LoongArch CPU Port Might Still Land For Linux 5.19

One of the open questions this merge window is whether the MIPS64-based LoongArch CPU architecture port of the kernel will manage to land for the Linux 5.19 cycle. There has been a discussion this holiday weekend by upstream kernel developers and looks like it may still land, but possibly without necessary hardware drivers included.

This LKML thread has the latest discussion over mainlining LoongArch, the Chinese CPU ISA developed by Loongson after their long-running MIPS64-based designs. With MIPS64 being a dead end and looking to boost their domestic CPU industry, LoongArch was developed and inspired by MIPS64 and RISC-V. In some areas of the kernel code at least, it's meant from the software perspective largely duplicating existing MIPS64 code.

LoongArch has gone through 10+ rounds of review for getting the Linux kernel support in place. One of its requirements was met in GCC 12 now having a LoongArch port for having the necessary compiler support in place before the kernel CPU architecture code can be merged.

But as detailed in that aforelinked kernel thread, there is an issues with the LoongArch PCI code not yet having the reviewed-by/signed-off tags in place. There is also an issue with the irqchip driver not passing review due to its non-standard way of integrating into ACPI and PCI. LoongArch's ACPI handling is a bit hairy but is being addressed with the next ACPI standards update. In any event the kernel developers are determining best how to proceed.

At least from the latest talk, it's sounding like the CPU port/architecture code may at least be sent in this week for the Linux 5.19 merge window while the necessary hardware drivers to boot such a system would land in a latest kernel release if necessary.

Getting the CPU port introduced at least for Linux 5.19 would unblock LoongArch developers for being able to submit their GNU C Library (Glibc) support upstream. Mainlining their Glibc support is contingent on first having the kernel support in place with a solidified user-space ABI. The developers are hoping to see LoongArch merged into Glibc 2.38 that is releasing later this summer and for that to happen they would need that arch code into Linux 5.19.

So we'll see how this plays out over the coming days particularly around the LoongArch drivers, but it's possible just the CPU/arch code will be merged this cycle while not actually able to boot a working LoongArch system until those necessary drivers land in a later kernel cycle.