On Fri, 2021-07-23 at 14:27 +0800, Chen-Yu Tsai wrote:
On Thu, Jul 22, 2021 at 4:56 PM Trevor Wu trevor.wu@mediatek.com wrote:
On Mon, 2021-07-19 at 18:05 +0800, Chen-Yu Tsai wrote:
Hi,
On Thu, Jul 15, 2021 at 7:05 PM Trevor Wu <trevor.wu@mediatek.com
wrote:
On Tue, 2021-07-13 at 14:00 +0800, Chen-Yu Tsai wrote:
On Mon, Jul 12, 2021 at 11:10 PM Trevor Wu < trevor.wu@mediatek.com> wrote:
On Mon, 2021-07-12 at 14:57 +0800, Chen-Yu Tsai wrote: > are all internal Hi, > > On Tue, Jun 29, 2021 at 9:49 AM Trevor Wu < > trevor.wu@mediatek.com > > > > wrote: > > > > This patch adds mt8195 platform and affiliated driver. > > > > Signed-off-by: Trevor Wu trevor.wu@mediatek.com > > --- > > sound/soc/mediatek/Kconfig | 9 > > + > > sound/soc/mediatek/Makefile | 1 + > > sound/soc/mediatek/mt8195/Makefile | 11 + > > sound/soc/mediatek/mt8195/mt8195-afe-clk.c | 899 > > +++++ > > sound/soc/mediatek/mt8195/mt8195-afe-clk.h | 201 + > > sound/soc/mediatek/mt8195/mt8195-afe-common.h | 200 + > > sound/soc/mediatek/mt8195/mt8195-afe-pcm.c | 3264 > > +++++++++++++++++ > > sound/soc/mediatek/mt8195/mt8195-reg.h | 2793 > > ++++++++++++++ > > 8 files changed, 7378 insertions(+) > > create mode 100644 sound/soc/mediatek/mt8195/Makefile > > create mode 100644 sound/soc/mediatek/mt8195/mt8195- > > afe- > > clk.c > > create mode 100644 sound/soc/mediatek/mt8195/mt8195- > > afe- > > clk.h > > create mode 100644 sound/soc/mediatek/mt8195/mt8195- > > afe- > > common.h > > create mode 100644 sound/soc/mediatek/mt8195/mt8195- > > afe- > > pcm.c > > create mode 100644 sound/soc/mediatek/mt8195/mt8195- > > reg.h > > > > diff --git a/sound/soc/mediatek/Kconfig > > b/sound/soc/mediatek/Kconfig > > index 74dae4332d17..3389f382be06 100644 > > --- a/sound/soc/mediatek/Kconfig > > +++ b/sound/soc/mediatek/Kconfig > > @@ -184,3 +184,12 @@ config > > SND_SOC_MT8192_MT6359_RT1015_RT5682 > > with the MT6359 RT1015 RT5682 audio codec. > > Select Y if you have such device. > > If unsure select "N". > > + > > +config SND_SOC_MT8195 > > + tristate "ASoC support for Mediatek MT8195 > > chip" > > + select SND_SOC_MEDIATEK > > + help > > + This adds ASoC platform driver support for > > Mediatek > > MT8195 chip > > + that can be used with other codecs. > > + Select Y if you have such device. > > + If unsure select "N". > > diff --git a/sound/soc/mediatek/Makefile > > b/sound/soc/mediatek/Makefile > > index f6cb6b8508e3..34778ca12106 100644 > > --- a/sound/soc/mediatek/Makefile > > +++ b/sound/soc/mediatek/Makefile > > @@ -5,3 +5,4 @@ obj-$(CONFIG_SND_SOC_MT6797) += mt6797/ > > obj-$(CONFIG_SND_SOC_MT8173) += mt8173/ > > obj-$(CONFIG_SND_SOC_MT8183) += mt8183/ > > obj-$(CONFIG_SND_SOC_MT8192) += mt8192/ > > +obj-$(CONFIG_SND_SOC_MT8195) += mt8195/ > > diff --git a/sound/soc/mediatek/mt8195/Makefile > > b/sound/soc/mediatek/mt8195/Makefile > > new file mode 100644 > > index 000000000000..b2c9fd88f39e > > --- /dev/null > > +++ b/sound/soc/mediatek/mt8195/Makefile > > @@ -0,0 +1,11 @@ > > +# SPDX-License-Identifier: GPL-2.0 > > + > > +# platform driver > > +snd-soc-mt8195-afe-objs := \ > > + mt8195-afe-clk.o \ > > + mt8195-afe-pcm.o \ > > + mt8195-dai-adda.o \ > > + mt8195-dai-etdm.o \ > > + mt8195-dai-pcm.o > > + > > +obj-$(CONFIG_SND_SOC_MT8195) += snd-soc-mt8195-afe.o > > diff --git a/sound/soc/mediatek/mt8195/mt8195-afe-clk.c > > b/sound/soc/mediatek/mt8195/mt8195-afe-clk.c > > new file mode 100644 > > index 