crypto: caam: map job ring registers without claiming region

[PATCH] crypto: caam: map job ring registers without claiming region

Posted by Rosen Penev 3 weeks, 3 days ago

devm_platform_ioremap_resource() ends up calling request_mem_region(),
which fails with -EBUSY on i.MX SoCs:

There the job rings are modelled as sub-regions of their parent
fsl,sec-v4.0 register window, and caam_probe() already reserves the
whole window exclusively via devm_of_iomap() before the children are
populated.  Every job ring therefore collides with its own parent and
fails to probe, taking the hardware RNG offline (seen on i.MX6, i.MX7
and i.MX8 boards such as colibri-imx7 and verdin-imx8mp).

Map the job ring registers with devm_ioremap() instead, which does not
claim the (already owned) region.

Fixes: 9a955c0a7d11 ("crypto: caam - simplify probe resource and IRQ handling")
Assisted-by: opencode:big-pickle
Signed-off-by: Rosen Penev <rosenp@gmail.com>
---
 drivers/crypto/caam/jr.c | 23 ++++++++++++++++++++---
 1 file changed, 20 insertions(+), 3 deletions(-)

diff --git a/drivers/crypto/caam/jr.c b/drivers/crypto/caam/jr.c
index f9e029945b51..2ed8229d4733 100644
--- a/drivers/crypto/caam/jr.c
+++ b/drivers/crypto/caam/jr.c
@@ -583,12 +583,29 @@ static int caam_jr_probe(struct platform_device *pdev)
 	struct caam_drv_private_jr *jrpriv;
 	static int total_jobrs;
 	void __iomem *ctrl;
+	struct resource *r;
 	int error;
 	int irq;
 
-	ctrl = devm_platform_ioremap_resource(pdev, 0);
-	if (IS_ERR(ctrl))
-		return PTR_ERR(ctrl);
+	/*
+	 * The job rings live inside the register window of their parent
+	 * fsl,sec-v4.0 node, which caam_probe() already reserves (and maps)
+	 * via devm_of_iomap().  A requested region that overlaps that
+	 * reservation, e.g. from devm_platform_ioremap_resource(), would
+	 * therefore fail with -EBUSY, so map the registers without claiming
+	 * the region here.
+	 */
+	r = platform_get_resource(pdev, IORESOURCE_MEM, 0);
+	if (!r) {
+		dev_err(&pdev->dev, "platform_get_resource() failed\n");
+		return -EINVAL;
+	}
+
+	ctrl = devm_ioremap(&pdev->dev, r->start, resource_size(r));
+	if (!ctrl) {
+		dev_err(&pdev->dev, "devm_ioremap() failed\n");
+		return -ENOMEM;
+	}
 
 	irq = platform_get_irq(pdev, 0);
 	if (irq < 0)
-- 
2.55.0

Re: [PATCH] crypto: caam: map job ring registers without claiming region

Posted by Herbert Xu 1 week, 3 days ago

On Tue, Sep 01, 2026 at 01:37:36PM -0700, Rosen Penev wrote:
> devm_platform_ioremap_resource() ends up calling request_mem_region(),
> which fails with -EBUSY on i.MX SoCs:
> 
> There the job rings are modelled as sub-regions of their parent
> fsl,sec-v4.0 register window, and caam_probe() already reserves the
> whole window exclusively via devm_of_iomap() before the children are
> populated.  Every job ring therefore collides with its own parent and
> fails to probe, taking the hardware RNG offline (seen on i.MX6, i.MX7
> and i.MX8 boards such as colibri-imx7 and verdin-imx8mp).
> 
> Map the job ring registers with devm_ioremap() instead, which does not
> claim the (already owned) region.
> 
> Fixes: 9a955c0a7d11 ("crypto: caam - simplify probe resource and IRQ handling")
> Assisted-by: opencode:big-pickle
> Signed-off-by: Rosen Penev <rosenp@gmail.com>
> ---
>  drivers/crypto/caam/jr.c | 23 ++++++++++++++++++++---
>  1 file changed, 20 insertions(+), 3 deletions(-)

Patch applied.  Thanks.
-- 
Email: Herbert Xu <herbert@gondor.apana.org.au>
Home Page: http://gondor.apana.org.au/~herbert/
PGP Key: http://gondor.apana.org.au/~herbert/pubkey.txt

Re: [PATCH] crypto: caam: map job ring registers without claiming region

Posted by Richard Leitner 2 weeks, 4 days ago

Hi Rosen,

On Tue, Sep 01, 2026 at 01:37:36PM -0700, Rosen Penev wrote:
> devm_platform_ioremap_resource() ends up calling request_mem_region(),
> which fails with -EBUSY on i.MX SoCs:
> 
> There the job rings are modelled as sub-regions of their parent
> fsl,sec-v4.0 register window, and caam_probe() already reserves the
> whole window exclusively via devm_of_iomap() before the children are
> populated.  Every job ring therefore collides with its own parent and
> fails to probe, taking the hardware RNG offline (seen on i.MX6, i.MX7
> and i.MX8 boards such as colibri-imx7 and verdin-imx8mp).
> 
> Map the job ring registers with devm_ioremap() instead, which does not
> claim the (already owned) region.
> 
> Fixes: 9a955c0a7d11 ("crypto: caam - simplify probe resource and IRQ handling")
> Assisted-by: opencode:big-pickle
> Signed-off-by: Rosen Penev <rosenp@gmail.com>

thanks for that fix!

