* clk-determine-rate: (120 commits)
clk: microchip: core: remove duplicate roclk_determine_rate()
clk: nxp: Fix pll0 rate check condition in LPC18xx CGU driver
clk: scmi: migrate round_rate() to determine_rate()
clk: ti: fapll: convert from round_rate() to determine_rate()
clk: ti: dra7-atl: convert from round_rate() to determine_rate()
clk: ti: divider: convert from round_rate() to determine_rate()
clk: ti: composite: convert from round_rate() to determine_rate()
clk: ti: dpll: convert from round_rate() to determine_rate()
clk: ti: dpll: change error return from ~0 to -EINVAL
clk: ti: dpll: remove round_rate() in favor of determine_rate()
clk: tegra: tegra210-emc: convert from round_rate() to determine_rate()
clk: tegra: super: convert from round_rate() to determine_rate()
clk: tegra: pll: convert from round_rate() to determine_rate()
clk: tegra: periph: divider: convert from round_rate() to determine_rate()
clk: tegra: divider: convert from round_rate() to determine_rate()
clk: tegra: audio-sync: convert from round_rate() to determine_rate()
clk: fixed-factor: drop round_rate() clk ops
clk: divider: remove round_rate() in favor of determine_rate()
clk: visconti: pll: convert from round_rate() to determine_rate()
clk: versatile: vexpress-osc: convert from round_rate() to determine_rate()
...
The round_rate() clk ops is deprecated, so migrate this driver from
round_rate() to determine_rate() using the Coccinelle semantic patch
on the cover letter of this series.
Signed-off-by: Brian Masney <bmasney@redhat.com>
The round_rate() clk ops is deprecated, so migrate this driver from
round_rate() to determine_rate() using the Coccinelle semantic patch
on the cover letter of this series.
Signed-off-by: Brian Masney <bmasney@redhat.com>
The round_rate() clk ops is deprecated, so migrate this driver from
round_rate() to determine_rate() using the Coccinelle semantic patch
on the cover letter of this series.
Signed-off-by: Brian Masney <bmasney@redhat.com>
Pull clk updates from Stephen Boyd:
"This has been a semi-quiet cycle. The core framework remains unchanged
this time around.
In terms of shiny new code though, we have support for the SpacemiT K1
SoC, Sophgo SG2044, and T-HEAD TH1520 VO clk drivers joining the usual
silicon players like Qualcomm, Samsung, Allwinner, and Renesas.
Surprisingly, the Qualcomm pile was smaller than usual but that is
likely because they put one SoC support inside a driver for a
different SoC that is very similar.
Other than all those new clk drivers there are the usual clk data
updates to fix parents, frequency tables, and add missing clks along
with some Kconfig changes to make compile testing simpler and even
more DT binding conversions to boot.
The exciting part is still the new SoC support like SpacemiT and
Sophgo support though, which really dominate the diffstat because they
introduce a whole new silicon vendor clk driver.
New Drivers:
- Camera clock controller driver for Qualcomm QCS8300
- DE (display engine) 3.3 clocks on Allwinner H616
- Samsung ExynosAutov920 CPU cluster CL0, CL1 and CL2 clock controllers
- Video Output (VO) subsystem clk controller in the T-HEAD TH1520 SoC
- Clock driver for Sophgo SG2044
- Clock driver for SpacemiT K1 SoC
- Renesas RZ/V2N (R9A09G056) SoC clk driver
Updates:
- Correct data in various SoC clk drivers
- Allow clkaN to be optional in the Qualcomm RPMh clock controller
driver if command db doesn't define it
- Change Kconfig options to not enable by default during compile
testing
- Add missing clks in various SoC clk drivers
- Remove some duplicate clk DT bindings and convert some more to
YAML"
* tag 'clk-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/clk/linux: (93 commits)
clk: qcom: gcc-x1e80100: Set FORCE MEM CORE for UFS clocks
clk: qcom: gcc: Set FORCE_MEM_CORE_ON for gcc_ufs_axi_clk for 8650/8750
clk: qcom: rpmh: make clkaN optional
clk: qcom: Add support for Camera Clock Controller on QCS8300
clk: rockchip: rk3528: add slab.h header include
clk: rockchip: rk3576: add missing slab.h include
clk: meson: Do not enable by default during compile testing
clk: meson-g12a: add missing fclk_div2 to spicc
clk: qcom: gcc-msm8939: Fix mclk0 & mclk1 for 24 MHz
clk: rockchip: rename gate-grf clk file
clk: rockchip: rename branch_muxgrf to branch_grf_mux
clk: sunxi-ng: ccu: add Display Engine 3.3 (DE33) support
dt-bindings: allwinner: add H616 DE33 clock binding
clk: samsung: correct clock summary for hsi1 block
dt-bindings: clock: add SM6350 QCOM video clock bindings
clk: rockchip: Pass NULL as reg pointer when registering GRF MMC clocks
clk: sunxi-ng: h616: Add LVDS reset for LCD TCON
dt-bindings: clock: sun50i-h616-ccu: Add LVDS reset
clk: rockchip: rk3036: mark ddrphy as critical
clk: rockchip: rk3036: fix implementation of usb480m clock mux
...
