Re: [PATCH V3 04/10] coresight: replicator: Move ACPI support from AMBA driver to platform driver

[Anshuman Khandual]

On 12/11/23 15:42, James Clark wrote:
> 
> 
> On 11/12/2023 07:51, Anshuman Khandual wrote:
>>
>>
>> On 12/8/23 11:09, Anshuman Khandual wrote:
>>> Add support for the dynamic replicator device in the platform driver, which
>>> can then be used on ACPI based platforms. This change would now allow
>>> runtime power management for repliacator devices on ACPI based systems.
>>>
>>> The driver would try to enable the APB clock if available. Also, rename the
>>> code to reflect the fact that it now handles both static and dynamic
>>> replicators.
>>>
>>> Cc: Lorenzo Pieralisi <lpieralisi@xxxxxxxxxx>
>>> Cc: Sudeep Holla <sudeep.holla@xxxxxxx>
>>> Cc: Suzuki K Poulose <suzuki.poulose@xxxxxxx>
>>> Cc: Mike Leach <mike.leach@xxxxxxxxxx>
>>> Cc: James Clark <james.clark@xxxxxxx>
>>> Cc: linux-acpi@xxxxxxxxxxxxxxx
>>> Cc: linux-arm-kernel@xxxxxxxxxxxxxxxxxxx
>>> Cc: linux-kernel@xxxxxxxxxxxxxxx
>>> Cc: coresight@xxxxxxxxxxxxxxxx
>>> Tested-by: Sudeep Holla <sudeep.holla@xxxxxxx> # Boot and driver probe only
>>> Acked-by: Sudeep Holla <sudeep.holla@xxxxxxx> # For ACPI related changes
>>> Signed-off-by: Anshuman Khandual <anshuman.khandual@xxxxxxx>
>>> ---
>>> Changes in V3:
>>>
>>> - Added commnets for 'drvdata->pclk'
>>> - Used coresight_init_driver()/coresight_remove_driver() helpers instead
>>> - Dropped pm_runtime_put() from replicator_probe()
>>> - Added pm_runtime_put() on success path in dynamic_replicator_probe()
>>> - Added pm_runtime_put() on success/error paths in
>>>   replicator_platform_probe()
>>>
>>>  drivers/acpi/arm64/amba.c                     |  1 -
>>>  .../coresight/coresight-replicator.c          | 81 ++++++++++---------
>>>  2 files changed, 42 insertions(+), 40 deletions(-)
>>>
>>> diff --git a/drivers/acpi/arm64/amba.c b/drivers/acpi/arm64/amba.c
>>> index 171b5c2c7edd..270f4e3819a2 100644
>>> --- a/drivers/acpi/arm64/amba.c
>>> +++ b/drivers/acpi/arm64/amba.c
>>> @@ -27,7 +27,6 @@ static const struct acpi_device_id amba_id_list[] = {
>>>  	{"ARMHC503", 0}, /* ARM CoreSight Debug */
>>>  	{"ARMHC979", 0}, /* ARM CoreSight TPIU */
>>>  	{"ARMHC97C", 0}, /* ARM CoreSight SoC-400 TMC, SoC-600 ETF/ETB */
>>> -	{"ARMHC98D", 0}, /* ARM CoreSight Dynamic Replicator */
>>>  	{"ARMHC9CA", 0}, /* ARM CoreSight CATU */
>>>  	{"ARMHC9FF", 0}, /* ARM CoreSight Dynamic Funnel */
>>>  	{"", 0},
>>> diff --git a/drivers/hwtracing/coresight/coresight-replicator.c b/drivers/hwtracing/coresight/coresight-replicator.c
>>> index b6be73034996..125b256cb8db 100644
>>> --- a/drivers/hwtracing/coresight/coresight-replicator.c
>>> +++ b/drivers/hwtracing/coresight/coresight-replicator.c
>>> @@ -31,6 +31,7 @@ DEFINE_CORESIGHT_DEVLIST(replicator_devs, "replicator");
>>>   * @base:	memory mapped base address for this component. Also indicates
>>>   *		whether this one is programmable or not.
>>>   * @atclk:	optional clock for the core parts of the replicator.
>>> + * @pclk:	APB clock if present, otherwise NULL
>>>   * @csdev:	component vitals needed by the framework
>>>   * @spinlock:	serialize enable/disable operations.
>>>   * @check_idfilter_val: check if the context is lost upon clock removal.
>>> @@ -38,6 +39,7 @@ DEFINE_CORESIGHT_DEVLIST(replicator_devs, "replicator");
>>>  struct replicator_drvdata {
>>>  	void __iomem		*base;
>>>  	struct clk		*atclk;
>>> +	struct clk		*pclk;
>>>  	struct coresight_device	*csdev;
>>>  	spinlock_t		spinlock;
>>>  	bool			check_idfilter_val;
