Hi Alexey, On 6/8/2026 3:56 PM, Alexey Charkov wrote: > On Mon, Jun 1, 2026 at 3:45 PM Jonas Karlman <[email protected]> wrote: >> >> Hi Alexey, >> >> On 6/1/2026 11:52 AM, Alexey Charkov wrote: >>> On Sun, May 31, 2026 at 10:26 PM Simon Glass <[email protected]> wrote: >>>> >>>> Hi Alexey, >>>> >>>> On Mon, 25 May 2026 at 01:57, Alexey Charkov <[email protected]> wrote: >>>>> >>>>> On Sat, May 23, 2026 at 4:41 AM Simon Glass <[email protected]> wrote: >>>>>> >>>>>> Hi Alexey, >>>>>> >>>>>> On Fri, 22 May 2026 at 00:24, Alexey Charkov <[email protected]> wrote: >>>>>>> >>>>>>> Hi Simon, >>>>>>> >>>>>>> On Fri, May 22, 2026 at 3:22 AM Simon Glass <[email protected]> wrote: >>>>>>>> >>>>>>>> Hi, >>>>>>>> >>>>>>>> I am trying to make Friendlyelec NanoPi M5 (RK3576) work with >>>>>>>> mainline. I found the note about needing a workaround so have tried >>>>>>>> kwiboo/rk3576 at >>>>>>>> >>>>>>>> https://github.com/Kwiboo/u-boot-rockchip.git >>>>>>>> >>>>>>>> I am building like this: >>>>>>>> >>>>>>>> $ >>>>>>>> ROCKCHIP_TPL=~/dev/rkbin/bin/rk35/rk3576_ddr_lp4_2112MHz_lp5_2736MHz_v1.09.bin >>>>>>>> BL31=~/dev/rkbin/bin/rk35/rk3576_bl31_v1.20.elf um build >>>>>>>> nanopi-m5-rk3576 >>>>>>> >>>>>>> FWIW, upstream TF-A master branch works well as BL31 on RK3576, though >>>>>>> it's most probably unrelated to the issue you're seeing >>>>>> >>>>>> OK, thanks. >>>>>> >>>>>>> >>>>>>>> Image 'simple-bin' is missing optional external blobs but is still >>>>>>>> functional: tee-os >>>>>>>> >>>>>>>> /binman/simple-bin/fit/images/@tee-SEQ/tee-os (tee-os): >>>>>>>> See the documentation for your board. You may need to build Open >>>>>>>> Portable >>>>>>>> Trusted Execution Environment (OP-TEE) and build with >>>>>>>> TEE=/path/to/tee.bin >>>>>>>> >>>>>>>> Image 'simple-bin-spi' is missing optional external blobs but is still >>>>>>>> functional: tee-os >>>>>>>> >>>>>>>> /binman/simple-bin-spi/fit/images/@tee-SEQ/tee-os (tee-os): >>>>>>>> See the documentation for your board. You may need to build Open >>>>>>>> Portable >>>>>>>> Trusted Execution Environment (OP-TEE) and build with >>>>>>>> TEE=/path/to/tee.bin >>>>>>>> >>>>>>>> $ dd if=/tmp/b/nanopi-m5-rk3576/u-boot-rockchip.bin of=/dev/sda seek=64 >>>>>>>> 18579+0 records in >>>>>>>> 18579+0 records out >>>>>>>> 9512448 bytes (9.5 MB, 9.1 MiB) copied, 1.99658 s, 4.8 MB/s >>>>>>>> >>>>>>>> I get no serial output at all booting from SD card. Is there a special >>>>>>>> trick I am missing? I have confirmed that the Alpine Linux image works >>>>>>>> OK. >>>>>>> >>>>>>> Dumb question: did you flip the switch on the back side of the board >>>>>>> from FSPI1 to UFS/SD? >>>>>> >>>>>> Yes it is set to UFS/SC. >>>>>>> >>>>>>> If yes, then having no serial output at all most likely means the >>>>>>> boost.bin trick from Jonas' WIP patch didn't apply, because the DDR >>>>>>> trainer unconditionally spits out some logs on the debug UART during >>>>>>> early init. >>>>>>> >>>>>>> The SD card might also be finicky, so if you happen to have an SPI >>>>>>> flash equipped board then you can flash U-boot there instead via >>>>>>> Maskrom. Just remove all other storage, connect a USB A-A cable to the >>>>>>> top USB 3.0 port, and power it up while holding the maskrom button. >>>>>> >>>>>> I am not sure how to power the board other than through the >>>>>>> >>>>>>> Then: >>>>>>> rockusb download-boot rk3576_loader_fspi1_v1.13.100.bin >>>>>>> # note the