dgilmore changed the topic of #fedora-riscv to: Fedora on RISC-V https://fedoraproject.org/wiki/Architectures/RISC-V || Logs: https://libera.irclog.whitequark.org/fedora-riscv || Alt Arch discussions are welcome in #fedora-alt-arches
bkeys has quit [Ping timeout: 240 seconds]
bkeys has joined #fedora-riscv
bkeys1 has joined #fedora-riscv
bkeys has quit [Ping timeout: 244 seconds]
bkeys1 is now known as bkeys
bkeys has quit [Quit: With every step we take, danger will follow closely]
bkeys has joined #fedora-riscv
bkeys has quit [Client Quit]
bkeys has joined #fedora-riscv
bkeys has quit [Read error: Connection reset by peer]
bkeys has joined #fedora-riscv
bkeys has quit [Quit: With every step we take, danger will follow closely]
<davidlt[m]> Finally NSS build finished (aka failed)
<davidlt[m]> 5 failures
<davidlt[m]> One failure seems to be performance issue, pwdecrypt failed to deliver 300 entries within <1 second.
<davidlt[m]> The rest are: ERROR -8181: Peer's Certificate has expired.
<davidlt[m]> Which seems to be just expired PayPalEE.cert
<davidlt[m]> I wonder if overclocking Unmatched would make sense :)
<davidlt[m]> The problem with Unmatched boards as-is is that libcurl timeouts trying to build SRPM if the source tarballs are large enough.
<davidlt[m]> Once NSS build finishes I will look into my board to see how I configured kojid on this new image.
jcajka has joined #fedora-riscv
<davidlt[m]> v5.18.3 soon to land
<davidlt[m]> stable-queue.git already got 5.18.3
<rwmjones> morning
<davidlt[m]> 5.18.3 seems to be 200-250 patches, nice
<davidlt[m]> rwmjones: do you want to pick hostname for the certificates or anything is fine? :)
<rwmjones> one sec
<rwmjones> the two new machines are called
<rwmjones> nufive.home.annexia.org
<rwmjones> nujive.home.annexia.org
<rwmjones> as I mentioned before if I'm going to set up the new machines, I'd like to retire the unleashed machines
<davidlt[m]> Yeah, I actually no longer have Unleashed :)
<davidlt[m]> I cannot even test on it anymore.
<davidlt[m]> I really hate NSS
<davidlt[m]> I mean the thing is compiled (and test are run) with -j1
<davidlt[m]> That's painful :)
<davidlt[m]> yeah, I will most likely wait for v.5.18.4
<davidlt[m]> I want to see how small or large it is
<davidlt[m]> v5.18.3 was quite big one
<davidlt[m]> and I also need to rebase libaudit :)
* rwmjones fishes out the toolless NVME enclosure ...
<davidlt[m]> those are the best
<davidlt[m]> I have one from sabrent
<davidlt[m]> Because NVMe kinda became a new microSD card for me
<davidlt[m]> This looks to be the fastest option to play with NVMe
<davidlt[m]> I would love to get rid of NMVe screw too, I got a rubber ones, but would prefer twist lock
<rwmjones> yeah I got one
<rwmjones> ok so to get this straight, Fedora-Developer-Rawhide-20211110.n.0.SiFive.Unmatched.firmware.raw.img.xz -> SD card (uncompressed)
<rwmjones> Fedora-Developer-Rawhide-20211110.n.0.SiFive.Unmatched.rootfs.raw.img.xz -> nvme drive (full drive??)
<rwmjones> let me see what's on the drives already, ISTR I had them booting a Fedora userspace
<davidlt[m]> Yes
<davidlt[m]> Note, that a newer image with v5.18.4 kernel will support SPI-NOR Flash.
<davidlt[m]> So that means you will be able to forget about that microSD card.
