[PATCH] coresight: Add coresight name support

[Jinlong Mao]

On 1/3/2024 7:33 PM, Suzuki K Poulose wrote:
> On 28/12/2023 11:26, James Clark wrote:
>>
>>
>> On 28/12/2023 09:33, Mao Jinlong wrote:
>>> Add coresight name support for custom names which will be
>>> easy to identify the device by the name.
>>>
>>
>> I suppose this is more of a V2 because the subject is the same as the
>> one sent earlier this year. But it looks like the discussion on the
>> previous one wasn't resolved.
>>
>> With the main issues to solve being:
>>
>>   * It would be nice to use the existing root node name instead of adding
>>     a new property. But at the same time DT nodes are supposed to have
>>     generic names.
>>
>>   * This only works for DT and not ACPI
>>
>> To me it seems like adding the new property is just a "cheat" to get
>> around not being allowed to have a specific name for the root node. But
>> if we admit that we need a name I don't see the benefit of not putting
>> the name where the node is already named.
>>
>> Using the root node name at this point would also undo the hard coded
>> per-cpu naming of the CTI and ETM devices, so maybe it would be nice,
>> but it's just too late. That means that a new field is necessary.
> 
> The CTI and ETM can be handled as special cases, like they are
> already done and fall back to the nodename for the rest ?
> But, I thought the node names must be generic (e.g, cti) and doesn't
> really solve the naming requirements for naming CTIs. (e.g,
> <device>_tpda, etr_cti). Is there something I missed ?
> 
>> Although that field could be a boolean like "use-root-name-for-display"
>> or something like that. In the end it probably doesn't really make a
>> difference whether it's that or a name string. >
>> And maybe the answer to the ACPI question is just that if anyone needs
>> it, they can add it in the future. It doesn't seem like it would
>> conflict with anything we do here.
>>
>>> Signed-off-by: Mao Jinlong <quic_jinlmao at quicinc.com>
>>> ---
>>>   .../hwtracing/coresight/coresight-cti-core.c  | 20 ++++++++------
>>>   drivers/hwtracing/coresight/coresight-dummy.c | 10 ++++---
>>>   .../hwtracing/coresight/coresight-platform.c  | 27 +++++++++++++++++++
>>>   drivers/hwtracing/coresight/coresight-tpdm.c  | 10 ++++---
>>>   include/linux/coresight.h                     |  1 +
>>>   5 files changed, 53 insertions(+), 15 deletions(-)
>>>
>>> diff --git a/drivers/hwtracing/coresight/coresight-cti-core.c 
>>> b/drivers/hwtracing/coresight/coresight-cti-core.c
>>> index 3999d0a2cb60..60a1e76064a9 100644
>>> --- a/drivers/hwtracing/coresight/coresight-cti-core.c
>>> +++ b/drivers/hwtracing/coresight/coresight-cti-core.c
>>> @@ -902,14 +902,18 @@ static int cti_probe(struct amba_device *adev, 
>>> const struct amba_id *id)
>>>       /* default to powered - could change on PM notifications */
>>>       drvdata->config.hw_powered = true;
>>> -    /* set up device name - will depend if cpu bound or otherwise */
>>> -    if (drvdata->ctidev.cpu >= 0)
>>> -        cti_desc.name = devm_kasprintf(dev, GFP_KERNEL, "cti_cpu%d",
>>> -                           drvdata->ctidev.cpu);
>>> -    else
>>> -        cti_desc.name = coresight_alloc_device_name(&cti_sys_devs, 
>>> dev);
>>
>> Can we put the new name stuff inside coresight_alloc_device_name()? Then
>> it happens by default for every device.
> 
> +1
> 
>>
>> I know Suzuki said previously to do it per-device, but the new DT
>> property is just "coresight-name", so it's generic. Rather than being
>> specific like "cti-name". So I don't see the benefit of duplicating the
>> code at this point if we do decide to do it.
> 
> My suggestion was to name the device based on the specific device rather
> than following a generic rule for all device. e.g., A TPDM connected to 
> modem, could be named as such based on the platform information. It
> could be any means, for e.g., tpdm nodes are always children nodes of
> the devices they are connected to ? or could have a phandle to point to 
> the device they are monitoring etc. And the name could be created from
> the "monitoring device name" + tpdm. Also, we do this for CPU bound CTI
> and ETMs already, where we name them based on the CPU.

