Hi, A small update regarding the SpacemiT K3 testing.
On 2026-06-24 19:34, Aurelien Jarno wrote: > On 2026-06-24 14:06, Emilio Pozuelo Monfort wrote: > > My preference would be to have some K3 boards, even if that means waiting a > > bit longer, so that we can further cut build time for large packages. > > I have been able to progress a bit on that front, thanks to SpacemiT, I > got access to a K3 board. On the good side, the performance are > impressive compared to the Unmatched boards we currently use, both in > terms of CPU and NVMe speed. I measured a speed improvement of up > between 6 to 8 compared to the current buildds. For instance the linux > package from experimental builds in 2h30. Here are a few more values: openjdk-25 3h41 qemu 0h52 rustc 3h40 scipy 2h44 > On the bad side upstream kernel support is not fully there, although I > am currently running a 7.1 kernel with ~50 patches, that translates to > ~25 patches against the future 7.2 kernel. Let's hope most of them can > be merged for 7.3. > > There are also a few additional issues that I have found, and which will > need to be fixed before we can consider using SpacemiT K3 based hardware > as buildd. I have already sent two fixes [1] [2], but there is also at > least reboot support The reset support can be added by backporting support for the SpacemiT P1 PMIC [1] [2] from OpenSBI v1.9 to the vendor version of OpenSBI (based on v1.6). > and a system crash with vector code that need to be > fixed. The system crash is fixed by this patch [3] that has been sent upstream. Regards Aurelien [1] https://github.com/riscv-software-src/opensbi/commit/b10e18ec854a80e938a1fc8cf23beb8f8577cbaf [2] https://github.com/riscv-software-src/opensbi/commit/d9637d00bf47a3cc17306fc62e71652c8a83c289 [3] https://lore.kernel.org/spacemit/[email protected]/ -- Aurelien Jarno GPG: 4096R/1DDD8C9B [email protected] http://aurel32.net