Tested-by: Richard Leitner <richard.leitner@linux.dev> # i.MX8MP

regards;rl

> ---
>  drivers/crypto/caam/jr.c | 23 ++++++++++++++++++++---
>  1 file changed, 20 insertions(+), 3 deletions(-)
> 
> diff --git a/drivers/crypto/caam/jr.c b/drivers/crypto/caam/jr.c
> index f9e029945b51..2ed8229d4733 100644
> --- a/drivers/crypto/caam/jr.c
> +++ b/drivers/crypto/caam/jr.c
> @@ -583,12 +583,29 @@ static int caam_jr_probe(struct platform_device *pdev)
>  	struct caam_drv_private_jr *jrpriv;
>  	static int total_jobrs;
>  	void __iomem *ctrl;
> +	struct resource *r;
>  	int error;
>  	int irq;
>  
> -	ctrl = devm_platform_ioremap_resource(pdev, 0);
> -	if (IS_ERR(ctrl))
> -		return PTR_ERR(ctrl);
> +	/*
> +	 * The job rings live inside the register window of their parent
> +	 * fsl,sec-v4.0 node, which caam_probe() already reserves (and maps)
> +	 * via devm_of_iomap().  A requested region that overlaps that
> +	 * reservation, e.g. from devm_platform_ioremap_resource(), would
> +	 * therefore fail with -EBUSY, so map the registers without claiming
> +	 * the region here.
> +	 */
> +	r = platform_get_resource(pdev, IORESOURCE_MEM, 0);
> +	if (!r) {
> +		dev_err(&pdev->dev, "platform_get_resource() failed\n");
> +		return -EINVAL;
> +	}
> +
> +	ctrl = devm_ioremap(&pdev->dev, r->start, resource_size(r));
> +	if (!ctrl) {
> +		dev_err(&pdev->dev, "devm_ioremap() failed\n");
> +		return -ENOMEM;
> +	}
>  
>  	irq = platform_get_irq(pdev, 0);
>  	if (irq < 0)
> -- 
> 2.55.0
> 
>

Re: [PATCH] crypto: caam: map job ring registers without claiming region

Posted by Richard Leitner 1 week, 3 days ago

Hi everybody,

just a friendly reminder that this patch is now on the mailing list for
14 days and was not included in the 2 rc's which were released since then.

As this essentially breaks CAAM on multiple i.MX SoC variants, is there
any possibility to get this included in v7.3-rc4?

Would be great :-)

Thanks!

