Re: [PATCH v11 06/10] soc: mediatek: Add subsys clock control for bus protection

[Matthias Brugger]

On 13/02/2020 03:46, Weiyi Lu wrote:
> On Wed, 2020-02-12 at 12:02 +0100, Matthias Brugger wrote:
>>
>> On 12/02/2020 03:55, Weiyi Lu wrote:
>>> On Tue, 2020-02-11 at 18:54 +0100, Matthias Brugger wrote:
>>>>
>>>> On 20/12/2019 04:46, Weiyi Lu wrote:
>>>>> Add subsys CG control flow before/after the bus protect control
>>>>> due to bus protection need SMI bus relative CGs enabled to feedback
>>>>> its ack.
>>>>>
>>>>
>>>> Sorry, I don't understand the commit message. Can you please rephrase and
>>>> explain better what this change is for.
>>>>
>>>
>>> OK! I'll reword it.
>>>
>>>>> Signed-off-by: Weiyi Lu <weiyi.lu@xxxxxxxxxxxx>
>>>>> Reviewed-by: Nicolas Boichat <drinkcat@xxxxxxxxxxxx>
>>>>> ---
>>>>>  drivers/soc/mediatek/mtk-scpsys.c | 72 +++++++++++++++++++++++++++++++++++++--
>>>>>  1 file changed, 70 insertions(+), 2 deletions(-)
>>>>>
>>>>> diff --git a/drivers/soc/mediatek/mtk-scpsys.c b/drivers/soc/mediatek/mtk-scpsys.c
>>>>> index 763ca58..32be4b3 100644
>>>>> --- a/drivers/soc/mediatek/mtk-scpsys.c
>>>>> +++ b/drivers/soc/mediatek/mtk-scpsys.c
>>>>> @@ -79,6 +79,7 @@
>>>>>  #define PWR_STATUS_WB			BIT(27)	/* MT7622 */
>>>>>  
>>>>>  #define MAX_CLKS	3
>>>>> +#define MAX_SUBSYS_CLKS 10
>>>>>  
>>>>>  /**
>>>>>   * struct scp_domain_data - scp domain data for power on/off flow
>>>>> @@ -88,6 +89,8 @@
>>>>>   * @sram_pdn_bits: The mask for sram power control bits.
>>>>>   * @sram_pdn_ack_bits: The mask for sram power control acked bits.
>>>>>   * @basic_clk_name: The basic clocks required by this power domain.
>>>>> + * @subsys_clk_prefix: The prefix name of the clocks need to be enabled
>>>>> + *                     before releasing bus protection.
>>>>>   * @caps: The flag for active wake-up action.
>>>>>   * @bp_table: The mask table for multiple step bus protection.
>>>>>   */
>>>>> @@ -98,6 +101,7 @@ struct scp_domain_data {
>>>>>  	u32 sram_pdn_bits;
>>>>>  	u32 sram_pdn_ack_bits;
>>>>>  	const char *basic_clk_name[MAX_CLKS];
>>>>> +	const char *subsys_clk_prefix;
>>>>>  	u8 caps;
>>>>>  	struct bus_prot bp_table[MAX_STEPS];
>>>>>  };
>>>>> @@ -108,6 +112,7 @@ struct scp_domain {
>>>>>  	struct generic_pm_domain genpd;
>>>>>  	struct scp *scp;
>>>>>  	struct clk *clk[MAX_CLKS];
>>>>> +	struct clk *subsys_clk[MAX_SUBSYS_CLKS];
>>>>>  	const struct scp_domain_data *data;
>>>>>  	struct regulator *supply;
>>>>>  };
>>>>> @@ -301,16 +306,22 @@ static int scpsys_power_on(struct generic_pm_domain *genpd)
>>>>>  	val |= PWR_RST_B_BIT;
>>>>>  	writel(val, ctl_addr);
>>>>>  
>>>>> -	ret = scpsys_sram_enable(scpd, ctl_addr);
>>>>> +	ret = scpsys_clk_enable(scpd->subsys_clk, MAX_SUBSYS_CLKS);
>>>>
>>>> Why can't we enable the subsystem clocks together with the rest just after
>>>> enabeling the regulator?
>>>>
>>>
>>> Subsys CG could only be enabled when its own power domain is already
>>> turned ON, and vice versa.
>>> In the dt-binding doc we mentioned there would have two groups of
>>> clocks.
>>> e.g.,
>>> BASIC clocks need to be enabled before enabling the corresponding power
>>> domain.
>>> SUBSYS clocks need to be enabled before releasing the bus protection.
>>>
>>
>> Do I understand correctly that we could enable/disable all clocks in the same
>> place as long as we make sure that the the basic clocks are turned on before we
>> turn on the subsys clocks, correct?
>>
> 
> simply, yes
> 
>> So why do we need to implement this logic in the power-controller driver?
>> Shouldn't that be part of the common clock driver?
>>
> 
> we implement the subsys clock logic here just due to we already
> implement the bus protection flow in the power-controller driver.

