Hello,
this is version 5 of the I2S/TDM driver patchset. A big thanks to everyone who has provided their valuable feedback so far.
Changes in v5: driver: - change comment style of the first comment to C++ style - make refcount non-atomic, as it's only ever used inside a spinlock - use newer SND_SOC_DAIFMT_CB* defines - change ternary statements to if/else conditions - make _clk_compensation_put return 1 if clock changed - implement set_bclk_ratio callback - always set half frame sync mode in TDM mode - automatically enable mclk-calibrate mode when the clocks for it are specified in the device tree bindings: - add Reviewed-by: Rob Herring - drop rockchip,frame-width property (done by set_bclk_ratio) - drop rockchip,fsync-half-frame property - drop rockchip,mclk-calibrate property dts: - drop empty codec block from Quartz64 device tree
Changes in v4: driver: - factor TDE/RDE enable/disable into their own inlined functions - add an RDE disable in a location where it looks like it was forgotten (rxctrl else), judging by corresponding TDE code - remove parentheses around CLK_PPM_MIN/MAX values - wording + titlecasing in the clock compensation control - use if statement in precious_reg instead - refactor rockchip_i2s_io_multiplex to have the switch statements in a function call to make the function less unwieldy - get rid of IS_ERR checks around clk enable/disable calls where already checked before by the probe - reworded some error message strings - fix potential deadlock in txrxctrl found by Sugar Zhang using spin_lock_irqsave - fix potential deadlock in trcm_mode found by Sugar Zhang using spin_lock_irqsave - use devm_platform_get_and_ioremap_resource in probe - only set DMA things if controller has capture/playback ability. Did not move this into init_dai because I'd then need to pass in the res and probe it earlier in the function, and it's also used elsewhere in the probe function - use _get_optional_exclusive for reset controls, as some controllers only have capture or playback capability bindings: - remove status = "okay" since that's the default - change the path configs to be an enum - rename "foo" to "bus" - make resets optional as controller may lack either playback or capture capability, and therefore also doesn't have a reset. At least one reset is still required, because a controller with no playback and no capture is not very useful
Changes in v3: driver: - alphabetically sort includes - check pm_runtime_get_sync return value, act on it - remove unnecessary initialisers in set_fmt - use udelay(15) in retry code: 10 retries * 15 = 150, so at worst we wait the full i2s register access delay - fix some weird returns to return directly - use __maybe_unused instead of #ifdef CONFIG_PM_SLEEP, also put __maybe_unused on the runtime callbacks - use (foo) instead of foo in header macros for precedence reasons - when using mclk-calibrate, also turn off/on those clocks during suspend and resume operations - remove mclk_tx and mclk_rx reenablement code in remove - move hclk enablement further down the probe, and disable it on probe failure - make reset controls mandatory, since the bindings state this too - use _exclusive for getting the reset controls - change reset assert/deassert delays to both be 10 usec (thank you Sugar Zhang!) - properly prepare and enable all mclks in probe, especially before calling clk_get_rate on them - if registering PCM fails, also use the cleanup error path instead of returning directly - bring back playback and capture only but in the way Sugar Zhang suggested it: set those modes depending on dma-names - rework clock enablement in general. Probe now always enables these, instead of relying on the pm resume thing - add myself to MAINTAINERS for this driver dt bindings: - fix a description still mentioning clk-trcm in the schema - document rockchip,io-multiplex, a property that describes the hardware as having multiplexed I2S GPIOs so direction needs to be changed dynamically - document rockchip,mclk-calibrate, which allows specifying different clocks for the two sample rate bases and switch between them as needed - dma-names now doesn't have a set order and items can be absent to indicate that the controller doesn't support this mode - add myself to MAINTAINERS for these bindings
Changes in v2: - remove ad-hoc writeq and needless (and broken) optimisation in reset assert/deassert. This wouldn't have worked on Big Endian, and would've been pointless on any other platform, as the overhead for saving one write was comparatively big - fix various checkpatch issues - get rid of leftover clk-trcm in schema - set status = "okay" in example in schema instead of "disabled" - change dma-names so rx is first, adjust device trees as necessary - properly reference uint32-array for rx-route and tx-route instead of uint32 - replace trcm-sync with two boolean properties, adjust DT changes accordingly and also get rid of the header file - get rid of rockchip,no-dmaengine. This was only needed for some downstream driver and shouldn't be in the DT - get rid of rockchip,capture-only/playback-only. Rationale being that I have no way to test whether they're needed, and unconditionally setting channels_min to 0 breaks everything - change hclk description in "clocks"
Nicolas Frattaroli (4): ASoC: rockchip: add support for i2s-tdm controller ASoC: dt-bindings: rockchip: add i2s-tdm bindings arm64: dts: rockchip: add i2s1 on rk356x arm64: dts: rockchip: add analog audio on Quartz64
.../bindings/sound/rockchip,i2s-tdm.yaml | 198 ++ MAINTAINERS | 7 + .../boot/dts/rockchip/rk3566-quartz64-a.dts | 31 +- arch/arm64/boot/dts/rockchip/rk356x.dtsi | 26 + sound/soc/rockchip/Kconfig | 11 + sound/soc/rockchip/Makefile | 2 + sound/soc/rockchip/rockchip_i2s_tdm.c | 1848 +++++++++++++++++ sound/soc/rockchip/rockchip_i2s_tdm.h | 398 ++++ 8 files changed, 2520 insertions(+), 1 deletion(-) create mode 100644 Documentation/devicetree/bindings/sound/rockchip,i2s-tdm.yaml create mode 100644 sound/soc/rockchip/rockchip_i2s_tdm.c create mode 100644 sound/soc/rockchip/rockchip_i2s_tdm.h