<rwmjones> davidlt[m]: the current situation is they are booting with extlinux, and have:
<rwmjones> actually foget about that, I'm going to redo them from scratch
<rwmjones> just want to see if there's anything in $HOME I want to save first
<rwmjones> nope
<davidlt[m]> yeah, still extlinux, but we should go into GRUB2 mode
<davidlt[m]> I believe Fu Wei already did this for Allwinner D1 builds
<rwmjones> $ nbdkit -r file Fedora-Developer-Rawhide-20211110.n.0.SiFive.Unmatched.rootfs.raw.img.xz --filter=xz --run 'guestfish --ro --format=raw -a $nbd'
<davidlt[m]> There are some issues with U-Boot GRUB2 IIRC (not exactly an issue)
<rwmjones> ><fs> run
<rwmjones> ><fs> list-filesystems
<rwmjones> /dev/sda1: ext4
<rwmjones> /dev/sda2: ext4
<rwmjones> (so that image is partitioned)
<davidlt[m]> boot and rootfs
<rwmjones> looks like vmlinuz-5.14.16-101.0.riscv64.fc33.riscv64
<davidlt[m]> I think, it's still ext4 (for rootfs), not btrfs yet
<rwmjones> ok so I'll put that on the NVME drives in a mo
* rwmjones checks the SD card image similarly
<rwmjones> ><fs> list-filesystems
<rwmjones> /dev/sda2: unknown
<rwmjones> ><fs> file /dev/sda1
<rwmjones> /dev/sda1: unknown
<rwmjones> data
<rwmjones> ><fs> file /dev/sda2
<rwmjones> Device Tree Blob version 17, size=734287, boot CPU=0, string block size=109, DT structure block size=733124
<davidlt[m]> Yeah, v5.14 is not the best, but it's way more stable build compared to the Unleashed boards.
<davidlt[m]> I will get LEDs working, DA9063 PMIC enabled (WDT, RTC, etc.), PCIe fixes, etc.
<davidlt[m]> That's just two partitions for the firmware, U-Boot SPL and proper.
<rwmjones> ok I'll put that on the SD cards
<rwmjones> let's go ...
<davidlt[m]> With newer U-Boot you could just directly that to SPI-NOR Flash, but this image didn't incl. those patches.
<rwmjones> do I really need big fan cooling? the fans I got run at full speed and are rather annoying
<rwmjones> there's no fan control header I think?
<davidlt[m]> There are 3 fan headers, 3-pin (just 2 wired).
<rwmjones> right, I mean they're not the ones that vary the speed?
<davidlt[m]> We use PWM to drive 2 out of 3, but just to get 12VDC.
<davidlt[m]> It's not a proper PWM 4-pin fan. It's not even a proper 3-pin fan header :)
<davidlt[m]> The one for SoC is directly wired to power rails.
<davidlt[m]> Yes, it's just 12VDC over 2-pins. You cannot even detect the RPM.
<davidlt[m]> Otherwise it's just random :)
<davidlt[m]> The rest of two requires software to enable them.
<davidlt[m]> Okay. Let's be more precise. The PWM is not initialized to some known state, it's up to U-Boot to initialized and enable those fan headers :)
<davidlt[m]> So if you want to be 100% sure nothing goes wrong, use the 2-pin header as that directly goes into 12VDC without software doing anything.
<rwmjones> such a waste using a 32G SD card for this :-(
<davidlt[m]> Actually lower capacity cards might be more expensive ;)
<davidlt[m]> IIRC going below 32-64GB things might be expensive and drop your performance. Anything above 64GB it's just capacity, you don't gain more performance.
<davidlt[m]> My advice is to cut a bracket for miniITX 120-140mm fan and use the SoC fan 2-pin header to blow from top-down.
<davidlt[m]> Technically that ASMedia PCIe wants some cooling too :)
<davidlt[m]> NVMe would be fine.
<davidlt[m]> Now WDT shuoldn't work properly on this board IIRC. It should cause poweroff.
<davidlt[m]> You might want to get some USB WDT (connect to some NUC) to trigger those front panel header pins.