zero below, because u-boot-rockchip-spi.bin already has a >>>>>>> 64-sector offset baked in, unlike u-boot-rockchip.bin: >>>>>>> rockusb write-file 0 u-boot-rockchip-spi.bin >>>>>>> rockusb reset-device >>>>>> >>>>>> With this I was able to get it to SPL: >>>>>> >>>>>> DDR 2f85f4b2d4 cym 24/11/07-19:07:28,fwver: v1.09 >>>>>> In >>>>>> ch0 ttot6 >>>>>> ch1 ttot6 >>>>>> ch0 ttot7 >>>>>> LPDDR5, 2736MHz >>>>>> channel[0] BW=16 Col=10 Bk=16 CS0 Row=16 CS=1 Die BW=16 Size=1536MB >>>>>> ch1 ttot7 >>>>>> channel[1] BW=16 Col=10 Bk=16 CS0 Row=16 CS=1 Die BW=16 Size=1536MB >>>>>> Manufacturer ID:0xff >>>>>> CH0 RX Vref:24.1%, RX DQS Vref:29.6%, TX Vref:19.0%,0.0% >>>>>> DQ roc: >>>>>> p4 n2, p6 n7, p5 n0, p1 n0, p0 n0, p0 n4, p3 n0, p7 n0, p3 n0, >>>>>> p4 n2, p5 n0, p2 n2, p0 n2, p7 n0, p2 n0, p6 n0, p7 n0, p6 n4, >>>>>> >>>>>> DQ rds:l0 l0 h1 l0 l0 l0 l0 l0, l0 l0 h1 l0 l0 l0 l0 l0 >>>>>> DQS roc: p2, n0, p1, n0 >>>>>> >>>>>> CH1 RX Vref:25.7%, RX DQS Vref:29.2%, TX Vref:20.0%,0.0% >>>>>> DQ roc: >>>>>> p2 n0, p7 n0, p2 n0, p1 n0, p2 n0, p0 n1, p3 n0, p5 n0, p7 n0, >>>>>> p6 n0, p3 n2, p3 n0, p2 n0, p2 n0, p6 n0, p0 n0, p0 n2, p0 n0, >>>>>> >>>>>> DQ rds:l0 h2 l0 l0 l0 l0 l0 h1, l0 l0 l0 l0 h1 l0 h1 l0 >>>>>> DQS roc: p2, n0, p2, n0 >>>>>> >>>>>> stride=0x3, ddr_config=0x2 >>>>>> hash bank_mask0-3 0x0 0x2100 0x44200 0x88400, rank_mask0 0x0 >>>>>> change to F1: 534MHz >>>>>> ch0 ttot6 >>>>>> ch1 ttot6 >>>>>> change to F2: 1320MHz >>>>>> ch0 ttot8 >>>>>> ch1 ttot8 >>>>>> change to F3: 1968MHz >>>>>> ch0 ttot6 >>>>>> ch1 ttot6 >>>>>> change to F0: 2736MHz >>>>>> ch0 ttot7 >>>>>> ch1 ttot7 >>>>>> out >>>>>> >>>>>> U-Boot SPL 2026.01-00014-gf6f1066339a5 (May 22 2026 - 18:32:58 -0600) >>>>>> Trying to boot from MMC1 >>>>>> Card did not respond to voltage select! : -110 >>>>>> spl: mmc init failed with error: -95 >>>>>> Error: -95 >>>>>> SPL: Unsupported Boot Device! >>>>>> SPL: failed to boot from all boot devices >>>>>> ### ERROR ### Please RESET the board ### >>>>>> >>>>>> >>>>>> It seems to boot from SPI but then try to switch to MMC? >>>>> >>>>> Sifting through my local tree I also noticed I've got the following >>>>> patch to set the pin config for SPI flash on NanoPi M5: >>>>> >>>>> https://github.com/flipperdevices/u-boot/commit/ffb0d6c9b4893d6a8c2fe51efc3dd5914857615e >>>>> >>>>> Would you mind giving it a try? >>>>> >>>>> The node itself and its properties are already in the upstream DTS, so >>>>> it _shouldn't_ really be required, but maybe the bootph-* markers >>>>> didn't propagate for me as-is and I didn't have the willpower to >>>>> investigate further :) >>>>> >>>>>> I also found that if I 'dd if=/tmp/image of=/dev/sda count=2000' from >>>>>> the good Alpine image, it ends up in the U-Boot I built (although of >>>>>> course I have not built the earlier parts). >>>>> >>>>> Another way to get it to boot without relying on a particular storage >>>>> configuration or the prebuilt image is to load an upstream binary >>>>> directly to RAM over Maskrom: >>>>> >>>>> export BL31=[...] ROCKCHIP_TPL=[...] >>>>> make nanopi-m5-rk3576_defconfig rockchip-ramboot.config >>>>> make -j$(nproc) >>>>> rockusb download-sram u-boot-rockchip-usb471.bin >>>>> rockusb download-ddr u-boot-rockchip-usb472.bin >>>>> >>>>> SPI boot is supposed to work fine though. Did you use the >>>>> u-boot-rockchip-spi.bin or u-boot-rockchip.bin image? >>>>> >>>>>>> Loader binary can be assembled from rkbin