Re: [EXTERNAL] Re: [PATCH v4 4/5] soc: ti: pruss: Add helper functions to set GPI mode, MII_RT_event and XFR

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



Hi,

On 16/03/2023 15:11, Md Danish Anwar wrote:
> 
> 
> On 16/03/23 17:49, Roger Quadros wrote:
>>
>>
>> On 16/03/2023 13:44, Md Danish Anwar wrote:
>>>
>>> On 16/03/23 17:06, Roger Quadros wrote:
>>>> Hi,
>>>>
>>>> On 16/03/2023 13:05, Md Danish Anwar wrote:
>>>>> Hi Roger,
>>>>>
>>>>> On 15/03/23 17:52, Roger Quadros wrote:
>>>>>>
>>>>>>
>>>>>> On 13/03/2023 13:11, MD Danish Anwar wrote:
>>>>>>> From: Suman Anna <s-anna@xxxxxx>
>>>>>>>
>>>>>>> The PRUSS CFG module is represented as a syscon node and is currently
>>>>>>> managed by the PRUSS platform driver. Add easy accessor functions to set
>>>>>>> GPI mode, MII_RT event enable/disable and XFR (XIN XOUT) enable/disable
>>>>>>> to enable the PRUSS Ethernet usecase. These functions reuse the generic
>>>>>>> pruss_cfg_update() API function.
>>>>>>>
>>>>>>> Signed-off-by: Suman Anna <s-anna@xxxxxx>
>>>>>>> Co-developed-by: Grzegorz Jaszczyk <grzegorz.jaszczyk@xxxxxxxxxx>
>>>>>>> Signed-off-by: Grzegorz Jaszczyk <grzegorz.jaszczyk@xxxxxxxxxx>
>>>>>>> Signed-off-by: Puranjay Mohan <p-mohan@xxxxxx>
>>>>>>> Signed-off-by: MD Danish Anwar <danishanwar@xxxxxx>
>>>>>>> ---
>>>>>>>  drivers/soc/ti/pruss.c           | 60 ++++++++++++++++++++++++++++++++
>>>>>>>  include/linux/remoteproc/pruss.h | 22 ++++++++++++
>>>>>>>  2 files changed, 82 insertions(+)
>>>>>>>
>>>>>>> diff --git a/drivers/soc/ti/pruss.c b/drivers/soc/ti/pruss.c
>>>>>>> index 26d8129b515c..2f04b7922ddb 100644
>>>>>>> --- a/drivers/soc/ti/pruss.c
>>>>>>> +++ b/drivers/soc/ti/pruss.c
>>>>>>> @@ -203,6 +203,66 @@ static int pruss_cfg_update(struct pruss *pruss, unsigned int reg,
>>>>>>>  	return regmap_update_bits(pruss->cfg_regmap, reg, mask, val);
>>>>>>>  }
>>>>>>>  
>>>>>>> +/**
>>>>>>> + * pruss_cfg_gpimode() - set the GPI mode of the PRU
>>>>>>> + * @pruss: the pruss instance handle
>>>>>>> + * @pru_id: id of the PRU core within the PRUSS
>>>>>>> + * @mode: GPI mode to set
>>>>>>> + *
>>>>>>> + * Sets the GPI mode for a given PRU by programming the
>>>>>>> + * corresponding PRUSS_CFG_GPCFGx register
>>>>>>> + *
>>>>>>> + * Return: 0 on success, or an error code otherwise
>>>>>>> + */
>>>>>>> +int pruss_cfg_gpimode(struct pruss *pruss, enum pruss_pru_id pru_id,
>>>>>>> +		      enum pruss_gpi_mode mode)
>>>>>>> +{
>>>>>>> +	if (pru_id < 0 || pru_id >= PRUSS_NUM_PRUS)
>>>>>>> +		return -EINVAL;
>>>>>>> +
>>>>>>> +	if (mode < 0 || mode > PRUSS_GPI_MODE_MAX)
>>>>>>> +		return -EINVAL;
>>>>>>> +
>>>>>>> +	return pruss_cfg_update(pruss, PRUSS_CFG_GPCFG(pru_id),
>>>>>>> +				PRUSS_GPCFG_PRU_GPI_MODE_MASK,
>>>>>>> +				mode << PRUSS_GPCFG_PRU_GPI_MODE_SHIFT);
>>>>>>> +}
>>>>>>> +EXPORT_SYMBOL_GPL(pruss_cfg_gpimode);
>>>>>>> +
>>>>>>> +/**
>>>>>>> + * pruss_cfg_miirt_enable() - Enable/disable MII RT Events
>>>>>>> + * @pruss: the pruss instance
>>>>>>> + * @enable: enable/disable
>>>>>>> + *
>>>>>>> + * Enable/disable the MII RT Events for the PRUSS.
>>>>>>> + *
>>>>>>> + * Return: 0 on success, or an error code otherwise
>>>>>>> + */
>>>>>>> +int pruss_cfg_miirt_enable(struct pruss *pruss, bool enable)
>>>>>>> +{
>>>>>>> +	u32 set = enable ? PRUSS_MII_RT_EVENT_EN : 0;
>>>>>>> +
>>>>>>> +	return pruss_cfg_update(pruss, PRUSS_CFG_MII_RT,
>>>>>>> +				PRUSS_MII_RT_EVENT_EN, set);
>>>>>>> +}
>>>>>>> +EXPORT_SYMBOL_GPL(pruss_cfg_miirt_enable);
>>>>>>> +
>>>>>>> +/**
>>>>>>> + * pruss_cfg_xfr_enable() - Enable/disable XIN XOUT shift functionality
>>>>>>> + * @pruss: the pruss instance
>>>>>>> + * @enable: enable/disable
>>>>>>> + * @mask: Mask for PRU / RTU
>>>>>>
>>>>>> You should not expect the user to provide the mask but only
>>>>>> the core type e.g. 
>>>>>>