<davidlt[m]> Anything that works with pcwd_usb driver.
<davidlt[m]> Oh, you might also want to reprogram FTDI chip if you plan to hook up the serial consoles to a single computer.
<davidlt[m]> Otherwise it's randomness party.
<rwmjones> ok writing the root disk of nufive
<rwmjones> I had to disassemble the case entirely to get a screwdriver in to release the nvme drive :-(
<davidlt[m]> Do you have RTC battery in?
<rwmjones> yes
<davidlt[m]> remember to increase rootfs partition to max size, disable /tmp on tmpfs ( systemctl mask tmp.mount ) and reboot
<rwmjones> 20:24 < davidlt[m]> sudo growpart -u auto /dev/nvme0n1 2; sudo resize2fs /dev/nvme0n1p2
<davidlt[m]> We cannot waste that RAM for /tmp crap.
<rwmjones> I can do the growpart *after* it's booted?
<davidlt[m]> That's what I have on my history on this box.
<davidlt[m]> Yeah
<rwmjones> ok
<rwmjones> I'm going to boot nufive first to check this all works before disassembling the other one
<rwmjones> booting ..
<rwmjones> ok boot good, growpart good, resize2fs good, mask tmp.mount good
<rwmjones> rebooting again
<rwmjones> doesn't reboot on its own
<rwmjones> also I notice now that I previously disabled the big fan ...
<davidlt[m]> reboot command?
<rwmjones> yup
<davidlt[m]> Yeah, that didn't work at that point.
<davidlt[m]> It requires newer OpenSBI and newer kernel, and PMIC being enabled.
<rwmjones> ok, let me plug the fan back in and reboot it via the front panel
<davidlt[m]> It doesn't do a proper reset, but we kinda mimic it.
<davidlt[m]> Basically we power down all the rails, but not PMIC itself and then bring everything up.
<rwmjones> actually the fan _is_ plugged in, but doesn't spin
<davidlt[m]> So you poweroff and RTC will ping and bring it back up.
<davidlt[m]> The proper reset works, but it requires two commands. You need to setup RTC to wake up the system IIRC.
<rwmjones> anyway it seems to have worked (except the fan) so I'm going to do the other one
<davidlt[m]> That's the only way to do a "proper" reset.
<davidlt[m]> As I said only the SoC fan is directly wired to 12VDC.
<davidlt[m]> The rest require software to initialize PWM to get the fans spinning.
<davidlt[m]> The new U-Boot (already built) does that.
<rwmjones> I see "chassis fan" and "asm fan"
<rwmjones> I'm plugged into chassis fan at the moment which is doing nothing
<davidlt[m]> and SoC, that small fan.
<davidlt[m]> Yeah, because it's all random :)
<rwmjones> oh I see, you mean the fan on the SoC already
<rwmjones> yes that's spinning
<davidlt[m]> As I said the hardware doesn't have initial state :) It'
<rwmjones> but the chassis fan isn't (using my added noctua fan)
<rwmjones> ok so it'll be kind of random if the fan comes up?
<davidlt[m]> It's up to U-Boot to initialise those something meaningful.
<davidlt[m]> Until you get a new image --> yes.
<rwmjones> did sifive spend more than 2 cents on the SoC fan or do we expect these ones to fail too :-?
<davidlt[m]> The SoC fan is the only directly wired to 12VDC and thus does not depend of FW/SW.
<davidlt[m]> *on
<davidlt[m]> Just remove it :)
<davidlt[m]> Let's not discuss those tiny fans, just remove it :)
<davidlt[m]> Plug your big fan to the 2-pin header and that'
<davidlt[m]> s it for now :)
<rwmjones> oh I see what you're talking about, let me try that one ...
<davidlt[m]> It's 2-pin, but technically all 3 fan headers have just 2-pins properly wired.
<davidlt[m]> Any of these 3 can drive a proper PC fan.