>>> @@ -243,6 +245,10 @@ static int replicator_probe(struct device *dev, struct resource *res)
>>>  			return ret;
>>>  	}
>>>  
>>> +	drvdata->pclk = coresight_get_enable_apb_pclk(dev);
>>> +	if (IS_ERR(drvdata->pclk))
>>> +		return -ENODEV;
>>> +
>>>  	/*
>>>  	 * Map the device base for dynamic-replicator, which has been
>>>  	 * validated by AMBA core
>>> @@ -285,7 +291,6 @@ static int replicator_probe(struct device *dev, struct resource *res)
>>>  	}
>>>  
>>>  	replicator_reset(drvdata);
>>> -	pm_runtime_put(dev);
>>>  
>>>  out_disable_clk:
>>>  	if (ret && !IS_ERR_OR_NULL(drvdata->atclk))
>>> @@ -301,29 +306,31 @@ static int replicator_remove(struct device *dev)
>>>  	return 0;
>>>  }
>>>  
>>> -static int static_replicator_probe(struct platform_device *pdev)
>>> +static int replicator_platform_probe(struct platform_device *pdev)
>>>  {
>>> +	struct resource *res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
>>>  	int ret;
>>>  
>>>  	pm_runtime_get_noresume(&pdev->dev);
>>>  	pm_runtime_set_active(&pdev->dev);
>>>  	pm_runtime_enable(&pdev->dev);
>>>  
>>> -	/* Static replicators do not have programming base */
>>> -	ret = replicator_probe(&pdev->dev, NULL);
>>> -
>>> -	if (ret) {
>>> -		pm_runtime_put_noidle(&pdev->dev);
>>> -		pm_runtime_disable(&pdev->dev);
>>> -	}
>>> +	ret = replicator_probe(&pdev->dev, res);
>>> +	pm_runtime_put(&pdev->dev);
>>
>> I believe pm_runtime_disable() would still be needed on the error path. Otherwise
>> pm_runtime_enable() will remain unbalanced on this error path when the replicator
>> module could not be loaded.
>>
>> --- a/drivers/hwtracing/coresight/coresight-replicator.c
>> +++ b/drivers/hwtracing/coresight/coresight-replicator.c
>> @@ -317,6 +317,8 @@ static int replicator_platform_probe(struct platform_device *pdev)
>>  
>>         ret = replicator_probe(&pdev->dev, res);
>>         pm_runtime_put(&pdev->dev);
>> +       if (ret)
>> +               pm_runtime_disable(&pdev->dev);
>>  
>>         return ret;
>>  }
>>
>> Similar constructs in this error path are also required in all other drivers (except
>> cpu debug) as well.
> 
> Would that not need to be done as a fixes commit first with more detail
> about the issue if that's true? Maybe simulate the error and paste any
> error logs. For example etm4 already has this:

There are existing inconsistencies e.g between etm4 and replicator regarding
whether pm_runtime_disable() is called or not. So the idea here was to get
all remaining devices in line with replicator and debug devices method.

I don't have continuous access to an ACPI based coresight platform, creating
some challenges - although still trying to get access to such a system. But
wondering - it might be possible to simulate these success and error paths,
without having real ACPI based coresight platform devices.

> 
>   	pm_runtime_get_noresume(&pdev->dev);
> 	pm_runtime_set_active(&pdev->dev);
> 	pm_runtime_enable(&pdev->dev);
> 
> 	ret = etm4_probe(&pdev->dev);
> 
> 	pm_runtime_put(&pdev->dev);
> 
> I'm wondering if the disable is already covered by the platform code if
> the probe fails so no change is required?

AFAICT pm_runtime_enable()/pm_runtime_disable() goes in a pair for them to
remain balanced, there is a increasing/decreasing counter to ensure such a
balance is established and so without a corresponding pm_runtime_disable()
above etm4 path looks problematic. I guess this might just need fixing, as
a pre-requisite.

> 
> 	if (drv->probe) {
> 		ret = drv->probe(dev);
> 		if (ret)
> 			dev_pm_domain_detach(_dev, true);
> 	}