piter75

  • Posts

    293
  • Joined

  • Last visited

Everything posted by piter75

  1. +100 😉 @wureka given this report it may either work or not, or... work intermittently. I don't have any sm2262en(g) based drive to verify...
  2. Armbian "current" (5.10.y) compiles without issues. I second @Igor's opinion that a change somewhere in this diff broke the eMMC. I tried reverting a few obvious parts of it, like the mmc driver changes, but without success. However I did find that with the unit I have the issue happens only in hs400{,es} modes. With those disabled my unit works fine and I can use nand-sata-install to transfer os from SD to eMMC successfully which is not possible with 5.10.60+ and hs400 enabled on eMMC. If anyone wants to check if switching eMMC to hs200 mode works also on their unit, here is how: Upgrade the kernel to 5.10.60, but don't reboot yet. Run: curl -o rk3399-kobol-helios64.dtb https://users.armbian.com/piter75/helios64/rk3399-kobol-helios64.dtb sudo cp rk3399-kobol-helios64.dtb /boot/dtb/rockchip/rk3399-kobol-helios64.dtb sudo reboot If this workaround works I will disable hs400{,es} (again) in Armbian until the underlying issue is found. There will be a performance penalty to that change but keep in mind that Helios64 was originally released with hs200 and only recently gained hs400 back ;-) Below you can find the comparison between hs400 and hs200 modes using iozone.
  3. You have probably used the option "Boot from SPI - system on SATA, USB or NVMe" which transfers the current system to SSD and needs to clean it. There is another option - better suited for your case - "Install/Update the bootloader to SPI Flash". It does not touch the existing partitions only writes to SPI Flash.
  4. @TCB13 I am a bit late to the party but I figured it may still be needed ;-) spi-spidev is a special / dynamic overlay. Besides enabling it you need to also configure it and armbian-config cannot currently do that. For configuration options have a look here: https://github.com/armbian/build/blob/master/patch/kernel/archive/rockchip64-5.10/general-rockchip-overlays.patch#L98-L126 You can also consult the local README file located in: /boot/dtb/rockchip/overlay/README.rockchip-overlays
  5. It will be fixed with https://github.com/armbian/build/pull/2877
  6. @Chalix You have done quite a research already Helios64 is Rockchip RK3399 based so you are better off with the theory found here http://opensource.rock-chips.com/wiki_Boot_option as a starter. U-boot and vendor blob binaries are found in "/usr/lib/linux-u-boot-$BRANCH-helios64_$RELEASE_arm64" folder on your system. You are probably using "current" $BRANCH and the latest $RELEASE is 21.02.3. The recipe to write the images to device (more or less) is to be found here.
  7. Something similar to this hat most probably ;-) https://www.aliexpress.com/item/33029451117.html
  8. With that setting you are actually disabling the DVFS after the short period of time during boot - before cpufrequtils kicks in. My experience is that you could also set it to min/max 2GHz and it would be just as good. Yes, this mod/tweak/hack ;-) fixes the behaviour of my boards with ondemand governor. It seems that the instability of little core's voltage during the voltage change was the issue. Rockchip has already limited max change per single step for this regulator to 100mV (because of "overshooting") and I did limit it further to 50mV (75mV was still unstable). It takes a bit longer to switch between frequencies now (at least 456uS vs at least 196uS for full swing between 408MHz and 1512MHz) but it is stable.
  9. +1 We need to do that from time to time. I am not particularly missing them ;p
  10. I am not sure of that. My understanding is that we are still building release images from master branch and the removal of this line from targets.conf: -rockpi-4b legacy focal cli stable yes means that focal legacy image for rockpi-4b will not be built. Lack of focal legacy image among 20.02.3 release images (built after the referred commit was merged) seems to corroborate that theory ;-)
  11. IMHO it's more than that ;-) There will be older releases in the archive but no new ones.
  12. Not that it is relevant to SPI/NVMe booting ;-) but it is no longer built for that combination automatically. You can build it yourself or download focal current and switch kernel to legacy with "armbian-config".
