Re: [PATCH v2 03/12] ASoC: mediatek: mt8188: support audsys clock

[AngeloGioacchino Del Regno <angelogioacchino.delregno@collabora.com>]

Il 21/10/22 11:58, Trevor Wu (吳文良) ha scritto:
> On Fri, 2022-10-21 at 10:41 +0200, AngeloGioacchino Del Regno wrote:
>> Il 21/10/22 10:27, Trevor Wu ha scritto:
>>> Add mt8188 audio cg clock control. Audio clock gates are registered
>>> to CCF
>>> for reference count and clock parent management.
>>>
>>> Signed-off-by: Trevor Wu <trevor.wu@mediatek.com>
>>> ---
>>>    sound/soc/mediatek/mt8188/mt8188-audsys-clk.c | 206
>>> ++++++++++++++++++
>>>    sound/soc/mediatek/mt8188/mt8188-audsys-clk.h |  15 ++
>>>    .../soc/mediatek/mt8188/mt8188-audsys-clkid.h |  83 +++++++
>>>    3 files changed, 304 insertions(+)
>>>    create mode 100644 sound/soc/mediatek/mt8188/mt8188-audsys-clk.c
>>>    create mode 100644 sound/soc/mediatek/mt8188/mt8188-audsys-clk.h
>>>    create mode 100644 sound/soc/mediatek/mt8188/mt8188-audsys-
>>> clkid.h
>>>
>>> diff --git a/sound/soc/mediatek/mt8188/mt8188-audsys-clk.c
>>> b/sound/soc/mediatek/mt8188/mt8188-audsys-clk.c
>>> new file mode 100644
>>> index 000000000000..1f294231d4c2
>>> --- /dev/null
>>> +++ b/sound/soc/mediatek/mt8188/mt8188-audsys-clk.c
>>> @@ -0,0 +1,206 @@
>>> +// SPDX-License-Identifier: GPL-2.0
>>> +/*
>>> + * mt8188-audsys-clk.c  --  MediaTek 8188 audsys clock control
>>> + *
>>> + * Copyright (c) 2022 MediaTek Inc.
>>> + * Author: Chun-Chia Chiu <chun-chia.chiu@mediatek.com>
>>> + */
>>> +
>>> +#include <linux/clk.h>
>>> +#include <linux/clk-provider.h>
>>> +#include <linux/clkdev.h>
>>> +#include "mt8188-afe-common.h"
>>> +#include "mt8188-audsys-clk.h"
>>> +#include "mt8188-audsys-clkid.h"
>>> +#include "mt8188-reg.h"
>>> +
>>> +struct afe_gate {
>>> +	int id;
>>> +	const char *name;
>>> +	const char *parent_name;
>>> +	int reg;
>>> +	u8 bit;
>>> +	const struct clk_ops *ops;
>>> +	unsigned long flags;
>>> +	u8 cg_flags;
>>> +};
>>> +
>>> +#define GATE_AFE_FLAGS(_id, _name, _parent, _reg, _bit, _flags,
>>> _cgflags) {\
>>> +		.id = _id,					\
>>> +		.name = _name,					\
>>> +		.parent_name = _parent,				\
>>> +		.reg = _reg,					\
>>> +		.bit = _bit,					\
>>> +		.flags = _flags,				\
>>> +		.cg_flags = _cgflags,				\
>>> +	}
>>> +
>>> +#define GATE_AFE(_id, _name, _parent, _reg, _bit)		\
>>> +	GATE_AFE_FLAGS(_id, _name, _parent, _reg, _bit,		\
>>> +		       CLK_SET_RATE_PARENT | CLK_IGNORE_UNUSED,
>>> CLK_GATE_SET_TO_DISABLE)
>>
>> Can you please explain what's the reason for CLK_IGNORE_UNUSED here?
>> Maybe we can solve some issue that you're facing in a cleaner way.
>>
>> Regards,
>> Angelo
> 
> Hi Angelo,
> 
> Because clk_disable_unused() calls clk_core_is_enabled(), register
> access happens in is_enabled() ops.
> At the moment, the power for register access is not enabled, so the
> register read results in CPU hang.
> 
> That's why I added CLK_IGNORE_UNUSED here, but it can't resolve all
> issues. Actually, we met same problem when "cat
> /sys/kernel/debug/clk/clk_summary" is used. We are still suffering the
> problem.
> 
> I'm not sure if I can implement clk ops by myself, and exclude the
> registration of is_enabled() ops.
> 

Is the power for register access enabled with a power domain?

Check drivers/clk/clk.c, grep for core->rpm_enabled.

If you enable runtime PM before registering the clocks, and you register them
with the right struct device, the clock API will enable power for you before
trying to read the clock enable status.

Regards,
Angelo