Re: [PATCH v5 18/22] crypto: qce: Defer probing if BAM dma channel is not yet initialized

[Bjorn Andersson]

On Wed 10 Nov 04:59 CST 2021, Bhupesh Sharma wrote:

> Since the Qualcomm qce crypto driver needs the BAM dma driver to be
> setup first (to allow crypto operations), it makes sense to defer
> the qce crypto driver probing in case the BAM dma driver is not yet
> probed.
> 

To me this sentence implies that qce_crypto_probe() doesn't return
-EPROBE_DEFER from qce_dma_request(), but looking at the code I don't
see why this would be...

> Move the code leg requesting dma channels earlier in the
> probe() flow. This fixes the qce probe failure issues when both qce
> and BMA dma are compiled as static part of the kernel.
> 

As far as I can tell the only actual difference is that you're moving
the qce_dma_request() above the icc_set_bw() call and the three
clk_prepare_enable() calls.

This is a very valid optimization, but where does qce_crypto_probe()
fail and why does it need the BAM driver to have probed before we try to
turn on the clocks and (our) interconnect vote?

> Cc: Thara Gopinath <thara.gopinath@xxxxxxxxxx>
> Cc: Bjorn Andersson <bjorn.andersson@xxxxxxxxxx>
> Signed-off-by: Bhupesh Sharma <bhupesh.sharma@xxxxxxxxxx>
> ---
>  drivers/crypto/qce/core.c | 23 ++++++++++++++---------
>  1 file changed, 14 insertions(+), 9 deletions(-)
> 
> diff --git a/drivers/crypto/qce/core.c b/drivers/crypto/qce/core.c
> index 7c90401a2ef1..84ed9e253d5d 100644
> --- a/drivers/crypto/qce/core.c
> +++ b/drivers/crypto/qce/core.c
> @@ -209,9 +209,19 @@ static int qce_crypto_probe(struct platform_device *pdev)
>  	if (ret < 0)
>  		return ret;
>  
> +	/* qce driver requires BAM dma driver to be setup first.
> +	 * In case the dma channel are not set yet, this check
> +	 * helps use to return -EPROBE_DEFER earlier.
> +	 */

This comment warrants the change, but I don't see that it will add value
in the code once the patch is merged. Please drop it.

> +	ret = qce_dma_request(qce->dev, &qce->dma);
> +	if (ret)

I presume this is where your added dev_err() in err: comes in handy. I
definitely think this warrants an error print; so return dev_err_probe()
here would be the right thing to do (in the previous patch).

> +		return ret;
> +
>  	qce->mem_path = devm_of_icc_get(qce->dev, "memory");
> -	if (IS_ERR(qce->mem_path))
> -		return PTR_ERR(qce->mem_path);

But I see no reason for moving it above devm_of_icc_get() and the
devm_clk_get*() calls - as they don't actually do anything, but moving
qce_dma_request() above them forces the introduction of goto here.

Regards,
Bjorn

> +	if (IS_ERR(qce->mem_path)) {
> +		ret = PTR_ERR(qce->mem_path);
> +		goto err;
> +	}
>  
>  	qce->core = devm_clk_get_optional(qce->dev, "core");
>  	if (IS_ERR(qce->core)) {
> @@ -247,10 +257,6 @@ static int qce_crypto_probe(struct platform_device *pdev)
>  	if (ret)
>  		goto err_clks_iface;
>  
> -	ret = qce_dma_request(qce->dev, &qce->dma);
> -	if (ret)
> -		goto err_clks;
> -
>  	ret = qce_check_version(qce);
>  	if (ret)
>  		goto err_clks;
> @@ -265,12 +271,10 @@ static int qce_crypto_probe(struct platform_device *pdev)
>  
>  	ret = qce_register_algs(qce);
>  	if (ret)
> -		goto err_dma;
> +		goto err_clks;
>  
>  	return 0;
>  
> -err_dma:
> -	qce_dma_release(&qce->dma);
>  err_clks:
>  	clk_disable_unprepare(qce->bus);
>  err_clks_iface:
> @@ -280,6 +284,7 @@ static int qce_crypto_probe(struct platform_device *pdev)
>  err_mem_path_disable:
>  	icc_set_bw(qce->mem_path, 0, 0);
>  err:
> +	qce_dma_release(&qce->dma);
>  	dev_err(dev, "%s failed : %d\n", __func__, ret);
>  	return ret;
>  }
> -- 
> 2.31.1
>