>>>>>> enum pru_type {
>>>>>>         PRU_TYPE_PRU = 0,
>>>>>>         PRU_TYPE_RTU,
>>>>>>         PRU_TYPE_TX_PRU,
>>>>>>         PRU_TYPE_MAX,
>>>>>> };
>>>>>>
>>>>>> Then you figure out the mask in the function.
>>>>>> Also check for invalid pru_type and return error if so.
>>>>>>
>>>>>
>>>>> Sure Roger, I will create a enum and take it as parameter in API. Based on
>>>>> these enum I will calculate mask and do XFR shifting inside the API
>>>>> pruss_cfg_xfr_enable().
>>>>>
>>>>> There are two registers for XFR shift.
>>>>>
>>>>> #define PRUSS_SPP_XFER_SHIFT_EN                 BIT(1)
>>>>> #define PRUSS_SPP_RTU_XFR_SHIFT_EN              BIT(3)
>>>>>
>>>>> For PRU XFR shifting, the mask should be PRUSS_SPP_XFER_SHIFT_EN,
>>>>> for RTU shifting mask should be PRUSS_SPP_RTU_XFR_SHIFT_EN and for PRU and RTU
>>>>> shifting mask should be (PRUSS_SPP_XFER_SHIFT_EN | PRUSS_SPP_RTU_XFR_SHIFT_EN)
>>>>>
>>>>> So the enum would be something like this.
>>>>>
>>>>> /**
>>>>>  * enum xfr_shift_type - XFR shift type
>>>>>  * @XFR_SHIFT_PRU: Enables XFR shift for PRU
>>>>>  * @XFR_SHIFT_RTU: Enables XFR shift for RTU
>>>>>  * @XFR_SHIFT_PRU_RTU: Enables XFR shift for both PRU and RTU
>>>>
>>>> This is not required. User can call the API twice. once for PRU and once for RTU.
>>>>
>>>>>  * @XFR_SHIFT_MAX: Total number of XFR shift types available.
>>>>>  *
>>>>>  */
>>>>>
>>>>> enum xfr_shift_type {
>>>>>         XFR_SHIFT_PRU = 0,
>>>>>         XFR_SHIFT_RTU,
>>>>>         XFR_SHIFT_PRU_RTU,
>>>>>         XFR_SHIFT_MAX,
>>>>> };
>>>>
>>>> Why do you need this new enum definition?
>>>> We already have pru_type defined somewhere. You can move it to a public header
>>>> if not there yet.
>>>>
>>>> enum pru_type {
>>>>          PRU_TYPE_PRU = 0,
>>>>          PRU_TYPE_RTU,
>>>>          PRU_TYPE_TX_PRU,
>>>>          PRU_TYPE_MAX,
>>>> };
>>>>
>>>
>>> This enum is present in drivers/remoteproc/pru_rproc.c file. But the problem
>>> with this enum is that in [1] we need to enable XFR shift for both PRU and RTU
>>> for which the mask will be OR of PRUSS_SPP_XFER_SHIFT_EN (mask for PRU) and
>>> PRUSS_SPP_RTU_XFR_SHIFT_EN (mask of RTU).
>>
>> Is there any limitation that you have to enable both simultaneously?
>> The driver can first do enable for PRU and then later for RTU.
>>
>> As you will do a read modify write, the previous enable state of register
>> shouldn't be affected.
>>
>>>
>>> Now this enum doesn't have a field for both PRU and RTU. Also we don't need
>>> need the XFR shift for PRU_TYPE_TX_PRU as only two XFR shift register bits are
>>> defined.
>>
>> That is OK. You can return error if not RTU or PRU.
>>
>>>
>>> That is why I thought of introducing new enum.
>>>
>>> [1] drivers/net/ethernet/ti/icssg_config.c
>>>
>>> /* enable XFR shift for PRU and RTU */
>>> 	mask = PRUSS_SPP_XFER_SHIFT_EN | PRUSS_SPP_RTU_XFR_SHIFT_EN;
>>
>> Driver can do like so
>>
>> 	pruss_cfg_xfr_enable(pruss, PRU_TYPE_PRU, true);
>> 	pruss_cfg_xfr_enable(pruss, PRU_TYPE_RTU, true);
>>
>> The second call should not disable the PRU XFR as you will do a
>> read-modify-write only affecting the RTU bit.
> 
> Sure, then I will use the existing enum pru_type.
> 
> The enum pru_type is currently in drivers/remoteproc/pruss.c I will move this
> enum definition from there to include/linux/remoteproc/pruss.h

There are 2 public pruss.h files.
	include/linux/remoteproc/pruss.h
and
	include/linux/pruss_driver.h

Why is that and when to use what?

cheers,
-roger



[Index of Archives]     [Linux Arm (vger)]     [ARM Kernel]     [ARM MSM]     [Linux Tegra]     [Linux WPAN Networking]     [Linux Wireless Networking]     [Maemo Users]     [Linux USB Devel]     [Video for Linux]     [Linux Audio Users]     [Yosemite Trails]     [Linux Kernel]     [Linux SCSI]

  Powered by Linux