blobs, or alternatively >>>>>>> taken from e.g. here: >>>>>>> https://dl-linux-images.flipp.dev/u-boot/u%3D1afdc52186635bb599642064ee7c23f539b73e3c/rk%3D495eec7a93220b1269e915e684c5667250462a3d__tfa%3Dfc3691b5022abd9feba6c37330005b7437cccf9f/nanopi-m5/rk3576_loader_fspi1_v1.13.100.bin >>>>>>> >>>>>>> rockusb tool: cargo install rockusb --example rockusb --features=nusb >>>>>> >>>>>> Thanks very much for taking the time to help with this! >>>>> >>>>> Always welcome! >>>> >>>> It was booting from UFS, but since U-Boot SPL didn't have it enabled, >>>> it didn't complete. >>> >>> Indeed, the boot-from-UFS series [1] is still pending Kever's >>> attention for the Rockchip-specific part. Neil merged the common UFS >>> bits a while ago, and I'm using it all the time on several RK3576 >>> boards. >>> >>> [1] >>> https://lore.kernel.org/u-boot/[email protected]/ >>> >>>> So I've erased that and am using your method of booting over USB - BTW >>>> it is very slow, doesn't seem to use bulk mode. >>> >>> I think there was some peculiarity about the boot ROM taking a while >>> to verify the checksum of the usb472 image, rather than the USB >>> transfer itself taking long. It might need further investigation. >> >> The very very slow boot time using ramboot on rk3576, around 10 min vs >> 30 seconds, is due to changes done in commit c6bba31dbdd4 ("rockchip: >> spl-boot-order: Defer probe of boot device"). >> >> The delay of probe done in that commit correctly result in clk driver >> not being probed in SPL when running ramboot, and because no clk driver >> is probed, cores run at very slow rates until OS change core rates :-) >> >> My ci branch contains a commit to fix this, and other, I just need to >> find some free time before it reaches the list, see commit "WIP: >> rockchip: clk: init clocks in spl for rk3576" at [2]. > > Hi Jonas, it does indeed look like painfully slow clocks, but > cherry-picking this patch alone doesn't seem to turn the tide for me. > It still takes about 6 minutes from the moment `rockusb download-ddr` > completes until there is any SPL output on the console. > > Is there any other change I need to apply for it to take effect?
6 minutes sounds very long and strange, is there some issue with rockusb? I did a quick test in my labgrid using close to v2026.07-rc3 + that patch and loading usb471.bin + usb472.bin takes around 2.1 seconds, and around 13-15 seconds from DDR init until SPL banner is printed. INFO StepLogger: → RKUSBMaskromDriver.load(filename=None) INFO SSHConnection(h: Created new SSH connection to noble INFO ManagedFile(loc: Synchronizing /srv/u-boot/rk3576-sige5/u-boot-rockchip-usb471.bin to noble INFO ManagedFile(loc: Synchronizing /srv/u-boot/rk3576-sige5/u-boot-rockchip-usb472.bin to noble INFO StepLogger: ← RKUSBMaskromDriver.load() [2.134s] No real change there, what took long before that commit was starting Linux, kernel boot could take around 5-10 minutes, now just 15-30 seconds, if I trust dmesg timestamps. I use labgrid PR1721 [3] to ramboot into my boards, when working on that I noticed that use of pre-calculated crc tables helped loading speed a lot, maybe rockusb is at fault or help slow things down for you? My test run above was on a ArmSoM Sige5 board. [3] https://github.com/labgrid-project/labgrid/pull/1721 Regards, Jonas > >> [2] https://source.denx.de/u-boot/contributors/kwiboo/u-boot/-/commits/ci > > Thanks a lot, > Alexey