000000000000..57aa799b4f41 > > --- /dev/null > > +++ b/sound/soc/mediatek/mt8195/mt8195-afe-clk.c > > @@ -0,0 +1,899 @@ > > +// SPDX-License-Identifier: GPL-2.0 > > +/* > > + * mt8195-afe-clk.c -- Mediatek 8195 afe clock ctrl > > + * > > + * Copyright (c) 2021 MediaTek Inc. > > + * Author: Bicycle Tsai bicycle.tsai@mediatek.com > > + * Trevor Wu trevor.wu@mediatek.com > > + */ > > + > > +#include <linux/clk.h> > > + > > +#include "mt8195-afe-common.h" > > +#include "mt8195-afe-clk.h" > > +#include "mt8195-reg.h" > > + > > +static const char *aud_clks[MT8195_CLK_NUM] = { > > Most of these clocks are not described in the device tree > binding. If > the driver needs to reference them, they should be > described. > We > should > not be hard-coding clock names across different drivers. >
Sorry, I didn't know I have to list all clocks in the dt- binding. Originally, I thought these clocks will be described in the clock binding, so I didn't add them to the binding of afe driver. I will add these clocks to mt8195-afe-pcm.yaml.
If the device consumes clocks, then the clocks that get consumed should be listed in the device's bindings. This is not related to the clock bindings, which is a clock provider.
Got it. Thanks.
> The more important question is, why does the driver need > to > reference > all of them? Maybe we should take a step back and draw > out a > clock > tree > diagram for the hardware? >
The clock structure is PLL -> MUX -> GATE. xtal, pll and divider are the possible clock inputs for MUX. Because we select the clock input of audio module based on the use case, we use clk_get to retrive all clocks which are possible to be used.
So I see a couple the driver is doing reparenting:
a. Reparent audio_h to standard oscillator when ADDA is not used, presumably to let the APLL be turned off
Why not just turn off audio_h? It looks like audio_h feeds a couple clock gates in the audio subsystem. Just a guess, but is this the AHB bus clock? Why not just have it parented to "univpll_d7" all the time then?
Sorry, I am not sure if it is the AHB bus clock. I only know how audio module uses the clock. audio_h feeds to some clock gate like aud_adc_hires, which is used when sampling rate is higher than 48kHz, and hardware designer suggests us use apll1_ck when AFE requrires the clock.
I see. So the simplified explanation is high clock rate for high res audio. Would high clock rate work for standard sample rates?
As far as I know, HW will switch clock to hires clock automatically when the required rate is high,(ex: aud_adc and aud_adc_hires) so it can't be controlled by driver.
I see. That might not be so friendly to the Linux clk driver.
Would using apll1 or univpll all the time work, instead of reparenting? What's the gain if we do reparenting?
As you said before, the gain is apll can be turned off when the clock is not requrired by ADDA. That's why we didn't use apll all the time.
Right, and what's the gain from turning it off? Lower power consumption?
Yes. Xtal_26m is supplied to most modules, but APLL1 is mainly used by afe. When audio feature is not used, we hope APLL1 can be turned off to lower power consumption.
As I know, DSP also requires audio_h. When we disable the clock in AFE driver, the ref count in CCF is not becoming zero if DSP still uses it. But only AFE requires higher clock rate, so we reparent audio_h to 26M when it's not required in adda module.
