linux-kernel - Re: [PATCH v5 13/22] dma: qcom: bam_dma: Add support to initialize interconnect path

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <YZKetkkXYR5Vdz1w@builder.lan>
Date:   Mon, 15 Nov 2021 11:53:58 -0600
From:   Bjorn Andersson <bjorn.andersson@...aro.org>
To:     Bhupesh Sharma <bhupesh.sharma@...aro.org>
Cc:     linux-arm-msm@...r.kernel.org, linux-crypto@...r.kernel.org,
        bhupesh.linux@...il.com, linux-kernel@...r.kernel.org,
        devicetree@...r.kernel.org, robh+dt@...nel.org, agross@...nel.org,
        herbert@...dor.apana.org.au, davem@...emloft.net,
        stephan@...hold.net, Thara Gopinath <thara.gopinath@...aro.org>
Subject: Re: [PATCH v5 13/22] dma: qcom: bam_dma: Add support to initialize
 interconnect path

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

> From: Thara Gopinath <thara.gopinath@...aro.org>
> 
> BAM dma engine associated with certain hardware blocks could require
> relevant interconnect pieces be initialized prior to the dma engine
> initialization. For e.g. crypto bam dma engine on sm8250. Such requirement
> is passed on to the bam dma driver from dt via the "interconnects"
> property.  Add support in bam_dma driver to check whether the interconnect
> path is accessible/enabled prior to attempting driver intializations.
> 

This patch acquires the path, presumably between BAM and DDR, but I
don't see that it actually do anything with it.

So if this makes sm8250 work I presume it's because you'll get probe
deferred until the interconnect provider is in place and it will hold
buses at max speed until sync_state - and then in runtime perhaps the
crypto driver hides the fact that BAM doesn't vote for its bandwidth?

I'm sceptical about hiding behind such circumstances, but at least the
commit message should be clear on what's going on - or you need cast
some bandwidth votes when the block is expected to transfer data.

> Cc: Bjorn Andersson <bjorn.andersson@...aro.org>
> Cc: Rob Herring <robh+dt@...nel.org>
> Signed-off-by: Bhupesh Sharma <bhupesh.sharma@...aro.org>
> [Make header file inclusion alphabetical and use 'devm_of_icc_get()']
> Signed-off-by: Thara Gopinath <thara.gopinath@...aro.org>

As Vladimir pointed out, the S-o-b of a patch should be read in
chronological order.

Please read the section about Developer's Certificate of Origin (link
below), it should make it clear why the specific order is important.

https://www.kernel.org/doc/html/latest/process/submitting-patches.html#developer-s-certificate-of-origin-1-1

> ---
>  drivers/dma/qcom/bam_dma.c | 11 +++++++++++
>  1 file changed, 11 insertions(+)
> 
> diff --git a/drivers/dma/qcom/bam_dma.c b/drivers/dma/qcom/bam_dma.c
> index c8a77b428b52..19fb17db467f 100644
> --- a/drivers/dma/qcom/bam_dma.c
> +++ b/drivers/dma/qcom/bam_dma.c
> @@ -26,6 +26,7 @@
>  #include <linux/kernel.h>
>  #include <linux/io.h>
>  #include <linux/init.h>
> +#include <linux/interconnect.h>
>  #include <linux/slab.h>
>  #include <linux/module.h>
>  #include <linux/interrupt.h>
> @@ -392,6 +393,7 @@ struct bam_device {
>  	const struct reg_offset_data *layout;
>  
>  	struct clk *bamclk;
> +	struct icc_path *mem_path;
>  	int irq;
>  
>  	/* dma start transaction tasklet */
> @@ -1284,6 +1286,15 @@ static int bam_dma_probe(struct platform_device *pdev)
>  		return ret;
>  	}
>  
> +	/* Ensure that interconnects are initialized */
> +	bdev->mem_path = devm_of_icc_get(bdev->dev, "memory");
> +

Please drop this empty line.

> +	if (IS_ERR(bdev->mem_path)) {
> +		ret = PTR_ERR(bdev->mem_path);
> +		dev_err(bdev->dev, "failed to acquire icc path %d\n", ret);

We typically want to avoid printing error messages when ret is
-EPROBE_DEFER. The best way around this is to utilize the
dev_err_probe() helper function.

Replace the two lines above with:

		ret = dev_err_probe(bdev->dev, PTR_ERR(bdev->mem_path),
				    "failed to acquire icc path\n")

> +		goto err_disable_clk;

If you move the devm_of_icc_get() right before clk_prepare_enable()
(still after devm_clk_get*()) you can simply return dev_err_probe() -
and will avoid toggling the CE clock unnecessarily on EPROBE_DEFER.

Regards,
Bjorn

> +	}
> +
>  	ret = bam_init(bdev);
>  	if (ret)
>  		goto err_disable_clk;
> -- 
> 2.31.1
>