The newly added GRF types introduced kzalloc usage into the rk3528.
At least for the similar rk3576 driver, the kernel-test-robot reported the
missing prototype, which warranted adding a slab.h include.
While it did not complain about the rk3528, so the header might be included
"accidentially" right now, add a real include to make sure we keep it
included in the future.
Fixes: 306d2f5dda ("clk: rockchip: rk3528: Add SD/SDIO tuning clocks in GRF region")
Signed-off-by: Heiko Stuebner <heiko@sntech.de>
Link: https://lore.kernel.org/r/20250515082652.2503063-2-heiko@sntech.de
We now have a number of new branch-types coming from the "General Register
Files" (gates and mmc phase clocks). Their naming as branch_grf_foo is
way nicer, so rename the old branch_muxgrf to a similar scheme.
Signed-off-by: Heiko Stuebner <heiko@sntech.de>
Link: https://lore.kernel.org/r/20250508182752.1925313-2-heiko@sntech.de
The ddrphy is supplied by the dpll, but due to the limited number of PLLs
on the rk3036, the dpll also is used for other periperhals, like the GPU.
So it happened, when the Lima driver turned off the gpu clock, this in
turn also disabled the dpll and thus the ram.
Signed-off-by: Heiko Stuebner <heiko@sntech.de>
Link: https://lore.kernel.org/r/20250503202532.992033-4-heiko@sntech.de
Contrary to how it is implemented right now, this mux is controllable via
a bit in CRU_MUSC_CON (same bit as on rk3128 even) and allows switching
between xin24m and the 480m output of the usb2phy.
So drop the hard-coded fixed-factor clock and implement the correct mux
instead.
Signed-off-by: Heiko Stuebner <heiko@sntech.de>
Link: https://lore.kernel.org/r/20250503202532.992033-3-heiko@sntech.de
These clocks locate in VO and VPU GRF, serving for SD/SDIO controller
tuning purpose. Add their definitions and register them in driver if
corresponding GRF is available.
GRFs are looked up by compatible to simplify devicetree binding.
Signed-off-by: Yao Zi <ziyao@disroot.org>
Link: https://lore.kernel.org/r/20250506092206.46143-4-ziyao@disroot.org
Signed-off-by: Heiko Stuebner <heiko@sntech.de>
Some rockchip SoCs, namely the RK3576, have bits in a General Register
File (GRF) that act just like clock gates. The downstream vendor kernel
simply maps over the already mapped GRF range with a generic clock gate
driver. This solution isn't suitable for upstream, as a memory range
will be in use by multiple drivers at the same time, and it leaks
implementation details into the device tree.
Instead, implement this with a new clock branch type in the Rockchip
clock driver: GRF gates. Somewhat akin to MUXGRF, this clock branch
depends on the type of GRF, but functions like a gate instead.
Signed-off-by: Nicolas Frattaroli <nicolas.frattaroli@collabora.com>
Link: https://lore.kernel.org/r/20250502-rk3576-sai-v3-3-376cef19dd7c@collabora.com
Signed-off-by: Heiko Stuebner <heiko@sntech.de>
The MUXGRF clock branch type depends on having access to some sort of
GRF as a regmap to be registered. So far, we could easily get away with
only ever having one GRF stowed away in the context.
However, newer Rockchip SoCs, such as the RK3576, have several GRFs
which are relevant for clock purposes. It already depends on the pmu0
GRF for MUXGRF reasons, but could get away with not refactoring this
because it didn't need the sysgrf at all, so could overwrite the pointer
in the clock provider to the pmu0 grf regmap handle.