<rwmjones> ok that's better .. do you think I should physically remove the SoC fan to get enough cooling?
masami has joined #fedora-riscv
<davidlt[m]> I haven't yet, but I will do that :)
<davidlt[m]> I have 2 Noctua fans thus I am lazy to remove them yet.
<rwmjones> I'm guessing there's no throttling on the SoC? it'll just halt and catch fire if it's not cooled properly?
<davidlt[m]> From my experience it will hang.
<rwmjones> ok
<davidlt[m]> It took heavy load for ~40 minutes for the system to become unresponsive.
<rwmjones> ok I'm going to do nujive now
<davidlt[m]> I actually ran the system with main fan disabled in a case for 1-2 weeks with just one 120mm fan :)
<davidlt[m]> It never reached the temperature to crash :)
<davidlt[m]> The newer U-Boot configured shutoff temp to be 80C
<davidlt[m]> Not sure if I had the patch applied to this particular U-Boot version, but it's in a new one.
<davidlt[m]> The sensor default is 125C.
<davidlt[m]> Technically sensors will be able to show the SoC temperature.
<davidlt[m]> (not yet)
<davidlt[m]> I am going to make those certificates for you
<rwmjones> bring up nujive now
<davidlt[m]> sending you the files in a minute
<davidlt[m]> rwmjones: sent it
<davidlt[m]> and entries created in koji server
<davidlt[m]> so technically you should be able to connect them
<davidlt[m]> there are two issues: I don't have firewalls rules updated and autorpmspec is still disabled, that's something to figure out later on once a new image is ready
bkeys has joined #fedora-riscv
<rwmjones> davidlt[m]: ok, give me a few mins to set up user accounts etc
<rwmjones> both machines are up now
<rwmjones> btw I've disconnected the unleashed boards so hope this works!!
<davidlt[m]> Once they are online I can send some food to test them out
<rwmjones> davidlt[m]: can you see them?
<davidlt[m]> yes, seems to be connected and ready according to koji
<rwmjones> send them some jobs?
<rwmjones> also, five & jive are gone now
<davidlt[m]> okay, let's try to see if there is any python package
<davidlt[m]> ok, some jobs are showing up on koji now
<davidlt[m]> so far unclaimed by the boards
<davidlt[m]> takes a bit of time
davidlt has joined #fedora-riscv
<davidlt[m]> both boards are now producing SRPMs
<davidlt[m]> djdelorie: nirik : if you want to join, you can pick hostname for the boards (does not have to a real thing, just needed for certificate CN)
<rwmjones> I see dnf running
<davidlt[m]> Yeah, my board was capable of doing 3 SRPMs at once!
pierosimonet has joined #fedora-riscv
pierosimonet has quit [Ping timeout: 255 seconds]
pierosimonet has joined #fedora-riscv
pierosimonet has quit [Ping timeout: 246 seconds]
pierosimonet has joined #fedora-riscv
<davidlt[m]> rwmjones: you boards are producing RPMs :)
<davidlt[m]> well, at least building :)
<davidlt[m]> Oh, seems I need to do some dust removal from my unmatched
<rwmjones> hopefully faster than the old boards
<davidlt[m]> not very much
<davidlt[m]> main gains will be from DRAM and faster IO (NVMe)
<rwmjones> there's no a temperature sensor on the SoC or board?
<rwmjones> oh the PCI adapter does
<davidlt[m]> there is
<davidlt[m]> the DT part and sensors config is missing in Fedora
<rwmjones> I'm apparently getting readings from something
<rwmjones> could be the NVMe drive?