I see. Wouldn't reparenting the clock while it is in use by another module cause glitches?
I checked with the DSP owner. audio_h clock is required for DSP bus, but the clock rate is not important. The only thing it cares is audio_h should be powered on, so reparenting is harmless for DSP.
OK.
Also, reparenting really should be done implicitly with clk_set_rate() with the clock driver supporting reparenting on rate changes.
b. Assignment of PLLs for I2S/PCM MCLK outputs
Is there a reason for explicit assignment, other than clock rate conflicts? CCF supports requesting and locking the clock rate. And again, implicit reparenting should be the norm. The clock driver's purpose is to fulfill any and all clock rate requirements from its consumers. The consumer should only need to ask for the clock rate, not a specific parent, unless there are details that are not yet covered by the CCF.
For MCLK output, we should configure divider to get the target rate, and it can only divide the clock from current parent source. So we should do reparent to divider's parent in case the parent rate is not a multiple of target rate.
Right. That is expected. What I'm saying is that the CCF provides the framework for automatically reparenting based on the requested clock rate. This is done in the clock driver's .determine_rate op.
When properly implemented, and also restricting or locking the clock rates of the PLLs, then you can simply request a clock rate on the leaf clock, in this case one of the MCLKs, and the CCF and clock driver would handle everything else. The consumer should not be reparenting clocks manually unless for a very good reason which cannot be satisfied by the CCF.
In some use cases, we really need to reparent clock manually. For example, spdif in(slave) -> .... -> i2s out(master)
APLL3/APLL4 are reserved for slave input like earc in or spdif in, which can refer to the external clock source.(APLL3 syncs with earc, and APLL4 syncs with spdif in.)
When i2s out selects the clock source to APLL4, this makes sure that spdif in and i2s out works in the same clock source. If we just use APLL1/APLL2 on i2s out, there is little rate mismatch between data input and output. Finally, it results in XRUN.
I see, that makes more sense.
If we only use set_rate, it's possible that it can't switch to the expected PLL source, because the rate of APLL3/APLL4 should be close to APLL1/APLL2.
Well, in theory the CCF should choose the one with the closest rate. And if APLL3/APLL4 is already tracking the external clock source, its clock rate should match.
If it's a static requirement, maybe we could replace the *-mclk- source DT properties with standard assigned-clocks and assigned-clock- parents? This would get handled by CCF directly, and then the only thing the clk driver has to do is make sure it doesn't get reparented again.
Or is there a need to do reparenting at runtime?
For the use case of APLL3/APLL4, static assignment should be ok.
But I checked with CCF owner, we can't just use clk_set_rate(divider, rate) to get expected MCLK output, because reparenting MUX automatically is not supported now. (PLL -> MUX -> divider)
We still have to call clk_set_parent() before clk_set_rate(). Which means some information should be got from DTS to check whether the PLL source can be switched or not, so *-mclk-source should be keeped to identify the case.
Thanks, Trevor
A related question: the chip has five APLLs. How many MCLK combinations does the application need to support? I assume this includes the standard 24.576 MHz and 22.5792 MHz clock rates.
APLL1 and APLL2 are used in most AFE modules, so their rate should be fixed. APLL1 is fixed to 196608000Hz. APLL2 is fixed to 180633600Hz. APLL is inputed to the divider(8bit), and MCLK is the output of divider. Other APLLs are reserved for some special usage which can't be supported by APLL1 & APLL2. But APLL3~APLL5 aren't used in the series, so I will remove them in v3.
Some of them are not used in this series, because some modules are still developing. Should I only keep the clocks that have been used in the series?
Yes please. Only add the ones that are used. Things that aren't used don't get tested and verified, and end up as dead code. If there are plans to extend them in the future, and you can leave comments stating that intent, and also mention it in the cover letter.
OK, I will remove the unused clock in v3.
