[PATCH 0/4] MIPS: Add support for the Siflower SF19A2890

[Joseph Telaak via B4 Relay <devnull+joseph.telaak.dev@kernel.org>]

This series adds basic support for the Siflower SF19A2890, a MIPS
interAptiv SoC (two cores, two VPEs each) for home routers, and for the
GL.iNet GL-SFT1200 ("Opal") travel router built on it. It boots to a
shell on the PL011 console with all four CPUs up.

Patch 2 is Chuanhong Guo's patch from January 2025 [1], rebased. Jiaxun
asked then why it does not use MIPS_GENERIC. It does: MACH_SIFLOWER_MIPS
selects MIPS_GENERIC and MACH_GENERIC_CORE and boots through
arch/mips/generic, the way MACH_REALTEK_RTL and MACH_INGENIC_SOC do. What
it avoids is MIPS_GENERIC_KERNEL, which selects MIPS_L1_CACHE_SHIFT_7 and
takes ARCH_DMA_MINALIGN 128 from mach-generic/kmalloc.h, so that with
DMA_NONCOHERENT no kmalloc cache below 128 bytes exists. The interAptiv's
L1 and L2 lines are 32 bytes (the boot log reports both).

On a GL-SFT1200 (128 MB) booted to an idle initramfs, 62 s after boot:

                                       MemFree   Slab   kmalloc-128 objs
  this series (MACH_SIFLOWER_MIPS)     107444    4832   1497
  MIPS_GENERIC_KERNEL                  106442    5252   5538
  MIPS_GENERIC_KERNEL, DMA_MINALIGN 32 107660    4732   1497

(kB, means of two boots.) So the whole difference, about 1 MB, comes
from the 128-byte kmalloc alignment; the larger L1_CACHE_SHIFT costs
nothing measurable. If the generic kernel sized its kmalloc alignment
from the probed cache lines at run time, as arm64 does, this board could
use MIPS_GENERIC_KERNEL, and I can send that change in place of a
platform of its own if you prefer.

The SoC's clock controller has a driver of its own, sent to the clk tree.
Until it is merged, the dtsi describes the clock rates the bootloader
programs with fixed clocks; a later patch replaces them. The pin
controller, GPIO, USB PHY, DWC2 and GMAC drivers go to their own trees.

Patch 1 adds a MAINTAINERS entry for the SoC that also covers those
drivers, by "N: sf19a2890" and one F: line for the GPIO driver, so that
they don't each need to touch MAINTAINERS. Until the GPIO driver lands,
get_maintainer.pl --self-test reports that F: line as matching no file.

Tested on a GL-SFT1200, booted from U-Boot over TFTP: the initramfs
shell on ttyAMA0; four CPUs, with CPUs 1-3 taken offline and back
online; the GIC timer within 0.06 s of a reference clock over 298 s; and
the UART divisors U-Boot programs (IBRD 0x43, FBRD 0x34), which confirm
the 125 MHz UART clock in the device tree.

[1] https://lore.kernel.org/r/20250120015254.124447-1-gch981213@gmail.com

---
Chuanhong Guo (1):
      MIPS: Add support for Siflower MIPS SoCs

Joseph Telaak (3):
      dt-bindings: mips: Add Siflower SF19A2890 based boards
      MIPS: dts: Add Siflower SF19A2890 and GL.iNet GL-SFT1200
      MIPS: configs: Add sf19a2890_defconfig

 .../devicetree/bindings/mips/siflower.yaml         | 26 ++++++
 MAINTAINERS                                        | 11 +++
 arch/mips/Kconfig                                  | 32 ++++++++
 arch/mips/boot/dts/Makefile                        |  1 +
 arch/mips/boot/dts/siflower/Makefile               |  2 +
 .../dts/siflower/sf19a2890-glinet-gl-sft1200.dts   | 27 ++++++
 arch/mips/boot/dts/siflower/sf19a2890.dtsi         | 96 ++++++++++++++++++++++
 arch/mips/configs/sf19a2890_defconfig              | 15 ++++
 arch/mips/generic/Platform                         |  1 +
 arch/mips/include/asm/mach-siflower/kmalloc.h      |  9 ++
 10 files changed, 220 insertions(+)
---
base-commit: 6628e7d4c6b25c8d18f1345e3a20d21829aae7e4
change-id: 20261102-sf19a2890-mips-ab72d99af797

Best regards,
--  
Joseph Telaak <joseph@telaak.dev>