  13. @Salvador Liébana I would be very cautious about comparing with results that I don't know how they were obtained. In fact I would avoid it ;-) We cannot say much about Radxa's benchmarking looking at the presented graph. Nonetheless below you will find the iozone run I performed with LK 5.10.20: piter@rockpi-4c:~$ uname -a Linux rockpi-4c 5.10.20-rockchip64 #1 SMP PREEMPT Fri Mar 5 10:47:39 CET 2021 aarch64 GNU/Linux It was performed using iozone (429) on EXT4 with ROCK Pi 4C and Corsair Force MP510B 480GB (yes, the model with degraded flash chips): As you can see it reaches 1GB+/s speeds both writing and reading with large enough block size, hovers around 1GB/s with medium block size and dives to 50-200MB/s with small blocks. pcie-gen2 overlay is not needed with ROCK Pi 4C as unsupported gen2 link speed was mainlined for all ROCK Pi 4 boards. BTW. Is the guy Jeff Geerling (https://github.com/geerlingguy)? If so then I am actually using the way he was using to compare different SD cards' performance with Raspberry Pi ;-) https://github.com/geerlingguy/raspberry-pi-dramble/issues/7
  14. RK3399 officially supports version 1.0 (2.5GT/s) and unofficially 2.0 (5GT/s). PCIE 3.0 support is needed for 8GT/s. https://en.wikipedia.org/wiki/PCI_Express#History_and_revisions
  15. True, it makes no difference because the unsupported by Rockchip Gen 2 link speed was mainlined into ROCK Pi 4(a/b/c) device trees so you don't need the overlay ;-) https://github.com/torvalds/linux/blob/master/arch/arm64/boot/dts/rockchip/rk3399-rock-pi-4.dtsi#L472 I am not sure it should be mainlined this way but it's another story. It confirms that you are indeed running at Gen2 speeds and have 1GiB/s physical bandwidth.
  16. Glad to hear that. BTW The fix is now included in Armbian v21.02.3 images
  17. Thanks for sharing drive details! I will add it to the opening post.
  18. Great. Looking forward to hear about your results.
  19. Great. Will it cover all the boards? If not please add NanoPi M4V2 images to the build. It also contains a long overdue stability issue fixed in current and dev.
  20. I am still thinking that this "fix" is actually only a workaround covering for some other issue. It sets "rootdev" to "/dev/nvme0n1p1" device but it should be overridden anyway with variables loaded from /boot/armbianEnv.txt. It also sets "ubootpart" in the kernel command line but this is only used by Armbian scripts for writing u-boot later on. IMHO it has nothing to do with the boot process. With those two facts combined - either /boot/armbianEnv.txt is not loaded correctly and does not use "rootenv=UUID=xxx" (I see in your armbianmonitor log that the variable is properly configured) or the fix is simply a placebo effect and it "works" because something else has also changed in the system. If you look at the last full dmesg of http://ix.io/2Rtz you see that the kernel command line uses "root=UUID=94xxx" which actually confirms that the fix is actually ineffective. If you look at the lsblk part of the log after this dmesg it shows that the system actually booted with nvme. What I suspect is that your drive is not always recognised by kernel. That would explain the missing UUID=5xxxx message in your screenshot. If you look at the screenshot there are actually no nvme* devices present in "ls /dev" output - that would corroborate my suspicion. What I would try would be a bit of testing: cold booting and rebooting at least a few times with SD card present and verify if the nvme device is actually present after booting.
  21. No problem Thanks for the testing of mdadm scenarios. I decided this was good enough (+ my memtester tests) and also got some time to verify if the fix did not affect other boards. Since it is merged it means that v21.05 should finally run stable on M4v2 in mainline... it may also be sooner if there is another revision of v21.02
  22. You are right. It does not depend on the way the board is powered. My electronics education is long diminished after spending years in software industry but I suppose it's either issue with rk808 pmic or the low pass filter in its output circuitry. The former is less likely as there are quite a few rk3399/rk808 boards that do not exhibit these issues so I guess it's more likely to be the output filter issue.
  23. Known issue with hdmi enabled in u-boot and rk3399. NanoPi M4V2 is long known to be unstable in current and dev branches - you can search the forum. Fortunately there is hope ;-) Try the image referred in the following message and report your results. It contains both stability fix and the "lying around patch" to disable hdmi in u-boot.
  24. @i5Js you can download these packages and install them with apt/dpkg: https://users.armbian.com/piter75/nanopim4v2-stability-fix/ You need both of them for the fix. When they are correctly installed and the board is rebooted you should see this message in your dmesg: piter@nanopim4v2-4:~$ dmesg | grep rk808-regulator.*buck [ 2.840331] rk808-regulator rk808-regulator: max buck steps per change: 4 The last "4" means you have the fix. No message or "8" means you don't have a fix.
  25. Could you test this image with your procedure? It should fix the issues with voltage scaling on little cores cluster that made my units unstable and fail in the first loop of "memtester 3280M". With the fix it run 150 loops without failures.