> > + /* xtal */ > > + [MT8195_CLK_XTAL_26M] = "clk26m", > > + /* pll */ > > + [MT8195_CLK_APMIXED_APLL1] = "apll1", > > + [MT8195_CLK_APMIXED_APLL2] = "apll2", > > + [MT8195_CLK_APMIXED_APLL3] = "apll3", > > + [MT8195_CLK_APMIXED_APLL4] = "apll4", > > + [MT8195_CLK_APMIXED_APLL5] = "apll5", > > + [MT8195_CLK_APMIXED_HDMIRX_APLL] = > > "hdmirx_apll", > > + /* divider */ > > + [MT8195_CLK_TOP_APLL1] = "apll1_ck", > > + [MT8195_CLK_TOP_APLL1_D4] = "apll1_d4", > > + [MT8195_CLK_TOP_APLL2] = "apll2_ck", > > + [MT8195_CLK_TOP_APLL2_D4] = "apll2_d4", > > + [MT8195_CLK_TOP_APLL3] = "apll3_ck", > > + [MT8195_CLK_TOP_APLL3_D4] = "apll3_d4", > > + [MT8195_CLK_TOP_APLL4] = "apll4_ck", > > + [MT8195_CLK_TOP_APLL4_D4] = "apll4_d4", > > + [MT8195_CLK_TOP_APLL5] = "apll5_ck", > > + [MT8195_CLK_TOP_APLL5_D4] = "apll5_d4", > > + [MT8195_CLK_TOP_APLL12_DIV0] = "apll12_div0", > > + [MT8195_CLK_TOP_APLL12_DIV1] = "apll12_div1", > > + [MT8195_CLK_TOP_APLL12_DIV2] = "apll12_div2", > > + [MT8195_CLK_TOP_APLL12_DIV3] = "apll12_div3", > > + [MT8195_CLK_TOP_APLL12_DIV4] = "apll12_div4", > > + [MT8195_CLK_TOP_APLL12_DIV9] = "apll12_div9", > > + [MT8195_CLK_TOP_HDMIRX_APLL] = > > "hdmirx_apll_ck", > > + [MT8195_CLK_TOP_MAINPLL_D4_D4] = > > "mainpll_d4_d4", > > + [MT8195_CLK_TOP_MAINPLL_D5_D2] = > > "mainpll_d5_d2", > > + [MT8195_CLK_TOP_MAINPLL_D7_D2] = > > "mainpll_d7_d2", > > + [MT8195_CLK_TOP_UNIVPLL_D4] = "univpll_d4", > > + /* mux */ > > + [MT8195_CLK_TOP_APLL1_SEL] = "apll1_sel", > > + [MT8195_CLK_TOP_APLL2_SEL] = "apll2_sel", > > + [MT8195_CLK_TOP_APLL3_SEL] = "apll3_sel", > > + [MT8195_CLK_TOP_APLL4_SEL] = "apll4_sel", > > + [MT8195_CLK_TOP_APLL5_SEL] = "apll5_sel", > > + [MT8195_CLK_TOP_A1SYS_HP_SEL] = "a1sys_hp_sel", > > + [MT8195_CLK_TOP_A2SYS_SEL] = "a2sys_sel", > > + [MT8195_CLK_TOP_A3SYS_SEL] = "a3sys_sel", > > + [MT8195_CLK_TOP_A4SYS_SEL] = "a4sys_sel", > > + [MT8195_CLK_TOP_ASM_H_SEL] = "asm_h_sel", > > + [MT8195_CLK_TOP_ASM_M_SEL] = "asm_m_sel", > > + [MT8195_CLK_TOP_ASM_L_SEL] = "asm_l_sel", > > + [MT8195_CLK_TOP_AUD_IEC_SEL] = "aud_iec_sel", > > + [MT8195_CLK_TOP_AUD_INTBUS_SEL] = > > "aud_intbus_sel", > > + [MT8195_CLK_TOP_AUDIO_H_SEL] = "audio_h_sel", > > + [MT8195_CLK_TOP_AUDIO_LOCAL_BUS_SEL] = > > "audio_local_bus_sel", > > + [MT8195_CLK_TOP_DPTX_M_SEL] = "dptx_m_sel", > > + [MT8195_CLK_TOP_INTDIR_SEL] = "intdir_sel", > > + [MT8195_CLK_TOP_I2SO1_M_SEL] = "i2so1_m_sel", > > + [MT8195_CLK_TOP_I2SO2_M_SEL] = "i2so2_m_sel", > > + [MT8195_CLK_TOP_I2SI1_M_SEL] = "i2si1_m_sel", > > + [MT8195_CLK_TOP_I2SI2_M_SEL] = "i2si2_m_sel", > > + /* clock