regards;rl


On Mon, Sep 07, 2026 at 05:35:51PM +0200, Richard Leitner wrote:
> Hi Rosen,
> 
> On Tue, Sep 01, 2026 at 01:37:36PM -0700, Rosen Penev wrote:
> > devm_platform_ioremap_resource() ends up calling request_mem_region(),
> > which fails with -EBUSY on i.MX SoCs:
> > 
> > There the job rings are modelled as sub-regions of their parent
> > fsl,sec-v4.0 register window, and caam_probe() already reserves the
> > whole window exclusively via devm_of_iomap() before the children are
> > populated.  Every job ring therefore collides with its own parent and
> > fails to probe, taking the hardware RNG offline (seen on i.MX6, i.MX7
> > and i.MX8 boards such as colibri-imx7 and verdin-imx8mp).
> > 
> > Map the job ring registers with devm_ioremap() instead, which does not
> > claim the (already owned) region.
> > 
> > Fixes: 9a955c0a7d11 ("crypto: caam - simplify probe resource and IRQ handling")
> > Assisted-by: opencode:big-pickle
> > Signed-off-by: Rosen Penev <rosenp@gmail.com>
> 
> thanks for that fix!
> 
> Tested-by: Richard Leitner <richard.leitner@linux.dev> # i.MX8MP
> 
> regards;rl
> 
> > ---
> >  drivers/crypto/caam/jr.c | 23 ++++++++++++++++++++---
> >  1 file changed, 20 insertions(+), 3 deletions(-)
> > 
> > diff --git a/drivers/crypto/caam/jr.c b/drivers/crypto/caam/jr.c
> > index f9e029945b51..2ed8229d4733 100644
> > --- a/drivers/crypto/caam/jr.c
> > +++ b/drivers/crypto/caam/jr.c
> > @@ -583,12 +583,29 @@ static int caam_jr_probe(struct platform_device *pdev)
> >  	struct caam_drv_private_jr *jrpriv;
> >  	static int total_jobrs;
> >  	void __iomem *ctrl;
> > +	struct resource *r;
> >  	int error;
> >  	int irq;
> >  
> > -	ctrl = devm_platform_ioremap_resource(pdev, 0);
> > -	if (IS_ERR(ctrl))
> > -		return PTR_ERR(ctrl);
> > +	/*
> > +	 * The job rings live inside the register window of their parent
> > +	 * fsl,sec-v4.0 node, which caam_probe() already reserves (and maps)
> > +	 * via devm_of_iomap().  A requested region that overlaps that
> > +	 * reservation, e.g. from devm_platform_ioremap_resource(), would
> > +	 * therefore fail with -EBUSY, so map the registers without claiming
> > +	 * the region here.
> > +	 */
> > +	r = platform_get_resource(pdev, IORESOURCE_MEM, 0);
> > +	if (!r) {
> > +		dev_err(&pdev->dev, "platform_get_resource() failed\n");
> > +		return -EINVAL;
> > +	}
> > +
> > +	ctrl = devm_ioremap(&pdev->dev, r->start, resource_size(r));
> > +	if (!ctrl) {
> > +		dev_err(&pdev->dev, "devm_ioremap() failed\n");
> > +		return -ENOMEM;
> > +	}
> >  
> >  	irq = platform_get_irq(pdev, 0);
> >  	if (irq < 0)
> > -- 
> > 2.55.0
> > 
> >

Re: [PATCH] crypto: caam: map job ring registers without claiming region

Posted by Francesco Dolcini 3 weeks, 3 days ago

+ Emanuele

On Tue, Sep 01, 2026 at 01:37:36PM -0700, Rosen Penev wrote:
> devm_platform_ioremap_resource() ends up calling request_mem_region(),
> which fails with -EBUSY on i.MX SoCs:
> 
> There the job rings are modelled as sub-regions of their parent
> fsl,sec-v4.0 register window, and caam_probe() already reserves the
> whole window exclusively via devm_of_iomap() before the children are
> populated.  Every job ring therefore collides with its own parent and
> fails to probe, taking the hardware RNG offline (seen on i.MX6, i.MX7
> and i.MX8 boards such as colibri-imx7 and verdin-imx8mp).
> 
> Map the job ring registers with devm_ioremap() instead, which does not
> claim the (already owned) region.
> 
> Fixes: 9a955c0a7d11 ("crypto: caam - simplify probe resource and IRQ handling")
> Assisted-by: opencode:big-pickle
> Signed-off-by: Rosen Penev <rosenp@gmail.com>

Reported-by: Emanuele Ghidoli <ghidoliemanuele@gmail.com>
Closes: https://lore.kernel.org/lkml/20260901194524.908240-1-ghidoliemanuele@gmail.com/

Thanks,
Francesco

Re: [PATCH] crypto: caam: map job ring registers without claiming region

Posted by Emanuele Ghidoli 3 weeks, 3 days ago


On 9/2/26 08:10, Francesco Dolcini wrote:
> + Emanuele
> 
> On Tue, Sep 01, 2026 at 01:37:36PM -0700, Rosen Penev wrote:
>> devm_platform_ioremap_resource() ends up calling request_mem_region(),
>> which fails with -EBUSY on i.MX SoCs:
>>
>> There the job rings are modelled as sub-regions of their parent
>> fsl,sec-v4.0 register window, and caam_probe() already reserves the
>> whole window exclusively via devm_of_iomap() before the children are
>> populated.  Every job ring therefore collides with its own parent and
>> fails to probe, taking the hardware RNG offline (seen on i.MX6, i.MX7
>> and i.MX8 boards such as colibri-imx7 and verdin-imx8mp).
>>
>> Map the job ring registers with devm_ioremap() instead, which does not
>> claim the (already owned) region.
>>
>> Fixes: 9a955c0a7d11 ("crypto: caam - simplify probe resource and IRQ handling")
>> Assisted-by: opencode:big-pickle
>> Signed-off-by: Rosen Penev <rosenp@gmail.com>
> 
> Reported-by: Emanuele Ghidoli <ghidoliemanuele@gmail.com>
> Closes: https://lore.kernel.org/lkml/20260901194524.908240-1-ghidoliemanuele@gmail.com/
> 
> Thanks,
> Francesco

Tested on colibri-imx7.

Tested-by: Emanuele Ghidoli <emanuele.ghidoli@toradex.com>

Thanks,
Emanuele