[PATCH v3 0/4] Add Tegra238 BPMP support and fix stale EMC clock rate

[Johnny Liu <johnliu@nvidia.com>]

EMC clock scaling is handled by the BPMP firmware without kernel
involvement, so clk_get_rate() on the EMC clock can return a stale value
cached by the common clock framework. This series makes the EMC clock
rate uncached on Tegra234, Tegra238 and Tegra264, and adds the pieces
needed to identify the EMC clock on each of these chips.

The EMC clock ID differs per chip (TEGRA234_CLK_EMC=31,
TEGRA238_CLK_EMC=43, TEGRA264_CLK_EMC=290), while the
"nvidia,tegra186-bpmp" fallback compatible and the tegra186_soc data are
shared across several chips. To avoid detecting the chip at runtime, the
EMC clock ID is carried in per-SoC BPMP data selected by the device tree
compatible.

Patch 1 adds the "nvidia,tegra238-bpmp" compatible to the binding,
together with the Tegra238 clock ID header. Patch 2 adds an emc_clk_id
field to struct tegra_bpmp_soc, together with dedicated SoC data for
Tegra234 and Tegra264 built on a common TEGRA186_BPMP_SOC_COMMON
initializer. Patch 3 adds Tegra238 support to the BPMP driver, including
its own SoC data and EMC clock ID. Patch 4 then sets
CLK_GET_RATE_NOCACHE on the clock matching that ID, so that the common
clock framework forwards rate queries to the BPMP instead of returning a
stale cached rate.

The series touches both the clock and the firmware code. Patch 4 depends
on patch 2, and patch 3 depends on patches 1 and 2, so I would suggest
taking the whole series through the Tegra tree with an Acked-by from the
clock maintainers.

Changes in v3:
- Move the TEGRA186_BPMP_SOC_COMMON refactoring out of the Tegra238
  support patch and into the "Add per-SoC EMC clock ID" patch, as the
  refactoring only makes sense together with the per-SoC emc_clk_id.
- Keep all Tegra238 specific changes (Makefile, CONFIG_ARCH_TEGRA_238_SOC
  guards, clock ID header include, tegra238_soc data and compatible match)
  in a single "Add Tegra238 support" patch, and reorder it after the
  per-SoC EMC clock ID patch.
- Reword the "Add per-SoC EMC clock ID" commit message to describe the
  intended use of emc_clk_id: the BPMP clock provider sets
  CLK_GET_RATE_NOCACHE on that clock, as it is completely managed by the
  firmware and the common clock framework must pass rate queries through
  to get the latest rate.
- No changes to patches 1 and 4 apart from the patch numbering.

Changes in v2:
- Squash "dt-bindings: clock: Add Tegra238 BPMP clock IDs" into
  "dt-bindings: firmware: tegra186-bpmp: Add Tegra238 compatible".
- Reword the binding commit message to state that the Tegra238 SoC is
  fully compatible with the Tegra186 BPMP.

v2: https://lore.kernel.org/all/20261028155000.1205739-1-johnliu@nvidia.com/
v1: https://lore.kernel.org/all/20261028084900.1047287-1-johnliu@nvidia.com/

Johnny Liu (4):
  dt-bindings: firmware: tegra186-bpmp: Add Tegra238 BPMP
  firmware: tegra: bpmp: Add per-SoC EMC clock ID
  firmware: tegra: bpmp: Add Tegra238 support
  clk: tegra: bpmp: Do not cache the EMC clock rate

 .../firmware/nvidia,tegra186-bpmp.yaml        |   1 +
 drivers/clk/tegra/clk-bpmp.c                  |   3 +
 drivers/firmware/tegra/Makefile               |   1 +
 drivers/firmware/tegra/bpmp.c                 |  61 ++--
 include/dt-bindings/clock/nvidia,tegra238.h   | 279 ++++++++++++++++++
 include/soc/tegra/bpmp.h                      |   1 +
 6 files changed, 329 insertions(+), 17 deletions(-)
 create mode 100644 include/dt-bindings/clock/nvidia,tegra238.h

-- 
2.43.0