I think in this driver is the correct place to implement bus protection flow.

> And If we don't enable the subsys clocks, bus protection cannot work.
> Hence, even the subsys power is on but the HW modules under this subsys
> power cannot access the bus though.

Ok, I understand that. But I understand that this should be fixed in the clock
driver. It seems the clock driver does not reflect the correct clock tree.
For example:
clocks CLK_MM_SMI_COMMON, CLK_MM_SMI_LARB0, CLK_MM_SMI_LARB1, CLK_MM_GALS_COMM0,
CLK_MM_GALS_COMM1, CLK_MM_GALS_CCU2MM, CLK_MM_GALS_IPU12MM, CLK_MM_GALS_IMG2MM,
CLK_MM_GALS_CAM2MM, CLK_MM_GALS_IPU2MM need the CLK_TOP_MUX_MM to be enabled
first. So I suppose CLK_TOP_MUX_MM is the parent clock of the other CLK_MM_*
clocks. If the clock tree is correctly described in the clock driver, then the
common clock framework will take care to enable CLK_TOP_MUX_MM when you try to
enable an CLK_MM_* clocks.

Why does that not work on mt8183? My impression after a quick look into the
clock driver is, that this should work.

Regards,
Matthias

> 
>> Regards,
>> Matthias
>>
>>>>>  	if (ret < 0)
>>>>>  		goto err_pwr_ack;
>>>>>  
>>>>> +	ret = scpsys_sram_enable(scpd, ctl_addr);
>>>>> +	if (ret < 0)
>>>>> +		goto err_sram;
>>>>> +
>>>>>  	ret = scpsys_bus_protect_disable(scpd);
>>>>>  	if (ret < 0)
>>>>> -		goto err_pwr_ack;
>>>>> +		goto err_sram;
>>>>>  
>>>>>  	return 0;
>>>>>  
>>>>> +err_sram:
>>>>> +	scpsys_clk_disable(scpd->subsys_clk, MAX_SUBSYS_CLKS);
>>>>>  err_pwr_ack:
>>>>>  	scpsys_clk_disable(scpd->clk, MAX_CLKS);
>>>>>  err_clk:
>>>>> @@ -337,6 +348,8 @@ static int scpsys_power_off(struct generic_pm_domain *genpd)
>>>>>  	if (ret < 0)
>>>>>  		goto out;
>>>>>  
>>>>> +	scpsys_clk_disable(scpd->subsys_clk, MAX_SUBSYS_CLKS);
>>>>> +
>>>>
>>>> Same here, why can't we disable the clocks in the scpsys_clk_disable call?
>>>>
>>>>>  	/* subsys power off */
>>>>>  	val = readl(ctl_addr);
>>>>>  	val |= PWR_ISO_BIT;
>>>>> @@ -374,6 +387,48 @@ static int scpsys_power_off(struct generic_pm_domain *genpd)
>>>>>  	return ret;
>>>>>  }
>>>>>  
>>>>> +static int init_subsys_clks(struct platform_device *pdev,
>>>>> +		const char *prefix, struct clk **clk)
>>>>> +{
>>>>> +	struct device_node *node = pdev->dev.of_node;
>>>>> +	u32 prefix_len, sub_clk_cnt = 0;
>>>>> +	struct property *prop;
>>>>> +	const char *clk_name;
>>>>> +
>>>>> +	if (!node) {
>>>>> +		dev_err(&pdev->dev, "Cannot find scpsys node: %ld\n",
>>>>> +			PTR_ERR(node));
>>>>> +		return PTR_ERR(node);
>>>>> +	}
>>>>> +
>>>>> +	prefix_len = strlen(prefix);
>>>>> +
>>>>> +	of_property_for_each_string(node, "clock-names", prop, clk_name) {
>>>>> +		if (!strncmp(clk_name, prefix, prefix_len) &&
>>>>> +				(clk_name[prefix_len] == '-')) {
>>>>> +			if (sub_clk_cnt >= MAX_SUBSYS_CLKS) {
>>>>> +				dev_err(&pdev->dev,
>>>>> +					"subsys clk out of range %d\n",
>>>>> +					sub_clk_cnt);
>>>>> +				return -ENOMEM;
>>>>
>>>> EINVAL maybe, ENOMEM seems wrong here.
>>>>
>>>
>>> OK, I'll fix with correct error.
>>>
>>>>> +			}
>>>>> +
>>>>> +			clk[sub_clk_cnt] = devm_clk_get(&pdev->dev,
