Re: [PATCH v6 2/4] dmaengine: tegra: Add tegra gpcdma driver

[Jon Hunter <jonathanh@nvidia.com>]

On 23/09/2021 13:51, Akhil R wrote:
>> On 22/09/2021 15:46, Akhil R wrote:
>>>
>>>> On 17/09/2021 15:11, Akhil R wrote:
>>>>> +static int tegra_dma_slave_config(struct dma_chan *dc,
>>>>> +				  struct dma_slave_config *sconfig) {
>>>>> +	struct tegra_dma_channel *tdc = to_tegra_dma_chan(dc);
>>>>> +
>>>>> +	if (tdc->dma_desc) {
>>>>> +		dev_err(tdc2dev(tdc), "Configuration not allowed\n");
>>>>> +		return -EBUSY;
>>>>> +	}
>>>>> +
>>>>> +	memcpy(&tdc->dma_sconfig, sconfig, sizeof(*sconfig));
>>>>> +	if (tdc->slave_id == -1)
>>>>> +		tdc->slave_id = sconfig->slave_id;
>>>>> +
>>>>> +	tdc->config_init = true;
>>>>> +	return 0;
>>>>> +}
>>>>
>>>> So you have a function to reserve a slave ID, but you don't check
>>>> here if it is already reserved.
>>> slave-id is reserved considering the direction as well.
>>> 'direction' is available only during prep_slave_sg function, I guess.
>>
>> Sorry I don't understand what you mean by that.
> I mean, it would not be possible to check if the sid is in use without knowing
> if the direction is MEM_TO_DEV or DEV_TO_MEM. The bitmask to check the
> sid reservation is separate for MEM_TO_DEV and DEV_TO_MEM.
> To get the direction parameter, we would need to wait till dma_prep_slave_sg
> is called, I guess. I saw in the documentation that the 'direction' element in
> dma_slave_config struct is deprecated and should use the value passed to
> dma_prep_slave_sg().

Do we even need to worry about slave_id here? Typically the slave_id is 
coming from device-tree and so is handled by tegra_dma_of_xlate(). I 
would drop this unless there is an actual use-case we need to support 
that uses this.

Jon

-- 
nvpublic