gate */ > > + [MT8195_CLK_TOP_MPHONE_SLAVE_B] = > > "mphone_slave_b", > > + [MT8195_CLK_TOP_CFG_26M_AUD] = "cfg_26m_aud", > > + [MT8195_CLK_INFRA_AO_AUDIO] = "infra_ao_audio", > > + [MT8195_CLK_INFRA_AO_AUDIO_26M_B] = > > "infra_ao_audio_26m_b", > > + [MT8195_CLK_SCP_ADSP_AUDIODSP] = > > "scp_adsp_audiodsp", > > > > + [MT8195_CLK_AUD_AFE] = "aud_afe", > > + [MT8195_CLK_AUD_LRCK_CNT] = "aud_lrck_cnt", > > + [MT8195_CLK_AUD_SPDIFIN_TUNER_APLL] = > > "aud_spdifin_tuner_apll", > > + [MT8195_CLK_AUD_SPDIFIN_TUNER_DBG] = > > "aud_spdifin_tuner_dbg", > > + [MT8195_CLK_AUD_UL_TML] = "aud_ul_tml", > > + [MT8195_CLK_AUD_APLL1_TUNER] = > > "aud_apll1_tuner", > > + [MT8195_CLK_AUD_APLL2_TUNER] = > > "aud_apll2_tuner", > > + [MT8195_CLK_AUD_TOP0_SPDF] = "aud_top0_spdf", > > + [MT8195_CLK_AUD_APLL] = "aud_apll", > > + [MT8195_CLK_AUD_APLL2] = "aud_apll2", > > + [MT8195_CLK_AUD_DAC] = "aud_dac", > > + [MT8195_CLK_AUD_DAC_PREDIS] = "aud_dac_predis", > > + [MT8195_CLK_AUD_TML] = "aud_tml", > > + [MT8195_CLK_AUD_ADC] = "aud_adc", > > + [MT8195_CLK_AUD_DAC_HIRES] = "aud_dac_hires", > > + [MT8195_CLK_AUD_A1SYS_HP] = "aud_a1sys_hp", > > + [MT8195_CLK_AUD_AFE_DMIC1] = "aud_afe_dmic1", > > + [MT8195_CLK_AUD_AFE_DMIC2] = "aud_afe_dmic2", > > + [MT8195_CLK_AUD_AFE_DMIC3] = "aud_afe_dmic3", > > + [MT8195_CLK_AUD_AFE_DMIC4] = "aud_afe_dmic4", > > + [MT8195_CLK_AUD_AFE_26M_DMIC_TM] = > > "aud_afe_26m_dmic_tm", > > + [MT8195_CLK_AUD_UL_TML_HIRES] = > > "aud_ul_tml_hires", > > + [MT8195_CLK_AUD_ADC_HIRES] = "aud_adc_hires", > > + [MT8195_CLK_AUD_ADDA6_ADC] = "aud_adda6_adc", > > + [MT8195_CLK_AUD_ADDA6_ADC_HIRES] = > > "aud_adda6_adc_hires", > > + [MT8195_CLK_AUD_LINEIN_TUNER] = > > "aud_linein_tuner", > > + [MT8195_CLK_AUD_EARC_TUNER] = "aud_earc_tuner", > > + [MT8195_CLK_AUD_I2SIN] = "aud_i2sin", > > + [MT8195_CLK_AUD_TDM_IN] = "aud_tdm_in", > > + [MT8195_CLK_AUD_I2S_OUT] = "aud_i2s_out", > > + [MT8195_CLK_AUD_TDM_OUT] = "aud_tdm_out", > > + [MT8195_CLK_AUD_HDMI_OUT] = "aud_hdmi_out", > > + [MT8195_CLK_AUD_ASRC11] = "aud_asrc11", > > + [MT8195_CLK_AUD_ASRC12] = "aud_asrc12", > > + [MT8195_CLK_AUD_MULTI_IN] = "aud_multi_in", > > + [MT8195_CLK_AUD_INTDIR] = "aud_intdir", > > + [MT8195_CLK_AUD_A1SYS] = "aud_a1sys", > > + [MT8195_CLK_AUD_A2SYS] = "aud_a2sys", > > + [MT8195_CLK_AUD_PCMIF] = "aud_pcmif", > > + [MT8195_CLK_AUD_A3SYS] = "aud_a3sys", > > + [MT8195_CLK_AUD_A4SYS] = "aud_a4sys", > > + [MT8195_CLK_AUD_MEMIF_UL1] = "aud_memif_ul1", > > + [MT8195_CLK_AUD_MEMIF_UL2] = "aud_memif_ul2", > > + [MT8195_CLK_AUD_MEMIF