/shim/pull/778 got merged

hrw gave a cookie to abologna. They now have 11 cookies, 1 of which were obtained in the Fedora 44 release cycle

so we might be looking at a proper shim package in the not-too-distant future

tool less than a year even. :)

took less than a year even. :)

davidlt gave a cookie to abologna. They now have 12 cookies, 2 of which were obtained in the Fedora 44 release cycle

kevin gave a cookie to abologna. They now have 13 cookies, 3 of which were obtained in the Fedora 44 release cycle

Does that mean it's time for us to rebuild shim ? 😉

We still running the one we hacked probably 1-2 years ago.

I would leave things as they are for now. I'll be working with the maintainers to figure out the Fedora packaging situation, so it will all be sorted out in due time

We do have a custom pacakge for it, `shim-unsigned-riscv64-15.8-3.1.riscv64`

So it's "standalone"

We don't use "shim" as that one is signed and has special build treatment.

yeah. the current build is not really holding us back in terms of not being able to boot boards or anything like that, so it's fine to keep using it for a while longer

in the long run things will probably look quite a bit different

work in progress :)

Actually to my surprise the last shim was built in F44 😄

but that's f46 material

anyway, that's all I have :)

yeah, there was a new major version along in f44 timeframe...

alrighty... thanks for coming everyone!