>>>>> +						clk_name);
>>>>
>>>> Here we get hit by the bad design of this driver in the first place. As we need
>>>> the subsystem-name (eg mm-0, mm-1) to group clocks to one scp_domain.
>>>> I think we should better try to model the domains and subdomains in DTS and add
>>>> their clocks to it. This way we can also get rid of the scp_subdomain which can
>>>> hit it's limit anytime soon when we have a chip with a sub-subdomain.
>>>> That will need a new driver, but as it seems the mt8183 and the mt6765 have a
>>>> more complex design I think it is worth it.
>>>>
>>>> That said, given that you are in v11 already I understand that your motivation
>>>> to start over isn't the biggest. The problem is, any new driver will have new
>>>> bindings and won't work with older DTS. So adding a lot of stuff on top of a not
>>>> really nice driver isn't something I'm very keen on. On the other hand you
>>>> already put a lot of work into this solution.
>>>>
>>>> My proposal, I'll try to bake up a new driver this week. If I fail to deliver,
>>>> it's up to you to decide if you want to go on with the approach in this series
>>>> or try to work on the new one.
>>>
>>>> Regards,
>>>> Matthias
>>>>
>>>
>>> Thanks for considering our request.
>>>
>>>>> +
>>>>> +			if (IS_ERR(clk[sub_clk_cnt])) {
>>>>> +				dev_err(&pdev->dev,
>>>>> +					"Subsys clk get fail %ld\n",
>>>>> +					PTR_ERR(clk[sub_clk_cnt]));
>>>>> +				return PTR_ERR(clk[sub_clk_cnt]);
>>>>> +			}
>>>>> +			sub_clk_cnt++;
>>>>> +		}
>>>>> +	}
>>>>> +
>>>>> +	return sub_clk_cnt;
>>>>> +}
>>>>> +
>>>>>  static int init_basic_clks(struct platform_device *pdev, struct clk **clk,
>>>>>  			const char * const *name)
>>>>>  {
>>>>> @@ -466,6 +521,7 @@ static struct scp *init_scp(struct platform_device *pdev,
>>>>>  		struct scp_domain *scpd = &scp->domains[i];
>>>>>  		struct generic_pm_domain *genpd = &scpd->genpd;
>>>>>  		const struct scp_domain_data *data = &scp_domain_data[i];
>>>>> +		int clk_cnt;
>>>>
>>>> clk_cnt sounds to me like clock count, but the variable actually is only used to
>>>> check the return value of init_subsys_clks. Please rename it to ret or something
>>>> like this.
>>>>
>>>
>>> OK, I'll fix it.
>>>
>>>>>  
>>>>>  		pd_data->domains[i] = genpd;
>>>>>  		scpd->scp = scp;
>>>>> @@ -476,6 +532,18 @@ static struct scp *init_scp(struct platform_device *pdev,
>>>>>  		if (ret)
>>>>>  			return ERR_PTR(ret);
>>>>>  
>>>>> +		if (data->subsys_clk_prefix) {
>>>>> +			clk_cnt = init_subsys_clks(pdev,
>>>>> +					data->subsys_clk_prefix,
>>>>> +					scpd->subsys_clk);
>>>>> +			if (clk_cnt < 0) {
>>>>> +				dev_err(&pdev->dev,
>>>>> +					"%s: subsys clk unavailable\n",
>>>>> +					data->name);
>>>>> +				return ERR_PTR(clk_cnt);
>>>>> +			}
>>>>> +		}
>>>>> +
>>>>>  		genpd->name = data->name;
>>>>>  		genpd->power_off = scpsys_power_off;
>>>>>  		genpd->power_on = scpsys_power_on;
>>>>>
>>>
>>
>> _______________________________________________
>> Linux-mediatek mailing list
>> Linux-mediatek@xxxxxxxxxxxxxxxxxxx
>> http://lists.infradead.org/mailman/listinfo/linux-mediatek
>