[PATCH v6 0/8] clk: sunxi-ng: Add support for Allwinner A733 CCU and PRCM

Enzo Adriano enzo.adriano.code at gmail.com
Mon Oct 12 08:41:16 PDT 2026


Hi Junhui, Vinicius,

I can add a scoped result from a 4 GiB Cubie A7S using the vendor boot
chain. With v6 unchanged, unused-clock cleanup enabled and no
clk_ignore_unused, all eight CPUs reach a RAM-root console.

Ten cycles of 15 seconds idle followed by a 64 MiB O_DIRECT read show
mmc0, bus-mmc0 and ahb-store disabled while idle and enabled during the
read. Two additional 4 MiB direct reads at each offset give matching
hashes. Card-write counters stay zero. GIC remains enabled at a reported
480 MHz; MMC operates at 50 MHz, 4-bit, 3.3 V. The board returns to vendor
Linux after the test. This agrees with the parent-gate behavior Vinicius
reported, on a different board and boot chain.

Correction to my earlier storage description: this is microSD on
SDMMC0, not eMMC. The eMMC footprint on this A7S test unit is unpopulated;
this result says nothing about eMMC on A7A/A7Z variants.

The test kernel is v7.3-rc1 plus the public dependencies and local
pinctrl/MMC prerequisites: IRQ-bank allocation, firmware bias preservation,
a read-only PF supply provider, MMC error handling and two DT pinctrl
retry/parser fixes. ITS is disabled in the board DT because enabling it
still stalls early initialization with this vendor boot chain. The GIC
clock is already marked critical in v6 and remained enabled throughout
these tests. I did not add critical flags to the storage clocks.

A further eight-CPU run read the full 119850 MiB card with O_DIRECT to
EOF, with zero card-write counters, and returned to vendor Linux. This
is a microSD/boot test report, not whole-CCU qualification. I have not
validated LPIs, voltage switching, native PSCI reboot, suspend or other
CCU consumers.

Thanks,
Enzo

Assisted-by: Codex:gpt-6



More information about the linux-arm-kernel mailing list