In preparation for needing to finally access more than one GRF per SoC,
let's untangle this. Introduce an auxiliary GRF hashmap, and a GRF type
enum. The hashmap is keyed by the enum, and clock branches now have a
struct member to store the value of that enum, which defaults to the
system GRF.
The SoC-specific _clk_init function can then insert pointers to GRF
regmaps into the hashmap based on the grf type.
During clock branch registration, we then pick the right GRF for each
branch from the hashmap if something other than the sys GRF is
requested.
The reason for doing it with this grf type indirection in the clock
branches is so that we don't need to define the MUXGRF branches in a
separate step, just to have a direct pointer to a regmap available
already.
Signed-off-by: Nicolas Frattaroli <nicolas.frattaroli@collabora.com>
Link: https://lore.kernel.org/r/20250502-rk3576-sai-v3-2-376cef19dd7c@collabora.com
Signed-off-by: Heiko Stuebner <heiko@sntech.de>
In the commit 5d0eb375e6 ("clk: rockchip: Add clock controller driver
for RK3528 SoC") only the dt-binding header was added for the reset
controller for the RK3528 SoC.
Add a reset lookup table generated from the SRST symbols used by vendor
linux-6.1-stan-rkr5 kernel to complete support for the reset controller.
Signed-off-by: Jonas Karlman <jonas@kwiboo.se>
Link: https://lore.kernel.org/r/20250227175302.2950788-1-jonas@kwiboo.se
Signed-off-by: Heiko Stuebner <heiko@sntech.de>
Add clock tree definition for RK3528. Similar to previous Rockchip
SoCs, clock controller of RK3528 is combined with the reset controller.
We omit the reset part for now since it's hard to test it without
support for other basic peripherals.
Signed-off-by: Yao Zi <ziyao@disroot.org>
Link: https://lore.kernel.org/r/20250217061142.38480-8-ziyao@disroot.org
Signed-off-by: Heiko Stuebner <heiko@sntech.de>
The clock 'pclk_vi_niu' has a dependency on 'hclk_vi_niu' according
to the Technical Reference Manual section '2.8.6 NIU Clock gating
reliance'. However, this kind of dependency cannot be addressed
properly at the moment (until the support for linked clocks is
implemented for the RK3568).
As an intermediate solution, mark the hclk_vi as critical on the
Rockchip RK3568.
Suggested-by: Nicolas Frattaroli <frattaroli.nicolas@gmail.com>
Signed-off-by: Michael Riesch <michael.riesch@wolfvision.net>
Link: https://lore.kernel.org/r/20250210-rk3568-hclk-vi-v1-1-9ade2626f638@wolfvision.net
Signed-off-by: Heiko Stuebner <heiko@sntech.de>
Ethernet phys normally need a 25MHz refclk input. On a lot of boards
this is done with a dedicated 25MHz crystal. But the rk3588 CRU also
provides a means for that via the refclko25m_ethX clock outputs that
can be used for that function.
The mdio bus normally probes devices on the bus at runtime, by reading
specific phy registers. This requires the phy to be running and thus
also being supplied by its reference clock.
While there exist the possibility and dt-binding to declare these
input clocks for each phy in the phy-dt-node, this is only relevant
_after_ the phy has been detected and during the drivers probe-run.
This results in a chicken-and-egg-problem. The refclks in the CRU are
running on boot of course, but phy-probing can very well happen after
clk_disable_unused has run.
In the past I tried to make clock-handling part of the mdio bus code [0]
but that wasn't very well received, due to it being specific to OF and
clocks with the consensus being that resources needed for detection
need to be enabled before.
So to make probing ethernet phys using the internal refclks possible,
make those 2 clocks critical.
[0] https://lore.kernel.org/netdev/13590315.F0gNSz5aLb@diego/T/
Signed-off-by: Heiko Stuebner <heiko.stuebner@cherry.de>
Link: https://lore.kernel.org/r/20241214224820.200665-1-heiko@sntech.de
Signed-off-by: Heiko Stuebner <heiko@sntech.de>
Recent Rockchip SoCs have a new hardware block called Native Interface
Unit (NIU), which gates clocks to devices behind them. These clock
gates will only have a running output clock when all of the following
conditions are met:
1. the parent clock is enabled
2. the enable bit is set correctly
3. the linked clock is enabled
To handle them this code registers them as a normal gate type clock,
which takes care of condition 1 + 2. The linked clock is handled by
using runtime PM clocks. Handling it via runtime PM requires setting
up a struct device for each of these clocks with a driver attached
to use the correct runtime PM operations. Thus the complete handling
of these clocks has been moved into its own driver.