<rwmjones> they look plausible:
<rwmjones> nvme-pci-0600
<rwmjones> Adapter: PCI adapter
<rwmjones> Composite: +43.9°C (low = -273.1°C, high = +84.8°C) (crit = +84.8°C)
<rwmjones> Sensor 1: +43.9°C (low = -273.1°C, high = +65261.8°C)
<rwmjones> Sensor 2: +49.9°C (low = -273.1°C, high = +65261.8°C)
<davidlt[m]> yeah
<davidlt[m]> that's NVMe
<davidlt[m]> there are in total 3-4 senors maybe
<davidlt[m]> PMIC has one IIRC, SoC and MB
<davidlt[m]> NVMe reports too
<davidlt[m]> Again, stuff for a new disk image
<rwmjones> it'd be nice to be able to monitor the SoC since I'm not using the supplied fan
pierosimonet has quit [Remote host closed the connection]
<djdelorie> davidlt[m]: unmatched.delorie.com
<davidlt[m]> djdelorie: ok
<davidlt[m]> djdelorie: sent you an email, entry created in koji instance, should be good to go
<djdelorie> got it. Will work on it this afternoon. Thanks!
<davidlt[m]> djdelorie: remember to grow rootfs partition to full drive and disable /tmp on tmpfs
<davidlt[m]> systemctl mask tmp.mount
<djdelorie> /dev/nvme0n1p4 457G 33G 420G 8% /
<davidlt[m]> p4? :) did you have additional partitions?
<davidlt[m]> The NVMe image has two partitions only (boot and root)
<djdelorie> it's an old install
<djdelorie> 5.14.16-101.0.riscv64.fc33.riscv64 - if I need to upgrade, point me at a new image...
<davidlt[m]> I am working on a new one to replace it.
<djdelorie> looks like I'm running Fedora-Developer-Rawhide-20211107.n.0-sda.raw
<djdelorie> no, I have your image locally too
<djdelorie> IIRC I manually built it up from the sdcard image
<davidlt[m]> Basically all instructions are in today's IRC chat above.
<davidlt[m]> But in general firmware goes into SD card, rootfs goes into NVMe drive.
<djdelorie> $ uptime
<djdelorie> 10:47:58 up 171 days, 15:37, 2 users, load average: 0.02, 0.01, 0.00
<djdelorie> yeah, it's all set :-)
<davidlt[m]> well that's impressive :)
<djdelorie> I set it up when I got it but you were about to move the koji instance so we never hooked it in
<davidlt[m]> we are still to move it :)
<davidlt[m]> there are so many moving parts :)
<djdelorie> ha!
<rwmjones> djdelorie: I just installed it if you need help
<djdelorie> davidlt[m]: installed and running
<davidlt[m]> Interesting
<davidlt[m]> I don't see all the messages on my matrix client
<davidlt[m]> e.g. the last djdelorie messages didn't show up :/
<djdelorie> davidlt[m]: ping?
<davidlt[m]> Yeah, don't see it on matrix :)
<davidlt> but I see it on IRC client :)
<djdelorie> I won't take it personally
<davidlt> I wonder if messages will show up later
<davidlt> heck
<davidlt> I don't even see my messages :)
<davidlt> I guess the whole channel is dead :)
<nirik> IRC is fine. Matrix is fine. The bridge between them is lagging by like forever...
<davidlt> nirik, just do double check, are you leading fedora infra team?
<nirik> yes, thats me. ;)
<davidlt> I have to run, but I have questions ;)
<davidlt> I will ping you later
<nirik> sure, happy to help. :) I've been swamped of late, but I do hope to help move riscv forward...
jcajka has quit [Quit: Leaving]
<rwmjones> nirik: if you have a hifive unmatched (hosted anywhere, even without incoming connections) it's possible to connect it to the koji instance
<rwmjones> dj, david & I have several you can see in the hosts list
<rwmjones> fedora.riscv.rocks/koji/hosts?start=150&state=enabled&order=name
<rwmjones> you'll have to ask david to set up a key for you
<nirik> yeah, I still need to connect the headers and put a power supply in it and get it installed... but then I will. ;)
<rwmjones> I can help with any of those things, done it for two unmatched now
<nirik> ok, I just need time. ;) perhpas tomorrow
davidlt has quit [Ping timeout: 240 seconds]
cmuellner has quit [Ping timeout: 258 seconds]