5 ms·

The really nice thing is that the RK3399, while an old and not particularly fast ARM, has almost entirely open documentation. A bit of googling should land both

The really nice thing is that the RK3399, while an old and not particularly fast ARM, has almost entirely open documentation. A bit of googling should land both parts of the TRM, which covers every bit of the SOC's registers and devices minus the Mali. The follow-on RK3588 announced last year with 4x A76's (and 4x A55s) in 8nm should completely crush the RK3399 performance wise. If it comes with open documentation too, it could be a killer hacker chip this year. (just to add to that) The RK3399 has 2x A72's with a peak frequency of 2Ghz and 4 A53's. The RPi's 4x A72s easily run at 2Ghz as well, given a heatsink. For most workloads the much faster A72s are a big advantage over the A53s.

sneak 7y ago
Does it use any proprietary blobs to boot? I have yet to see a GPU without them, and you said "almost", so I assume the answer is yes?
- StillBored 7y ago
  I don't think there are any binary blobs in the normal atf+uboot+kernel boot path. The display will run without the mali, so. AFAIK it can be run entirely open source. There could be some additional hidden burned in place firmware to handle the early SPI/etc boot flow, but they actually document the cortex M0 mgmt processor as well. So that puts them ahead of basically everyone. Not all the documentation is "good" the DDR controller registers are in the docs AFAIK, but lack a theory of operation section. OTOH, there is example code in TFA for that, https://git.trustedfirmware.org/TF-A/trusted-firmware-a.git/tree/plat/rockchip/rk3399/drivers/dram/dfs.c https://git.trustedfirmware.org/TF-A/trusted-firmware-a.git/... So, really as far as I know this is literally the most open A72 platform out there when it comes to docs you can get via google. The MALI being the problem, which means that one could consider one of the reverse engineered QC chips or the rpi better understood (because of the Adreno or vp9 efforts) but even then the documentation quality is worse.
  - zozbot234 7y ago
    If there are no blobs in the "normal" path, why isn't this device supported by mainline Debian without special patches? Is this code making its way into kernel/bootloader upstream so that distros can pick it up?
- LargoLasskhyfv 7y ago
  WRT GPU there is [1] https://bootlin.com/blog/wrapping-up-the-allwinner-vpu-crowdfunded-linux-driver-work/ https://bootlin.com/blog/wrapping-up-the-allwinner-vpu-crowd... The rest i don't know exactly. But all in all this seems like a viable replacement for the obsolete 12/13' ThinkPads coveted in some crowds. Could have more RAM, better Displays, but hey, all Open Source, so this really could be new 'ecosystem' for that. Edit: I know this is for Allwinner, but this applies to other SOCs which embed MALI also. I'm watching this space, and it slowly 'trickles down' there. [2] https://bootlin.com/?s=Mali https://bootlin.com/?s=Mali
  - LargoLasskhyfv 7y ago
    [2] https://www.collabora.com/news-and-blog/blog/2019/12/20/opening-up-mali-t720/ https://www.collabora.com/news-and-blog/blog/2019/12/20/open... [3] https://stikonas.eu/wordpress/2019/09/15/blobless-boot-with-rockpro64/ https://stikonas.eu/wordpress/2019/09/15/blobless-boot-with-...

sneak 7y ago

Does it use any proprietary blobs to boot? I have yet to see a GPU without them, and you said "almost", so I assume the answer is yes?

StillBored 7y ago
I don't think there are any binary blobs in the normal atf+uboot+kernel boot path. The display will run without the mali, so. AFAIK it can be run entirely open source. There could be some additional hidden burned in place firmware to handle the early SPI/etc boot flow, but they actually document the cortex M0 mgmt processor as well. So that puts them ahead of basically everyone. Not all the documentation is "good" the DDR controller registers are in the docs AFAIK, but lack a theory of operation section. OTOH, there is example code in TFA for that, https://git.trustedfirmware.org/TF-A/trusted-firmware-a.git/tree/plat/rockchip/rk3399/drivers/dram/dfs.c https://git.trustedfirmware.org/TF-A/trusted-firmware-a.git/... So, really as far as I know this is literally the most open A72 platform out there when it comes to docs you can get via google. The MALI being the problem, which means that one could consider one of the reverse engineered QC chips or the rpi better understood (because of the Adreno or vp9 efforts) but even then the documentation quality is worse.
- zozbot234 7y ago
  If there are no blobs in the "normal" path, why isn't this device supported by mainline Debian without special patches? Is this code making its way into kernel/bootloader upstream so that distros can pick it up?
LargoLasskhyfv 7y ago
WRT GPU there is [1] https://bootlin.com/blog/wrapping-up-the-allwinner-vpu-crowdfunded-linux-driver-work/ https://bootlin.com/blog/wrapping-up-the-allwinner-vpu-crowd... The rest i don't know exactly. But all in all this seems like a viable replacement for the obsolete 12/13' ThinkPads coveted in some crowds. Could have more RAM, better Displays, but hey, all Open Source, so this really could be new 'ecosystem' for that. Edit: I know this is for Allwinner, but this applies to other SOCs which embed MALI also. I'm watching this space, and it slowly 'trickles down' there. [2] https://bootlin.com/?s=Mali https://bootlin.com/?s=Mali
- LargoLasskhyfv 7y ago
  [2] https://www.collabora.com/news-and-blog/blog/2019/12/20/opening-up-mali-t720/ https://www.collabora.com/news-and-blog/blog/2019/12/20/open... [3] https://stikonas.eu/wordpress/2019/09/15/blobless-boot-with-rockpro64/ https://stikonas.eu/wordpress/2019/09/15/blobless-boot-with-...