Signed-off-by: Sebastian Reichel <sebastian.reichel@collabora.com>
Link: https://lore.kernel.org/r/20241211165957.94922-5-sebastian.reichel@collabora.com
Signed-off-by: Heiko Stuebner <heiko@sntech.de>
The proper GATE_LINK implementation will use runtime PM to handle the
linked gate clocks, which requires device context. Currently all clocks
are registered early via CLK_OF_DECLARE, which is before the kernel
knows about devices.
Moving the full clocks registration to the probe routine does not work,
since the clocks needed for timers must be registered early.
To work around this issue, most of the clock tree is registered early,
but GATE_LINK clocks are handled in the probe routine. Since the resets
are not needed early either, they have also been moved to the probe
routine.
Signed-off-by: Sebastian Reichel <sebastian.reichel@collabora.com>
Link: https://lore.kernel.org/r/20241211165957.94922-3-sebastian.reichel@collabora.com
Signed-off-by: Heiko Stuebner <heiko@sntech.de>
When some clocks are registered late and some clocks are registered
early we need to make sure the late registered clocks report probe defer
until the final registration has happened.
But we do not want to keep reporting probe defer after the late
registration has happened. Also not all Rockchip SoCs have late
registered clocks and may not need to report probe defer at all.
This restructures code a bit, so that there is a new function
rockchip_clk_init_early(), which should be used for initializing the CRU
structure on SoCs making use of late initialization in addition to the
early init. These platforms should call rockchip_clk_finalize()
once all clocks have been registered.
Signed-off-by: Sebastian Reichel <sebastian.reichel@collabora.com>
[added EXPORT_SYMBOL_GPL(rockchip_clk_finalize) to match the early function]
Link: https://lore.kernel.org/r/20241211165957.94922-2-sebastian.reichel@collabora.com
Signed-off-by: Heiko Stuebner <heiko@sntech.de>
When -Wunused-const-variable is enabled (not the default),
there is a warning about two definitions in this file:
In file included from drivers/clk/rockchip/clk-rk3576.c:14:
drivers/clk/rockchip/clk-rk3576.c:334:7: error: 'mclk_pdm0_p' defined but not used [-Werror=unused-const-variable=]
334 | PNAME(mclk_pdm0_p) = { "mclk_pdm0_src_top", "xin24m" };
| ^~~~~~~~~~~
drivers/clk/rockchip/clk.h:564:43: note: in definition of macro 'PNAME'
564 | #define PNAME(x) static const char *const x[] __initconst
| ^
drivers/clk/rockchip/clk-rk3576.c:333:7: error: 'pdm0_p' defined but not used [-Werror=unused-const-variable=]
333 | PNAME(pdm0_p) = { "clk_pdm0_src_top", "xin24m" };
| ^~~~~~
drivers/clk/rockchip/clk.h:564:43: note: in definition of macro 'PNAME'
564 | #define PNAME(x) static const char *const x[] __initconst
| ^
Remove them for the moment. If they are needed later, they can
be added back at that point.
Fixes: cc40f5baa9 ("clk: rockchip: Add clock controller for the RK3576")
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Link: https://lore.kernel.org/r/20240909121116.254036-1-arnd@kernel.org
Signed-off-by: Stephen Boyd <sboyd@kernel.org>
All clocks are registered early using CLK_OF_DECLARE(), which marks
the DT node as processed. For the processed DT node the probe routine
is never called. Thus this whole code is never executed. This could
be "fixed" by using CLK_OF_DECLARE_DRIVER, which avoids marking the
DT node as processed. But then the probe routine would re-register
all the clocks by calling rk3588_clk_init() again.
Signed-off-by: Sebastian Reichel <sebastian.reichel@collabora.com>
Link: https://lore.kernel.org/r/20240325193609.237182-2-sebastian.reichel@collabora.com
Signed-off-by: Heiko Stuebner <heiko@sntech.de>
