Re: [PATCH v4 1/3] dmaengine: add dma_get_slave_sg_limits()

[Lars-Peter Clausen <lars@metafoo.de>]

[...]
>   *	struct with auxiliary transfer status information, otherwise the call
>   *	will just return a simple status code
>   * @device_issue_pending: push pending transactions to hardware
> + * @device_slave_sg_limits: return the slave SG capabilities
>   */
>  struct dma_device {
>  
> @@ -602,6 +615,9 @@ struct dma_device {
>  					    dma_cookie_t cookie,
>  					    struct dma_tx_state *txstate);
>  	void (*device_issue_pending)(struct dma_chan *chan);
> +	struct dma_slave_sg_limits *(*device_slave_sg_limits)(
> +		struct dma_chan *chan, enum dma_slave_buswidth addr_width,
> +		u32 maxburst);

In my opinion it is better to pass in a pointer to a dma_slave_sg_limits
struct and let the driver fill it. Instead of passing back a pointer to an
internal structure. It is kind of problematic because you never really know
when the user is done using the struct and you don't know when it is safe to
free or reuse it. E.g. in your implementation for the edma driver if the
function is called with different parameters for the same channel, the
previous result will also be overwritten.

- Lars