_UL3] = "aud_memif_ul3", > > + [MT8195_CLK_AUD_MEMIF_UL4] = "aud_memif_ul4", > > + [MT8195_CLK_AUD_MEMIF_UL5] = "aud_memif_ul5", > > + [MT8195_CLK_AUD_MEMIF_UL6] = "aud_memif_ul6", > > + [MT8195_CLK_AUD_MEMIF_UL8] = "aud_memif_ul8", > > + [MT8195_CLK_AUD_MEMIF_UL9] = "aud_memif_ul9", > > + [MT8195_CLK_AUD_MEMIF_UL10] = "aud_memif_ul10", > > + [MT8195_CLK_AUD_MEMIF_DL2] = "aud_memif_dl2", > > + [MT8195_CLK_AUD_MEMIF_DL3] = "aud_memif_dl3", > > + [MT8195_CLK_AUD_MEMIF_DL6] = "aud_memif_dl6", > > + [MT8195_CLK_AUD_MEMIF_DL7] = "aud_memif_dl7", > > + [MT8195_CLK_AUD_MEMIF_DL8] = "aud_memif_dl8", > > + [MT8195_CLK_AUD_MEMIF_DL10] = "aud_memif_dl10", > > + [MT8195_CLK_AUD_MEMIF_DL11] = "aud_memif_dl11", > > + [MT8195_CLK_AUD_GASRC0] = "aud_gasrc0", > > + [MT8195_CLK_AUD_GASRC1] = "aud_gasrc1", > > + [MT8195_CLK_AUD_GASRC2] = "aud_gasrc2", > > + [MT8195_CLK_AUD_GASRC3] = "aud_gasrc3", > > + [MT8195_CLK_AUD_GASRC4] = "aud_gasrc4", > > + [MT8195_CLK_AUD_GASRC5] = "aud_gasrc5", > > + [MT8195_CLK_AUD_GASRC6] = "aud_gasrc6", > > + [MT8195_CLK_AUD_GASRC7] = "aud_gasrc7", > > + [MT8195_CLK_AUD_GASRC8] = "aud_gasrc8", > > + [MT8195_CLK_AUD_GASRC9] = "aud_gasrc9", > > + [MT8195_CLK_AUD_GASRC10] = "aud_gasrc10", > > + [MT8195_CLK_AUD_GASRC11] = "aud_gasrc11", > > + [MT8195_CLK_AUD_GASRC12] = "aud_gasrc12", > > + [MT8195_CLK_AUD_GASRC13] = "aud_gasrc13", > > + [MT8195_CLK_AUD_GASRC14] = "aud_gasrc14", > > + [MT8195_CLK_AUD_GASRC15] = "aud_gasrc15", > > + [MT8195_CLK_AUD_GASRC16] = "aud_gasrc16", > > + [MT8195_CLK_AUD_GASRC17] = "aud_gasrc17", > > + [MT8195_CLK_AUD_GASRC18] = "aud_gasrc18", > > + [MT8195_CLK_AUD_GASRC19] = "aud_gasrc19", > > The MT8195_CLK_AUD_* clocks are all internal to the audio > subsystem: > the bits that control these clock gates are in the same > address > space > as the audio parts. Would it be possible to model them as > internal > ASoC SUPPLY widgets? The external ones could be modeled > using > ASoC > CLK_SUPPLY widgets, and the dependencies could be modeled > with > ASoC > routes. The ASoC core could then handle power sequencing, > which > the > driver currently does manually. > > IMO this is better than having two drivers handling two > aspects > of > the same piece of hardware, while the two aspects are > intertwined. >
Yes, it's ok to use the CLK_SUPPLY and SUPPLY to model such clocks. But those clocks are managed by CCF in the preceding SOCs like mt2701, mt6779 and mt8183. Additionally, in some audio modules, clocks should
This being a new driver, we have some more freedom to improve the design.