TPDM can only connect to the funnel or TPDA. The system TPDM is 
monitoring may not have the device node in DT.

> 
> But then the "nodename" is something we explored and it looks like
> may not be an option.
> 
>>
>>> -    if (!cti_desc.name)
>>> -        return -ENOMEM;
>>> +    cti_desc.name = coresight_get_device_name(dev);
>>> +    if (!cti_desc.name) {
>>> +        /* set up device name - will depend if cpu bound or 
>>> otherwise */
>>> +        if (drvdata->ctidev.cpu >= 0)
>>> +            cti_desc.name = devm_kasprintf(dev, GFP_KERNEL, 
>>> "cti_cpu%d",
>>> +                               drvdata->ctidev.cpu);
>>> +        else {
>>> +            cti_desc.name = 
>>> coresight_alloc_device_name(&cti_sys_devs, dev);
>>> +            if (!cti_desc.name)
>>> +                return -ENOMEM;
>>> +        }
>>> +    }
> 
> For these special cases, i.e., CPU bound, we should handle them with 
> priority.
> 
> if (drvdata->ctidev.cpu >= 0)
>      name = devm_kasprintf(... "cti_cpu%d", .. cpu);
> else
>          name = coresight_alloc_device_name(...);
> 
>>
>>>       /* setup CPU power management handling for CPU bound CTI 
>>> devices. */
>>>       ret = cti_pm_setup(drvdata);
>>> diff --git a/drivers/hwtracing/coresight/coresight-dummy.c 
>>> b/drivers/hwtracing/coresight/coresight-dummy.c
>>> index e4deafae7bc2..b19cd400df79 100644
>>> --- a/drivers/hwtracing/coresight/coresight-dummy.c
>>> +++ b/drivers/hwtracing/coresight/coresight-dummy.c
>>> @@ -76,10 +76,12 @@ static int dummy_probe(struct platform_device *pdev)
>>>       struct coresight_desc desc = { 0 };
>>>       if (of_device_is_compatible(node, "arm,coresight-dummy-source")) {
>>> -
>>> -        desc.name = coresight_alloc_device_name(&source_devs, dev);
>>> -        if (!desc.name)
>>> -            return -ENOMEM;
>>> +        desc.name = coresight_get_device_name(dev);
>>> +        if (!desc.name) {
>>> +            desc.name = coresight_alloc_device_name(&source_devs, dev);
>>> +            if (!desc.name)
>>> +                return -ENOMEM;
>>> +        }
>>>           desc.type = CORESIGHT_DEV_TYPE_SOURCE;
>>>           desc.subtype.source_subtype =
>>> diff --git a/drivers/hwtracing/coresight/coresight-platform.c 
>>> b/drivers/hwtracing/coresight/coresight-platform.c
>>> index 9d550f5697fa..284aa22a06b7 100644
>>> --- a/drivers/hwtracing/coresight/coresight-platform.c
>>> +++ b/drivers/hwtracing/coresight/coresight-platform.c
>>> @@ -183,6 +183,18 @@ static int of_coresight_get_cpu(struct device *dev)
>>>       return cpu;
>>>   }
>>> +static const char *of_coresight_get_device_name(struct device *dev)
>>> +{
>>> +    const char *name = NULL;
>>> +
>>> +    if (!dev->of_node)
>>> +        return NULL;
>>> +
>>> +    of_property_read_string(dev->of_node, "coresight-name", &name);
>>
>> Do you need to update the binding docs with this new property?
>>
>> Also a minor nit: Maybe "display-name" is better? "Coresight" is
>> implied, and the node is already named, although that node name isn't
>> used for display purposes, but this one is.
> 
> On that front, the name is used as a "device" name and not simply 
> display. So, even "device-name" sounds more appropriate.
> 
I will use the device-name.

Thanks
Jinlong Mao
> Suzuki
> 
>>
>> Thanks
>> James
>