be enabled before configuring parameters(hw_params). As far as I know, if we use CLK_SUPPLY or SUPPLY to model clocks, the power sequence is controlled by DAPM. It seems to be impossible to fulfill all use cases. That's why we just keep the manual control sequence and CCF seems to be the best choice to model such clock gatess.
I see. So yes, using CCF does give you reference counting, dependency tracking and other advantages. And using DAPM supplies means you can't enable the clock gates outside of DAPM without both pieces of code fighting for control.
Can we at least move the audio clock gates into the audio driver though? The arbitrary separation into two devices and drivers is fishy. And with the move the external references to the audio clock gates can be removed.
Because DAPM SUPPLY can't fit our control scenario. Did you suggest us implement the simple logic control(including ref count, clock dependency) for clock gate(MT8195_CLK_AUD_*) in afe driver instead of using CCF?
I meant simply moving the CCF-based clk driver code (clk-mt8516- aud.c) from `drivers/clk` and incorporating it into the audio driver, likely in `mt8195-afe-clk.c` or maybe as a separate file. So the audio driver would be a clock provider, and a clock consumer. It will directly use the clocks it provides, internally, and you could remove all those clock references from the device tree.
The goal is to have one hardware representation (device node) only, so that it matches the hardware, which is one single unified block.
After the driver is completed, we can look for opportunities to improve it, if resources are available.
Thanks for your detailed information. I will try to move the CCF-based clk driver code to AFE driver. If there are no other internal concerns and blocking problems, I will include the changes in v3.
Great.
And regarding the clock requirements for different modules, could we have that information put in comments somewhere, so if someone were to revisit it later, they would have the information needed to understand and possibly improve it? Because right now there's just a bunch of clocks enabled and disabled and nothing to explain why that's needed.
For example, MT8195_CLK_AUD_ADC(clock gate) is one of the clock feeding to ADDA module. Did you want me show the clock gate list feeding to ADDA? On the other hand, I didn't know how to show the information properly in comments. Could you kindly share me an example for reference?
For example, in `mt8195_afe_enable_reg_rw_clk()` in mt8195-afe- clk.c:
unsigned int clk_array[] = { MT8195_CLK_SCP_ADSP_AUDIODSP, MT8195_CLK_TOP_AUDIO_LOCAL_BUS_SEL, MT8195_CLK_TOP_CFG_26M_AUD, MT8195_CLK_INFRA_AO_AUDIO, MT8195_CLK_INFRA_AO_AUDIO_26M_B, MT8195_CLK_TOP_AUD_INTBUS_SEL, MT8195_CLK_TOP_A1SYS_HP_SEL, MT8195_CLK_AUD_A1SYS_HP, MT8195_CLK_AUD_A1SYS, MT8195_CLK_TOP_AUDIO_H_SEL, };You could add a comment after each line stating why that clock needs to be enabled. A simple note like "bus access clock" or "internal logic clock" would suffice.
OK, I will add short notes to such clock lists.
The above list also has some redundancies that could be eliminated. MT8195_CLK_TOP_A1SYS_HP_SEL is parent to both MT8195_CLK_AUD_A1SYS_HP and MT8195_CLK_AUD_A1SYS. When clocks are enabled, their parents are also enabled by CCF, so there's no need to enable them explicitly, unless that clock also directly feeds the clock consumer.
OK, I will review all clock usages and remove the unnecessary clocks.
Another thing I wanted to bring up: is any of the code after
struct mt8195_afe_tuner_cfg {used? It looks like it is used to configure the five extra PLLs in the audio subsystem, but the exposed (non-static) functions don't seem to be called anywhere. Are they for modules not yet supported?
Yes, tuners are not supported now. I will remove the code and add them back when tuners are required in the future.
Thanks.
ChenYu